2018-04-23 16:09 GMT+08:00 Michael Schmitz <schmitzmic@xxxxxxxxx>:
> Hi
>
> Am 23.04.2018 um 18:49 schrieb Greentime Hu:
>> 2018-04-20 16:03 GMT+08:00 Christoph Hellwig <hch@xxxxxx>:
>>> Switch to the generic noncoherent direct mapping implementation.
>>>
>>> This makes sure kmap_atomic_pfn is consistently used for access to
>>> virtual addresses instead of either using the slower plain kmap
>>> or blindly expecting page_address() to work.
>>>
>>> This makes sure the cache_sync routines is called in the unmap_sg
>>> case, to match the unmap_single and sync_{single,sg}_to_cpu cases.
>>>
>>> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
>>
>> CC arch/nds32/kernel/dma.o
>> arch/nds32/kernel/dma.c: In function 'arch_sync_dma_for_device':
>> arch/nds32/kernel/dma.c:333:10: error: 'direction' undeclared (first
>> use in this function)
>> switch (direction) {
>> ^~~~~~~~~
>> arch/nds32/kernel/dma.c:333:10: note: each undeclared identifier is
>> reported only once for each function it appears in
>> arch/nds32/kernel/dma.c: In function 'arch_sync_dma_for_cpu':
>> arch/nds32/kernel/dma.c:353:10: error: 'direction' undeclared (first
>> use in this function)
>> switch (direction) {
>> ^~~~~~~~~
>> arch/nds32/kernel/dma.c:358:30: error: 'end' undeclared (first use in
>> this function)
>> cpu_dma_inval_range(start, end);
>> ^~~
>> make[1]: *** [arch/nds32/kernel/dma.o] Error 1
>> make: *** [arch/nds32/kernel] Error 2
>>
>> After this building error, the ftmac100.c driver is broken. Not sure
>> what happened.
>
> Christoph used 'direction' as in the original consistent_sync(), but the
> parameter is now named 'dir' in the new function. Changing
>
> +void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
> + size_t size, enum dma_data_direction dir)
>
> to
>
> +void arch_sync_dma_for_device(struct device *dev, phys_addr_t paddr,
> + size_t size, enum dma_data_direction direction)
>
> (or using 'dir' in place of 'direction' in the switch clauses) should
> fix that.
>
> Christoph also forgot to declare and assign 'end'. Adding back the
>
> unsigned long end = start + size;
>
> after
>
> + unsigned long start = (unsigned long)addr;
>
> should fix the second error.
>
> If your ftmac100 driver relies on the DMA API, it wouldn't be expected
> to work if kernel and module views of the DMA API disagree.
>
> HTH,
>
> Michael
>
Hi, Micchael:
Thank you for helping fixing this build error.
What I mean before is that I fixed this build error just like what you
did and I built-in the ftmac100 driver as default in defconfig.
However the dhcpc failed, it works in Linux 4.17-rc1. I am still
debugging it. I use ftmac100 to test because it's the only driver I
can use for now and which will use DMA.
