Re: [PATCH] drivers/base: wrong return value of dmam_declare_coherent_memory

From: Vyacheslav Yurkov
Date: Thu Apr 28 2016 - 11:12:51 EST


On 28.04.2016 16:31, Andrey Gursky wrote:
> Vyacheslav,
>
> thanks for your patch.
>
> For now it introduces a new bug.
>
> Vyacheslav Yurkov <uvv.mail <at> gmail.com> writes:
>
>>
>> Hi guys,
>> I found an issue in managed version of dma_declare_coherent_memory,
>> i.e. in dmam_declare_coherent_memory. It looks like the return value
>> of dma_declare_coherent_memory is zero in case of error and a
>> requested flag on success,
>
> You should take this into account and not ignore the value of rc at all
> by using flags instead, which is not being altered by
> dma_declare_coherent_memory() making the latter appear to succeed
> always.
>
>

Hi Andrey,
Thanks for spotting this. My intention was to compare rc of course,
not the flags. While thinking it over I came up with improved version,
which is inlined below.
---
Yours sincerely,
Vyacheslav V. Yurkov

P.S.: I'm not subscribed to the maillist, so please include me in CC
when responding to this Email.

---
--- linux-4.5.2/drivers/base/dma-mapping.c.orig 2016-04-20 08:46:35.000000000 +0200
+++ linux-4.5.2/drivers/base/dma-mapping.c 2016-04-28 16:55:13.211945276 +0200
@@ -198,10 +198,14 @@ int dmam_declare_coherent_memory(struct

rc = dma_declare_coherent_memory(dev, phys_addr, device_addr, size,
flags);
- if (rc == 0)
+
+ if ((rc & (DMA_MEMORY_MAP | DMA_MEMORY_IO)) != 0) {
devres_add(dev, res);
- else
+ rc = 0;
+ } else {
devres_free(res);
+ rc = -ENOMEM;
+ }

return rc;
}

Signed-off-by: Vyacheslav Yurkov <uvv.mail@xxxxxxxxx>