Re: [PATCH v2 00/33] locking/atomic: convert all architectures to ARCH_ATOMIC

[Randy Dunlap]

On 6/27/21 2:47 PM, Randy Dunlap wrote:
> On 6/18/21 1:48 AM, Mark Rutland wrote:
>> On Fri, Jun 04, 2021 at 10:56:16PM -0700, Randy Dunlap wrote:
>>> On 5/25/21 7:01 AM, Mark Rutland wrote:
>>>> This series (based on v5.13-rc2) converts all architectures to
>>>> ARCH_ATOMIC. This will allow the use of instrumented atomics on all
>>>> architectures (e.g. for KASAN and similar), and simplifies the core
>>>> atomic code (which should allow for easier rework of the fallbacks and
>>>> other bits in future).
>>
>> [...]
>>
>>> Hi Mark,
>>> Sorry for the late reply. 
>>
>> Hi Randy,
>>
>> Likewise, apologies in the delay in getting to this!
>>
>>> I was just trying to update a patch
>>> to arch/sh/include/asm/cmpxchg.h, in its xchg() macro:
>>>
>>> https://lore.kernel.org/lkml/20210602231443.4670-2-rdunlap@xxxxxxxxxxxxx/
>>>
>>> The patch simply converts xchg() to a GCC statement expression to
>>> eliminate a build warning.
> 
> Hm, with your locking/atomic patch series applied (in linux-next), I can
> no longer make arch/sh/ get this build warning:
> 
> ../fs/ocfs2/file.c: In function 'ocfs2_file_write_iter':
> ../arch/sh/include/asm/cmpxchg.h:49:3: warning: value computed is not used [-Wunused-value]
>    49 |  ((__typeof__(*(ptr)))__xchg((ptr),(unsigned long)(x), sizeof(*(ptr))))
> 
> 
> so I will go ahead with the rest of my arch/sh/ patches and then contemplate
> what to do about this one.
> 
> 
>>> Arnd has done this for m68k and I have done it for sparc in the past.
>>>
>>> Is there any (good) reason that all versions of arch_xchg() are not
>>> statement expressions?  In this patch series, they seem to be quite
>>> mixed (as they were before this patch series). I count 11 arches
>>> that use a statement expression and 4 that do not (including arch/sh/).
>>
>> Largely I tried to make the minimal change from what was there before,
>> and I didn't have any specific reason to either use or avoid statement
>> expressions.
>>
>> This series has been queued in the tip tree's locking/core branch for a
>> while now, but we could spin a patch atop. Do you want to spin a patch
>> to convert the remaining 4 architectures in one go?
> 
> I'll look at the 4 remaining arches later..
> 

Hi Mark,

I checked xchg(), __xchg(), and cmpxchg() in all
arch/*/include/asm/cmpxchg.h.  They are use static inline
functions or statement expressions so I don't see any need
for follow-ups to fix warnings like this (old) one, which
I cannot cause with your series applied:

> ../fs/ocfs2/file.c: In function 'ocfs2_file_write_iter':
> ../arch/sh/include/asm/cmpxchg.h:49:3: warning: value computed is not used [-Wunused-value]
>    49 |  ((__typeof__(*(ptr)))__xchg((ptr),(unsigned long)(x), sizeof(*(ptr))))


However, something in arch/arc/ did look suspicious so I decided to
try an ARC allmodconfig build, where I did see a few errors FYI:


  CC      drivers/iommu/io-pgtable-arm.o
In file included from ../include/linux/atomic.h:80,
                 from ../drivers/iommu/io-pgtable-arm.c:12:
../drivers/iommu/io-pgtable-arm.c: In function 'arm_lpae_install_table':
../include/linux/atomic-arch-fallback.h:60:32: error: implicit declaration of function 'arch_cmpxchg64'; did you mean 'arch_cmpxchg'? [-Werror=implicit-function-declaration]
   60 | #define arch_cmpxchg64_relaxed arch_cmpxchg64
      |                                ^~~~~~~~~~~~~~
../include/asm-generic/atomic-instrumented.h:1261:2: note: in expansion of macro 'arch_cmpxchg64_relaxed'
 1261 |  arch_cmpxchg64_relaxed(__ai_ptr, __VA_ARGS__); \
      |  ^~~~~~~~~~~~~~~~~~~~~~
../drivers/iommu/io-pgtable-arm.c:320:8: note: in expansion of macro 'cmpxchg64_relaxed'
  320 |  old = cmpxchg64_relaxed(ptep, curr, new);
      |        ^~~~~~~~~~~~~~~~~



-- 
~Randy