Skip to contentSkip to conversation

mm: improve the output of print_page_info()

PATCH In review Yafang Shao 2 participants 9 messages Jan 15, 2021

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover mm: improve the output of print_page_info()

Before that patchset, the output of print_page_info() is

[ 6155.716018] INFO: Slab 0x000000004027dd4f objects=33 used=3 fp=0x000000008cd1579c flags=0x17ffffc0010200

While after it, the output is

[ 8843.757770] INFO: Slab 0x00000000f0e98335 objects=33 used=3 fp=0x00000000b643c7d8 flags=0x10200(slab|head)

The new one is more friendly and human readable.

Yafang Shao (2):
mm, slub: use pGp to print page flags
mm: introduce PAGE_FLAGS() to make output of page flags better

 include/linux/page-flags.h | 2 ++
 mm/slub.c                  | 5 +++--
 2 files changed, 5 insertions(+), 2 deletions(-)
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/2 mm, slub: use pGp to print page flags

As pGp has been already introduced in printk, we'd better use it to make the output human readable.

Before this change, the output is,

[ 6155.716018] INFO: Slab 0x000000004027dd4f objects=33 used=3 fp=0x000000008cd1579c flags=0x17ffffc0010200

While after this change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)
---
 mm/slub.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
mm/slub.c +3 -2
index dc5b42e..901a45a 100644
@@ -641,8 +641,9 @@ void print_tracking(struct kmem_cache *s, void *object)
 
 static void print_page_info(struct page *page)
 {
-	pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
-	       page, page->objects, page->inuse, page->freelist, page->flags);
+	pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n",
+	       page, page->objects, page->inuse, page->freelist,
+	       page->flags, &page->flags);
 
 }
 
Signature · 1 line
1.8.3.1
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] mm, slub: use pGp to print page flags

Quoted 21 lines · Yafang Shao

As pGp has been already introduced in printk, we'd better use it to make the output human readable.

Before this change, the output is,

[ 6155.716018] INFO: Slab 0x000000004027dd4f objects=33 used=3 fp=0x000000008cd1579c flags=0x17ffffc0010200

While after this change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)
---
 mm/slub.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
mm/slub.c +0 -0
index dc5b42e..901a45a 100644
@@ -641,8 +641,9 @@ void print_tracking(struct kmem_cache *s, void *object)
 

mm/slub.c void print_tracking(struct kmem_cache *s, void *object) · line 649

Quoted 10 lines
mm/slub.c +3 -2
 static void print_page_info(struct page *page)
 {
-	pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
-	       page, page->objects, page->inuse, page->freelist, page->flags);
+	pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n",
+	       page, page->objects, page->inuse, page->freelist,
+	       page->flags, &page->flags);
 
 }
 

Sounds helpful to me.

"pr_err" combined with "INFO" looks a little weird - "Oh my, something serious just happened!!!! - but, hey, just for your information ... " :)

Signature · 3 lines
Thanks,

David / dhildenb
#3 lore raw

Re: [PATCH 1/2] mm, slub: use pGp to print page flags

Quoted 30 lines · David Hildenbrand
Quoted 28 lines · Yafang Shao

As pGp has been already introduced in printk, we'd better use it to make the output human readable.

Before this change, the output is,

[ 6155.716018] INFO: Slab 0x000000004027dd4f objects=33 used=3 fp=0x000000008cd1579c flags=0x17ffffc0010200

While after this change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)
---
 mm/slub.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
mm/slub.c +3 -2
index dc5b42e..901a45a 100644
@@ -641,8 +641,9 @@ void print_tracking(struct kmem_cache *s, void *object)
 
 static void print_page_info(struct page *page)
 {
-     pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
-            page, page->objects, page->inuse, page->freelist, page->flags);
+     pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n",
+            page, page->objects, page->inuse, page->freelist,
+            page->flags, &page->flags);
 
Quoted 9 lines

mm/slub.c void print_tracking(struct kmem_cache *s, void *object) · line 649

Quoted 3 lines
mm/slub.c +0 -0
 }
 

Sounds helpful to me.

"pr_err" combined with "INFO" looks a little weird - "Oh my, something serious just happened!!!! - but, hey, just for your information ... " :)

Ah, it seems we should improve it.
There are many pr_err("INFO) in mm/slub.c...
I will change them all in a separate patch.

Quoted 1 line

Thanks for the review.

Signature · 2 lines
Thanks
Yafang
#4 lore raw

PATCH 2/2 mm: introduce PAGE_FLAGS() to make output of page flags better

In reply to Yafang Shao

There're totally __NR_PAGEFLAGS page flags, but the type of page->flags is unsigned long, that makes the value of page->flags a little misleading when it is printed to the user. We'd better print the real pages flags, instead of the whole 64bits including the random values in the useless high bits.

There're two choices to achieve that, one of which is clear the useless high bits when we initlize the page->flags, the other is don't print the high bits when it is showed to the user. The latter one is better because it is in the slow path and the performance won't be impacted.

Before that change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)

After that change, the output is,

[ 8843.757770] INFO: Slab 0x00000000f0e98335 objects=33 used=3 fp=0x00000000b643c7d8 flags=0x10200(slab|head)
---
 include/linux/page-flags.h | 2 ++
 mm/slub.c                  | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)
include/linux/page-flags.h +2 -0
index ec5d029..db5c017 100644
@@ -175,6 +175,8 @@ enum pageflags {
 	PG_reported = PG_uptodate,
 };
 
+#define PAGE_FLAGS(flags) (flags & ((1 << __NR_PAGEFLAGS) - 1))
+
 #ifndef __GENERATING_BOUNDS_H
 
 struct page;	/* forward declaration */
mm/slub.c +1 -1
index 901a45a..a93a03c 100644
@@ -643,7 +643,7 @@ static void print_page_info(struct page *page)
 {
 	pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n",
 	       page, page->objects, page->inuse, page->freelist,
-	       page->flags, &page->flags);
+	       PAGE_FLAGS(page->flags), &page->flags);
 
 }
 
Signature · 1 line
1.8.3.1
#5 lore raw

Re: [PATCH 2/2] mm: introduce PAGE_FLAGS() to make output of page flags better

Quoted 4 lines · Yafang Shao

There're totally __NR_PAGEFLAGS page flags, but the type of page->flags is unsigned long, that makes the value of page->flags a little misleading when it is printed to the user. We'd better print the real pages flags, instead of the whole 64bits including the random values in the useless high bits.

No, these are *not* random values. They include the nid, zid, and section_nr - which are helpful to have at hand when debugging, or detecting that something might be messed up there.

Quoted 2 lines

There're two choices to achieve that, one of which is clear the useless

Again, not useless.

Quoted 10 lines

high bits when we initlize the page->flags, the other is don't print the high bits when it is showed to the user. The latter one is better because it is in the slow path and the performance won't be impacted.

Before that change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)

After that change, the output is,

[ 8843.757770] INFO: Slab 0x00000000f0e98335 objects=33 used=3 fp=0x00000000b643c7d8 flags=0x10200(slab|head)

Nack to the current approach. If you're going to strip this information, you should expose it differently. E.g., printing page_zonenum() or page_to_nid(). But still, then we might lose valuable information of bits stored in there that shouldn't have been set.

Quoted 33 lines
---
 include/linux/page-flags.h | 2 ++
 mm/slub.c                  | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)
include/linux/page-flags.h +2 -0
index ec5d029..db5c017 100644
@@ -175,6 +175,8 @@ enum pageflags {
 	PG_reported = PG_uptodate,
 };
 
+#define PAGE_FLAGS(flags) (flags & ((1 << __NR_PAGEFLAGS) - 1))
+
 #ifndef __GENERATING_BOUNDS_H
 
 struct page;	/* forward declaration */
mm/slub.c +1 -1
index 901a45a..a93a03c 100644
@@ -643,7 +643,7 @@ static void print_page_info(struct page *page)
 {
 	pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n",
 	       page, page->objects, page->inuse, page->freelist,
-	       page->flags, &page->flags);
+	       PAGE_FLAGS(page->flags), &page->flags);
 
 }
 
Signature · 3 lines
Thanks,

David / dhildenb
#6 lore raw

Re: [PATCH 2/2] mm: introduce PAGE_FLAGS() to make output of page flags better

Quoted 11 lines · David Hildenbrand
Quoted 4 lines · Yafang Shao

There're totally __NR_PAGEFLAGS page flags, but the type of page->flags is unsigned long, that makes the value of page->flags a little misleading when it is printed to the user. We'd better print the real pages flags, instead of the whole 64bits including the random values in the useless high bits.

No, these are *not* random values. They include the nid, zid, and section_nr - which are helpful to have at hand when debugging, or detecting that something might be messed up there.

Thanks for the explanation. I just noticed the page-flags layout in page-flags-layout.h.

Quoted 12 lines
Quoted 2 lines

There're two choices to achieve that, one of which is clear the useless

Again, not useless.

Quoted 7 lines

high bits when we initlize the page->flags, the other is don't print the high bits when it is showed to the user. The latter one is better because it is in the slow path and the performance won't be impacted.

Before that change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)
Quoted 9 lines
Quoted 3 lines

After that change, the output is,

[ 8843.757770] INFO: Slab 0x00000000f0e98335 objects=33 used=3 fp=0x00000000b643c7d8 flags=0x10200(slab|head)

Nack to the current approach. If you're going to strip this information, you should expose it differently. E.g., printing page_zonenum() or page_to_nid(). But still, then we might lose valuable information of bits stored in there that shouldn't have been set.

How about changing the implementation of pGp in printk() ? In the new implementation of pGp we can dump the full information of page->flags, rather than the flag's name only.

For example,
0xXXXXXXXX(node n, nid n, ..., slab|head)

That will make it easier to understand, as it is not easy to look into the detail of page-flags layout.

Quoted 40 lines
Quoted 25 lines
---
 include/linux/page-flags.h | 2 ++
 mm/slub.c                  | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)
include/linux/page-flags.h +2 -0
index ec5d029..db5c017 100644
@@ -175,6 +175,8 @@ enum pageflags {
      PG_reported = PG_uptodate,
 };
 
+#define PAGE_FLAGS(flags) (flags & ((1 << __NR_PAGEFLAGS) - 1))
+
 #ifndef __GENERATING_BOUNDS_H
 
 struct page; /* forward declaration */
mm/slub.c +0 -0
index 901a45a..a93a03c 100644

mm/slub.c static void print_page_info(struct page *page) · line 649

Quoted 8 lines
mm/slub.c +1 -1
@@ -643,7 +643,7 @@ static void print_page_info(struct page *page)
 {
      pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n",
             page, page->objects, page->inuse, page->freelist,
-            page->flags, &page->flags);
+            PAGE_FLAGS(page->flags), &page->flags);
 
 }
 

--
Thanks,

David / dhildenb

Signature · 2 lines
Thanks
Yafang
#7 lore raw

Re: [PATCH 2/2] mm: introduce PAGE_FLAGS() to make output of page flags better

Quoted 42 lines · Yafang Shao
Quoted 11 lines · David Hildenbrand
Quoted 4 lines · Yafang Shao

There're totally __NR_PAGEFLAGS page flags, but the type of page->flags is unsigned long, that makes the value of page->flags a little misleading when it is printed to the user. We'd better print the real pages flags, instead of the whole 64bits including the random values in the useless high bits.

No, these are *not* random values. They include the nid, zid, and section_nr - which are helpful to have at hand when debugging, or detecting that something might be messed up there.

Thanks for the explanation. I just noticed the page-flags layout in page-flags-layout.h.

Quoted 12 lines
Quoted 2 lines

There're two choices to achieve that, one of which is clear the useless

Again, not useless.

Quoted 7 lines

high bits when we initlize the page->flags, the other is don't print the high bits when it is showed to the user. The latter one is better because it is in the slow path and the performance won't be impacted.

Before that change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)
Quoted 9 lines
Quoted 3 lines

After that change, the output is,

[ 8843.757770] INFO: Slab 0x00000000f0e98335 objects=33 used=3 fp=0x00000000b643c7d8 flags=0x10200(slab|head)

Nack to the current approach. If you're going to strip this information, you should expose it differently. E.g., printing page_zonenum() or page_to_nid(). But still, then we might lose valuable information of bits stored in there that shouldn't have been set.

How about changing the implementation of pGp in printk() ? In the new implementation of pGp we can dump the full information of page->flags, rather than the flag's name only.

Quoted 5 lines

For example,
0xXXXXXXXX(node n, nid n, ..., slab|head)

That will make it easier to understand, as it is not easy to look into the detail of page-flags layout.

Not completely opposed to this. Just keep in mind that the information stored in the high bits differs per configuration. See include/linux/page-flags-layout.h

So to dump reliably, you would have to consider all different flavors, looking at SECTIONS_WIDTH, NODES_WIDTH, ZONES_WIDTH, LAST_CPUPID_WIDTH, KASAN_TAG_WIDTH

Might be helpful for other users as well, like in mm/memory-failure.c

Signature · 3 lines
Thanks,

David / dhildenb
#8 lore raw

Re: [PATCH 2/2] mm: introduce PAGE_FLAGS() to make output of page flags better

Quoted 54 lines · David Hildenbrand
Quoted 42 lines · Yafang Shao
Quoted 11 lines · David Hildenbrand
Quoted 4 lines · Yafang Shao

There're totally __NR_PAGEFLAGS page flags, but the type of page->flags is unsigned long, that makes the value of page->flags a little misleading when it is printed to the user. We'd better print the real pages flags, instead of the whole 64bits including the random values in the useless high bits.

No, these are *not* random values. They include the nid, zid, and section_nr - which are helpful to have at hand when debugging, or detecting that something might be messed up there.

Thanks for the explanation. I just noticed the page-flags layout in page-flags-layout.h.

Quoted 12 lines
Quoted 2 lines

There're two choices to achieve that, one of which is clear the useless

Again, not useless.

Quoted 7 lines

high bits when we initlize the page->flags, the other is don't print the high bits when it is showed to the user. The latter one is better because it is in the slow path and the performance won't be impacted.

Before that change, the output is,

[ 8846.517809] INFO: Slab 0x00000000f42a2c60 objects=33 used=3 fp=0x0000000060d32ca8 flags=0x17ffffc0010200(slab|head)
Quoted 9 lines
Quoted 3 lines

After that change, the output is,

[ 8843.757770] INFO: Slab 0x00000000f0e98335 objects=33 used=3 fp=0x00000000b643c7d8 flags=0x10200(slab|head)

Nack to the current approach. If you're going to strip this information, you should expose it differently. E.g., printing page_zonenum() or page_to_nid(). But still, then we might lose valuable information of bits stored in there that shouldn't have been set.

How about changing the implementation of pGp in printk() ? In the new implementation of pGp we can dump the full information of page->flags, rather than the flag's name only.

Quoted 5 lines

For example,
0xXXXXXXXX(node n, nid n, ..., slab|head)

That will make it easier to understand, as it is not easy to look into the detail of page-flags layout.

Not completely opposed to this. Just keep in mind that the information stored in the high bits differs per configuration. See include/linux/page-flags-layout.h

Quoted 6 lines

So to dump reliably, you would have to consider all different flavors, looking at SECTIONS_WIDTH, NODES_WIDTH, ZONES_WIDTH, LAST_CPUPID_WIDTH, KASAN_TAG_WIDTH

Might be helpful for other users as well, like in mm/memory-failure.c

Thanks for the suggestion.
I will think about it.

Signature · 2 lines
Thanks
Yafang