[PATCH v2 4/6] spi: davinci: flush caches when performing DMA

[Frode Isaksen]

On 20/02/2017 10:47, Vignesh R wrote:
>
> On Monday 20 February 2017 02:56 PM, Frode Isaksen wrote:
>>
>> On 20/02/2017 07:55, Vignesh R wrote:
>>> On Friday 17 February 2017 05:37 PM, Russell King - ARM Linux wrote:
>>> [...]
>>>> SPI is rather another special case - rather than SPI following the 
>>>> established mechanism of passing data references via scatterlists or 
>>>> similar, it also passes them via virtual addresses, which means SPI 
>>>> can directly access the vmalloc area when performing PIO.  This 
>>>> really makes the problem more complex, because it means that if you 
>>>> do have a SPI driver that does that, it's going to be
>>>> reading/writing direct from vmalloc space.
>>>>
>>>> That's not a problem as long as the data is only accessed via
>>>> vmalloc space, but it will definitely go totally wrong if the data
>>>> is subsequently mapped into userspace.
>>>>
>>>> The other important thing to realise is that the interfaces in 
>>>> cachetlb.txt assume that it's the lowmem mapping that will be
>>>> accessed, and the IO device will push that data out to physical
>>>> memory (either via the DMA API, or flush_kernel_dcache_page()).
>>>> That's not true of SPI, as it passes virtual addresses around.
>>>>
>>>> So... overall, I'm not sure that this problem is properly solvable
>>>> given SPIs insistance on passing virtual addresses and the
>>>> differences in this area between SPI and block.
>>>>
>>> I am debugging another issue with UBIFS wherein pages allocated by
>>> vmalloc are in highmem region that are not addressable using 32 bit
>>> addresses and is backed by LPAE. So, a 32 bit DMA cannot access these
>>> buffers at all.
>>> When dma_map_sg() is called to map these pages by spi_map_buf() the
>>> physical address is just truncated to 32 bit in pfn_to_dma() (as part of
>>> dma_map_sg() call). This results in random crashes as DMA starts
>>> accessing random memory during SPI read.
>>>
>>> Given, the above problem and also issue surrounding VIVT caches, I am
>>> thinking of may be using pre-allocated fixed size bounce buffer to
>>> handle buffers not in lowmem mapping.
>>> I have tried using 64KB pre-allocated buffer on TI DRA74 EVM with QSPI
>>> running at 76.8MHz and do not see any significant degradation in
>>> performance with UBIFS. Mainly because UBIFS seems to use vmalloc'd
>>> buffers only during initial preparing and mounting phase and not during
>>> file read/write.
>> I am seeing a bug caused by VIVT cache in 'read_ltab()' function. In this function, the vmalloc'ed buffer is of size 11. Isn't it better to use kmalloc in this case ?
> read_ltab() isn't the only place where vmalloc() is used. A quick grep
> for vmalloc on fs/ubifs/ shows about ~19 occurrence. I guess every
> vmalloc() call can potentially allocate memory from highmem and might
> potentially cause issue for VIVT and such aliasing caches.
> Fixing just one such case isn't going to help IMHO.
Of course fixing it only in one place is of course not going to help..
For the moment there are 3 solutions to the UBIFS DMA problem:
1) Always use bounce buffer for vmalloc'ed buffers - impacts everyone.
2) Remove use of vmalloc'ed buffers in UBIFS - is it possible ?
3) Flush cache in UBIFS when reading/wring vmalloc'ed buffers - does not fix the higmem case.
I will try to see if option 2) is possible.

Thanks,
Frode
>
>