On Tue, 2024-09-03 at 17:14 +0200, Christian König wrote:
> Am 03.09.24 um 15:38 schrieb Thomas Hellström:
> > Some user-space APIs distinguison between graphics memory OOMs and
> > system (host) memory OOMs. To aid UMDs in determining the type of
> > OOM, allow forwarding the ENOSPC from resource managers to drivers
> > on calls to ttm_bo_validate().
> >
> > Cc: Christian König <christian.koenig@xxxxxxx>
> > Cc: Matthew Brost <matthew.brost@xxxxxxxxx>
> > Cc: dri-devel@xxxxxxxxxxxxxxxxxxxxx
> > Signed-off-by: Thomas Hellström <thomas.hellstrom@xxxxxxxxxxxxxxx>
>
> Ah yes that was on my TODO list as well.
>
> > ---
> > drivers/gpu/drm/ttm/ttm_bo.c | 2 +-
> > include/drm/ttm/ttm_bo.h | 3 +++
> > 2 files changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
> > b/drivers/gpu/drm/ttm/ttm_bo.c
> > index dd867b5e744c..d9a320dc8130 100644
> > --- a/drivers/gpu/drm/ttm/ttm_bo.c
> > +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> > @@ -836,7 +836,7 @@ int ttm_bo_validate(struct ttm_buffer_object
> > *bo,
> > } while (ret && force_space);
> >
> > /* For backward compatibility with userspace */
> > - if (ret == -ENOSPC)
> > + if (ret == -ENOSPC && !ctx->forward_enospc)
>
> Mhm, couldn't we put that into the bdev? I would rather like to keep
> the
> UAPI consistent at least per driver.
Yes, we could probably do that, although that means changing
ttm_device_init() in all drivers.
And if doing that, then I think we should coalesce all bool arguments
to a flags argument to make the callers more readable.
What do you think?
/Thomas
>
> Christian.
>
> > return -ENOMEM;
> >
> > /*
> > diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h
> > index 5804408815be..d3e12318d336 100644
> > --- a/include/drm/ttm/ttm_bo.h
> > +++ b/include/drm/ttm/ttm_bo.h
> > @@ -174,6 +174,8 @@ struct ttm_bo_kmap_obj {
> > * BOs share the same reservation object.
> > * @force_alloc: Don't check the memory account during suspend or
> > CPU page
> > * faults. Should only be used by TTM internally.
> > + * @forward_enospc: Don't translate -ENOSPC errors from resource
> > managers to
> > + * -ENOMEM, but forward them to the driver.
> > * @resv: Reservation object to allow reserved evictions with.
> > * @bytes_moved: Statistics on how many bytes have been moved.
> > *
> > @@ -185,6 +187,7 @@ struct ttm_operation_ctx {
> > bool no_wait_gpu;
> > bool gfp_retry_mayfail;
> > bool allow_res_evict;
> > + bool forward_enospc;
> > bool force_alloc;
> > struct dma_resv *resv;
> > uint64_t bytes_moved;
>
