Hi all,
I've finally gotten around to polish of my lockdep anntotation patches from a while ago:
https://lore.kernel.org/dri-devel/20200610194101.1668038-1-daniel.vetter@ffw...
That patch has been in -mm for a few days already, but it immediately hit some issues with xfs.
Changes since v2: - Now hopefully the bug that bombed xfs fixed. - With unit-tests (that's the part I really wanted and never got to) - might_alloc() helper thrown in for good.
The unit test stuff was the major drag until I figured out how to make this very easy with the locking selftests.
Comments, review, testing all very much welcome.
Cheers, Daniel
Daniel Vetter (3): mm: Track mmu notifiers in fs_reclaim_acquire/release mm: Extract might_alloc() debug check locking/selftests: Add testcases for fs_reclaim
include/linux/sched/mm.h | 16 ++++++++++++++ lib/locking-selftest.c | 47 ++++++++++++++++++++++++++++++++++++++++ mm/mmu_notifier.c | 7 ------ mm/page_alloc.c | 31 ++++++++++++++++---------- mm/slab.h | 5 +---- mm/slob.c | 6 ++--- 6 files changed, 86 insertions(+), 26 deletions(-)
fs_reclaim_acquire/release nicely catch recursion issues when allocating GFP_KERNEL memory against shrinkers (which gpu drivers tend to use to keep the excessive caches in check). For mmu notifier recursions we do have lockdep annotations since 23b68395c7c7 ("mm/mmu_notifiers: add a lockdep map for invalidate_range_start/end").
But these only fire if a path actually results in some pte invalidation - for most small allocations that's very rarely the case. The other trouble is that pte invalidation can happen any time when __GFP_RECLAIM is set. Which means only really GFP_ATOMIC is a safe choice, GFP_NOIO isn't good enough to avoid potential mmu notifier recursion.
I was pondering whether we should just do the general annotation, but there's always the risk for false positives. Plus I'm assuming that the core fs and io code is a lot better reviewed and tested than random mmu notifier code in drivers. Hence why I decide to only annotate for that specific case.
Furthermore even if we'd create a lockdep map for direct reclaim, we'd still need to explicit pull in the mmu notifier map - there's a lot more places that do pte invalidation than just direct reclaim, these two contexts arent the same.
Note that the mmu notifiers needing their own independent lockdep map is also the reason we can't hold them from fs_reclaim_acquire to fs_reclaim_release - it would nest with the acquistion in the pte invalidation code, causing a lockdep splat. And we can't remove the annotations from pte invalidation and all the other places since they're called from many other places than page reclaim. Hence we can only do the equivalent of might_lock, but on the raw lockdep map.
With this we can also remove the lockdep priming added in 66204f1d2d1b ("mm/mmu_notifiers: prime lockdep") since the new annotations are strictly more powerful.
v2: Review from Thomas Hellstrom: - unbotch the fs_reclaim context check, I accidentally inverted it, but it didn't blow up because I inverted it immediately - fix compiling for !CONFIG_MMU_NOTIFIER
v3: Unbreak the PF_MEMALLOC_ context flags. Thanks to Qian for the report and Dave for explaining what I failed to see.
Cc: linux-fsdevel@vger.kernel.org Cc: Dave Chinner david@fromorbit.com Cc: Qian Cai cai@lca.pw Cc: linux-xfs@vger.kernel.org Cc: Thomas Hellström (Intel) thomas_os@shipmail.org Cc: Andrew Morton akpm@linux-foundation.org Cc: Jason Gunthorpe jgg@mellanox.com Cc: linux-mm@kvack.org Cc: linux-rdma@vger.kernel.org Cc: Maarten Lankhorst maarten.lankhorst@linux.intel.com Cc: Christian König christian.koenig@amd.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Signed-off-by: Daniel Vetter daniel.vetter@intel.com --- mm/mmu_notifier.c | 7 ------- mm/page_alloc.c | 31 ++++++++++++++++++++----------- 2 files changed, 20 insertions(+), 18 deletions(-)
diff --git a/mm/mmu_notifier.c b/mm/mmu_notifier.c index 5654dd19addc..61ee40ed804e 100644 --- a/mm/mmu_notifier.c +++ b/mm/mmu_notifier.c @@ -612,13 +612,6 @@ int __mmu_notifier_register(struct mmu_notifier *subscription, mmap_assert_write_locked(mm); BUG_ON(atomic_read(&mm->mm_users) <= 0);
- if (IS_ENABLED(CONFIG_LOCKDEP)) { - fs_reclaim_acquire(GFP_KERNEL); - lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); - lock_map_release(&__mmu_notifier_invalidate_range_start_map); - fs_reclaim_release(GFP_KERNEL); - } - if (!mm->notifier_subscriptions) { /* * kmalloc cannot be called under mm_take_all_locks(), but we diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 23f5066bd4a5..ff0f9a84b8de 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -57,6 +57,7 @@ #include <trace/events/oom.h> #include <linux/prefetch.h> #include <linux/mm_inline.h> +#include <linux/mmu_notifier.h> #include <linux/migrate.h> #include <linux/hugetlb.h> #include <linux/sched/rt.h> @@ -4264,10 +4265,8 @@ should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_fla static struct lockdep_map __fs_reclaim_map = STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
-static bool __need_fs_reclaim(gfp_t gfp_mask) +static bool __need_reclaim(gfp_t gfp_mask) { - gfp_mask = current_gfp_context(gfp_mask); - /* no reclaim without waiting on it */ if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) return false; @@ -4276,10 +4275,6 @@ static bool __need_fs_reclaim(gfp_t gfp_mask) if (current->flags & PF_MEMALLOC) return false;
- /* We're only interested __GFP_FS allocations for now */ - if (!(gfp_mask & __GFP_FS)) - return false; - if (gfp_mask & __GFP_NOLOCKDEP) return false;
@@ -4298,15 +4293,29 @@ void __fs_reclaim_release(void)
void fs_reclaim_acquire(gfp_t gfp_mask) { - if (__need_fs_reclaim(gfp_mask)) - __fs_reclaim_acquire(); + gfp_mask = current_gfp_context(gfp_mask); + + if (__need_reclaim(gfp_mask)) { + if (gfp_mask & __GFP_FS) + __fs_reclaim_acquire(); + +#ifdef CONFIG_MMU_NOTIFIER + lock_map_acquire(&__mmu_notifier_invalidate_range_start_map); + lock_map_release(&__mmu_notifier_invalidate_range_start_map); +#endif + + } } EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
void fs_reclaim_release(gfp_t gfp_mask) { - if (__need_fs_reclaim(gfp_mask)) - __fs_reclaim_release(); + gfp_mask = current_gfp_context(gfp_mask); + + if (__need_reclaim(gfp_mask)) { + if (gfp_mask & __GFP_FS) + __fs_reclaim_release(); + } } EXPORT_SYMBOL_GPL(fs_reclaim_release); #endif
On Fri, Nov 20, 2020 at 10:54:42AM +0100, Daniel Vetter wrote:
fs_reclaim_acquire/release nicely catch recursion issues when allocating GFP_KERNEL memory against shrinkers (which gpu drivers tend to use to keep the excessive caches in check). For mmu notifier recursions we do have lockdep annotations since 23b68395c7c7 ("mm/mmu_notifiers: add a lockdep map for invalidate_range_start/end").
But these only fire if a path actually results in some pte invalidation - for most small allocations that's very rarely the case. The other trouble is that pte invalidation can happen any time when __GFP_RECLAIM is set. Which means only really GFP_ATOMIC is a safe choice, GFP_NOIO isn't good enough to avoid potential mmu notifier recursion.
I was pondering whether we should just do the general annotation, but there's always the risk for false positives. Plus I'm assuming that the core fs and io code is a lot better reviewed and tested than random mmu notifier code in drivers. Hence why I decide to only annotate for that specific case.
Furthermore even if we'd create a lockdep map for direct reclaim, we'd still need to explicit pull in the mmu notifier map - there's a lot more places that do pte invalidation than just direct reclaim, these two contexts arent the same.
Note that the mmu notifiers needing their own independent lockdep map is also the reason we can't hold them from fs_reclaim_acquire to fs_reclaim_release - it would nest with the acquistion in the pte invalidation code, causing a lockdep splat. And we can't remove the annotations from pte invalidation and all the other places since they're called from many other places than page reclaim. Hence we can only do the equivalent of might_lock, but on the raw lockdep map.
With this we can also remove the lockdep priming added in 66204f1d2d1b ("mm/mmu_notifiers: prime lockdep") since the new annotations are strictly more powerful.
v2: Review from Thomas Hellstrom:
- unbotch the fs_reclaim context check, I accidentally inverted it, but it didn't blow up because I inverted it immediately
- fix compiling for !CONFIG_MMU_NOTIFIER
v3: Unbreak the PF_MEMALLOC_ context flags. Thanks to Qian for the report and Dave for explaining what I failed to see.
Cc: linux-fsdevel@vger.kernel.org Cc: Dave Chinner david@fromorbit.com Cc: Qian Cai cai@lca.pw Cc: linux-xfs@vger.kernel.org Cc: Thomas Hellström (Intel) thomas_os@shipmail.org Cc: Andrew Morton akpm@linux-foundation.org Cc: Jason Gunthorpe jgg@mellanox.com Cc: linux-mm@kvack.org Cc: linux-rdma@vger.kernel.org Cc: Maarten Lankhorst maarten.lankhorst@linux.intel.com Cc: Christian König christian.koenig@amd.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Signed-off-by: Daniel Vetter daniel.vetter@intel.com
mm/mmu_notifier.c | 7 ------- mm/page_alloc.c | 31 ++++++++++++++++++++----------- 2 files changed, 20 insertions(+), 18 deletions(-)
Reviewed-by: Jason Gunthorpe jgg@nvidia.com
Jason
Extracted from slab.h, which seems to have the most complete version including the correct might_sleep() check. Roll it out to slob.c.
Motivated by a discussion with Paul about possibly changing call_rcu behaviour to allocate memory, but only roughly every 500th call.
There are a lot fewer places in the kernel that care about whether allocating memory is allowed or not (due to deadlocks with reclaim code) than places that care whether sleeping is allowed. But debugging these also tends to be a lot harder, so nice descriptive checks could come in handy. I might have some use eventually for annotations in drivers/gpu.
Note that unlike fs_reclaim_acquire/release gfpflags_allow_blocking does not consult the PF_MEMALLOC flags. But there is no flag equivalent for GFP_NOWAIT, hence this check can't go wrong due to memalloc_no*_save/restore contexts. Willy is working on a patch series which might change this:
https://lore.kernel.org/linux-mm/20200625113122.7540-7-willy@infradead.org/
I think best would be if that updates gfpflags_allow_blocking(), since there's a ton of callers all over the place for that already.
Acked-by: Vlastimil Babka vbabka@suse.cz Acked-by: Paul E. McKenney paulmck@kernel.org Cc: Paul E. McKenney paulmck@kernel.org Cc: Christoph Lameter cl@linux.com Cc: Pekka Enberg penberg@kernel.org Cc: David Rientjes rientjes@google.com Cc: Joonsoo Kim iamjoonsoo.kim@lge.com Cc: Andrew Morton akpm@linux-foundation.org Cc: Peter Zijlstra peterz@infradead.org Cc: Ingo Molnar mingo@kernel.org Cc: Vlastimil Babka vbabka@suse.cz Cc: Mathieu Desnoyers mathieu.desnoyers@efficios.com Cc: Sebastian Andrzej Siewior bigeasy@linutronix.de Cc: Michel Lespinasse walken@google.com Cc: Daniel Vetter daniel.vetter@ffwll.ch Cc: Waiman Long longman@redhat.com Cc: Thomas Gleixner tglx@linutronix.de Cc: Randy Dunlap rdunlap@infradead.org Cc: linux-mm@kvack.org Cc: linux-fsdevel@vger.kernel.org Cc: Dave Chinner david@fromorbit.com Cc: Qian Cai cai@lca.pw Cc: linux-xfs@vger.kernel.org Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Signed-off-by: Daniel Vetter daniel.vetter@intel.com --- include/linux/sched/mm.h | 16 ++++++++++++++++ mm/slab.h | 5 +---- mm/slob.c | 6 ++---- 3 files changed, 19 insertions(+), 8 deletions(-)
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index d5ece7a9a403..f94405d43fd1 100644 --- a/include/linux/sched/mm.h +++ b/include/linux/sched/mm.h @@ -180,6 +180,22 @@ static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } static inline void fs_reclaim_release(gfp_t gfp_mask) { } #endif
+/** + * might_alloc - Marks possible allocation sites + * @gfp_mask: gfp_t flags that would be use to allocate + * + * Similar to might_sleep() and other annotations this can be used in functions + * that might allocate, but often dont. Compiles to nothing without + * CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking. + */ +static inline void might_alloc(gfp_t gfp_mask) +{ + fs_reclaim_acquire(gfp_mask); + fs_reclaim_release(gfp_mask); + + might_sleep_if(gfpflags_allow_blocking(gfp_mask)); +} + /** * memalloc_noio_save - Marks implicit GFP_NOIO allocation scope. * diff --git a/mm/slab.h b/mm/slab.h index 6d7c6a5056ba..37b981247e5d 100644 --- a/mm/slab.h +++ b/mm/slab.h @@ -500,10 +500,7 @@ static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, { flags &= gfp_allowed_mask;
- fs_reclaim_acquire(flags); - fs_reclaim_release(flags); - - might_sleep_if(gfpflags_allow_blocking(flags)); + might_alloc(flags);
if (should_failslab(s, flags)) return NULL; diff --git a/mm/slob.c b/mm/slob.c index 7cc9805c8091..8d4bfa46247f 100644 --- a/mm/slob.c +++ b/mm/slob.c @@ -474,8 +474,7 @@ __do_kmalloc_node(size_t size, gfp_t gfp, int node, unsigned long caller)
gfp &= gfp_allowed_mask;
- fs_reclaim_acquire(gfp); - fs_reclaim_release(gfp); + might_alloc(gfp);
if (size < PAGE_SIZE - minalign) { int align = minalign; @@ -597,8 +596,7 @@ static void *slob_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
flags &= gfp_allowed_mask;
- fs_reclaim_acquire(flags); - fs_reclaim_release(flags); + might_alloc(flags);
if (c->size < PAGE_SIZE) { b = slob_alloc(c->size, flags, c->align, node, 0);
Hi,
On 11/20/20 1:54 AM, Daniel Vetter wrote:
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index d5ece7a9a403..f94405d43fd1 100644 --- a/include/linux/sched/mm.h +++ b/include/linux/sched/mm.h @@ -180,6 +180,22 @@ static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } static inline void fs_reclaim_release(gfp_t gfp_mask) { } #endif
+/**
- might_alloc - Marks possible allocation sites
Mark
- @gfp_mask: gfp_t flags that would be use to allocate
used
- Similar to might_sleep() and other annotations this can be used in functions
annotations,
- that might allocate, but often dont. Compiles to nothing without
don't.
- CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
? might_sleep_if() if
- */
+static inline void might_alloc(gfp_t gfp_mask) +{
- fs_reclaim_acquire(gfp_mask);
- fs_reclaim_release(gfp_mask);
- might_sleep_if(gfpflags_allow_blocking(gfp_mask));
+}
On Fri, Nov 20, 2020 at 6:20 PM Randy Dunlap rdunlap@infradead.org wrote:
Hi,
On 11/20/20 1:54 AM, Daniel Vetter wrote:
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index d5ece7a9a403..f94405d43fd1 100644 --- a/include/linux/sched/mm.h +++ b/include/linux/sched/mm.h @@ -180,6 +180,22 @@ static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } static inline void fs_reclaim_release(gfp_t gfp_mask) { } #endif
+/**
- might_alloc - Marks possible allocation sites
Mark
- @gfp_mask: gfp_t flags that would be use to allocate
used
- Similar to might_sleep() and other annotations this can be used in functions
annotations,
- that might allocate, but often dont. Compiles to nothing without
don't.
- CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
? might_sleep_if() if
That's one if too many, I'll do the others for next round. Thanks for taking a look. -Daniel
- */
+static inline void might_alloc(gfp_t gfp_mask) +{
fs_reclaim_acquire(gfp_mask);fs_reclaim_release(gfp_mask);might_sleep_if(gfpflags_allow_blocking(gfp_mask));+}
-- ~Randy
On Fri, Nov 20, 2020 at 10:54:43AM +0100, Daniel Vetter wrote:
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index d5ece7a9a403..f94405d43fd1 100644 --- a/include/linux/sched/mm.h +++ b/include/linux/sched/mm.h @@ -180,6 +180,22 @@ static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } static inline void fs_reclaim_release(gfp_t gfp_mask) { } #endif
+/**
- might_alloc - Marks possible allocation sites
- @gfp_mask: gfp_t flags that would be use to allocate
- Similar to might_sleep() and other annotations this can be used in functions
- that might allocate, but often dont. Compiles to nothing without
- CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
- */
+static inline void might_alloc(gfp_t gfp_mask) +{
- fs_reclaim_acquire(gfp_mask);
- fs_reclaim_release(gfp_mask);
- might_sleep_if(gfpflags_allow_blocking(gfp_mask));
+}
Reviewed-by: Jason Gunthorpe jgg@nvidia.com
Oh, I just had a another thread with Matt about xarray, this would be perfect to add before xas_nomem():
diff --git a/lib/idr.c b/lib/idr.c index f4ab4f4aa3c7f5..722d9ddff53221 100644 --- a/lib/idr.c +++ b/lib/idr.c @@ -391,6 +391,8 @@ int ida_alloc_range(struct ida *ida, unsigned int min, unsigned int max, if ((int)max < 0) max = INT_MAX;
+ might_alloc(gfp); + retry: xas_lock_irqsave(&xas, flags); next: diff --git a/lib/xarray.c b/lib/xarray.c index 5fa51614802ada..dd260ee7dcae9a 100644 --- a/lib/xarray.c +++ b/lib/xarray.c @@ -1534,6 +1534,8 @@ void *__xa_store(struct xarray *xa, unsigned long index, void *entry, gfp_t gfp) XA_STATE(xas, xa, index); void *curr;
+ might_alloc(gfp); + if (WARN_ON_ONCE(xa_is_advanced(entry))) return XA_ERROR(-EINVAL); if (xa_track_free(xa) && !entry) @@ -1600,6 +1602,8 @@ void *__xa_cmpxchg(struct xarray *xa, unsigned long index, XA_STATE(xas, xa, index); void *curr;
+ might_alloc(gfp); + if (WARN_ON_ONCE(xa_is_advanced(entry))) return XA_ERROR(-EINVAL);
@@ -1637,6 +1641,8 @@ int __xa_insert(struct xarray *xa, unsigned long index, void *entry, gfp_t gfp) XA_STATE(xas, xa, index); void *curr;
+ might_alloc(gfp); + if (WARN_ON_ONCE(xa_is_advanced(entry))) return -EINVAL; if (!entry) @@ -1806,6 +1812,8 @@ int __xa_alloc(struct xarray *xa, u32 *id, void *entry, { XA_STATE(xas, xa, 0);
+ might_alloc(gfp); + if (WARN_ON_ONCE(xa_is_advanced(entry))) return -EINVAL; if (WARN_ON_ONCE(!xa_track_free(xa)))
On Fri, Nov 20, 2020 at 02:07:19PM -0400, Jason Gunthorpe wrote:
On Fri, Nov 20, 2020 at 10:54:43AM +0100, Daniel Vetter wrote:
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index d5ece7a9a403..f94405d43fd1 100644 --- a/include/linux/sched/mm.h +++ b/include/linux/sched/mm.h @@ -180,6 +180,22 @@ static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } static inline void fs_reclaim_release(gfp_t gfp_mask) { } #endif
+/**
- might_alloc - Marks possible allocation sites
- @gfp_mask: gfp_t flags that would be use to allocate
- Similar to might_sleep() and other annotations this can be used in functions
- that might allocate, but often dont. Compiles to nothing without
- CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
- */
+static inline void might_alloc(gfp_t gfp_mask) +{
- fs_reclaim_acquire(gfp_mask);
- fs_reclaim_release(gfp_mask);
- might_sleep_if(gfpflags_allow_blocking(gfp_mask));
+}
Reviewed-by: Jason Gunthorpe jgg@nvidia.com
Oh, I just had a another thread with Matt about xarray, this would be perfect to add before xas_nomem():
Yeah I think there's plenty of places where this will be useful. Want to slap a sob onto this diff so I can include it for the next round, or will you or Matt send this out when my might_alloc has landed? -Daniel
diff --git a/lib/idr.c b/lib/idr.c index f4ab4f4aa3c7f5..722d9ddff53221 100644 --- a/lib/idr.c +++ b/lib/idr.c @@ -391,6 +391,8 @@ int ida_alloc_range(struct ida *ida, unsigned int min, unsigned int max, if ((int)max < 0) max = INT_MAX;
- might_alloc(gfp);
retry: xas_lock_irqsave(&xas, flags); next: diff --git a/lib/xarray.c b/lib/xarray.c index 5fa51614802ada..dd260ee7dcae9a 100644 --- a/lib/xarray.c +++ b/lib/xarray.c @@ -1534,6 +1534,8 @@ void *__xa_store(struct xarray *xa, unsigned long index, void *entry, gfp_t gfp) XA_STATE(xas, xa, index); void *curr;
- might_alloc(gfp);
- if (WARN_ON_ONCE(xa_is_advanced(entry))) return XA_ERROR(-EINVAL); if (xa_track_free(xa) && !entry)
@@ -1600,6 +1602,8 @@ void *__xa_cmpxchg(struct xarray *xa, unsigned long index, XA_STATE(xas, xa, index); void *curr;
- might_alloc(gfp);
- if (WARN_ON_ONCE(xa_is_advanced(entry))) return XA_ERROR(-EINVAL);
@@ -1637,6 +1641,8 @@ int __xa_insert(struct xarray *xa, unsigned long index, void *entry, gfp_t gfp) XA_STATE(xas, xa, index); void *curr;
- might_alloc(gfp);
- if (WARN_ON_ONCE(xa_is_advanced(entry))) return -EINVAL; if (!entry)
@@ -1806,6 +1812,8 @@ int __xa_alloc(struct xarray *xa, u32 *id, void *entry, { XA_STATE(xas, xa, 0);
- might_alloc(gfp);
- if (WARN_ON_ONCE(xa_is_advanced(entry))) return -EINVAL; if (WARN_ON_ONCE(!xa_track_free(xa)))
On Tue, Nov 24, 2020 at 03:34:11PM +0100, Daniel Vetter wrote:
On Fri, Nov 20, 2020 at 02:07:19PM -0400, Jason Gunthorpe wrote:
On Fri, Nov 20, 2020 at 10:54:43AM +0100, Daniel Vetter wrote:
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h index d5ece7a9a403..f94405d43fd1 100644 +++ b/include/linux/sched/mm.h @@ -180,6 +180,22 @@ static inline void fs_reclaim_acquire(gfp_t gfp_mask) { } static inline void fs_reclaim_release(gfp_t gfp_mask) { } #endif
+/**
- might_alloc - Marks possible allocation sites
- @gfp_mask: gfp_t flags that would be use to allocate
- Similar to might_sleep() and other annotations this can be used in functions
- that might allocate, but often dont. Compiles to nothing without
- CONFIG_LOCKDEP. Includes a conditional might_sleep() if @gfp allows blocking.
- */
+static inline void might_alloc(gfp_t gfp_mask) +{
- fs_reclaim_acquire(gfp_mask);
- fs_reclaim_release(gfp_mask);
- might_sleep_if(gfpflags_allow_blocking(gfp_mask));
+}
Reviewed-by: Jason Gunthorpe jgg@nvidia.com
Oh, I just had a another thread with Matt about xarray, this would be perfect to add before xas_nomem():
Yeah I think there's plenty of places where this will be useful. Want to slap a sob onto this diff so I can include it for the next round, or will you or Matt send this out when my might_alloc has landed?
When this is merged I can do this - just wanted to point out the API is good and useful
Jason
Since I butchered this I figured better to make sure we have testcases for this now. Since we only have a locking context for __GFP_FS that's the only thing we're testing right now.
Cc: linux-fsdevel@vger.kernel.org Cc: Dave Chinner david@fromorbit.com Cc: Qian Cai cai@lca.pw Cc: linux-xfs@vger.kernel.org Cc: Thomas Hellström (Intel) thomas_os@shipmail.org Cc: Andrew Morton akpm@linux-foundation.org Cc: Jason Gunthorpe jgg@mellanox.com Cc: linux-mm@kvack.org Cc: linux-rdma@vger.kernel.org Cc: Maarten Lankhorst maarten.lankhorst@linux.intel.com Cc: Christian König christian.koenig@amd.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Signed-off-by: Daniel Vetter daniel.vetter@intel.com Cc: Peter Zijlstra peterz@infradead.org Cc: Ingo Molnar mingo@redhat.com Cc: Will Deacon will@kernel.org Cc: linux-kernel@vger.kernel.org --- lib/locking-selftest.c | 47 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 47 insertions(+)
diff --git a/lib/locking-selftest.c b/lib/locking-selftest.c index a899b3f0e2e5..ad47c3358e30 100644 --- a/lib/locking-selftest.c +++ b/lib/locking-selftest.c @@ -15,6 +15,7 @@ #include <linux/mutex.h> #include <linux/ww_mutex.h> #include <linux/sched.h> +#include <linux/sched/mm.h> #include <linux/delay.h> #include <linux/lockdep.h> #include <linux/spinlock.h> @@ -2357,6 +2358,50 @@ static void queued_read_lock_tests(void) pr_cont("\n"); }
+static void fs_reclaim_correct_nesting(void) +{ + fs_reclaim_acquire(GFP_KERNEL); + might_alloc(GFP_NOFS); + fs_reclaim_release(GFP_KERNEL); +} + +static void fs_reclaim_wrong_nesting(void) +{ + fs_reclaim_acquire(GFP_KERNEL); + might_alloc(GFP_KERNEL); + fs_reclaim_release(GFP_KERNEL); +} + +static void fs_reclaim_protected_nesting(void) +{ + unsigned int flags; + + fs_reclaim_acquire(GFP_KERNEL); + flags = memalloc_nofs_save(); + might_alloc(GFP_KERNEL); + memalloc_nofs_restore(flags); + fs_reclaim_release(GFP_KERNEL); +} + +static void fs_reclaim_tests(void) +{ + printk(" --------------------\n"); + printk(" | fs_reclaim tests |\n"); + printk(" --------------------\n"); + + print_testname("correct nesting"); + dotest(fs_reclaim_correct_nesting, SUCCESS, 0); + pr_cont("\n"); + + print_testname("wrong nesting"); + dotest(fs_reclaim_wrong_nesting, FAILURE, 0); + pr_cont("\n"); + + print_testname("protected nesting"); + dotest(fs_reclaim_protected_nesting, SUCCESS, 0); + pr_cont("\n"); +} + void locking_selftest(void) { /* @@ -2478,6 +2523,8 @@ void locking_selftest(void) if (IS_ENABLED(CONFIG_QUEUED_RWLOCKS)) queued_read_lock_tests();
+ fs_reclaim_tests(); + if (unexpected_testcase_failures) { printk("-----------------------------------------------------------------\n"); debug_locks = 0;
On Fri, Nov 20, 2020 at 10:54:44AM +0100, Daniel Vetter wrote:
Since I butchered this I figured better to make sure we have testcases for this now. Since we only have a locking context for __GFP_FS that's the only thing we're testing right now.
Cc: linux-fsdevel@vger.kernel.org Cc: Dave Chinner david@fromorbit.com Cc: Qian Cai cai@lca.pw Cc: linux-xfs@vger.kernel.org Cc: Thomas Hellström (Intel) thomas_os@shipmail.org Cc: Andrew Morton akpm@linux-foundation.org Cc: Jason Gunthorpe jgg@mellanox.com Cc: linux-mm@kvack.org Cc: linux-rdma@vger.kernel.org Cc: Maarten Lankhorst maarten.lankhorst@linux.intel.com Cc: Christian König christian.koenig@amd.com Cc: "Matthew Wilcox (Oracle)" willy@infradead.org Signed-off-by: Daniel Vetter daniel.vetter@intel.com Cc: Peter Zijlstra peterz@infradead.org Cc: Ingo Molnar mingo@redhat.com Cc: Will Deacon will@kernel.org Cc: linux-kernel@vger.kernel.org
lib/locking-selftest.c | 47 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 47 insertions(+)
I have a few changes pending for this file, I don't think the conflicts will be bad, but..
In any case:
Acked-by: Peter Zijlstra (Intel) peterz@infradead.org
Random observation while trying to review Christian's patch series to stop looking at struct page for dma-buf imports.
This was originally added in
commit 58aa6622d32af7d2c08d45085f44c54554a16ed7 Author: Thomas Hellstrom thellstrom@vmware.com Date: Fri Jan 3 11:47:23 2014 +0100
drm/ttm: Correctly set page mapping and -index members
Needed for some vm operations; most notably unmap_mapping_range() with even_cows = 0.
Signed-off-by: Thomas Hellstrom thellstrom@vmware.com Reviewed-by: Brian Paul brianp@vmware.com
but we do not have a single caller of unmap_mapping_range with even_cows == 0. And all the gem drivers don't do this, so another small thing we could standardize between drm and ttm drivers.
Plus I don't really see a need for unamp_mapping_range where we don't want to indiscriminately shoot down all ptes.
Cc: Thomas Hellstrom thellstrom@vmware.com Cc: Brian Paul brianp@vmware.com Signed-off-by: Daniel Vetter daniel.vetter@intel.com Cc: Christian Koenig christian.koenig@amd.com Cc: Huang Rui ray.huang@amd.com --- drivers/gpu/drm/ttm/ttm_tt.c | 12 ------------ 1 file changed, 12 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index da9eeffe0c6d..5b2eb6d58bb7 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -284,17 +284,6 @@ int ttm_tt_swapout(struct ttm_bo_device *bdev, struct ttm_tt *ttm) return ret; }
-static void ttm_tt_add_mapping(struct ttm_bo_device *bdev, struct ttm_tt *ttm) -{ - pgoff_t i; - - if (ttm->page_flags & TTM_PAGE_FLAG_SG) - return; - - for (i = 0; i < ttm->num_pages; ++i) - ttm->pages[i]->mapping = bdev->dev_mapping; -} - int ttm_tt_populate(struct ttm_bo_device *bdev, struct ttm_tt *ttm, struct ttm_operation_ctx *ctx) { @@ -313,7 +302,6 @@ int ttm_tt_populate(struct ttm_bo_device *bdev, if (ret) return ret;
- ttm_tt_add_mapping(bdev, ttm); ttm->page_flags |= TTM_PAGE_FLAG_PRIV_POPULATED; if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { ret = ttm_tt_swapin(ttm);
Am 20.11.20 um 10:54 schrieb Daniel Vetter:
Random observation while trying to review Christian's patch series to stop looking at struct page for dma-buf imports.
This was originally added in
commit 58aa6622d32af7d2c08d45085f44c54554a16ed7 Author: Thomas Hellstrom thellstrom@vmware.com Date: Fri Jan 3 11:47:23 2014 +0100
drm/ttm: Correctly set page mapping and -index members Needed for some vm operations; most notably unmap_mapping_range() with even_cows = 0. Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com> Reviewed-by: Brian Paul <brianp@vmware.com>but we do not have a single caller of unmap_mapping_range with even_cows == 0. And all the gem drivers don't do this, so another small thing we could standardize between drm and ttm drivers.
Plus I don't really see a need for unamp_mapping_range where we don't want to indiscriminately shoot down all ptes.
Cc: Thomas Hellstrom thellstrom@vmware.com Cc: Brian Paul brianp@vmware.com Signed-off-by: Daniel Vetter daniel.vetter@intel.com Cc: Christian Koenig christian.koenig@amd.com Cc: Huang Rui ray.huang@amd.com
This is still a NAK as long as we can't come up with a better way to track TTMs page allocations.
Additional to that page_mapping() is used quite extensively in the mm code and I'm not sure if that isn't needed for other stuff as well.
Regards, Christian.
drivers/gpu/drm/ttm/ttm_tt.c | 12 ------------ 1 file changed, 12 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index da9eeffe0c6d..5b2eb6d58bb7 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -284,17 +284,6 @@ int ttm_tt_swapout(struct ttm_bo_device *bdev, struct ttm_tt *ttm) return ret; }
-static void ttm_tt_add_mapping(struct ttm_bo_device *bdev, struct ttm_tt *ttm) -{
- pgoff_t i;
- if (ttm->page_flags & TTM_PAGE_FLAG_SG)
return;- for (i = 0; i < ttm->num_pages; ++i)
ttm->pages[i]->mapping = bdev->dev_mapping;-}
- int ttm_tt_populate(struct ttm_bo_device *bdev, struct ttm_tt *ttm, struct ttm_operation_ctx *ctx) {
@@ -313,7 +302,6 @@ int ttm_tt_populate(struct ttm_bo_device *bdev, if (ret) return ret;
- ttm_tt_add_mapping(bdev, ttm); ttm->page_flags |= TTM_PAGE_FLAG_PRIV_POPULATED; if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { ret = ttm_tt_swapin(ttm);
On Fri, Nov 20, 2020 at 11:04 AM Christian König christian.koenig@amd.com wrote:
Am 20.11.20 um 10:54 schrieb Daniel Vetter:
Random observation while trying to review Christian's patch series to stop looking at struct page for dma-buf imports.
This was originally added in
commit 58aa6622d32af7d2c08d45085f44c54554a16ed7 Author: Thomas Hellstrom thellstrom@vmware.com Date: Fri Jan 3 11:47:23 2014 +0100
drm/ttm: Correctly set page mapping and -index members Needed for some vm operations; most notably unmap_mapping_range() with even_cows = 0. Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com> Reviewed-by: Brian Paul <brianp@vmware.com>but we do not have a single caller of unmap_mapping_range with even_cows == 0. And all the gem drivers don't do this, so another small thing we could standardize between drm and ttm drivers.
Plus I don't really see a need for unamp_mapping_range where we don't want to indiscriminately shoot down all ptes.
Cc: Thomas Hellstrom thellstrom@vmware.com Cc: Brian Paul brianp@vmware.com Signed-off-by: Daniel Vetter daniel.vetter@intel.com Cc: Christian Koenig christian.koenig@amd.com Cc: Huang Rui ray.huang@amd.com
This is still a NAK as long as we can't come up with a better way to track TTMs page allocations.
Additional to that page_mapping() is used quite extensively in the mm code and I'm not sure if that isn't needed for other stuff as well.
Apologies, I'm honestly not quite sure how this lone patch here ended up in this submission. I didn't want to send it out. -Daniel
Regards, Christian.
drivers/gpu/drm/ttm/ttm_tt.c | 12 ------------ 1 file changed, 12 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index da9eeffe0c6d..5b2eb6d58bb7 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -284,17 +284,6 @@ int ttm_tt_swapout(struct ttm_bo_device *bdev, struct ttm_tt *ttm) return ret; }
-static void ttm_tt_add_mapping(struct ttm_bo_device *bdev, struct ttm_tt *ttm) -{
pgoff_t i;if (ttm->page_flags & TTM_PAGE_FLAG_SG)return;for (i = 0; i < ttm->num_pages; ++i)ttm->pages[i]->mapping = bdev->dev_mapping;-}
- int ttm_tt_populate(struct ttm_bo_device *bdev, struct ttm_tt *ttm, struct ttm_operation_ctx *ctx) {
@@ -313,7 +302,6 @@ int ttm_tt_populate(struct ttm_bo_device *bdev, if (ret) return ret;
ttm_tt_add_mapping(bdev, ttm); ttm->page_flags |= TTM_PAGE_FLAG_PRIV_POPULATED; if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { ret = ttm_tt_swapin(ttm);
Am 20.11.20 um 11:05 schrieb Daniel Vetter:
On Fri, Nov 20, 2020 at 11:04 AM Christian König christian.koenig@amd.com wrote:
Am 20.11.20 um 10:54 schrieb Daniel Vetter:
Random observation while trying to review Christian's patch series to stop looking at struct page for dma-buf imports.
This was originally added in
commit 58aa6622d32af7d2c08d45085f44c54554a16ed7 Author: Thomas Hellstrom thellstrom@vmware.com Date: Fri Jan 3 11:47:23 2014 +0100
drm/ttm: Correctly set page mapping and -index members Needed for some vm operations; most notably unmap_mapping_range() with even_cows = 0. Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com> Reviewed-by: Brian Paul <brianp@vmware.com>but we do not have a single caller of unmap_mapping_range with even_cows == 0. And all the gem drivers don't do this, so another small thing we could standardize between drm and ttm drivers.
Plus I don't really see a need for unamp_mapping_range where we don't want to indiscriminately shoot down all ptes.
Cc: Thomas Hellstrom thellstrom@vmware.com Cc: Brian Paul brianp@vmware.com Signed-off-by: Daniel Vetter daniel.vetter@intel.com Cc: Christian Koenig christian.koenig@amd.com Cc: Huang Rui ray.huang@amd.com
This is still a NAK as long as we can't come up with a better way to track TTMs page allocations.
Additional to that page_mapping() is used quite extensively in the mm code and I'm not sure if that isn't needed for other stuff as well.
Apologies, I'm honestly not quite sure how this lone patch here ended up in this submission. I didn't want to send it out.
No problem.
But looking a bit deeper into the mm code that other drm drivers don't set this correctly and still use unmap_mapping_range() sounds like quite a bug to me.
Going to track down what exactly that is used for.
Christian.
-Daniel
Regards, Christian.
drivers/gpu/drm/ttm/ttm_tt.c | 12 ------------ 1 file changed, 12 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index da9eeffe0c6d..5b2eb6d58bb7 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -284,17 +284,6 @@ int ttm_tt_swapout(struct ttm_bo_device *bdev, struct ttm_tt *ttm) return ret; }
-static void ttm_tt_add_mapping(struct ttm_bo_device *bdev, struct ttm_tt *ttm) -{
pgoff_t i;if (ttm->page_flags & TTM_PAGE_FLAG_SG)return;for (i = 0; i < ttm->num_pages; ++i)ttm->pages[i]->mapping = bdev->dev_mapping;-}
- int ttm_tt_populate(struct ttm_bo_device *bdev, struct ttm_tt *ttm, struct ttm_operation_ctx *ctx) {
@@ -313,7 +302,6 @@ int ttm_tt_populate(struct ttm_bo_device *bdev, if (ret) return ret;
ttm_tt_add_mapping(bdev, ttm); ttm->page_flags |= TTM_PAGE_FLAG_PRIV_POPULATED; if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { ret = ttm_tt_swapin(ttm);
On Fri, Nov 20, 2020 at 11:08:31AM +0100, Christian König wrote:
Am 20.11.20 um 11:05 schrieb Daniel Vetter:
On Fri, Nov 20, 2020 at 11:04 AM Christian König christian.koenig@amd.com wrote:
Am 20.11.20 um 10:54 schrieb Daniel Vetter:
Random observation while trying to review Christian's patch series to stop looking at struct page for dma-buf imports.
This was originally added in
commit 58aa6622d32af7d2c08d45085f44c54554a16ed7 Author: Thomas Hellstrom thellstrom@vmware.com Date: Fri Jan 3 11:47:23 2014 +0100
drm/ttm: Correctly set page mapping and -index members Needed for some vm operations; most notably unmap_mapping_range() with even_cows = 0. Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com> Reviewed-by: Brian Paul <brianp@vmware.com>but we do not have a single caller of unmap_mapping_range with even_cows == 0. And all the gem drivers don't do this, so another small thing we could standardize between drm and ttm drivers.
Plus I don't really see a need for unamp_mapping_range where we don't want to indiscriminately shoot down all ptes.
Cc: Thomas Hellstrom thellstrom@vmware.com Cc: Brian Paul brianp@vmware.com Signed-off-by: Daniel Vetter daniel.vetter@intel.com Cc: Christian Koenig christian.koenig@amd.com Cc: Huang Rui ray.huang@amd.com
This is still a NAK as long as we can't come up with a better way to track TTMs page allocations.
Additional to that page_mapping() is used quite extensively in the mm code and I'm not sure if that isn't needed for other stuff as well.
Apologies, I'm honestly not quite sure how this lone patch here ended up in this submission. I didn't want to send it out.
No problem.
But looking a bit deeper into the mm code that other drm drivers don't set this correctly and still use unmap_mapping_range() sounds like quite a bug to me.
Going to track down what exactly that is used for.
Pagecache shootdown. unmap_mapping_range only shoots down from the virtual side. Since that's all we care about, we don't need to set up the address_space in the page. -Daniel
Christian.
-Daniel
Regards, Christian.
drivers/gpu/drm/ttm/ttm_tt.c | 12 ------------ 1 file changed, 12 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index da9eeffe0c6d..5b2eb6d58bb7 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -284,17 +284,6 @@ int ttm_tt_swapout(struct ttm_bo_device *bdev, struct ttm_tt *ttm) return ret; }
-static void ttm_tt_add_mapping(struct ttm_bo_device *bdev, struct ttm_tt *ttm) -{
pgoff_t i;if (ttm->page_flags & TTM_PAGE_FLAG_SG)return;for (i = 0; i < ttm->num_pages; ++i)ttm->pages[i]->mapping = bdev->dev_mapping;-}
- int ttm_tt_populate(struct ttm_bo_device *bdev, struct ttm_tt *ttm, struct ttm_operation_ctx *ctx) {
@@ -313,7 +302,6 @@ int ttm_tt_populate(struct ttm_bo_device *bdev, if (ret) return ret;
ttm_tt_add_mapping(bdev, ttm); ttm->page_flags |= TTM_PAGE_FLAG_PRIV_POPULATED; if (unlikely(ttm->page_flags & TTM_PAGE_FLAG_SWAPPED)) { ret = ttm_tt_swapin(ttm);
dri-devel@lists.freedesktop.org
-
Christian König -
Daniel Vetter -
Daniel Vetter -
Jason Gunthorpe -
Peter Zijlstra -
Randy Dunlap