Re: [PATCH 4/4] iommu/arm-smmu-v3: Remove cmpxchg() in arm_smmu_cmdq_issue_cmdlist()

[John Garry <john.garry@huawei.com>]

On 23/06/2020 14:55, Rikard Falkeborn wrote:
> Den tis 23 juni 2020 12:21John Garry <john.garry@huawei.com 
> <mailto:john.garry@huawei.com>> skrev:
> 
>     On 23/06/2020 10:35, Rikard Falkeborn wrote:
>      >
>      >     I'd say that GENMASK_INPUT_CHECK() should be able to handle a
>     l=0 and
>      >     h=unsigned value, so I doubt this warn.
>      >
>      >     Using GENMASK((int)cmdq->q.llq.max_n_shift, 0) resolves it,
>     but it
>      >     looks
>      >     like GENMASK_INPUT_CHECK() could be improved.
>      >
>      >
>      > Indeed it could, it is fixed in -next.
> 
>     ok, thanks for the pointer, but I still see this on today's -next with
>     this patch:
> 
>     make W=1 drivers/iommu/arm-smmu-v3.o
> 
> 
> Oh, ok thanks for reporting. I guess different gcc versions have 
> different behaviour. I guess we'll have to change the comparison to 
> (!((h) == (l) || (h) > (l))) instead (not sure I got all parenthesis and 
> logic correct but you get the idea).
> 

Yeah, so this looks to fix it:

--- a/include/linux/bits.h
+++ b/include/linux/bits.h
@@ -23,7 +23,8 @@
#include <linux/build_bug.h>
#define GENMASK_INPUT_CHECK(h, l) \
        (BUILD_BUG_ON_ZERO(__builtin_choose_expr( \
-               __builtin_constant_p((l) > (h)), (l) > (h), 0)))
+               __builtin_constant_p(!((h) == (l) ||(h) > (l))), !((h) 
== (l) ||(h) > (l)), 0)))
+

We may be able to just use (h) == (l) as the const expr to make it more 
concise, but that may be confusing.

I only tested with my toolchain based on 7.5.0

Thanks,
John