Re: [PATCH 5/5] drm/amdgpu: skip GFX FED error in page fault handling

[Christian König]

Am 19.02.24 um 09:15 schrieb Tao Zhou:
Let kfd interrupt handler process it.
Signed-off-by: Tao Zhou <tao.zh...@amd.com>
---
  drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 10 +++++++++-
  1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 773725a92cf1..70defc394b7b 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -552,7 +552,7 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device 
*adev,
  {
        bool retry_fault = !!(entry->src_data[1] & 0x80);
        bool write_fault = !!(entry->src_data[1] & 0x20);
-       uint32_t status = 0, cid = 0, rw = 0;
+       uint32_t status = 0, cid = 0, rw = 0, fed = 0;
        struct amdgpu_task_info task_info;
        struct amdgpu_vmhub *hub;
        const char *mmhub_cid;
@@ -663,6 +663,14 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device 
*adev,
        status = RREG32(hub->vm_l2_pro_fault_status);
        cid = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, CID);
        rw = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, RW);
+       fed = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, FED);
+
+       /* for gfx fed error, kfd will handle it, return directly */
+       if (fed && amdgpu_ras_is_poison_mode_supported(adev) &&
+           amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(9, 4, 2) &&
+           !strcmp(hub_name, "gfxhub0"))

Please never ever use strcmp() to make a decision like that, 
*especially* not in an interrupt handler.

Regards,
Christian.

+               return 1;
+
        WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
  #ifdef HAVE_STRUCT_XARRAY
        amdgpu_vm_update_fault_cache(adev, entry->pasid, addr, status, vmhub);