Re: [PATCH v4] tools/dma: move dma_map_benchmark from selftests to tools/dma

[Qinxin Xia <xiaqinxin@huawei.com>]

On 2025/10/22 12:23:04, Barry Song <21cnbao@gmail.com> wrote:
>>
>>   #define DMA_MAP_BENCHMARK       _IOWR('d', 1, struct map_benchmark)
>>   #define DMA_MAP_MAX_THREADS     1024
>> @@ -27,5 +29,7 @@ struct map_benchmark {
>>          __u32 dma_dir; /* DMA data direction */
>>          __u32 dma_trans_ns; /* time for DMA transmission in ns */
>>          __u32 granule;  /* how many PAGE_SIZE will do map/unmap once a time */
>> +       __u8 expansion[76]; /* For future use */
>>   };
> 
> I didn’t realize you were hiding a bugfix within the code cleanup.
> 
> We agreed this is a hotfix that should be merged promptly[1]. Before sending
> v5, please submit the hotfix first using the existing folder (without moving
> the code from selftest to tools). Once it is merged, resend v5 based on it.
> 
> [1] https://lore.kernel.org/lkml/f8c0cbdc-9c68-46bb-8dad-1d7c80f3f741@samsung.com/
> 
> Thanks
> Barry

Hello Barry:
For the previous patch has not merged, I have carefully considered this.
(I didn't include this in v1. You can see it in the 'Changes')

Previously, struct map_benchmark is defined in map_benchmark.h. The
header file in include/linux/ generally does not guarantee stable uABI.
(Although it does get used in userspace, I thought that's because we
include the kernel header in userspace.Referencing kernel headers in
userspace programs is inherently risky)

This patch moves map_benchmark.h from include/linux to include/uapi
/linux. I think maybe it's reasonable to restore that here...

I'm sure you have more experience in this area, and if you still insist
on separation, I will prepare v5 based on merged version.

Thanks,
Qinxin