Re: swiotlb/caamjr regression (Was: [GIT PULL] (swiotlb) stable/for-linus-5.12)

[Horia Geantă]

On 6/11/2021 1:35 PM, Konrad Rzeszutek Wilk wrote:
> On Fri, Jun 11, 2021 at 08:21:53AM +0200, Christoph Hellwig wrote:
>> On Thu, Jun 10, 2021 at 05:52:07PM +0300, Horia Geantă wrote:
>>> I've noticed the failure also in v5.10 and v5.11 stable kernels,
>>> since the patch set has been backported.
>>
>> FYI, there has been a patch on the list that should have fixed this
>> for about a month:
>>
>> https://lore.kernel.org/linux-iommu/20210510091816.GA2084@xxxxxx/T/#m0d0df6490350a08dcc24c9086c8edc165b402d6f
>>
>> but it seems like it never got picked up.
> 
> Yikes!
> 
> Dominique,
> 
> Would you be up to testing the attached (and inline) patch please?
> 
> Linus,
> 
> Would you be terribly offended if I took your code (s/unsigned
> long/unsigned int), and used Chanho's description of the problem (see below)?
> 
Both patches work for my case.

However, there's yet another, possibly significant, difference b/w the two:
	offset = (tlb_addr - mem->start) & (IO_TLB_SIZE - 1);
vs.
	offset = (tlb_addr & (IO_TLB_SIZE - 1)) -
		 swiotlb_align_offset(dev, orig_addr);

I think accounting for the alignment offset (swiotlb_align_offset())
has to be kept.

Horia