mainline build failure for loongarch allmodconfig with gcc-12

Hi All,

I have been trying to build loongarch as part of my nightly builds, and I can build loongson3_defconfig without any error. But allmodconfig fails with the error:

In function '__cmpxchg',
inlined from 'ssh_seq_next' at drivers/platform/surface/aggregator/controller.c:61:9,
inlined from 'ssam_request_write_data' at drivers/platform/surface/aggregator/controller.c:1483:8, inlined from 'ssam_request_sync_with_buffer' at drivers/platform/surface/aggregator/controller.c:1761:8:
././include/linux/compiler_types.h:354:45: error: call to '__compiletime_assert_9' declared with attribute error: BUILD_BUG failed
354 | _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
| ^
././include/linux/compiler_types.h:335:25: note: in definition of macro '__compiletime_assert'

  335 |                         prefix ## suffix();                             \
      |                         ^~~~~~
././include/linux/compiler_types.h:354:9: note: in expansion of macro '_compiletime_assert'
  354 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
      |         ^~~~~~~~~~~~~~~~~~~
./include/linux/build_bug.h:39:37: note: in expansion of macro 'compiletime_assert'
   39 | #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
      |                                     ^~~~~~~~~~~~~~~~~~
./include/linux/build_bug.h:59:21: note: in expansion of macro 'BUILD_BUG_ON_MSG'
   59 | #define BUILD_BUG() BUILD_BUG_ON_MSG(1, "BUILD_BUG failed")
      |                     ^~~~~~~~~~~~~~~~
./arch/loongarch/include/asm/cmpxchg.h:83:17: note: in expansion of macro 'BUILD_BUG'
   83 |                 BUILD_BUG();
      |                 ^~~~~~~~~

I will be happy to test any patch or provide any extra log if needed.

Regards
Sudip

Looks like ssh_seq_next() wants to do an atomic cmpxchg() on a single byte value, and the Loongarch implementation only does 4- and 8-byte versions.

It looks like loongarch - from its MIPS heritage - inherited the "we can't do atomics on byte variables", so that it needs the same strange "do bytes as word accesses with mask-and-shifts".

For MIPS, the code is in __xchg_small() in arch/mips/kernel/cmpxchg.c.

Alpha has something similar, except it's all done in inline asm in arch/alpha/include/asm/xchg.h (look for "____cmpxchg(_u8," in there.

Of course, we could just add a Kconfig variable like "ARCH_LACKS_BYTE_ATOMICS" and make that driver depend on it not being true, and just have Loongarch set it.

But I think loongarch should just implement the byte masking stuff. Particularly since I suspect it can just copy the MIPS code as-is.

Linus

Hi, Linus,

Yes, I agree. There is already a proposed patch to do this [1]. It was discussed for a long time about the "forward progress", but I think that problem is solved because LoongArch has "exclusive access (with timeout) of ll" and "random delay of sc" in hardware. So I hope this patch can be accepted now.

https://lore.kernel.org/loongarch/CAJF2gTQBjetiA1eDaXhBsiEmRYCdOAAWWAGcwVjhZTBYH5BpGQ@mail.gmail.com/T/#m9df7f86d9509cec23a66a0304152377e1070b62e

Huacai