2024-04-05 14:30:28

by Arnd Bergmann

Subject: [PATCH] mtd: diskonchip: work around ubsan link failure

From: Arnd Bergmann <[email protected]>

I ran into a randconfig build failure with UBSAN using gcc-13.2:

arm-linux-gnueabi-ld: error: unplaced orphan section `.bss..Lubsan_data31' from `drivers/mtd/nand/raw/diskonchip.o'

I'm not entirely sure what is going on here, but I suspect this has something
to do with the check for the end of the doc_locations[] array that contains
an (unsigned long)0xffffffff element, which is compared against the signed
(int)0xffffffff. If this is the case, we should get a runtime check for
undefined behavior, but we instead get an unexpected build-time error.

I would have expected this to work fine on 32-bit architectures despite the
signed integer overflow, though on 64-bit architectures this likely won't
ever work.

Changing the contition to instead check for the size of the array makes the
code safe everywhere and avoids the ubsan check that leads to the link
error. The loop code goes back to before 2.6.12.

Cc: [email protected]
Signed-off-by: Arnd Bergmann <[email protected]>
---
drivers/mtd/nand/raw/diskonchip.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/mtd/nand/raw/diskonchip.c b/drivers/mtd/nand/raw/diskonchip.c
index 5243fab9face..8db7fc424571 100644
--- a/drivers/mtd/nand/raw/diskonchip.c
+++ b/drivers/mtd/nand/raw/diskonchip.c
@@ -53,7 +53,7 @@ static unsigned long doc_locations[] __initdata = {
0xe8000, 0xea000, 0xec000, 0xee000,
#endif
#endif
- 0xffffffff };
+};

static struct mtd_info *doclist = NULL;

@@ -1554,7 +1554,7 @@ static int __init init_nanddoc(void)
if (ret < 0)
return ret;
} else {
- for (i = 0; (doc_locations[i] != 0xffffffff); i++) {
+ for (i = 0; i < ARRAY_SIZE(doc_locations); i++) {
doc_probe(doc_locations[i]);
}
}
--
2.39.2

2024-04-09 06:41:49

by Miquel Raynal

Subject: Re: [PATCH] mtd: diskonchip: work around ubsan link failure

On Fri, 2024-04-05 at 14:30:04 UTC, Arnd Bergmann wrote:
> From: Arnd Bergmann <[email protected]>
>
> I ran into a randconfig build failure with UBSAN using gcc-13.2:
>
> arm-linux-gnueabi-ld: error: unplaced orphan section `.bss..Lubsan_data31' from `drivers/mtd/nand/raw/diskonchip.o'
>
> I'm not entirely sure what is going on here, but I suspect this has something
> to do with the check for the end of the doc_locations[] array that contains
> an (unsigned long)0xffffffff element, which is compared against the signed
> (int)0xffffffff. If this is the case, we should get a runtime check for
> undefined behavior, but we instead get an unexpected build-time error.
>
> I would have expected this to work fine on 32-bit architectures despite the
> signed integer overflow, though on 64-bit architectures this likely won't
> ever work.
>
> Changing the contition to instead check for the size of the array makes the
> code safe everywhere and avoids the ubsan check that leads to the link
> error. The loop code goes back to before 2.6.12.
>
> Cc: [email protected]
> Signed-off-by: Arnd Bergmann <[email protected]>

Applied to https://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux.git mtd/fixes, thanks.

Miquel