On 2025-03-04 22:54, Emily Deng wrote:
While the entry get in svm_range_unmap_from_cpu is the last entry, and the entry is page fault, it also need to be dropped. So for equal case, it also need to be dropped. v2: Only modify the svm_range_restore_pages. Signed-off-by: Emily Deng <Emily.Deng@xxxxxxx> --- drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h | 3 +++ drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 2 +- 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h index 7d4395a5d8ac..b0a88f92cd82 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h @@ -78,6 +78,9 @@ struct amdgpu_ih_ring { #define amdgpu_ih_ts_after(t1, t2) \ (((int64_t)((t2) << 16) - (int64_t)((t1) << 16)) > 0LL)+#define amdgpu_ih_ts_after_or_equal(t1, t2) \+ (((int64_t)((t2) << 16) - (int64_t)((t1) << 16)) >= 0LL) + /* provided by the ih block */ struct amdgpu_ih_funcs { /* ring read/write ptr handling, called from interrupt context */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index bd3e20d981e0..d04725583f19 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -3010,7 +3010,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,/* check if this page fault time stamp is before svms->checkpoint_ts */if (svms->checkpoint_ts[gpuidx] != 0) { - if (amdgpu_ih_ts_after(ts, svms->checkpoint_ts[gpuidx])) { + if (amdgpu_ih_ts_after_or_equal(ts, svms->checkpoint_ts[gpuidx])) { pr_debug("draining retry fault, drop fault 0x%llx\n", addr); r = 0;
Looks reasonable. Should we also set r = -EAGAIN, so that the next fault for this address doesn't get rejected by the SW filter? See the code at the end of svm_range_restore_pages that handles -EAGAIN and calls amdgpu_gmc_filter_faults_remove. I guess that only matters on chips where the CAM is not enabled.
Regards, Felix
goto out;
