Message ID | bee6ac5945df3db876a553dbf6d0dd546f145aa7.1737701962.git.wqu@suse.com (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | btrfs: zlib: refactor S390x HW acceleration buffer preparation | expand |
Hello Qu, Sorry for the long silence. I've tested this patch on s390 using our hardware compression test suite. No issues discovered. Here is my Acked-by: Mikhail Zaslonko <zaslonko@linux.ibm.com> Tested-by: Mikhail Zaslonko <zaslonko@linux.ibm.com> On 24.01.2025 07:59, Qu Wenruo wrote: > Currently for s390x HW zlib compression, to get the best performance we > need a buffer size which is larger than a page. > > This means we need to copy multiple pages into workspace->buf, then use > that buffer as zlib compression input. > > Currently it's hardcoded using page sized folio, and all the handling > are deep inside a loop. > > Refactor the code by: > > - Introduce a dedicated helper to do the buffer copy > The new helper will be called copy_data_into_buffer(). > > - Add extra ASSERT()s > * Make sure we only go into the function for hardware acceleration > * Make sure we still get page sized folio > > - Prepare for future larger folios > This means we will rely on the folio size, other than PAGE_SIZE to do > the copy. > > - Handle the folio mapping and unmapping inside the helper function > For S390x hardware acceleration case, it never utilize the @data_in > pointer, thus we can do folio mapping/unmapping all inside the function. > > Cc: Mikhail Zaslonko <zaslonko@linux.ibm.com> > Signed-off-by: Qu Wenruo <wqu@suse.com> > --- > Hi Mikhail, > > Mind to give this patch a test on S390x? As I do not have easy access to > such system. > > And I hope I won't need to bother you again when we enable larger folios > for btrfs in the near future. > > Thanks, > Qu > --- > fs/btrfs/zlib.c | 82 ++++++++++++++++++++++++++++++++----------------- > 1 file changed, 54 insertions(+), 28 deletions(-) > > diff --git a/fs/btrfs/zlib.c b/fs/btrfs/zlib.c > index c9e92c6941ec..7d81d9a0b3a0 100644 > --- a/fs/btrfs/zlib.c > +++ b/fs/btrfs/zlib.c > @@ -94,6 +94,47 @@ struct list_head *zlib_alloc_workspace(unsigned int level) > return ERR_PTR(-ENOMEM); > } > > +/* > + * Helper for S390x with hardware zlib compression support. > + * > + * That hardware acceleration requires a buffer size larger than a single page > + * to get ideal performance, thus we need to do the memory copy other than > + * use the page cache directly as input buffer. > + */ > +static int copy_data_into_buffer(struct address_space *mapping, > + struct workspace *workspace, u64 filepos, > + unsigned long length) > +{ > + u64 cur = filepos; > + > + /* It's only for hardware accelerated zlib code. */ > + ASSERT(zlib_deflate_dfltcc_enabled()); > + > + while (cur < filepos + length) { > + struct folio *folio; > + void *data_in; > + unsigned int offset; > + unsigned long copy_length; > + int ret; > + > + ret = btrfs_compress_filemap_get_folio(mapping, cur, &folio); > + if (ret < 0) > + return ret; > + /* No larger folio support yet. */ > + ASSERT(!folio_test_large(folio)); > + > + offset = offset_in_folio(folio, cur); > + copy_length = min(folio_size(folio) - offset, > + filepos + length - cur); > + > + data_in = kmap_local_folio(folio, offset); > + memcpy(workspace->buf + cur - filepos, data_in, copy_length); > + kunmap_local(data_in); > + cur += copy_length; > + } > + return 0; > +} > + > int zlib_compress_folios(struct list_head *ws, struct address_space *mapping, > u64 start, struct folio **folios, unsigned long *out_folios, > unsigned long *total_in, unsigned long *total_out) > @@ -105,8 +146,6 @@ int zlib_compress_folios(struct list_head *ws, struct address_space *mapping, > int nr_folios = 0; > struct folio *in_folio = NULL; > struct folio *out_folio = NULL; > - unsigned long bytes_left; > - unsigned int in_buf_folios; > unsigned long len = *total_out; > unsigned long nr_dest_folios = *out_folios; > const unsigned long max_out = nr_dest_folios * PAGE_SIZE; > @@ -150,34 +189,21 @@ int zlib_compress_folios(struct list_head *ws, struct address_space *mapping, > * the workspace buffer if required. > */ > if (workspace->strm.avail_in == 0) { > - bytes_left = len - workspace->strm.total_in; > - in_buf_folios = min(DIV_ROUND_UP(bytes_left, PAGE_SIZE), > - workspace->buf_size / PAGE_SIZE); > - if (in_buf_folios > 1) { > - int i; > + unsigned long bytes_left = len - workspace->strm.total_in; > + unsigned int copy_length = min(bytes_left, workspace->buf_size); > > - /* S390 hardware acceleration path, not subpage. */ > - ASSERT(!btrfs_is_subpage( > - inode_to_fs_info(mapping->host), > - mapping)); > - for (i = 0; i < in_buf_folios; i++) { > - if (data_in) { > - kunmap_local(data_in); > - folio_put(in_folio); > - data_in = NULL; > - } > - ret = btrfs_compress_filemap_get_folio(mapping, > - start, &in_folio); > - if (ret < 0) > - goto out; > - data_in = kmap_local_folio(in_folio, 0); > - copy_page(workspace->buf + i * PAGE_SIZE, > - data_in); > - start += PAGE_SIZE; > - } > + /* > + * This can only happen when hardware zlib compression is > + * enabled. > + */ > + if (copy_length > PAGE_SIZE) { > + ret = copy_data_into_buffer(mapping, workspace, > + start, copy_length); > + if (ret < 0) > + goto out; > + start += copy_length; > workspace->strm.next_in = workspace->buf; > - workspace->strm.avail_in = min(bytes_left, > - in_buf_folios << PAGE_SHIFT); > + workspace->strm.avail_in = copy_length; > } else { > unsigned int pg_off; > unsigned int cur_len;
diff --git a/fs/btrfs/zlib.c b/fs/btrfs/zlib.c index c9e92c6941ec..7d81d9a0b3a0 100644 --- a/fs/btrfs/zlib.c +++ b/fs/btrfs/zlib.c @@ -94,6 +94,47 @@ struct list_head *zlib_alloc_workspace(unsigned int level) return ERR_PTR(-ENOMEM); } +/* + * Helper for S390x with hardware zlib compression support. + * + * That hardware acceleration requires a buffer size larger than a single page + * to get ideal performance, thus we need to do the memory copy other than + * use the page cache directly as input buffer. + */ +static int copy_data_into_buffer(struct address_space *mapping, + struct workspace *workspace, u64 filepos, + unsigned long length) +{ + u64 cur = filepos; + + /* It's only for hardware accelerated zlib code. */ + ASSERT(zlib_deflate_dfltcc_enabled()); + + while (cur < filepos + length) { + struct folio *folio; + void *data_in; + unsigned int offset; + unsigned long copy_length; + int ret; + + ret = btrfs_compress_filemap_get_folio(mapping, cur, &folio); + if (ret < 0) + return ret; + /* No larger folio support yet. */ + ASSERT(!folio_test_large(folio)); + + offset = offset_in_folio(folio, cur); + copy_length = min(folio_size(folio) - offset, + filepos + length - cur); + + data_in = kmap_local_folio(folio, offset); + memcpy(workspace->buf + cur - filepos, data_in, copy_length); + kunmap_local(data_in); + cur += copy_length; + } + return 0; +} + int zlib_compress_folios(struct list_head *ws, struct address_space *mapping, u64 start, struct folio **folios, unsigned long *out_folios, unsigned long *total_in, unsigned long *total_out) @@ -105,8 +146,6 @@ int zlib_compress_folios(struct list_head *ws, struct address_space *mapping, int nr_folios = 0; struct folio *in_folio = NULL; struct folio *out_folio = NULL; - unsigned long bytes_left; - unsigned int in_buf_folios; unsigned long len = *total_out; unsigned long nr_dest_folios = *out_folios; const unsigned long max_out = nr_dest_folios * PAGE_SIZE; @@ -150,34 +189,21 @@ int zlib_compress_folios(struct list_head *ws, struct address_space *mapping, * the workspace buffer if required. */ if (workspace->strm.avail_in == 0) { - bytes_left = len - workspace->strm.total_in; - in_buf_folios = min(DIV_ROUND_UP(bytes_left, PAGE_SIZE), - workspace->buf_size / PAGE_SIZE); - if (in_buf_folios > 1) { - int i; + unsigned long bytes_left = len - workspace->strm.total_in; + unsigned int copy_length = min(bytes_left, workspace->buf_size); - /* S390 hardware acceleration path, not subpage. */ - ASSERT(!btrfs_is_subpage( - inode_to_fs_info(mapping->host), - mapping)); - for (i = 0; i < in_buf_folios; i++) { - if (data_in) { - kunmap_local(data_in); - folio_put(in_folio); - data_in = NULL; - } - ret = btrfs_compress_filemap_get_folio(mapping, - start, &in_folio); - if (ret < 0) - goto out; - data_in = kmap_local_folio(in_folio, 0); - copy_page(workspace->buf + i * PAGE_SIZE, - data_in); - start += PAGE_SIZE; - } + /* + * This can only happen when hardware zlib compression is + * enabled. + */ + if (copy_length > PAGE_SIZE) { + ret = copy_data_into_buffer(mapping, workspace, + start, copy_length); + if (ret < 0) + goto out; + start += copy_length; workspace->strm.next_in = workspace->buf; - workspace->strm.avail_in = min(bytes_left, - in_buf_folios << PAGE_SHIFT); + workspace->strm.avail_in = copy_length; } else { unsigned int pg_off; unsigned int cur_len;
Currently for s390x HW zlib compression, to get the best performance we need a buffer size which is larger than a page. This means we need to copy multiple pages into workspace->buf, then use that buffer as zlib compression input. Currently it's hardcoded using page sized folio, and all the handling are deep inside a loop. Refactor the code by: - Introduce a dedicated helper to do the buffer copy The new helper will be called copy_data_into_buffer(). - Add extra ASSERT()s * Make sure we only go into the function for hardware acceleration * Make sure we still get page sized folio - Prepare for future larger folios This means we will rely on the folio size, other than PAGE_SIZE to do the copy. - Handle the folio mapping and unmapping inside the helper function For S390x hardware acceleration case, it never utilize the @data_in pointer, thus we can do folio mapping/unmapping all inside the function. Cc: Mikhail Zaslonko <zaslonko@linux.ibm.com> Signed-off-by: Qu Wenruo <wqu@suse.com> --- Hi Mikhail, Mind to give this patch a test on S390x? As I do not have easy access to such system. And I hope I won't need to bother you again when we enable larger folios for btrfs in the near future. Thanks, Qu --- fs/btrfs/zlib.c | 82 ++++++++++++++++++++++++++++++++----------------- 1 file changed, 54 insertions(+), 28 deletions(-)