On 13/05/15 19:49, Kevin Cernekee wrote:
> On Wed, May 13, 2015 at 6:49 PM, Leonid Yegoshin
> <Leonid.Yegoshin@xxxxxxxxxx> wrote:
>> Some MIPS CPUs have an aggressive speculative load and may erroneuosly load
>> some cache line in the middle of DMA transaction. CPU discards result but cache
>> doesn't. If DMA happens from device then additional cache invalidation is needed
>> on that CPU's after DMA.
>>
>> Found in test.
>>
>> Signed-off-by: Leonid Yegoshin <Leonid.Yegoshin@xxxxxxxxxx>
>> ---
>> arch/mips/mm/dma-default.c | 10 ++++++----
>> 1 file changed, 6 insertions(+), 4 deletions(-)
>>
>> diff --git a/arch/mips/mm/dma-default.c b/arch/mips/mm/dma-default.c
>> index 609d1241b0c4..ccf49ecfbf8c 100644
>> --- a/arch/mips/mm/dma-default.c
>> +++ b/arch/mips/mm/dma-default.c
>> @@ -67,11 +67,13 @@ static inline struct page *dma_addr_to_page(struct device *dev,
>> * systems and only the R10000 and R12000 are used in such systems, the
>> * SGI IP28 Indigo² rsp. SGI IP32 aka O2.
>> */
>> -static inline int cpu_needs_post_dma_flush(struct device *dev)
>> +static inline int cpu_needs_post_dma_flush(struct device *dev,
>> + enum dma_data_direction direction)
>> {
>> return !plat_device_is_coherent(dev) &&
>> (boot_cpu_type() == CPU_R10000 ||
>> boot_cpu_type() == CPU_R12000 ||
>> + (cpu_has_maar && (direction != DMA_TO_DEVICE)) ||
>> boot_cpu_type() == CPU_BMIPS5000);
>
> Can all of these CPUs safely skip the post_dma_flush on DMA_TO_DEVICE
> (not just maar)?
Agreed that would seem like the logical thing to do. You could then just
skip the call to cpu_needs_post_dma_flush() and move the direction test
in arch/mips/mm/dma-default.c for instance?
--
Florian
