Skip to contentSkip to conversation

blk-iocost stringop-overread warning workaround

RFC PATCH In review Arnd Bergmann 2 participants 2 messages Sep 20, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 blk-iocost stringop-overread warning workaround

From: Arnd Bergmann <[email protected]>

In some randconfig builds with gcc-11, I get a warning from the fortified string helpers:

In function 'memcpy',
inlined from 'ioc_cost_model_write' at block/blk-iocost.c:3345:2:
include/linux/fortify-string.h:20:33: error: '__builtin_memcpy' reading 48 bytes from a region of size 0 [-Werror=stringop-overread]

   20 | #define __underlying_memcpy     __builtin_memcpy
      |                                 ^
include/linux/fortify-string.h:191:16: note: in expansion of macro '__underlying_memcpy'
  191 |         return __underlying_memcpy(p, q, size);
      |                ^~~~~~~~~~~~~~~~~~~

I don't see anything wrong in the code itself, so I suspect it's gcc doing something weird again. The only way I could find to make this warning go away is to hide the object using the RELOC_HIDE() macro, but this is really ugly and I hope someone has a better idea.

---
 block/blk-iocost.c | 2 ++
 1 file changed, 2 insertions(+)
block/blk-iocost.c +2 -0
index b3880e4ba22a..51f641840ed9 100644
@@ -3173,6 +3173,7 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 		ioc = q_to_ioc(bdev->bd_disk->queue);
 	}
 
+	ioc = RELOC_HIDE(ioc, 0);
 	spin_lock_irq(&ioc->lock);
 	memcpy(qos, ioc->params.qos, sizeof(qos));
 	enable = ioc->enabled;
@@ -3340,6 +3341,7 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
 		ioc = q_to_ioc(bdev->bd_disk->queue);
 	}
 
+	ioc = RELOC_HIDE(ioc, 0);
 	spin_lock_irq(&ioc->lock);
 	memcpy(u, ioc->params.i_lcoefs, sizeof(u));
 	user = ioc->user_cost_model;
Signature · 1 line
2.29.2
#1 lore raw
Quoted 15 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

In some randconfig builds with gcc-11, I get a warning from the fortified string helpers:

In function 'memcpy',
inlined from 'ioc_cost_model_write' at block/blk-iocost.c:3345:2:
include/linux/fortify-string.h:20:33: error: '__builtin_memcpy' reading 48 bytes from a region of size 0 [-Werror=stringop-overread]

   20 | #define __underlying_memcpy     __builtin_memcpy
      |                                 ^
include/linux/fortify-string.h:191:16: note: in expansion of macro '__underlying_memcpy'
  191 |         return __underlying_memcpy(p, q, size);
      |                ^~~~~~~~~~~~~~~~~~~
Quoted 8 lines

I don't see anything wrong in the code itself, so I suspect it's gcc doing something weird again. The only way I could find to make this warning go away is to hide the object using the RELOC_HIDE() macro, but this is really ugly and I hope someone has a better idea.

Please disregard my patch for now. After looking closer, I found that I had applied a patch that accidentally removed the

KBUILD_CFLAGS += -fno-delete-null-pointer-checks

line from the top-level Makefile. If I put that line back, the warning disappears.

Arnd