| Message ID | 20210113131325.223511-1-christian.koenig@amd.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/ttm: stop using GFP_TRANSHUGE_LIGHT | expand |
On Wed, Jan 13, 2021 at 02:13:25PM +0100, Christian König wrote: > The only flag we really need is __GFP_NOMEMALLOC, highmem depends on > dma32 and moveable/compound should never be set in the first place. > > Signed-off-by: Christian König <christian.koenig@amd.com> Needs Fixes: line and References to the GFP_TRANSHUGE_LIGHT report from Hans and for the chromium report for the mmap fail. With those: Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> Also I guess really time for me to stitch together that hack to catch dma-buf sglist struct page abusers ... -Daniel > --- > drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- > 1 file changed, 6 insertions(+), 5 deletions(-) > > diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c > index 8cd776adc592..11e0313db0ea 100644 > --- a/drivers/gpu/drm/ttm/ttm_pool.c > +++ b/drivers/gpu/drm/ttm/ttm_pool.c > @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, > struct page *p; > void *vaddr; > > - if (order) { > - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | > + /* Don't set the __GFP_COMP flag for higher order allocations. > + * Mapping pages directly into an userspace process and calling > + * put_page() on a TTM allocated page is illegal. > + */ > + if (order) > + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | > __GFP_KSWAPD_RECLAIM; > - gfp_flags &= ~__GFP_MOVABLE; > - gfp_flags &= ~__GFP_COMP; > - } > > if (!pool->use_dma_alloc) { > p = alloc_pages(gfp_flags, order); > -- > 2.25.1 >
Am 13.01.21 um 14:27 schrieb Daniel Vetter: > On Wed, Jan 13, 2021 at 02:13:25PM +0100, Christian König wrote: >> The only flag we really need is __GFP_NOMEMALLOC, highmem depends on >> dma32 and moveable/compound should never be set in the first place. >> >> Signed-off-by: Christian König <christian.koenig@amd.com> > Needs Fixes: line and References to the GFP_TRANSHUGE_LIGHT report from > Hans and for the chromium report for the mmap fail. Which tag should I use for this? Just Link:? > With those: > > Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> > > Also I guess really time for me to stitch together that hack to catch > dma-buf sglist struct page abusers ... Oh, yes please. Christian. > -Daniel > >> --- >> drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- >> 1 file changed, 6 insertions(+), 5 deletions(-) >> >> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c >> index 8cd776adc592..11e0313db0ea 100644 >> --- a/drivers/gpu/drm/ttm/ttm_pool.c >> +++ b/drivers/gpu/drm/ttm/ttm_pool.c >> @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, >> struct page *p; >> void *vaddr; >> >> - if (order) { >> - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | >> + /* Don't set the __GFP_COMP flag for higher order allocations. >> + * Mapping pages directly into an userspace process and calling >> + * put_page() on a TTM allocated page is illegal. >> + */ >> + if (order) >> + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | >> __GFP_KSWAPD_RECLAIM; >> - gfp_flags &= ~__GFP_MOVABLE; >> - gfp_flags &= ~__GFP_COMP; >> - } >> >> if (!pool->use_dma_alloc) { >> p = alloc_pages(gfp_flags, order); >> -- >> 2.25.1 >>
On Wed, Jan 13, 2021 at 3:46 PM Christian König <ckoenig.leichtzumerken@gmail.com> wrote: > > Am 13.01.21 um 14:27 schrieb Daniel Vetter: > > On Wed, Jan 13, 2021 at 02:13:25PM +0100, Christian König wrote: > >> The only flag we really need is __GFP_NOMEMALLOC, highmem depends on > >> dma32 and moveable/compound should never be set in the first place. > >> > >> Signed-off-by: Christian König <christian.koenig@amd.com> > > Needs Fixes: line and References to the GFP_TRANSHUGE_LIGHT report from > > Hans and for the chromium report for the mmap fail. > > Which tag should I use for this? Just Link:? Link: is for the patch submission itself (to link to archives with the review&discussions), usually it's References is for bug reports and anything else. I think at least. You can have multiple Link: tags too, if you record all the previous submissions, that's a bit up to subsystems and how they all roll. -Daniel > > With those: > > > > Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> > > > > Also I guess really time for me to stitch together that hack to catch > > dma-buf sglist struct page abusers ... > > Oh, yes please. > > Christian. > > > -Daniel > > > >> --- > >> drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- > >> 1 file changed, 6 insertions(+), 5 deletions(-) > >> > >> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c > >> index 8cd776adc592..11e0313db0ea 100644 > >> --- a/drivers/gpu/drm/ttm/ttm_pool.c > >> +++ b/drivers/gpu/drm/ttm/ttm_pool.c > >> @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, > >> struct page *p; > >> void *vaddr; > >> > >> - if (order) { > >> - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | > >> + /* Don't set the __GFP_COMP flag for higher order allocations. > >> + * Mapping pages directly into an userspace process and calling > >> + * put_page() on a TTM allocated page is illegal. > >> + */ > >> + if (order) > >> + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | > >> __GFP_KSWAPD_RECLAIM; > >> - gfp_flags &= ~__GFP_MOVABLE; > >> - gfp_flags &= ~__GFP_COMP; > >> - } > >> > >> if (!pool->use_dma_alloc) { > >> p = alloc_pages(gfp_flags, order); > >> -- > >> 2.25.1 > >> >
Hans do you have any more comments or a tested-by? Otherwise I push it to drm-misc-fixes today. Thanks, Christian. Am 13.01.21 um 14:13 schrieb Christian König: > The only flag we really need is __GFP_NOMEMALLOC, highmem depends on > dma32 and moveable/compound should never be set in the first place. > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- > 1 file changed, 6 insertions(+), 5 deletions(-) > > diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c > index 8cd776adc592..11e0313db0ea 100644 > --- a/drivers/gpu/drm/ttm/ttm_pool.c > +++ b/drivers/gpu/drm/ttm/ttm_pool.c > @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, > struct page *p; > void *vaddr; > > - if (order) { > - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | > + /* Don't set the __GFP_COMP flag for higher order allocations. > + * Mapping pages directly into an userspace process and calling > + * put_page() on a TTM allocated page is illegal. > + */ > + if (order) > + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | > __GFP_KSWAPD_RECLAIM; > - gfp_flags &= ~__GFP_MOVABLE; > - gfp_flags &= ~__GFP_COMP; > - } > > if (!pool->use_dma_alloc) { > p = alloc_pages(gfp_flags, order);
Hi, On 1/15/21 1:14 PM, Christian König wrote: > Hans do you have any more comments or a tested-by? Sorry, I've been busy chasing after another 5.11 regression, no comments, also no tested-by, but I do fully expect this to solve the issue. > Otherwise I push it to drm-misc-fixes today. That sounds good to me. Regards, Hans > > Thanks, > Christian. > > Am 13.01.21 um 14:13 schrieb Christian König: >> The only flag we really need is __GFP_NOMEMALLOC, highmem depends on >> dma32 and moveable/compound should never be set in the first place. >> >> Signed-off-by: Christian König <christian.koenig@amd.com> >> --- >> drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- >> 1 file changed, 6 insertions(+), 5 deletions(-) >> >> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c >> index 8cd776adc592..11e0313db0ea 100644 >> --- a/drivers/gpu/drm/ttm/ttm_pool.c >> +++ b/drivers/gpu/drm/ttm/ttm_pool.c >> @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, >> struct page *p; >> void *vaddr; >> - if (order) { >> - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | >> + /* Don't set the __GFP_COMP flag for higher order allocations. >> + * Mapping pages directly into an userspace process and calling >> + * put_page() on a TTM allocated page is illegal. >> + */ >> + if (order) >> + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | >> __GFP_KSWAPD_RECLAIM; >> - gfp_flags &= ~__GFP_MOVABLE; >> - gfp_flags &= ~__GFP_COMP; >> - } >> if (!pool->use_dma_alloc) { >> p = alloc_pages(gfp_flags, order); >
Am 15.01.21 um 13:54 schrieb Hans de Goede: > Hi, > > On 1/15/21 1:14 PM, Christian König wrote: >> Hans do you have any more comments or a tested-by? > Sorry, I've been busy chasing after another 5.11 regression, > no comments, also no tested-by, but I do fully expect this to solve > the issue. Yeah, I know what you mean :) > >> Otherwise I push it to drm-misc-fixes today. > That sounds good to me. Thanks, going to do so any moment. Regards, Christian. > > Regards, > > Hans > > > > >> Thanks, >> Christian. >> >> Am 13.01.21 um 14:13 schrieb Christian König: >>> The only flag we really need is __GFP_NOMEMALLOC, highmem depends on >>> dma32 and moveable/compound should never be set in the first place. >>> >>> Signed-off-by: Christian König <christian.koenig@amd.com> >>> --- >>> drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- >>> 1 file changed, 6 insertions(+), 5 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c >>> index 8cd776adc592..11e0313db0ea 100644 >>> --- a/drivers/gpu/drm/ttm/ttm_pool.c >>> +++ b/drivers/gpu/drm/ttm/ttm_pool.c >>> @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, >>> struct page *p; >>> void *vaddr; >>> - if (order) { >>> - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | >>> + /* Don't set the __GFP_COMP flag for higher order allocations. >>> + * Mapping pages directly into an userspace process and calling >>> + * put_page() on a TTM allocated page is illegal. >>> + */ >>> + if (order) >>> + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | >>> __GFP_KSWAPD_RECLAIM; >>> - gfp_flags &= ~__GFP_MOVABLE; >>> - gfp_flags &= ~__GFP_COMP; >>> - } >>> if (!pool->use_dma_alloc) { >>> p = alloc_pages(gfp_flags, order);
On 2021-01-13 2:13 p.m., Christian König wrote: > The only flag we really need is __GFP_NOMEMALLOC, highmem depends on > dma32 and moveable/compound should never be set in the first place. > > Signed-off-by: Christian König <christian.koenig@amd.com> > --- > drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- > 1 file changed, 6 insertions(+), 5 deletions(-) > > diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c > index 8cd776adc592..11e0313db0ea 100644 > --- a/drivers/gpu/drm/ttm/ttm_pool.c > +++ b/drivers/gpu/drm/ttm/ttm_pool.c > @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, > struct page *p; > void *vaddr; > > - if (order) { > - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | > + /* Don't set the __GFP_COMP flag for higher order allocations. > + * Mapping pages directly into an userspace process and calling > + * put_page() on a TTM allocated page is illegal. > + */ > + if (order) > + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | > __GFP_KSWAPD_RECLAIM; > - gfp_flags &= ~__GFP_MOVABLE; > - gfp_flags &= ~__GFP_COMP; > - } > > if (!pool->use_dma_alloc) { > p = alloc_pages(gfp_flags, order); > I picked up this change today, and got the attached splat while running piglit. scripts/faddr2line drivers/gpu/drm/ttm/ttm.ko ttm_pool_alloc+0x2e4/0x5e0 gives: ttm_pool_alloc+0x2e4/0x5e0: alloc_pages at /home/daenzer/src/linux-git/linux/./include/linux/gfp.h:547 (inlined by) ttm_pool_alloc_page at /home/daenzer/src/linux-git/linux/drivers/gpu/drm//ttm/ttm_pool.c:91 (inlined by) ttm_pool_alloc at /home/daenzer/src/linux-git/linux/drivers/gpu/drm//ttm/ttm_pool.c:398 I suspect we need __GFP_NOWARN as well to avoid these splats.
diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c index 8cd776adc592..11e0313db0ea 100644 --- a/drivers/gpu/drm/ttm/ttm_pool.c +++ b/drivers/gpu/drm/ttm/ttm_pool.c @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags, struct page *p; void *vaddr; - if (order) { - gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY | + /* Don't set the __GFP_COMP flag for higher order allocations. + * Mapping pages directly into an userspace process and calling + * put_page() on a TTM allocated page is illegal. + */ + if (order) + gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM; - gfp_flags &= ~__GFP_MOVABLE; - gfp_flags &= ~__GFP_COMP; - } if (!pool->use_dma_alloc) { p = alloc_pages(gfp_flags, order);
The only flag we really need is __GFP_NOMEMALLOC, highmem depends on dma32 and moveable/compound should never be set in the first place. Signed-off-by: Christian König <christian.koenig@amd.com> --- drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-)