Re: [PATCH v2 10/15] lib/lz4: switch the compressor to the vendored sources

[sashiko-bot@kernel.org]

> Replace the forked compressor with thin entry points over the vendored
> lz4.c. LZ4_compress_default(), LZ4_compress_fast() and
> LZ4_compress_destSize() keep the trailing wrkmem argument and forward to
> upstream's *_extState() functions; the rest are plain forwarders.
> 
> Upstream's LZ4_initStream() returns NULL for a misaligned workspace,
> which LZ4_compress_fast_extState() dereferences and
> LZ4_compress_destSize_extState() ignores. Check the alignment in the
> entry points, warn once and fail the call. In-tree callers allocate with
> kmalloc() or vmalloc() and are not affected.
> 
> Compressed output differs from the fork's on a small fraction of inputs
> by a byte or two; both decode each other's output. Compressing 4K
> blocks, upstream is 2-4% faster on kernel source and binaries and up to
> 9% slower on datagen.
> 
> LZ4_stream_t becomes an incomplete type, as upstream declares it, so
> <linux/lz4.h> and upstream's lz4.h can share a translation unit. The
> macros that laid out the forked structure go, and LZ4_MEM_COMPRESS keeps
> its value, 16416. zram, the only caller using sizeof(LZ4_stream_t), now
> allocates LZ4_MEM_COMPRESS bytes.
> 
> Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261106-lz4-vendor-upstream-v2-0-ee04c3f0d7dc@samsung.com?part=10