On Thu, Aug 21, 2025 at 11:28 AM Shaoyun Liu <shaoyun....@amd.com> wrote:
>
> From MES version 0x81, it provide the new API INV_TLBS that support
> invalidate tlbs with PASID.
>
> Signed-off-by: Shaoyun Liu <shaoyun....@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 27 ++++++++++++++++
> drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c | 20 ++++++++++++
> drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 43 +++++++++++++++++++++++++
> 3 files changed, 90 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
> index c0d2c195fe2e..46235b8726f3 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
> @@ -280,6 +280,30 @@ struct mes_reset_queue_input {
> bool is_kq;
> };
>
> +enum amdgpu_mes_hub_id {
> + AMDGPU_MES_GC_HUB0 = 0,
> + AMDGPU_MES_GC_HUB1 = 1,
> + AMDGPU_MES_GC_HUB2 = 2,
> + AMDGPU_MES_GC_HUB3 = 3,
> + AMDGPU_MES_GC_HUB4 = 4,
> + AMDGPU_MES_GC_HUB5 = 5,
> + AMDGPU_MES_GC_HUB6 = 6,
> + AMDGPU_MES_GC_HUB7 = 7,
> + AMDGPU_MES_MM_HUB00 = 8,
> + AMDGPU_MES_MM_HUB01 = 9,
> + AMDGPU_MES_MM_HUB02= 10,
> + AMDGPU_MES_MM_HUB03 = 11,
> + AMDGPU_MES_MM_HUB10 = 12,
> + AMDGPU_MES_MM_HUB11 = 13,
> +};
If these are the same as the hub definitions in amdgpu_vm.h you can
just use those directly and skip these.
> +
> +struct mes_inv_tlbs_pasid_input {
> + uint32_t xcc_id;
> + uint16_t pasid;
> + uint8_t hub_id;
> + uint8_t flush_type;
> +};
> +
> enum mes_misc_opcode {
> MES_MISC_OP_WRITE_REG,
> MES_MISC_OP_READ_REG,
> @@ -367,6 +391,9 @@ struct amdgpu_mes_funcs {
>
> int (*reset_hw_queue)(struct amdgpu_mes *mes,
> struct mes_reset_queue_input *input);
> +
> + int (*invalidate_tlbs_pasid)(struct amdgpu_mes *mes,
> + struct mes_inv_tlbs_pasid_input *input);
> };
>
> #define amdgpu_mes_kiq_hw_init(adev) (adev)->mes.kiq_hw_init((adev))
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c
> b/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c
> index feb92e107af8..b5be6c7838aa 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c
> @@ -339,6 +339,26 @@ static void gmc_v12_0_flush_gpu_tlb_pasid(struct
> amdgpu_device *adev,
> uint16_t queried;
> int vmid, i;
>
> + if (adev->enable_uni_mes &&
> adev->mes.ring[AMDGPU_MES_SCHED_PIPE].sched.ready &&
> + (adev->mes.sched_version & AMDGPU_MES_VERSION_MASK) >= 0x81) {
> + struct mes_inv_tlbs_pasid_input input = {0};
> + input.pasid = pasid;
> + input.flush_type = flush_type;
> + input.hub_id = AMDGPU_GFXHUB(0);
> + /* MES will invalidate all gc_hub for the device from master
> */
> + adev->mes.funcs->invalidate_tlbs_pasid(&adev->mes, &input);
> + if (all_hub) {
> + /* Only need to invalidate mm_hub now */
> + for_each_set_bit(i, adev->vmhubs_mask,
> AMDGPU_MAX_VMHUBS) {
> + if (i < AMDGPU_MMHUB0_START)
> + continue;
I think you can drop the loop here and just have:
input.hub_id = AMDGPU_MMHUB(0);
adev->mes.funcs->invalidate_tlbs_pasid(&adev->mes, &input);
in the all_hub case since gfx12 only supports one mmhub as well.
> + input.hub_id = i;
> +
> adev->mes.funcs->invalidate_tlbs_pasid(&adev->mes, &input);
> + }
> + }
> + return;
> + }
> +
> for (vmid = 1; vmid < 16; vmid++) {
> bool valid;
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
> b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
> index 6b222630f3fa..6740383f7721 100644
> --- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
> @@ -108,6 +108,7 @@ static const char *mes_v12_0_opcodes[] = {
> "SET_SE_MODE",
> "SET_GANG_SUBMIT",
> "SET_HW_RSRC_1",
> + "INVALIDATE_TLBS",
> };
>
> static const char *mes_v12_0_misc_opcodes[] = {
> @@ -879,6 +880,47 @@ static int mes_v12_0_reset_hw_queue(struct amdgpu_mes
> *mes,
> offsetof(union MESAPI__RESET, api_status));
> }
>
> +static int mes_v12_inv_tlb_convert_hub_id(enum amdgpu_mes_hub_id id)
> +{
> + int mes_hub_id = id;
> + /*
> + * MES doesn't support invalidate gc_hub on slave xcc individually
> + * master xcc will invalidate all gc_hub for the partition
> + */
> + if (id == 0)
> + return 0;
> + if (id < AMDGPU_MMHUB0_START)
> + return -EINVAL;
> +
> + mes_hub_id -= AMDGPU_MMHUB0_START - 1;
> + /* gfx12 only support maximum one mmhub */
> + return (mes_hub_id > 1) ? -EINVAL: mes_hub_id;
It would be cleaner to do something like this:
if (AMDGPU_IS_GFXHUB(id))
return 0;
else if (AMDGPU_IS_MMHUB0(id))
return 1;
else
return -EINVAL;
> +}
add a new line here.
Alex
> +static int mes_v12_0_inv_tlbs_pasid(struct amdgpu_mes *mes,
> + struct mes_inv_tlbs_pasid_input *input)
> +{
> + union MESAPI__INV_TLBS mes_inv_tlbs;
> +
> + memset(&mes_inv_tlbs, 0, sizeof(mes_inv_tlbs));
> +
> + mes_inv_tlbs.header.type = MES_API_TYPE_SCHEDULER;
> + mes_inv_tlbs.header.opcode = MES_SCH_API_INV_TLBS;
> + mes_inv_tlbs.header.dwsize = API_FRAME_SIZE_IN_DWORDS;
> +
> + mes_inv_tlbs.invalidate_tlbs.inv_sel = 0;
> + mes_inv_tlbs.invalidate_tlbs.flush_type = input->flush_type;
> + mes_inv_tlbs.invalidate_tlbs.inv_sel_id = input->pasid;
> +
> + /*convert amdgpu_mes_hub_id to mes expected hub_id */
> + mes_inv_tlbs.invalidate_tlbs.hub_id =
> mes_v12_inv_tlb_convert_hub_id(input->hub_id);
> + if (mes_inv_tlbs.invalidate_tlbs.hub_id < 0)
> + return -EINVAL;
> + return mes_v12_0_submit_pkt_and_poll_completion(mes,
> AMDGPU_MES_KIQ_PIPE,
> + &mes_inv_tlbs, sizeof(mes_inv_tlbs),
> + offsetof(union MESAPI__INV_TLBS, api_status));
> +
> +}
> +
> static const struct amdgpu_mes_funcs mes_v12_0_funcs = {
> .add_hw_queue = mes_v12_0_add_hw_queue,
> .remove_hw_queue = mes_v12_0_remove_hw_queue,
> @@ -888,6 +930,7 @@ static const struct amdgpu_mes_funcs mes_v12_0_funcs = {
> .resume_gang = mes_v12_0_resume_gang,
> .misc_op = mes_v12_0_misc_op,
> .reset_hw_queue = mes_v12_0_reset_hw_queue,
> + .invalidate_tlbs_pasid = mes_v12_0_inv_tlbs_pasid,
> };
>
> static int mes_v12_0_allocate_ucode_buffer(struct amdgpu_device *adev,
> --
> 2.34.1
>
