Hi Jess,
On Wed, Apr 05, 2023 at 05:19:35AM +0100, Jessica Clarke wrote:
> On 4 Apr 2023, at 19:20, Sunil V L <sunilvl@xxxxxxxxxxxxxxxx> wrote:
> >
> > With CONFIG_ACPI enabled for RISC-V, this driver gets enabled
> > in allmodconfig build. However, RISC-V doesn't support sub-word
> > atomics which is used by this driver.
>
> Why not? Compilers and libatomic do, so surely the Linux kernel should
> too.
>
I think you are probably right. But I don't want to combine that
activity with this series. IMO, that should be separate activity.
> > Due to this, the build fails
> > with below error.
> >
> > In function ‘ssh_seq_next’,
> > inlined from ‘ssam_request_write_data’ at drivers/platform/surface/aggregator/controller.c:1483:8:
> > ././include/linux/compiler_types.h:399:45: error: call to ‘__compiletime_assert_335’ declared with attribute error: BUILD_BUG failed
> > 399 | _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
> > | ^
> > ./include/linux/compiler.h:78:45: note: in definition of macro ‘unlikely’
> > 78 | # define unlikely(x) __builtin_expect(!!(x), 0)
> > | ^
> > ././include/linux/compiler_types.h:387:9: note: in expansion of macro ‘__compiletime_assert’
> > 387 | __compiletime_assert(condition, msg, prefix, suffix)
> > | ^~~~~~~~~~~~~~~~~~~~
> > ././include/linux/compiler_types.h:399:9: note: in expansion of macro ‘_compiletime_assert’
> > 399 | _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/riscv/include/asm/cmpxchg.h:335:17: note: in expansion of macro ‘BUILD_BUG’
> > 335 | BUILD_BUG(); \
> > | ^~~~~~~~~
> > ./arch/riscv/include/asm/cmpxchg.h:344:30: note: in expansion of macro ‘__cmpxchg’
> > 344 | (__typeof__(*(ptr))) __cmpxchg((ptr), \
> > | ^~~~~~~~~
> > ./include/linux/atomic/atomic-instrumented.h:1916:9: note: in expansion of macro ‘arch_cmpxchg’
> > 1916 | arch_cmpxchg(__ai_ptr, __VA_ARGS__); \
> > | ^~~~~~~~~~~~
> > drivers/platform/surface/aggregator/controller.c:61:32: note: in expansion of macro ‘cmpxchg’
> > 61 | while (unlikely((ret = cmpxchg(&c->value, old, new)) != old)) {
> > | ^~~~~~~
> >
> > So, disable this driver for RISC-V even when ACPI is enabled for now.
> >
> > Signed-off-by: Sunil V L <sunilvl@xxxxxxxxxxxxxxxx>
> > ---
> > drivers/platform/surface/aggregator/Kconfig | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/platform/surface/aggregator/Kconfig b/drivers/platform/surface/aggregator/Kconfig
> > index c114f9dd5fe1..88afc38ffdc5 100644
> > --- a/drivers/platform/surface/aggregator/Kconfig
> > +++ b/drivers/platform/surface/aggregator/Kconfig
> > @@ -4,7 +4,7 @@
> > menuconfig SURFACE_AGGREGATOR
> > tristate "Microsoft Surface System Aggregator Module Subsystem and Drivers"
> > depends on SERIAL_DEV_BUS
> > - depends on ACPI
> > + depends on ACPI && !RISCV
>
> If you insist on doing this, at least make it some new config variable
> that’s self-documenting and means this automatically gets re-enabled
> when arch/riscv fixes this deficiency? Hard-coding arch lists like this
> seems like a terrible anti-pattern.
>
I understand your point. But given that this is currently only issue with
a single driver from Microsoft and that too only in COMPILE_TEST builds,
I think introducing a new config variable is overkill. If we support
sub-word atomics in kernel, the option may not be useful much anyway.
There are patterns to disable an architecture for COMPILE_TEST builds.
Thanks,
Sunil
