Skip to contentSkip to conversation

5.14-rc1 mm/page_alloc.c stray patches

PATCH In review Mel Gorman 4 participants 10 messages Jul 13, 2021 – Jul 16, 2021

Subsystems (4)
lore
Share & tools
lore

Mel Gorman

author
#0 lore raw

v2 cover 5.14-rc1 mm/page_alloc.c stray patches

(This v2 is because I didn't refresh the patches from my git tree properly before sending, sorry for the noise)

This series is some fixes that would have likely have been included in the 5.14-rc1 merge window if they were on time. Mail indicates that some may already be picked up for mmotm but the tree is not up to date yet so I'm including them just in case.

Three are fixes to the bulk memory allocator and one is a fallout from cleaning up warnings that trips BTF that expected a symbol to be global.

 mm/page_alloc.c | 28 +++++++++++++++++++++-------
 1 file changed, 21 insertions(+), 7 deletions(-)
Signature · 1 line
2.26.2
author
#1 lore raw

PATCH v2 1/4 mm/page_alloc: Avoid page allocator recursion with pagesets.lock held

Syzbot is reporting potential deadlocks due to pagesets.lock when PAGE_OWNER is enabled. One example from Desmond Cheong Zhi Xi is as follows

  __alloc_pages_bulk()
    local_lock_irqsave(&pagesets.lock, flags) <---- outer lock here
    prep_new_page():
      post_alloc_hook():
        set_page_owner():
          __set_page_owner():
            save_stack():
              stack_depot_save():
                alloc_pages():
                  alloc_page_interleave():
                    __alloc_pages():
                      get_page_from_freelist():
                        rm_queue():
                          rm_queue_pcplist():
                            local_lock_irqsave(&pagesets.lock, flags);
                            *** DEADLOCK ***

Zhang, Qiang also reported

  BUG: sleeping function called from invalid context at mm/page_alloc.c:5179
  in_atomic(): 0, irqs_disabled(): 1, non_block: 0, pid: 1, name: swapper/0
  .....
  __dump_stack lib/dump_stack.c:79 [inline]
  dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:96
  ___might_sleep.cold+0x1f1/0x237 kernel/sched/core.c:9153
  prepare_alloc_pages+0x3da/0x580 mm/page_alloc.c:5179
  __alloc_pages+0x12f/0x500 mm/page_alloc.c:5375
  alloc_page_interleave+0x1e/0x200 mm/mempolicy.c:2147
  alloc_pages+0x238/0x2a0 mm/mempolicy.c:2270
  stack_depot_save+0x39d/0x4e0 lib/stackdepot.c:303
  save_stack+0x15e/0x1e0 mm/page_owner.c:120
  __set_page_owner+0x50/0x290 mm/page_owner.c:181
  prep_new_page mm/page_alloc.c:2445 [inline]
  __alloc_pages_bulk+0x8b9/0x1870 mm/page_alloc.c:5313
  alloc_pages_bulk_array_node include/linux/gfp.h:557 [inline]
  vm_area_alloc_pages mm/vmalloc.c:2775 [inline]
  __vmalloc_area_node mm/vmalloc.c:2845 [inline]
  __vmalloc_node_range+0x39d/0x960 mm/vmalloc.c:2947
  __vmalloc_node mm/vmalloc.c:2996 [inline]
  vzalloc+0x67/0x80 mm/vmalloc.c:3066

There are a number of ways it could be fixed. The page owner code could be audited to strip GFP flags that allow sleeping but it'll impair the functionality of PAGE_OWNER if allocations fail. The bulk allocator could add a special case to release/reacquire the lock for prep_new_page and lookup PCP after the lock is reacquired at the cost of performance. The pages requiring prep could be tracked using the least significant bit and looping through the array although it is more complicated for the list interface. The options are relatively complex and the second one still incurs a performance penalty when PAGE_OWNER is active so this patch takes the simple approach -- disable bulk allocation if PAGE_OWNER is active. The caller will be forced to allocate one page at a time incurring a performance penalty but PAGE_OWNER is already a performance penalty.

Fixes dbbee9d5cd83 ("mm/page_alloc: convert per-cpu list protection to local_lock")
Reported-and-tested-by [email protected]
---
 mm/page_alloc.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
mm/page_alloc.c +12 -0
index 3b97e17806be..6ef86f338151 100644
@@ -5239,6 +5239,18 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	if (nr_pages - nr_populated == 1)
 		goto failed;
 
+#ifdef CONFIG_PAGE_OWNER
+	/*
+	 * PAGE_OWNER may recurse into the allocator to allocate space to
+	 * save the stack with pagesets.lock held. Releasing/reacquiring
+	 * removes much of the performance benefit of bulk allocation so
+	 * force the caller to allocate one page at a time as it'll have
+	 * similar performance to added complexity to the bulk allocator.
+	 */
+	if (static_branch_unlikely(&page_owner_inited))
+		goto failed;
+#endif
+
 	/* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */
 	gfp &= gfp_allowed_mask;
 	alloc_gfp = gfp;
Signature · 1 line
2.26.2
author
#2 lore raw

PATCH v2 2/4 mm/page_alloc: correct return value when failing at preparing

In reply to Mel Gorman

From: Yanfei Xu <[email protected]>

If the array passed in is already partially populated, we should return "nr_populated" even failing at preparing arguments stage.

---
 mm/page_alloc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/page_alloc.c +1 -1
index 6ef86f338151..803414ce9264 100644
@@ -5255,7 +5255,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	gfp &= gfp_allowed_mask;
 	alloc_gfp = gfp;
 	if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags))
-		return 0;
+		return nr_populated;
 	gfp = alloc_gfp;
 
 	/* Find an allowed local zone that meets the low watermark. */
Signature · 1 line
2.26.2
author
#3 lore raw

PATCH v2 3/4 mm/page_alloc: Further fix __alloc_pages_bulk() return value

In reply to Mel Gorman

From: Chuck Lever <[email protected]>

The author of commit b3b64ebd3822 ("mm/page_alloc: do bulk array bounds check after checking populated elements") was possibly confused by the mixture of return values throughout the function.

The API contract is clear that the function "Returns the number of pages on the list or array." It does not list zero as a unique return value with a special meaning. Therefore zero is a plausible return value only if @nr_pages is zero or less.

Clean up the return logic to make it clear that the returned value is always the total number of pages in the array/list, not the number of pages that were allocated during this call.

The only change in behavior with this patch is the value returned if prepare_alloc_pages() fails. To match the API contract, the number of pages currently in the array/list is returned in this case.

The call site in __page_pool_alloc_pages_slow() also seems to be confused on this matter. It should be attended to by someone who is familiar with that code.

[[email protected]: Return nr_populated if 0 pages are requested]
---
 mm/page_alloc.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)
mm/page_alloc.c +8 -6
index 803414ce9264..c66f1e6204c2 100644
@@ -5221,9 +5221,6 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
 	int nr_populated = 0, nr_account = 0;
 
-	if (unlikely(nr_pages <= 0))
-		return 0;
-
 	/*
 	 * Skip populated array elements to determine if any pages need
 	 * to be allocated before disabling IRQs.
@@ -5231,9 +5228,13 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	while (page_array && nr_populated < nr_pages && page_array[nr_populated])
 		nr_populated++;
 
+	/* No pages requested? */
+	if (unlikely(nr_pages <= 0))
+		goto out;
+
 	/* Already populated array? */
 	if (unlikely(page_array && nr_pages - nr_populated == 0))
-		return nr_populated;
+		goto out;
 
 	/* Use the single page allocator for one page. */
 	if (nr_pages - nr_populated == 1)
@@ -5255,7 +5256,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	gfp &= gfp_allowed_mask;
 	alloc_gfp = gfp;
 	if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags))
-		return nr_populated;
+		goto out;
 	gfp = alloc_gfp;
 
 	/* Find an allowed local zone that meets the low watermark. */
@@ -5323,6 +5324,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	__count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
 	zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
 
+out:
 	return nr_populated;
 
 failed_irq:
@@ -5338,7 +5340,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 		nr_populated++;
 	}
 
-	return nr_populated;
+	goto out;
 }
 EXPORT_SYMBOL_GPL(__alloc_pages_bulk);
 
Signature · 1 line
2.26.2
#4 lore raw

Re: [PATCH 3/4] mm/page_alloc: Further fix __alloc_pages_bulk() return value

Quoted 66 lines · Mel Gorman

From: Chuck Lever <[email protected]>

The author of commit b3b64ebd3822 ("mm/page_alloc: do bulk array bounds check after checking populated elements") was possibly confused by the mixture of return values throughout the function.

The API contract is clear that the function "Returns the number of pages on the list or array." It does not list zero as a unique return value with a special meaning. Therefore zero is a plausible return value only if @nr_pages is zero or less.

Clean up the return logic to make it clear that the returned value is always the total number of pages in the array/list, not the number of pages that were allocated during this call.

The only change in behavior with this patch is the value returned if prepare_alloc_pages() fails. To match the API contract, the number of pages currently in the array/list is returned in this case.

The call site in __page_pool_alloc_pages_slow() also seems to be confused on this matter. It should be attended to by someone who is familiar with that code.

[[email protected]: Return nr_populated if 0 pages are requested]

---
mm/page_alloc.c | 14 ++++++++------

1 file changed, 8 insertions(+), 6 deletions(-)
mm/page_alloc.c +5 -4
index 803414ce9264..c66f1e6204c2 100644
@@ -5221,9 +5221,6 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
 	int nr_populated = 0, nr_account = 0;
 
-	if (unlikely(nr_pages <= 0))
-		return 0;
-
 	/*
 	 * Skip populated array elements to determine if any pages need
 	 * to be allocated before disabling IRQs.
@@ -5231,9 +5228,13 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	while (page_array && nr_populated < nr_pages && page_array[nr_populated])
 		nr_populated++;
 
+	/* No pages requested? */
+	if (unlikely(nr_pages <= 0))
+		goto out;
+
 	/* Already populated array? */
 	if (unlikely(page_array && nr_pages - nr_populated == 0))
-		return nr_populated;
+		goto out;
 
 	/* Use the single page allocator for one page. */
 	if (nr_pages - nr_populated == 1)

mm/page_alloc.c unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, · line 5259

Quoted 5 lines
mm/page_alloc.c +1 -1
@@ -5255,7 +5256,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	gfp &= gfp_allowed_mask;
 	alloc_gfp = gfp;
 	if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags))
-		return nr_populated;
+		goto out;

:thumbsup: Thanks!

Quoted 23 lines
mm/page_alloc.c +1 -0
 	gfp = alloc_gfp;
 
 	/* Find an allowed local zone that meets the low watermark. */
@@ -5323,6 +5324,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	__count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
 	zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
 
+out:
 	return nr_populated;
 

failed_irq:

mm/page_alloc.c +1 -1
@@ -5338,7 +5340,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 		nr_populated++;
 	}
 
-	return nr_populated;
+	goto out;
 }
 EXPORT_SYMBOL_GPL(__alloc_pages_bulk);
 
Signature · 1 line
2.26.2

--
Chuck Lever

author
#5 lore raw

PATCH v2 4/4 Revert "mm/page_alloc: make should_fail_alloc_page() static"

In reply to Mel Gorman

From: Matteo Croce <[email protected]>

This reverts commit f7173090033c70886d925995e9dfdfb76dbb2441.

Fix an unresolved symbol error when CONFIG_DEBUG_INFO_BTF=y:

LD vmlinux
BTFIDS vmlinux
FAILED unresolved symbol should_fail_alloc_page
make: *** [Makefile:1199: vmlinux] Error 255
make: *** Deleting file 'vmlinux'

Fixes f7173090033c ("mm/page_alloc: make should_fail_alloc_page() static")
---
 mm/page_alloc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/page_alloc.c +1 -1
index c66f1e6204c2..3e97e68aef7a 100644
@@ -3820,7 +3820,7 @@ static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
 
 #endif /* CONFIG_FAIL_PAGE_ALLOC */
 
-static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
+noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
 {
 	return __should_fail_alloc_page(gfp_mask, order);
 }
Signature · 1 line
2.26.2
Tested-by on v2 4/4
#6 lore raw

Re: [PATCH 4/4] Revert "mm/page_alloc: make should_fail_alloc_page() static"

Quoted 11 lines · Mel Gorman

From: Matteo Croce <[email protected]>

This reverts commit f7173090033c70886d925995e9dfdfb76dbb2441.

Fix an unresolved symbol error when CONFIG_DEBUG_INFO_BTF=y:

LD vmlinux
BTFIDS vmlinux
FAILED unresolved symbol should_fail_alloc_page
make: *** [Makefile:1199: vmlinux] Error 255
make: *** Deleting file 'vmlinux'

Yes! I ran into this yesterday. Your patch fixes this build failure for me, so feel free to add:

However, I should add that I'm still seeing another build failure, after fixing the above:

LD vmlinux
BTFIDS vmlinux
FAILED elf_update(WRITE): no error
make: *** [Makefile:1176: vmlinux] Error 255
make: *** Deleting file 'vmlinux'

...and un-setting CONFIG_DEBUG_INFO_BTF makes that disappear. Maybe someone who is understands the BTFIDS build step can shed some light on that; I'm not there yet. :)

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
Quoted 24 lines
Fixes f7173090033c ("mm/page_alloc: make should_fail_alloc_page() static")
---
  mm/page_alloc.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
mm/page_alloc.c +1 -1
index c66f1e6204c2..3e97e68aef7a 100644
@@ -3820,7 +3820,7 @@ static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
 
  #endif /* CONFIG_FAIL_PAGE_ALLOC */
 
-static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
+noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  {
  	return __should_fail_alloc_page(gfp_mask, order);
  }
#7 lore raw

Re: [PATCH 4/4] Revert "mm/page_alloc: make should_fail_alloc_page() static"

Cc. Jiri Olsa + Arnaldo

Quoted 19 lines · John Hubbard
Quoted 11 lines · Mel Gorman

From: Matteo Croce <[email protected]>

This reverts commit f7173090033c70886d925995e9dfdfb76dbb2441.

Fix an unresolved symbol error when CONFIG_DEBUG_INFO_BTF=y:

   LD      vmlinux
   BTFIDS  vmlinux
FAILED unresolved symbol should_fail_alloc_page
make: *** [Makefile:1199: vmlinux] Error 255
make: *** Deleting file 'vmlinux'

Yes! I ran into this yesterday. Your patch fixes this build failure for me, so feel free to add:

Quoted 6 lines

However, I should add that I'm still seeing another build failure, after fixing the above:

LD      vmlinux
BTFIDS  vmlinux
FAILED elf_update(WRITE): no error

This elf_update(WRITE) error is new to me.

Quoted 2 lines

make: *** [Makefile:1176: vmlinux] Error 255
make: *** Deleting file 'vmlinux'

It is annoying that vmlinux is deleted in this case, because I usually give Jiri the output from 'resolve_btfids -v' on vmlinux.

 $ ./tools/bpf/resolve_btfids/resolve_btfids -v vmlinux.failed

You can do:

$ git diff
scripts/link-vmlinux.sh +2 -1
index 3b261b0f74f0..02dec10a7d75 100755
@@ -302,7 +302,8 @@ cleanup()
        rm -f .tmp_symversions.lds
        rm -f .tmp_vmlinux*
        rm -f System.map
-       rm -f vmlinux
+       # rm -f vmlinux
+       mv vmlinux vmlinux.failed
        rm -f vmlinux.o
 }
 
Quoted 5 lines

...and un-setting CONFIG_DEBUG_INFO_BTF makes that disappear. Maybe someone who is understands the BTFIDS build step can shed some light on that; I'm not there yet. :)

I'm just a user/consume of output from the BTFIDS build step, I think Jiri Olsa own the tool resolve_btfids, and ACME pahole. I've hit a number of issues in the past that Jiri and ACME help resolve quickly. The most efficient solution I've found was to upgrade pahole to a newer version.

What version of pahole does your build system have?

What is your GCC version?

--Jesper

#8 lore raw

Re: [PATCH 4/4] Revert "mm/page_alloc: make should_fail_alloc_page() static"

...

Quoted 36 lines
Quoted 3 lines

LD      vmlinux
BTFIDS  vmlinux
FAILED elf_update(WRITE): no error

This elf_update(WRITE) error is new to me.

Quoted 2 lines

make: *** [Makefile:1176: vmlinux] Error 255
make: *** Deleting file 'vmlinux'

It is annoying that vmlinux is deleted in this case, because I usually give Jiri the output from 'resolve_btfids -v' on vmlinux.

  $ ./tools/bpf/resolve_btfids/resolve_btfids -v vmlinux.failed

You can do:

$ git diff
scripts/link-vmlinux.sh +2 -1
index 3b261b0f74f0..02dec10a7d75 100755
@@ -302,7 +302,8 @@ cleanup()
         rm -f .tmp_symversions.lds
         rm -f .tmp_vmlinux*
         rm -f System.map
-       rm -f vmlinux
+       # rm -f vmlinux
+       mv vmlinux vmlinux.failed
         rm -f vmlinux.o
  }
 
Quoted 6 lines

...and un-setting CONFIG_DEBUG_INFO_BTF makes that disappear. Maybe someone who is understands the BTFIDS build step can shed some light on that; I'm not there yet. :)

Quoted 7 lines

I'm just a user/consume of output from the BTFIDS build step, I think Jiri Olsa own the tool resolve_btfids, and ACME pahole.  I've hit a number of issues in the past that Jiri and ACME help resolve quickly. The most efficient solution I've found was to upgrade pahole to a newer version.

What version of pahole does your build system have?

What is your GCC version?

Just a quick answer first on the versions: this is an up to date Arch Linux system:

gcc: 11.1.0
pahole: 1.21

I'll try to get the other step done later this evening.

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
#9 lore raw

Re: [PATCH 4/4] Revert "mm/page_alloc: make should_fail_alloc_page() static"

On 7/15/21 5:04 PM, John Hubbard wrote:
...

Quoted 12 lines
Quoted 11 lines
Quoted 3 lines

...and un-setting CONFIG_DEBUG_INFO_BTF makes that disappear. Maybe someone who is understands the BTFIDS build step can shed some light on that; I'm not there yet. :)

I'm just a user/consume of output from the BTFIDS build step, I think Jiri Olsa own the tool resolve_btfids, and ACME pahole.  I've hit a number of issues in the past that Jiri and ACME help resolve quickly. The most efficient solution I've found was to upgrade pahole to a newer version.

What version of pahole does your build system have?

What is your GCC version?

Quoted 6 lines

Just a quick answer first on the versions: this is an up to date Arch Linux system:

gcc: 11.1.0
pahole: 1.21

I'll try to get the other step done later this evening.

...and...I've lost the repro completely. The only thing I changed was that I attempted to update pahole. This caused Arch Linux reinstall pahole, claiming that 1.21 is already the current version.

It acts as if there was something wrong with the pahole installation. This seems unlikely, given that the system is merely on a routine update schedule. However, that's the data I have.

If it ever comes up again I'll be able to run resolve_btfids, using your steps here, so thanks for posting those!

thanks,

Signature · 2 lines
John Hubbard
NVIDIA