Re: [PATCH 2/3] drm/amdgpu: add AMDGPU_VM_NOALLOC
Ok that sounds like a rather simple bug. I will try to take a look. Thanks, Christian. Am 17.05.22 um 02:12 schrieb Marek Olšák: Dmesg doesn't contain anything. There is no backtrace because it's not a crash. The VA map ioctl just fails with the new flag. It looks like the flag is considered invalid. Marek On Mon., May 16, 2022, 12:13 Christian König, wrote: I don't have access to any gfx10 hardware. Can you give me a dmesg and/or backtrace, etc..? I can't push this unless it's working properly. Christian. Am 16.05.22 um 14:56 schrieb Marek Olšák: Reproduction steps: - use mesa/main on gfx10.3 (not sure what other GPUs do) - run: radeonsi_mall_noalloc=true glxgears Marek On Mon, May 16, 2022 at 7:53 AM Christian König wrote: Crap, do you have a link to the failure? Am 16.05.22 um 13:10 schrieb Marek Olšák: I forgot to say: The NOALLOC flag causes an allocation failure, so there is a kernel bug somewhere. Marek On Mon, May 16, 2022 at 7:06 AM Marek Olšák wrote: FYI, I think it's time to merge this because the Mesa commits are going to be merged in ~30 minutes if Gitlab CI is green, and that includes updated amdgpu_drm.h. Marek On Wed, May 11, 2022 at 2:55 PM Marek Olšák wrote: Ok sounds good. Marek On Wed., May 11, 2022, 03:43 Christian König, wrote: It really *is* a NOALLOC feature. In other words there is no latency improvement on reads because the cache is always checked, even with the noalloc flag set. The only thing it affects is that misses not enter the cache and so don't cause any additional pressure on evicting cache lines. You might want to double check with the hardware guys, but I'm something like 95% sure that it works this way. Christian. Am 11.05.22 um 09:22 schrieb Marek Olšák: Bypass means that the contents of the cache are ignored, which decreases latency at the cost of no coherency between bypassed and normal memory requests. NOA (noalloc) means that the cache is checked and can give you cache hits, but misses are not cached and the overall latency is higher. I don't know what the hw does, but I hope it was misnamed and it really means bypass because there is no point in doing cache lookups on every memory request if the driver wants to disable caching to *decrease* latency in the situations when the cache isn't helping. Marek On Wed, May 11, 2022 at 2:15 AM Lazar, Lijo wrote: On 5/11/2022 11:36 AM, Christian König wrote: > Mhm, it doesn't really bypass MALL. It just doesn't allocate any MALL > entries on write. > > How about AMDGPU_VM_PAGE_NO_MALL ? One more - AMDGPU_VM_PAGE_LLC_* [ LLC = last level cache, * = some sort of attribute which decides LLC behaviour] Thanks, Lijo > > Christian. > > Am 10.05.22 um 23:21 schrieb Marek Olšák: >> A better name would be: >> AMDGPU_VM_PAGE_BYPASS_MALL >> >> Marek >> >> On Fri, May 6, 2022 at 7:23 AM Christian König >> wrote: >> >> Add the AMDGPU_VM_NOALLOC flag to let userspace control MALL >> allocation. >> >> Only compile tested! >> >> Signed-off-by: Christian König >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++ >> drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +++ >> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 3 +++ >> include/uapi/drm/amdgpu_
[PATCH] drm/amdgpu: Set CP_HQD_PQ_CONTROL_RPTR_BLOCK_SIZE correctly in gfx_v8-v11.
From: Haohui Mai Remove the accidental shifts on the values of RPTR_BLOCK_SIZE in gfx_v8-v11. The bug essentially always programs the corresponding fields to zero instead of the correct value. Signed-off-by: Haohui Mai --- drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 2 +- 4 files changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c index 12f63d6aab12..dd8f4344eeb8 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c @@ -6898,7 +6898,7 @@ static int gfx_v10_0_compute_mqd_init(struct amdgpu_device *adev, void *m, tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, (order_base_2(prop->queue_size / 4) - 1)); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, - ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8)); + (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1)); #ifdef __BIG_ENDIAN tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index 8a1bec70c719..82c98e59a13e 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -4082,7 +4082,7 @@ static int gfx_v11_0_compute_mqd_init(struct amdgpu_device *adev, void *m, tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, (order_base_2(prop->queue_size / 4) - 1)); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, - ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8)); + (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1)); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 90f64219d291..fb9302910742 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -4490,7 +4490,7 @@ static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring) tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, (order_base_2(ring->ring_size / 4) - 1)); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, - ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8)); + (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1)); #ifdef __BIG_ENDIAN tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index f49a2dd89ee7..f12ae6e2359a 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -3514,7 +3514,7 @@ static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring) tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, (order_base_2(ring->ring_size / 4) - 1)); tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, - ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8)); + (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1)); #ifdef __BIG_ENDIAN tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1); #endif -- 2.25.1
[PATCH 6/6] drm/amd/pm: enable memory temp reading for SMU 13.0.0
With the latest vbios, the memory temp reading is working. Signed-off-by: Evan Quan Reviewed-by: Hawking Zhang Change-Id: I3b4679f03e5ff7cf8e0b68c095d210e608149fcb --- drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c index 7bb2923eb819..7432b3e76d3d 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c @@ -275,9 +275,7 @@ smu_v13_0_0_get_allowed_feature_mask(struct smu_context *smu, *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_VDDIO_MEM_SCALING_BIT); } -#if 0 *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_MEM_TEMP_READ_BIT); -#endif if (adev->pm.pp_feature & PP_SCLK_DEEP_SLEEP_MASK) *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_DS_GFXCLK_BIT); -- 2.29.0
[PATCH 5/6] drm/amd/pm: enable more dpm features for SMU 13.0.0
Enable OOB Monitor and SOC CG which are ready since 78.38.0. Signed-off-by: Evan Quan Reviewed-by: Hawking Zhang Acked-by: Alex Deucher Change-Id: I0d7334106917ac83fff2b673ec7e9eb096089afe --- drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c index 7bfceca246ae..7bb2923eb819 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c @@ -299,6 +299,9 @@ smu_v13_0_0_get_allowed_feature_mask(struct smu_context *smu, *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_DPM_MP0CLK_BIT); *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_FW_DSTATE_BIT); + *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_OUT_OF_BAND_MONITOR_BIT); + *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_SOC_CG_BIT); + return 0; } -- 2.29.0
[PATCH 3/6] drm/amd/pm: update SMU 13.0.0 driver_if header
To align with 78.37.0 and later PMFWs. Signed-off-by: Evan Quan Reviewed-by: Hawking Zhang Change-Id: I17c3a6b466c921cb5ffe5025a94023fae478c80e --- .../inc/pmfw_if/smu13_driver_if_v13_0_0.h | 22 +-- drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h | 2 +- 2 files changed, 11 insertions(+), 13 deletions(-) diff --git a/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_v13_0_0.h b/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_v13_0_0.h index ecc6411dfc8d..c1f76236da26 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_v13_0_0.h +++ b/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_v13_0_0.h @@ -671,8 +671,8 @@ typedef struct { uint16_t reserved[2]; //Frequency changes - uint16_t GfxclkFmin; // MHz - uint16_t GfxclkFmax; // MHz + int16_tGfxclkFmin; // MHz + int16_tGfxclkFmax; // MHz uint16_t UclkFmin; // MHz uint16_t UclkFmax; // MHz @@ -683,15 +683,14 @@ typedef struct { //Fan control uint8_tFanLinearPwmPoints[NUM_OD_FAN_MAX_POINTS]; uint8_tFanLinearTempPoints[NUM_OD_FAN_MAX_POINTS]; - uint16_t FanMaximumRpm; uint16_t FanMinimumPwm; - uint16_t FanAcousticLimitRpm; + uint16_t AcousticTargetRpmThreshold; + uint16_t AcousticLimitRpmThreshold; uint16_t FanTargetTemperature; // Degree Celcius uint8_tFanZeroRpmEnable; uint8_tFanZeroRpmStopTemp; uint8_tFanMode; - uint8_tPadding[1]; - + uint8_tMaxOpTemp; uint32_t Spare[13]; uint32_t MmHubPadding[8]; // SMU internal use. Adding here instead of external as a workaround @@ -719,15 +718,14 @@ typedef struct { uint8_tFanLinearPwmPoints; uint8_tFanLinearTempPoints; - uint16_t FanMaximumRpm; uint16_t FanMinimumPwm; - uint16_t FanAcousticLimitRpm; + uint16_t AcousticTargetRpmThreshold; + uint16_t AcousticLimitRpmThreshold; uint16_t FanTargetTemperature; // Degree Celcius uint8_tFanZeroRpmEnable; uint8_tFanZeroRpmStopTemp; uint8_tFanMode; - uint8_tPadding[1]; - + uint8_tMaxOpTemp; uint32_t Spare[13]; @@ -997,7 +995,8 @@ typedef struct { uint16_t SocketPowerLimitAcTau[PPT_THROTTLER_COUNT]; // Time constant of LPF in ms uint16_t SocketPowerLimitDcTau[PPT_THROTTLER_COUNT]; // Time constant of LPF in ms - uint32_t SpareVmin[12]; + QuadraticInt_t Vmin_droop; + uint32_t SpareVmin[9]; //SECTION: DPM Configuration 1 @@ -1286,7 +1285,6 @@ typedef struct { uint32_tPostVoltageSetBacoDelay; // in microseconds. Amount of time FW will wait after power good is established or PSI0 command is issued uint32_tBacoEntryDelay; // in milliseconds. Amount of time FW will wait to trigger BACO entry after receiving entry notification from OS - // SECTION: Board Reserved uint32_t BoardSpare[64]; diff --git a/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h b/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h index 2b44d41a5157..afa1991e26f9 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h +++ b/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h @@ -30,7 +30,7 @@ #define SMU13_DRIVER_IF_VERSION_ALDE 0x08 #define SMU13_DRIVER_IF_VERSION_SMU_V13_0_4 0x04 #define SMU13_DRIVER_IF_VERSION_SMU_V13_0_5 0x04 -#define SMU13_DRIVER_IF_VERSION_SMU_V13_0_0 0x27 +#define SMU13_DRIVER_IF_VERSION_SMU_V13_0_0 0x28 #define SMU13_DRIVER_IF_VERSION_SMU_V13_0_7 0x28 #define SMU13_MODE1_RESET_WAIT_TIME_IN_MS 500 //500ms -- 2.29.0
[PATCH 2/6] drm/amd/pm: skip dpm disablement on suspend for SMU 13.0.0
Since PMFW will handle this properly. Driver involvement is unnecessary. Signed-off-by: Evan Quan Reviewed-by: Hawking Zhang Change-Id: I19be26eed090d57040553f5cdff9534072f08106 --- drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c index 6016b325b6b5..a601024ba4de 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c +++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c @@ -1436,6 +1436,7 @@ static int smu_disable_dpms(struct smu_context *smu) case IP_VERSION(11, 0, 0): case IP_VERSION(11, 0, 5): case IP_VERSION(11, 0, 9): + case IP_VERSION(13, 0, 0): return 0; default: break; -- 2.29.0
[PATCH 4/6] drm/amd/pm: correct the softpptable ids used for SMU 13.0.0
To better match with the pptable_id settings from VBIOS. Signed-off-by: Evan Quan Reviewed-by: Hawking Zhang Acked-by: Alex Deucher Change-Id: I3379443067a5df3a2fb04ff1bc48e8c8f28e1c66 --- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c| 47 ++- 1 file changed, 35 insertions(+), 12 deletions(-) diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c index ae6321af9d88..7be4f6875a7b 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c @@ -218,13 +218,25 @@ int smu_v13_0_init_pptable_microcode(struct smu_context *smu) pptable_id == 3688) pptable_id = 36881; /* -* Temporary solution for SMU V13.0.0: -* - use 1 signed pptable when SCPM enabled -* TODO: drop this when the pptable carried in vbios -* is ready. +* Temporary solution for SMU V13.0.0 with SCPM enabled: +* - use 36831 signed pptable when pp_table_id is 3683 +* - use 36641 signed pptable when pp_table_id is 3664 or 0 +* TODO: drop these when the pptable carried in vbios is ready. */ - if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 0)) - pptable_id = 1; + if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 0)) { + switch (pptable_id) { + case 0: + case 3664: + pptable_id = 36641; + break; + case 3683: + pptable_id = 36831; + break; + default: + dev_err(adev->dev, "Unsupported pptable id %d\n", pptable_id); + return -EINVAL; + } + } } /* "pptable_id == 0" means vbios carries the pptable. */ @@ -448,13 +460,24 @@ int smu_v13_0_setup_pptable(struct smu_context *smu) pptable_id = smu->smu_table.boot_values.pp_table_id; /* -* Temporary solution for SMU V13.0.0: -* - use unsigned pptable when SCPM disabled -* TODO: drop this when the pptable carried in vbios -* is ready. +* Temporary solution for SMU V13.0.0 with SCPM disabled: +* - use 3664 or 3683 on request +* - use 3664 when pptable_id is 0 +* TODO: drop these when the pptable carried in vbios is ready. */ - if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 0)) - pptable_id = ; + if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 0)) { + switch (pptable_id) { + case 0: + pptable_id = 3664; + break; + case 3664: + case 3683: + break; + default: + dev_err(adev->dev, "Unsupported pptable id %d\n", pptable_id); + return -EINVAL; + } + } } /* force using vbios pptable in sriov mode */ -- 2.29.0
[PATCH 1/6] drm/amd/pm: enable more dpm features for SMU 13.0.0
Enable MP0CLK DPM and FW Dstate since they are already supported by latest 78.36.0 PMFW. Signed-off-by: Evan Quan Reviewed-by: Hawking Zhang Change-Id: I75b6129dab509a51ccaf92bbc0b094eae41ea20f --- drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c index 197a0e2ff063..7bfceca246ae 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c @@ -296,6 +296,9 @@ smu_v13_0_0_get_allowed_feature_mask(struct smu_context *smu, *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_BACO_BIT); + *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_DPM_MP0CLK_BIT); + *(uint64_t *)feature_mask |= FEATURE_MASK(FEATURE_FW_DSTATE_BIT); + return 0; } -- 2.29.0
RE: [PATCH] drm/amdgpu/discovery: validate VCN and SDMA instances
Reviewed-by: Guchun Chen Regards, Guchun -Original Message- From: amd-gfx On Behalf Of Alex Deucher Sent: Tuesday, May 17, 2022 2:17 AM To: amd-gfx@lists.freedesktop.org Cc: Deucher, Alexander Subject: [PATCH] drm/amdgpu/discovery: validate VCN and SDMA instances Validate the VCN and SDMA instances against the driver structure sizes to make sure we don't get into a situation where the firmware reports more instances than the driver supports. Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 17 ++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c index 1f4e07a32445..2c4f1adb5343 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c @@ -1137,13 +1137,24 @@ static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev) adev->vcn.vcn_config[adev->vcn.num_vcn_inst] = ip->revision & 0xc0; ip->revision &= ~0xc0; - adev->vcn.num_vcn_inst++; + if (adev->vcn.num_vcn_inst < AMDGPU_MAX_VCN_INSTANCES) + adev->vcn.num_vcn_inst++; + else + dev_err(adev->dev, "Too many VCN instances: %d vs %d\n", + adev->vcn.num_vcn_inst + 1, + AMDGPU_MAX_VCN_INSTANCES); } if (le16_to_cpu(ip->hw_id) == SDMA0_HWID || le16_to_cpu(ip->hw_id) == SDMA1_HWID || le16_to_cpu(ip->hw_id) == SDMA2_HWID || - le16_to_cpu(ip->hw_id) == SDMA3_HWID) - adev->sdma.num_instances++; + le16_to_cpu(ip->hw_id) == SDMA3_HWID) { + if (adev->sdma.num_instances < AMDGPU_MAX_SDMA_INSTANCES) + adev->sdma.num_instances++; + else + dev_err(adev->dev, "Too many SDMA instances: %d vs %d\n", + adev->sdma.num_instances + 1, + AMDGPU_MAX_SDMA_INSTANCES); + } if (le16_to_cpu(ip->hw_id) == UMC_HWID) adev->gmc.num_umc++; -- 2.35.3
[linux-next:master] BUILD REGRESSION 3f7bdc402fb06f897ff1f492a2d42e1f7c2efedb
tree/branch: https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git master branch HEAD: 3f7bdc402fb06f897ff1f492a2d42e1f7c2efedb Add linux-next specific files for 20220516 Error/Warning reports: https://lore.kernel.org/linux-mm/202204291924.vtgzmeri-...@intel.com https://lore.kernel.org/linux-mm/202205041248.wgcwpcev-...@intel.com https://lore.kernel.org/linux-mm/202205122113.ulkzd3sz-...@intel.com https://lore.kernel.org/linux-mm/202205150051.3rzuooag-...@intel.com https://lore.kernel.org/linux-mm/202205150117.sd6hzbvm-...@intel.com https://lore.kernel.org/llvm/202205170327.tvbbish2-...@intel.com https://lore.kernel.org/llvm/202205170352.5yjubp5h-...@intel.com Error/Warning: (recently discovered and may have been fixed) : fatal error: ./include/generated/utsrelease.h: No such file or directory ERROR: modpost: "__udivdi3" [drivers/mtd/parsers/scpart.ko] undefined! arch/riscv/include/asm/tlbflush.h:23:2: error: expected assembly-time absolute expression arch/x86/kvm/pmu.h:20:32: warning: 'vmx_icl_pebs_cpu' defined but not used [-Wunused-const-variable=] drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1364:5: warning: no previous prototype for 'amdgpu_discovery_get_mall_info' [-Wmissing-prototypes] drivers/gpu/drm/amd/amdgpu/soc21.c:171:6: warning: no previous prototype for 'soc21_grbm_select' [-Wmissing-prototypes] drivers/gpu/drm/solomon/ssd130x-spi.c:154:35: warning: 'ssd130x_spi_table' defined but not used [-Wunused-const-variable=] drivers/video/fbdev/omap/hwa742.c:492:5: warning: no previous prototype for 'hwa742_update_window_async' [-Wmissing-prototypes] fs/buffer.c:2254:5: warning: stack frame size (2152) exceeds limit (1024) in 'block_read_full_folio' [-Wframe-larger-than] fs/ntfs/aops.c:378:12: warning: stack frame size (2216) exceeds limit (1024) in 'ntfs_read_folio' [-Wframe-larger-than] kernel/trace/fgraph.c:37:12: warning: no previous prototype for 'ftrace_enable_ftrace_graph_caller' [-Wmissing-prototypes] kernel/trace/fgraph.c:46:12: warning: no previous prototype for 'ftrace_disable_ftrace_graph_caller' [-Wmissing-prototypes] Unverified Error/Warning (likely false positive, please contact us if interested): Error: Section .bss not empty in prom_init.c Makefile:686: arch/h8300/Makefile: No such file or directory arch/Kconfig:10: can't open file "arch/h8300/Kconfig" arch/arm64/kvm/pmu.c:9:46: warning: tentative definition of variable with internal linkage has incomplete non-array type 'typeof(struct kvm_pmu_events)' (aka 'struct kvm_pmu_events') [-Wtentative-definition-incomplete-type] drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_link_dp.c:5102:14: warning: variable 'allow_lttpr_non_transparent_mode' set but not used [-Wunused-but-set-variable] drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_link_dp.c:5147:6: warning: no previous prototype for 'dp_parse_lttpr_mode' [-Wmissing-prototypes] drivers/gpu/drm/bridge/adv7511/adv7511.h:229:17: warning: 'ADV7511_REG_CEC_RX_FRAME_HDR' defined but not used [-Wunused-const-variable=] drivers/gpu/drm/bridge/adv7511/adv7511.h:235:17: warning: 'ADV7511_REG_CEC_RX_FRAME_LEN' defined but not used [-Wunused-const-variable=] drivers/gpu/drm/i915/gt/intel_gt_sysfs_pm.c:46 sysfs_gt_attribute_w_func() error: uninitialized symbol 'ret'. drivers/opp/core.c:1499:13: warning: Uninitialized variable: iter->rate [uninitvar] drivers/opp/core.c:1533:14: warning: Uninitialized variable: temp->removed [uninitvar] drivers/opp/core.c:2682:16: warning: Uninitialized variable: tmp_opp->rate [uninitvar] drivers/opp/core.c:365:12: warning: Uninitialized variable: opp->available [uninitvar] drivers/opp/core.c:442:17: warning: Uninitialized variable: temp_opp->available [uninitvar] drivers/opp/core.c:491:17: warning: Uninitialized variable: temp_opp->level [uninitvar] drivers/opp/core.c:60:26: warning: Uninitialized variables: opp_table.node, opp_table.lazy, opp_table.head, opp_table.dev_list, opp_table.opp_list, opp_table.kref, opp_table.lock, opp_table.np, opp_table.clock_latency_ns_max, opp_table.voltage_tolerance_v1, opp_table.parsed_static_opps, opp_table.shared_opp, opp_table.current_rate, opp_table.current_opp, opp_table.suspend_opp, opp_table.genpd_virt_dev_lock, opp_table.genpd_virt_devs, opp_table.required_opp_tables, opp_table.required_opp_count, opp_table.supported_hw, opp_table.supported_hw_count, opp_table.prop_name, opp_table.clk, opp_table.regulators, opp_table.regulator_count, opp_table.paths, opp_table.path_count, opp_table.enabled, opp_table.genpd_performance_state, opp_table.is_genpd, opp_table.set_opp, opp_table.sod_supplies, opp_table.set_opp_data [uninitvar] kernel/bpf/verifier.c:5354 process_kptr_func() warn: passing zero to 'PTR_ERR' make[1]: *** No rule to make target 'arch/h830
Re: [PATCH 2/3] drm/amdgpu: add AMDGPU_VM_NOALLOC
Dmesg doesn't contain anything. There is no backtrace because it's not a crash. The VA map ioctl just fails with the new flag. It looks like the flag is considered invalid. Marek On Mon., May 16, 2022, 12:13 Christian König, < ckoenig.leichtzumer...@gmail.com> wrote: > I don't have access to any gfx10 hardware. > > Can you give me a dmesg and/or backtrace, etc..? > > I can't push this unless it's working properly. > > Christian. > > Am 16.05.22 um 14:56 schrieb Marek Olšák: > > Reproduction steps: > - use mesa/main on gfx10.3 (not sure what other GPUs do) > - run: radeonsi_mall_noalloc=true glxgears > > Marek > > On Mon, May 16, 2022 at 7:53 AM Christian König < > ckoenig.leichtzumer...@gmail.com> wrote: > >> Crap, do you have a link to the failure? >> >> Am 16.05.22 um 13:10 schrieb Marek Olšák: >> >> I forgot to say: The NOALLOC flag causes an allocation failure, so there >> is a kernel bug somewhere. >> >> Marek >> >> On Mon, May 16, 2022 at 7:06 AM Marek Olšák wrote: >> >>> FYI, I think it's time to merge this because the Mesa commits are going >>> to be merged in ~30 minutes if Gitlab CI is green, and that includes >>> updated amdgpu_drm.h. >>> >>> Marek >>> >>> On Wed, May 11, 2022 at 2:55 PM Marek Olšák wrote: >>> Ok sounds good. Marek On Wed., May 11, 2022, 03:43 Christian König, < ckoenig.leichtzumer...@gmail.com> wrote: > It really *is* a NOALLOC feature. In other words there is no latency > improvement on reads because the cache is always checked, even with the > noalloc flag set. > > The only thing it affects is that misses not enter the cache and so > don't cause any additional pressure on evicting cache lines. > > You might want to double check with the hardware guys, but I'm > something like 95% sure that it works this way. > > Christian. > > Am 11.05.22 um 09:22 schrieb Marek Olšák: > > Bypass means that the contents of the cache are ignored, which > decreases latency at the cost of no coherency between bypassed and normal > memory requests. NOA (noalloc) means that the cache is checked and can > give > you cache hits, but misses are not cached and the overall latency is > higher. I don't know what the hw does, but I hope it was misnamed and it > really means bypass because there is no point in doing cache lookups on > every memory request if the driver wants to disable caching to *decrease* > latency in the situations when the cache isn't helping. > > Marek > > On Wed, May 11, 2022 at 2:15 AM Lazar, Lijo > wrote: > >> >> >> On 5/11/2022 11:36 AM, Christian König wrote: >> > Mhm, it doesn't really bypass MALL. It just doesn't allocate any >> MALL >> > entries on write. >> > >> > How about AMDGPU_VM_PAGE_NO_MALL ? >> >> One more - AMDGPU_VM_PAGE_LLC_* [ LLC = last level cache, * = some >> sort >> of attribute which decides LLC behaviour] >> >> Thanks, >> Lijo >> >> > >> > Christian. >> > >> > Am 10.05.22 um 23:21 schrieb Marek Olšák: >> >> A better name would be: >> >> AMDGPU_VM_PAGE_BYPASS_MALL >> >> >> >> Marek >> >> >> >> On Fri, May 6, 2022 at 7:23 AM Christian König >> >> wrote: >> >> >> >> Add the AMDGPU_VM_NOALLOC flag to let userspace control MALL >> >> allocation. >> >> >> >> Only compile tested! >> >> >> >> Signed-off-by: Christian König >> >> --- >> >> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++ >> >> drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +++ >> >> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 3 +++ >> >> include/uapi/drm/amdgpu_drm.h | 2 ++ >> >> 4 files changed, 10 insertions(+) >> >> >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> >> index bf97d8f07f57..d8129626581f 100644 >> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> >> @@ -650,6 +650,8 @@ uint64_t amdgpu_gem_va_map_flags(struct >> >> amdgpu_device *adev, uint32_t flags) >> >> pte_flag |= AMDGPU_PTE_WRITEABLE; >> >> if (flags & AMDGPU_VM_PAGE_PRT) >> >> pte_flag |= AMDGPU_PTE_PRT; >> >> + if (flags & AMDGPU_VM_PAGE_NOALLOC) >> >> + pte_flag |= AMDGPU_PTE_NOALLOC; >> >> >> >> if (adev->gmc.gmc_funcs->map_mtype) >> >> pte_flag |= amdgpu_gmc_map_mtype(adev, >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> >> b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> >> index b8c79789e1e4..9077dfccaf3c 100644 >> >> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> >> +++ b/drivers/gpu/drm/amd/amdgpu/gmc
[PATCH v2 11/13] mm: handling Non-LRU pages returned by vm_normal_pages
With DEVICE_COHERENT, we'll soon have vm_normal_pages() return device-managed anonymous pages that are not LRU pages. Although they behave like normal pages for purposes of mapping in CPU page, and for COW. They do not support LRU lists, NUMA migration or THP. We also introduced a FOLL_LRU flag that adds the same behaviour to follow_page and related APIs, to allow callers to specify that they expect to put pages on an LRU list. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling --- fs/proc/task_mmu.c | 2 +- include/linux/mm.h | 3 ++- mm/gup.c | 2 ++ mm/huge_memory.c | 2 +- mm/khugepaged.c| 9 ++--- mm/ksm.c | 6 +++--- mm/madvise.c | 4 ++-- mm/memory.c| 9 - mm/mempolicy.c | 2 +- mm/migrate.c | 4 ++-- mm/mlock.c | 2 +- mm/mprotect.c | 2 +- 12 files changed, 30 insertions(+), 17 deletions(-) diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c index f46060eb91b5..5d620733f173 100644 --- a/fs/proc/task_mmu.c +++ b/fs/proc/task_mmu.c @@ -1785,7 +1785,7 @@ static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, return NULL; page = vm_normal_page(vma, addr, pte); - if (!page) + if (!page || is_zone_device_page(page)) return NULL; if (PageReserved(page)) diff --git a/include/linux/mm.h b/include/linux/mm.h index 9f44254af8ce..d7f253a0c41e 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -601,7 +601,7 @@ struct vm_operations_struct { #endif /* * Called by vm_normal_page() for special PTEs to find the -* page for @addr. This is useful if the default behavior +* page for @addr. This is useful if the default behavior * (using pte_page()) would not find the correct page. */ struct page *(*find_special_page)(struct vm_area_struct *vma, @@ -2929,6 +2929,7 @@ struct page *follow_page(struct vm_area_struct *vma, unsigned long address, #define FOLL_NUMA 0x200 /* force NUMA hinting page fault */ #define FOLL_MIGRATION 0x400 /* wait for page to replace migration entry */ #define FOLL_TRIED 0x800 /* a retry, previous pass started an IO */ +#define FOLL_LRU0x1000 /* return only LRU (anon or page cache) */ #define FOLL_REMOTE0x2000 /* we are working on non-current tsk/mm */ #define FOLL_COW 0x4000 /* internal GUP flag */ #define FOLL_ANON 0x8000 /* don't do file mappings */ diff --git a/mm/gup.c b/mm/gup.c index a214c8df7140..13bf04bce132 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -479,6 +479,8 @@ static struct page *follow_page_pte(struct vm_area_struct *vma, } page = vm_normal_page(vma, address, pte); + if ((flags & FOLL_LRU) && page && is_zone_device_page(page)) + page = NULL; if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { /* * Only return device mapping pages in the FOLL_GET or FOLL_PIN diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 910a138e9859..eed80696c5fd 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -2856,7 +2856,7 @@ static int split_huge_pages_pid(int pid, unsigned long vaddr_start, } /* FOLL_DUMP to ignore special (like zero) pages */ - page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP); + page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP | FOLL_LRU); if (IS_ERR(page)) continue; diff --git a/mm/khugepaged.c b/mm/khugepaged.c index a4e5eaf3eb01..eb3cfd679800 100644 --- a/mm/khugepaged.c +++ b/mm/khugepaged.c @@ -627,7 +627,7 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma, goto out; } page = vm_normal_page(vma, address, pteval); - if (unlikely(!page)) { + if (unlikely(!page) || unlikely(is_zone_device_page(page))) { result = SCAN_PAGE_NULL; goto out; } @@ -1276,7 +1276,7 @@ static int khugepaged_scan_pmd(struct mm_struct *mm, writable = true; page = vm_normal_page(vma, _address, pteval); - if (unlikely(!page)) { + if (unlikely(!page) || unlikely(is_zone_device_page(page))) { result = SCAN_PAGE_NULL; goto out_unmap; } @@ -1484,7 +1484,8 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr) goto abort; page = vm_normal_page(vma, addr, *pte); - + if (page && is_zone_device_page(page)) + page = NULL; /* * Note that uprobe, debugger, or MAP_PRIVATE may change the * page table, but the new page will not be a subpage of hpage. @@ -1502,6 +1503,8 @@ void collapse_p
[PATCH v2 13/13] tools: add selftests to hmm for COW in device memory
The objective is to test device migration mechanism in pages marked as COW, for private and coherent device type. In case of writing to COW private page(s), a page fault will migrate pages back to system memory first. Then, these pages will be duplicated. In case of COW device coherent type, pages are duplicated directly from device memory. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling --- tools/testing/selftests/vm/hmm-tests.c | 80 ++ 1 file changed, 80 insertions(+) diff --git a/tools/testing/selftests/vm/hmm-tests.c b/tools/testing/selftests/vm/hmm-tests.c index 61a02a6a3dea..76570030ba5b 100644 --- a/tools/testing/selftests/vm/hmm-tests.c +++ b/tools/testing/selftests/vm/hmm-tests.c @@ -1871,4 +1871,84 @@ TEST_F(hmm, hmm_gup_test) close(gup_fd); hmm_buffer_free(buffer); } + +/* + * Test copy-on-write in device pages. + * In case of writing to COW private page(s), a page fault will migrate pages + * back to system memory first. Then, these pages will be duplicated. In case + * of COW device coherent type, pages are duplicated directly from device + * memory. + */ +TEST_F(hmm, hmm_cow_in_device) +{ + struct hmm_buffer *buffer; + unsigned long npages; + unsigned long size; + unsigned long i; + int *ptr; + int ret; + unsigned char *m; + pid_t pid; + int status; + + npages = 4; + size = npages << self->page_shift; + + buffer = malloc(sizeof(*buffer)); + ASSERT_NE(buffer, NULL); + + buffer->fd = -1; + buffer->size = size; + buffer->mirror = malloc(size); + ASSERT_NE(buffer->mirror, NULL); + + buffer->ptr = mmap(NULL, size, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, + buffer->fd, 0); + ASSERT_NE(buffer->ptr, MAP_FAILED); + + /* Initialize buffer in system memory. */ + for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i) + ptr[i] = i; + + /* Migrate memory to device. */ + + ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages); + ASSERT_EQ(ret, 0); + ASSERT_EQ(buffer->cpages, npages); + + pid = fork(); + if (pid == -1) + ASSERT_EQ(pid, 0); + if (!pid) { + /* Child process waitd for SIGTERM from the parent. */ + while (1) { + } + perror("Should not reach this\n"); + exit(0); + } + /* Parent process writes to COW pages(s) and gets a +* new copy in system. In case of device private pages, +* this write causes a migration to system mem first. +*/ + for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i) + ptr[i] = i; + + /* Terminate child and wait */ + EXPECT_EQ(0, kill(pid, SIGTERM)); + EXPECT_EQ(pid, waitpid(pid, &status, 0)); + EXPECT_NE(0, WIFSIGNALED(status)); + EXPECT_EQ(SIGTERM, WTERMSIG(status)); + + /* Take snapshot to CPU pagetables */ + ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_SNAPSHOT, buffer, npages); + ASSERT_EQ(ret, 0); + ASSERT_EQ(buffer->cpages, npages); + m = buffer->mirror; + for (i = 0; i < npages; i++) + ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[i]); + + hmm_buffer_free(buffer); +} TEST_HARNESS_MAIN -- 2.32.0
[PATCH v2 08/13] lib: add support for device coherent type in test_hmm
Device Coherent type uses device memory that is coherently accesible by the CPU. This could be shown as SP (special purpose) memory range at the BIOS-e820 memory enumeration. If no SP memory is supported in system, this could be faked by setting CONFIG_EFI_FAKE_MEMMAP. Currently, test_hmm only supports two different SP ranges of at least 256MB size. This could be specified in the kernel parameter variable efi_fake_mem. Ex. Two SP ranges of 1GB starting at 0x1 & 0x14000 physical address. Ex. efi_fake_mem=1G@0x1:0x4,1G@0x14000:0x4 Private and coherent device mirror instances can be created in the same probed. This is done by passing the module parameters spm_addr_dev0 & spm_addr_dev1. In this case, it will create four instances of device_mirror. The first two correspond to private device type, the last two to coherent type. Then, they can be easily accessed from user space through /dev/hmm_mirror. Usually num_device 0 and 1 are for private, and 2 and 3 for coherent types. If no module parameters are passed, two instances of private type device_mirror will be created only. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling Reviewed-by: Alistair Poppple --- lib/test_hmm.c | 253 +--- lib/test_hmm_uapi.h | 15 ++- 2 files changed, 202 insertions(+), 66 deletions(-) diff --git a/lib/test_hmm.c b/lib/test_hmm.c index 15747f70c5bc..361a026c5d21 100644 --- a/lib/test_hmm.c +++ b/lib/test_hmm.c @@ -32,11 +32,22 @@ #include "test_hmm_uapi.h" -#define DMIRROR_NDEVICES 2 +#define DMIRROR_NDEVICES 4 #define DMIRROR_RANGE_FAULT_TIMEOUT1000 #define DEVMEM_CHUNK_SIZE (256 * 1024 * 1024U) #define DEVMEM_CHUNKS_RESERVE 16 +/* + * For device_private pages, dpage is just a dummy struct page + * representing a piece of device memory. dmirror_devmem_alloc_page + * allocates a real system memory page as backing storage to fake a + * real device. zone_device_data points to that backing page. But + * for device_coherent memory, the struct page represents real + * physical CPU-accessible memory that we can use directly. + */ +#define BACKING_PAGE(page) (is_device_private_page((page)) ? \ + (page)->zone_device_data : (page)) + static unsigned long spm_addr_dev0; module_param(spm_addr_dev0, long, 0644); MODULE_PARM_DESC(spm_addr_dev0, @@ -125,6 +136,21 @@ static int dmirror_bounce_init(struct dmirror_bounce *bounce, return 0; } +static bool dmirror_is_private_zone(struct dmirror_device *mdevice) +{ + return (mdevice->zone_device_type == + HMM_DMIRROR_MEMORY_DEVICE_PRIVATE) ? true : false; +} + +static enum migrate_vma_direction +dmirror_select_device(struct dmirror *dmirror) +{ + return (dmirror->mdevice->zone_device_type == + HMM_DMIRROR_MEMORY_DEVICE_PRIVATE) ? + MIGRATE_VMA_SELECT_DEVICE_PRIVATE : + MIGRATE_VMA_SELECT_DEVICE_COHERENT; +} + static void dmirror_bounce_fini(struct dmirror_bounce *bounce) { vfree(bounce->ptr); @@ -575,16 +601,19 @@ static int dmirror_allocate_chunk(struct dmirror_device *mdevice, static struct page *dmirror_devmem_alloc_page(struct dmirror_device *mdevice) { struct page *dpage = NULL; - struct page *rpage; + struct page *rpage = NULL; /* -* This is a fake device so we alloc real system memory to store -* our device memory. +* For ZONE_DEVICE private type, this is a fake device so we allocate +* real system memory to store our device memory. +* For ZONE_DEVICE coherent type we use the actual dpage to store the +* data and ignore rpage. */ - rpage = alloc_page(GFP_HIGHUSER); - if (!rpage) - return NULL; - + if (dmirror_is_private_zone(mdevice)) { + rpage = alloc_page(GFP_HIGHUSER); + if (!rpage) + return NULL; + } spin_lock(&mdevice->lock); if (mdevice->free_pages) { @@ -603,7 +632,8 @@ static struct page *dmirror_devmem_alloc_page(struct dmirror_device *mdevice) return dpage; error: - __free_page(rpage); + if (rpage) + __free_page(rpage); return NULL; } @@ -629,12 +659,16 @@ static void dmirror_migrate_alloc_and_copy(struct migrate_vma *args, * unallocated pte_none() or read-only zero page. */ spage = migrate_pfn_to_page(*src); + if (WARN(spage && is_zone_device_page(spage), +"page already in device spage pfn: 0x%lx\n", +page_to_pfn(spage))) + continue; dpage = dmirror_devmem_alloc_page(mdevice); if (!dpage) continue; - rpage = dpage->zone_device_data; + rpage = BACKING_PAGE(dpage);
[PATCH v2 12/13] tools: add hmm gup tests for device coherent type
The intention is to test hmm device coherent type under different get user pages paths. Also, test gup with FOLL_LONGTERM flag set in device coherent pages. These pages should get migrated back to system memory. Signed-off-by: Alex Sierra Reviewed-by: Alistair Popple --- tools/testing/selftests/vm/hmm-tests.c | 105 + 1 file changed, 105 insertions(+) diff --git a/tools/testing/selftests/vm/hmm-tests.c b/tools/testing/selftests/vm/hmm-tests.c index 84ec8c4a1dc7..61a02a6a3dea 100644 --- a/tools/testing/selftests/vm/hmm-tests.c +++ b/tools/testing/selftests/vm/hmm-tests.c @@ -36,6 +36,7 @@ * in the usual include/uapi/... directory. */ #include "../../../../lib/test_hmm_uapi.h" +#include "../../../../mm/gup_test.h" struct hmm_buffer { void*ptr; @@ -60,6 +61,8 @@ enum { #define NTIMES 10 #define ALIGN(x, a) (((x) + (a - 1)) & (~((a) - 1))) +/* Just the flags we need, copied from mm.h: */ +#define FOLL_WRITE 0x01/* check pte is writable */ FIXTURE(hmm) { @@ -1766,4 +1769,106 @@ TEST_F(hmm, exclusive_cow) hmm_buffer_free(buffer); } +static int gup_test_exec(int gup_fd, unsigned long addr, +int cmd, int npages, int size) +{ + struct gup_test gup = { + .nr_pages_per_call = npages, + .addr = addr, + .gup_flags = FOLL_WRITE, + .size = size, + }; + + if (ioctl(gup_fd, cmd, &gup)) { + perror("ioctl on error\n"); + return errno; + } + + return 0; +} + +/* + * Test get user device pages through gup_test. Setting PIN_LONGTERM flag. + * This should trigger a migration back to system memory for both, private + * and coherent type pages. + * This test makes use of gup_test module. Make sure GUP_TEST_CONFIG is added + * to your configuration before you run it. + */ +TEST_F(hmm, hmm_gup_test) +{ + struct hmm_buffer *buffer; + int gup_fd; + unsigned long npages; + unsigned long size; + unsigned long i; + int *ptr; + int ret; + unsigned char *m; + + gup_fd = open("/sys/kernel/debug/gup_test", O_RDWR); + if (gup_fd == -1) + SKIP(return, "Skipping test, could not find gup_test driver"); + + npages = 3; + size = npages << self->page_shift; + + buffer = malloc(sizeof(*buffer)); + ASSERT_NE(buffer, NULL); + + buffer->fd = -1; + buffer->size = size; + buffer->mirror = malloc(size); + ASSERT_NE(buffer->mirror, NULL); + + buffer->ptr = mmap(NULL, size, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, + buffer->fd, 0); + ASSERT_NE(buffer->ptr, MAP_FAILED); + + /* Initialize buffer in system memory. */ + for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i) + ptr[i] = i; + + /* Migrate memory to device. */ + ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages); + ASSERT_EQ(ret, 0); + ASSERT_EQ(buffer->cpages, npages); + /* Check what the device read. */ + for (i = 0, ptr = buffer->mirror; i < size / sizeof(*ptr); ++i) + ASSERT_EQ(ptr[i], i); + + ASSERT_EQ(gup_test_exec(gup_fd, + (unsigned long)buffer->ptr, + GUP_BASIC_TEST, 1, self->page_size), 0); + ASSERT_EQ(gup_test_exec(gup_fd, + (unsigned long)buffer->ptr + 1 * self->page_size, + GUP_FAST_BENCHMARK, 1, self->page_size), 0); + ASSERT_EQ(gup_test_exec(gup_fd, + (unsigned long)buffer->ptr + 2 * self->page_size, + PIN_LONGTERM_BENCHMARK, 1, self->page_size), 0); + + /* Take snapshot to CPU pagetables */ + ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_SNAPSHOT, buffer, npages); + ASSERT_EQ(ret, 0); + ASSERT_EQ(buffer->cpages, npages); + m = buffer->mirror; + if (hmm_is_coherent_type(variant->device_number)) { + ASSERT_EQ(HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL | HMM_DMIRROR_PROT_WRITE, m[0]); + ASSERT_EQ(HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL | HMM_DMIRROR_PROT_WRITE, m[1]); + } else { + ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[0]); + ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[1]); + } + ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[2]); + /* +* Check again the content on the pages. Make sure there's no +* corrupted data. +*/ + for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i) + ASSERT_EQ(ptr[i], i); + + close(gup_fd); + hmm_buffer_free(buffer); +} TEST_HARNESS_MAIN -- 2.32.0
[PATCH v2 10/13] tools: update test_hmm script to support SP config
Add two more parameters to set spm_addr_dev0 & spm_addr_dev1 addresses. These two parameters configure the start SP addresses for each device in test_hmm driver. Consequently, this configures zone device type as coherent. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling Reviewed-by: Alistair Popple Signed-off-by: Christoph Hellwig --- tools/testing/selftests/vm/test_hmm.sh | 24 +--- 1 file changed, 21 insertions(+), 3 deletions(-) diff --git a/tools/testing/selftests/vm/test_hmm.sh b/tools/testing/selftests/vm/test_hmm.sh index 0647b525a625..539c9371e592 100755 --- a/tools/testing/selftests/vm/test_hmm.sh +++ b/tools/testing/selftests/vm/test_hmm.sh @@ -40,11 +40,26 @@ check_test_requirements() load_driver() { - modprobe $DRIVER > /dev/null 2>&1 + if [ $# -eq 0 ]; then + modprobe $DRIVER > /dev/null 2>&1 + else + if [ $# -eq 2 ]; then + modprobe $DRIVER spm_addr_dev0=$1 spm_addr_dev1=$2 + > /dev/null 2>&1 + else + echo "Missing module parameters. Make sure pass"\ + "spm_addr_dev0 and spm_addr_dev1" + usage + fi + fi if [ $? == 0 ]; then major=$(awk "\$2==\"HMM_DMIRROR\" {print \$1}" /proc/devices) mknod /dev/hmm_dmirror0 c $major 0 mknod /dev/hmm_dmirror1 c $major 1 + if [ $# -eq 2 ]; then + mknod /dev/hmm_dmirror2 c $major 2 + mknod /dev/hmm_dmirror3 c $major 3 + fi fi } @@ -58,7 +73,7 @@ run_smoke() { echo "Running smoke test. Note, this test provides basic coverage." - load_driver + load_driver $1 $2 $(dirname "${BASH_SOURCE[0]}")/hmm-tests unload_driver } @@ -75,6 +90,9 @@ usage() echo "# Smoke testing" echo "./${TEST_NAME}.sh smoke" echo + echo "# Smoke testing with SPM enabled" + echo "./${TEST_NAME}.sh smoke " + echo exit 0 } @@ -84,7 +102,7 @@ function run_test() usage else if [ "$1" = "smoke" ]; then - run_smoke + run_smoke $2 $3 else usage fi -- 2.32.0
[PATCH v2 05/13] drm/amdkfd: add SPM support for SVM
When CPU is connected throug XGMI, it has coherent access to VRAM resource. In this case that resource is taken from a table in the device gmc aperture base. This resource is used along with the device type, which could be DEVICE_PRIVATE or DEVICE_COHERENT to create the device page map region. Also, MIGRATE_VMA_SELECT_DEVICE_COHERENT flag is selected for coherent type case during migration to device. Signed-off-by: Alex Sierra Reviewed-by: Felix Kuehling Signed-off-by: Christoph Hellwig --- drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 33 +++- 1 file changed, 21 insertions(+), 12 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 7e3a7fcb9fe6..25c9f7a4325d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -673,9 +673,12 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange, migrate.vma = vma; migrate.start = start; migrate.end = end; - migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE; migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev); + if (adev->gmc.xgmi.connected_to_cpu) + migrate.flags = MIGRATE_VMA_SELECT_DEVICE_COHERENT; + else + migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE; size = 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t); size *= npages; buf = kvmalloc(size, GFP_KERNEL | __GFP_ZERO); @@ -948,7 +951,7 @@ int svm_migrate_init(struct amdgpu_device *adev) { struct kfd_dev *kfddev = adev->kfd.dev; struct dev_pagemap *pgmap; - struct resource *res; + struct resource *res = NULL; unsigned long size; void *r; @@ -963,28 +966,34 @@ int svm_migrate_init(struct amdgpu_device *adev) * should remove reserved size */ size = ALIGN(adev->gmc.real_vram_size, 2ULL << 20); - res = devm_request_free_mem_region(adev->dev, &iomem_resource, size); - if (IS_ERR(res)) - return -ENOMEM; + if (adev->gmc.xgmi.connected_to_cpu) { + pgmap->range.start = adev->gmc.aper_base; + pgmap->range.end = adev->gmc.aper_base + adev->gmc.aper_size - 1; + pgmap->type = MEMORY_DEVICE_COHERENT; + } else { + res = devm_request_free_mem_region(adev->dev, &iomem_resource, size); + if (IS_ERR(res)) + return -ENOMEM; + pgmap->range.start = res->start; + pgmap->range.end = res->end; + pgmap->type = MEMORY_DEVICE_PRIVATE; + } - pgmap->type = MEMORY_DEVICE_PRIVATE; pgmap->nr_range = 1; - pgmap->range.start = res->start; - pgmap->range.end = res->end; pgmap->ops = &svm_migrate_pgmap_ops; pgmap->owner = SVM_ADEV_PGMAP_OWNER(adev); - pgmap->flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE; - + pgmap->flags = 0; /* Device manager releases device-specific resources, memory region and * pgmap when driver disconnects from device. */ r = devm_memremap_pages(adev->dev, pgmap); if (IS_ERR(r)) { pr_err("failed to register HMM device memory\n"); - /* Disable SVM support capability */ pgmap->type = 0; - devm_release_mem_region(adev->dev, res->start, resource_size(res)); + if (pgmap->type == MEMORY_DEVICE_PRIVATE) + devm_release_mem_region(adev->dev, res->start, + res->end - res->start + 1); return PTR_ERR(r); } -- 2.32.0
[PATCH v2 09/13] tools: update hmm-test to support device coherent type
Test cases such as migrate_fault and migrate_multiple, were modified to explicit migrate from device to sys memory without the need of page faults, when using device coherent type. Snapshot test case updated to read memory device type first and based on that, get the proper returned results migrate_ping_pong test case added to test explicit migration from device to sys memory for both private and coherent zone types. Helpers to migrate from device to sys memory and vicerversa were also added. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling Reviewed-by: Alistair Popple Signed-off-by: Christoph Hellwig --- tools/testing/selftests/vm/hmm-tests.c | 123 - 1 file changed, 102 insertions(+), 21 deletions(-) diff --git a/tools/testing/selftests/vm/hmm-tests.c b/tools/testing/selftests/vm/hmm-tests.c index 203323967b50..84ec8c4a1dc7 100644 --- a/tools/testing/selftests/vm/hmm-tests.c +++ b/tools/testing/selftests/vm/hmm-tests.c @@ -44,6 +44,14 @@ struct hmm_buffer { int fd; uint64_tcpages; uint64_tfaults; + int zone_device_type; +}; + +enum { + HMM_PRIVATE_DEVICE_ONE, + HMM_PRIVATE_DEVICE_TWO, + HMM_COHERENCE_DEVICE_ONE, + HMM_COHERENCE_DEVICE_TWO, }; #define TWOMEG (1 << 21) @@ -60,6 +68,21 @@ FIXTURE(hmm) unsigned intpage_shift; }; +FIXTURE_VARIANT(hmm) +{ + int device_number; +}; + +FIXTURE_VARIANT_ADD(hmm, hmm_device_private) +{ + .device_number = HMM_PRIVATE_DEVICE_ONE, +}; + +FIXTURE_VARIANT_ADD(hmm, hmm_device_coherent) +{ + .device_number = HMM_COHERENCE_DEVICE_ONE, +}; + FIXTURE(hmm2) { int fd0; @@ -68,6 +91,24 @@ FIXTURE(hmm2) unsigned intpage_shift; }; +FIXTURE_VARIANT(hmm2) +{ + int device_number0; + int device_number1; +}; + +FIXTURE_VARIANT_ADD(hmm2, hmm2_device_private) +{ + .device_number0 = HMM_PRIVATE_DEVICE_ONE, + .device_number1 = HMM_PRIVATE_DEVICE_TWO, +}; + +FIXTURE_VARIANT_ADD(hmm2, hmm2_device_coherent) +{ + .device_number0 = HMM_COHERENCE_DEVICE_ONE, + .device_number1 = HMM_COHERENCE_DEVICE_TWO, +}; + static int hmm_open(int unit) { char pathname[HMM_PATH_MAX]; @@ -81,12 +122,19 @@ static int hmm_open(int unit) return fd; } +static bool hmm_is_coherent_type(int dev_num) +{ + return (dev_num >= HMM_COHERENCE_DEVICE_ONE); +} + FIXTURE_SETUP(hmm) { self->page_size = sysconf(_SC_PAGE_SIZE); self->page_shift = ffs(self->page_size) - 1; - self->fd = hmm_open(0); + self->fd = hmm_open(variant->device_number); + if (self->fd < 0 && hmm_is_coherent_type(variant->device_number)) + SKIP(exit(0), "DEVICE_COHERENT not available"); ASSERT_GE(self->fd, 0); } @@ -95,9 +143,11 @@ FIXTURE_SETUP(hmm2) self->page_size = sysconf(_SC_PAGE_SIZE); self->page_shift = ffs(self->page_size) - 1; - self->fd0 = hmm_open(0); + self->fd0 = hmm_open(variant->device_number0); + if (self->fd0 < 0 && hmm_is_coherent_type(variant->device_number0)) + SKIP(exit(0), "DEVICE_COHERENT not available"); ASSERT_GE(self->fd0, 0); - self->fd1 = hmm_open(1); + self->fd1 = hmm_open(variant->device_number1); ASSERT_GE(self->fd1, 0); } @@ -144,6 +194,7 @@ static int hmm_dmirror_cmd(int fd, } buffer->cpages = cmd.cpages; buffer->faults = cmd.faults; + buffer->zone_device_type = cmd.zone_device_type; return 0; } @@ -211,6 +262,20 @@ static void hmm_nanosleep(unsigned int n) nanosleep(&t, NULL); } +static int hmm_migrate_sys_to_dev(int fd, + struct hmm_buffer *buffer, + unsigned long npages) +{ + return hmm_dmirror_cmd(fd, HMM_DMIRROR_MIGRATE_TO_DEV, buffer, npages); +} + +static int hmm_migrate_dev_to_sys(int fd, + struct hmm_buffer *buffer, + unsigned long npages) +{ + return hmm_dmirror_cmd(fd, HMM_DMIRROR_MIGRATE_TO_SYS, buffer, npages); +} + /* * Simple NULL test of device open/close. */ @@ -875,7 +940,7 @@ TEST_F(hmm, migrate) ptr[i] = i; /* Migrate memory to device. */ - ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_MIGRATE, buffer, npages); + ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages); ASSERT_EQ(ret, 0); ASSERT_EQ(buffer->cpages, npages); @@ -923,7 +988,7 @@ TEST_F(hmm, migrate_fault) ptr[i] = i; /* Migrate memory to device. */ - ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_MIGRATE, buffer, npages); + ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages); ASSERT_EQ(ret, 0); ASSERT_EQ(buffer->cpages, npages); @@ -936,7 +1001,7 @@ TEST_F(hmm, migrate_fault) ASSERT_EQ(ptr[i], i);
[PATCH v2 07/13] lib: test_hmm add module param for zone device type
In order to configure device coherent in test_hmm, two module parameters should be passed, which correspond to the SP start address of each device (2) spm_addr_dev0 & spm_addr_dev1. If no parameters are passed, private device type is configured. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling Reviewed-by: Alistair Poppple Signed-off-by: Christoph Hellwig --- lib/test_hmm.c | 73 - lib/test_hmm_uapi.h | 1 + 2 files changed, 53 insertions(+), 21 deletions(-) diff --git a/lib/test_hmm.c b/lib/test_hmm.c index 7a27584484ce..15747f70c5bc 100644 --- a/lib/test_hmm.c +++ b/lib/test_hmm.c @@ -37,6 +37,16 @@ #define DEVMEM_CHUNK_SIZE (256 * 1024 * 1024U) #define DEVMEM_CHUNKS_RESERVE 16 +static unsigned long spm_addr_dev0; +module_param(spm_addr_dev0, long, 0644); +MODULE_PARM_DESC(spm_addr_dev0, + "Specify start address for SPM (special purpose memory) used for device 0. By setting this Coherent device type will be used. Make sure spm_addr_dev1 is set too. Minimum SPM size should be DEVMEM_CHUNK_SIZE."); + +static unsigned long spm_addr_dev1; +module_param(spm_addr_dev1, long, 0644); +MODULE_PARM_DESC(spm_addr_dev1, + "Specify start address for SPM (special purpose memory) used for device 1. By setting this Coherent device type will be used. Make sure spm_addr_dev0 is set too. Minimum SPM size should be DEVMEM_CHUNK_SIZE."); + static const struct dev_pagemap_ops dmirror_devmem_ops; static const struct mmu_interval_notifier_ops dmirror_min_ops; static dev_t dmirror_dev; @@ -455,28 +465,44 @@ static int dmirror_write(struct dmirror *dmirror, struct hmm_dmirror_cmd *cmd) return ret; } -static bool dmirror_allocate_chunk(struct dmirror_device *mdevice, +static int dmirror_allocate_chunk(struct dmirror_device *mdevice, struct page **ppage) { struct dmirror_chunk *devmem; - struct resource *res; + struct resource *res = NULL; unsigned long pfn; unsigned long pfn_first; unsigned long pfn_last; void *ptr; + int ret = -ENOMEM; devmem = kzalloc(sizeof(*devmem), GFP_KERNEL); if (!devmem) - return false; + return ret; - res = request_free_mem_region(&iomem_resource, DEVMEM_CHUNK_SIZE, - "hmm_dmirror"); - if (IS_ERR(res)) + switch (mdevice->zone_device_type) { + case HMM_DMIRROR_MEMORY_DEVICE_PRIVATE: + res = request_free_mem_region(&iomem_resource, DEVMEM_CHUNK_SIZE, + "hmm_dmirror"); + if (IS_ERR_OR_NULL(res)) + goto err_devmem; + devmem->pagemap.range.start = res->start; + devmem->pagemap.range.end = res->end; + devmem->pagemap.type = MEMORY_DEVICE_PRIVATE; + break; + case HMM_DMIRROR_MEMORY_DEVICE_COHERENT: + devmem->pagemap.range.start = (MINOR(mdevice->cdevice.dev) - 2) ? + spm_addr_dev0 : + spm_addr_dev1; + devmem->pagemap.range.end = devmem->pagemap.range.start + + DEVMEM_CHUNK_SIZE - 1; + devmem->pagemap.type = MEMORY_DEVICE_COHERENT; + break; + default: + ret = -EINVAL; goto err_devmem; + } - devmem->pagemap.type = MEMORY_DEVICE_PRIVATE; - devmem->pagemap.range.start = res->start; - devmem->pagemap.range.end = res->end; devmem->pagemap.nr_range = 1; devmem->pagemap.ops = &dmirror_devmem_ops; devmem->pagemap.owner = mdevice; @@ -497,10 +523,14 @@ static bool dmirror_allocate_chunk(struct dmirror_device *mdevice, mdevice->devmem_capacity = new_capacity; mdevice->devmem_chunks = new_chunks; } - ptr = memremap_pages(&devmem->pagemap, numa_node_id()); - if (IS_ERR(ptr)) + if (IS_ERR_OR_NULL(ptr)) { + if (ptr) + ret = PTR_ERR(ptr); + else + ret = -EFAULT; goto err_release; + } devmem->mdevice = mdevice; pfn_first = devmem->pagemap.range.start >> PAGE_SHIFT; @@ -529,15 +559,17 @@ static bool dmirror_allocate_chunk(struct dmirror_device *mdevice, } spin_unlock(&mdevice->lock); - return true; + return 0; err_release: mutex_unlock(&mdevice->devmem_lock); - release_mem_region(devmem->pagemap.range.start, range_len(&devmem->pagemap.range)); + if (res && devmem->pagemap.type == MEMORY_DEVICE_PRIVATE) + release_mem_region(devmem->pagemap.range.start, + range_len(&devmem->pagemap.range));
[PATCH v2 03/13] mm: remove the vma check in migrate_vma_setup()
From: Alistair Popple migrate_vma_setup() checks that a valid vma is passed so that the page tables can be walked to find the pfns associated with a given address range. However in some cases the pfns are already known, such as when migrating device coherent pages during pin_user_pages() meaning a valid vma isn't required. Signed-off-by: Alistair Popple Acked-by: Felix Kuehling Signed-off-by: Christoph Hellwig --- mm/migrate_device.c | 34 +- 1 file changed, 17 insertions(+), 17 deletions(-) diff --git a/mm/migrate_device.c b/mm/migrate_device.c index a0b997935cf9..b399c498f0aa 100644 --- a/mm/migrate_device.c +++ b/mm/migrate_device.c @@ -467,24 +467,24 @@ int migrate_vma_setup(struct migrate_vma *args) args->start &= PAGE_MASK; args->end &= PAGE_MASK; - if (!args->vma || is_vm_hugetlb_page(args->vma) || - (args->vma->vm_flags & VM_SPECIAL) || vma_is_dax(args->vma)) - return -EINVAL; - if (nr_pages <= 0) - return -EINVAL; - if (args->start < args->vma->vm_start || - args->start >= args->vma->vm_end) - return -EINVAL; - if (args->end <= args->vma->vm_start || args->end > args->vma->vm_end) - return -EINVAL; if (!args->src || !args->dst) return -EINVAL; - - memset(args->src, 0, sizeof(*args->src) * nr_pages); - args->cpages = 0; - args->npages = 0; - - migrate_vma_collect(args); + if (args->vma) { + if (is_vm_hugetlb_page(args->vma) || + (args->vma->vm_flags & VM_SPECIAL) || vma_is_dax(args->vma)) + return -EINVAL; + if (args->start < args->vma->vm_start || + args->start >= args->vma->vm_end) + return -EINVAL; + if (args->end <= args->vma->vm_start || + args->end > args->vma->vm_end) + return -EINVAL; + memset(args->src, 0, sizeof(*args->src) * nr_pages); + args->cpages = 0; + args->npages = 0; + + migrate_vma_collect(args); + } if (args->cpages) migrate_vma_unmap(args); @@ -666,7 +666,7 @@ void migrate_vma_pages(struct migrate_vma *migrate) continue; } - if (!page) { + if (!page && migrate->vma) { if (!(migrate->src[i] & MIGRATE_PFN_MIGRATE)) continue; if (!notified) { -- 2.32.0
[PATCH v2 06/13] lib: test_hmm add ioctl to get zone device type
new ioctl cmd added to query zone device type. This will be used once the test_hmm adds zone device coherent type. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling Reviewed-by: Alistair Poppple Signed-off-by: Christoph Hellwig --- lib/test_hmm.c | 23 +-- lib/test_hmm_uapi.h | 8 2 files changed, 29 insertions(+), 2 deletions(-) diff --git a/lib/test_hmm.c b/lib/test_hmm.c index cfe632047839..7a27584484ce 100644 --- a/lib/test_hmm.c +++ b/lib/test_hmm.c @@ -87,6 +87,7 @@ struct dmirror_chunk { struct dmirror_device { struct cdev cdevice; struct hmm_devmem *devmem; + unsigned intzone_device_type; unsigned intdevmem_capacity; unsigned intdevmem_count; @@ -1026,6 +1027,15 @@ static int dmirror_snapshot(struct dmirror *dmirror, return ret; } +static int dmirror_get_device_type(struct dmirror *dmirror, + struct hmm_dmirror_cmd *cmd) +{ + mutex_lock(&dmirror->mutex); + cmd->zone_device_type = dmirror->mdevice->zone_device_type; + mutex_unlock(&dmirror->mutex); + + return 0; +} static long dmirror_fops_unlocked_ioctl(struct file *filp, unsigned int command, unsigned long arg) @@ -1076,6 +1086,9 @@ static long dmirror_fops_unlocked_ioctl(struct file *filp, ret = dmirror_snapshot(dmirror, &cmd); break; + case HMM_DMIRROR_GET_MEM_DEV_TYPE: + ret = dmirror_get_device_type(dmirror, &cmd); + break; default: return -EINVAL; } @@ -1260,14 +1273,20 @@ static void dmirror_device_remove(struct dmirror_device *mdevice) static int __init hmm_dmirror_init(void) { int ret; - int id; + int id = 0; + int ndevices = 0; ret = alloc_chrdev_region(&dmirror_dev, 0, DMIRROR_NDEVICES, "HMM_DMIRROR"); if (ret) goto err_unreg; - for (id = 0; id < DMIRROR_NDEVICES; id++) { + memset(dmirror_devices, 0, DMIRROR_NDEVICES * sizeof(dmirror_devices[0])); + dmirror_devices[ndevices++].zone_device_type = + HMM_DMIRROR_MEMORY_DEVICE_PRIVATE; + dmirror_devices[ndevices++].zone_device_type = + HMM_DMIRROR_MEMORY_DEVICE_PRIVATE; + for (id = 0; id < ndevices; id++) { ret = dmirror_device_init(dmirror_devices + id, id); if (ret) goto err_chrdev; diff --git a/lib/test_hmm_uapi.h b/lib/test_hmm_uapi.h index f14dea5dcd06..17f842f1aa02 100644 --- a/lib/test_hmm_uapi.h +++ b/lib/test_hmm_uapi.h @@ -19,6 +19,7 @@ * @npages: (in) number of pages to read/write * @cpages: (out) number of pages copied * @faults: (out) number of device page faults seen + * @zone_device_type: (out) zone device memory type */ struct hmm_dmirror_cmd { __u64 addr; @@ -26,6 +27,7 @@ struct hmm_dmirror_cmd { __u64 npages; __u64 cpages; __u64 faults; + __u64 zone_device_type; }; /* Expose the address space of the calling process through hmm device file */ @@ -35,6 +37,7 @@ struct hmm_dmirror_cmd { #define HMM_DMIRROR_SNAPSHOT _IOWR('H', 0x03, struct hmm_dmirror_cmd) #define HMM_DMIRROR_EXCLUSIVE _IOWR('H', 0x04, struct hmm_dmirror_cmd) #define HMM_DMIRROR_CHECK_EXCLUSIVE_IOWR('H', 0x05, struct hmm_dmirror_cmd) +#define HMM_DMIRROR_GET_MEM_DEV_TYPE _IOWR('H', 0x06, struct hmm_dmirror_cmd) /* * Values returned in hmm_dmirror_cmd.ptr for HMM_DMIRROR_SNAPSHOT. @@ -62,4 +65,9 @@ enum { HMM_DMIRROR_PROT_DEV_PRIVATE_REMOTE = 0x30, }; +enum { + /* 0 is reserved to catch uninitialized type fields */ + HMM_DMIRROR_MEMORY_DEVICE_PRIVATE = 1, +}; + #endif /* _LIB_TEST_HMM_UAPI_H */ -- 2.32.0
[PATCH v2 04/13] mm/gup: migrate device coherent pages when pinning instead of failing
From: Alistair Popple Currently any attempts to pin a device coherent page will fail. This is because device coherent pages need to be managed by a device driver, and pinning them would prevent a driver from migrating them off the device. However this is no reason to fail pinning of these pages. These are coherent and accessible from the CPU so can be migrated just like pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin them first try migrating them out of ZONE_DEVICE. Signed-off-by: Alistair Popple Acked-by: Felix Kuehling [hch: rebased to the split device memory checks, moved migrate_device_page to migrate_device.c] Signed-off-by: Christoph Hellwig --- mm/gup.c| 47 +++- mm/internal.h | 1 + mm/migrate_device.c | 53 + 3 files changed, 96 insertions(+), 5 deletions(-) diff --git a/mm/gup.c b/mm/gup.c index f598a037eb04..a214c8df7140 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -1786,9 +1786,43 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages, continue; prev_folio = folio; - if (folio_is_pinnable(folio)) + /* +* Device private pages will get faulted in during gup so it +* shouldn't be possible to see one here. +*/ + if (WARN_ON_ONCE(folio_is_device_private(folio))) { + ret = -EFAULT; + goto unpin_pages; + } + + /* +* Device coherent pages are managed by a driver and should not +* be pinned indefinitely as it prevents the driver moving the +* page. So when trying to pin with FOLL_LONGTERM instead try +* to migrate the page out of device memory. +*/ + if (folio_is_device_coherent(folio)) { + WARN_ON_ONCE(PageCompound(&folio->page)); + + /* +* Migration will fail if the page is pinned, so convert +* the pin on the source page to a normal reference. +*/ + if (gup_flags & FOLL_PIN) { + get_page(&folio->page); + unpin_user_page(&folio->page); + } + + pages[i] = migrate_device_page(&folio->page, gup_flags); + if (!pages[i]) { + ret = -EBUSY; + goto unpin_pages; + } continue; + } + if (folio_is_pinnable(folio)) + continue; /* * Try to move out any movable page before pinning the range. */ @@ -1824,10 +1858,13 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages, return nr_pages; unpin_pages: - if (gup_flags & FOLL_PIN) { - unpin_user_pages(pages, nr_pages); - } else { - for (i = 0; i < nr_pages; i++) + for (i = 0; i < nr_pages; i++) { + if (!pages[i]) + continue; + + if (gup_flags & FOLL_PIN) + unpin_user_page(pages[i]); + else put_page(pages[i]); } diff --git a/mm/internal.h b/mm/internal.h index cf16280ce132..32c621416966 100644 --- a/mm/internal.h +++ b/mm/internal.h @@ -750,6 +750,7 @@ int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, unsigned long addr, int page_nid, int *flags); void free_zone_device_page(struct page *page); +struct page *migrate_device_page(struct page *page, unsigned int gup_flags); /* * mm/gup.c diff --git a/mm/migrate_device.c b/mm/migrate_device.c index b399c498f0aa..1e74aa7ab580 100644 --- a/mm/migrate_device.c +++ b/mm/migrate_device.c @@ -775,3 +775,56 @@ void migrate_vma_finalize(struct migrate_vma *migrate) } } EXPORT_SYMBOL(migrate_vma_finalize); + +/* + * Migrate a device coherent page back to normal memory. The caller should have + * a reference on page which will be copied to the new page if migration is + * successful or dropped on failure. + */ +struct page *migrate_device_page(struct page *page, unsigned int gup_flags) +{ + unsigned long src_pfn, dst_pfn = 0; + struct migrate_vma args; + struct page *dpage; + + lock_page(page); + src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE; + args.src = &src_pfn; + args.dst = &dst_pfn; + args.cpages = 1; + args.npages = 1; + args.vma = NULL; + migrate_vma_setup(&args); + if (!(src_pfn & MIGRATE_PFN_MIGRATE)) + return NULL; + + dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0); +
[PATCH v2 01/13] mm: add zone device coherent type memory support
Device memory that is cache coherent from device and CPU point of view. This is used on platforms that have an advanced system bus (like CAPI or CXL). Any page of a process can be migrated to such memory. However, no one should be allowed to pin such memory so that it can always be evicted. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling Reviewed-by: Alistair Popple [hch: rebased ontop of the refcount changes, removed is_dev_private_or_coherent_page] Signed-off-by: Christoph Hellwig --- include/linux/memremap.h | 19 +++ mm/memcontrol.c | 7 --- mm/memory-failure.c | 8 ++-- mm/memremap.c| 10 ++ mm/migrate_device.c | 16 +++- mm/rmap.c| 5 +++-- 6 files changed, 49 insertions(+), 16 deletions(-) diff --git a/include/linux/memremap.h b/include/linux/memremap.h index 8af304f6b504..9f752ebed613 100644 --- a/include/linux/memremap.h +++ b/include/linux/memremap.h @@ -41,6 +41,13 @@ struct vmem_altmap { * A more complete discussion of unaddressable memory may be found in * include/linux/hmm.h and Documentation/vm/hmm.rst. * + * MEMORY_DEVICE_COHERENT: + * Device memory that is cache coherent from device and CPU point of view. This + * is used on platforms that have an advanced system bus (like CAPI or CXL). A + * driver can hotplug the device memory using ZONE_DEVICE and with that memory + * type. Any page of a process can be migrated to such memory. However no one + * should be allowed to pin such memory so that it can always be evicted. + * * MEMORY_DEVICE_FS_DAX: * Host memory that has similar access semantics as System RAM i.e. DMA * coherent and supports page pinning. In support of coordinating page @@ -61,6 +68,7 @@ struct vmem_altmap { enum memory_type { /* 0 is reserved to catch uninitialized type fields */ MEMORY_DEVICE_PRIVATE = 1, + MEMORY_DEVICE_COHERENT, MEMORY_DEVICE_FS_DAX, MEMORY_DEVICE_GENERIC, MEMORY_DEVICE_PCI_P2PDMA, @@ -143,6 +151,17 @@ static inline bool folio_is_device_private(const struct folio *folio) return is_device_private_page(&folio->page); } +static inline bool is_device_coherent_page(const struct page *page) +{ + return is_zone_device_page(page) && + page->pgmap->type == MEMORY_DEVICE_COHERENT; +} + +static inline bool folio_is_device_coherent(const struct folio *folio) +{ + return is_device_coherent_page(&folio->page); +} + static inline bool is_pci_p2pdma_page(const struct page *page) { return IS_ENABLED(CONFIG_PCI_P2PDMA) && diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 598fece89e2b..3e1f97c9fdc6 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -5629,8 +5629,8 @@ static int mem_cgroup_move_account(struct page *page, * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a * target for charge migration. if @target is not NULL, the entry is stored * in target->ent. - * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE - * (so ZONE_DEVICE page and thus not on the lru). + * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is device memory and + * thus not on the lru. * For now we such page is charge like a regular page would be as for all * intent and purposes it is just special memory taking the place of a * regular page. @@ -5664,7 +5664,8 @@ static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, */ if (page_memcg(page) == mc.from) { ret = MC_TARGET_PAGE; - if (is_device_private_page(page)) + if (is_device_private_page(page) || + is_device_coherent_page(page)) ret = MC_TARGET_DEVICE; if (target) target->page = page; diff --git a/mm/memory-failure.c b/mm/memory-failure.c index d4a4adcca01f..c8a0e9fcdf42 100644 --- a/mm/memory-failure.c +++ b/mm/memory-failure.c @@ -1695,12 +1695,16 @@ static int memory_failure_dev_pagemap(unsigned long pfn, int flags, goto unlock; } - if (pgmap->type == MEMORY_DEVICE_PRIVATE) { + switch (pgmap->type) { + case MEMORY_DEVICE_PRIVATE: + case MEMORY_DEVICE_COHERENT: /* -* TODO: Handle HMM pages which may need coordination +* TODO: Handle device pages which may need coordination * with device-side memory. */ goto unlock; + default: + break; } /* diff --git a/mm/memremap.c b/mm/memremap.c index af0223605e69..471ec84ed82b 100644 --- a/mm/memremap.c +++ b/mm/memremap.c @@ -314,6 +314,16 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid) return ERR_PTR(-EINVAL); } break;
[PATCH v2 02/13] mm: add device coherent vma selection for memory migration
This case is used to migrate pages from device memory, back to system memory. Device coherent type memory is cache coherent from device and CPU point of view. Signed-off-by: Alex Sierra Acked-by: Felix Kuehling Reviewed-by: Alistair Poppple Signed-off-by: Christoph Hellwig --- include/linux/migrate.h | 1 + mm/migrate_device.c | 12 +--- 2 files changed, 10 insertions(+), 3 deletions(-) diff --git a/include/linux/migrate.h b/include/linux/migrate.h index 90e75d5a54d6..a4a336fd81fc 100644 --- a/include/linux/migrate.h +++ b/include/linux/migrate.h @@ -138,6 +138,7 @@ static inline unsigned long migrate_pfn(unsigned long pfn) enum migrate_vma_direction { MIGRATE_VMA_SELECT_SYSTEM = 1 << 0, MIGRATE_VMA_SELECT_DEVICE_PRIVATE = 1 << 1, + MIGRATE_VMA_SELECT_DEVICE_COHERENT = 1 << 2, }; struct migrate_vma { diff --git a/mm/migrate_device.c b/mm/migrate_device.c index 7e267142ea34..a0b997935cf9 100644 --- a/mm/migrate_device.c +++ b/mm/migrate_device.c @@ -148,15 +148,21 @@ static int migrate_vma_collect_pmd(pmd_t *pmdp, if (is_writable_device_private_entry(entry)) mpfn |= MIGRATE_PFN_WRITE; } else { - if (!(migrate->flags & MIGRATE_VMA_SELECT_SYSTEM)) - goto next; pfn = pte_pfn(pte); - if (is_zero_pfn(pfn)) { + if (is_zero_pfn(pfn) && + (migrate->flags & MIGRATE_VMA_SELECT_SYSTEM)) { mpfn = MIGRATE_PFN_MIGRATE; migrate->cpages++; goto next; } page = vm_normal_page(migrate->vma, addr, pte); + if (page && !is_zone_device_page(page) && + !(migrate->flags & MIGRATE_VMA_SELECT_SYSTEM)) + goto next; + else if (page && is_device_coherent_page(page) && + (!(migrate->flags & MIGRATE_VMA_SELECT_DEVICE_COHERENT) || +page->pgmap->owner != migrate->pgmap_owner)) + goto next; mpfn = migrate_pfn(pfn) | MIGRATE_PFN_MIGRATE; mpfn |= pte_write(pte) ? MIGRATE_PFN_WRITE : 0; } -- 2.32.0
[PATCH v2 00/13] Add MEMORY_DEVICE_COHERENT for coherent device memory mapping
This is our MEMORY_DEVICE_COHERENT patch series rebased and updated for current 5.18-rc7. Changes since the last version: - Fixed problems with migration during long-term pinning in get_user_pages - Open coded vm_normal_lru_pages as suggested in previous code review - Update hmm_gup_test with more get_user_pages calls, include hmm_cow_in_device in hmm-test. This patch series introduces MEMORY_DEVICE_COHERENT, a type of memory owned by a device that can be mapped into CPU page tables like MEMORY_DEVICE_GENERIC and can also be migrated like MEMORY_DEVICE_PRIVATE. This patch series is mostly self-contained except for a few places where it needs to update other subsystems to handle the new memory type. System stability and performance are not affected according to our ongoing testing, including xfstests. How it works: The system BIOS advertises the GPU device memory (aka VRAM) as SPM (special purpose memory) in the UEFI system address map. The amdgpu driver registers the memory with devmap as MEMORY_DEVICE_COHERENT using devm_memremap_pages. The initial user for this hardware page migration capability is the Frontier supercomputer project. This functionality is not AMD-specific. We expect other GPU vendors to find this functionality useful, and possibly other hardware types in the future. Our test nodes in the lab are similar to the Frontier configuration, with .5 TB of system memory plus 256 GB of device memory split across 4 GPUs, all in a single coherent address space. Page migration is expected to improve application efficiency significantly. We will report empirical results as they become available. Coherent device type pages at gup are now migrated back to system memory if they are being pinned long-term (FOLL_LONGTERM). The reason is, that long-term pinning would interfere with the device memory manager owning the device-coherent pages (e.g. evictions in TTM). These series incorporate Alistair Popple patches to do this migration from pin_user_pages() calls. hmm_gup_test has been added to hmm-test to test different get user pages calls. This series includes handling of device-managed anonymous pages returned by vm_normal_pages. Although they behave like normal pages for purposes of mapping in CPU page tables and for COW, they do not support LRU lists, NUMA migration or THP. We also introduced a FOLL_LRU flag that adds the same behaviour to follow_page and related APIs, to allow callers to specify that they expect to put pages on an LRU list. v2: - Rebase to latest 5.18-rc7. - Drop patch "mm: add device coherent checker to remove migration pte" and modify try_to_migrate_one, to let DEVICE_COHERENT pages fall through to normal page path. Based on Alistair Popple's comment. - Fix comment formatting. - Reword comment in vm_normal_page about pte_devmap(). - Merge "drm/amdkfd: coherent type as sys mem on migration to ram" to "drm/amdkfd: add SPM support for SVM". Alex Sierra (11): mm: add zone device coherent type memory support mm: add device coherent vma selection for memory migration drm/amdkfd: add SPM support for SVM lib: test_hmm add ioctl to get zone device type lib: test_hmm add module param for zone device type lib: add support for device coherent type in test_hmm tools: update hmm-test to support device coherent type tools: update test_hmm script to support SP config mm: handling Non-LRU pages returned by vm_normal_pages tools: add hmm gup tests for device coherent type tools: add selftests to hmm for COW in device memory Alistair Popple (2): mm: remove the vma check in migrate_vma_setup() mm/gup: migrate device coherent pages when pinning instead of failing drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 33 ++- fs/proc/task_mmu.c | 2 +- include/linux/memremap.h | 19 ++ include/linux/migrate.h | 1 + include/linux/mm.h | 3 +- lib/test_hmm.c | 349 +-- lib/test_hmm_uapi.h | 22 +- mm/gup.c | 49 +++- mm/huge_memory.c | 2 +- mm/internal.h| 1 + mm/khugepaged.c | 9 +- mm/ksm.c | 6 +- mm/madvise.c | 4 +- mm/memcontrol.c | 7 +- mm/memory-failure.c | 8 +- mm/memory.c | 9 +- mm/mempolicy.c | 2 +- mm/memremap.c| 10 + mm/migrate.c | 4 +- mm/migrate_device.c | 115 ++-- mm/mlock.c | 2 +- mm/mprotect.c| 2 +- mm/rmap.c| 5 +- tools/testing/selftests/vm/hmm-tests.c | 308 ++-- tools/testing/selftests/vm/test_hmm.sh | 24 +- 25 file
[PATCH v2 27/27] dyndbg/drm: POC add tracebits sysfs-knob
clone DRM.debug interface to DRM.tracebits: ie map bits to drm-debug-categories, except this interface enables messages to tracefs, not to syslog. This reuses dyndbg's register-classes API to add the new sysfs-knob: drm_drv.h: [A] 2nd use of DYNAMIC_DEBUG_CLASSES(drm_trace_classes), which declares and initializes a known-classes map with the same literal classnames as in the 1st use. This "shares" the classnames for both sysfs-knobs, which is necessary because they're manipulating a shared dyndbg callsite, toggling "p" and "T" flags respectively. This incurs a tiny waste of constant strings, but its worth it for the simpler declarative macro. drm_print.c: 1- declare and export a 2nd bit-vector: __drm_trace, like __drm_debug 2- declare and init a struct ddebug_classes_bitmap_param with: __drm_trace [1], drm_trace_classes [A]. 3- module_param_cb([2]) - does the sysfs part drm_drv.c: register and unregister [A], and missed unregister on drm_debug_classes. Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_drv.c | 3 +++ drivers/gpu/drm/drm_print.c | 10 ++ include/drm/drm_drv.h | 14 +- 3 files changed, 26 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c index 16683fb169aa..ad141c5e29ca 100644 --- a/drivers/gpu/drm/drm_drv.c +++ b/drivers/gpu/drm/drm_drv.c @@ -1039,6 +1039,8 @@ static void drm_core_exit(void) drm_sysfs_destroy(); idr_destroy(&drm_minors_idr); drm_connector_ida_destroy(); + dynamic_debug_unregister_classes(&drm_debug_classes); + dynamic_debug_unregister_classes(&drm_trace_classes); } static int __init drm_core_init(void) @@ -1046,6 +1048,7 @@ static int __init drm_core_init(void) int ret; dynamic_debug_register_classes(&drm_debug_classes); + dynamic_debug_register_classes(&drm_trace_classes); drm_connector_ida_init(); idr_init(&drm_minors_idr); diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index 9afb676bda4d..970b6dd46c42 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -44,6 +44,9 @@ unsigned long __drm_debug; EXPORT_SYMBOL(__drm_debug); +unsigned long __drm_trace; +EXPORT_SYMBOL(__drm_trace); + MODULE_PARM_DESC(debug, "Enable debug output, where each bit enables a debug category.\n" "\t\tBit 0 (0x01) will enable CORE messages (drm core code)\n" "\t\tBit 1 (0x02) will enable DRIVER messages (drm controller code)\n" @@ -63,6 +66,13 @@ static struct ddebug_classes_bitmap_param drm_debug_bitmap = { .map = &drm_debug_classes, }; module_param_cb(debug, ¶m_ops_dyndbg_classes, &drm_debug_bitmap, 0600); + +static struct ddebug_classes_bitmap_param drm_trace_bitmap = { + .bits = &__drm_trace, + .flags = "T", + .map = &drm_trace_classes, +}; +module_param_cb(tracecats, ¶m_ops_dyndbg_classes, &drm_trace_bitmap, 0600); #endif void __drm_puts_coredump(struct drm_printer *p, const char *str) diff --git a/include/drm/drm_drv.h b/include/drm/drm_drv.h index c2ffe12161b8..596de0addfd5 100644 --- a/include/drm/drm_drv.h +++ b/include/drm/drm_drv.h @@ -45,7 +45,19 @@ struct drm_printer; struct sg_table; /* these must comport with enum drm_debug_category values */ -DYNAMIC_DEBUG_CLASSES(drm_debug_classes, "*", 0, +DYNAMIC_DEBUG_CLASSES(drm_debug_classes, 0, + "DRM_UT_CORE", + "DRM_UT_DRIVER", + "DRM_UT_KMS", + "DRM_UT_PRIME", + "DRM_UT_ATOMIC", + "DRM_UT_VBL", + "DRM_UT_STATE", + "DRM_UT_LEASE", + "DRM_UT_DP", + "DRM_UT_DRMRES"); + +DYNAMIC_DEBUG_CLASSES(drm_trace_classes, 0, "DRM_UT_CORE", "DRM_UT_DRIVER", "DRM_UT_KMS", -- 2.35.3
[PATCH v2 21/27] drm_print: prefer bare printk KERN_DEBUG on generic fn
drm_print.c calls pr_debug() just once, from __drm_printfn_debug(), which is a generic/service fn. The callsite is compile-time enabled by DEBUG in both DYNAMIC_DEBUG=y/n builds. For dyndbg builds, reverting this callsite back to bare printk is correcting a few anti-features: 1- callsite is generic, serves multiple drm users. its hardwired on currently could accidentally: #> echo -p > /proc/dynamic_debug/control 2- optional "decorations" by dyndbg are unhelpful/misleading they describe only the generic site, not end users IOW, 1,2 are unhelpful at best, and possibly confusing. reverting yields a nominal data and text shrink: textdata bss dec hex filename 462583 36604 54592 553779 87333 /kernel/drivers/gpu/drm/drm.ko 462515 36532 54592 553639 872a7 -dirty/kernel/drivers/gpu/drm/drm.ko Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_print.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index 22b7ab1c5c40..1b11a15366b7 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -23,8 +23,6 @@ * Rob Clark */ -#define DEBUG /* for pr_debug() */ - #include #include @@ -172,7 +170,8 @@ EXPORT_SYMBOL(__drm_printfn_info); void __drm_printfn_debug(struct drm_printer *p, struct va_format *vaf) { - pr_debug("%s %pV", p->prefix, vaf); + /* pr_debug callsite decorations are unhelpful here */ + printk(KERN_DEBUG "%s %pV", p->prefix, vaf); } EXPORT_SYMBOL(__drm_printfn_debug); -- 2.35.3
[PATCH v2 22/27] drm_print: add _ddebug desc to drm_*dbg prototypes
Add a struct _ddebug ptr to drm_dbg() and drm_dev_dbg() protos, and upgrade their callers - the interposed macros added recently, to use _dynamic_func_call_no_desc(); ie drop the '_no_desc', since the factory macro's callees (these 2 functions) are now expecting the arg. This makes those functions controllable like pr_debug(). It also means that they don't get to use the decorations implemented by pr_debug(). DRM already has standards for logging/messaging; tossing optional decorations on top may not help. For CONFIG_DRM_USE_DYNAMIC_DEBUG=N, just pass null. Note: desc->class_id is redundant with category parameter, but its availability is dependent on desc. Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_print.c | 23 +-- include/drm/drm_print.h | 22 +++--- 2 files changed, 24 insertions(+), 21 deletions(-) diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index 1b11a15366b7..1752b40b0189 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -265,8 +265,8 @@ void drm_dev_printk(const struct device *dev, const char *level, } EXPORT_SYMBOL(drm_dev_printk); -void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, - const char *format, ...) +void __drm_dev_dbg(struct _ddebug *desc, const struct device *dev, + enum drm_debug_category category, const char *format, ...) { struct va_format vaf; va_list args; @@ -274,22 +274,25 @@ void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, if (!__drm_debug_enabled(category)) return; + /* we know we are printing for either syslog, tracefs, or both */ va_start(args, format); vaf.fmt = format; vaf.va = &args; - if (dev) - dev_printk(KERN_DEBUG, dev, "[" DRM_NAME ":%ps] %pV", - __builtin_return_address(0), &vaf); - else - printk(KERN_DEBUG "[" DRM_NAME ":%ps] %pV", - __builtin_return_address(0), &vaf); - + if (dev) { + if (desc->flags && _DPRINTK_FLAGS_PRINT) + dev_printk(KERN_DEBUG, dev, "[" DRM_NAME ":%ps] %pV", + __builtin_return_address(0), &vaf); + } else { + if (desc->flags && _DPRINTK_FLAGS_PRINT) + printk(KERN_DEBUG "[" DRM_NAME ":%ps] %pV", + __builtin_return_address(0), &vaf); + } va_end(args); } EXPORT_SYMBOL(__drm_dev_dbg); -void ___drm_dbg(enum drm_debug_category category, const char *format, ...) +void ___drm_dbg(struct _ddebug *desc, enum drm_debug_category category, const char *format, ...) { struct va_format vaf; va_list args; diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h index ccd177236ab3..fc0f2f6c2b91 100644 --- a/include/drm/drm_print.h +++ b/include/drm/drm_print.h @@ -366,9 +366,9 @@ static inline bool drm_debug_enabled(enum drm_debug_category category) __printf(3, 4) void drm_dev_printk(const struct device *dev, const char *level, const char *format, ...); -__printf(3, 4) -void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, -const char *format, ...); +__printf(4, 5) +void __drm_dev_dbg(struct _ddebug *desc, const struct device *dev, + enum drm_debug_category category, const char *format, ...); /** * DRM_DEV_ERROR() - Error output. @@ -418,11 +418,11 @@ void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, #if !defined(CONFIG_DRM_USE_DYNAMIC_DEBUG) #define drm_dev_dbg(dev, cat, fmt, ...)\ - __drm_dev_dbg(dev, cat, fmt, ##__VA_ARGS__) + __drm_dev_dbg(NULL, dev, cat, fmt, ##__VA_ARGS__) #else #define drm_dev_dbg(dev, cat, fmt, ...)\ - _dynamic_func_call_no_desc(fmt, __drm_dev_dbg, \ - dev, cat, fmt, ##__VA_ARGS__) + _dynamic_func_call_cls(cat, fmt, __drm_dev_dbg, \ + dev, cat, fmt, ##__VA_ARGS__) #endif /** @@ -525,17 +525,17 @@ void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, * Prefer drm_device based logging over device or prink based logging. */ -__printf(2, 3) -void ___drm_dbg(enum drm_debug_category category, const char *format, ...); +__printf(3, 4) +void ___drm_dbg(struct _ddebug *desc, enum drm_debug_category category, const char *format, ...); __printf(1, 2) void __drm_err(const char *format, ...); #if !defined(CONFIG_DRM_USE_DYNAMIC_DEBUG) -#define __drm_dbg(fmt, ...)___drm_dbg(fmt, ##__VA_ARGS__) +#define __drm_dbg(fmt, ...)___drm_dbg(NULL, fmt, ##__VA_ARGS__) #else #define __drm_dbg(cat, fmt, ...) \ - _d
[PATCH v2 24/27] dyndbg: add _DPRINTK_FLAGS_TRACE
add new flag, and OR it into _DPRINTK_FLAGS_ENABLED definition CC: vincent.whitchu...@axis.com Signed-off-by: Jim Cromie --- include/linux/dynamic_debug.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index 0a81a3582281..7581fe79980a 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -42,7 +42,9 @@ struct _ddebug { (_DPRINTK_FLAGS_INCL_MODNAME | _DPRINTK_FLAGS_INCL_FUNCNAME |\ _DPRINTK_FLAGS_INCL_LINENO | _DPRINTK_FLAGS_INCL_TID) -#define _DPRINTK_FLAGS_ENABLED _DPRINTK_FLAGS_PRINTK +#define _DPRINTK_FLAGS_TRACE (1 << 5) +#define _DPRINTK_FLAGS_ENABLED (_DPRINTK_FLAGS_PRINTK | \ +_DPRINTK_FLAGS_TRACE) #if defined DEBUG #define _DPRINTK_FLAGS_DEFAULT _DPRINTK_FLAGS_PRINTK -- 2.35.3
[PATCH v2 25/27] dyndbg: add write-events-to-tracefs code
adds: ddebug_trace() uses trace_console() temporarily to issue printk:console event uses internal-ish __ftrace_trace_stack code: 4-context buffer stack, barriers per Steve Rostedt call it from new funcs: ddebug_printk() - print to both syslog/tracefs ddebug_dev_printk() - dev-print to both syslog/tracefs These handle both _DPRINTK_FLAGS_PRINTK and _DPRINTK_FLAGS_TRACE cases, allowing to vsnprintf the message once and use it for both, skipping past the KERN_DEBUG character for tracing. Finally, adjust the callers: __ddebug_{pr_debug,{,net,ib}dev_dbg}, replacing printk and dev_printk with the new funcs above. The _DPRINTK_FLAGS_TRACE flag character is 'T', so the following finds all callsites enabled for tracing: grep -P =p?T /proc/dynamic_debug/control This patch,~1,~2 are basically copies of: https://lore.kernel.org/lkml/20200825153338.17061-1-vincent.whitchu...@axis.com/ with a few differences: - s/dynamic_/ddebug_/ on Vincent's additions - __printf attrs on the _printk funcs - reuses trace_console() event, not adding a new "printk:dyndbg" event. next patch replaces this with 2 new events CC: vincent.whitchu...@axis.com Signed-off-by: Jim Cromie --- .../admin-guide/dynamic-debug-howto.rst | 1 + drivers/gpu/drm/drm_print.c | 4 +- lib/dynamic_debug.c | 156 +++--- 3 files changed, 133 insertions(+), 28 deletions(-) diff --git a/Documentation/admin-guide/dynamic-debug-howto.rst b/Documentation/admin-guide/dynamic-debug-howto.rst index 01ca6f635dcc..b8c6772b88ac 100644 --- a/Documentation/admin-guide/dynamic-debug-howto.rst +++ b/Documentation/admin-guide/dynamic-debug-howto.rst @@ -239,6 +239,7 @@ of the characters:: The flags are:: penables the pr_debug() callsite. + Tenables callsite to issue a dyndbg:* trace-event fInclude the function name in the printed message lInclude line number in the printed message mInclude module name in the printed message diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index 77b56399147a..63125a507577 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -280,11 +280,11 @@ void __drm_dev_dbg(struct _ddebug *desc, const struct device *dev, vaf.va = &args; if (dev) { - if (desc->flags && _DPRINTK_FLAGS_PRINTK) + if (desc->flags & _DPRINTK_FLAGS_PRINTK) dev_printk(KERN_DEBUG, dev, "[" DRM_NAME ":%ps] %pV", __builtin_return_address(0), &vaf); } else { - if (desc->flags && _DPRINTK_FLAGS_PRINTK) + if (desc->flags & _DPRINTK_FLAGS_PRINTK) printk(KERN_DEBUG "[" DRM_NAME ":%ps] %pV", __builtin_return_address(0), &vaf); } diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index 5682ee477677..81cfe6e21d15 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -36,6 +36,7 @@ #include #include #include +#include #include @@ -88,6 +89,7 @@ static inline const char *trim_prefix(const char *path) static struct { unsigned flag:8; char opt_char; } opt_array[] = { { _DPRINTK_FLAGS_PRINTK, 'p' }, + { _DPRINTK_FLAGS_TRACE, 'T' }, { _DPRINTK_FLAGS_INCL_MODNAME, 'm' }, { _DPRINTK_FLAGS_INCL_FUNCNAME, 'f' }, { _DPRINTK_FLAGS_INCL_LINENO, 'l' }, @@ -770,6 +772,98 @@ static inline char *dynamic_emit_prefix(struct _ddebug *desc, char *buf) return buf; } +/* + * This code is heavily based on __ftrace_trace_stack(). + * + * Allow 4 levels of nesting: normal, softirq, irq, NMI. + */ +#define DYNAMIC_TRACE_NESTING 4 + +struct ddebug_trace_buf { + char buf[256]; +}; + +struct ddebug_trace_bufs { + struct ddebug_trace_buf bufs[DYNAMIC_TRACE_NESTING]; +}; + +static DEFINE_PER_CPU(struct ddebug_trace_bufs, ddebug_trace_bufs); +static DEFINE_PER_CPU(int, ddebug_trace_reserve); + +static void ddebug_trace(const char *fmt, va_list args) +{ + struct ddebug_trace_buf *buf; + int bufidx; + int len; + + preempt_disable_notrace(); + + bufidx = __this_cpu_inc_return(ddebug_trace_reserve) - 1; + + if (WARN_ON_ONCE(bufidx > DYNAMIC_TRACE_NESTING)) + goto out; + + /* For the same reasons as in __ftrace_trace_stack(). */ + barrier(); + + buf = this_cpu_ptr(ddebug_trace_bufs.bufs) + bufidx; + + len = vscnprintf(buf->buf, sizeof(buf->buf), fmt, args); + trace_console(buf->buf, len); + +out: + /* As above. */ + barrier(); + __this_cpu_dec(ddebug_trace_reserve); + preempt_enable_notrace(); +} + +__printf(2, 3) +static void ddebug_printk(unsigned int flags, const char *fmt, ...) +{ + if (flags & _DPRINTK_FLAGS_TRACE) { + va_list args; + + va_start(args, fmt); + /* +* All callers include the KERN
[PATCH v2 23/27] dyndbg: add _DPRINTK_FLAGS_ENABLED
Distinguish the condition: _DPRINTK_FLAGS_ENABLED from the bit: _DPRINTK_FLAGS_PRINT (and define former as latter), in preparation to add another bit next: _DPRINTK_FLAGS_TRACE And change JUMP_LABEL code block to use the more general _DPRINTK_FLAGS_ENABLED symbol. Also add a 'K' to get new symbol _DPRINTK_FLAGS_PRINTK, in order to break any stale uses. CC: vincent.whitchu...@axis.com Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_print.c | 4 ++-- include/linux/dynamic_debug.h | 10 ++ lib/dynamic_debug.c | 8 3 files changed, 12 insertions(+), 10 deletions(-) diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index 1752b40b0189..77b56399147a 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -280,11 +280,11 @@ void __drm_dev_dbg(struct _ddebug *desc, const struct device *dev, vaf.va = &args; if (dev) { - if (desc->flags && _DPRINTK_FLAGS_PRINT) + if (desc->flags && _DPRINTK_FLAGS_PRINTK) dev_printk(KERN_DEBUG, dev, "[" DRM_NAME ":%ps] %pV", __builtin_return_address(0), &vaf); } else { - if (desc->flags && _DPRINTK_FLAGS_PRINT) + if (desc->flags && _DPRINTK_FLAGS_PRINTK) printk(KERN_DEBUG "[" DRM_NAME ":%ps] %pV", __builtin_return_address(0), &vaf); } diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index abf43e28d8d3..0a81a3582281 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -32,7 +32,7 @@ struct _ddebug { * writes commands to /dynamic_debug/control */ #define _DPRINTK_FLAGS_NONE0 -#define _DPRINTK_FLAGS_PRINT (1<<0) /* printk() a message using the format */ +#define _DPRINTK_FLAGS_PRINTK (1 << 0) /* printk() a message using the format */ #define _DPRINTK_FLAGS_INCL_MODNAME(1<<1) #define _DPRINTK_FLAGS_INCL_FUNCNAME (1<<2) #define _DPRINTK_FLAGS_INCL_LINENO (1<<3) @@ -42,8 +42,10 @@ struct _ddebug { (_DPRINTK_FLAGS_INCL_MODNAME | _DPRINTK_FLAGS_INCL_FUNCNAME |\ _DPRINTK_FLAGS_INCL_LINENO | _DPRINTK_FLAGS_INCL_TID) +#define _DPRINTK_FLAGS_ENABLED _DPRINTK_FLAGS_PRINTK + #if defined DEBUG -#define _DPRINTK_FLAGS_DEFAULT _DPRINTK_FLAGS_PRINT +#define _DPRINTK_FLAGS_DEFAULT _DPRINTK_FLAGS_PRINTK #else #define _DPRINTK_FLAGS_DEFAULT 0 #endif @@ -161,10 +163,10 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, #ifdef DEBUG #define DYNAMIC_DEBUG_BRANCH(descriptor) \ - likely(descriptor.flags & _DPRINTK_FLAGS_PRINT) + likely(descriptor.flags & _DPRINTK_FLAGS_ENABLED) #else #define DYNAMIC_DEBUG_BRANCH(descriptor) \ - unlikely(descriptor.flags & _DPRINTK_FLAGS_PRINT) + unlikely(descriptor.flags & _DPRINTK_FLAGS_ENABLED) #endif #endif /* CONFIG_JUMP_LABEL */ diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index f9c5bbf9d43b..5682ee477677 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -87,7 +87,7 @@ static inline const char *trim_prefix(const char *path) } static struct { unsigned flag:8; char opt_char; } opt_array[] = { - { _DPRINTK_FLAGS_PRINT, 'p' }, + { _DPRINTK_FLAGS_PRINTK, 'p' }, { _DPRINTK_FLAGS_INCL_MODNAME, 'm' }, { _DPRINTK_FLAGS_INCL_FUNCNAME, 'f' }, { _DPRINTK_FLAGS_INCL_LINENO, 'l' }, @@ -238,10 +238,10 @@ static int ddebug_change(const struct ddebug_query *query, if (newflags == dp->flags) continue; #ifdef CONFIG_JUMP_LABEL - if (dp->flags & _DPRINTK_FLAGS_PRINT) { - if (!(newflags & _DPRINTK_FLAGS_PRINT)) + if (dp->flags & _DPRINTK_FLAGS_ENABLED) { + if (!(newflags & _DPRINTK_FLAGS_ENABLED)) static_branch_disable(&dp->key.dd_key_true); - } else if (newflags & _DPRINTK_FLAGS_PRINT) { + } else if (newflags & _DPRINTK_FLAGS_ENABLED) { static_branch_enable(&dp->key.dd_key_true); } #endif -- 2.35.3
[PATCH v2 26/27] dyndbg: 4 new trace-events: pr_debug, dev_dbg, drm_{, dev}debug
ddebug_trace() currently issues a single printk:console event. Replace that, adding include/trace/events/dyndbg.h, which defines 2 new events: - dyndbg:prdbg - from trace_prdbg() - if !dev - dyndbg:devdbg - from trace_devdbg() - if !!dev This links the legacy pr_debug API to tracefs, via dyndbg, allowing pr_debug()s etc to add just a little more user-context to the trace-logs, and then at users option, less syslog. The 2 new trace_*() calls accept their caller's args respectively, keeping the available info w/o alteration; we can't improve on transparency. The args: 1- struct _ddebug *descriptor, giving tracefs all of dyndbg's info. 2- struct device *dev, used by trace_devdbg(), if !!dev The trace_*() calls need the descriptor arg, the callchain prototypes above them are extended to provide it. The existing category param in this callchain is partially redundant; when descriptor is available, it has the class_id. dev_dbg(desc, dev...), if dev is true, issues a trace_devdbg(), otherwise trace_prdbg(). This way we dont consume buffer space storing nulls. Otherwise the events are equivalent. Also add include/trace/events/drm.h, to create 2 events for use in drm_dbg() and drm_devdbg(), and call them. This recapitulates the changes described above, connecting 3-10K drm.debug callsites to tracefs. Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_print.c | 9 + include/trace/events/drm.h| 68 include/trace/events/dyndbg.h | 74 +++ lib/dynamic_debug.c | 73 +- 4 files changed, 188 insertions(+), 36 deletions(-) create mode 100644 include/trace/events/drm.h create mode 100644 include/trace/events/dyndbg.h diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index 63125a507577..9afb676bda4d 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -34,6 +34,9 @@ #include #include +#define CREATE_TRACE_POINTS +#include + /* * __drm_debug: Enable debug output. * Bitmask of DRM_UT_x. See include/drm/drm_print.h for details. @@ -283,10 +286,14 @@ void __drm_dev_dbg(struct _ddebug *desc, const struct device *dev, if (desc->flags & _DPRINTK_FLAGS_PRINTK) dev_printk(KERN_DEBUG, dev, "[" DRM_NAME ":%ps] %pV", __builtin_return_address(0), &vaf); + if (desc->flags & _DPRINTK_FLAGS_TRACE) + trace_drm_devdbg(dev, category, &vaf); } else { if (desc->flags & _DPRINTK_FLAGS_PRINTK) printk(KERN_DEBUG "[" DRM_NAME ":%ps] %pV", __builtin_return_address(0), &vaf); + if (desc->flags & _DPRINTK_FLAGS_TRACE) + trace_drm_debug(category, &vaf); } va_end(args); } @@ -307,6 +314,8 @@ void ___drm_dbg(struct _ddebug *desc, enum drm_debug_category category, const ch printk(KERN_DEBUG "[" DRM_NAME ":%ps] %pV", __builtin_return_address(0), &vaf); + trace_drm_debug(category, &vaf); + va_end(args); } EXPORT_SYMBOL(___drm_dbg); diff --git a/include/trace/events/drm.h b/include/trace/events/drm.h new file mode 100644 index ..6de80dd68620 --- /dev/null +++ b/include/trace/events/drm.h @@ -0,0 +1,68 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#undef TRACE_SYSTEM +#define TRACE_SYSTEM drm + +#if !defined(_TRACE_DRM_H) || defined(TRACE_HEADER_MULTI_READ) +#define _TRACE_DRM_H + +#include + +/* drm_debug() was called, pass its args */ +TRACE_EVENT(drm_debug, + TP_PROTO(int drm_debug_category, struct va_format *vaf), + + TP_ARGS(drm_debug_category, vaf), + + TP_STRUCT__entry( + __field(int, drm_debug_category) + __dynamic_array(char, msg, 256) + ), + + TP_fast_assign( + int len; + + __entry->drm_debug_category = drm_debug_category; + vsnprintf(__get_str(msg), 256, vaf->fmt, *vaf->va); + + len = strlen(__get_str(msg)); + if (len > 0 && (__get_str(msg)[len - 1] == '\n')) + len -= 1; + __get_str(msg)[len] = 0; + ), + + TP_printk("%s", __get_str(msg)) +); + +/* drm_devdbg() was called, pass its args, preserving order */ +TRACE_EVENT(drm_devdbg, + TP_PROTO(const struct device *dev, int drm_debug_category, struct va_format *vaf), + + TP_ARGS(dev, drm_debug_category, vaf), + + TP_STRUCT__entry( + __field(const struct device*, dev) + __field(int, drm_debug_category) + __dynamic_array(char, msg, 256) + ), + + TP_fast_assign( + int len; + + __entry->drm_debug_category = drm_debug_categor
[PATCH v2 20/27] drm_print: refine drm_debug_enabled for jump-label
In order to use dynamic-debug's jump-label optimization in drm-debug, its clarifying to refine drm_debug_enabled into 3 uses: 1. drm_debug_enabled - legacy, public 2. __drm_debug_enabled - optimized for dyndbg jump-label enablement. 3. _drm_debug_enabled - pr_debug instrumented, observable 1. The legacy version always checks the bits. 2. is privileged, for use by __drm_dbg(), __drm_dev_dbg(), which do an early return unless the category is enabled (free of call/NOOP side effects). For dyndbg builds, debug callsites are selectively "pre-enabled", so __drm_debug_enabled() short-circuits to true there. Remaining callers of 1 may be able to use 2, case by case. 3. is 1st wrapped in a macro, with a pr_debug, which reports each usage in /proc/dynamic_debug/control, making it observable in the logs. The macro lets the pr_debug see the real caller, not an inline function. When plugged into 1, it identified ~10 remaining callers of the function, leading to the follow-on cleanup patch, and would allow activating the pr_debugs, estimating the callrate, and the potential savings by using the wrapper macro. It is unused ATM, but it fills out the picture. Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_print.c | 4 ++-- include/drm/drm_print.h | 28 2 files changed, 30 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index 5192533794a2..22b7ab1c5c40 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -272,7 +272,7 @@ void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, struct va_format vaf; va_list args; - if (!drm_debug_enabled(category)) + if (!__drm_debug_enabled(category)) return; va_start(args, format); @@ -295,7 +295,7 @@ void ___drm_dbg(enum drm_debug_category category, const char *format, ...) struct va_format vaf; va_list args; - if (!drm_debug_enabled(category)) + if (!__drm_debug_enabled(category)) return; va_start(args, format); diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h index 5c39bacac2b3..ccd177236ab3 100644 --- a/include/drm/drm_print.h +++ b/include/drm/drm_print.h @@ -324,11 +324,39 @@ enum drm_debug_category { DRM_UT_DRMRES }; +/* + * 3 name flavors of drm_debug_enabled: + * drm_debug_enabled - public/legacy, always checks bits + * _drm_debug_enabled - instrumented to observe call-rates, est overheads. + * __drm_debug_enabled - privileged - knows jump-label state, can short-circuit + */ static inline bool drm_debug_enabled(enum drm_debug_category category) { return unlikely(__drm_debug & BIT(category)); } +/* + * Wrap fn in macro, so that the pr_debug sees the actual caller, not + * the inline fn. Using this name creates a callsite entry / control + * point in /proc/dynamic_debug/control. + */ +#define _drm_debug_enabled(category) \ + ({ \ + pr_debug("todo: maybe avoid via dyndbg\n"); \ + drm_debug_enabled(category);\ + }) + +#if defined(CONFIG_DRM_USE_DYNAMIC_DEBUG) +/* + * dyndbg is wrapping the drm.debug API, so as to avoid the runtime + * bit-test overheads of drm_debug_enabled() in those api calls. + * In this case, executed callsites are known enabled, so true. + */ +#define __drm_debug_enabled(category) true +#else +#define __drm_debug_enabled(category) drm_debug_enabled(category) +#endif + /* * struct device based logging * -- 2.35.3
[PATCH v2 18/27] drm_print: interpose drm_*dbg with forwarding macros
change drm_dev_dbg & drm_dbg to macros, which forward to the renamed functions (with __ prefix added). Those functions sit below the categorized layer of macros implementing the DRM debug.category API, and implement most of it. These are good places to insert dynamic-debug jump-label mechanics, which will allow DRM to avoid the runtime cost of drm_debug_enabled(). no functional changes. memory cost baseline: (unchanged) bash-5.1# drms_load [9.220389] dyndbg: 1 debug prints in module drm [9.224426] ACPI: bus type drm_connector registered [9.302192] dyndbg: 2 debug prints in module ttm [9.305033] dyndbg: 8 debug prints in module video [9.627563] dyndbg: 127 debug prints in module i915 [9.721505] AMD-Vi: AMD IOMMUv2 functionality not available on this system - This is not a bug. [ 10.091345] dyndbg: 2196 debug prints in module amdgpu [ 10.106589] [drm] amdgpu kernel modesetting enabled. [ 10.107270] amdgpu: CRAT table not found [ 10.107926] amdgpu: Virtual CRAT table created for CPU [ 10.108398] amdgpu: Topology: Add CPU node [ 10.168507] dyndbg: 3 debug prints in module wmi [ 10.329587] dyndbg: 3 debug prints in module nouveau Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_print.c | 10 +- include/drm/drm_print.h | 9 +++-- 2 files changed, 12 insertions(+), 7 deletions(-) diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index e49ffda0a47e..e77b49c7725b 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -266,8 +266,8 @@ void drm_dev_printk(const struct device *dev, const char *level, } EXPORT_SYMBOL(drm_dev_printk); -void drm_dev_dbg(const struct device *dev, enum drm_debug_category category, -const char *format, ...) +void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, + const char *format, ...) { struct va_format vaf; va_list args; @@ -288,9 +288,9 @@ void drm_dev_dbg(const struct device *dev, enum drm_debug_category category, va_end(args); } -EXPORT_SYMBOL(drm_dev_dbg); +EXPORT_SYMBOL(__drm_dev_dbg); -void __drm_dbg(enum drm_debug_category category, const char *format, ...) +void ___drm_dbg(enum drm_debug_category category, const char *format, ...) { struct va_format vaf; va_list args; @@ -307,7 +307,7 @@ void __drm_dbg(enum drm_debug_category category, const char *format, ...) va_end(args); } -EXPORT_SYMBOL(__drm_dbg); +EXPORT_SYMBOL(___drm_dbg); void __drm_err(const char *format, ...) { diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h index a157485bf573..383feb0d34fe 100644 --- a/include/drm/drm_print.h +++ b/include/drm/drm_print.h @@ -339,7 +339,7 @@ __printf(3, 4) void drm_dev_printk(const struct device *dev, const char *level, const char *format, ...); __printf(3, 4) -void drm_dev_dbg(const struct device *dev, enum drm_debug_category category, +void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, const char *format, ...); /** @@ -388,6 +388,9 @@ void drm_dev_dbg(const struct device *dev, enum drm_debug_category category, } \ }) +#define drm_dev_dbg(dev, cat, fmt, ...)\ + __drm_dev_dbg(dev, cat, fmt, ##__VA_ARGS__) + /** * DRM_DEV_DEBUG() - Debug output for generic drm code * @@ -489,10 +492,12 @@ void drm_dev_dbg(const struct device *dev, enum drm_debug_category category, */ __printf(2, 3) -void __drm_dbg(enum drm_debug_category category, const char *format, ...); +void ___drm_dbg(enum drm_debug_category category, const char *format, ...); __printf(1, 2) void __drm_err(const char *format, ...); +#define __drm_dbg(fmt, ...)___drm_dbg(fmt, ##__VA_ARGS__) + /* Macros to make printk easier */ #define _DRM_PRINTK(once, level, fmt, ...) \ -- 2.35.3
[PATCH v2 17/27] drm_print: condense enum drm_debug_category
enum drm_debug_category has 10 categories, but is initialized with bitmasks which require 10 bits of underlying storage. By using natural enumeration, and moving the BIT(cat) into drm_debug_enabled(), the enum fits in 4 bits, allowing the category to be represented directly in pr_debug callsites, via the ddebug.class_id field. While this slightly pessimizes the bit-test in drm_debug_enabled(), using dyndbg with JUMP_LABEL will avoid the function entirely. NOTE: this change forecloses the possibility of doing: drm_dbg(DRM_UT_CORE|DRM_UT_KMS, "weird 2-cat experiment") but thats already strongly implied by the use of the enum itself; its not a normal enum if it can be 2 values simultaneously. Signed-off-by: Jim Cromie --- include/drm/drm_print.h | 22 +++--- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h index 0597137bac75..a157485bf573 100644 --- a/include/drm/drm_print.h +++ b/include/drm/drm_print.h @@ -284,49 +284,49 @@ enum drm_debug_category { * @DRM_UT_CORE: Used in the generic drm code: drm_ioctl.c, drm_mm.c, * drm_memory.c, ... */ - DRM_UT_CORE = 0x01, + DRM_UT_CORE, /** * @DRM_UT_DRIVER: Used in the vendor specific part of the driver: i915, * radeon, ... macro. */ - DRM_UT_DRIVER = 0x02, + DRM_UT_DRIVER, /** * @DRM_UT_KMS: Used in the modesetting code. */ - DRM_UT_KMS = 0x04, + DRM_UT_KMS, /** * @DRM_UT_PRIME: Used in the prime code. */ - DRM_UT_PRIME= 0x08, + DRM_UT_PRIME, /** * @DRM_UT_ATOMIC: Used in the atomic code. */ - DRM_UT_ATOMIC = 0x10, + DRM_UT_ATOMIC, /** * @DRM_UT_VBL: Used for verbose debug message in the vblank code. */ - DRM_UT_VBL = 0x20, + DRM_UT_VBL, /** * @DRM_UT_STATE: Used for verbose atomic state debugging. */ - DRM_UT_STATE= 0x40, + DRM_UT_STATE, /** * @DRM_UT_LEASE: Used in the lease code. */ - DRM_UT_LEASE= 0x80, + DRM_UT_LEASE, /** * @DRM_UT_DP: Used in the DP code. */ - DRM_UT_DP = 0x100, + DRM_UT_DP, /** * @DRM_UT_DRMRES: Used in the drm managed resources code. */ - DRM_UT_DRMRES = 0x200, + DRM_UT_DRMRES }; static inline bool drm_debug_enabled(enum drm_debug_category category) { - return unlikely(__drm_debug & category); + return unlikely(__drm_debug & BIT(category)); } /* -- 2.35.3
[PATCH v2 15/27] drm: POC drm on dyndbg - map class-names to drm_debug_category's
Invoke DYNAMIC_DEBUG_CLASSES from drm_drv.h. This declares a maybe-unused struct ddebug_known_classes_map var, initialized with: . var: passed to dynamic_debug_register_classes() . class-names: "DRM_UT_CORE", "DRM_UT_DRIVER", "DRM_UT_KMS", etc. These names map to .class_id's by their index, ie: 0-30. Then in 4 test-case drm-drivers (drm, i915, amdgpu, nouveau); call dynamic_debug_register_classes(var). i915 also gets an adaptor func, and calls it 1st in the array of initialization helpers, since early logging might be valuable for diagnosing setup problems. Since these modules all use the same class-names, they all will respond together to class FOO changes: #> echo class DRM_UT_KMS +p > /proc/dynamic_debug/control NOTES: DRM uses enum drm_debug_category across modules and common core, so our class-names => index map must apply across them too, hence drm_drv.h invokes the macro once for everyone. DRM's enum drm_debug_category values need to sync with the index of their respective class-names here. Then .class_id == category, and dyndbg's class FOO mechanisms will work. Once enum drm_debug_category is naturalized (value inits dropped, yielding 0..N), then this condition holds true: assert(!strcmp(classes[DRM_UT_KMS],"DRM_UT_KMS")); Though DRM needs consistent categories across all modules, thats not generally needed; modules X and Y could define FOO differently (ie different corresponding .class_id) and things would work. No callsites are actually selected here, since none are class'd yet. bash-5.1# dmesg | grep register_class [7.095579] dyndbg: register_classes: drm [7.557109] dyndbg: register_classes: i915 [8.096818] dyndbg: register_classes: amdgpu Signed-off-by: Jim Cromie --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 +++ drivers/gpu/drm/drm_drv.c | 2 ++ drivers/gpu/drm/i915/i915_module.c | 11 +++ drivers/gpu/drm/nouveau/nouveau_drm.c | 4 include/drm/drm_drv.h | 14 ++ include/drm/drm_print.h | 4 6 files changed, 38 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 7fd0277b2805..addb991b4663 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -2765,6 +2765,8 @@ static int __init amdgpu_init(void) if (r) goto error_fence; + dynamic_debug_register_classes(&drm_debug_classes); + DRM_INFO("amdgpu kernel modesetting enabled.\n"); amdgpu_register_atpx_handler(); amdgpu_acpi_detect(); @@ -2787,6 +2789,7 @@ static void __exit amdgpu_exit(void) amdgpu_amdkfd_fini(); pci_unregister_driver(&amdgpu_kms_pci_driver); amdgpu_unregister_atpx_handler(); + dynamic_debug_unregister_classes(&drm_debug_classes); amdgpu_sync_fini(); amdgpu_fence_slab_fini(); mmu_notifier_synchronize(); diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c index 8214a0b1ab7f..16683fb169aa 100644 --- a/drivers/gpu/drm/drm_drv.c +++ b/drivers/gpu/drm/drm_drv.c @@ -1045,6 +1045,8 @@ static int __init drm_core_init(void) { int ret; + dynamic_debug_register_classes(&drm_debug_classes); + drm_connector_ida_init(); idr_init(&drm_minors_idr); drm_memcpy_init_early(); diff --git a/drivers/gpu/drm/i915/i915_module.c b/drivers/gpu/drm/i915/i915_module.c index 65acd7bf75d0..31f00f973866 100644 --- a/drivers/gpu/drm/i915/i915_module.c +++ b/drivers/gpu/drm/i915/i915_module.c @@ -44,10 +44,21 @@ static int i915_check_nomodeset(void) return 0; } +static int i915_ddebug_classes_register(void) +{ + return dynamic_debug_register_classes(&drm_debug_classes); +} +static void i915_ddebug_classes_unregister(void) +{ + dynamic_debug_unregister_classes(&drm_debug_classes); +} + static const struct { int (*init)(void); void (*exit)(void); } init_funcs[] = { + { .init = i915_ddebug_classes_register, + .exit = i915_ddebug_classes_unregister }, { .init = i915_check_nomodeset }, { .init = i915_active_module_init, .exit = i915_active_module_exit }, diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c index 561309d447e0..9a780b6d4796 100644 --- a/drivers/gpu/drm/nouveau/nouveau_drm.c +++ b/drivers/gpu/drm/nouveau/nouveau_drm.c @@ -1355,6 +1355,8 @@ nouveau_drm_init(void) driver_pci = driver_stub; driver_platform = driver_stub; + dynamic_debug_register_classes(&drm_debug_classes); + nouveau_display_options(); if (nouveau_modeset == -1) { @@ -1391,6 +1393,8 @@ nouveau_drm_exit(void) nouveau_backlight_dtor(); nouveau_unregister_dsm_handler(); + dynamic_debug_unregister_classes(&drm_debug_classes); + #ifdef CONFIG_NOUVEAU_PLATFORM_DRIVER platform_driver_unregister(&nouveau_platform_dr
[PATCH v2 19/27] drm_print: wrap drm_*_dbg in dyndbg descriptor factory macro
For CONFIG_DRM_USE_DYNAMIC_DEBUG=y, wrap __drm_dbg() & __drm_dev_dbg() in one of dyndbg's Factory macros: _dynamic_func_call_no_desc(). Next, those functions are adapted to accept a descriptor arg, and we drop the _no_desc suffix, then the (~4000) drm.debug callsites will be controllable by their class-names: #> echo class DRM_UT_ATOMIC +p > /proc/dynamic_debug/control CONFIG_DRM_USE_DYNAMIC_DEBUG=y/n is configurable because of the .data footprint cost of per-callsite control; 56 bytes/site * ~2k for i915, ~4k callsites for amdgpu. This is large enough that a kernel builder might not want it. Signed-off-by: Jim Cromie --- drivers/gpu/drm/Kconfig | 12 drivers/gpu/drm/Makefile| 2 ++ drivers/gpu/drm/drm_print.c | 2 +- include/drm/drm_print.h | 12 4 files changed, 27 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig index f1422bee3dcc..5ab31753c25d 100644 --- a/drivers/gpu/drm/Kconfig +++ b/drivers/gpu/drm/Kconfig @@ -63,6 +63,18 @@ config DRM_DEBUG_MM If in doubt, say "N". +config DRM_USE_DYNAMIC_DEBUG + bool "use dynamic debug to implement drm.debug" + default y + depends on DRM + depends on DYNAMIC_DEBUG || DYNAMIC_DEBUG_CORE + depends on JUMP_LABEL + help + Use dynamic-debug to avoid drm_debug_enabled() runtime overheads. + Due to callsite counts in DRM drivers (~4k in amdgpu) and 56 + bytes per callsite, the .data costs can be substantial, and + are therefore configurable. + config DRM_DEBUG_SELFTEST tristate "kselftests for DRM" depends on DRM diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile index c2ef5f9fce54..e93b7902994a 100644 --- a/drivers/gpu/drm/Makefile +++ b/drivers/gpu/drm/Makefile @@ -3,6 +3,8 @@ # Makefile for the drm device driver. This driver provides support for the # Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher. +CFLAGS-$(CONFIG_DRM_USE_DYNAMIC_DEBUG) += -DDYNAMIC_DEBUG_MODULE + drm-y := drm_aperture.o drm_auth.o drm_cache.o \ drm_file.o drm_gem.o drm_ioctl.o \ drm_drv.o \ diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index e77b49c7725b..5192533794a2 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -53,7 +53,7 @@ MODULE_PARM_DESC(debug, "Enable debug output, where each bit enables a debug cat "\t\tBit 7 (0x80) will enable LEASE messages (leasing code)\n" "\t\tBit 8 (0x100) will enable DP messages (displayport code)"); -#if !defined(CONFIG_DYNAMIC_DEBUG_CORE) +#if !defined(CONFIG_DRM_USE_DYNAMIC_DEBUG) module_param_named(debug, __drm_debug, ulong, 0600); #else static struct ddebug_classes_bitmap_param drm_debug_bitmap = { diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h index 383feb0d34fe..5c39bacac2b3 100644 --- a/include/drm/drm_print.h +++ b/include/drm/drm_print.h @@ -388,8 +388,14 @@ void __drm_dev_dbg(const struct device *dev, enum drm_debug_category category, } \ }) +#if !defined(CONFIG_DRM_USE_DYNAMIC_DEBUG) #define drm_dev_dbg(dev, cat, fmt, ...)\ __drm_dev_dbg(dev, cat, fmt, ##__VA_ARGS__) +#else +#define drm_dev_dbg(dev, cat, fmt, ...)\ + _dynamic_func_call_no_desc(fmt, __drm_dev_dbg, \ + dev, cat, fmt, ##__VA_ARGS__) +#endif /** * DRM_DEV_DEBUG() - Debug output for generic drm code @@ -496,7 +502,13 @@ void ___drm_dbg(enum drm_debug_category category, const char *format, ...); __printf(1, 2) void __drm_err(const char *format, ...); +#if !defined(CONFIG_DRM_USE_DYNAMIC_DEBUG) #define __drm_dbg(fmt, ...)___drm_dbg(fmt, ##__VA_ARGS__) +#else +#define __drm_dbg(cat, fmt, ...) \ + _dynamic_func_call_no_desc(fmt, ___drm_dbg, \ + cat, fmt, ##__VA_ARGS__) +#endif /* Macros to make printk easier */ -- 2.35.3
[PATCH v2 16/27] drm/print: POC drm on dyndbg - use bitmap
POC: adapt drm_print to use/test the bitmap callback support. with dynamic_debug.verbose=1: bash-5.1# echo 1 > /sys/module/drm/parameters/debug [ 33.697039] dyndbg: set_dyndbg_classes: new 0x1 current 0x0 [ 33.697571] dyndbg: query 0: "class DRM_UT_CORE +p" mod:* [ 33.698072] dyndbg: no matches for query [ 33.698498] dyndbg: total matches: 0 bash-5.1# echo 2 > /sys/module/drm/parameters/debug [ 46.687012] dyndbg: set_dyndbg_classes: new 0x2 current 0x1 [ 46.687655] dyndbg: query 0: "class DRM_UT_CORE -p" mod:* [ 46.688280] dyndbg: no matches for query [ 46.688632] dyndbg: query 0: "class DRM_UT_DRIVER +p" mod:* [ 46.689122] dyndbg: no matches for query [ 46.689441] dyndbg: total matches: 0 This changes typeof __drm_debug to unsigned long from unsigned int, which dyndbg requires so it can use BIT(). It is currently dependent on CONFIG_DYNAMIC_DEBUG_CORE, which is only aproximately correct, it will need proper DRM dependence too. Signed-off-by: Jim Cromie --- drivers/gpu/drm/drm_print.c | 14 -- include/drm/drm_print.h | 3 ++- 2 files changed, 14 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c index f783d4963d4b..e49ffda0a47e 100644 --- a/drivers/gpu/drm/drm_print.c +++ b/drivers/gpu/drm/drm_print.c @@ -40,7 +40,7 @@ * __drm_debug: Enable debug output. * Bitmask of DRM_UT_x. See include/drm/drm_print.h for details. */ -unsigned int __drm_debug; +unsigned long __drm_debug; EXPORT_SYMBOL(__drm_debug); MODULE_PARM_DESC(debug, "Enable debug output, where each bit enables a debug category.\n" @@ -52,7 +52,17 @@ MODULE_PARM_DESC(debug, "Enable debug output, where each bit enables a debug cat "\t\tBit 5 (0x20) will enable VBL messages (vblank code)\n" "\t\tBit 7 (0x80) will enable LEASE messages (leasing code)\n" "\t\tBit 8 (0x100) will enable DP messages (displayport code)"); -module_param_named(debug, __drm_debug, int, 0600); + +#if !defined(CONFIG_DYNAMIC_DEBUG_CORE) +module_param_named(debug, __drm_debug, ulong, 0600); +#else +static struct ddebug_classes_bitmap_param drm_debug_bitmap = { + .bits = &__drm_debug, + .flags = "p", + .map = &drm_debug_classes, +}; +module_param_cb(debug, ¶m_ops_dyndbg_classes, &drm_debug_bitmap, 0600); +#endif void __drm_puts_coredump(struct drm_printer *p, const char *str) { diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h index 5b7eedb0f477..0597137bac75 100644 --- a/include/drm/drm_print.h +++ b/include/drm/drm_print.h @@ -31,11 +31,12 @@ #include #include #include +#include #include /* Do *not* use outside of drm_print.[ch]! */ -extern unsigned int __drm_debug; +extern unsigned long __drm_debug; /** * DOC: print -- 2.35.3
[PATCH v2 14/27] dyndbg: add test_dynamic_debug module
Demonstrate dyndbg's "class FOO" and bitmap-to-classes support. This support is meant to plug into drm.debug system, and largely replace the use of drm_debug_enabled(category) with JUMP_LABELs. Recap: #> echo class DRM_UT_CORE +p > /proc/dynamic_debug/control This is made "safe" because dyndbg skips it for any modules which don't know that class (havent called dynamic_debug_register_classes). Other modules may use the same .class_id for a separate classified debug scheme. Use the API: - DYNAMIC_DEBUG_CLASSES(_var, classes), to declare static _var by name - dynamic_debug_register_classes(_var) - dynamic_debug_unregister_classes(_var) Use these 3 times; with base = 0,8,16 respectively, to demonstrate the segmenting of the module's .class_id range [0..30] For each of those 3 class-name-sets, add 2 sysfs-node-bitmaps, one each for p-syslog, and T-tracefs, the latter will work once dyndbg gets that patchset. #> modprobe test_dynamic_debug dyndbg=+pfm #> cat /sys/module/test_dynamic_debug/parameters/do_prints #> echo class FOO +pf > /proc/dynamic_debug/control #> echo class Foo +pfm > /proc/dynamic_debug/control #> cat /sys/module/test_dynamic_debug/parameters/do_prints RFC: This use case exposes a weak point in the api; the 2nd query command given in the dyndbg option will not work like the 1st: #> modprobe test_dynamic_debug dyndbg='+pfm; class FOO +pfm' This is because the option is processed early in module-load, well before the registration can attach the class-map to the module's ddebug_table entry. Signed-off-by: Jim Cromie --- MAINTAINERS | 1 + lib/Kconfig.debug| 11 +++ lib/Makefile | 1 + lib/test_dynamic_debug.c | 172 +++ 4 files changed, 185 insertions(+) create mode 100644 lib/test_dynamic_debug.c diff --git a/MAINTAINERS b/MAINTAINERS index e8c52d0192a6..bf615853be47 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -6918,6 +6918,7 @@ M:Jason Baron S: Maintained F: include/linux/dynamic_debug.h F: lib/dynamic_debug.c +F: lib/test_dynamic_debug.c DYNAMIC INTERRUPT MODERATION M: Tal Gilboa diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug index 075cd25363ac..c88d691d3df1 100644 --- a/lib/Kconfig.debug +++ b/lib/Kconfig.debug @@ -2601,6 +2601,17 @@ config TEST_STATIC_KEYS If unsure, say N. +config TEST_DYNAMIC_DEBUG + tristate "Test DYNAMIC_DEBUG" + depends on m + depends on DYNAMIC_DEBUG + help + This module registers a tracer callback to count enabled + pr_debugs in a 'do_debugging' function, then alters their + enablements, calls the function, and compares counts. + + If unsure, say N. + config TEST_KMOD tristate "kmod stress tester" depends on m diff --git a/lib/Makefile b/lib/Makefile index 6b9ffc1bd1ee..e5727fbbfc7d 100644 --- a/lib/Makefile +++ b/lib/Makefile @@ -82,6 +82,7 @@ obj-$(CONFIG_TEST_SORT) += test_sort.o obj-$(CONFIG_TEST_USER_COPY) += test_user_copy.o obj-$(CONFIG_TEST_STATIC_KEYS) += test_static_keys.o obj-$(CONFIG_TEST_STATIC_KEYS) += test_static_key_base.o +obj-$(CONFIG_TEST_DYNAMIC_DEBUG) += test_dynamic_debug.o obj-$(CONFIG_TEST_PRINTF) += test_printf.o obj-$(CONFIG_TEST_SCANF) += test_scanf.o obj-$(CONFIG_TEST_BITMAP) += test_bitmap.o diff --git a/lib/test_dynamic_debug.c b/lib/test_dynamic_debug.c new file mode 100644 index ..65c37ba6c0da --- /dev/null +++ b/lib/test_dynamic_debug.c @@ -0,0 +1,172 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Kernel module for testing dynamic_debug + * + * Authors: + * Jim Cromie + */ + +#define pr_fmt(fmt) "test_dd: " fmt + +#include + +static void do_prints(void); /* device under test */ + +/* run tests by reading or writing sysfs node */ + +int param_set_do_prints(const char *instr, const struct kernel_param *kp) +{ + do_prints(); + return 0; +} +EXPORT_SYMBOL(param_set_do_prints); + +int param_get_do_prints(char *buffer, const struct kernel_param *kp) +{ + do_prints(); + return scnprintf(buffer, PAGE_SIZE, "did do_prints\n"); +} +EXPORT_SYMBOL(param_get_do_prints); + +const struct kernel_param_ops param_ops_do_prints = { + .set = param_set_do_prints, + .get = param_get_do_prints, +}; +EXPORT_SYMBOL(param_ops_do_prints); + +module_param_cb(do_prints, ¶m_ops_do_prints, NULL, 0600); + +/* + * Declare 3 groups of classes, with different .class_id[] ranges, + * each with 2 sysfs-node bitmaps controlling p,T flags respectively + * for those named classes. This example is rather more involved than + * anyone will likely use. + + * The T-bitmap sysfs-node functionality will need a few patches which + * add trace-events to dyndbg. + + * Rules: + * - enum symbols must match/correlate with class-name strings + * - base must equal enum's 1st value + */ + +enum cat1 { FOO, BAR, BUZZ }; +DYNAMIC_DEBUG_CLASSES(ddt_classes1, 0, + "FOO",
[PATCH v2 13/27] dyndbg: add __pr_debug_cls(class, fmt, ...)
For selftest purposes, add __pr_debug_cls(class, fmt, ...) I didn't think we'd need to define this, since DRM effectively has it already. But test_dynamic_debug needs it in order to demonstrate all the moving parts. Note the __ prefix; its not intended for general use, and doesn't include any builtin-constant checks that could pertain. I'd prefer to see a use-case where copying the drm.debug model isn't better. Signed-off-by: Jim Cromie --- include/linux/dynamic_debug.h | 7 +++ 1 file changed, 7 insertions(+) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index e9e6d0f503f3..abf43e28d8d3 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -216,6 +216,10 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, _DPRINTK_CLASS_DFLT,\ fmt, func, ##__VA_ARGS__) +#define dynamic_pr_debug_cls(cls, fmt, ...)\ + _dynamic_func_call_cls(cls, fmt, __dynamic_pr_debug,\ + pr_fmt(fmt), ##__VA_ARGS__) + #define dynamic_pr_debug(fmt, ...) \ _dynamic_func_call(fmt, __dynamic_pr_debug, \ pr_fmt(fmt), ##__VA_ARGS__) @@ -246,6 +250,9 @@ struct kernel_param; int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp); int param_get_dyndbg_classes(char *buffer, const struct kernel_param *kp); +#define __pr_debug_cls(cls, fmt, ...) \ + dynamic_pr_debug_cls(cls, fmt, ##__VA_ARGS__) + #else /* !CONFIG_DYNAMIC_DEBUG_CORE */ #include -- 2.35.3
[PATCH v2 12/27] dyndbg: change zero-or-one classes-map to maps list
Upgrade single classes-map to list of them: This allows multiple DYNAMIC_DEBUG_CLASSES(class-map)s per module, using _base to segment the 0..30 classid space. alter struct ddebug table: replace .classes (a &map) with maps (list-head) dynamic_debug_register_classes(map) - adds new map to maps list. dynamic_debug_unregister_classes(map) - deletes map after ID-check. ddebug_validate_classname() - check all maps in list before failing. ddebug_class_name() - which supports ```cat control``` now walks maps list, finds the map whose sub-range of .class_id's spans the one in the callsite, and returns that class-name. Signed-off-by: Jim Cromie --- . split out validate_classnames() . fold in fixes for multi class-maps --- lib/dynamic_debug.c | 76 +++-- 1 file changed, 52 insertions(+), 24 deletions(-) diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index 8e1b9159e881..f9c5bbf9d43b 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -43,10 +43,8 @@ extern struct _ddebug __start___dyndbg[]; extern struct _ddebug __stop___dyndbg[]; struct ddebug_table { - struct list_head link; + struct list_head link, maps; const char *mod_name; - /* a module can have multiple class-sets eventually, but not yet */ - struct ddebug_known_classes_map const *map; unsigned int num_ddebugs; struct _ddebug *ddebugs; }; @@ -149,28 +147,18 @@ static void vpr_info_dq(const struct ddebug_query *query, const char *msg) /* return <0 if class-name is unknown/invalid, 0..CLASS_DFLT otherwise */ static int ddebug_validate_classname(struct ddebug_table *dt, const char *class_string) { - int query_class = -ENOENT; + struct ddebug_known_classes_map *map; int idx; if (!class_string) - /* all queries w/o class given work only on default class */ return _DPRINTK_CLASS_DFLT; - /* -* XXX single list will need to be a for-list -* so that modules can have 2 sets of class-decls -*/ - if (!dt->map) - return -ENOENT; - - idx = match_string(dt->map->classes, dt->map->length, class_string); - if (idx < 0) { - v3pr_info("class: %s.%s unknown\n", dt->mod_name, class_string); - return -ENOENT; + list_for_each_entry(map, &dt->maps, link) { + idx = match_string(map->classes, map->length, class_string); + if (idx >= 0) + return idx + map->base; } - query_class = idx + dt->map->base; - - return query_class; + return -ENOENT; } /* @@ -1032,8 +1020,14 @@ static void *ddebug_proc_next(struct seq_file *m, void *p, loff_t *pos) static const char *ddebug_class_name(struct ddebug_iter *iter, struct _ddebug *dp) { - if (iter->table->map) - return iter->table->map->classes[dp->class_id]; + struct ddebug_known_classes_map *map; + + list_for_each_entry(map, &iter->table->maps, link) { + if (dp->class_id < map->base || + dp->class_id >= map->base + map->length) + continue; + return map->classes[dp->class_id - map->base]; + } return NULL; } @@ -1124,6 +1118,7 @@ int dynamic_debug_register_classes(struct ddebug_known_classes_map *map) struct ddebug_table *dt; int rc = -ENOENT; + INIT_LIST_HEAD(&map->link); mutex_lock(&ddebug_lock); #ifdef CONFIG_MODULES if (map->mod) { @@ -1131,7 +1126,7 @@ int dynamic_debug_register_classes(struct ddebug_known_classes_map *map) list_for_each_entry(dt, &ddebug_tables, link) { if (dt->mod_name == map->mod->name) { rc = 0; - dt->map = map; + list_add(&map->link, &dt->maps); break; } } @@ -1142,7 +1137,7 @@ int dynamic_debug_register_classes(struct ddebug_known_classes_map *map) list_for_each_entry(dt, &ddebug_tables, link) { if (!strcmp(dt->mod_name, map->mod_name)) { rc = 0; - dt->map = map; + list_add(&map->link, &dt->maps); break; } } @@ -1159,8 +1154,38 @@ EXPORT_SYMBOL(dynamic_debug_register_classes); void dynamic_debug_unregister_classes(struct ddebug_known_classes_map *map) { - vpr_info("unregister_classes: %s\n", map->mod_name); + int rc = -ENOENT; + + mutex_lock(&ddebug_lock); +#ifdef CONFIG_MODULES + if (map->mod) { + struct ddebug_known_classes_map *dmap; + struct ddebug_table *dt; + + list_for_each_entry(dt, &ddebug_tables, link) { +
[PATCH v2 10/27] dyndbg: let query-modname override defaulting modname
dyndbg's control-parser: ddebug_parse_query(), requires that search terms: module, func, file, lineno, are not used 2x in a query; a thing cannot be named both foo and bar (not even wildcards, no OR is contemplated). Amend the treatment of module; while still enforcing the 2x rule on it, set the default module at end, if none was given in the query itself, so the preset doesn't spoil the check. Background: ddebug_parse_query() got a modname arg to support boot-args with compound queries, all with "module aMod" added by default. aMod.dyndbg="func foo +p; func bar +p" With default module at end, this becomes possible; "module *" in the query overrides the default "module aMod". aMod.dyndbg="module * class FOO +p" This means aMod can modify (by explicit means) any FOO class'd pr_debugs in other modules. It is intended for use by DRM, which generally would want sub-system wide coordination. Signed-off-by: Jim Cromie --- lib/dynamic_debug.c | 11 +++ 1 file changed, 7 insertions(+), 4 deletions(-) diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index b67bf9efec07..42dce2e76014 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -425,10 +425,6 @@ static int ddebug_parse_query(char *words[], int nwords, return -EINVAL; } - if (modname) - /* support $modname.dyndbg= */ - query->module = modname; - for (i = 0; i < nwords; i += 2) { char *keyword = words[i]; char *arg = words[i+1]; @@ -471,6 +467,13 @@ static int ddebug_parse_query(char *words[], int nwords, if (rc) return rc; } + if (!query->module && modname) + /* +* support $modname.dyndbg=, +* allowing query.module to override $modname +*/ + query->module = modname; + vpr_info_dq(query, "parsed"); return 0; } -- 2.35.3
[PATCH v2 11/27] dyndbg: support symbolic class-names in bitmap
Extend the sysfs-node bitmap support to accept class-names registered by the module, allowing: #> echo DRM_UT_CORE,-DRM_UT_ATOMIC,+DRM_UT_KMS \ > /sys/module/drm/parameters/debug Do this in param_set_dyndbg_class_strings(), which is called from param_set_dyndbg_classes() when the input string isn't a integer. Signed-off-by: Jim Cromie --- . continue/skip on unknown class, so following symbols are processed. . better loadable module handling, no more KP_MOD_NAME . add do_callback_changes --- lib/dynamic_debug.c | 63 ++--- 1 file changed, 59 insertions(+), 4 deletions(-) diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index 42dce2e76014..8e1b9159e881 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -618,6 +618,61 @@ static int do_callback_changes(const struct ddebug_classes_bitmap_param *dcp, return matches; } +/* support for [+-] symbolic-name boolean list */ +static int param_set_dyndbg_class_strings(const char *instr, const struct kernel_param *kp) +{ + const struct ddebug_classes_bitmap_param *dcp = kp->arg; + unsigned long inbits; + int idx_rc, totct = 0; + bool wanted; + char *cls, *p; + + if (!dcp || !dcp->map) { + pr_err("set_dyndbg_classes: no bits=>queries map\n"); + return -EINVAL; + } + + cls = kstrdup(instr, GFP_KERNEL); + p = strchr(cls, '\n'); + if (p) + *p = '\0'; + + inbits = *dcp->bits; + vpr_info("set_dyndbg_class_strings: %s on 0x%lx\n", cls, inbits); + + for (; cls; cls = p) { + p = strchr(cls, ','); + if (p) + *p++ = '\0'; + + if (*cls == '-') { + wanted = false; + cls++; + } else { + wanted = true; + if (*cls == '+') + cls++; + } + idx_rc = match_string(dcp->map->classes, dcp->map->length, cls); + if (idx_rc < 0) { + pr_err("%s not found for module: %s\n", cls, dcp->map->mod_name); + continue; + } + if (test_bit(idx_rc, &inbits) == wanted) { + v3pr_info("no change on %s\n", cls); + continue; + } + + vpr_info("set_dyndbg_classes: bit %d: %s\n", idx_rc, dcp->map->classes[idx_rc]); + inbits ^= BIT(idx_rc); + totct += do_callback_changes(dcp, idx_rc, &inbits); + } + kfree(cls); + *dcp->bits = inbits; + vpr_info("total matches: %d\n", totct); + return 0; +} + /** * param_set_dyndbg_classes - bits => categories >control setter * @instr: string echo>d to sysfs @@ -639,10 +694,9 @@ int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp) return -EINVAL; } rc = kstrtoul(instr, 0, &inbits); - if (rc) { - pr_err("set_dyndbg_classes: expecting bits/integer\n"); - return -EINVAL; - } + if (rc) + return param_set_dyndbg_class_strings(instr, kp); + vpr_info("set_dyndbg_classes: new 0x%lx current 0x%lx\n", inbits, *dcp->bits); for (i = 0; i < dcp->map->length; i++) { @@ -650,6 +704,7 @@ int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp) if (test_bit(i, &inbits) == test_bit(i, dcp->bits)) continue; + vpr_info("set_dyndbg_classes: bit %d: %s\n", i, dcp->map->classes[i]); totct += do_callback_changes(dcp, i, &inbits); } *dcp->bits = inbits; -- 2.35.3
[PATCH v2 08/27] dyndbg: add drm.debug style bitmap support
Add kernel_param_ops and callbacks to implement a bitmap in a sysfs-node. IE, add these: - int param_set_dyndbg_classes() - int param_get_dyndbg_classes() - struct kernel_param_ops param_ops_dyndbg_classes Following the model of kernel/params.c STANDARD_PARAM_DEFS, these are non-static and exported. get/set use an augmented kernel_param; the arg refs a new struct dyndbg_classes_param, initialized by the macro, containing: BITS: a pointer to the user module's ulong holding the bits/state. By ref'g the client's bit-state, we coordinate with existing code that uses it, so it works unchanged; for example drm_debug_enabled(). The change to ulong allows use of BIT() etc. FLAGS: dyndbg.flags toggled by bit-changes. Usually just "p". MAP: a pointer to struct ddebug_known_classes, which maps those class-names to pr_debug.class_ids 0..N. Using the sys-node, with dynamic_debug.verbose=1: bash-5.1# echo 1 > /sys/module/drm/parameters/debug [ 29.821298] dyndbg: set_dyndbg_classes: new 0x1 current 0x0 [ 29.822841] dyndbg: query 0: "class DRM_UT_CORE +p" mod:* [ 29.824348] dyndbg: no matches for query [ 29.825428] dyndbg: total matches: 0 Signed-off-by: Jim Cromie --- include/linux/dynamic_debug.h | 17 lib/dynamic_debug.c | 81 +++ 2 files changed, 98 insertions(+) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index 328722ba2d8e..e9e6d0f503f3 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -85,6 +85,11 @@ struct ddebug_known_classes_map { .classes = { __VA_ARGS__ } \ } +struct ddebug_classes_bitmap_param { + unsigned long *bits; + char flags[8]; + const struct ddebug_known_classes_map *map; +}; #if defined(CONFIG_DYNAMIC_DEBUG_CORE) @@ -237,6 +242,10 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, int dynamic_debug_register_classes(struct ddebug_known_classes_map *map); void dynamic_debug_unregister_classes(struct ddebug_known_classes_map *map); +struct kernel_param; +int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp); +int param_get_dyndbg_classes(char *buffer, const struct kernel_param *kp); + #else /* !CONFIG_DYNAMIC_DEBUG_CORE */ #include @@ -283,6 +292,14 @@ static inline int dynamic_debug_register_classes(const struct ddebug_known_class static inline void dynamic_debug_unregister_classes(struct ddebug_known_classes_map *map) {} +struct kernel_param; +static inline int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp) +{ return 0; } +static inline int param_get_dyndbg_classes(char *buffer, const struct kernel_param *kp) +{ return 0; } + #endif /* !CONFIG_DYNAMIC_DEBUG_CORE */ +extern const struct kernel_param_ops param_ops_dyndbg_classes; + #endif diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index b1aaf8893cdf..b67bf9efec07 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -597,6 +597,87 @@ static int ddebug_exec_queries(char *query, const char *modname) return nfound; } +static int do_callback_changes(const struct ddebug_classes_bitmap_param *dcp, + int bitpos, unsigned long *inbits) +{ +#define QUERY_SIZE 128 + char query[QUERY_SIZE]; + int matches; + + snprintf(query, QUERY_SIZE, "class %s %c%s", dcp->map->classes[bitpos], +test_bit(bitpos, inbits) ? '+' : '-', dcp->flags); + + matches = ddebug_exec_queries(query, NULL); + + v2pr_info("bit-%d: %d matches on class:%s\n", bitpos, + matches, dcp->map->classes[bitpos]); + + return matches; +} + +/** + * param_set_dyndbg_classes - bits => categories >control setter + * @instr: string echo>d to sysfs + * @kp:kp->arg has state: bits, map + * + * Enable/disable prdbgs by their "category", as specified in the + * arguments to DEFINE_DYNAMIC_DEBUG_CLASSES. + * + * Returns: 0 or <0 if error. + */ +int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp) +{ + const struct ddebug_classes_bitmap_param *dcp = kp->arg; + unsigned long inbits; + int rc, i, totct = 0; + + if (!dcp || !dcp->map) { + pr_err("set_dyndbg_classes: no bits=>queries map\n"); + return -EINVAL; + } + rc = kstrtoul(instr, 0, &inbits); + if (rc) { + pr_err("set_dyndbg_classes: expecting bits/integer\n"); + return -EINVAL; + } + vpr_info("set_dyndbg_classes: new 0x%lx current 0x%lx\n", inbits, *dcp->bits); + + for (i = 0; i < dcp->map->length; i++) { + + if (test_bit(i, &inbits) == test_bit(i, dcp->bits)) + continue; + + totct += do_callback_changes(dcp, i, &inbits); + } + *dcp->bits = inbits; + vpr_info("total matches: %d\n", totct); + return 0; +} +EXPORT_SYMBOL(param_se
[PATCH v2 07/27] dyndbg: validate class FOO on module
Add module-to-class validation, in #> echo class DRM_UT_KMS +p > /proc/dynamic_debug/control If a query has "class FOO", ddebug_validate_classname (called from ddebug_change) requires that FOO is known to module X, otherwize X is skipped entirely. The choice of FOO is highly selective, giving isolation and/or coordinated sharing of FOOs. For example, only DRM modules should know and respond to DRM_UT_KMS. So this, combined with module's opt-in declaration of known classes, effectively privatizes the .class_id space for each module (or coordinated set of modules). Ignoring a bad query, with dynamic_debug.verbose=3: (lots of modules will not know other module's classes) bash-5.1# echo module drm class FOO +p > /proc/dynamic_debug/control [ 54.507983] dyndbg: read 24 bytes from userspace [ 54.509549] dyndbg: query 0: "module drm class FOO +p" mod:* [ 54.511502] dyndbg: split into words: "module" "drm" "class" "FOO" "+p" [ 54.513672] dyndbg: op='+' [ 54.514575] dyndbg: flags=0x1 [ 54.515547] dyndbg: *flagsp=0x1 *maskp=0x [ 54.517112] dyndbg: parsed: func="" file="" module="drm" format="" lineno=0-0 class=FOO [ 54.519707] dyndbg: class: drm.FOO unknown [ 54.521302] dyndbg: no matches for query [ 54.522581] dyndbg: no-match: func="" file="" module="drm" format="" lineno=0-0 class=FOO [ 54.525236] dyndbg: processed 1 queries, with 0 matches, 0 errs Also add a new column to control-file output, displaying the class-name when its not default. If a module has pr_debugs with non-default .class_id's, and has not registered them, "" is issued. Signed-off-by: Jim Cromie --- . split out validate_classnames() --- lib/dynamic_debug.c | 70 +++-- 1 file changed, 62 insertions(+), 8 deletions(-) diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index 8f600c13048a..b1aaf8893cdf 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -56,6 +56,7 @@ struct ddebug_query { const char *module; const char *function; const char *format; + const char *class_string; unsigned int first_lineno, last_lineno; }; @@ -136,13 +137,40 @@ static void vpr_info_dq(const struct ddebug_query *query, const char *msg) fmtlen--; } - v3pr_info("%s: func=\"%s\" file=\"%s\" module=\"%s\" format=\"%.*s\" lineno=%u-%u\n", -msg, -query->function ?: "", -query->filename ?: "", -query->module ?: "", -fmtlen, query->format ?: "", -query->first_lineno, query->last_lineno); + v3pr_info("%s: func=\"%s\" file=\"%s\" module=\"%s\" format=\"%.*s\" lineno=%u-%u class=%s\n", + msg, + query->function ?: "", + query->filename ?: "", + query->module ?: "", + fmtlen, query->format ?: "", + query->first_lineno, query->last_lineno, query->class_string); +} + +/* return <0 if class-name is unknown/invalid, 0..CLASS_DFLT otherwise */ +static int ddebug_validate_classname(struct ddebug_table *dt, const char *class_string) +{ + int query_class = -ENOENT; + int idx; + + if (!class_string) + /* all queries w/o class given work only on default class */ + return _DPRINTK_CLASS_DFLT; + + /* +* XXX single list will need to be a for-list +* so that modules can have 2 sets of class-decls +*/ + if (!dt->map) + return -ENOENT; + + idx = match_string(dt->map->classes, dt->map->length, class_string); + if (idx < 0) { + v3pr_info("class: %s.%s unknown\n", dt->mod_name, class_string); + return -ENOENT; + } + query_class = idx + dt->map->base; + + return query_class; } /* @@ -159,6 +187,7 @@ static int ddebug_change(const struct ddebug_query *query, unsigned int newflags; unsigned int nfound = 0; struct flagsbuf fbuf, nbuf; + int query_class; /* search for matching ddebugs */ mutex_lock(&ddebug_lock); @@ -169,9 +198,18 @@ static int ddebug_change(const struct ddebug_query *query, !match_wildcard(query->module, dt->mod_name)) continue; + /* validate class-string against module's known classes */ + query_class = ddebug_validate_classname(dt, query->class_string); + if (query_class < 0) + continue; + for (i = 0; i < dt->num_ddebugs; i++) { struct _ddebug *dp = &dt->ddebugs[i]; + /* match against query-class, either valid input or default */ + if (query_class != dp->class_id) + continue; + /* match against the source filename */ if (query->filename
[PATCH v2 09/27] Doc/dyndbg: document new class class_name query support
The added paragraph is slightly process oriented, rather than in language of guarantees; I thought the implications were clear enough. It does perhaps undersell the selectivity gained with string class_names; only drm/* would sanely register DRM_UT_CORE etc, so doing multiple "module {drm*,amdgpu,i915}" queries is unnecessary. Signed-off-by: Jim Cromie --- Documentation/admin-guide/dynamic-debug-howto.rst | 11 +++ 1 file changed, 11 insertions(+) diff --git a/Documentation/admin-guide/dynamic-debug-howto.rst b/Documentation/admin-guide/dynamic-debug-howto.rst index a89cfa083155..01ca6f635dcc 100644 --- a/Documentation/admin-guide/dynamic-debug-howto.rst +++ b/Documentation/admin-guide/dynamic-debug-howto.rst @@ -35,6 +35,7 @@ Dynamic debug has even more useful features: - line number (including ranges of line numbers) - module name - format string + - class string (as known by each module) * Provides a debugfs control file: ``/dynamic_debug/control`` which can be read to display the complete list of known debug @@ -142,6 +143,7 @@ against. Possible keywords are::: 'file' string | 'module' string | 'format' string | +'class' string | 'line' line-range line-range ::= lineno | @@ -203,6 +205,15 @@ format format "nfsd: SETATTR" // a neater way to match a format with whitespace format 'nfsd: SETATTR' // yet another way to match a format with whitespace +class +The given class_name is validated against each module, which may +have registered a list of known class_names. If the class_name is +found for a module, callsite matching and adjustment proceeds. +Examples:: + + class DRM_UT_KMS# unless DRM wants different + class JUNK # silent non-match + line The given line number or range of line numbers is compared against the line number of each ``pr_debug()`` callsite. A single -- 2.35.3
[PATCH v2 05/27] dyndbg: add exclusive class_id to pr_debug callsites
DRM issues ~10 exclusive categories of debug messages; to represent this directly in dyndbg, add a new field: struct _ddebug.class_id:5. We only need 4 bits for drm, and with that reserved, we have 2 to spare on 32 bit builds; lets take one extra (5 bits total), and keep a spare bit. 32 classes-per-bitmap is a practical usability limit anyway with a bitmap interface: #> echo 0x012345678 > /sys/module/drm/parameters/debug All existing callsites are initialized with _DPRINTK_CLASS_DFLT, which is 2^5-1. This reserves 0-30 for use in new categorized/class'd pr_debugs, which fits perfectly with natural enums (ints: 0..N). To achieve this, DEFINE_DYNAMIC_DEBUG_METADATA_CLS(cls,...) is added, and DEFINE_DYNAMIC_DEBUG_METADATA is altered to call it with the default. The factory macro chain between there and pr_debug is adjusted similarly. No behavior change. Signed-off-by: Jim Cromie --- include/linux/dynamic_debug.h | 54 ++- 1 file changed, 41 insertions(+), 13 deletions(-) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index dd20e825c36f..39550fefcf0f 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -6,6 +6,8 @@ #include #endif +#include + /* * An instance of this structure is created in a special * ELF section at every dynamic debug callsite. At runtime, @@ -21,6 +23,9 @@ struct _ddebug { const char *filename; const char *format; unsigned int lineno:18; +#define CLS_BITS 5 + unsigned int class_id:CLS_BITS; +#define _DPRINTK_CLASS_DFLT((1 << CLS_BITS) - 1) /* * The flags field controls the behaviour at the callsite. * The bits here are changed dynamically when the user @@ -84,7 +89,7 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, const struct ib_device *ibdev, const char *fmt, ...); -#define DEFINE_DYNAMIC_DEBUG_METADATA(name, fmt) \ +#define DEFINE_DYNAMIC_DEBUG_METADATA_CLS(name, cls, fmt) \ static struct _ddebug __aligned(8) \ __section("__dyndbg") name = { \ .modname = KBUILD_MODNAME, \ @@ -93,8 +98,14 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, .format = (fmt),\ .lineno = __LINE__, \ .flags = _DPRINTK_FLAGS_DEFAULT,\ + .class_id = cls,\ _DPRINTK_KEY_INIT \ - } + }; \ + BUILD_BUG_ON_MSG(cls > _DPRINTK_CLASS_DFLT, \ +"classid value overflow") + +#define DEFINE_DYNAMIC_DEBUG_METADATA(name, fmt) \ + DEFINE_DYNAMIC_DEBUG_METADATA_CLS(name, _DPRINTK_CLASS_DFLT, fmt) #ifdef CONFIG_JUMP_LABEL @@ -125,18 +136,26 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, #endif /* CONFIG_JUMP_LABEL */ -#define __dynamic_func_call(id, fmt, func, ...) do { \ - DEFINE_DYNAMIC_DEBUG_METADATA(id, fmt); \ - if (DYNAMIC_DEBUG_BRANCH(id)) \ - func(&id, ##__VA_ARGS__); \ +#define __dynamic_func_call_cls(id, cls, fmt, func, ...) do { \ + DEFINE_DYNAMIC_DEBUG_METADATA_CLS(id, cls, fmt);\ + if (DYNAMIC_DEBUG_BRANCH(id)) \ + func(&id, ##__VA_ARGS__); \ } while (0) -#define __dynamic_func_call_no_desc(id, fmt, func, ...) do { \ - DEFINE_DYNAMIC_DEBUG_METADATA(id, fmt); \ - if (DYNAMIC_DEBUG_BRANCH(id)) \ - func(__VA_ARGS__); \ +#define __dynamic_func_call_no_desc_cls(id, cls, fmt, func, ...) do { \ + DEFINE_DYNAMIC_DEBUG_METADATA_CLS(id, cls, fmt);\ + if (DYNAMIC_DEBUG_BRANCH(id)) \ + func(__VA_ARGS__); \ } while (0) +#define __dynamic_func_call(id, fmt, func, ...) \ + __dynamic_func_call_cls(id, _DPRINTK_CLASS_DFLT,\ + fmt, func, ##__VA_ARGS__) + +#define __dynamic_func_call_no_desc(id, fmt, func, ...) \ + __dynamic_func_call_no_desc_cls(id, _DPRINTK_CLASS_DFLT,\ + fmt, func, ##__VA_ARGS__) + /* * "Factory macro" for generating a call to func, guarded by a * DYNAMIC_DEBUG_BRANCH. The dynamic debug descriptor will be @@ -145,15 +164,24 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, * the varargs. Note that fmt is repeated in invocations of this * macro. */ +#define _dynamic_fu
[PATCH v2 06/27] dyndbg: add dynamic_debug_(un)register_classes
Add dynamic_debug_register_classes() to API, allowing user modules (builtin and loadable) to register class_names for the .class_id's they are using. Knowing classes is 1st step to validating with them. Add dynamic_debug_unregister_classes() also. Add struct ddebug_known_classes_map: maps known class_name strings to .class_id[0..N], where N<31. .base allows sharing of that range. A wrapper macro: DYNAMIC_DEBUG_CLASSES(_var, _modmatch, _base, classes) defines and initializes that struct var, _maybe_unused is added so the decl doesn't draw warnings when dyndbg is not enabled; a small list of classnames is tolerable wasted space. _var: user passes this into dynamic_debug_register_classes(var). _base: usually 0, it allows splitting 31 classes into subranges, so that multiple sysfs-nodes can share the module's class-id space. classes: list of strings with classnames, mapped to class-id=idx(+_base) mod_name: KBUILD_MODNAME, available for builtins, loadables mod: ref to loadable module. allows ==, distinguishing loadables. When modules register known classnames, they opt-in to permit dyndbg to allow "class " queries to manipulate their class'd pr_debugs (if any), default class_id pr_debugs are still controllable as before. dynamic_debug_register_classes(&map) finds the module's ddebug_table record, and attaches the map to it. This makes it available to ddebug_change(), which will use it to validate class'd commands. Sharing class-names across multiple modules is how those modules coordinate; all drm* and drivers would respond to: #> echo class DRM_UT_CORE +p > /proc/dynamic_debug/control Therefore no class-name uniqueness check is useful. TODO: Eventually we need a list of registered classes, not just the zero-or-one implemented here. This will support multiple sysfs-nodes, one each for print-to-syslog, print-to-tracefs, or subranges using _base. Signed-off-by: Jim Cromie --- . fix register_classes() for !CONFIG_MODULES . add maybe-unused to var decl in DYNAMIC_DEBUG_CLASSES . kdoc for DYNAMIC_DEBUG_CLASSES . add dynamic_debug_unregister_classes --- include/linux/dynamic_debug.h | 36 lib/dynamic_debug.c | 53 +++ 2 files changed, 89 insertions(+) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index 39550fefcf0f..328722ba2d8e 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -56,6 +56,34 @@ struct _ddebug { #endif } __attribute__((aligned(8))); +struct ddebug_known_classes_map { + struct list_head link; + struct module *mod; + const char *mod_name; /* needed for builtins */ + const int base; /* index of 1st .class_id, allows split/shared space */ + const int length; + const char *classes[]; /* index maps .class_id */ +}; + +#define NUM_TYPE_ARGS(eltype, ...) \ + (sizeof((eltype[]) {__VA_ARGS__}) / sizeof(eltype)) +/** + * DYNAMIC_DEBUG_CLASSES - declare classnames known by a module + * @_var: passed to dynamic_debug_register_classes(_var) + * @_base: offset of 1st class-name. splits .class_id space + * @classes: class-names known/used by of .class_ids[_base.._base+length] + * + * @classes specifies names for the classids used by a module; dyndbg + * accepts "class " commands if is known and registered. + */ +#define DYNAMIC_DEBUG_CLASSES(_var, _base, ...) \ + static __maybe_unused struct ddebug_known_classes_map _var = { \ + .mod = THIS_MODULE, \ + .mod_name = KBUILD_MODNAME, \ + .base = _base, \ + .length = NUM_TYPE_ARGS(char*, __VA_ARGS__),\ + .classes = { __VA_ARGS__ } \ + } #if defined(CONFIG_DYNAMIC_DEBUG_CORE) @@ -206,6 +234,9 @@ void __dynamic_ibdev_dbg(struct _ddebug *descriptor, KERN_DEBUG, prefix_str, prefix_type, \ rowsize, groupsize, buf, len, ascii) +int dynamic_debug_register_classes(struct ddebug_known_classes_map *map); +void dynamic_debug_unregister_classes(struct ddebug_known_classes_map *map); + #else /* !CONFIG_DYNAMIC_DEBUG_CORE */ #include @@ -247,6 +278,11 @@ static inline int ddebug_dyndbg_module_param_cb(char *param, char *val, } while (0) +static inline int dynamic_debug_register_classes(const struct ddebug_known_classes_map *map) +{ return 0; } +static inline void dynamic_debug_unregister_classes(struct ddebug_known_classes_map *map) +{} + #endif /* !CONFIG_DYNAMIC_DEBUG_CORE */ #endif diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index cdc0b03b1148..8f600c13048a 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -45,6 +45,8 @@ extern struct _ddebug __stop___dyndbg[]; struct ddebug_table { str
[PATCH v2 03/27] dyndbg: fix module.dyndbg handling
For CONFIG_DYNAMIC_DEBUG=N, the ddebug_dyndbg_module_param_cb() stub-fn is too permissive: bash-5.1# modprobe drm JUNKdyndbg bash-5.1# modprobe drm dyndbgJUNK [ 42.933220] dyndbg param is supported only in CONFIG_DYNAMIC_DEBUG builds [ 42.937484] ACPI: bus type drm_connector registered This caused no ill effects, because unknown parameters are either ignored by default (with an "unknown parameter" warning, see below), or ignored because dyndbg allows its no-effect use on non-dyndbg builds. That said, the code has an explicit feedback message, which should be issued accurately. Fix with strcmp, for exact param-name match. Here is fixed behavior: bash-5.1# modprobe drm dyndbgJUNK [ 20.127473] drm: unknown parameter 'dyndbgJUNK' ignored [ 20.128835] ACPI: bus type drm_connector registered bash-5.1# rmmod drm [ 37.961656] ACPI: bus type drm_connector unregistered bash-5.1# modprobe drm dyndbg [ 42.933220] dyndbg param is supported only in CONFIG_DYNAMIC_DEBUG builds [ 42.937484] ACPI: bus type drm_connector registered Reported-by: Rasmus Villemoes Signed-off-by: Jim Cromie --- include/linux/dynamic_debug.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index dce631e678dd..f30b01aa9fa4 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -201,7 +201,7 @@ static inline int ddebug_remove_module(const char *mod) static inline int ddebug_dyndbg_module_param_cb(char *param, char *val, const char *modname) { - if (strstr(param, "dyndbg")) { + if (!strcmp(param, "dyndbg")) { /* avoid pr_warn(), which wants pr_fmt() fully defined */ printk(KERN_WARNING "dyndbg param is supported only in " "CONFIG_DYNAMIC_DEBUG builds\n"); -- 2.35.3
[PATCH v2 04/27] dyndbg: drop EXPORTed dynamic_debug_exec_queries
This exported fn is unused, and is effectively obsoleted by a forthcoming commit, so remove it. The export was added to let drm control pr_debugs, as part of using them to avoid drm_debug_enabled overheads. But following patches implement the drm.debug interface, and adapt drm to just use it, so nobody will never need the export. This also drops the CONFIG_DYNAMIC_DEBUG=N stub-func, and its pr_warn(), which I avoided in 2012, then added in 2020 :-/ Fixes: a2d375eda771 ("dyndbg: refine export, rename to dynamic_debug_exec_queries()") Fixes: 4c0d77828d4f ("dyndbg: export ddebug_exec_queries") Signed-off-by: Jim Cromie --- include/linux/dynamic_debug.h | 8 lib/dynamic_debug.c | 29 - 2 files changed, 37 deletions(-) diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h index f30b01aa9fa4..dd20e825c36f 100644 --- a/include/linux/dynamic_debug.h +++ b/include/linux/dynamic_debug.h @@ -55,9 +55,6 @@ struct _ddebug { #if defined(CONFIG_DYNAMIC_DEBUG_CORE) -/* exported for module authors to exercise >control */ -int dynamic_debug_exec_queries(const char *query, const char *modname); - int ddebug_add_module(struct _ddebug *tab, unsigned int n, const char *modname); extern int ddebug_remove_module(const char *mod_name); @@ -221,11 +218,6 @@ static inline int ddebug_dyndbg_module_param_cb(char *param, char *val, rowsize, groupsize, buf, len, ascii); \ } while (0) -static inline int dynamic_debug_exec_queries(const char *query, const char *modname) -{ - pr_warn("kernel not built with CONFIG_DYNAMIC_DEBUG_CORE\n"); - return 0; -} #endif /* !CONFIG_DYNAMIC_DEBUG_CORE */ diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index ca2a28f1d51c..cdc0b03b1148 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -555,35 +555,6 @@ static int ddebug_exec_queries(char *query, const char *modname) return nfound; } -/** - * dynamic_debug_exec_queries - select and change dynamic-debug prints - * @query: query-string described in admin-guide/dynamic-debug-howto - * @modname: string containing module name, usually &module.mod_name - * - * This uses the >/proc/dynamic_debug/control reader, allowing module - * authors to modify their dynamic-debug callsites. The modname is - * canonically struct module.mod_name, but can also be null or a - * module-wildcard, for example: "drm*". - */ -int dynamic_debug_exec_queries(const char *query, const char *modname) -{ - int rc; - char *qry; /* writable copy of query */ - - if (!query) { - pr_err("non-null query/command string expected\n"); - return -EINVAL; - } - qry = kstrndup(query, PAGE_SIZE, GFP_KERNEL); - if (!qry) - return -ENOMEM; - - rc = ddebug_exec_queries(qry, modname); - kfree(qry); - return rc; -} -EXPORT_SYMBOL_GPL(dynamic_debug_exec_queries); - #define PREFIX_SIZE 64 static int remaining(int wrote) -- 2.35.3
[PATCH v2 02/27] dyndbg: show both old and new in change-info
print "old -> new" flag values in the info("change") message. no functional change. Signed-off-by: Jim Cromie --- lib/dynamic_debug.c | 11 ++- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index a56c1286ffa4..ca2a28f1d51c 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -156,7 +156,7 @@ static int ddebug_change(const struct ddebug_query *query, struct ddebug_table *dt; unsigned int newflags; unsigned int nfound = 0; - struct flagsbuf fbuf; + struct flagsbuf fbuf, nbuf; /* search for matching ddebugs */ mutex_lock(&ddebug_lock); @@ -217,11 +217,12 @@ static int ddebug_change(const struct ddebug_query *query, static_branch_enable(&dp->key.dd_key_true); } #endif + v4pr_info("changed %s:%d [%s]%s %s -> %s\n", + trim_prefix(dp->filename), dp->lineno, + dt->mod_name, dp->function, + ddebug_describe_flags(dp->flags, &fbuf), + ddebug_describe_flags(newflags, &nbuf)); dp->flags = newflags; - v4pr_info("changed %s:%d [%s]%s =%s\n", -trim_prefix(dp->filename), dp->lineno, -dt->mod_name, dp->function, -ddebug_describe_flags(dp->flags, &fbuf)); } } mutex_unlock(&ddebug_lock); -- 2.35.3
[PATCH v2 01/27] dyndbg: fix static_branch manipulation
In https://lore.kernel.org/lkml/20211209150910.ga23...@axis.com/ Vincent's patch commented on, and worked around, a bug toggling static_branch's, when a 2nd PRINTK-ish flag was added. The bug results in a premature static_branch_disable when the 1st of 2 flags was disabled. The cited commit computed newflags, but then in the JUMP_LABEL block, failed to use that result, instead using just one of the terms in it. Using newflags instead made the code work properly. This is Vincents test-case, reduced. It needs the 2nd flag to demonstrate the bug, but it's explanatory here. pt_test() { echo 5 > /sys/module/dynamic_debug/verbose site="module tcp" # just one callsite echo " $site =_ " > /proc/dynamic_debug/control # clear it # A B ~A ~B for flg in +T +p "-T #broke here" -p; do echo " $site $flg " > /proc/dynamic_debug/control done; # A B ~B ~A for flg in +T +p "-p #broke here" -T; do echo " $site $flg " > /proc/dynamic_debug/control done } pt_test Fixes: 84da83a6ffc0 dyndbg: combine flags & mask into a struct, simplify with it CC: vincent.whitchu...@axis.com Signed-off-by: Jim Cromie Acked-by: Jason Baron --- .drop @stable, no exposed bug. .add jbaron ack --- lib/dynamic_debug.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c index dd7f56af9aed..a56c1286ffa4 100644 --- a/lib/dynamic_debug.c +++ b/lib/dynamic_debug.c @@ -211,10 +211,11 @@ static int ddebug_change(const struct ddebug_query *query, continue; #ifdef CONFIG_JUMP_LABEL if (dp->flags & _DPRINTK_FLAGS_PRINT) { - if (!(modifiers->flags & _DPRINTK_FLAGS_PRINT)) + if (!(newflags & _DPRINTK_FLAGS_PRINT)) static_branch_disable(&dp->key.dd_key_true); - } else if (modifiers->flags & _DPRINTK_FLAGS_PRINT) + } else if (newflags & _DPRINTK_FLAGS_PRINT) { static_branch_enable(&dp->key.dd_key_true); + } #endif dp->flags = newflags; v4pr_info("changed %s:%d [%s]%s =%s\n", -- 2.35.3
[RFC PATCH v2 00/27] DRM.debug on DYNAMIC_DEBUG, add trace events
DRM.debug API is 23 macros, issuing 10 exclusive categories of debug messages. By rough count, they are used 5140 times in the kernel. These all call drm_dbg or drm_devdbg, which call drm_debug_enabled(), which checks bits in global __drm_debug. Some of these are page-flips and vblanks, and get called often. DYNAMIC_DEBUG (with CONFIG_JUMP_LABEL) is built to avoid this kind of work, with NOOPd jump/callsites. This patchset is RFC because: - it touches 2.5 subsystems: dyndbg, drm, tracefs (new events) - dyndbg class support is built for drm, needs it for validation - new api, used by drm - big memory impact, with 5100 new pr-debug callsites. - drm class bikeshedding opportunities - others, names etc. DYNAMIC_DEBUG: Adapt to directly represent 32 exclusive classes / categories. All existing users get _CLASS_DFLT:31, reserving 0-30 for opt-in. dynamic_debug_register_classes(): Tells dyndbg that caller/client is using .class_id's 0..N, and wants them exposed and manipulatable with names: DRM_UT_KMS etc. Client invokes DYNAMIC_DEBUG_CLASSES to define the name->id map, then registers it during its initialization. Then, when a dyndbg command-query has "class FOO": echo class DRM_UT_KMS +p > /proc/dynamic_debug/control ddebug_change() will 1st ddebug_validate_classname() for each module, using its registered class-maps; only modules which know DRM_UT_KMS will get those callsite adjustments. This "protects" each module's class-id space, unlike "class 4" query syntax. Add bitmap support for: echo 0x02 > /sys/module/drm/parameters/debug Add symbolic bitmap, because we have the class-names: echo +DRM_UT_KMS,+DRM_UT_ATOMIC > /sys/module/drm/parameters/debug Add test_dynamic_debug, to prove out the API. which is (used 3x to prove class-id ranges): - DYNAMIC_DEBUG_CLASSES(var, ... [3 class-name-literals each]) - dynamic_debug_register_classes(var) - dynamic_debug_unregister_classes(var) also (6x): +module_param_cb(, ¶m_ops_dyndbg_classes, /[pt][123]_bitmap/, 0600); Bits 0,1,2 of each of the 6 bitmaps control the 3 named classes in each of 3 groups, toggling their p,T bits respectively. RFC: dynamic_debug_register_classes() cannot act early enough to be in effect at module-load. So this will not work as you'd reasonably expect: modprobe test_dynamic_debug dyndbg='+pfm; class FOO +pfmlt' The 1st query:+pfm will be enabled during load, but in the 2nd query, "class FOO" will be unknown at load time. Early class enablement would be nice. DYNAMIC_DEBUG_CLASSES is a static initializer, which is certainly early enough, but Im missing a trick, suggestions? Wildcarding on classname is possible, maybe useful: echo +DRM_UT_\* > /sys/module/drm/parameters/debug If contemplated in earnest, it should consider the difference between "DRM_*" and "*_KMS", wrt observing across class hierarchies, as well as decide whether that echo means 1st match, or all matching names. __pr_debug_cls(cls, fmt,...) is just for testing, drm doesnt need it. DRM adaptation: Basically, these patches add another layer of macro indirection, which by CONFIG_DRM_USE_DYNAMIC_DEBUG=y, wraps drm_*dbg() in dyndbg's callsite Factory macro, and adds the new descriptor arg to the funcs. With these patches, drm.debugs appear as controllable *pr-dbg* callsites: [ 11.804352] dyndbg: register_classes: drm [ 11.920715] dyndbg: 234 debug prints in module drm_kms_helper [ 11.932135] dyndbg: 2 debug prints in module ttm [ 11.936119] dyndbg: 8 debug prints in module video [ 12.537543] dyndbg: 1756 debug prints in module i915 [ 12.555045] dyndbg: register_classes: i915 [ 12.666727] AMD-Vi: AMD IOMMUv2 functionality not available on this .. [ 13.735310] dyndbg: 3859 debug prints in module amdgpu amdgpu's large count includes: 582 direct uses/mentions of pr_debug 43 defns of DC_LOG_*, which use either pr_debug or DRM_DEBUG_* ~1000 uses of DC_LOG_* 1116 uses of dml_print in drivers/gpu/drm/amd 15 drm_dbg_\\w+ drivers/gpu/drm/amd 273 DRM_DEBUG_\\w+ drivers/gpu/drm/amd i915 has: 1072 drm_dbg_\\w+ drivers/gpu/drm/i915/ | wc -l 200 DRM_DEBUG_\\w+ drivers/gpu/drm/i915/ 46 pr_debug drivers/gpu/drm/i915/ 144 gvt_dbg_\\w+ drivers/gpu/drm/i915/gvt, using pr_debug This impacts the data footprint, so everything new/drm is dependent on DRM_USE_DYNAMIC_DEBUG=y. RFC for DRM: - decoration flags "fmlt" do not work on drm_*dbg(). (drm_*dbg() dont use pr_debug, they *become* one flavor of them) this could (should?) be added, and maybe tailored for drm. some of the device prefixes are very long, a "d" flag could optionalize them. - api use needs review wrt drm life-cycle. enum drm_debug_category and DYNAMIC_DEBUG_CLASSES could be together? - class-names could stand review, perhaps extension "drm:core:" etc have appeared (maybe just from me) or a "plan" to look at it later - i915 & amdgpu have pr_debugs (DC_LOG_*, gvt_dbg_*) that have class-ish prefixes that are separ
[PATCH] drm/amdgpu: Unmap legacy queue when MES is enabled
This fixes a kernel oops when MES is not enabled. Reported-by: Kenny Ho Suggested-by: Jack Xiao Reviewed-by: Alex Deucher Signed-off-by: Luben Tuikov Fixes: d69e773ca75b8f ("drm/amdgpu: add mes unmap legacy queue routine") Fixes: 36d7a7094750fc ("drm/amdgpu: add init support for GFX11 (v2)") --- drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c index 4b66b9c93754ba..65a4126135b0e2 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c @@ -3568,7 +3568,7 @@ static void gfx10_kiq_unmap_queues(struct amdgpu_ring *kiq_ring, struct amdgpu_device *adev = kiq_ring->adev; uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; - if (!adev->gfx.kiq.ring.sched.ready) { + if (adev->enable_mes && !adev->gfx.kiq.ring.sched.ready) { amdgpu_mes_unmap_legacy_queue(adev, ring, action, gpu_addr, seq); return; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index 8a1bec70c71932..8773cbd1f03b54 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -191,7 +191,7 @@ static void gfx11_kiq_unmap_queues(struct amdgpu_ring *kiq_ring, struct amdgpu_device *adev = kiq_ring->adev; uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; - if (!adev->gfx.kiq.ring.sched.ready) { + if (adev->enable_mes && !adev->gfx.kiq.ring.sched.ready) { amdgpu_mes_unmap_legacy_queue(adev, ring, action, gpu_addr, seq); return; } base-commit: 2092fd3ff7e91c9f408c98799f4cdcaa93a06c3d -- 2.36.1.74.g277cf0bc36
Re: [PATCH] drm/amdkfd: simplify cpu hive assignment
Am 2022-05-16 um 10:43 schrieb Jonathan Kim: CPU hive assignment currently assumes when a GPU hive is connected_to_cpu, there is only one hive in the system. Only assign CPUs to the hive if they are explicitly directly connected to the GPU hive to get rid of the need for this assumption. It's more efficient to do this when querying IO links since other non-CRAT info has to be filled in anyways. Also, stop re-assigning the same CPU to the same GPU hive if it has already been done before. Signed-off-by: Jonathan Kim Reviewed-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 22 ++ 1 file changed, 6 insertions(+), 16 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 2e20f54bb147..8d50d207cf66 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -1271,6 +1271,12 @@ static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev) if (!peer_dev) continue; + /* Include the CPU peer in GPU hive if connected over xGMI. */ + if (!peer_dev->gpu && !peer_dev->node_props.hive_id && + dev->node_props.hive_id && + dev->gpu->adev->gmc.xgmi.connected_to_cpu) + peer_dev->node_props.hive_id = dev->node_props.hive_id; + list_for_each_entry(inbound_link, &peer_dev->io_link_props, list) { if (inbound_link->node_to != link->node_from) @@ -1302,22 +1308,6 @@ int kfd_topology_add_device(struct kfd_dev *gpu) pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id); - /* Include the CPU in xGMI hive if xGMI connected by assigning it the hive ID. */ - if (gpu->hive_id && gpu->adev->gmc.xgmi.connected_to_cpu) { - struct kfd_topology_device *top_dev; - - down_read(&topology_lock); - - list_for_each_entry(top_dev, &topology_device_list, list) { - if (top_dev->gpu) - break; - - top_dev->node_props.hive_id = gpu->hive_id; - } - - up_read(&topology_lock); - } - /* Check to see if this gpu device exists in the topology_device_list. * If so, assign the gpu to that device, * else create a Virtual CRAT for this gpu device and then parse that
Re: VCN_INFO_TABLE_MAX_NUM_INSTANCES vs AMDGPU_MAX_VCN_INSTANCES
Am 16.05.22 um 20:15 schrieb Alex Deucher: On Mon, May 16, 2022 at 2:10 PM Christian König wrote: Am 16.05.22 um 19:49 schrieb Ernst Sjöstrand: Den mån 16 maj 2022 kl 17:13 skrev Alex Deucher : On Sun, May 15, 2022 at 11:46 AM Ernst Sjöstrand wrote: smatch found this problem on amd-staging-drm-next: drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1443 amdgpu_discovery_get_vcn_info() error: buffer overflow 'adev->vcn.vcn_codec_disable_mask' 2 <= 3 This is caused by: #define AMDGPU_MAX_VCN_INSTANCES 2 #define VCN_INFO_TABLE_MAX_NUM_INSTANCES 4 Can we just drop VCN_INFO_TABLE_MAX_NUM_INSTANCES completely and use AMDGPU_MAX_VCN_INSTANCES everywhere instead (and bump it to 4)? We should be able to bump AMDGPU_MAX_VCN_INSTANCES to 4 (although it would waste some memory in the places it is used at this point). VCN_INFO_TABLE_MAX_NUM_INSTANCES is part of a firmware structure so we can't change that without breaking the firmware structure. Alex It would be nice to get rid of this pattern and make sure it doesn't happen again when the VCN info table is raised to 5. It's very similar to the HWIP_MAX_INSTANCE issue. No, as Alex explained that distinction is intentional. The firmware definition is 4 for future extensions, that doesn't mean that this is currently used. There is currently simply no need to set AMDGPU_MAX_VCN_INSTANCES to more than 2. Right. The attached patch should protect against the scenario you are envisioning. Acked-by: Christian König Alex Regards, Christian. //E
[PATCH 1/2] drm/amdkfd: port cwsr trap handler from dkms branch
It is to simplify trap handler support for new asics in the future. Signed-off-by: Eric Huang --- .../gpu/drm/amd/amdkfd/cwsr_trap_handler.h| 2527 + .../amd/amdkfd/cwsr_trap_handler_gfx10.asm| 325 ++- .../drm/amd/amdkfd/cwsr_trap_handler_gfx9.asm | 244 +- 3 files changed, 1596 insertions(+), 1500 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h index 475f89700c74..8cbdc7f519c6 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h @@ -166,7 +166,7 @@ static const uint32_t cwsr_trap_gfx8_hex[] = { 0x807c847c, 0x806eff6e, 0x0400, 0xbf0a757c, 0xbf85ffef, 0xbf9c, - 0xbf8200cd, 0xbef8007e, + 0xbf8200ce, 0xbef8007e, 0x8679ff7f, 0x, 0x8779ff79, 0x0004, 0xbefa0080, 0xbefb00ff, @@ -212,304 +212,310 @@ static const uint32_t cwsr_trap_gfx8_hex[] = { 0x761e, 0xe0524100, 0x761e0100, 0xe0524200, 0x761e0200, 0xe0524300, - 0x761e0300, 0xb8f22a05, - 0x80728172, 0x8e728a72, - 0xb8f61605, 0x80768176, - 0x8e768676, 0x80727672, - 0x80f2c072, 0xb8f31605, - 0x80738173, 0x8e738473, - 0x8e7a8273, 0xbefa00ff, - 0x0100, 0xbefc0073, - 0xc031003c, 0x0072, - 0x80f2c072, 0xbf8c007f, - 0x80fc907c, 0xbe802d00, - 0xbe822d02, 0xbe842d04, - 0xbe862d06, 0xbe882d08, - 0xbe8a2d0a, 0xbe8c2d0c, - 0xbe8e2d0e, 0xbf06807c, - 0xbf84fff1, 0xb8f22a05, - 0x80728172, 0x8e728a72, - 0xb8f61605, 0x80768176, - 0x8e768676, 0x80727672, - 0xbefa0084, 0xbefa00ff, - 0x0100, 0xc0211cfc, + 0x761e0300, 0xbf8c0f70, + 0xb8f22a05, 0x80728172, + 0x8e728a72, 0xb8f61605, + 0x80768176, 0x8e768676, + 0x80727672, 0x80f2c072, + 0xb8f31605, 0x80738173, + 0x8e738473, 0x8e7a8273, + 0xbefa00ff, 0x0100, + 0xbefc0073, 0xc031003c, + 0x0072, 0x80f2c072, + 0xbf8c007f, 0x80fc907c, + 0xbe802d00, 0xbe822d02, + 0xbe842d04, 0xbe862d06, + 0xbe882d08, 0xbe8a2d0a, + 0xbe8c2d0c, 0xbe8e2d0e, + 0xbf06807c, 0xbf84fff1, + 0xb8f22a05, 0x80728172, + 0x8e728a72, 0xb8f61605, + 0x80768176, 0x8e768676, + 0x80727672, 0xbefa0084, + 0xbefa00ff, 0x0100, + 0xc0211cfc, 0x0072, + 0x80728472, 0xc0211c3c, 0x0072, 0x80728472, - 0xc0211c3c, 0x0072, - 0x80728472, 0xc0211c7c, + 0xc0211c7c, 0x0072, + 0x80728472, 0xc0211bbc, 0x0072, 0x80728472, - 0xc0211bbc, 0x0072, - 0x80728472, 0xc0211bfc, + 0xc0211bfc, 0x0072, + 0x80728472, 0xc0211d3c, 0x0072, 0x80728472, - 0xc0211d3c, 0x0072, - 0x80728472, 0xc0211d7c, + 0xc0211d7c, 0x0072, + 0x80728472, 0xc0211a3c, 0x0072, 0x80728472, - 0xc0211a3c, 0x0072, - 0x80728472, 0xc0211a7c, + 0xc0211a7c, 0x0072, + 0x80728472, 0xc0211dfc, 0x0072, 0x80728472, - 0xc0211dfc, 0x0072, - 0x80728472, 0xc0211b3c, + 0xc0211b3c, 0x0072, + 0x80728472, 0xc0211b7c, 0x0072, 0x80728472, - 0xc0211b7c, 0x0072, - 0x80728472, 0xbf8c007f, - 0xbefc0073, 0xbefe006e, - 0xbeff006f, 0x867375ff, - 0x03ff, 0xb9734803, - 0x867375ff, 0xf800, - 0x8f738b73, 0xb973a2c3, - 0xb977f801, 0x8673ff71, - 0xf000, 0x8f739c73, - 0x8e739073, 0xbef60080, - 0x87767376, 0x8673ff71, - 0x0800, 0x8f739b73, - 0x8e738f73, 0x87767376, - 0x8673ff74, 0x0080, - 0x8f739773, 0xb976f807, - 0x8671ff71, 0x, - 0x86fe7e7e, 0x86ea6a6a, - 0x8f768374, 0xb976e0c2, - 0xbf82, 0xb9740002, - 0xbf8a, 0x95807370, - 0xbf81, 0x, + 0xbf8c007f, 0xbefc0073, + 0xbefe006e, 0xbeff006f, + 0x867375ff, 0x03ff, + 0xb9734803, 0x867375ff, + 0xf800, 0x8f738b73, + 0xb973a2c3, 0xb977f801, + 0x8673ff71, 0xf000, + 0x8f739c73, 0x8e739073, + 0xbef60080, 0x87767376, + 0x8673ff71, 0x0800, + 0x8f739b73, 0x8e738f73, + 0x87767376, 0x8673ff74, + 0x0080, 0x8f739773, + 0xb976f807, 0x8671ff71, + 0x, 0x86fe7e7e, + 0x86ea6a6a, 0x8f768374, + 0xb976e0c2, 0xbf82, + 0xb9740002, 0xbf8a, + 0x95807370, 0xbf81, }; static const uint32_t cwsr_trap_gfx9_hex[] = { - 0xbf820001, 0xbf820248, - 0xb8f8f802, 0x89788678, - 0xb8eef801, 0x866eff6e, - 0x0800, 0xbf840003, + 0xbf820001, 0xbf820254, + 0xb8f8f802, 0x8978ff78, + 0x00020006, 0xb8fbf803, 0x866eff78, 0x2000, - 0xbf840016, 0xb8fbf803, + 0xbf840009, 0x866eff6d, + 0x00f
[PATCH 2/2] drm/amdkfd: Add gfx11 trap handler
From: Jay Cornwall Based on gfx10 with following changes: - GPR_ALLOC.VGPR_SIZE field moved (and size corrected in gfx10) - s_sendmsg_rtn_b64 replaces some s_sendmsg/s_getreg - Buffer instructions no longer have direct-to-LDS modifier Signed-off-by: Jay Cornwall Reviewed-by: Laurent Morichetti --- .../gpu/drm/amd/amdkfd/cwsr_trap_handler.h| 463 +- .../amd/amdkfd/cwsr_trap_handler_gfx10.asm| 69 ++- drivers/gpu/drm/amd/amdkfd/kfd_device.c | 6 +- 3 files changed, 507 insertions(+), 31 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h index 8cbdc7f519c6..60a81649cf12 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h @@ -776,7 +776,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0xe0704100, 0x705d0100, 0xe0704200, 0x705d0200, 0xe0704300, 0x705d0300, - 0xb9702a05, 0x80708170, + 0xb9703a05, 0x80708170, 0xbf0d9973, 0xbf850002, 0x8f708970, 0xbf820001, 0x8f708a70, 0xb97a1e06, @@ -855,7 +855,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0x877aff6d, 0x8000, 0xbf840040, 0x8f7b867b, 0x8f7b827b, 0xbef6037b, - 0xb9702a05, 0x80708170, + 0xb9703a05, 0x80708170, 0xbf0d9973, 0xbf850002, 0x8f708970, 0xbf820001, 0x8f708a70, 0xb97a1e06, @@ -891,7 +891,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0xbef003ff, 0x0200, 0xbeff0380, 0xbf820003, 0xbef003ff, 0x0400, - 0xbeff03c1, 0xb97b2a05, + 0xbeff03c1, 0xb97b3a05, 0x807b817b, 0x8f7b827b, 0x907c9973, 0x877c817c, 0xbf06817c, 0xbf850017, @@ -939,7 +939,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0xb96f4306, 0x876fc16f, 0xbf840029, 0x8f6f866f, 0x8f6f826f, 0xbef6036f, - 0xb9782a05, 0x80788178, + 0xb9783a05, 0x80788178, 0xbf0d9972, 0xbf850002, 0x8f788978, 0xbf820001, 0x8f788a78, 0xb96e1e06, @@ -962,7 +962,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0x907c9972, 0x877c817c, 0xbf06817c, 0xbf850002, 0xbeff0380, 0xbf820001, - 0xbeff03c1, 0xb96f2a05, + 0xbeff03c1, 0xb96f3a05, 0x806f816f, 0x8f6f826f, 0x907c9972, 0x877c817c, 0xbf06817c, 0xbf850024, @@ -1010,7 +1010,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0x6e5d0100, 0xe0304200, 0x6e5d0200, 0xe0304300, 0x6e5d0300, 0xbf8c3f70, - 0xb9782a05, 0x80788178, + 0xb9783a05, 0x80788178, 0xbf0d9972, 0xbf850002, 0x8f788978, 0xbf820001, 0x8f788a78, 0xb96e1e06, @@ -1037,7 +1037,7 @@ static const uint32_t cwsr_trap_nv1x_hex[] = { 0xbe8c310c, 0xbe8e310e, 0xbf06807c, 0xbf84fff0, 0xba80f801, 0x, - 0xbf8a, 0xb9782a05, + 0xbf8a, 0xb9783a05, 0x80788178, 0xbf0d9972, 0xbf850002, 0x8f788978, 0xbf820001, 0x8f788a78, @@ -2261,7 +2261,7 @@ static const uint32_t cwsr_trap_gfx10_hex[] = { 0xbf8a, 0x877aff6d, 0x8000, 0xbf840040, 0x8f7b867b, 0x8f7b827b, - 0xbef6037b, 0xb9702a05, + 0xbef6037b, 0xb9703a05, 0x80708170, 0xbf0d9973, 0xbf850002, 0x8f708970, 0xbf820001, 0x8f708a70, @@ -2298,7 +2298,7 @@ static const uint32_t cwsr_trap_gfx10_hex[] = { 0x0200, 0xbeff0380, 0xbf820003, 0xbef003ff, 0x0400, 0xbeff03c1, - 0xb97b2a05, 0x807b817b, + 0xb97b3a05, 0x807b817b, 0x8f7b827b, 0x907c9973, 0x877c817c, 0xbf06817c, 0xbf850017, 0xbef603ff, @@ -2345,7 +2345,7 @@ static const uint32_t cwsr_trap_gfx10_hex[] = { 0xbeff03c1, 0xb96f4306, 0x876fc16f, 0xbf840029, 0x8f6f866f, 0x8f6f826f, - 0xbef6036f, 0xb9782a05, + 0xbef6036f, 0xb9783a05, 0x80788178, 0xbf0d9972, 0xbf850002, 0x8f788978, 0xbf820001, 0x8f788a78, @@ -2369,7 +2369,7 @@ static const uint32_t cwsr_trap_gfx10_hex[] = { 0x877c817c, 0xbf06817c, 0xbf850002, 0xbeff0380, 0xbf820001, 0xbeff03c1, - 0xb96f2a05, 0x806f816f, + 0xb96f3a05, 0x806f816f, 0x8f6f826f, 0x907c9972, 0x877c817c, 0xbf06817c, 0xbf850024, 0xbef603ff, @@ -2416,7 +2416,7 @@ static const uint32_t cwsr_trap_gfx10_hex[] = { 0xe0304100, 0x6e5d0100, 0xe0304200, 0x6e5d0200, 0xe0304300, 0x6e5d0300, - 0xbf8c3f70, 0xb9782a05, + 0xbf8c3f70, 0xb9783a05, 0x80788178, 0xbf0d9972, 0xbf850002, 0x8f788978, 0xbf820001, 0x8f788a78, @@ -2444,7 +2444,7 @@ static const uint32_t cwsr_trap_gfx10_hex[] = { 0xbe8e310e, 0xbf06807c, 0xbf84fff0, 0xba80f801, 0x, 0xbf8a, - 0xb9782a05, 0x80788178, + 0xb9783a05, 0x80788178, 0xbf0d9972, 0xbf850002, 0x8f788978, 0xbf820001, 0x
Re: [linux-next:master] BUILD REGRESSION 1e1b28b936aed946122b4e0991e7144fdbbfd77e
On Sun, 15 May 2022 21:00:21 +0800 kernel test robot wrote: > tree/branch: > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git master > branch HEAD: 1e1b28b936aed946122b4e0991e7144fdbbfd77e Add linux-next > specific files for 20220513 > > Error/Warning reports: > > https://lore.kernel.org/linux-mm/202204181931.klac6fwo-...@intel.com > https://lore.kernel.org/linux-mm/202204291924.vtgzmeri-...@intel.com > https://lore.kernel.org/linux-mm/202205031017.4twman3l-...@intel.com > https://lore.kernel.org/linux-mm/202205041248.wgcwpcev-...@intel.com > https://lore.kernel.org/linux-mm/202205122113.ulkzd3sz-...@intel.com > https://lore.kernel.org/linux-mm/202205150051.3rzuooag-...@intel.com > https://lore.kernel.org/linux-mm/202205150117.sd6hzbvm-...@intel.com > https://lore.kernel.org/lkml/202205100617.5uum3uet-...@intel.com > https://lore.kernel.org/llvm/202204210555.dnvfhvib-...@intel.com > https://lore.kernel.org/llvm/202205060132.uhqyux1l-...@intel.com > https://lore.kernel.org/llvm/202205120010.zwbednzm-...@intel.com > https://lore.kernel.org/llvm/202205141122.qihfguem-...@intel.com > > Error/Warning: (recently discovered and may have been fixed) > > : fatal error: ./include/generated/utsrelease.h: No such file > or directory > arch/arm/mach-versatile/versatile.c:56:14: warning: no previous prototype for > function 'mmc_status' [-Wmissing-prototypes] > arch/x86/kvm/pmu.h:20:32: warning: 'vmx_icl_pebs_cpu' defined but not used > [-Wunused-const-variable=] > drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_link_dp.c:5102:7: warning: > variable 'allow_lttpr_non_transparent_mode' set but not used > [-Wunused-but-set-variable] > drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_link_dp.c:5147:6: warning: > no previous prototype for function 'dp_parse_lttpr_mode' > [-Wmissing-prototypes] > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1364:5: warning: no previous > prototype for 'amdgpu_discovery_get_mall_info' [-Wmissing-prototypes] > drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c:1983:6: warning: no previous prototype > for function 'gfx_v11_0_rlc_stop' [-Wmissing-prototypes] > drivers/gpu/drm/amd/amdgpu/soc21.c:171:6: warning: no previous prototype for > 'soc21_grbm_select' [-Wmissing-prototypes] > drivers/gpu/drm/solomon/ssd130x-spi.c:154:35: warning: 'ssd130x_spi_table' > defined but not used [-Wunused-const-variable=] > drivers/hwmon/nct6775-platform.c:199:9: sparse:unsigned char > drivers/hwmon/nct6775-platform.c:199:9: sparse:void > drivers/video/fbdev/omap/hwa742.c:492:5: warning: no previous prototype for > 'hwa742_update_window_async' [-Wmissing-prototypes] > fs/buffer.c:2254:5: warning: stack frame size (2144) exceeds limit (1024) in > 'block_read_full_folio' [-Wframe-larger-than] > fs/ntfs/aops.c:378:12: warning: stack frame size (2224) exceeds limit (1024) > in 'ntfs_read_folio' [-Wframe-larger-than] > kernel/trace/fgraph.c:37:12: warning: no previous prototype for > 'ftrace_enable_ftrace_graph_caller' [-Wmissing-prototypes] > kernel/trace/fgraph.c:46:12: warning: no previous prototype for > 'ftrace_disable_ftrace_graph_caller' [-Wmissing-prototypes] Is this report CCed everywhere or there's a reason why netdev@ is CCed? I'm trying to figure out we need to care and it's not obvious..
Re: [PATCH] drm/amdgpu: Clean up of initializing doorbells for gfx_v9 and gfx_v10
Applied. Thanks! Alex On Mon, May 16, 2022 at 8:01 AM wrote: > > From: Haohui Mai > > Clean up redundant, copy-paste code blocks during the intialization of > the doorbells in mqd_init(). > > Signed-off-by: Haohui Mai > --- > drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 17 - > drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 17 - > 2 files changed, 34 deletions(-) > > diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c > b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c > index 4b66b9c93754..12f63d6aab12 100644 > --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c > +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c > @@ -6919,23 +6919,6 @@ static int gfx_v10_0_compute_mqd_init(struct > amdgpu_device *adev, void *m, > mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffc; > mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & > 0x; > > - tmp = 0; > - /* enable the doorbell if requested */ > - if (prop->use_doorbell) { > - tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL); > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > - DOORBELL_OFFSET, prop->doorbell_index); > - > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > - DOORBELL_EN, 1); > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > - DOORBELL_SOURCE, 0); > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > - DOORBELL_HIT, 0); > - } > - > - mqd->cp_hqd_pq_doorbell_control = tmp; > - > /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ > mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR); > > diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c > b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c > index 83639b5ea6a9..f49a2dd89ee7 100644 > --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c > +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c > @@ -3535,23 +3535,6 @@ static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring) > mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffc; > mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & > 0x; > > - tmp = 0; > - /* enable the doorbell if requested */ > - if (ring->use_doorbell) { > - tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL); > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > - DOORBELL_OFFSET, ring->doorbell_index); > - > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > -DOORBELL_EN, 1); > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > -DOORBELL_SOURCE, 0); > - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, > -DOORBELL_HIT, 0); > - } > - > - mqd->cp_hqd_pq_doorbell_control = tmp; > - > /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ > ring->wptr = 0; > mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR); > -- > 2.25.1 >
Re: VCN_INFO_TABLE_MAX_NUM_INSTANCES vs AMDGPU_MAX_VCN_INSTANCES
Looks good to me! Den mån 16 maj 2022 kl 20:15 skrev Alex Deucher : > On Mon, May 16, 2022 at 2:10 PM Christian König > wrote: > > > > Am 16.05.22 um 19:49 schrieb Ernst Sjöstrand: > > > > Den mån 16 maj 2022 kl 17:13 skrev Alex Deucher : > >> > >> On Sun, May 15, 2022 at 11:46 AM Ernst Sjöstrand > wrote: > >> > > >> > smatch found this problem on amd-staging-drm-next: > >> > > >> > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1443 > amdgpu_discovery_get_vcn_info() error: buffer overflow > 'adev->vcn.vcn_codec_disable_mask' 2 <= 3 > >> > > >> > This is caused by: > >> > #define AMDGPU_MAX_VCN_INSTANCES 2 > >> > #define VCN_INFO_TABLE_MAX_NUM_INSTANCES 4 > >> > > >> > Can we just drop VCN_INFO_TABLE_MAX_NUM_INSTANCES completely and use > AMDGPU_MAX_VCN_INSTANCES everywhere instead (and bump it to 4)? > >> > >> We should be able to bump AMDGPU_MAX_VCN_INSTANCES to 4 (although it > >> would waste some memory in the places it is used at this point). > >> VCN_INFO_TABLE_MAX_NUM_INSTANCES is part of a firmware structure so we > >> can't change that without breaking the firmware structure. > >> > >> Alex > > > > > > It would be nice to get rid of this pattern and make sure it doesn't > happen again when the VCN info table is raised to 5. > > It's very similar to the HWIP_MAX_INSTANCE issue. > > > > > > No, as Alex explained that distinction is intentional. > > > > The firmware definition is 4 for future extensions, that doesn't mean > that this is currently used. > > > > There is currently simply no need to set AMDGPU_MAX_VCN_INSTANCES to > more than 2. > > Right. The attached patch should protect against the scenario you are > envisioning. > > Alex > > > > > Regards, > > Christian. > > > > > > //E > > > > >
[PATCH] drm/amdgpu/discovery: validate VCN and SDMA instances
Validate the VCN and SDMA instances against the driver structure sizes to make sure we don't get into a situation where the firmware reports more instances than the driver supports. Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 17 ++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c index 1f4e07a32445..2c4f1adb5343 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c @@ -1137,13 +1137,24 @@ static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev) adev->vcn.vcn_config[adev->vcn.num_vcn_inst] = ip->revision & 0xc0; ip->revision &= ~0xc0; - adev->vcn.num_vcn_inst++; + if (adev->vcn.num_vcn_inst < AMDGPU_MAX_VCN_INSTANCES) + adev->vcn.num_vcn_inst++; + else + dev_err(adev->dev, "Too many VCN instances: %d vs %d\n", + adev->vcn.num_vcn_inst + 1, + AMDGPU_MAX_VCN_INSTANCES); } if (le16_to_cpu(ip->hw_id) == SDMA0_HWID || le16_to_cpu(ip->hw_id) == SDMA1_HWID || le16_to_cpu(ip->hw_id) == SDMA2_HWID || - le16_to_cpu(ip->hw_id) == SDMA3_HWID) - adev->sdma.num_instances++; + le16_to_cpu(ip->hw_id) == SDMA3_HWID) { + if (adev->sdma.num_instances < AMDGPU_MAX_SDMA_INSTANCES) + adev->sdma.num_instances++; + else + dev_err(adev->dev, "Too many SDMA instances: %d vs %d\n", + adev->sdma.num_instances + 1, + AMDGPU_MAX_SDMA_INSTANCES); + } if (le16_to_cpu(ip->hw_id) == UMC_HWID) adev->gmc.num_umc++; -- 2.35.3
Re: VCN_INFO_TABLE_MAX_NUM_INSTANCES vs AMDGPU_MAX_VCN_INSTANCES
On Mon, May 16, 2022 at 2:10 PM Christian König wrote: > > Am 16.05.22 um 19:49 schrieb Ernst Sjöstrand: > > Den mån 16 maj 2022 kl 17:13 skrev Alex Deucher : >> >> On Sun, May 15, 2022 at 11:46 AM Ernst Sjöstrand wrote: >> > >> > smatch found this problem on amd-staging-drm-next: >> > >> > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1443 >> > amdgpu_discovery_get_vcn_info() error: buffer overflow >> > 'adev->vcn.vcn_codec_disable_mask' 2 <= 3 >> > >> > This is caused by: >> > #define AMDGPU_MAX_VCN_INSTANCES 2 >> > #define VCN_INFO_TABLE_MAX_NUM_INSTANCES 4 >> > >> > Can we just drop VCN_INFO_TABLE_MAX_NUM_INSTANCES completely and use >> > AMDGPU_MAX_VCN_INSTANCES everywhere instead (and bump it to 4)? >> >> We should be able to bump AMDGPU_MAX_VCN_INSTANCES to 4 (although it >> would waste some memory in the places it is used at this point). >> VCN_INFO_TABLE_MAX_NUM_INSTANCES is part of a firmware structure so we >> can't change that without breaking the firmware structure. >> >> Alex > > > It would be nice to get rid of this pattern and make sure it doesn't happen > again when the VCN info table is raised to 5. > It's very similar to the HWIP_MAX_INSTANCE issue. > > > No, as Alex explained that distinction is intentional. > > The firmware definition is 4 for future extensions, that doesn't mean that > this is currently used. > > There is currently simply no need to set AMDGPU_MAX_VCN_INSTANCES to more > than 2. Right. The attached patch should protect against the scenario you are envisioning. Alex > > Regards, > Christian. > > > //E > > From 6fecd3c86fbb7b89d59d16cffea438e5b2a5915c Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Mon, 16 May 2022 14:12:33 -0400 Subject: [PATCH] drm/amdgpu/discovery: validate VCN and SDMA instances Validate the VCN and SDMA instances against the driver structure sizes to make sure we don't get into a situation where the firmware reports more instances than the driver supports. Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 17 ++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c index 1f4e07a32445..2c4f1adb5343 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c @@ -1137,13 +1137,24 @@ static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev) adev->vcn.vcn_config[adev->vcn.num_vcn_inst] = ip->revision & 0xc0; ip->revision &= ~0xc0; -adev->vcn.num_vcn_inst++; +if (adev->vcn.num_vcn_inst < AMDGPU_MAX_VCN_INSTANCES) + adev->vcn.num_vcn_inst++; +else + dev_err(adev->dev, "Too many VCN instances: %d vs %d\n", + adev->vcn.num_vcn_inst + 1, + AMDGPU_MAX_VCN_INSTANCES); } if (le16_to_cpu(ip->hw_id) == SDMA0_HWID || le16_to_cpu(ip->hw_id) == SDMA1_HWID || le16_to_cpu(ip->hw_id) == SDMA2_HWID || - le16_to_cpu(ip->hw_id) == SDMA3_HWID) -adev->sdma.num_instances++; + le16_to_cpu(ip->hw_id) == SDMA3_HWID) { +if (adev->sdma.num_instances < AMDGPU_MAX_SDMA_INSTANCES) + adev->sdma.num_instances++; +else + dev_err(adev->dev, "Too many SDMA instances: %d vs %d\n", + adev->sdma.num_instances + 1, + AMDGPU_MAX_SDMA_INSTANCES); + } if (le16_to_cpu(ip->hw_id) == UMC_HWID) adev->gmc.num_umc++; -- 2.35.3
Re: VCN_INFO_TABLE_MAX_NUM_INSTANCES vs AMDGPU_MAX_VCN_INSTANCES
Am 16.05.22 um 19:49 schrieb Ernst Sjöstrand: Den mån 16 maj 2022 kl 17:13 skrev Alex Deucher : On Sun, May 15, 2022 at 11:46 AM Ernst Sjöstrand wrote: > > smatch found this problem on amd-staging-drm-next: > > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1443 amdgpu_discovery_get_vcn_info() error: buffer overflow 'adev->vcn.vcn_codec_disable_mask' 2 <= 3 > > This is caused by: > #define AMDGPU_MAX_VCN_INSTANCES 2 > #define VCN_INFO_TABLE_MAX_NUM_INSTANCES 4 > > Can we just drop VCN_INFO_TABLE_MAX_NUM_INSTANCES completely and use AMDGPU_MAX_VCN_INSTANCES everywhere instead (and bump it to 4)? We should be able to bump AMDGPU_MAX_VCN_INSTANCES to 4 (although it would waste some memory in the places it is used at this point). VCN_INFO_TABLE_MAX_NUM_INSTANCES is part of a firmware structure so we can't change that without breaking the firmware structure. Alex It would be nice to get rid of this pattern and make sure it doesn't happen again when the VCN info table is raised to 5. It's very similar to the HWIP_MAX_INSTANCE issue. No, as Alex explained that distinction is intentional. The firmware definition is 4 for future extensions, that doesn't mean that this is currently used. There is currently simply no need to set AMDGPU_MAX_VCN_INSTANCES to more than 2. Regards, Christian. //E
Re: VCN_INFO_TABLE_MAX_NUM_INSTANCES vs AMDGPU_MAX_VCN_INSTANCES
Den mån 16 maj 2022 kl 17:13 skrev Alex Deucher : > On Sun, May 15, 2022 at 11:46 AM Ernst Sjöstrand wrote: > > > > smatch found this problem on amd-staging-drm-next: > > > > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1443 > amdgpu_discovery_get_vcn_info() error: buffer overflow > 'adev->vcn.vcn_codec_disable_mask' 2 <= 3 > > > > This is caused by: > > #define AMDGPU_MAX_VCN_INSTANCES 2 > > #define VCN_INFO_TABLE_MAX_NUM_INSTANCES 4 > > > > Can we just drop VCN_INFO_TABLE_MAX_NUM_INSTANCES completely and use > AMDGPU_MAX_VCN_INSTANCES everywhere instead (and bump it to 4)? > > We should be able to bump AMDGPU_MAX_VCN_INSTANCES to 4 (although it > would waste some memory in the places it is used at this point). > VCN_INFO_TABLE_MAX_NUM_INSTANCES is part of a firmware structure so we > can't change that without breaking the firmware structure. > > Alex > It would be nice to get rid of this pattern and make sure it doesn't happen again when the VCN info table is raised to 5. It's very similar to the HWIP_MAX_INSTANCE issue. //E
Re: [PATCH v2 00/19] DC/DM changes needed for amdgpu PSR-SU
On Mon, 16 May 2022 at 18:23, Leo Li wrote: > > > > On 2022-05-12 13:39, Daniel Vetter wrote: > > On Thu, 12 May 2022 at 19:22, Zhang, Dingchen (David) > > wrote: > >> > >> [AMD Official Use Only - General] > >> > >> Hi Daniel > >> > >> Thanks for your comments and explanations. I replied in-line and look > >> forward to more discussion. > >> > >> regards > >> David > >> > >> > >> From: Daniel Vetter > >> Sent: Thursday, May 12, 2022 7:22 AM > >> To: Alex Deucher > >> Cc: Zhang, Dingchen (David) ; amd-gfx list > >> ; Wang, Chao-kai (Stylon) > >> ; Li, Sun peng (Leo) ; Wentland, > >> Harry ; Zhuo, Qingqing (Lillian) > >> ; Siqueira, Rodrigo ; Li, > >> Roman ; Chiu, Solomon ; Zuo, Jerry > >> ; Pillai, Aurabindo ; Lin, > >> Wayne ; Lakha, Bhawanpreet ; > >> Gutierrez, Agustin ; Kotarac, Pavle > >> > >> Subject: Re: [PATCH v2 00/19] DC/DM changes needed for amdgpu PSR-SU > >> > >> On Wed, 11 May 2022 at 17:35, Alex Deucher wrote: > >>> > >>> On Tue, May 10, 2022 at 4:45 PM David Zhang > >>> wrote: > > changes in v2: > --- > - set vsc_packet_rev2 for PSR1 which is safer > - add exposure of AMD specific DPCD regs for PSR-SU-RC (rate-control) > - add DC/DM change related to amdgpu PSR-SU-RC > > > David Zhang (18): > drm/amd/display: align dmub cmd header to latest dmub FW to support > PSR-SU > drm/amd/display: feed PSR-SU as psr version to dmub FW > drm/amd/display: combine dirty rectangles in DMUB FW > drm/amd/display: update GSP1 generic info packet for PSRSU > drm/amd/display: revise Start/End SDP data > drm/amd/display: program PSR2 DPCD Configuration > drm/amd/display: Passing Y-granularity to dmub fw > drm/amd/display: Set default value of line_capture_indication > drm/amd/display: add vline time in micro sec to PSR context > drm/amd/display: fix system hang when PSR exits > drm/amd/display: Set PSR level to enable ALPM by default > drm/amd/display: use HW lock mgr for PSR-SU > drm/amd/display: PSRSU+DSC WA for specific TCON > drm/amd/display: add shared helpers to update psr config fields to > power module > drm/amd/display: calculate psr config settings in runtime in DM > drm/amd/display: update cursor position to DMUB FW > drm/amd/display: expose AMD source specific DPCD for FreeSync PSR > support > drm/amd/display: PSR-SU rate control support in DC > > Leo Li (1): > drm/amd/display: Implement MPO PSR SU > >>> > >>> A couple of suggestions from Daniel on IRC: > >>> 1. Might be good to extract the "calculate total crtc damage" code > >>> from i915 in intel_psr2_sel_fetch_update, stuff that into damage > >>> helpers and reuse for i915 and amdgpu > >> > >> To expand a bit on this. There is currently a helper for total damage, > >> but it's at the fb/plane level for drivers which need to upload > >> buffers (usb/spi or virtual) drm_atomic_helper_damage_merged(). That > >> one probably needs to be renamed to signify it's about the plane, and > >> then we need a new drm_atomic_helper_crtc_damage_merged() which > >> (extract from i915 code ideally) which computes total crtc damage for > >> stuff like psr2/su or the command mode dsi panels (unfortunately none > >> of the drivers for android for these panels have been upstreamed yet). > >> > >> <<< > >> Checked the DRM comment for the helper `drm_atomic_helper_damage_merged()` > >> and > >> quoted below: > >> * > >> Drivers might want to use the helper functions > >> drm_atomic_helper_damage_iter_init() > >> and drm_atomic_helper_damage_iter_next() or > >> drm_atomic_helper_damage_merged() > >> if the driver can only handle a single damage region at most. > >> * > >> Currently for amdgpu, the multiple damage clips combination (merging) is > >> handled in > >> DMUB firmware. And the DRM comment shows that the usage of > >> "damage_merged()" > >> helper is for the driver which can only handle single damage region at > >> most. > >> > >> Since AMDGPU is capable of handling multiple damaged clip (in DMUB FW), > >> can I > >> understand that the group of helpers of `damage_merged()` in DRM is not > >> mandatory > >> but optional? > > > > Ah I didn't see from a quick read that this was possible. How does > > this work when the plane is enabled/disabled or resized or moved? > > -Daniel > > Hi Daniel, > > When a plane is disabled, enabled, and/or resized(*), PSR is temporarily > disabled. The mechanism to do so isn't in this patchset, but was added > when PSR1 was first introduced to amdgpu_dm. > > In short, amdgpu_dm will disable PSR whenever DC requires a full update > to program an atomic state (needs bandwidth recalculations and/or > shuffling hw resources). For DC, enabling, disabling, and changing the > scaling of a plane are considered full updates. > > When the plane is moved,
Re: [PATCH v2 00/19] DC/DM changes needed for amdgpu PSR-SU
On 2022-05-12 13:39, Daniel Vetter wrote: On Thu, 12 May 2022 at 19:22, Zhang, Dingchen (David) wrote: [AMD Official Use Only - General] Hi Daniel Thanks for your comments and explanations. I replied in-line and look forward to more discussion. regards David From: Daniel Vetter Sent: Thursday, May 12, 2022 7:22 AM To: Alex Deucher Cc: Zhang, Dingchen (David) ; amd-gfx list ; Wang, Chao-kai (Stylon) ; Li, Sun peng (Leo) ; Wentland, Harry ; Zhuo, Qingqing (Lillian) ; Siqueira, Rodrigo ; Li, Roman ; Chiu, Solomon ; Zuo, Jerry ; Pillai, Aurabindo ; Lin, Wayne ; Lakha, Bhawanpreet ; Gutierrez, Agustin ; Kotarac, Pavle Subject: Re: [PATCH v2 00/19] DC/DM changes needed for amdgpu PSR-SU On Wed, 11 May 2022 at 17:35, Alex Deucher wrote: On Tue, May 10, 2022 at 4:45 PM David Zhang wrote: changes in v2: --- - set vsc_packet_rev2 for PSR1 which is safer - add exposure of AMD specific DPCD regs for PSR-SU-RC (rate-control) - add DC/DM change related to amdgpu PSR-SU-RC David Zhang (18): drm/amd/display: align dmub cmd header to latest dmub FW to support PSR-SU drm/amd/display: feed PSR-SU as psr version to dmub FW drm/amd/display: combine dirty rectangles in DMUB FW drm/amd/display: update GSP1 generic info packet for PSRSU drm/amd/display: revise Start/End SDP data drm/amd/display: program PSR2 DPCD Configuration drm/amd/display: Passing Y-granularity to dmub fw drm/amd/display: Set default value of line_capture_indication drm/amd/display: add vline time in micro sec to PSR context drm/amd/display: fix system hang when PSR exits drm/amd/display: Set PSR level to enable ALPM by default drm/amd/display: use HW lock mgr for PSR-SU drm/amd/display: PSRSU+DSC WA for specific TCON drm/amd/display: add shared helpers to update psr config fields to power module drm/amd/display: calculate psr config settings in runtime in DM drm/amd/display: update cursor position to DMUB FW drm/amd/display: expose AMD source specific DPCD for FreeSync PSR support drm/amd/display: PSR-SU rate control support in DC Leo Li (1): drm/amd/display: Implement MPO PSR SU A couple of suggestions from Daniel on IRC: 1. Might be good to extract the "calculate total crtc damage" code from i915 in intel_psr2_sel_fetch_update, stuff that into damage helpers and reuse for i915 and amdgpu To expand a bit on this. There is currently a helper for total damage, but it's at the fb/plane level for drivers which need to upload buffers (usb/spi or virtual) drm_atomic_helper_damage_merged(). That one probably needs to be renamed to signify it's about the plane, and then we need a new drm_atomic_helper_crtc_damage_merged() which (extract from i915 code ideally) which computes total crtc damage for stuff like psr2/su or the command mode dsi panels (unfortunately none of the drivers for android for these panels have been upstreamed yet). <<< Checked the DRM comment for the helper `drm_atomic_helper_damage_merged()` and quoted below: * Drivers might want to use the helper functions drm_atomic_helper_damage_iter_init() and drm_atomic_helper_damage_iter_next() or drm_atomic_helper_damage_merged() if the driver can only handle a single damage region at most. * Currently for amdgpu, the multiple damage clips combination (merging) is handled in DMUB firmware. And the DRM comment shows that the usage of "damage_merged()" helper is for the driver which can only handle single damage region at most. Since AMDGPU is capable of handling multiple damaged clip (in DMUB FW), can I understand that the group of helpers of `damage_merged()` in DRM is not mandatory but optional? Ah I didn't see from a quick read that this was possible. How does this work when the plane is enabled/disabled or resized or moved? -Daniel Hi Daniel, When a plane is disabled, enabled, and/or resized(*), PSR is temporarily disabled. The mechanism to do so isn't in this patchset, but was added when PSR1 was first introduced to amdgpu_dm. In short, amdgpu_dm will disable PSR whenever DC requires a full update to program an atomic state (needs bandwidth recalculations and/or shuffling hw resources). For DC, enabling, disabling, and changing the scaling of a plane are considered full updates. When the plane is moved, both the old and new plane bounds are given to FW as dirty rectangles. (*)Resize should fall under the same bucket, though DC would consider that as a full update. I think disabling PSR would take precedence... will give this another spin to check. Thanks, Leo I also think that the split between dc and kms is a bit funny, I'd put only the resulting damage rect into dc_pipe and do the computation of that in the drm/kms linux code outside of dc functions (or in the glue code for dc), since I'm assuming on windows it's completely different approach in how you compute damage. Especially once we have the crtc damage helper on linux. 2. The co
Re: [PATCH 2/3] drm/amdgpu: add AMDGPU_VM_NOALLOC
I don't have access to any gfx10 hardware. Can you give me a dmesg and/or backtrace, etc..? I can't push this unless it's working properly. Christian. Am 16.05.22 um 14:56 schrieb Marek Olšák: Reproduction steps: - use mesa/main on gfx10.3 (not sure what other GPUs do) - run: radeonsi_mall_noalloc=true glxgears Marek On Mon, May 16, 2022 at 7:53 AM Christian König wrote: Crap, do you have a link to the failure? Am 16.05.22 um 13:10 schrieb Marek Olšák: I forgot to say: The NOALLOC flag causes an allocation failure, so there is a kernel bug somewhere. Marek On Mon, May 16, 2022 at 7:06 AM Marek Olšák wrote: FYI, I think it's time to merge this because the Mesa commits are going to be merged in ~30 minutes if Gitlab CI is green, and that includes updated amdgpu_drm.h. Marek On Wed, May 11, 2022 at 2:55 PM Marek Olšák wrote: Ok sounds good. Marek On Wed., May 11, 2022, 03:43 Christian König, wrote: It really *is* a NOALLOC feature. In other words there is no latency improvement on reads because the cache is always checked, even with the noalloc flag set. The only thing it affects is that misses not enter the cache and so don't cause any additional pressure on evicting cache lines. You might want to double check with the hardware guys, but I'm something like 95% sure that it works this way. Christian. Am 11.05.22 um 09:22 schrieb Marek Olšák: Bypass means that the contents of the cache are ignored, which decreases latency at the cost of no coherency between bypassed and normal memory requests. NOA (noalloc) means that the cache is checked and can give you cache hits, but misses are not cached and the overall latency is higher. I don't know what the hw does, but I hope it was misnamed and it really means bypass because there is no point in doing cache lookups on every memory request if the driver wants to disable caching to *decrease* latency in the situations when the cache isn't helping. Marek On Wed, May 11, 2022 at 2:15 AM Lazar, Lijo wrote: On 5/11/2022 11:36 AM, Christian König wrote: > Mhm, it doesn't really bypass MALL. It just doesn't allocate any MALL > entries on write. > > How about AMDGPU_VM_PAGE_NO_MALL ? One more - AMDGPU_VM_PAGE_LLC_* [ LLC = last level cache, * = some sort of attribute which decides LLC behaviour] Thanks, Lijo > > Christian. > > Am 10.05.22 um 23:21 schrieb Marek Olšák: >> A better name would be: >> AMDGPU_VM_PAGE_BYPASS_MALL >> >> Marek >> >> On Fri, May 6, 2022 at 7:23 AM Christian König >> wrote: >> >> Add the AMDGPU_VM_NOALLOC flag to let userspace control MALL >> allocation. >> >> Only compile tested! >> >> Signed-off-by: Christian König >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++ >> drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +++ >> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 3 +++ >> include/uapi/drm/amdgpu_drm.h | 2 ++ >> 4 files changed, 10 insertions(+) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> index bf97d8f07f57..d8129626581f 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> @@ -650,6 +650,8 @@ uint64_t amdgpu_gem_va_map_flags(struct >> amdgpu_device *adev, uint32_t flags) >> pte_flag |= AMDGPU_PTE_WRITEABLE; >> if (flags & AMDGPU_VM_PAGE_PRT) >> pte_flag |= AMDGPU_PTE_PRT; >> + if (fl
Re: [PATCH] drm/amdgpu: Fix multiple GPU resets in XGMI hive.
On 2022-05-16 11:08, Christian König wrote: Am 16.05.22 um 16:12 schrieb Andrey Grodzovsky: Ping Ah, yes sorry. Andrey On 2022-05-13 11:41, Andrey Grodzovsky wrote: Yes, exactly that's the idea. Basically the reset domain knowns which amdgpu devices it needs to reset together. If you then represent that so that you always have a hive even when you only have one device in it, or if you put an array of devices which needs to be reset together into the reset domain doesn't matter. Maybe go for the later approach, that is probably a bit cleaner and less code to change. Christian. Unfortunately this approach raises also a few difficulties - First - if holding array of devices in reset_domain then when you come to GPU reset function you don't really know which adev is the one triggered the reset and this is actually essential to some procedures like emergency restart. What is "emergency restart"? That's not some requirement I know about. Emergency restart is something u can see at the beginning of amdgpu_gpu_recover function - it's a historical work around for some type of ASICs who weren't able to do full reset I think. We should eventually remove it bu for now I thin it's still in use. Second - in XGMI case we must take into account that one of the hive members might go away in runtime (i could do echo 1 > /sysfs/pci_id/remove on it for example at any moment) - so now we need to maintain this array and mark such entry with NULL probably on XGMI node removal , and then there might be hot insertion and all this adds more complications. I now tend to prefer your initial solution for it's simplicity and the result will be what we need - "E.g. in the reset code (either before or after the reset, that's debatable) you do something like this: for (i = 0; i < num_ring; ++i) cancel_delayed_work(ring[i]->scheduler) cancel_work(adev->ras_work); cancel_work(adev->iofault_work); cancel_work(adev->debugfs_work); " Works for me. I already expected that switching over the reset to be based on the reset context wouldn't be that easy. Regards, Christian. Ok - i will resend a patch. Andrey Let me know what you think. Andrey
Re: VCN_INFO_TABLE_MAX_NUM_INSTANCES vs AMDGPU_MAX_VCN_INSTANCES
On Sun, May 15, 2022 at 11:46 AM Ernst Sjöstrand wrote: > > smatch found this problem on amd-staging-drm-next: > > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c:1443 > amdgpu_discovery_get_vcn_info() error: buffer overflow > 'adev->vcn.vcn_codec_disable_mask' 2 <= 3 > > This is caused by: > #define AMDGPU_MAX_VCN_INSTANCES 2 > #define VCN_INFO_TABLE_MAX_NUM_INSTANCES 4 > > Can we just drop VCN_INFO_TABLE_MAX_NUM_INSTANCES completely and use > AMDGPU_MAX_VCN_INSTANCES everywhere instead (and bump it to 4)? We should be able to bump AMDGPU_MAX_VCN_INSTANCES to 4 (although it would waste some memory in the places it is used at this point). VCN_INFO_TABLE_MAX_NUM_INSTANCES is part of a firmware structure so we can't change that without breaking the firmware structure. Alex > > //E >
Re: [PATCH] drm/amdgpu: Fix multiple GPU resets in XGMI hive.
Am 16.05.22 um 16:12 schrieb Andrey Grodzovsky: Ping Ah, yes sorry. Andrey On 2022-05-13 11:41, Andrey Grodzovsky wrote: Yes, exactly that's the idea. Basically the reset domain knowns which amdgpu devices it needs to reset together. If you then represent that so that you always have a hive even when you only have one device in it, or if you put an array of devices which needs to be reset together into the reset domain doesn't matter. Maybe go for the later approach, that is probably a bit cleaner and less code to change. Christian. Unfortunately this approach raises also a few difficulties - First - if holding array of devices in reset_domain then when you come to GPU reset function you don't really know which adev is the one triggered the reset and this is actually essential to some procedures like emergency restart. What is "emergency restart"? That's not some requirement I know about. Second - in XGMI case we must take into account that one of the hive members might go away in runtime (i could do echo 1 > /sysfs/pci_id/remove on it for example at any moment) - so now we need to maintain this array and mark such entry with NULL probably on XGMI node removal , and then there might be hot insertion and all this adds more complications. I now tend to prefer your initial solution for it's simplicity and the result will be what we need - "E.g. in the reset code (either before or after the reset, that's debatable) you do something like this: for (i = 0; i < num_ring; ++i) cancel_delayed_work(ring[i]->scheduler) cancel_work(adev->ras_work); cancel_work(adev->iofault_work); cancel_work(adev->debugfs_work); " Works for me. I already expected that switching over the reset to be based on the reset context wouldn't be that easy. Regards, Christian. Let me know what you think. Andrey
Re: [PATCH v12] drm/amdgpu: add drm buddy support to amdgpu
On Mon, May 16, 2022 at 8:40 AM Mike Lothian wrote: > > Hi > > The merge window for 5.19 will probably be opening next week, has > there been any progress with this bug? It took a while to find a combination of GPUs that would repro the issue, but now that we can, it is still being investigated. Alex > > Thanks > > Mike > > On Mon, 2 May 2022 at 17:31, Mike Lothian wrote: > > > > On Mon, 2 May 2022 at 16:54, Arunpravin Paneer Selvam > > wrote: > > > > > > > > > > > > On 5/2/2022 8:41 PM, Mike Lothian wrote: > > > > On Wed, 27 Apr 2022 at 12:55, Mike Lothian wrote: > > > >> On Tue, 26 Apr 2022 at 17:36, Christian König > > > >> wrote: > > > >>> Hi Mike, > > > >>> > > > >>> sounds like somehow stitching together the SG table for PRIME doesn't > > > >>> work any more with this patch. > > > >>> > > > >>> Can you try with P2P DMA disabled? > > > >> -CONFIG_PCI_P2PDMA=y > > > >> +# CONFIG_PCI_P2PDMA is not set > > > >> > > > >> If that's what you're meaning, then there's no difference, I'll upload > > > >> my dmesg to the gitlab issue > > > >> > > > >>> Apart from that can you take a look Arun? > > > >>> > > > >>> Thanks, > > > >>> Christian. > > > > Hi > > > > > > > > Have you had any success in replicating this? > > > Hi Mike, > > > I couldn't replicate on my Raven APU machine. I see you have 2 cards > > > initialized, one is Renoir > > > and the other is Navy Flounder. Could you give some more details, are > > > you running Gravity Mark > > > on Renoir and what is your system RAM configuration? > > > > > > > > Cheers > > > > > > > > Mike > > > > > Hi > > > > It's a PRIME laptop, it failed on the RENOIR too, it caused a lockup, > > but systemd managed to capture it, I'll attach it to the issue > > > > I've got 64GB RAM, the 6800M has 12GB VRAM > > > > Cheers > > > > Mike
[PATCH] drm/amdkfd: simplify cpu hive assignment
CPU hive assignment currently assumes when a GPU hive is connected_to_cpu, there is only one hive in the system. Only assign CPUs to the hive if they are explicitly directly connected to the GPU hive to get rid of the need for this assumption. It's more efficient to do this when querying IO links since other non-CRAT info has to be filled in anyways. Also, stop re-assigning the same CPU to the same GPU hive if it has already been done before. Signed-off-by: Jonathan Kim --- drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 22 ++ 1 file changed, 6 insertions(+), 16 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 2e20f54bb147..8d50d207cf66 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -1271,6 +1271,12 @@ static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev) if (!peer_dev) continue; + /* Include the CPU peer in GPU hive if connected over xGMI. */ + if (!peer_dev->gpu && !peer_dev->node_props.hive_id && + dev->node_props.hive_id && + dev->gpu->adev->gmc.xgmi.connected_to_cpu) + peer_dev->node_props.hive_id = dev->node_props.hive_id; + list_for_each_entry(inbound_link, &peer_dev->io_link_props, list) { if (inbound_link->node_to != link->node_from) @@ -1302,22 +1308,6 @@ int kfd_topology_add_device(struct kfd_dev *gpu) pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id); - /* Include the CPU in xGMI hive if xGMI connected by assigning it the hive ID. */ - if (gpu->hive_id && gpu->adev->gmc.xgmi.connected_to_cpu) { - struct kfd_topology_device *top_dev; - - down_read(&topology_lock); - - list_for_each_entry(top_dev, &topology_device_list, list) { - if (top_dev->gpu) - break; - - top_dev->node_props.hive_id = gpu->hive_id; - } - - up_read(&topology_lock); - } - /* Check to see if this gpu device exists in the topology_device_list. * If so, assign the gpu to that device, * else create a Virtual CRAT for this gpu device and then parse that -- 2.25.1
Re: [PATCH] drm/amdgpu: Fix multiple GPU resets in XGMI hive.
Ping Andrey On 2022-05-13 11:41, Andrey Grodzovsky wrote: Yes, exactly that's the idea. Basically the reset domain knowns which amdgpu devices it needs to reset together. If you then represent that so that you always have a hive even when you only have one device in it, or if you put an array of devices which needs to be reset together into the reset domain doesn't matter. Maybe go for the later approach, that is probably a bit cleaner and less code to change. Christian. Unfortunately this approach raises also a few difficulties - First - if holding array of devices in reset_domain then when you come to GPU reset function you don't really know which adev is the one triggered the reset and this is actually essential to some procedures like emergency restart. Second - in XGMI case we must take into account that one of the hive members might go away in runtime (i could do echo 1 > /sysfs/pci_id/remove on it for example at any moment) - so now we need to maintain this array and mark such entry with NULL probably on XGMI node removal , and then there might be hot insertion and all this adds more complications. I now tend to prefer your initial solution for it's simplicity and the result will be what we need - "E.g. in the reset code (either before or after the reset, that's debatable) you do something like this: for (i = 0; i < num_ring; ++i) cancel_delayed_work(ring[i]->scheduler) cancel_work(adev->ras_work); cancel_work(adev->iofault_work); cancel_work(adev->debugfs_work); " Let me know what you think. Andrey
RE: [PATCH v2 00/11] DC Patches May 16, 2022
[Public] Hi all, This week this patchset was tested on the following systems: Sapphire Pulse RX5700XT Reference AMD RX6800 Engineering board with Ryzen 9 5900H These systems were tested on the following display types: eDP, (1080p 60hz [5900H]) VGA and DVI (1680x1050 60HZ [DP to VGA/DVI, USB-C to DVI/VGA]) DP/HDMI/USB-C (1440p 170hz, 4k 60hz, 4k 144hz [Includes USB-C to DP/HDMI adapters]) MST tested with Startech MST14DP123DP and 2x 4k 60Hz displays DSC tested with Cable Matters 101075 (DP to 3x DP), and 201375 (USB-C to 3x DP) with 3x 4k60 displays The testing is a mix of automated and manual tests. Manual testing includes (but is not limited to): Changing display configurations and settings Benchmark testing Feature testing (Freesync, etc.) Automated testing includes (but is not limited to): Script testing (scripts to automate some of the manual checks) IGT testing The patchset consists of the branch amd-staging-drm-next (Head commit - da38a66ac46e334f198afcd1b4d4554b4ddca0df -> drm/amdgpu: Ensure the DMA engine is deactivated during set ups) with a selection of patches added on top of it. This goes for both Ubuntu testing and Chrome OS testing. Tested on Ubuntu 22.04 Tested-by: Daniel Wheeler Thank you, Dan Wheeler Technologist | AMD SW Display -- 1 Commerce Valley Dr E, Thornhill, ON L3T 7X6 amd.com -Original Message- From: amd-gfx On Behalf Of Qingqing Zhuo Sent: May 13, 2022 5:17 PM To: amd-gfx@lists.freedesktop.org Cc: Wang, Chao-kai (Stylon) ; Li, Sun peng (Leo) ; Wentland, Harry ; Zhuo, Qingqing (Lillian) ; Siqueira, Rodrigo ; Li, Roman ; Chiu, Solomon ; Pillai, Aurabindo ; Lin, Wayne ; Lakha, Bhawanpreet ; Gutierrez, Agustin ; Kotarac, Pavle Subject: [PATCH v2 00/11] DC Patches May 16, 2022 This DC patchset brings improvements in multiple areas. In summary, we highlight: * Improvements in link training fallback * Adding individual edp hotplug support * Fixes in DPIA HPD status, display clock change hang, etc. * FPU isolation work for DCN30 --- Alvin Lee (1): drm/amd/display: Clean up code in dc Aric Cyr (1): drm/amd/display: 3.2.186 Bhawanpreet Lakha (1): drm/amd/display: Fic incorrect pipe being used for clk update David Galiffi (1): drm/amd/display: Check if modulo is 0 before dividing. Derek Lai (1): drm/amd/display: Allow individual control of eDP hotplug support Jasdeep Dhillon (1): drm/amd/display: Move FPU associated DCN30 code to DML folder Jimmy Kizito (2): drm/amd/display: Update link training fallback behaviour. drm/amd/display: Query DPIA HPD status. Michael Strauss (1): Revert "drm/amd/display: Refactor LTTPR cap retrieval" Nicholas Kazlauskas (1): drm/amd/display: Check zero planes for OTG disable W/A on clock change Paul Hsieh (1): drm/amd/display: clear request when release aux engine .../amd/display/dc/clk_mgr/dcn21/rn_clk_mgr.c | 2 +- .../dc/clk_mgr/dcn315/dcn315_clk_mgr.c| 3 +- .../dc/clk_mgr/dcn316/dcn316_clk_mgr.c| 3 +- drivers/gpu/drm/amd/display/dc/core/dc.c | 15 +- drivers/gpu/drm/amd/display/dc/core/dc_link.c | 24 +- .../gpu/drm/amd/display/dc/core/dc_link_dp.c | 248 --- .../drm/amd/display/dc/core/dc_link_dpia.c| 19 + drivers/gpu/drm/amd/display/dc/dc.h | 4 +- drivers/gpu/drm/amd/display/dc/dc_link.h | 2 - drivers/gpu/drm/amd/display/dc/dce/dce_aux.c | 3 +- .../drm/amd/display/dc/dce/dce_clock_source.c | 9 +- .../drm/amd/display/dc/dcn20/dcn20_hwseq.c| 1 - .../gpu/drm/amd/display/dc/dcn30/dcn30_optc.c | 10 + .../drm/amd/display/dc/dcn30/dcn30_resource.c | 516 ++- .../drm/amd/display/dc/dcn30/dcn30_resource.h | 5 + .../amd/display/dc/dcn301/dcn301_resource.c | 2 + .../amd/display/dc/dcn302/dcn302_resource.c | 2 + .../amd/display/dc/dcn303/dcn303_resource.c | 2 + .../drm/amd/display/dc/dcn31/dcn31_resource.c | 2 + drivers/gpu/drm/amd/display/dc/dml/Makefile | 3 +- .../drm/amd/display/dc/dml/dcn30/dcn30_fpu.c | 617 ++ .../drm/amd/display/dc/dml/dcn30/dcn30_fpu.h | 67 ++ .../gpu/drm/amd/display/dc/inc/core_types.h | 7 + .../gpu/drm/amd/display/dc/inc/dc_link_dp.h | 3 +- .../gpu/drm/amd/display/dc/inc/dc_link_dpia.h | 5 + .../amd/display/include/link_service_types.h | 6 - 26 files changed, 947 insertions(+), 633 deletions(-) create mode 100644 drivers/gpu/drm/amd/display/dc/dml/dcn30/dcn30_fpu.c create mode 100644 drivers/gpu/drm/amd/display/dc/dml/dcn30/dcn30_fpu.h -- 2.25.1
Re: Screen corruption using radeon kernel driver
On Mon, Apr 25, 2022 at 01:22:04PM -0400, Alex Deucher wrote: > + dri-devel > > On Mon, Apr 25, 2022 at 3:33 AM Krylov Michael wrote: > > > > Hello! > > > > After updating my Linux kernel from version 4.19 (Debian 10 version) to > > 5.10 (packaged with Debian 11), I've noticed that the image > > displayed on my older computer, 32-bit Pentium 4 using ATI Radeon X1950 > > AGP video card is severely corrupted in the graphical (Xorg and Wayland) > > mode: all kinds of black and white stripes across the screen, some > > letters missing, etc. > > > > I've checked several options (Xorg drivers, Wayland instead of > > Xorg, radeon.agpmode=-1 in kernel command line and so on), but the > > problem persisted. I've managed to find that the problem was in the > > kernel, as everything worked well with 4.19 kernel with everything > > else being from Debian 11. > > > > I have managed to find the culprit of that corruption, that is the > > commit 33b3ad3788aba846fc8b9a065fe2685a0b64f713 on the linux kernel. > > Reverting this commit and building the kernel with that commit reverted > > fixes the problem. Disabling HIMEM also gets rid of that problem. But it > > also leaves the system with less that 1G of RAM, which is, of course, > > undesirable. > > > > Apparently this problem is somewhat known, as I can tell after googling > > for the commit id, see this link for example: > > https://lkml.org/lkml/2020/1/9/518 > > > > Mageia distro, for example, reverted this commit in the kernel they are > > building: > > > > http://sophie.zarb.org/distrib/Mageia/7/i586/by-pkgid/b9193a4f85192bc57f4d770fb9bb399c/files/32 > > > > I've reported this bug to Debian bugtracker, checked the recent verion > > of the kernel (5.17), bug still persists. Here's a link to the Debian > > bug page: > > > > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=993670 > > > > I'm not sure if reverting this commit is the correct way to go, so if > > you need to check any changes/patches that I could apply and test on > > the real hardware, I'll be glad to do that (but please keep in mind > > that testing could take some time, I don't have access to this computer > > 24/7, but I'll do my best to respond ASAP). > > I would be happy to revert that commit. I attempted to revert it a > year or so ago, but Christoph didn't want to. He was going to look > further into it. I was not able to repro the issue. It seemed to be > related to highmem support. You might try disabling that. Here is > the previous thread for reference: > https://lists.freedesktop.org/archives/amd-gfx/2020-September/053922.html > > Alex Yeah, I tried to disable HIMEM, and that indeed fixes the problem, but it leaves me with less than 1G of available memory which is undesirable. signature.asc Description: PGP signature
Re: [PATCH 2/3] drm/amdgpu: add AMDGPU_VM_NOALLOC
Reproduction steps: - use mesa/main on gfx10.3 (not sure what other GPUs do) - run: radeonsi_mall_noalloc=true glxgears Marek On Mon, May 16, 2022 at 7:53 AM Christian König < ckoenig.leichtzumer...@gmail.com> wrote: > Crap, do you have a link to the failure? > > Am 16.05.22 um 13:10 schrieb Marek Olšák: > > I forgot to say: The NOALLOC flag causes an allocation failure, so there > is a kernel bug somewhere. > > Marek > > On Mon, May 16, 2022 at 7:06 AM Marek Olšák wrote: > >> FYI, I think it's time to merge this because the Mesa commits are going >> to be merged in ~30 minutes if Gitlab CI is green, and that includes >> updated amdgpu_drm.h. >> >> Marek >> >> On Wed, May 11, 2022 at 2:55 PM Marek Olšák wrote: >> >>> Ok sounds good. >>> >>> Marek >>> >>> On Wed., May 11, 2022, 03:43 Christian König, < >>> ckoenig.leichtzumer...@gmail.com> wrote: >>> It really *is* a NOALLOC feature. In other words there is no latency improvement on reads because the cache is always checked, even with the noalloc flag set. The only thing it affects is that misses not enter the cache and so don't cause any additional pressure on evicting cache lines. You might want to double check with the hardware guys, but I'm something like 95% sure that it works this way. Christian. Am 11.05.22 um 09:22 schrieb Marek Olšák: Bypass means that the contents of the cache are ignored, which decreases latency at the cost of no coherency between bypassed and normal memory requests. NOA (noalloc) means that the cache is checked and can give you cache hits, but misses are not cached and the overall latency is higher. I don't know what the hw does, but I hope it was misnamed and it really means bypass because there is no point in doing cache lookups on every memory request if the driver wants to disable caching to *decrease* latency in the situations when the cache isn't helping. Marek On Wed, May 11, 2022 at 2:15 AM Lazar, Lijo wrote: > > > On 5/11/2022 11:36 AM, Christian König wrote: > > Mhm, it doesn't really bypass MALL. It just doesn't allocate any > MALL > > entries on write. > > > > How about AMDGPU_VM_PAGE_NO_MALL ? > > One more - AMDGPU_VM_PAGE_LLC_* [ LLC = last level cache, * = some > sort > of attribute which decides LLC behaviour] > > Thanks, > Lijo > > > > > Christian. > > > > Am 10.05.22 um 23:21 schrieb Marek Olšák: > >> A better name would be: > >> AMDGPU_VM_PAGE_BYPASS_MALL > >> > >> Marek > >> > >> On Fri, May 6, 2022 at 7:23 AM Christian König > >> wrote: > >> > >> Add the AMDGPU_VM_NOALLOC flag to let userspace control MALL > >> allocation. > >> > >> Only compile tested! > >> > >> Signed-off-by: Christian König > >> --- > >> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++ > >> drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +++ > >> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 3 +++ > >> include/uapi/drm/amdgpu_drm.h | 2 ++ > >> 4 files changed, 10 insertions(+) > >> > >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c > >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c > >> index bf97d8f07f57..d8129626581f 100644 > >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c > >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c > >> @@ -650,6 +650,8 @@ uint64_t amdgpu_gem_va_map_flags(struct > >> amdgpu_device *adev, uint32_t flags) > >> pte_flag |= AMDGPU_PTE_WRITEABLE; > >> if (flags & AMDGPU_VM_PAGE_PRT) > >> pte_flag |= AMDGPU_PTE_PRT; > >> + if (flags & AMDGPU_VM_PAGE_NOALLOC) > >> + pte_flag |= AMDGPU_PTE_NOALLOC; > >> > >> if (adev->gmc.gmc_funcs->map_mtype) > >> pte_flag |= amdgpu_gmc_map_mtype(adev, > >> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c > >> b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c > >> index b8c79789e1e4..9077dfccaf3c 100644 > >> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c > >> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c > >> @@ -613,6 +613,9 @@ static void gmc_v10_0_get_vm_pte(struct > >> amdgpu_device *adev, > >> *flags &= ~AMDGPU_PTE_MTYPE_NV10_MASK; > >> *flags |= (mapping->flags & AMDGPU_PTE_MTYPE_NV10_MASK); > >> > >> + *flags &= ~AMDGPU_PTE_NOALLOC; > >> + *flags |= (mapping->flags & AMDGPU_PTE_NOALLOC); > >> + > >> if (mapping->flags & AMDGPU_PTE_PRT) { > >> *flags |= AMDGPU_PTE_PRT; > >> *flags |= AMDGPU_PTE_SNOOPED; > >> diff --git a/drivers/gpu/drm/amd/amdgpu
Re: [PATCH v1 14/15] tools: add hmm gup tests for device coherent type
Alex Sierra writes: > The intention is to test hmm device coherent type under different get > user pages paths. Also, test gup with FOLL_LONGTERM flag set in > device coherent pages. These pages should get migrated back to system > memory. > > Signed-off-by: Alex Sierra > --- > tools/testing/selftests/vm/hmm-tests.c | 104 + > 1 file changed, 104 insertions(+) > > diff --git a/tools/testing/selftests/vm/hmm-tests.c > b/tools/testing/selftests/vm/hmm-tests.c > index 84ec8c4a1dc7..65e30ab6494c 100644 > --- a/tools/testing/selftests/vm/hmm-tests.c > +++ b/tools/testing/selftests/vm/hmm-tests.c > @@ -36,6 +36,7 @@ > * in the usual include/uapi/... directory. > */ > #include "../../../../lib/test_hmm_uapi.h" > +#include "../../../../mm/gup_test.h" > > struct hmm_buffer { > void*ptr; > @@ -60,6 +61,8 @@ enum { > #define NTIMES 10 > > #define ALIGN(x, a) (((x) + (a - 1)) & (~((a) - 1))) > +/* Just the flags we need, copied from mm.h: */ > +#define FOLL_WRITE 0x01/* check pte is writable */ > > FIXTURE(hmm) > { > @@ -1766,4 +1769,105 @@ TEST_F(hmm, exclusive_cow) > hmm_buffer_free(buffer); > } > > +static int gup_test_exec(int gup_fd, unsigned long addr, > + int cmd, int npages, int size) > +{ > + struct gup_test gup = { > + .nr_pages_per_call = npages, > + .addr = addr, > + .gup_flags = FOLL_WRITE, > + .size = size, > + }; > + > + if (ioctl(gup_fd, cmd, &gup)) { > + perror("ioctl on error\n"); > + return errno; > + } > + > + return 0; > +} > + > +/* > + * Test get user device pages through gup_test. Setting PIN_LONGTERM flag. > + * This should trigger a migration back to system memory for both, private > + * and coherent type pages. > + * This test makes use of gup_test module. Make sure GUP_TEST_CONFIG is added > + * to your configuration before you run it. > + */ > +TEST_F(hmm, hmm_gup_test) > +{ > + struct hmm_buffer *buffer; > + int gup_fd; > + unsigned long npages; > + unsigned long size; > + unsigned long i; > + int *ptr; > + int ret; > + unsigned char *m; > + > + gup_fd = open("/sys/kernel/debug/gup_test", O_RDWR); > + if (gup_fd == -1) > + SKIP(return, "Skipping test, could not find gup_test driver"); > + > + npages = 3; > + size = npages << self->page_shift; > + > + buffer = malloc(sizeof(*buffer)); > + ASSERT_NE(buffer, NULL); > + > + buffer->fd = -1; > + buffer->size = size; > + buffer->mirror = malloc(size); > + ASSERT_NE(buffer->mirror, NULL); > + > + buffer->ptr = mmap(NULL, size, > +PROT_READ | PROT_WRITE, > +MAP_PRIVATE | MAP_ANONYMOUS, > +buffer->fd, 0); > + ASSERT_NE(buffer->ptr, MAP_FAILED); > + > + /* Initialize buffer in system memory. */ > + for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i) > + ptr[i] = i; > + > + /* Migrate memory to device. */ > + ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages); > + ASSERT_EQ(ret, 0); > + ASSERT_EQ(buffer->cpages, npages); > + /* Check what the device read. */ > + for (i = 0, ptr = buffer->mirror; i < size / sizeof(*ptr); ++i) > + ASSERT_EQ(ptr[i], i); > + > + ASSERT_EQ(gup_test_exec(gup_fd, > + (unsigned long)buffer->ptr, > + GUP_BASIC_TEST, 1, self->page_size), 0); > + ASSERT_EQ(gup_test_exec(gup_fd, > + (unsigned long)buffer->ptr + 1 * > self->page_size, > + GUP_FAST_BENCHMARK, 1, self->page_size), 0); > + ASSERT_EQ(gup_test_exec(gup_fd, > + (unsigned long)buffer->ptr + 2 * > self->page_size, > + PIN_LONGTERM_BENCHMARK, 1, self->page_size), 0); > + > + /* Take snapshot to CPU pagetables */ > + ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_SNAPSHOT, buffer, npages); > + ASSERT_EQ(ret, 0); > + ASSERT_EQ(buffer->cpages, npages); > + m = buffer->mirror; > + if (hmm_is_coherent_type(variant->device_number)) { > + ASSERT_EQ(HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL | > HMM_DMIRROR_PROT_WRITE, m[0]); > + ASSERT_EQ(HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL | > HMM_DMIRROR_PROT_WRITE, m[1]); > + } else { > + ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[0]); > + ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[1]); > + } > + ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[2]); > + /* Check again the content on the pages. Make sure there's no Very minor nit-pick, but if you happen to be re-spinning anyway the comment formatting needs fixing and a new line before the comment would be nice. Otherwise: Reviewed-by: Alistair Popple > + * corrupted data. > +
Re: [PATCH v12] drm/amdgpu: add drm buddy support to amdgpu
Hi The merge window for 5.19 will probably be opening next week, has there been any progress with this bug? Thanks Mike On Mon, 2 May 2022 at 17:31, Mike Lothian wrote: > > On Mon, 2 May 2022 at 16:54, Arunpravin Paneer Selvam > wrote: > > > > > > > > On 5/2/2022 8:41 PM, Mike Lothian wrote: > > > On Wed, 27 Apr 2022 at 12:55, Mike Lothian wrote: > > >> On Tue, 26 Apr 2022 at 17:36, Christian König > > >> wrote: > > >>> Hi Mike, > > >>> > > >>> sounds like somehow stitching together the SG table for PRIME doesn't > > >>> work any more with this patch. > > >>> > > >>> Can you try with P2P DMA disabled? > > >> -CONFIG_PCI_P2PDMA=y > > >> +# CONFIG_PCI_P2PDMA is not set > > >> > > >> If that's what you're meaning, then there's no difference, I'll upload > > >> my dmesg to the gitlab issue > > >> > > >>> Apart from that can you take a look Arun? > > >>> > > >>> Thanks, > > >>> Christian. > > > Hi > > > > > > Have you had any success in replicating this? > > Hi Mike, > > I couldn't replicate on my Raven APU machine. I see you have 2 cards > > initialized, one is Renoir > > and the other is Navy Flounder. Could you give some more details, are > > you running Gravity Mark > > on Renoir and what is your system RAM configuration? > > > > > > Cheers > > > > > > Mike > > > Hi > > It's a PRIME laptop, it failed on the RENOIR too, it caused a lockup, > but systemd managed to capture it, I'll attach it to the issue > > I've got 64GB RAM, the 6800M has 12GB VRAM > > Cheers > > Mike
[PATCH] drm/amdgpu: Clean up of initializing doorbells for gfx_v9 and gfx_v10
From: Haohui Mai Clean up redundant, copy-paste code blocks during the intialization of the doorbells in mqd_init(). Signed-off-by: Haohui Mai --- drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 17 - drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 17 - 2 files changed, 34 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c index 4b66b9c93754..12f63d6aab12 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c @@ -6919,23 +6919,6 @@ static int gfx_v10_0_compute_mqd_init(struct amdgpu_device *adev, void *m, mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffc; mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0x; - tmp = 0; - /* enable the doorbell if requested */ - if (prop->use_doorbell) { - tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL); - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, - DOORBELL_OFFSET, prop->doorbell_index); - - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, - DOORBELL_EN, 1); - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, - DOORBELL_SOURCE, 0); - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, - DOORBELL_HIT, 0); - } - - mqd->cp_hqd_pq_doorbell_control = tmp; - /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 83639b5ea6a9..f49a2dd89ee7 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -3535,23 +3535,6 @@ static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring) mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffc; mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0x; - tmp = 0; - /* enable the doorbell if requested */ - if (ring->use_doorbell) { - tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL); - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, - DOORBELL_OFFSET, ring->doorbell_index); - - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, -DOORBELL_EN, 1); - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, -DOORBELL_SOURCE, 0); - tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, -DOORBELL_HIT, 0); - } - - mqd->cp_hqd_pq_doorbell_control = tmp; - /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ ring->wptr = 0; mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR); -- 2.25.1
Re: [PATCH 2/3] drm/amdgpu: add AMDGPU_VM_NOALLOC
Crap, do you have a link to the failure? Am 16.05.22 um 13:10 schrieb Marek Olšák: I forgot to say: The NOALLOC flag causes an allocation failure, so there is a kernel bug somewhere. Marek On Mon, May 16, 2022 at 7:06 AM Marek Olšák wrote: FYI, I think it's time to merge this because the Mesa commits are going to be merged in ~30 minutes if Gitlab CI is green, and that includes updated amdgpu_drm.h. Marek On Wed, May 11, 2022 at 2:55 PM Marek Olšák wrote: Ok sounds good. Marek On Wed., May 11, 2022, 03:43 Christian König, wrote: It really *is* a NOALLOC feature. In other words there is no latency improvement on reads because the cache is always checked, even with the noalloc flag set. The only thing it affects is that misses not enter the cache and so don't cause any additional pressure on evicting cache lines. You might want to double check with the hardware guys, but I'm something like 95% sure that it works this way. Christian. Am 11.05.22 um 09:22 schrieb Marek Olšák: Bypass means that the contents of the cache are ignored, which decreases latency at the cost of no coherency between bypassed and normal memory requests. NOA (noalloc) means that the cache is checked and can give you cache hits, but misses are not cached and the overall latency is higher. I don't know what the hw does, but I hope it was misnamed and it really means bypass because there is no point in doing cache lookups on every memory request if the driver wants to disable caching to *decrease* latency in the situations when the cache isn't helping. Marek On Wed, May 11, 2022 at 2:15 AM Lazar, Lijo wrote: On 5/11/2022 11:36 AM, Christian König wrote: > Mhm, it doesn't really bypass MALL. It just doesn't allocate any MALL > entries on write. > > How about AMDGPU_VM_PAGE_NO_MALL ? One more - AMDGPU_VM_PAGE_LLC_* [ LLC = last level cache, * = some sort of attribute which decides LLC behaviour] Thanks, Lijo > > Christian. > > Am 10.05.22 um 23:21 schrieb Marek Olšák: >> A better name would be: >> AMDGPU_VM_PAGE_BYPASS_MALL >> >> Marek >> >> On Fri, May 6, 2022 at 7:23 AM Christian König >> wrote: >> >> Add the AMDGPU_VM_NOALLOC flag to let userspace control MALL >> allocation. >> >> Only compile tested! >> >> Signed-off-by: Christian König >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++ >> drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +++ >> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 3 +++ >> include/uapi/drm/amdgpu_drm.h | 2 ++ >> 4 files changed, 10 insertions(+) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> index bf97d8f07f57..d8129626581f 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> @@ -650,6 +650,8 @@ uint64_t amdgpu_gem_va_map_flags(struct >> amdgpu_device *adev, uint32_t flags) >> pte_flag |= AMDGPU_PTE_WRITEABLE; >> if (flags & AMDGPU_VM_PAGE_PRT) >> pte_flag |= AMDGPU_PTE_PRT; >> + if (flags & AMDGPU_VM_PAGE_NOALLOC) >> + pte_flag |= AMDGPU_PTE_NOALLOC; >> >> if (adev->gmc.gmc_funcs->map_mtype) >> pte_flag |= amdgpu_gmc_map_mtype(adev, >> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> index b8c79789e1e4..9077dfccaf3c 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> @@ -613,6 +613,9 @@ static void gmc_v10_0_get_vm_pte(struct
Re: [PATCH 2/3] drm/amdgpu: add AMDGPU_VM_NOALLOC
I forgot to say: The NOALLOC flag causes an allocation failure, so there is a kernel bug somewhere. Marek On Mon, May 16, 2022 at 7:06 AM Marek Olšák wrote: > FYI, I think it's time to merge this because the Mesa commits are going to > be merged in ~30 minutes if Gitlab CI is green, and that includes updated > amdgpu_drm.h. > > Marek > > On Wed, May 11, 2022 at 2:55 PM Marek Olšák wrote: > >> Ok sounds good. >> >> Marek >> >> On Wed., May 11, 2022, 03:43 Christian König, < >> ckoenig.leichtzumer...@gmail.com> wrote: >> >>> It really *is* a NOALLOC feature. In other words there is no latency >>> improvement on reads because the cache is always checked, even with the >>> noalloc flag set. >>> >>> The only thing it affects is that misses not enter the cache and so >>> don't cause any additional pressure on evicting cache lines. >>> >>> You might want to double check with the hardware guys, but I'm something >>> like 95% sure that it works this way. >>> >>> Christian. >>> >>> Am 11.05.22 um 09:22 schrieb Marek Olšák: >>> >>> Bypass means that the contents of the cache are ignored, which decreases >>> latency at the cost of no coherency between bypassed and normal memory >>> requests. NOA (noalloc) means that the cache is checked and can give you >>> cache hits, but misses are not cached and the overall latency is higher. I >>> don't know what the hw does, but I hope it was misnamed and it really means >>> bypass because there is no point in doing cache lookups on every memory >>> request if the driver wants to disable caching to *decrease* latency in the >>> situations when the cache isn't helping. >>> >>> Marek >>> >>> On Wed, May 11, 2022 at 2:15 AM Lazar, Lijo wrote: >>> On 5/11/2022 11:36 AM, Christian König wrote: > Mhm, it doesn't really bypass MALL. It just doesn't allocate any MALL > entries on write. > > How about AMDGPU_VM_PAGE_NO_MALL ? One more - AMDGPU_VM_PAGE_LLC_* [ LLC = last level cache, * = some sort of attribute which decides LLC behaviour] Thanks, Lijo > > Christian. > > Am 10.05.22 um 23:21 schrieb Marek Olšák: >> A better name would be: >> AMDGPU_VM_PAGE_BYPASS_MALL >> >> Marek >> >> On Fri, May 6, 2022 at 7:23 AM Christian König >> wrote: >> >> Add the AMDGPU_VM_NOALLOC flag to let userspace control MALL >> allocation. >> >> Only compile tested! >> >> Signed-off-by: Christian König >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++ >> drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +++ >> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 3 +++ >> include/uapi/drm/amdgpu_drm.h | 2 ++ >> 4 files changed, 10 insertions(+) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> index bf97d8f07f57..d8129626581f 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >> @@ -650,6 +650,8 @@ uint64_t amdgpu_gem_va_map_flags(struct >> amdgpu_device *adev, uint32_t flags) >> pte_flag |= AMDGPU_PTE_WRITEABLE; >> if (flags & AMDGPU_VM_PAGE_PRT) >> pte_flag |= AMDGPU_PTE_PRT; >> + if (flags & AMDGPU_VM_PAGE_NOALLOC) >> + pte_flag |= AMDGPU_PTE_NOALLOC; >> >> if (adev->gmc.gmc_funcs->map_mtype) >> pte_flag |= amdgpu_gmc_map_mtype(adev, >> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> index b8c79789e1e4..9077dfccaf3c 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >> @@ -613,6 +613,9 @@ static void gmc_v10_0_get_vm_pte(struct >> amdgpu_device *adev, >> *flags &= ~AMDGPU_PTE_MTYPE_NV10_MASK; >> *flags |= (mapping->flags & AMDGPU_PTE_MTYPE_NV10_MASK); >> >> + *flags &= ~AMDGPU_PTE_NOALLOC; >> + *flags |= (mapping->flags & AMDGPU_PTE_NOALLOC); >> + >> if (mapping->flags & AMDGPU_PTE_PRT) { >> *flags |= AMDGPU_PTE_PRT; >> *flags |= AMDGPU_PTE_SNOOPED; >> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >> b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >> index 8d733eeac556..32ee56adb602 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >> @@ -508,6 +508,9 @@ static void gmc_v11_0_get_vm_pte(struct >> amdgpu_device *adev, >> *flags &= ~AMDGPU_PTE_MTYPE_NV10_MASK; >> *flags |= (mapping->flags & AMD
Re: [PATCH 2/3] drm/amdgpu: add AMDGPU_VM_NOALLOC
FYI, I think it's time to merge this because the Mesa commits are going to be merged in ~30 minutes if Gitlab CI is green, and that includes updated amdgpu_drm.h. Marek On Wed, May 11, 2022 at 2:55 PM Marek Olšák wrote: > Ok sounds good. > > Marek > > On Wed., May 11, 2022, 03:43 Christian König, < > ckoenig.leichtzumer...@gmail.com> wrote: > >> It really *is* a NOALLOC feature. In other words there is no latency >> improvement on reads because the cache is always checked, even with the >> noalloc flag set. >> >> The only thing it affects is that misses not enter the cache and so don't >> cause any additional pressure on evicting cache lines. >> >> You might want to double check with the hardware guys, but I'm something >> like 95% sure that it works this way. >> >> Christian. >> >> Am 11.05.22 um 09:22 schrieb Marek Olšák: >> >> Bypass means that the contents of the cache are ignored, which decreases >> latency at the cost of no coherency between bypassed and normal memory >> requests. NOA (noalloc) means that the cache is checked and can give you >> cache hits, but misses are not cached and the overall latency is higher. I >> don't know what the hw does, but I hope it was misnamed and it really means >> bypass because there is no point in doing cache lookups on every memory >> request if the driver wants to disable caching to *decrease* latency in the >> situations when the cache isn't helping. >> >> Marek >> >> On Wed, May 11, 2022 at 2:15 AM Lazar, Lijo wrote: >> >>> >>> >>> On 5/11/2022 11:36 AM, Christian König wrote: >>> > Mhm, it doesn't really bypass MALL. It just doesn't allocate any MALL >>> > entries on write. >>> > >>> > How about AMDGPU_VM_PAGE_NO_MALL ? >>> >>> One more - AMDGPU_VM_PAGE_LLC_* [ LLC = last level cache, * = some sort >>> of attribute which decides LLC behaviour] >>> >>> Thanks, >>> Lijo >>> >>> > >>> > Christian. >>> > >>> > Am 10.05.22 um 23:21 schrieb Marek Olšák: >>> >> A better name would be: >>> >> AMDGPU_VM_PAGE_BYPASS_MALL >>> >> >>> >> Marek >>> >> >>> >> On Fri, May 6, 2022 at 7:23 AM Christian König >>> >> wrote: >>> >> >>> >> Add the AMDGPU_VM_NOALLOC flag to let userspace control MALL >>> >> allocation. >>> >> >>> >> Only compile tested! >>> >> >>> >> Signed-off-by: Christian König >>> >> --- >>> >> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++ >>> >> drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 +++ >>> >> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 3 +++ >>> >> include/uapi/drm/amdgpu_drm.h | 2 ++ >>> >> 4 files changed, 10 insertions(+) >>> >> >>> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >>> >> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >>> >> index bf97d8f07f57..d8129626581f 100644 >>> >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >>> >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c >>> >> @@ -650,6 +650,8 @@ uint64_t amdgpu_gem_va_map_flags(struct >>> >> amdgpu_device *adev, uint32_t flags) >>> >> pte_flag |= AMDGPU_PTE_WRITEABLE; >>> >> if (flags & AMDGPU_VM_PAGE_PRT) >>> >> pte_flag |= AMDGPU_PTE_PRT; >>> >> + if (flags & AMDGPU_VM_PAGE_NOALLOC) >>> >> + pte_flag |= AMDGPU_PTE_NOALLOC; >>> >> >>> >> if (adev->gmc.gmc_funcs->map_mtype) >>> >> pte_flag |= amdgpu_gmc_map_mtype(adev, >>> >> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >>> >> b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >>> >> index b8c79789e1e4..9077dfccaf3c 100644 >>> >> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >>> >> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c >>> >> @@ -613,6 +613,9 @@ static void gmc_v10_0_get_vm_pte(struct >>> >> amdgpu_device *adev, >>> >> *flags &= ~AMDGPU_PTE_MTYPE_NV10_MASK; >>> >> *flags |= (mapping->flags & AMDGPU_PTE_MTYPE_NV10_MASK); >>> >> >>> >> + *flags &= ~AMDGPU_PTE_NOALLOC; >>> >> + *flags |= (mapping->flags & AMDGPU_PTE_NOALLOC); >>> >> + >>> >> if (mapping->flags & AMDGPU_PTE_PRT) { >>> >> *flags |= AMDGPU_PTE_PRT; >>> >> *flags |= AMDGPU_PTE_SNOOPED; >>> >> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >>> >> b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >>> >> index 8d733eeac556..32ee56adb602 100644 >>> >> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >>> >> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c >>> >> @@ -508,6 +508,9 @@ static void gmc_v11_0_get_vm_pte(struct >>> >> amdgpu_device *adev, >>> >> *flags &= ~AMDGPU_PTE_MTYPE_NV10_MASK; >>> >> *flags |= (mapping->flags & AMDGPU_PTE_MTYPE_NV10_MASK); >>> >> >>> >> + *flags &= ~AMDGPU_PTE_NOALLOC; >>> >> + *flags |= (mapping->flags & AMDGPU_PTE_NOALLOC); >>> >> + >>> >> if (mapping->flags & AMDGPU_PTE_PRT) { >>> >> *flags |= AMDGPU_PTE_PRT; >>> >>
RE: [Intel-gfx] [V2 2/3] drm/i915/display/debug: Expose crtc current bpc via debugfs
On Tue, 12 Apr 2022, "Murthy, Arun R" wrote: >> -Original Message- >> From: Intel-gfx On Behalf Of >> Bhanuprakash Modem >> Sent: Monday, April 11, 2022 3:21 PM >> To: intel-...@lists.freedesktop.org; dri-de...@lists.freedesktop.org; amd- >> g...@lists.freedesktop.org; jani.nik...@linux.intel.com; >> ville.syrj...@linux.intel.com; harry.wentl...@amd.com; Sharma, Swati2 >> >> Subject: [Intel-gfx] [V2 2/3] drm/i915/display/debug: Expose crtc current bpc >> via debugfs >> >> This new debugfs will expose the currently using bpc by crtc. >> It is very useful for verifying whether we enter the correct output color >> depth >> from IGT. >> >> This patch will also add the connector's max supported bpc to >> "i915_display_info" debugfs. >> >> Example: >> cat /sys/kernel/debug/dri/0/crtc-0/i915_current_bpc >> Current: 8 >> >> Cc: Jani Nikula >> Cc: Ville Syrjälä >> Cc: Uma Shankar >> Signed-off-by: Bhanuprakash Modem > > Reviewed-by: Arun R Murthy Acked-by: Jani Nikula for merging this via drm-misc-next. -- Jani Nikula, Intel Open Source Graphics Center
Re: [Intel-gfx] [V2 3/3] drm/amd/display: Move connector debugfs to drm
On Mon, 02 May 2022, Harry Wentland wrote: > Both the kernel and IGT series look good to me. > > I recommend you merge the entire kernel set as one into drm-next. We > can pull it into amd-staging-drm-next so as not to break our CI once > the IGT patches land. Can we read that as an ack to merge via drm-misc-next? :) BR, Jani. -- Jani Nikula, Intel Open Source Graphics Center
[PATCH] drm/amdgpu: Off by one in dm_dmub_outbox1_low_irq()
The > ARRAY_SIZE() should be >= ARRAY_SIZE() to prevent an out of bounds access. Fixes: e27c41d5b068 ("drm/amd/display: Support for DMUB HPD interrupt handling") Signed-off-by: Dan Carpenter --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c index 6edfc0963dab..c4f122c6faec 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -770,7 +770,7 @@ static void dm_dmub_outbox1_low_irq(void *interrupt_params) do { dc_stat_get_dmub_notification(adev->dm.dc, ¬ify); - if (notify.type > ARRAY_SIZE(dm->dmub_thread_offload)) { + if (notify.type >= ARRAY_SIZE(dm->dmub_thread_offload)) { DRM_ERROR("DM: notify type %d invalid!", notify.type); continue; } -- 2.35.1