2018-04-03 20:23:33 +03:00
// SPDX-License-Identifier: GPL-2.0
2007-10-16 00:18:56 +04:00
/*
* Copyright ( C ) 2007 Oracle . All rights reserved .
*/
2012-07-10 06:22:35 +04:00
# include <asm/unaligned.h>
2022-10-19 17:50:49 +03:00
# include "messages.h"
2012-07-10 06:22:35 +04:00
# include "ctree.h"
2022-10-19 17:50:59 +03:00
# include "accessors.h"
2012-07-10 06:22:35 +04:00
2020-05-01 00:38:11 +03:00
static bool check_setget_bounds ( const struct extent_buffer * eb ,
const void * ptr , unsigned off , int size )
{
const unsigned long member_offset = ( unsigned long ) ptr + off ;
btrfs: remove redundant check in up check_setget_bounds
There are two separate checks in the bounds checker, the first one being
a special case of the second. As this function is performance critical
due to checking access to any eb member, reducing the size can slightly
improve performance.
On a release build on x86_64 the helper is completely inlined so the
function call overhead is also gone.
There was a report of 5% performance drop on metadata heavy workload,
that disappeared after disabling asserts. The most significant part of
that is the bounds checker.
https://lore.kernel.org/linux-btrfs/20200724164147.39925-1-josef@toxicpanda.com/
After the analysis, the optimized code removes the worst overhead which
is the function call and the performance was restored.
https://lore.kernel.org/linux-btrfs/20200730110943.GE3703@twin.jikos.cz/
1. baseline, asserts on, setget check on
run time: 46s
run time with perf: 48s
2. asserts on, comment out setget check
run time: 44s
run time with perf: 47s
So this is confirms the 5% difference
3. asserts on, optimized seget check
run time: 44s
run time with perf: 47s
The optimizations are reducing the number of ifs to 1 and inlining the
hot path. Low-level stuff, gets the performance back. Patch below.
4. asserts off, no setget check
run time: 44s
run time with perf: 45s
This verifies that asserts other than the setget check have negligible
impact on performance and it's not harmful to keep them on.
Analysis where the performance is lost:
* check_setget_bounds is short function, but it's still a function call,
changing the flow of instructions and given how many times it's
called the overhead adds up
* there are two conditions, one to check if the range is
completely outside (member_offset > eb->len) or partially inside
(member_offset + size > eb->len)
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-07-27 21:59:20 +03:00
if ( unlikely ( member_offset + size > eb - > len ) ) {
2020-05-01 00:38:11 +03:00
btrfs_warn ( eb - > fs_info ,
btrfs: remove redundant check in up check_setget_bounds
There are two separate checks in the bounds checker, the first one being
a special case of the second. As this function is performance critical
due to checking access to any eb member, reducing the size can slightly
improve performance.
On a release build on x86_64 the helper is completely inlined so the
function call overhead is also gone.
There was a report of 5% performance drop on metadata heavy workload,
that disappeared after disabling asserts. The most significant part of
that is the bounds checker.
https://lore.kernel.org/linux-btrfs/20200724164147.39925-1-josef@toxicpanda.com/
After the analysis, the optimized code removes the worst overhead which
is the function call and the performance was restored.
https://lore.kernel.org/linux-btrfs/20200730110943.GE3703@twin.jikos.cz/
1. baseline, asserts on, setget check on
run time: 46s
run time with perf: 48s
2. asserts on, comment out setget check
run time: 44s
run time with perf: 47s
So this is confirms the 5% difference
3. asserts on, optimized seget check
run time: 44s
run time with perf: 47s
The optimizations are reducing the number of ifs to 1 and inlining the
hot path. Low-level stuff, gets the performance back. Patch below.
4. asserts off, no setget check
run time: 44s
run time with perf: 45s
This verifies that asserts other than the setget check have negligible
impact on performance and it's not harmful to keep them on.
Analysis where the performance is lost:
* check_setget_bounds is short function, but it's still a function call,
changing the flow of instructions and given how many times it's
called the overhead adds up
* there are two conditions, one to check if the range is
completely outside (member_offset > eb->len) or partially inside
(member_offset + size > eb->len)
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-07-27 21:59:20 +03:00
" bad eb member %s: ptr 0x%lx start %llu member offset %lu size %d " ,
( member_offset > eb - > len ? " start " : " end " ) ,
2020-05-01 00:38:11 +03:00
( unsigned long ) ptr , eb - > start , member_offset , size ) ;
return false ;
}
return true ;
}
2022-10-19 17:50:59 +03:00
void btrfs_init_map_token ( struct btrfs_map_token * token , struct extent_buffer * eb )
{
token - > eb = eb ;
token - > kaddr = page_address ( eb - > pages [ 0 ] ) ;
token - > offset = 0 ;
}
2012-07-10 06:22:35 +04:00
/*
2020-05-06 21:54:13 +03:00
* Macro templates that define helpers to read / write extent buffer data of a
* given size , that are also used via ctree . h for access to item members by
* specialized helpers .
2008-09-29 23:18:18 +04:00
*
2020-05-06 21:54:13 +03:00
* Generic helpers :
* - btrfs_set_8 ( for 8 / 16 / 32 / 64 )
* - btrfs_get_8 ( for 8 / 16 / 32 / 64 )
2008-09-29 23:18:18 +04:00
*
2020-05-06 21:54:13 +03:00
* Generic helpers with a token ( cached address of the most recently accessed
* page ) :
* - btrfs_set_token_8 ( for 8 / 16 / 32 / 64 )
* - btrfs_get_token_8 ( for 8 / 16 / 32 / 64 )
2008-09-29 23:18:18 +04:00
*
2020-05-06 21:54:13 +03:00
* The set / get functions handle data spanning two pages transparently , in case
* metadata block size is larger than page . Every pointer to metadata items is
* an offset into the extent buffer page array , cast to a specific type . This
* gives us all the type checking .
2019-08-09 18:12:38 +03:00
*
2020-05-06 21:54:13 +03:00
* The extent buffer pages stored in the array pages do not form a contiguous
* phyusical range , but the API functions assume the linear offset to the range
* from 0 to metadata node size .
2008-09-29 23:18:18 +04:00
*/
2012-07-10 06:22:35 +04:00
# define DEFINE_BTRFS_SETGET_BITS(bits) \
2020-04-29 03:15:56 +03:00
u # # bits btrfs_get_token_ # # bits ( struct btrfs_map_token * token , \
const void * ptr , unsigned long off ) \
2007-10-16 00:18:56 +04:00
{ \
2020-04-29 18:45:33 +03:00
const unsigned long member_offset = ( unsigned long ) ptr + off ; \
2020-12-02 09:48:04 +03:00
const unsigned long idx = get_eb_page_index ( member_offset ) ; \
const unsigned long oip = get_eb_offset_in_page ( token - > eb , \
member_offset ) ; \
2020-04-29 18:45:33 +03:00
const int size = sizeof ( u # # bits ) ; \
2020-04-30 18:57:55 +03:00
u8 lebytes [ sizeof ( u # # bits ) ] ; \
const int part = PAGE_SIZE - oip ; \
2012-07-10 06:22:35 +04:00
\
2019-08-09 18:30:23 +03:00
ASSERT ( token ) ; \
2020-04-29 20:29:04 +03:00
ASSERT ( token - > kaddr ) ; \
2020-05-01 00:38:11 +03:00
ASSERT ( check_setget_bounds ( token - > eb , ptr , off , size ) ) ; \
2020-04-29 18:45:33 +03:00
if ( token - > offset < = member_offset & & \
member_offset + size < = token - > offset + PAGE_SIZE ) { \
return get_unaligned_le # # bits ( token - > kaddr + oip ) ; \
2012-07-10 06:22:35 +04:00
} \
2020-04-30 18:57:55 +03:00
token - > kaddr = page_address ( token - > eb - > pages [ idx ] ) ; \
token - > offset = idx < < PAGE_SHIFT ; \
2021-06-23 16:48:53 +03:00
if ( INLINE_EXTENT_BUFFER_PAGES = = 1 | | oip + size < = PAGE_SIZE ) \
2020-04-29 18:45:33 +03:00
return get_unaligned_le # # bits ( token - > kaddr + oip ) ; \
2020-04-30 18:57:55 +03:00
\
memcpy ( lebytes , token - > kaddr + oip , part ) ; \
2020-04-29 18:45:33 +03:00
token - > kaddr = page_address ( token - > eb - > pages [ idx + 1 ] ) ; \
token - > offset = ( idx + 1 ) < < PAGE_SHIFT ; \
2020-04-30 18:57:55 +03:00
memcpy ( lebytes + part , token - > kaddr , size - part ) ; \
return get_unaligned_le # # bits ( lebytes ) ; \
2007-10-16 00:18:56 +04:00
} \
2019-08-09 18:12:38 +03:00
u # # bits btrfs_get_ # # bits ( const struct extent_buffer * eb , \
const void * ptr , unsigned long off ) \
{ \
2020-04-29 17:04:44 +03:00
const unsigned long member_offset = ( unsigned long ) ptr + off ; \
2020-12-02 09:48:04 +03:00
const unsigned long oip = get_eb_offset_in_page ( eb , member_offset ) ; \
const unsigned long idx = get_eb_page_index ( member_offset ) ; \
2020-04-30 18:57:55 +03:00
char * kaddr = page_address ( eb - > pages [ idx ] ) ; \
2020-04-29 17:04:44 +03:00
const int size = sizeof ( u # # bits ) ; \
2020-04-30 18:57:55 +03:00
const int part = PAGE_SIZE - oip ; \
u8 lebytes [ sizeof ( u # # bits ) ] ; \
2019-08-09 18:12:38 +03:00
\
2020-05-01 00:38:11 +03:00
ASSERT ( check_setget_bounds ( eb , ptr , off , size ) ) ; \
2021-06-23 16:48:53 +03:00
if ( INLINE_EXTENT_BUFFER_PAGES = = 1 | | oip + size < = PAGE_SIZE ) \
2020-04-29 17:04:44 +03:00
return get_unaligned_le # # bits ( kaddr + oip ) ; \
2020-04-30 18:57:55 +03:00
\
memcpy ( lebytes , kaddr + oip , part ) ; \
kaddr = page_address ( eb - > pages [ idx + 1 ] ) ; \
memcpy ( lebytes + part , kaddr , size - part ) ; \
return get_unaligned_le # # bits ( lebytes ) ; \
2019-08-09 18:12:38 +03:00
} \
2020-04-29 03:15:56 +03:00
void btrfs_set_token_ # # bits ( struct btrfs_map_token * token , \
2017-06-29 06:56:53 +03:00
const void * ptr , unsigned long off , \
2020-04-29 03:15:56 +03:00
u # # bits val ) \
2007-10-16 00:18:56 +04:00
{ \
2020-04-29 19:23:37 +03:00
const unsigned long member_offset = ( unsigned long ) ptr + off ; \
2020-12-02 09:48:04 +03:00
const unsigned long idx = get_eb_page_index ( member_offset ) ; \
const unsigned long oip = get_eb_offset_in_page ( token - > eb , \
member_offset ) ; \
2020-04-29 19:23:37 +03:00
const int size = sizeof ( u # # bits ) ; \
2020-04-30 18:57:55 +03:00
u8 lebytes [ sizeof ( u # # bits ) ] ; \
const int part = PAGE_SIZE - oip ; \
2012-07-10 06:22:35 +04:00
\
2019-08-09 18:30:23 +03:00
ASSERT ( token ) ; \
2020-04-29 20:29:04 +03:00
ASSERT ( token - > kaddr ) ; \
2020-05-01 00:38:11 +03:00
ASSERT ( check_setget_bounds ( token - > eb , ptr , off , size ) ) ; \
2020-04-29 19:23:37 +03:00
if ( token - > offset < = member_offset & & \
member_offset + size < = token - > offset + PAGE_SIZE ) { \
put_unaligned_le # # bits ( val , token - > kaddr + oip ) ; \
2012-07-10 06:22:35 +04:00
return ; \
} \
2020-04-30 18:57:55 +03:00
token - > kaddr = page_address ( token - > eb - > pages [ idx ] ) ; \
token - > offset = idx < < PAGE_SHIFT ; \
2021-06-23 16:48:53 +03:00
if ( INLINE_EXTENT_BUFFER_PAGES = = 1 | | oip + size < = PAGE_SIZE ) { \
2020-04-29 19:23:37 +03:00
put_unaligned_le # # bits ( val , token - > kaddr + oip ) ; \
2012-07-10 06:22:35 +04:00
return ; \
} \
2020-04-30 18:57:55 +03:00
put_unaligned_le # # bits ( val , lebytes ) ; \
memcpy ( token - > kaddr + oip , lebytes , part ) ; \
2020-04-29 19:23:37 +03:00
token - > kaddr = page_address ( token - > eb - > pages [ idx + 1 ] ) ; \
token - > offset = ( idx + 1 ) < < PAGE_SHIFT ; \
2020-04-30 18:57:55 +03:00
memcpy ( token - > kaddr , lebytes + part , size - part ) ; \
2019-08-09 18:12:38 +03:00
} \
2020-04-29 04:04:10 +03:00
void btrfs_set_ # # bits ( const struct extent_buffer * eb , void * ptr , \
2019-08-09 18:12:38 +03:00
unsigned long off , u # # bits val ) \
{ \
2020-04-29 19:07:04 +03:00
const unsigned long member_offset = ( unsigned long ) ptr + off ; \
2020-12-02 09:48:04 +03:00
const unsigned long oip = get_eb_offset_in_page ( eb , member_offset ) ; \
const unsigned long idx = get_eb_page_index ( member_offset ) ; \
2020-04-30 18:57:55 +03:00
char * kaddr = page_address ( eb - > pages [ idx ] ) ; \
2020-04-29 19:07:04 +03:00
const int size = sizeof ( u # # bits ) ; \
2020-04-30 18:57:55 +03:00
const int part = PAGE_SIZE - oip ; \
u8 lebytes [ sizeof ( u # # bits ) ] ; \
2019-08-09 18:12:38 +03:00
\
2020-05-01 00:38:11 +03:00
ASSERT ( check_setget_bounds ( eb , ptr , off , size ) ) ; \
2021-06-23 16:48:53 +03:00
if ( INLINE_EXTENT_BUFFER_PAGES = = 1 | | oip + size < = PAGE_SIZE ) { \
2020-04-29 19:07:04 +03:00
put_unaligned_le # # bits ( val , kaddr + oip ) ; \
2019-08-09 18:12:38 +03:00
return ; \
} \
2020-04-30 18:57:55 +03:00
\
put_unaligned_le # # bits ( val , lebytes ) ; \
memcpy ( kaddr + oip , lebytes , part ) ; \
kaddr = page_address ( eb - > pages [ idx + 1 ] ) ; \
memcpy ( kaddr , lebytes + part , size - part ) ; \
2012-07-10 06:22:35 +04:00
}
2007-10-16 00:18:56 +04:00
2012-07-10 06:22:35 +04:00
DEFINE_BTRFS_SETGET_BITS ( 8 )
DEFINE_BTRFS_SETGET_BITS ( 16 )
DEFINE_BTRFS_SETGET_BITS ( 32 )
DEFINE_BTRFS_SETGET_BITS ( 64 )
2007-10-16 00:18:56 +04:00
2017-06-29 06:56:53 +03:00
void btrfs_node_key ( const struct extent_buffer * eb ,
2007-11-06 23:09:29 +03:00
struct btrfs_disk_key * disk_key , int nr )
{
unsigned long ptr = btrfs_node_key_ptr_offset ( nr ) ;
read_eb_member ( eb , ( struct btrfs_key_ptr * ) ptr ,
struct btrfs_key_ptr , key , disk_key ) ;
}