Skip to contentSkip to conversation

btrfs: avoid link error with CONFIG_NO_AUTO_INLINE

PATCH In review Arnd Bergmann 5 participants 10 messages Nov 3, 2018 – Nov 6, 2018 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 btrfs: avoid link error with CONFIG_NO_AUTO_INLINE

On 32-bit ARM with gcc-8, I see a link error with the addition of the CONFIG_NO_AUTO_INLINE option:

fs/btrfs/super.o: In function `btrfs_statfs':
super.c:(.text+0x67b8): undefined reference to __aeabi_uldivmod' super.c:(.text+0x67fc): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6858): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6920): undefined reference to __aeabi_uldivmod' super.c:(.text+0x693c): undefined reference to __aeabi_uldivmod' fs/btrfs/super.o:super.c:(.text+0x6958): more undefined references to __aeabi_uldivmod' follow

So far this is the only file that shows the behavior, so I'd propose to just work around it by marking the functions as 'static inline' that normally get inlined here.

The reference to __aeabi_uldivmod comes from a div_u64() which has an optimization for a constant division that uses a straight '/' operator when the result should be known to the compiler. My interpretation is that as we turn off inlining, gcc still expects the result to be constant but fails to use that constant value.

Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable compiler auto-inlining")
---
 fs/btrfs/super.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/btrfs/super.c +3 -3
index c3c1e7bee49d..b7af0b8936ad 100644
@@ -1922,7 +1922,7 @@ static int btrfs_remount(struct super_block *sb, int *flags,
 }
 
 /* Used to sort the devices by max_avail(descending sort) */
-static int btrfs_cmp_device_free_bytes(const void *dev_info1,
+static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
 				       const void *dev_info2)
 {
 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
@@ -1951,8 +1951,8 @@ static inline void btrfs_descending_sort_devices(
  * The helper to calc the free space on the devices that can be used to store
  * file data.
  */
-static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
-				       u64 *free_bytes)
+static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
+					      u64 *free_bytes)
 {
 	struct btrfs_device_info *devices_info;
 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
Signature · 1 line
2.18.0
#1 lore raw
Quoted 15 lines · Arnd Bergmann

On 32-bit ARM with gcc-8, I see a link error with the addition of the CONFIG_NO_AUTO_INLINE option:

fs/btrfs/super.o: In function `btrfs_statfs':
super.c:(.text+0x67b8): undefined reference to __aeabi_uldivmod' super.c:(.text+0x67fc): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6858): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6920): undefined reference to __aeabi_uldivmod' super.c:(.text+0x693c): undefined reference to __aeabi_uldivmod' fs/btrfs/super.o:super.c:(.text+0x6958): more undefined references to __aeabi_uldivmod' follow

So far this is the only file that shows the behavior, so I'd propose to just work around it by marking the functions as 'static inline' that normally get inlined here.

Quoted 5 lines

The reference to __aeabi_uldivmod comes from a div_u64() which has an optimization for a constant division that uses a straight '/' operator when the result should be known to the compiler. My interpretation is that as we turn off inlining, gcc still expects the result to be constant but fails to use that constant value.

I read this as "this is a compiler bug", no ? So you are providing a hack around a compiler bug?

Quoted 33 lines
Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable compiler auto-inlining")
---
 fs/btrfs/super.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/btrfs/super.c +3 -3
index c3c1e7bee49d..b7af0b8936ad 100644
@@ -1922,7 +1922,7 @@ static int btrfs_remount(struct super_block *sb, int *flags,
 }
 
 /* Used to sort the devices by max_avail(descending sort) */
-static int btrfs_cmp_device_free_bytes(const void *dev_info1,
+static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
 				       const void *dev_info2)
 {
 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
@@ -1951,8 +1951,8 @@ static inline void btrfs_descending_sort_devices(
  * The helper to calc the free space on the devices that can be used to store
  * file data.
  */
-static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
-				       u64 *free_bytes)
+static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
+					      u64 *free_bytes)
 {
 	struct btrfs_device_info *devices_info;
 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
#2 lore raw
Quoted 17 lines · Nikolay Borisov
Quoted 16 lines · Arnd Bergmann

On 32-bit ARM with gcc-8, I see a link error with the addition of the CONFIG_NO_AUTO_INLINE option:

fs/btrfs/super.o: In function `btrfs_statfs':
super.c:(.text+0x67b8): undefined reference to __aeabi_uldivmod' super.c:(.text+0x67fc): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6858): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6920): undefined reference to __aeabi_uldivmod' super.c:(.text+0x693c): undefined reference to __aeabi_uldivmod' fs/btrfs/super.o:super.c:(.text+0x6958): more undefined references to __aeabi_uldivmod' follow

So far this is the only file that shows the behavior, so I'd propose to just work around it by marking the functions as 'static inline' that normally get inlined here.

Quoted 8 lines
Quoted 5 lines

The reference to __aeabi_uldivmod comes from a div_u64() which has an optimization for a constant division that uses a straight '/' operator when the result should be known to the compiler. My interpretation is that as we turn off inlining, gcc still expects the result to be constant but fails to use that constant value.

I read this as "this is a compiler bug", no ? So you are providing a hack around a compiler bug?

Mostly, yes. The do_div() macro is really pushing the boundaries of what we can expect the compiler to do in terms of optimizations, and we've had problems with it in the past.

IIRC the gcc developers would not classify this as a bug because the result of __builtin_constant_p() is not guaranteed to work the way we expect, it just does so most of the time.

Arnd

#3 lore raw
In reply to Arnd Bergmann
Quoted 11 lines · Arnd Bergmann

On 32-bit ARM with gcc-8, I see a link error with the addition of the CONFIG_NO_AUTO_INLINE option:

fs/btrfs/super.o: In function `btrfs_statfs':
super.c:(.text+0x67b8): undefined reference to __aeabi_uldivmod' super.c:(.text+0x67fc): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6858): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6920): undefined reference to __aeabi_uldivmod' super.c:(.text+0x693c): undefined reference to __aeabi_uldivmod' fs/btrfs/super.o:super.c:(.text+0x6958): more undefined references to __aeabi_uldivmod' follow

Quoted 3 lines

So far this is the only file that shows the behavior, so I'd propose to just work around it by marking the functions as 'static inline' that normally get inlined here.

As a workaround it looks OK, but it's definitely not the root cause.

Quoted 6 lines

The reference to __aeabi_uldivmod comes from a div_u64() which has an optimization for a constant division that uses a straight '/' operator when the result should be known to the compiler. My interpretation is that as we turn off inlining, gcc still expects the result to be constant but fails to use that constant value.

It looks more like a bug in div_u64() optimization.

Despite this file in btrfs, did you hit the same problem for any other file?

Quoted 3 lines
Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable compiler auto-inlining")

I can't find it in the mainline kernel, is the commit hash correct? If not merged, we should still has a chance to further polish that patch.

Thanks,
Qu

Quoted 30 lines
---
 fs/btrfs/super.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/btrfs/super.c +3 -3
index c3c1e7bee49d..b7af0b8936ad 100644
@@ -1922,7 +1922,7 @@ static int btrfs_remount(struct super_block *sb, int *flags,
 }
 
 /* Used to sort the devices by max_avail(descending sort) */
-static int btrfs_cmp_device_free_bytes(const void *dev_info1,
+static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
 				       const void *dev_info2)
 {
 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
@@ -1951,8 +1951,8 @@ static inline void btrfs_descending_sort_devices(
  * The helper to calc the free space on the devices that can be used to store
  * file data.
  */
-static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
-				       u64 *free_bytes)
+static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
+					      u64 *free_bytes)
 {
 	struct btrfs_device_info *devices_info;
 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
#4 lore raw
Quoted 28 lines · Qu Wenruo
Quoted 12 lines · Arnd Bergmann

On 32-bit ARM with gcc-8, I see a link error with the addition of the CONFIG_NO_AUTO_INLINE option:

fs/btrfs/super.o: In function `btrfs_statfs':
super.c:(.text+0x67b8): undefined reference to __aeabi_uldivmod' super.c:(.text+0x67fc): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6858): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6920): undefined reference to __aeabi_uldivmod' super.c:(.text+0x693c): undefined reference to __aeabi_uldivmod' fs/btrfs/super.o:super.c:(.text+0x6958): more undefined references to __aeabi_uldivmod' follow

Quoted 3 lines

So far this is the only file that shows the behavior, so I'd propose to just work around it by marking the functions as 'static inline' that normally get inlined here.

As a workaround it looks OK, but it's definitely not the root cause.

Quoted 6 lines

The reference to __aeabi_uldivmod comes from a div_u64() which has an optimization for a constant division that uses a straight '/' operator when the result should be known to the compiler. My interpretation is that as we turn off inlining, gcc still expects the result to be constant but fails to use that constant value.

Quoted 4 lines

It looks more like a bug in div_u64() optimization.

Despite this file in btrfs, did you hit the same problem for any other file?

Not this time. I've done a creduce on the file and got to this code

struct kstatfs {
u64 f_bfree;
};
btrfs_calc_avail_data_space(p1) {}
btrfs_statfs(struct kstatfs *p1) {
u64 d = 0;
unsigned e = 1;
for (; a;)

    e = btrfs_bg_type_to_factor();

p1->f_bfree = div_u64(0, e) >> c;
__asm__("");
div_u64(d, e);
b = btrfs_calc_avail_data_space(&d);
}

Looking at the assembler code produced by this, it seems to be the same thing that we dealt with in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72785

Quoted 6 lines
Quoted 3 lines
Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable

compiler auto-inlining")

I can't find it in the mainline kernel, is the commit hash correct? If not merged, we should still has a chance to further polish that patch.

It's in linux-next.

Arnd

#5 lore raw
Quoted 8 lines · Arnd Bergmann
Quoted 6 lines
Quoted 3 lines
Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable

compiler auto-inlining")

I can't find it in the mainline kernel, is the commit hash correct? If not merged, we should still has a chance to further polish that patch.

It's in linux-next.

I can't find it in current linux-next either, the final reference in
Fixes: must refer to a commit in Linus' tree.

You can take this fix with the patch that introduces the config option (ack for that) in case merging through the btrfs tree would be too late for it (ie. no common base for the git trees containg the new code and fix).

Or I can take it through btrfs tree in 4.20-rc cycle. In both cases the
Fixes: does not need to be there.

#6 lore raw
Quoted 5 lines · David Sterba
Quoted 4 lines · Arnd Bergmann
Quoted 4 lines
Quoted 3 lines
Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable

compiler auto-inlining")

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

I can't find it in the mainline kernel, is the commit hash correct? If not merged, we should still has a chance to further polish that patch.

It's in linux-next.

I can't find it in current linux-next either, the final reference in
Fixes: must refer to a commit in Linus' tree.

Ah, right, it got rebased. The commit ID in today's linux-next is 917fad29febd. Most trees in linux-next don't rebase so this would not be an issue, but you are right that this one clearly did, so the line is wrong here.

The commit is now delayed until 4.21 I assume.

Quoted 7 lines

You can take this fix with the patch that introduces the config option (ack for that) in case merging through the btrfs tree would be too late for it (ie. no common base for the git trees containg the new code and fix).

Or I can take it through btrfs tree in 4.20-rc cycle. In both cases the
Fixes: does not need to be there.

Please take it through the btrfs tree. Let me know if you need me to extend the changelog to explain that situation better.

Arnd

#7 lore raw
Quoted 27 lines · Arnd Bergmann
Quoted 5 lines · David Sterba
Quoted 4 lines · Arnd Bergmann
Quoted 4 lines
Quoted 3 lines
Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable

compiler auto-inlining")

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

I can't find it in the mainline kernel, is the commit hash correct? If not merged, we should still has a chance to further polish that patch.

It's in linux-next.

I can't find it in current linux-next either, the final reference in
Fixes: must refer to a commit in Linus' tree.

Ah, right, it got rebased. The commit ID in today's linux-next is 917fad29febd. Most trees in linux-next don't rebase so this would not be an issue, but you are right that this one clearly did, so the line is wrong here.

The commit is now delayed until 4.21 I assume.

Quoted 5 lines

You can take this fix with the patch that introduces the config option (ack for that) in case merging through the btrfs tree would be too late for it (ie. no common base for the git trees containg the new code and fix).

Quoted 5 lines
Quoted 2 lines

Or I can take it through btrfs tree in 4.20-rc cycle. In both cases the
Fixes: does not need to be there.

Please take it through the btrfs tree. Let me know if you need me to extend the changelog to explain that situation better.

Ok, I'll add it to my tree, the changelog is good, thanks.

Reviewed-by on v1 1/1
#8 lore raw
In reply to Arnd Bergmann
Quoted 21 lines · Arnd Bergmann

On 32-bit ARM with gcc-8, I see a link error with the addition of the CONFIG_NO_AUTO_INLINE option:

fs/btrfs/super.o: In function `btrfs_statfs':
super.c:(.text+0x67b8): undefined reference to __aeabi_uldivmod' super.c:(.text+0x67fc): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6858): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6920): undefined reference to __aeabi_uldivmod' super.c:(.text+0x693c): undefined reference to __aeabi_uldivmod' fs/btrfs/super.o:super.c:(.text+0x6958): more undefined references to __aeabi_uldivmod' follow

So far this is the only file that shows the behavior, so I'd propose to just work around it by marking the functions as 'static inline' that normally get inlined here.

The reference to __aeabi_uldivmod comes from a div_u64() which has an optimization for a constant division that uses a straight '/' operator when the result should be known to the compiler. My interpretation is that as we turn off inlining, gcc still expects the result to be constant but fails to use that constant value.

Quoted 3 lines
Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable compiler auto-inlining")

I spoke with Arnd on irc and am fine with taking this patch as-is if btrfs is the sole offender. So:

Quoted 29 lines
---
 fs/btrfs/super.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/btrfs/super.c +3 -3
index c3c1e7bee49d..b7af0b8936ad 100644
@@ -1922,7 +1922,7 @@ static int btrfs_remount(struct super_block *sb, int *flags,
 }
 
 /* Used to sort the devices by max_avail(descending sort) */
-static int btrfs_cmp_device_free_bytes(const void *dev_info1,
+static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
 				       const void *dev_info2)
 {
 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
@@ -1951,8 +1951,8 @@ static inline void btrfs_descending_sort_devices(
  * The helper to calc the free space on the devices that can be used to store
  * file data.
  */
-static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
-				       u64 *free_bytes)
+static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
+					      u64 *free_bytes)
 {
 	struct btrfs_device_info *devices_info;
 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
Reviewed-by on v1 1/1
#9 lore raw
In reply to Arnd Bergmann

Thanks for this fix.

Quoted 55 lines · Arnd Bergmann

On 32-bit ARM with gcc-8, I see a link error with the addition of the CONFIG_NO_AUTO_INLINE option:

fs/btrfs/super.o: In function `btrfs_statfs':
super.c:(.text+0x67b8): undefined reference to __aeabi_uldivmod' super.c:(.text+0x67fc): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6858): undefined reference to __aeabi_uldivmod' super.c:(.text+0x6920): undefined reference to __aeabi_uldivmod' super.c:(.text+0x693c): undefined reference to __aeabi_uldivmod' fs/btrfs/super.o:super.c:(.text+0x6958): more undefined references to __aeabi_uldivmod' follow

So far this is the only file that shows the behavior, so I'd propose to just work around it by marking the functions as 'static inline' that normally get inlined here.

The reference to __aeabi_uldivmod comes from a div_u64() which has an optimization for a constant division that uses a straight '/' operator when the result should be known to the compiler. My interpretation is that as we turn off inlining, gcc still expects the result to be constant but fails to use that constant value.

Cc Changbin Du <[email protected]>
Fixes 943b8435c3bd ("kernel hacking: add a config option to disable compiler auto-inlining")
---
 fs/btrfs/super.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/btrfs/super.c +3 -3
index c3c1e7bee49d..b7af0b8936ad 100644
@@ -1922,7 +1922,7 @@ static int btrfs_remount(struct super_block *sb, int *flags,
 }
 
 /* Used to sort the devices by max_avail(descending sort) */
-static int btrfs_cmp_device_free_bytes(const void *dev_info1,
+static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
 				       const void *dev_info2)
 {
 	if (((struct btrfs_device_info *)dev_info1)->max_avail >
@@ -1951,8 +1951,8 @@ static inline void btrfs_descending_sort_devices(
  * The helper to calc the free space on the devices that can be used to store
  * file data.
  */
-static int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
-				       u64 *free_bytes)
+static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
+					      u64 *free_bytes)
 {
 	struct btrfs_device_info *devices_info;
 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
Signature · 1 line
2.18.0
Signature · 2 lines
Thanks,
Changbin Du