* [PATCH] drm/ttm: use compound pages for THP
@ 2021-01-13 8:47 David Stevens
2021-01-13 8:58 ` Christian König
0 siblings, 1 reply; 4+ messages in thread
From: David Stevens @ 2021-01-13 8:47 UTC (permalink / raw)
To: Christian Koenig, dri-devel; +Cc: David Stevens
Compound page metadata is necessary for page reference counting to work
properly on pages besides the head page. Without it, put_page
corresponding to the last outstanding get_page call on a tail page will
end up freeing that page, even if the bo still references the page.
Signed-off-by: David Stevens <stevensd@chromium.org>
---
drivers/gpu/drm/ttm/ttm_pool.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
index 7b2f60616750..09239b93dc2c 100644
--- a/drivers/gpu/drm/ttm/ttm_pool.c
+++ b/drivers/gpu/drm/ttm/ttm_pool.c
@@ -83,7 +83,6 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags,
gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY |
__GFP_KSWAPD_RECLAIM;
gfp_flags &= ~__GFP_MOVABLE;
- gfp_flags &= ~__GFP_COMP;
}
if (!pool->use_dma_alloc) {
--
2.30.0.284.gd98b1dd5eaa7-goog
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH] drm/ttm: use compound pages for THP
2021-01-13 8:47 [PATCH] drm/ttm: use compound pages for THP David Stevens
@ 2021-01-13 8:58 ` Christian König
2021-01-13 9:33 ` David Stevens
0 siblings, 1 reply; 4+ messages in thread
From: Christian König @ 2021-01-13 8:58 UTC (permalink / raw)
To: David Stevens, dri-devel
Am 13.01.21 um 09:47 schrieb David Stevens:
> Compound page metadata is necessary for page reference counting to work
> properly on pages besides the head page. Without it, put_page
> corresponding to the last outstanding get_page call on a tail page will
> end up freeing that page, even if the bo still references the page.
NAK, I should add a comment since I can't count any more how many times
people came up with this.
Calling put_page() on a TTM allocated page is strictly illegal in the
first place since they are not reference counted.
Please don't tell me somebody once more tried to mmap() pages from a
DMA-buf created sg_table into a process address space?
Regards,
Christian.
>
> Signed-off-by: David Stevens <stevensd@chromium.org>
> ---
> drivers/gpu/drm/ttm/ttm_pool.c | 1 -
> 1 file changed, 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
> index 7b2f60616750..09239b93dc2c 100644
> --- a/drivers/gpu/drm/ttm/ttm_pool.c
> +++ b/drivers/gpu/drm/ttm/ttm_pool.c
> @@ -83,7 +83,6 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags,
> gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY |
> __GFP_KSWAPD_RECLAIM;
> gfp_flags &= ~__GFP_MOVABLE;
> - gfp_flags &= ~__GFP_COMP;
> }
>
> if (!pool->use_dma_alloc) {
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] drm/ttm: use compound pages for THP
2021-01-13 8:58 ` Christian König
@ 2021-01-13 9:33 ` David Stevens
2021-01-13 12:44 ` Christian König
0 siblings, 1 reply; 4+ messages in thread
From: David Stevens @ 2021-01-13 9:33 UTC (permalink / raw)
To: Christian König; +Cc: ML dri-devel, Gurchetan Singh
On Wed, Jan 13, 2021 at 5:58 PM Christian König
<christian.koenig@amd.com> wrote:
>
> Am 13.01.21 um 09:47 schrieb David Stevens:
> > Compound page metadata is necessary for page reference counting to work
> > properly on pages besides the head page. Without it, put_page
> > corresponding to the last outstanding get_page call on a tail page will
> > end up freeing that page, even if the bo still references the page.
>
> NAK, I should add a comment since I can't count any more how many times
> people came up with this.
>
> Calling put_page() on a TTM allocated page is strictly illegal in the
> first place since they are not reference counted.
>
> Please don't tell me somebody once more tried to mmap() pages from a
> DMA-buf created sg_table into a process address space?
I ran into this on a system using the currently in development
virtio-gpu blob resources [1]. The implementation passes dma buffers
allocated by the host gpu to KVM_SET_USER_MEMORY_REGION, so the guest
can directly access the buffers without the need for an intermediate
copy.
[1] https://patchwork.kernel.org/project/dri-devel/cover/20200814024000.2485-1-gurchetansingh@chromium.org/
> Regards,
> Christian.
>
> >
> > Signed-off-by: David Stevens <stevensd@chromium.org>
> > ---
> > drivers/gpu/drm/ttm/ttm_pool.c | 1 -
> > 1 file changed, 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
> > index 7b2f60616750..09239b93dc2c 100644
> > --- a/drivers/gpu/drm/ttm/ttm_pool.c
> > +++ b/drivers/gpu/drm/ttm/ttm_pool.c
> > @@ -83,7 +83,6 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags,
> > gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY |
> > __GFP_KSWAPD_RECLAIM;
> > gfp_flags &= ~__GFP_MOVABLE;
> > - gfp_flags &= ~__GFP_COMP;
> > }
> >
> > if (!pool->use_dma_alloc) {
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] drm/ttm: use compound pages for THP
2021-01-13 9:33 ` David Stevens
@ 2021-01-13 12:44 ` Christian König
0 siblings, 0 replies; 4+ messages in thread
From: Christian König @ 2021-01-13 12:44 UTC (permalink / raw)
To: David Stevens; +Cc: ML dri-devel, Gurchetan Singh
Am 13.01.21 um 10:33 schrieb David Stevens:
> On Wed, Jan 13, 2021 at 5:58 PM Christian König
> <christian.koenig@amd.com> wrote:
>> Am 13.01.21 um 09:47 schrieb David Stevens:
>>> Compound page metadata is necessary for page reference counting to work
>>> properly on pages besides the head page. Without it, put_page
>>> corresponding to the last outstanding get_page call on a tail page will
>>> end up freeing that page, even if the bo still references the page.
>> NAK, I should add a comment since I can't count any more how many times
>> people came up with this.
>>
>> Calling put_page() on a TTM allocated page is strictly illegal in the
>> first place since they are not reference counted.
>>
>> Please don't tell me somebody once more tried to mmap() pages from a
>> DMA-buf created sg_table into a process address space?
> I ran into this on a system using the currently in development
> virtio-gpu blob resources [1]. The implementation passes dma buffers
> allocated by the host gpu to KVM_SET_USER_MEMORY_REGION, so the guest
> can directly access the buffers without the need for an intermediate
> copy.
Yeah, that is completely illegal. They risk memory corruption with that.
You should probably report back to the author of those patches and not
that the dma_buf_mmap() functions need to be used for this use case.
Regards,
Christian.
>
> [1] https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatchwork.kernel.org%2Fproject%2Fdri-devel%2Fcover%2F20200814024000.2485-1-gurchetansingh%40chromium.org%2F&data=04%7C01%7Cchristian.koenig%40amd.com%7C4d93e46264e54f7d047908d8b7a64991%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637461272261264257%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=XH4UM65P6tNwJ4CUxt12TmBemEgl6q8gQYY0EqdBtyc%3D&reserved=0
>
>> Regards,
>> Christian.
>>
>>> Signed-off-by: David Stevens <stevensd@chromium.org>
>>> ---
>>> drivers/gpu/drm/ttm/ttm_pool.c | 1 -
>>> 1 file changed, 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
>>> index 7b2f60616750..09239b93dc2c 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_pool.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_pool.c
>>> @@ -83,7 +83,6 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags,
>>> gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY |
>>> __GFP_KSWAPD_RECLAIM;
>>> gfp_flags &= ~__GFP_MOVABLE;
>>> - gfp_flags &= ~__GFP_COMP;
>>> }
>>>
>>> if (!pool->use_dma_alloc) {
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2021-01-14 9:29 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-01-13 8:47 [PATCH] drm/ttm: use compound pages for THP David Stevens
2021-01-13 8:58 ` Christian König
2021-01-13 9:33 ` David Stevens
2021-01-13 12:44 ` Christian König
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).