bybrooklyn/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-04-18 14:53:58 -04:00
Code Issues Packages Projects Releases Wiki Activity

drm/amdgpu: fix GDS/GWS/OA switch handling

Browse Source 
Bas pointed out that this isn't working as expected and could cause
crashes. Fix the handling by storing the marker that a switch is needed
inside the job instead.

Reported-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Christian König
2022-11-25 16:04:25 +01:00
committed by Alex Deucher
parent e0607c10eb
commit 56b0989e29
3 changed files with 53 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
View File

@@ -165,6 +165,26 @@ bool amdgpu_vmid_had_gpu_reset(struct amdgpu_device *adev,
atomic_read(&adev->gpu_reset_counter);
}
/* Check if we need to switch to another set of resources */
static bool amdgpu_vmid_gds_switch_needed(struct amdgpu_vmid *id,
struct amdgpu_job *job)
{
return id->gds_base != job->gds_base ||
id->gds_size != job->gds_size ||
id->gws_base != job->gws_base ||
id->gws_size != job->gws_size ||
id->oa_base != job->oa_base ||
id->oa_size != job->oa_size;
}
/* Check if the id is compatible with the job */
static bool amdgpu_vmid_compatible(struct amdgpu_vmid *id,
struct amdgpu_job *job)
{
return id->pd_gpu_addr == job->vm_pd_addr &&
!amdgpu_vmid_gds_switch_needed(id, job);
}
/**
* amdgpu_vmid_grab_idle - grab idle VMID
*
@@ -265,7 +285,7 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm,
*id = vm->reserved_vmid[vmhub];
if ((*id)->owner != vm->immediate.fence_context ||
(*id)->pd_gpu_addr != job->vm_pd_addr ||
!amdgpu_vmid_compatible(*id, job) ||
(*id)->flushed_updates < updates ||
!(*id)->last_flush ||
((*id)->last_flush->context != fence_context &&
@@ -294,7 +314,6 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm,
if (r)
return r;
(*id)->flushed_updates = updates;
job->vm_needs_flush = needs_flush;
return 0;
}
@@ -333,7 +352,7 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm,
if ((*id)->owner != vm->immediate.fence_context)
continue;
if ((*id)->pd_gpu_addr != job->vm_pd_addr)
if (!amdgpu_vmid_compatible(*id, job))
continue;
if (!(*id)->last_flush ||
@@ -355,7 +374,6 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm,
if (r)
return r;
(*id)->flushed_updates = updates;
job->vm_needs_flush |= needs_flush;
return 0;
}
@@ -408,22 +426,30 @@ int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
if (r)
goto error;
id->flushed_updates = amdgpu_vm_tlb_seq(vm);
job->vm_needs_flush = true;
}
list_move_tail(&id->list, &id_mgr->ids_lru);
}
id->pd_gpu_addr = job->vm_pd_addr;
id->owner = vm->immediate.fence_context;
job->gds_switch_needed = amdgpu_vmid_gds_switch_needed(id, job);
if (job->vm_needs_flush) {
id->flushed_updates = amdgpu_vm_tlb_seq(vm);
dma_fence_put(id->last_flush);
id->last_flush = NULL;
}
job->vmid = id - id_mgr->ids;
job->pasid = vm->pasid;
id->gds_base = job->gds_base;
id->gds_size = job->gds_size;
id->gws_base = job->gws_base;
id->gws_size = job->gws_size;
id->oa_base = job->oa_base;
id->oa_size = job->oa_size;
id->pd_gpu_addr = job->vm_pd_addr;
id->owner = vm->immediate.fence_context;
trace_amdgpu_vm_grab_id(vm, ring, job);
error: