Re: Sporadic regmap unit test failure in raw_sync

[Guenter Roeck]

On 2/8/24 14:10, Mark Brown wrote:
> On Thu, Feb 08, 2024 at 01:45:13PM -0800, Guenter Roeck wrote:
>
> >      # raw_sync: EXPECTATION FAILED at drivers/base/regmap/regmap-kunit.c:1305
> >      Expected &hw_buf[6] != val, but
> >          &hw_buf[6] ==
> >           57  30
> >          val ==
> >           57  30
> >          not ok 4 rbtree-big
> >          ok 5 maple-little
> >          ok 6 maple-big
> >      # raw_sync: pass:5 fail:1 skip:0 total:6
> >      not ok 25 raw_sync
>
> > This is with regmap: 'kunit: fix raw noinc write test wrapping'
> > applied on top of the upstream kernel (v6.8-rc3-47-g047371968ffc).
> > So far I have seen it only once, with the x86_64:q35 emulation in qemu.
>
> I guess it's possible that we randomly generated the same value for the
> initial and modified values here?


I think the diffs below should fix the problem. Would that do, or do you
have a better idea ?

Thanks,
Guenter

---
diff --git a/drivers/base/regmap/regmap-kunit.c b/drivers/base/regmap/regmap-kunit.c
index 4eb18f5d3265..666ece18726f 100644
--- a/drivers/base/regmap/regmap-kunit.c
+++ b/drivers/base/regmap/regmap-kunit.c
@@ -1266,7 +1266,11 @@ static void raw_sync(struct kunit *test)

        hw_buf = (u16 *)data->vals;

-       get_random_bytes(&val, sizeof(val));
+       /* values in val[] must be different from those in hw_buf[] */
+       val[0] = ~hw_buf[2];
+       if (val[0] == hw_buf[6])
+               val[0]++
+       val[1] = ~hw_buf[3];

        /* Do a regular write and a raw write in cache only mode */
        regcache_cache_only(map, true);