Re: [PATCH v2 0/8] btrfs: do not poking into the implementation details of bio_vec

[Josef Bacik <josef@toxicpanda.com>]

On Sat, Apr 19, 2025 at 04:47:07PM +0930, Qu Wenruo wrote:
> - Fix a crash caused by incorrectly index sector_ptrs
>   The 5th patch, caused by missing increasement of the @offset variable.
> 
> - Use physical addresses instead of virtual addresses to handle HIGHMEM
>   The 6th patch, as we can still get sector_ptr from bio_sectors[] which
>   is using pages from the bio, and that can be highmem.
> 
>   However I can not do a proper test on this, as the latest kernel has
>   already rejected HIGHMEM64G option, thus even if my VM has extra 3GB
>   for HIGHMEM (total 6GB), I'm not sure if the kernel can really utilize
>   those high memories.
> 
>   Furthermore there seems to be other bugs in mm layer related to
>   HIGHMEM + PAGE, resulting zswap crash when try compressing a page to
>   be swapped out.
>   But at least several scrub/balance related test cases passed on x86
>   32bit with HIGHMEM and PAGE.
> 
>   I have tested with x86_64 (64 bit, no HIGHMEM), aarch64 (64bit, 64K
>   page size, no HIGHMEM) with no regression.
> 
> - Fix a incorrect __bio_add_page() usage in scrub_bio_add_sector()
>   The 6th patch, as bio_add_page() do extra bvec merging before
>   allocating a new bvec, but __bio_add_page() does not.
> 
>   This triggers WARN_ON_ONCE() in __bio_add_page() checking if the bio
>   is full.
> 
>   Fixing it by do the old bio_add_page() and ASSERT(), with extra
>   comment on we can not use __bio_add_page().
> 
> - Various minor commit message update
>   And full proper test runs.
> 

Reviewed-by: Josef Bacik <josef@toxicpanda.com>

Thanks,

Josef