Re: [PATCH v7] crypto: scompress: remove memcpy if sg_nents is 1 and pages are lowmem

[Guenter Roeck <linux@roeck-us.net>]

On 3/17/24 17:21, Barry Song wrote:
> On Mon, Mar 18, 2024 at 8:14 AM Guenter Roeck <linux@roeck-us.net> wrote:
>>
>> On 3/17/24 16:48, Barry Song wrote:
>>> On Mon, Mar 18, 2024 at 7:13 AM Guenter Roeck <linux@roeck-us.net> wrote:
>>>>
>>>> Hi,
>>>>
>>>> On Sat, Mar 02, 2024 at 08:27:45AM +1300, Barry Song wrote:
>>>> [ ... ]
>>>>> @@ -152,8 +165,17 @@ static int scomp_acomp_comp_decomp(struct acomp_req *req, int dir)
>>>>>                         ret = -ENOSPC;
>>>>>                         goto out;
>>>>>                 }
>>>>> -             scatterwalk_map_and_copy(scratch->dst, req->dst, 0, req->dlen,
>>>>> -                                      1);
>>>>> +             if (dst == scratch->dst) {
>>>>> +                     scatterwalk_map_and_copy(scratch->dst, req->dst, 0,
>>>>> +                                              req->dlen, 1);
>>>>> +             } else {
>>>>> +                     int nr_pages = DIV_ROUND_UP(req->dst->offset + req->dlen, PAGE_SIZE);
>>>>> +                     int i;
>>>>> +                     struct page *dst_page = sg_page(req->dst);
>>>>> +
>>>>> +                     for (i = 0; i < nr_pages; i++)
>>>>> +                             flush_dcache_page(dst_page + i);
>>>>
>>>> flush_dcache_page() is an empty macro on some architectures
>>>> such as xtensa. This results in
>>>
>>> Hi Guenter,
>>>
>>> this is a bug of xtensa, could you test the patch:
>>
>> Thanks for the update.
>>
>>> https://lore.kernel.org/all/20240313045036.51065-1-21cnbao@gmail.com/
>>>
>>
>> That doesn't build for me.
>>
>>     CC      arch/xtensa/kernel/asm-offsets.s
>> In file included from ./include/linux/highmem.h:8,
>>                    from ./include/linux/bvec.h:10,
>>                    from ./include/linux/blk_types.h:10,
>>                    from ./include/linux/writeback.h:13,
>>                    from ./include/linux/memcontrol.h:23,
>>                    from ./include/linux/swap.h:9,
>>                    from ./include/linux/suspend.h:5,
>>                    from arch/xtensa/kernel/asm-offsets.c:24:
>> ./include/linux/cacheflush.h:9:5: error: "ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE" is not defined, evaluates to 0 [-Werror=undef]
>>       9 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
>>
>> A similar works for loongarch, so I don't know what is wrong.
>> Maybe some context patch is missing.
> 
> this is weird as include/asm-generic/cacheflush.h will define it to 0
> 
>   #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
>   static inline void flush_dcache_page(struct page *page)
>   {
>   }
> 
>   #define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 0
>   #endif
> 
> Maybe somewhere else also needs to be fixed.
> Can you tell me your toolchain version and toolchain name? and defconfig name?
> 

config is allmodconfig - GCC_PLUGINS.

Toolchains are self-built from gcc source using the buildall system. I tried
gcc 11.4, 12.3, and 13.2. I don't think that matters, though, since
"asm-generic/cacheflush.h" isn't included from arch/xtensa. The diff below
seems to fix the problem, but I have not fully tested it.

Guenter

---
diff --git a/arch/xtensa/include/asm/cacheflush.h b/arch/xtensa/include/asm/cacheflush.h
index 11efdc8eb262..62662919cbbc 100644
--- a/arch/xtensa/include/asm/cacheflush.h
+++ b/arch/xtensa/include/asm/cacheflush.h
@@ -183,4 +183,6 @@ extern void copy_from_user_page(struct vm_area_struct*, struct page*,

  #endif

+#include <asm-generic/cacheflush.h>
+
  #endif /* _XTENSA_CACHEFLUSH_H */