[PATCH v1 1/1] ath10k: convert kmemdup to dma_alloc_coherent

[Jared Bents]

Hi Kalle,

On Sat, Apr 21, 2018 at 2:10 AM, Kalle Valo <kvalo at codeaurora.org> wrote:
> Jared Bents <jared.bents at rockwellcollins.com> writes:
>
>> Update to convert the use of kmemdup to dma_alloc_coherent as
>> dma_alloc_coherent will consider DMA region limits such as
>> those seen with CONFIG_FSL_PCI && CONFIG_ZONE_DMA32 whereas
>> kmemdup does not take those limitations into account.
>>
>> Signed-off-by: Jared Bents <jared.bents at rockwellcollins.com>
>
> Does this fix a real bug or is it just a theoretical issue you noticed
> while looking at the code? If this fixes a real bug please give a
> description of it.

It fixes a real bug that I have experienced.  My processor is a qoriq
T1042 powerpc and I am running the fsl-sdk-v2.0-1703 kernel (mainline
v4.1.35-rt41).  In 32 bit, the ath10k and ath9k work without issue.
After migrating to 64 bit, we experienced dma mapping errors with both
devices.  This patch is the only one that applies to a driver, the
rest of the updates performed to fix the dma mapping issues deal with
the sk_buffs are handled in skbuff, mac80211, and netlink.  We worked
with NXP to resolve the issue and I believe they will be handling the
ones that don't apply to the ath10k driver through their process.  If
we don't see them after a while, we will be sending them to the kernel
as well.

Here is the output from the dma mapping problem:

ath10k_pci 0000:01:00.0: unable to get target info from device
ath10k_pci 0000:01:00.0: could not get target info (-5)
ath10k_pci 0000:01:00.0: could not probe fw (-5)
>
> https://wireless.wiki.kernel.org/en/users/drivers/ath10k/submittingpatches#guidelines
>
> --
> Kalle Valo

Jared Bents