Re: [PATCH] crypto: gcm - fix cacheline sharing

[Horia Geanta]

On 5/31/2019 8:43 AM, Herbert Xu wrote:
> On Fri, May 31, 2019 at 05:22:50AM +0000, Horia Geanta wrote:
>>
>> Unless it's clearly defined *which* virtual addresses mustn't be accessed,
>> things won't work properly.
>> In theory, any two objects could share a cache line. We can't just stop all
>> memory accesses from CPU while a peripheral is busy.
> 
> The user obviously can't touch the memory areas potentially under
> DMA.  But in this case it's not the user that's doing it, it's
> the driver.
> 
> So the driver must not touch any virtual pointers given to it
> as input/output while the DMA areas are mapped.
> 
Driver is not touching the DMA mapped areas, the DMA API conventions are
carefully followed.
It's touching a virtual pointer that is not DMA mapped, that just happens to be
on the same cache line with a DMA mapped buffer.

Thanks,
Horia