Re: [PATCH 1/1] drm/amdgpu: fix amdgpu_vm_pt_free warning

[Philip Yang <yangp@xxxxxxx>]

On 2022-09-12 06:22, Christian König wrote:
> Am 09.09.22 um 20:25 schrieb Philip Yang:
> > Free page table BO from vm resv unlocked context generate below
> > warnings.
> >
> > Add a free_work in vm_bo base of each page table BO, pass vm resv
> > unlock status from page table update caller, and schedule the
> > work if calling with vm resv unlocked.
> >
> >   WARNING: CPU: 12 PID: 3238 at
> >   drivers/gpu/drm/ttm/ttm_bo.c:106 ttm_bo_set_bulk_move+0xa1/0xc0
> >   Call Trace:
> >    amdgpu_vm_pt_free+0x42/0xd0 [amdgpu]
> >    amdgpu_vm_pt_free_dfs+0xb3/0xf0 [amdgpu]
> >    amdgpu_vm_ptes_update+0x52d/0x850 [amdgpu]
> >    amdgpu_vm_update_range+0x2a6/0x640 [amdgpu]
> >    svm_range_unmap_from_gpus+0x110/0x300 [amdgpu]
> >    svm_range_cpu_invalidate_pagetables+0x535/0x600 [amdgpu]
> >   __mmu_notifier_invalidate_range_start+0x1cd/0x230
> >    unmap_vmas+0x9d/0x140
> >    unmap_region+0xa8/0x110
> >
> >   WARNING: CPU: 0 PID: 1475 at
> >   drivers/dma-buf/dma-resv.c:483 dma_resv_iter_next
> >   Call Trace:
> >    dma_resv_iter_first+0x43/0xa0
> >    amdgpu_vm_sdma_update+0x69/0x2d0 [amdgpu]
> >    amdgpu_vm_ptes_update+0x29c/0x870 [amdgpu]
> >    amdgpu_vm_update_range+0x2f6/0x6c0 [amdgpu]
> >    svm_range_unmap_from_gpus+0x115/0x300 [amdgpu]
> >    svm_range_cpu_invalidate_pagetables+0x510/0x5e0 [amdgpu]
> >    __mmu_notifier_invalidate_range_start+0x1d3/0x230
> >    unmap_vmas+0x140/0x150
> >    unmap_region+0xa8/0x110
> >
> > Signed-off-by: Philip Yang <Philip.Yang@xxxxxxx>
> > ---
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h    |  3 ++
> >   drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c | 40 +++++++++++++++++++----
> >   2 files changed, 37 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> > index 9ecb7f663e19..ac7cd2c738e5 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> > @@ -150,6 +150,9 @@ struct amdgpu_vm_bo_base {
> >         /* protected by the BO being reserved */
> >       bool                moved;
> > +
> > +    /* work to free pt bo if vm resv is not hold */
> > +    struct work_struct              free_work;
>
> Oh, this is a pretty clear NAK. We can't increase the size of this 
> struct.
>
> Instead just come up with a single worker in the VM and a new state 
> for the page tables.

Good idea, I agree it is too much overhead to add a worker in each pt 
bo. I will send v2 patch.

Regards,

Philip

> Regards,
> Christian.
>
> >   };
> >     /* provided by hw blocks that can write ptes, e.g., sdma */
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c 
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
> > index 88de9f0d4728..e6f6d7e6368f 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
> > @@ -37,6 +37,8 @@ struct amdgpu_vm_pt_cursor {
> >       unsigned int level;
> >   };
> >   +static void amdgpu_vm_pt_free_work(struct work_struct *work);
> > +
> >   /**
> >    * amdgpu_vm_pt_level_shift - return the addr shift for each level
> >    *
> > @@ -607,6 +609,7 @@ static int amdgpu_vm_pt_alloc(struct 
> > amdgpu_device *adev,
> >       pt_bo = &pt->bo;
> >       pt_bo->parent = amdgpu_bo_ref(cursor->parent->bo);
> >       amdgpu_vm_bo_base_init(entry, vm, pt_bo);
> > +    INIT_WORK(&entry->free_work, amdgpu_vm_pt_free_work);
> >       r = amdgpu_vm_pt_clear(adev, vm, pt, immediate);
> >       if (r)
> >           goto error_free_pt;
> > @@ -624,23 +627,46 @@ static int amdgpu_vm_pt_alloc(struct 
> > amdgpu_device *adev,
> >    *
> >    * @entry: PDE to free
> >    */
> > -static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry)
> > +static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry, bool 
> > unlocked)
> >   {
> >       struct amdgpu_bo *shadow;
> >         if (!entry->bo)
> >           return;
> > +
> > +    if (unlocked) {
> > +        schedule_work(&entry->free_work);
> > +        return;
> > +    }
> > +
> >       shadow = amdgpu_bo_shadowed(entry->bo);
> >       if (shadow) {
> >           ttm_bo_set_bulk_move(&shadow->tbo, NULL);
> >           amdgpu_bo_unref(&shadow);
> >       }
> >       ttm_bo_set_bulk_move(&entry->bo->tbo, NULL);
> > +
> >       entry->bo->vm_bo = NULL;
> >       list_del(&entry->vm_status);
> >       amdgpu_bo_unref(&entry->bo);
> >   }
> >   +static void amdgpu_vm_pt_free_work(struct work_struct *work)
> > +{
> > +    struct amdgpu_vm_bo_base *entry;
> > +    struct amdgpu_bo *root;
> > +
> > +    entry = container_of(work, struct amdgpu_vm_bo_base, delayed_work);
> > +
> > +    root = amdgpu_bo_ref(entry->vm->root.bo);
> > +    amdgpu_bo_reserve(root, true);
> > +
> > +    amdgpu_vm_pt_free(entry, true);
> > +
> > +    amdgpu_bo_unreserve(root);
> > +    amdgpu_bo_unref(&root);
> > +}
> > +
> >   /**
> >    * amdgpu_vm_pt_free_dfs - free PD/PT levels
> >    *
> > @@ -652,16 +678,17 @@ static void amdgpu_vm_pt_free(struct 
> > amdgpu_vm_bo_base *entry)
> >    */
> >   static void amdgpu_vm_pt_free_dfs(struct amdgpu_device *adev,
> >                     struct amdgpu_vm *vm,
> > -                  struct amdgpu_vm_pt_cursor *start)
> > +                  struct amdgpu_vm_pt_cursor *start,
> > +                  bool unlocked)
> >   {
> >       struct amdgpu_vm_pt_cursor cursor;
> >       struct amdgpu_vm_bo_base *entry;
> >         for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry)
> > -        amdgpu_vm_pt_free(entry);
> > +        amdgpu_vm_pt_free(entry, unlocked);
> >         if (start)
> > -        amdgpu_vm_pt_free(start->entry);
> > +        amdgpu_vm_pt_free(start->entry, unlocked);
> >   }
> >     /**
> > @@ -673,7 +700,7 @@ static void amdgpu_vm_pt_free_dfs(struct 
> > amdgpu_device *adev,
> >    */
> >   void amdgpu_vm_pt_free_root(struct amdgpu_device *adev, struct 
> > amdgpu_vm *vm)
> >   {
> > -    amdgpu_vm_pt_free_dfs(adev, vm, NULL);
> > +    amdgpu_vm_pt_free_dfs(adev, vm, NULL, false);
> >   }
> >     /**
> > @@ -966,7 +993,8 @@ int amdgpu_vm_ptes_update(struct 
> > amdgpu_vm_update_params *params,
> >                   if (cursor.entry->bo) {
> >                       params->table_freed = true;
> >                       amdgpu_vm_pt_free_dfs(adev, params->vm,
> > -                                  &cursor);
> > +                                  &cursor,
> > +                                  params->unlocked);
> >                   }
> >                   amdgpu_vm_pt_next(adev, &cursor);
> >               }