Short cleanups
Subsystems (2)
David Sterba
maintainercover Short cleanups
Collection of short clenups that improve code that's somehow performance sensitive. The improvements are on instruction level or reduce stack consumption.
David Sterba (12):
btrfs: remove duplicate calculation of eb offset in btrfs_bin_search()
btrfs: unify types for binary search variables
btrfs: rename local variable for offset in folio
btrfs: read eb folio index right before loops
btrfs: use common eb range validation in
read_extent_buffer_to_user_nofault()
btrfs: lzo: inline read/write length helpers
btrfs: zlib: drop redundant folio address variable
btrfs: zlib: don't cache sectorsize in a local variable
btrfs: zlib: remove local variable nr_dest_folios in
zlib_compress_folios()
btrfs: zstd: reuse total in and out parameters for calculations
btrfs: zstd: don't cache sectorsize in a local variable
btrfs: zstd: remove local variable nr_dest_folios in
zstd_compress_folios()
fs/btrfs/ctree.c | 11 +++++------ fs/btrfs/extent_io.c | 23 ++++++++++++----------- fs/btrfs/lzo.c | 28 ++++++---------------------- fs/btrfs/zlib.c | 25 +++++++++---------------- fs/btrfs/zstd.c | 43 +++++++++++++++++-------------------------- 5 files changed, 49 insertions(+), 81 deletions(-)
Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 1/12 btrfs: remove duplicate calculation of eb offset in btrfs_bin_search()
In the main search loop the variable 'oil' (offset in folio) is set twice, one duplicated when the key fits completely to the contiguous range. We can remove it and while it's just a simple calculation, the binary search loop is executed many times so micro optimizations add up.
The code size is reduced by 64 bytes on release config, the loop is reorganized a bit and a few instructions shorter.
--- fs/btrfs/ctree.c | 1 - 1 file changed, 1 deletion(-)
fs/btrfs/ctree.c +0 -1
@@ -776,7 +776,6 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,const unsigned long idx =get_eb_folio_index (eb, offset);char *kaddr =folio_address (eb->folios[idx]); - oil =get_eb_offset_in_folio (eb, offset); tmp = (struct btrfs_disk_key *)(kaddr + oil); }else {read_extent_buffer (eb, &unaligned, offset, key_size);
Signature · 1 line
2.51.1
Re: [PATCH 01/12] btrfs: remove duplicate calculation of eb offset in btrfs_bin_search()
Quoted 9 lines · David Sterba
In the main search loop the variable 'oil' (offset in folio) is set twice, one duplicated when the key fits completely to the contiguous range. We can remove it and while it's just a simple calculation, the binary search loop is executed many times so micro optimizations add up.
The code size is reduced by 64 bytes on release config, the loop is reorganized a bit and a few instructions shorter.
Signed-off-by David Sterba <[email protected]>
Quoted 19 lines
--- fs/btrfs/ctree.c | 1 - 1 file changed, 1 deletion(-)fs/btrfs/ctree.c +0
-1@@ -776,7 +776,6 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,const unsigned long idx =get_eb_folio_index (eb, offset);char *kaddr =folio_address (eb->folios[idx]); - oil =get_eb_offset_in_folio (eb, offset); tmp = (struct btrfs_disk_key *)(kaddr + oil); }else {read_extent_buffer (eb, &unaligned, offset, key_size);Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 2/12 btrfs: unify types for binary search variables
In reply to David SterbaThe variables calculating where to jump next are useing mixed in types which requires some conversions on the instruction level. Using 'u32' removes one call to 'movslq', making the main loop shorter.
This complements type conversion done in a724f313f84beb ("btrfs: do unsigned integer division in the extent buffer binary search loop")
--- fs/btrfs/ctree.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/btrfs/ctree.c +1 -1
@@ -766,7 +766,7 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,unsigned long offset;struct btrfs_disk_key *tmp;struct btrfs_disk_key unaligned; -int mid; +u32 mid; mid = (low + high) /2 ; offset = p + mid * item_size;
Signature · 1 line
2.51.1
Re: [PATCH 02/12] btrfs: unify types for binary search variables
fs/btrfs/ctree.c int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,
Quoted 17 lines · David Sterba
The variables calculating where to jump next are useing mixed in types which requires some conversions on the instruction level. Using 'u32' removes one call to 'movslq', making the main loop shorter.
This complements type conversion done in
a724f313f84beb("btrfs: do unsigned integer division in the extent buffer binary search loop")Signed-off-by David Sterba <[email protected]>--- fs/btrfs/ctree.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)fs/btrfs/ctree.c +0
-0@@ -766,7 +766,7 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,
Does converting first_slot to u32 also add a movslq?
Also, why the "q"? Isn't everything 32 bit sized here? surprised to see it go to quad.
fs/btrfs/ctree.c int btrfs_bin_search(const struct extent_buffer *eb, int first_slot, · line 769
Quoted 5 lines
fs/btrfs/ctree.c +1
-1unsigned long offset;struct btrfs_disk_key *tmp;struct btrfs_disk_key unaligned; -int mid; +u32 mid;
shouldn't it be unsigned long long to theoretically avoid overflow? Not really sure why we are storing "nritems" as a u32, I doubt we ever want 4 billion items in a leaf :)
If there was a format enforcable limit, I suppose we could make the low/high smaller unsigned types?
Quoted 6 lines
fs/btrfs/ctree.c +0
-0mid = (low + high) /2 ; offset = p + mid * item_size;Signature · 1 line
2.51.1
Re: [PATCH 02/12] btrfs: unify types for binary search variables
Quoted 13 lines · Boris Burkov
Quoted 12 lines · David Sterba
The variables calculating where to jump next are useing mixed in types which requires some conversions on the instruction level. Using 'u32' removes one call to 'movslq', making the main loop shorter.
This complements type conversion done in
a724f313f84beb("btrfs: do unsigned integer division in the extent buffer binary search loop")Signed-off-by David Sterba <[email protected]>--- fs/btrfs/ctree.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 7 lines
fs/btrfs/ctree.c int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,
Quoted 5 lines
fs/btrfs/ctree.c +0
-0@@ -766,7 +766,7 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,Does converting
first_slotto u32 also add a movslq?
No, still a plain mov but there's some other effect that removes 2 reloads of the 2nd parameter to some other temporary register. I did not expect that, there's more instruction level tuning ahead.
Quoted 5 lines
Also, why the "q"? Isn't everything 32 bit sized here? surprised to see it go to quad.
Quoted 9 lines
fs/btrfs/ctree.c int btrfs_bin_search(const struct extent_buffer *eb, int first_slot, · line 769
Quoted 5 lines
fs/btrfs/ctree.c +1
-1unsigned long offset;struct btrfs_disk_key *tmp;struct btrfs_disk_key unaligned; -int mid; +u32 mid;shouldn't it be unsigned long long to theoretically avoid overflow? Not really sure why we are storing "nritems" as a u32, I doubt we ever want 4 billion items in a leaf :)
I don't think overflow can happen as the bounds are checked in many place before it gets to the bin search. The type width (btrfs_header::nritems) could be u16 and still cover the 64k leaves as the item isu >=4 for this to work.
Quoted 2 lines
If there was a format enforcable limit, I suppose we could make the low/high smaller unsigned types?
For the format it could be smaller in case we'd want to squeeze every byte out of that, but this is for the time when the format is designed and it's not now. At that time one does not know what would be needed so the types are chosen to be enough, while today we know well how it's used but can't go back to change the format.
The header size is 101 bytes, what could be reduced in size is flags and nritems, potentialy reducing it by 8 bytes to 93 (92%). This is per node/leaf, there are thousands per filesystem. On the biggest one I have next to me (20T) there are about 1.2m nodes, the metadata size is 21G. The savings translate to something like 160MiB. So yes it not zero.
For the calculations u32 is more convenient as it matches the native register size on many architectures.
David Sterba
maintainerPATCH 3/12 btrfs: rename local variable for offset in folio
In reply to David SterbaUse proper abbreviation of the 'offset in folio' in the variable name, same as we have in accessors.c.
--- fs/btrfs/ctree.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
fs/btrfs/ctree.c +4 -4
@@ -762,7 +762,7 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,while (low < high) {const int unit_size = eb->folio_size; -unsigned long oil; +unsigned long oif;unsigned long offset;struct btrfs_disk_key *tmp;struct btrfs_disk_key unaligned;
@@ -770,13 +770,13 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot, mid = (low + high) /2 ; offset = p + mid * item_size; - oil =get_eb_offset_in_folio (eb, offset); + oif =get_eb_offset_in_folio (eb, offset); -if (oil + key_size <= unit_size) { +if (oif + key_size <= unit_size) {const unsigned long idx =get_eb_folio_index (eb, offset);char *kaddr =folio_address (eb->folios[idx]); - tmp = (struct btrfs_disk_key *)(kaddr + oil); + tmp = (struct btrfs_disk_key *)(kaddr + oif); }else {read_extent_buffer (eb, &unaligned, offset, key_size); tmp = &unaligned;
Signature · 1 line
2.51.1
Re: [PATCH 03/12] btrfs: rename local variable for offset in folio
Quoted 4 lines · David Sterba
Use proper abbreviation of the 'offset in folio' in the variable name, same as we have in
accessors.c.Signed-off-by David Sterba <[email protected]>
Quoted 37 lines
--- fs/btrfs/ctree.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)fs/btrfs/ctree.c +4
-4@@ -762,7 +762,7 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot,while (low < high) {const int unit_size = eb->folio_size; -unsigned long oil; +unsigned long oif;unsigned long offset;struct btrfs_disk_key *tmp;struct btrfs_disk_key unaligned;@@ -770,13 +770,13 @@ int btrfs_bin_search(const struct extent_buffer *eb, int first_slot, mid = (low + high) /2 ; offset = p + mid * item_size; - oil =get_eb_offset_in_folio (eb, offset); + oif =get_eb_offset_in_folio (eb, offset); -if (oil + key_size <= unit_size) { +if (oif + key_size <= unit_size) {const unsigned long idx =get_eb_folio_index (eb, offset);char *kaddr =folio_address (eb->folios[idx]); - tmp = (struct btrfs_disk_key *)(kaddr + oil); + tmp = (struct btrfs_disk_key *)(kaddr + oif); }else {read_extent_buffer (eb, &unaligned, offset, key_size); tmp = &unaligned;Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 4/12 btrfs: read eb folio index right before loops
In reply to David SterbaThere are generic helpers to access extent buffer folio data of any length, potentially iterating over a few of them. This is a slow path, either we use the type based accessors or the eb folio allocation is contiguous and we can use the memcpy/memcmp helpers.
The initialization of 'i' is done at the beginning though it may not be needed. Move it right before the folio loop, this has minor effect on generated code in __write_extent_buffer().
--- fs/btrfs/extent_io.c | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-)
fs/btrfs/extent_io.c +10 -9
@@ -3960,7 +3960,7 @@ void read_extent_buffer(const struct extent_buffer *eb, void *dstv,size_t cur;size_t offset;char *dst = (char *)dstv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;if (check_eb_range (eb, start, len)) {/*
@@ -3977,7 +3977,7 @@ void read_extent_buffer(const struct extent_buffer *eb, void *dstv, } offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) {char *kaddr;
@@ -4000,7 +4000,7 @@ int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,size_t cur;size_t offset;char __user *dst = (char __user *)dstv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;int ret =0 ;WARN_ON (start > eb->len);
@@ -4013,7 +4013,7 @@ int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, } offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) {char *kaddr;
@@ -4041,7 +4041,7 @@ int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,size_t offset;char *kaddr;char *ptr = (char *)ptrv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;int ret =0 ;if (check_eb_range (eb, start, len))
@@ -4051,7 +4051,7 @@ int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,return memcmp (ptrv, eb->addr + start, len); offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) { cur =min (len, unit_size - offset); kaddr =folio_address (eb->folios[i]);
@@ -4111,7 +4111,7 @@ static void __write_extent_buffer(const struct extent_buffer *eb,size_t offset;char *kaddr;const char *src = (const char *)srcv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;/* For unmapped (dummy) ebs, no need to check their uptodate status. */ const bool check_uptodate = !test_bit (EXTENT_BUFFER_UNMAPPED , &eb->bflags);
@@ -4127,7 +4127,7 @@ static void __write_extent_buffer(const struct extent_buffer *eb, } offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) {if (check_uptodate)assert_eb_folio_uptodate (eb, i);
@@ -4213,7 +4213,7 @@ void copy_extent_buffer(const struct extent_buffer *dst,size_t cur;size_t offset;char *kaddr; -unsigned long i =get_eb_folio_index (dst, dst_offset); +unsigned long i;if (check_eb_range (dst, dst_offset, len) ||check_eb_range (src, src_offset, len))
@@ -4223,6 +4223,7 @@ void copy_extent_buffer(const struct extent_buffer *dst, offset =get_eb_offset_in_folio (dst, dst_offset); + i =get_eb_folio_index (dst, dst_offset);while (len >0 ) {assert_eb_folio_uptodate (dst, i);
Signature · 1 line
2.51.1
Re: [PATCH 04/12] btrfs: read eb folio index right before loops
Quoted 8 lines · David Sterba
There are generic helpers to access extent buffer folio data of any length, potentially iterating over a few of them. This is a slow path, either we use the type based accessors or the eb folio allocation is contiguous and we can use the memcpy/memcmp helpers.
The initialization of 'i' is done at the beginning though it may not be needed. Move it right before the folio loop, this has minor effect on generated code in
__write_extent_buffer().
This seems fine, but also pointless. One right shift that the compiler *can* move is causing us real pain? How often are we reading/writing extent_buffers? Should I be constantly worried about any arithmetic I ever do a hair early just to make code easier to read?
Anyway, if you do think it's worth it, you can add
Quoted 102 lines
Signed-off-by David Sterba <[email protected]>--- fs/btrfs/extent_io.c | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-)fs/btrfs/extent_io.c +10
-9@@ -3960,7 +3960,7 @@ void read_extent_buffer(const struct extent_buffer *eb, void *dstv,size_t cur;size_t offset;char *dst = (char *)dstv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;if (check_eb_range (eb, start, len)) {/* @@ -3977,7 +3977,7 @@ void read_extent_buffer(const struct extent_buffer *eb, void *dstv, } offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) {char *kaddr;@@ -4000,7 +4000,7 @@ int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,size_t cur;size_t offset;char __user *dst = (char __user *)dstv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;int ret =0 ;WARN_ON (start > eb->len);@@ -4013,7 +4013,7 @@ int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, } offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) {char *kaddr;@@ -4041,7 +4041,7 @@ int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,size_t offset;char *kaddr;char *ptr = (char *)ptrv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;int ret =0 ;if (check_eb_range (eb, start, len))@@ -4051,7 +4051,7 @@ int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,return memcmp (ptrv, eb->addr + start, len); offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) { cur =min (len, unit_size - offset); kaddr =folio_address (eb->folios[i]);@@ -4111,7 +4111,7 @@ static void __write_extent_buffer(const struct extent_buffer *eb,size_t offset;char *kaddr;const char *src = (const char *)srcv; -unsigned long i =get_eb_folio_index (eb, start); +unsigned long i;/* For unmapped (dummy) ebs, no need to check their uptodate status. */ const bool check_uptodate = !test_bit (EXTENT_BUFFER_UNMAPPED , &eb->bflags);@@ -4127,7 +4127,7 @@ static void __write_extent_buffer(const struct extent_buffer *eb, } offset =get_eb_offset_in_folio (eb, start); - + i =get_eb_folio_index (eb, start);while (len >0 ) {if (check_uptodate)assert_eb_folio_uptodate (eb, i);@@ -4213,7 +4213,7 @@ void copy_extent_buffer(const struct extent_buffer *dst,size_t cur;size_t offset;char *kaddr; -unsigned long i =get_eb_folio_index (dst, dst_offset); +unsigned long i;if (check_eb_range (dst, dst_offset, len) ||check_eb_range (src, src_offset, len))@@ -4223,6 +4223,7 @@ void copy_extent_buffer(const struct extent_buffer *dst, offset =get_eb_offset_in_folio (dst, dst_offset); + i =get_eb_folio_index (dst, dst_offset);while (len >0 ) {assert_eb_folio_uptodate (dst, i);Signature · 1 line
2.51.1
Re: [PATCH 04/12] btrfs: read eb folio index right before loops
Quoted 6 lines · Boris Burkov
Quoted 5 lines · David Sterba
There are generic helpers to access extent buffer folio data of any length, potentially iterating over a few of them. This is a slow path, either we use the type based accessors or the eb folio allocation is contiguous and we can use the memcpy/memcmp helpers.
Quoted 6 lines
Quoted 3 lines
The initialization of 'i' is done at the beginning though it may not be needed. Move it right before the folio loop, this has minor effect on generated code in
__write_extent_buffer().This seems fine, but also pointless. One right shift that the compiler *can* move is causing us real pain?
Compiler can move it but it does not, probably because there's a memory access. I'm not saying it's causing pain but that there's a fast path when the eb->addr is valid, then the get_eb_folio_index() is simply not done. The fast path also will apply to the large folios as this means the address range covering extent pages is contiguous.
Quoted 2 lines
How often are we reading/writing
extent_buffers?
For metadata heavy operations all the time, often enough that extent buffer access shows up in the perf top profiles at the top. The function read_extent_buffer() itself not as it's for unbounded length reads from extent buffer but otherwise it's accessing the shared structures.
Quoted 2 lines
Should I be constantly worried about any arithmetic I ever do a hair early just to make code easier to read?
No you don't have to worry about that (too much), I assume everybody writing C is somehow aware of various low level effects of the code but first it should work then we go optimize that. Doing separate optimization passes works better as there can be patterns spotted, hot functions identified etc. We're almost out of possible algorithmic improvements in the extent buffers realm so instruction level it is.
Quoted 3 lines
Anyway, if you do think it's worth it, you can add
Reviewed-by Boris Burkov <[email protected]>
Thanks. I unfortunatelly would be bothered by the code left around because I cannot unsee it. In the long run chipping away anything that does not need to be there would add up so we have fast code.
David Sterba
maintainerPATCH 5/12 btrfs: use common eb range validation in read_extent_buffer_to_user_nofault()
In reply to David SterbaThe extent buffer access is checked in other helpers by check_eb_range(), which validates the requested start, length against the extent buffer. While this almost never fails we should still handle it as an error and not just warn.
--- fs/btrfs/extent_io.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
fs/btrfs/extent_io.c +2 -2
@@ -4003,8 +4003,8 @@ int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,unsigned long i;int ret =0 ; -WARN_ON (start > eb->len); -WARN_ON (start + len > eb->start + eb->len); +if (check_eb_range (eb, start, len)) +return -EINVAL ;if (eb->addr) {if (copy_to_user_nofault (dstv, eb->addr + start, len))
Signature · 1 line
2.51.1
Re: [PATCH 05/12] btrfs: use common eb range validation in read_extent_buffer_to_user_nofault()
Quoted 6 lines · David Sterba
The extent buffer access is checked in other helpers by
check_eb_range(), which validates the requested start, length against the extent buffer. While this almost never fails we should still handle it as an error and not just warn.Signed-off-by David Sterba <[email protected]>
Quoted 22 lines
--- fs/btrfs/extent_io.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)fs/btrfs/extent_io.c +2
-2@@ -4003,8 +4003,8 @@ int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,unsigned long i;int ret =0 ; -WARN_ON (start > eb->len); -WARN_ON (start + len > eb->start + eb->len); +if (check_eb_range (eb, start, len)) +return -EINVAL ;if (eb->addr) {if (copy_to_user_nofault (dstv, eb->addr + start, len))Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 6/12 btrfs: lzo: inline read/write length helpers
In reply to David SterbaThe LZO_LEN read/write helpers are supposed to be trivial and we're duplicating the put/get unaligned helpers so use them directly.
--- fs/btrfs/lzo.c | 28 ++++++---------------------- 1 file changed, 6 insertions(+), 22 deletions(-)
fs/btrfs/lzo.c +6 -22
@@ -106,22 +106,6 @@ struct list_head *lzo_alloc_workspace(struct btrfs_fs_info *fs_info)return ERR_PTR (-ENOMEM ); } -static inline void write_compress_length (char *buf,size_t len) -{ -__le32 dlen; - - dlen =cpu_to_le32 (len); -memcpy (buf, &dlen,LZO_LEN ); -} - -static inline size_t read_compress_length (const char *buf) -{ -__le32 dlen; - -memcpy (&dlen, buf,LZO_LEN ); -return le32_to_cpu (dlen); -} -/* * Will do: *
@@ -165,7 +149,7 @@ static int copy_compressed_data_to_page(struct btrfs_fs_info *fs_info, } kaddr =kmap_local_folio (cur_folio,offset_in_folio (cur_folio, *cur_out)); -write_compress_length (kaddr, compressed_size); +put_unaligned_le32 (compressed_size, kaddr); *cur_out +=LZO_LEN ; orig_out = *cur_out;
@@ -297,7 +281,7 @@ int lzo_compress_folios(struct list_head *ws, struct btrfs_inode *inode,/* Store the size of all chunks of compressed data */ sizes_ptr =kmap_local_folio (folios[0 ],0 ); -write_compress_length (sizes_ptr, cur_out); +put_unaligned_le32 (cur_out, sizes_ptr);kunmap_local (sizes_ptr); ret =0 ;
@@ -352,7 +336,7 @@ int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb)u32 cur_out =0 ; kaddr =kmap_local_folio (cb->compressed_folios[0 ],0 ); - len_in =read_compress_length (kaddr); + len_in =get_unaligned_le32 (kaddr);kunmap_local (kaddr); cur_in +=LZO_LEN ;
@@ -391,7 +375,7 @@ int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb) cur_folio = cb->compressed_folios[cur_in >> min_folio_shift];ASSERT (cur_folio); kaddr =kmap_local_folio (cur_folio,0 ); - seg_len =read_compress_length (kaddr +offset_in_folio (cur_folio, cur_in)); + seg_len =get_unaligned_le32 (kaddr +offset_in_folio (cur_folio, cur_in));kunmap_local (kaddr); cur_in +=LZO_LEN ;
@@ -461,12 +445,12 @@ int lzo_decompress(struct list_head *ws, const u8 *data_in,if (unlikely (srclen <LZO_LEN || srclen > max_segment_len +LZO_LEN *2 ))return -EUCLEAN ; - in_len =read_compress_length (data_in); + in_len =get_unaligned_le32 (data_in);if (unlikely (in_len != srclen))return -EUCLEAN ; data_in +=LZO_LEN ; - in_len =read_compress_length (data_in); + in_len =get_unaligned_le32 (data_in);if (unlikely (in_len != srclen -LZO_LEN *2 )) { ret = -EUCLEAN ; goto out;
Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 7/12 btrfs: zlib: drop redundant folio address variable
In reply to David SterbaWe're caching the current output folio address but it's not really necessary as we store it in the variable and then pass it to the stream context. We can read the folio address directly.
--- fs/btrfs/zlib.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-)
fs/btrfs/zlib.c +3 -7
@@ -155,7 +155,6 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode,const u32 min_folio_size =btrfs_min_folio_size (fs_info);int ret;char *data_in =NULL ; -char *cfolio_out;int nr_folios =0 ;struct folio *in_folio =NULL ;struct folio *out_folio =NULL ;
@@ -186,13 +185,12 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, ret = -ENOMEM ; goto out; } - cfolio_out =folio_address (out_folio); folios[0 ] = out_folio; nr_folios =1 ; workspace->strm.next_in = workspace->buf; workspace->strm.avail_in =0 ; - workspace->strm.next_out = cfolio_out; + workspace->strm.next_out =folio_address (out_folio); workspace->strm.avail_out = min_folio_size;while (workspace->strm.total_in < len) {
@@ -270,11 +268,10 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, ret = -ENOMEM ; goto out; } - cfolio_out =folio_address (out_folio); folios[nr_folios] = out_folio; nr_folios++; workspace->strm.avail_out = min_folio_size; - workspace->strm.next_out = cfolio_out; + workspace->strm.next_out =folio_address (out_folio); }/* we're all done */ if (workspace->strm.total_in >= len)
@@ -306,11 +303,10 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, ret = -ENOMEM ; goto out; } - cfolio_out =folio_address (out_folio); folios[nr_folios] = out_folio; nr_folios++; workspace->strm.avail_out = min_folio_size; - workspace->strm.next_out = cfolio_out; + workspace->strm.next_out =folio_address (out_folio); } }zlib_deflateEnd (&workspace->strm);
Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 8/12 btrfs: zlib: don't cache sectorsize in a local variable
In reply to David SterbaThe sectorsize is used once or at most twice in the callbacks, no need to cache it on stack.
--- fs/btrfs/zlib.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-)
fs/btrfs/zlib.c +3 -5
@@ -71,7 +71,6 @@ static bool need_special_buffer(struct btrfs_fs_info *fs_info)struct list_head *zlib_alloc_workspace (struct btrfs_fs_info *fs_info,unsigned int level) { -const u32 blocksize = fs_info->sectorsize;struct workspace *workspace;int workspacesize;
@@ -91,8 +90,8 @@ struct list_head *zlib_alloc_workspace(struct btrfs_fs_info *fs_info, unsigned i workspace->buf_size =ZLIB_DFLTCC_BUF_SIZE ; }if (!workspace->buf) { - workspace->buf =kmalloc (blocksize,GFP_KERNEL ); - workspace->buf_size = blocksize; + workspace->buf =kmalloc (fs_info->sectorsize,GFP_KERNEL ); + workspace->buf_size = fs_info->sectorsize; }if (!workspace->strm.workspace || !workspace->buf) goto fail;
@@ -161,7 +160,6 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode,unsigned long len = *total_out;unsigned long nr_dest_folios = *out_folios;const unsigned long max_out = nr_dest_folios << min_folio_shift; -const u32 blocksize = fs_info->sectorsize;const u64 orig_end = start + len; *out_folios =0 ;
@@ -248,7 +246,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, }/* we're making it bigger, give up */ -if (workspace->strm.total_in > blocksize *2 && +if (workspace->strm.total_in > fs_info->sectorsize *2 && workspace->strm.total_in < workspace->strm.total_out) { ret = -E2BIG ;
Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 9/12 btrfs: zlib: remove local variable nr_dest_folios in zlib_compress_folios()
In reply to David SterbaThe value of *out_folios does not change and nr_dest_folios is only a local copy, we can remove it. This saves 8 bytes of stack.
--- fs/btrfs/zlib.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
fs/btrfs/zlib.c +3 -4
@@ -158,8 +158,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode,struct folio *in_folio =NULL ;struct folio *out_folio =NULL ;unsigned long len = *total_out; -unsigned long nr_dest_folios = *out_folios; -const unsigned long max_out = nr_dest_folios << min_folio_shift; +const unsigned long max_out = *out_folios << min_folio_shift;const u64 orig_end = start + len; *out_folios =0 ;
@@ -257,7 +256,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode,* the stream end if required */ if (workspace->strm.avail_out ==0 ) { -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }
@@ -292,7 +291,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, goto out; }else if (workspace->strm.avail_out ==0 ) {/* Get another folio for the stream end. */ -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }
Signature · 1 line
2.51.1
Re: [PATCH 09/12] btrfs: zlib: remove local variable nr_dest_folios in zlib_compress_folios()
Quoted 13 lines · David Sterba
The value of
*out_foliosdoes not change andnr_dest_foliosis only a local copy, we can remove it. This saves 8 bytes of stack.Signed-off-by David Sterba <[email protected]>--- fs/btrfs/zlib.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)fs/btrfs/zlib.c +0
-0
fs/btrfs/zlib.c int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, · line 164
Quoted 9 lines
fs/btrfs/zlib.c +1
-2@@ -158,8 +158,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode,struct folio *in_folio =NULL ;struct folio *out_folio =NULL ;unsigned long len = *total_out; -unsigned long nr_dest_folios = *out_folios; -const unsigned long max_out = nr_dest_folios << min_folio_shift; +const unsigned long max_out = *out_folios << min_folio_shift;const u64 orig_end = start + len; *out_folios =0 ;
I may be missing something, but it looks like it does change here? Then it only gets set to nr_folios at the out: label. So in the other two uses of nr_dest_folios, *out_folios is wrong.
Quoted 21 lines
fs/btrfs/zlib.c +2
-2@@ -257,7 +256,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode,* the stream end if required */ if (workspace->strm.avail_out ==0 ) { -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }@@ -292,7 +291,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, goto out; }else if (workspace->strm.avail_out ==0 ) {/* Get another folio for the stream end. */ -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }Signature · 1 line
2.51.1
Re: [PATCH 09/12] btrfs: zlib: remove local variable nr_dest_folios in zlib_compress_folios()
Quoted 22 lines · Boris Burkov
Quoted 21 lines · David Sterba
The value of
*out_foliosdoes not change andnr_dest_foliosis only a local copy, we can remove it. This saves 8 bytes of stack.Signed-off-by David Sterba <[email protected]>--- fs/btrfs/zlib.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)fs/btrfs/zlib.c +1
-2@@ -158,8 +158,7 @@ int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode,struct folio *in_folio =NULL ;struct folio *out_folio =NULL ;unsigned long len = *total_out; -unsigned long nr_dest_folios = *out_folios; -const unsigned long max_out = nr_dest_folios << min_folio_shift; +const unsigned long max_out = *out_folios << min_folio_shift;const u64 orig_end = start + len;
Quoted 5 lines
fs/btrfs/zlib.c int zlib_compress_folios(struct list_head *ws, struct btrfs_inode *inode, · line 164
Quoted 1 line
fs/btrfs/zlib.c +0
-0*out_folios =0 ;I may be missing something, but it looks like it does change here? Then it only gets set to
nr_foliosat the out: label. So in the other two uses ofnr_dest_folios,*out_foliosis wrong.
You're right, the "*out_folios = 0" would have to be removed as well. The wording was incorect, what I meant is that it was not changed until the out label, as you say. The nr_folios and valid entries in folios are always in sync (regarding the code flow) so removing the out_folios initialization should be sufficient.
David Sterba
maintainerPATCH 10/12 btrfs: zstd: reuse total in and out parameters for calculations
In reply to David SterbaReduce the stack consumption which is 240 bytes on release config by 16 bytes. The local variables are not adding anything on top of the parameters. As a calling convention if the compression helper returns an error the parameters are considered invalid.
--- fs/btrfs/zstd.c | 26 +++++++++++--------------- 1 file changed, 11 insertions(+), 15 deletions(-)
fs/btrfs/zstd.c +11 -15
@@ -408,8 +408,6 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,int nr_folios =0 ;struct folio *in_folio =NULL ;/* The current folio to read. */ struct folio *out_folio =NULL ;/* The current folio to write to. */ -unsigned long tot_in =0 ; -unsigned long tot_out =0 ;unsigned long len = *total_out;const unsigned long nr_dest_folios = *out_folios;const u64 orig_end = start + len;
@@ -471,23 +469,23 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, }/* Check to see if we are making it bigger */ -if (tot_in + workspace->in_buf.pos > blocksize *2 && - tot_in + workspace->in_buf.pos < - tot_out + workspace->out_buf.pos) { +if (*total_in + workspace->in_buf.pos > blocksize *2 && + *total_in + workspace->in_buf.pos < + *total_out + workspace->out_buf.pos) { ret = -E2BIG ; goto out; }/* We've reached the end of our output range */ if (workspace->out_buf.pos >= max_out) { - tot_out += workspace->out_buf.pos; + *total_out += workspace->out_buf.pos; ret = -E2BIG ; goto out; }/* Check if we need more output space */ if (workspace->out_buf.pos == workspace->out_buf.size) { - tot_out += min_folio_size; + *total_out += min_folio_size; max_out -= min_folio_size;if (nr_folios == nr_dest_folios) { ret = -E2BIG ;
@@ -506,13 +504,13 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,/* We've reached the end of the input */ if (workspace->in_buf.pos >= len) { - tot_in += workspace->in_buf.pos; + *total_in += workspace->in_buf.pos;break ; }/* Check if we need more input */ if (workspace->in_buf.pos == workspace->in_buf.size) { - tot_in += workspace->in_buf.size; + *total_in += workspace->in_buf.size;kunmap_local (workspace->in_buf.src); workspace->in_buf.src =NULL ;folio_put (in_folio);
@@ -542,16 +540,16 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, goto out; }if (ret2 ==0 ) { - tot_out += workspace->out_buf.pos; + *total_out += workspace->out_buf.pos;break ; }if (workspace->out_buf.pos >= max_out) { - tot_out += workspace->out_buf.pos; + *total_out += workspace->out_buf.pos; ret = -E2BIG ; goto out; } - tot_out += min_folio_size; + *total_out += min_folio_size; max_out -= min_folio_size;if (nr_folios == nr_dest_folios) { ret = -E2BIG ;
@@ -568,14 +566,12 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, workspace->out_buf.size =min_t (size_t, max_out, min_folio_size); } -if (tot_out >= tot_in) { +if (*total_out >= *total_in) { ret = -E2BIG ; goto out; } ret =0 ; - *total_in = tot_in; - *total_out = tot_out; out: *out_folios = nr_folios;if (workspace->in_buf.src) {
Signature · 1 line
2.51.1
Re: [PATCH 10/12] btrfs: zstd: reuse total in and out parameters for calculations
Quoted 4 lines · David Sterba
Reduce the stack consumption which is 240 bytes on release config by 16 bytes. The local variables are not adding anything on top of the parameters. As a calling convention if the compression helper returns an error the parameters are considered invalid.
I don't think that's technically true, as btrfs_compress_folios() does an ASSERT(*total_in < orig_len) unconditionally.
However, I don't think it makes your change invalid.
Quoted 103 lines
Signed-off-by David Sterba <[email protected]>--- fs/btrfs/zstd.c | 26 +++++++++++--------------- 1 file changed, 11 insertions(+), 15 deletions(-)fs/btrfs/zstd.c +11
-15@@ -408,8 +408,6 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,int nr_folios =0 ;struct folio *in_folio =NULL ;/* The current folio to read. */ struct folio *out_folio =NULL ;/* The current folio to write to. */ -unsigned long tot_in =0 ; -unsigned long tot_out =0 ;unsigned long len = *total_out;const unsigned long nr_dest_folios = *out_folios;const u64 orig_end = start + len;@@ -471,23 +469,23 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, }/* Check to see if we are making it bigger */ -if (tot_in + workspace->in_buf.pos > blocksize *2 && - tot_in + workspace->in_buf.pos < - tot_out + workspace->out_buf.pos) { +if (*total_in + workspace->in_buf.pos > blocksize *2 && + *total_in + workspace->in_buf.pos < + *total_out + workspace->out_buf.pos) { ret = -E2BIG ; goto out; }/* We've reached the end of our output range */ if (workspace->out_buf.pos >= max_out) { - tot_out += workspace->out_buf.pos; + *total_out += workspace->out_buf.pos; ret = -E2BIG ; goto out; }/* Check if we need more output space */ if (workspace->out_buf.pos == workspace->out_buf.size) { - tot_out += min_folio_size; + *total_out += min_folio_size; max_out -= min_folio_size;if (nr_folios == nr_dest_folios) { ret = -E2BIG ;@@ -506,13 +504,13 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,/* We've reached the end of the input */ if (workspace->in_buf.pos >= len) { - tot_in += workspace->in_buf.pos; + *total_in += workspace->in_buf.pos;break ; }/* Check if we need more input */ if (workspace->in_buf.pos == workspace->in_buf.size) { - tot_in += workspace->in_buf.size; + *total_in += workspace->in_buf.size;kunmap_local (workspace->in_buf.src); workspace->in_buf.src =NULL ;folio_put (in_folio);@@ -542,16 +540,16 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, goto out; }if (ret2 ==0 ) { - tot_out += workspace->out_buf.pos; + *total_out += workspace->out_buf.pos;break ; }if (workspace->out_buf.pos >= max_out) { - tot_out += workspace->out_buf.pos; + *total_out += workspace->out_buf.pos; ret = -E2BIG ; goto out; } - tot_out += min_folio_size; + *total_out += min_folio_size; max_out -= min_folio_size;if (nr_folios == nr_dest_folios) { ret = -E2BIG ;@@ -568,14 +566,12 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, workspace->out_buf.size =min_t (size_t, max_out, min_folio_size); } -if (tot_out >= tot_in) { +if (*total_out >= *total_in) { ret = -E2BIG ; goto out; } ret =0 ; - *total_in = tot_in; - *total_out = tot_out; out: *out_folios = nr_folios;if (workspace->in_buf.src) {Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 11/12 btrfs: zstd: don't cache sectorsize in a local variable
In reply to David SterbaThe sectorsize is used once or at most twice in the callbacks, no need to cache it on stack. Minor effect on zstd_compress_folios() where it saves 8 bytes of stack.
--- fs/btrfs/zstd.c | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-)
fs/btrfs/zstd.c +4 -8
@@ -370,7 +370,6 @@ void zstd_free_workspace(struct list_head *ws)struct list_head *zstd_alloc_workspace (struct btrfs_fs_info *fs_info,int level) { -const u32 blocksize = fs_info->sectorsize;struct workspace *workspace; workspace =kzalloc (sizeof (*workspace),GFP_KERNEL );
@@ -383,7 +382,7 @@ struct list_head *zstd_alloc_workspace(struct btrfs_fs_info *fs_info, int level) workspace->req_level = level; workspace->last_used = jiffies; workspace->mem =kvmalloc (workspace->size,GFP_KERNEL | __GFP_NOWARN); - workspace->buf =kmalloc (blocksize,GFP_KERNEL ); + workspace->buf =kmalloc (fs_info->sectorsize,GFP_KERNEL );if (!workspace->mem || !workspace->buf) goto fail;
@@ -411,7 +410,6 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,unsigned long len = *total_out;const unsigned long nr_dest_folios = *out_folios;const u64 orig_end = start + len; -const u32 blocksize = fs_info->sectorsize;const u32 min_folio_size =btrfs_min_folio_size (fs_info);unsigned long max_out = nr_dest_folios * min_folio_size;unsigned int cur_len;
@@ -469,7 +467,7 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, }/* Check to see if we are making it bigger */ -if (*total_in + workspace->in_buf.pos > blocksize *2 && +if (*total_in + workspace->in_buf.pos > fs_info->sectorsize *2 && *total_in + workspace->in_buf.pos < *total_out + workspace->out_buf.pos) { ret = -E2BIG ;
@@ -589,7 +587,6 @@ int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)size_t srclen = cb->compressed_len;zstd_dstream *stream;int ret =0 ; -const u32 blocksize = fs_info->sectorsize;const unsigned int min_folio_size =btrfs_min_folio_size (fs_info);unsigned long folio_in_index =0 ;unsigned long total_folios_in =DIV_ROUND_UP (srclen, min_folio_size);
@@ -614,7 +611,7 @@ int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb) workspace->out_buf.dst = workspace->buf; workspace->out_buf.pos =0 ; - workspace->out_buf.size = blocksize; + workspace->out_buf.size = fs_info->sectorsize;while (1 ) {size_t ret2;
@@ -675,7 +672,6 @@ int zstd_decompress(struct list_head *ws, const u8 *data_in, { struct workspace *workspace = list_entry(ws, struct workspace, list);struct btrfs_fs_info *fs_info =btrfs_sb (folio_inode (dest_folio)->i_sb); -const u32 sectorsize = fs_info->sectorsize;zstd_dstream *stream;int ret =0 ;unsigned long to_copy =0 ;
@@ -699,7 +695,7 @@ int zstd_decompress(struct list_head *ws, const u8 *data_in, workspace->out_buf.dst = workspace->buf; workspace->out_buf.pos =0 ; - workspace->out_buf.size = sectorsize; + workspace->out_buf.size = fs_info->sectorsize;/* * Since both input and output buffers should not exceed one sector,
Signature · 1 line
2.51.1
Re: [PATCH 11/12] btrfs: zstd: don't cache sectorsize in a local variable
Quoted 4 lines · David Sterba
The sectorsize is used once or at most twice in the callbacks, no need to cache it on stack. Minor effect on
zstd_compress_folios()where it saves 8 bytes of stack.
Quoted 80 lines
Signed-off-by David Sterba <[email protected]>--- fs/btrfs/zstd.c | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-)fs/btrfs/zstd.c +4
-8@@ -370,7 +370,6 @@ void zstd_free_workspace(struct list_head *ws)struct list_head *zstd_alloc_workspace (struct btrfs_fs_info *fs_info,int level) { -const u32 blocksize = fs_info->sectorsize;struct workspace *workspace; workspace =kzalloc (sizeof (*workspace),GFP_KERNEL );@@ -383,7 +382,7 @@ struct list_head *zstd_alloc_workspace(struct btrfs_fs_info *fs_info, int level) workspace->req_level = level; workspace->last_used = jiffies; workspace->mem =kvmalloc (workspace->size,GFP_KERNEL | __GFP_NOWARN); - workspace->buf =kmalloc (blocksize,GFP_KERNEL ); + workspace->buf =kmalloc (fs_info->sectorsize,GFP_KERNEL );if (!workspace->mem || !workspace->buf) goto fail;@@ -411,7 +410,6 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,unsigned long len = *total_out;const unsigned long nr_dest_folios = *out_folios;const u64 orig_end = start + len; -const u32 blocksize = fs_info->sectorsize;const u32 min_folio_size =btrfs_min_folio_size (fs_info);unsigned long max_out = nr_dest_folios * min_folio_size;unsigned int cur_len;@@ -469,7 +467,7 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, }/* Check to see if we are making it bigger */ -if (*total_in + workspace->in_buf.pos > blocksize *2 && +if (*total_in + workspace->in_buf.pos > fs_info->sectorsize *2 && *total_in + workspace->in_buf.pos < *total_out + workspace->out_buf.pos) { ret = -E2BIG ;@@ -589,7 +587,6 @@ int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)size_t srclen = cb->compressed_len;zstd_dstream *stream;int ret =0 ; -const u32 blocksize = fs_info->sectorsize;const unsigned int min_folio_size =btrfs_min_folio_size (fs_info);unsigned long folio_in_index =0 ;unsigned long total_folios_in =DIV_ROUND_UP (srclen, min_folio_size);@@ -614,7 +611,7 @@ int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb) workspace->out_buf.dst = workspace->buf; workspace->out_buf.pos =0 ; - workspace->out_buf.size = blocksize; + workspace->out_buf.size = fs_info->sectorsize;while (1 ) {size_t ret2;@@ -675,7 +672,6 @@ int zstd_decompress(struct list_head *ws, const u8 *data_in, { struct workspace *workspace = list_entry(ws, struct workspace, list);struct btrfs_fs_info *fs_info =btrfs_sb (folio_inode (dest_folio)->i_sb); -const u32 sectorsize = fs_info->sectorsize;zstd_dstream *stream;int ret =0 ;unsigned long to_copy =0 ;@@ -699,7 +695,7 @@ int zstd_decompress(struct list_head *ws, const u8 *data_in, workspace->out_buf.dst = workspace->buf; workspace->out_buf.pos =0 ; - workspace->out_buf.size = sectorsize; + workspace->out_buf.size = fs_info->sectorsize;/* * Since both input and output buffers should not exceed one sector, Signature · 1 line
2.51.1
David Sterba
maintainerPATCH 12/12 btrfs: zstd: remove local variable nr_dest_folios in zstd_compress_folios()
In reply to David SterbaThe value of *out_folios does not change and nr_dest_folios is only a local copy, we can remove it. This saves 8 bytes of stack.
--- fs/btrfs/zstd.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
fs/btrfs/zstd.c +3 -4
@@ -408,10 +408,9 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,struct folio *in_folio =NULL ;/* The current folio to read. */ struct folio *out_folio =NULL ;/* The current folio to write to. */ unsigned long len = *total_out; -const unsigned long nr_dest_folios = *out_folios;const u64 orig_end = start + len;const u32 min_folio_size =btrfs_min_folio_size (fs_info); -unsigned long max_out = nr_dest_folios * min_folio_size; +unsigned long max_out = *out_folios * min_folio_size;unsigned int cur_len; workspace->params =zstd_get_btrfs_parameters (workspace->req_level, len);
@@ -485,7 +484,7 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,if (workspace->out_buf.pos == workspace->out_buf.size) { *total_out += min_folio_size; max_out -= min_folio_size; -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }
@@ -549,7 +548,7 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, *total_out += min_folio_size; max_out -= min_folio_size; -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }
Signature · 1 line
2.51.1
Re: [PATCH 12/12] btrfs: zstd: remove local variable nr_dest_folios in zstd_compress_folios()
Quoted 2 lines · David Sterba
The value of
*out_foliosdoes not change andnr_dest_foliosis only a local copy, we can remove it. This saves 8 bytes of stack.
To my eye, this one has the same bug as the zlib one.
Quoted 43 lines
Signed-off-by David Sterba <[email protected]>--- fs/btrfs/zstd.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)fs/btrfs/zstd.c +3
-4@@ -408,10 +408,9 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,struct folio *in_folio =NULL ;/* The current folio to read. */ struct folio *out_folio =NULL ;/* The current folio to write to. */ unsigned long len = *total_out; -const unsigned long nr_dest_folios = *out_folios;const u64 orig_end = start + len;const u32 min_folio_size =btrfs_min_folio_size (fs_info); -unsigned long max_out = nr_dest_folios * min_folio_size; +unsigned long max_out = *out_folios * min_folio_size;unsigned int cur_len; workspace->params =zstd_get_btrfs_parameters (workspace->req_level, len);@@ -485,7 +484,7 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode,if (workspace->out_buf.pos == workspace->out_buf.size) { *total_out += min_folio_size; max_out -= min_folio_size; -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }@@ -549,7 +548,7 @@ int zstd_compress_folios(struct list_head *ws, struct btrfs_inode *inode, *total_out += min_folio_size; max_out -= min_folio_size; -if (nr_folios == nr_dest_folios) { +if (nr_folios == *out_folios) { ret = -E2BIG ; goto out; }Signature · 1 line
2.51.1