Robin Murphy <robin.murphy@xxxxxxx> writes:
> On 2022-03-24 10:25, Oleksandr Natalenko wrote:
>> Hello.
>>
>> On čtvrtek 24. března 2022 6:57:32 CET Christoph Hellwig wrote:
>>> On Wed, Mar 23, 2022 at 08:54:08PM +0000, Robin Murphy wrote:
>>>> I'll admit I still never quite grasped the reason for also adding the
>>>> override to swiotlb_sync_single_for_device() in aa6f8dcbab47, but I think
>>>> by that point we were increasingly tired and confused and starting to
>>>> second-guess ourselves (well, I was, at least). I don't think it's wrong
>>>> per se, but as I said I do think it can bite anyone who's been doing
>>>> dma_sync_*() wrong but getting away with it until now. If ddbd89deb7d3
>>>> alone turns out to work OK then I'd be inclined to try a partial revert of
>>>> just that one hunk.
>>>
>>> Agreed. Let's try that first.
>>>
>>> Oleksandr, can you try the patch below:
>>>
>>>
>>> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
>>> index 6db1c475ec827..6c350555e5a1c 100644
>>> --- a/kernel/dma/swiotlb.c
>>> +++ b/kernel/dma/swiotlb.c
>>> @@ -701,13 +701,10 @@ void swiotlb_tbl_unmap_single(struct device *dev, phys_addr_t tlb_addr,
>>> void swiotlb_sync_single_for_device(struct device *dev, phys_addr_t tlb_addr,
>>> size_t size, enum dma_data_direction dir)
>>> {
>>> - /*
>>> - * Unconditional bounce is necessary to avoid corruption on
>>> - * sync_*_for_cpu or dma_ummap_* when the device didn't overwrite
>>> - * the whole lengt of the bounce buffer.
>>> - */
>>> - swiotlb_bounce(dev, tlb_addr, size, DMA_TO_DEVICE);
>>> - BUG_ON(!valid_dma_direction(dir));
>>> + if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL)
>>> + swiotlb_bounce(dev, tlb_addr, size, DMA_TO_DEVICE);
>>> + else
>>> + BUG_ON(dir != DMA_FROM_DEVICE);
>>> }
>>>
>>> void swiotlb_sync_single_for_cpu(struct device *dev, phys_addr_t tlb_addr,
>>>
>>
>> With this patch the AP works for me.
>
> Cool, thanks for confirming. So I think ath9k probably is doing
> something dodgy with dma_sync_*(), but if Linus prefers to make the
> above change rather than wait for that to get figured out, I believe
> that should be fine.
I'm looking into this; but in the interest of a speedy resolution of the
regression I would be in favour of merging that partial revert and
reinstating it if/when we identify (and fix) any bugs in ath9k :)
-Toke
