bootmem speedup from the IA64 tree
PATCH 1/1 bootmem speedup from the IA64 tree
This patch is from the IA64 tree, with some minor cleanups by me. David described it as:
This is a performance speed up and some minor indendation fixups.
The problem is that the bootmem code is (a) hugely slow and (b) has execution that grow quadratically with the size of the bootmap bitmap. This causes noticable slowdowns, especially on machines with (relatively) large holes in the physical memory map. Issue (b) is addressed by maintaining the "last_success" cache, so that we start the next search from the place where we last found some memory (this part of the patch could stand additional reviewing/testing). Issue (a) is addressed by using find_next_zero_bit() instead of the slow bit-by-bit testing.
mm/bootmem.c +41 -35
@@ -135,26 +135,24 @@* is not a problem. * * On low memory boxes we get it right in 100% of the cases. -*/ - -/* +* * alignment has to be a power of 2 value. +* +* NOTE: This function is _not_ reenetrant. */ -static void *__init __alloc_bootmem_core (bootmem_data_t *bdata, -unsigned long size,unsigned long align,unsigned long goal) +static void *__init +__alloc_bootmem_core (struct bootmem_data *bdata,unsigned long size, +unsigned long align,unsigned long goal) { -unsigned long i, start =0 ; +unsigned long offset, remaining_size, areasize, preferred; +unsigned long i, start =0 , incr, eidx; +static unsigned long last_success;void *ret; -unsigned long offset, remaining_size; -unsigned long areasize, preferred, incr; -unsigned long eidx = bdata->node_low_pfn - (bdata->node_boot_start >> -PAGE_SHIFT ); - -if (!size)BUG (); -if (align & (align-1 )) -BUG (); +BUG_ON (!size); +BUG_ON (align & (align-1 )); + eidx = bdata->node_low_pfn - (bdata->node_boot_start >>PAGE_SHIFT ); offset =0 ;if (align && (bdata->node_boot_start & (align -1UL )) !=0 )
@@ -166,8 +164,11 @@* first, then we try to allocate lower pages. */ if (goal && (goal >= bdata->node_boot_start) && - ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { + ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { preferred = goal - bdata->node_boot_start; + +if (last_success >= preferred) + preferred = last_success; }else preferred =0 ;
@@ -179,6 +180,8 @@ restart_scan:for (i = preferred; i < eidx; i += incr) {unsigned long j; + i =find_next_zero_bit (bdata->node_bootmem_map, eidx, i); + i = (i + incr -1 ) & -incr;if (test_bit (i, bdata->node_bootmem_map))continue ;for (j = i +1 ; j < i + areasize; ++j) {
@@ -189,31 +192,33 @@ } start = i; goto found; - fail_block:; + fail_block: + ; } +if (preferred) { preferred = offset; goto restart_scan; }return NULL ; + found: -if (start >= eidx) -BUG (); + last_success = start <<PAGE_SHIFT ; +BUG_ON (start >= eidx);/* * Is the next page of the previous allocation-end the start * of this allocation's buffer? If yes then we can 'merge' * the previous partial page with this allocation. */ -if (align <PAGE_SIZE - && bdata->last_offset && bdata->last_pos+1 == start) { +if (align <PAGE_SIZE && + bdata->last_offset && bdata->last_pos+1 == start) { offset = (bdata->last_offset+align-1 ) & ~(align-1 ); -if (offset >PAGE_SIZE ) -BUG (); +BUG_ON (offset >PAGE_SIZE ); remaining_size =PAGE_SIZE -offset;if (size < remaining_size) { areasize =0 ; -// last_pos unchanged +/* last_pos unchanged */ bdata->last_offset = offset+size; ret =phys_to_virt (bdata->last_pos*PAGE_SIZE + offset + bdata->node_boot_start);
@@ -231,11 +236,12 @@ bdata->last_offset = size & ~PAGE_MASK ; ret =phys_to_virt (start *PAGE_SIZE + bdata->node_boot_start); } +/* * Reserve the area now: */ for (i = start; i < start+areasize; i++) -if (test_and_set_bit (i, bdata->node_bootmem_map)) +if (unlikely (test_and_set_bit (i, bdata->node_bootmem_map)))BUG ();memset (ret,0 , size);return ret;
@@ -256,21 +262,21 @@ map = bdata->node_bootmem_map;for (i =0 ; i < idx; ) {unsigned long v = ~map[i /BITS_PER_LONG ]; -if (v) { +if (v) {unsigned long m; -for (m =1 ; m && i < idx; m<<=1 , page++, i++) { +for (m =1 ; m && i < idx; m<<=1 , page++, i++) {if (v & m) { - count++; -ClearPageReserved (page); -set_page_count (page,1 ); -__free_page (page); - } - } + count++; +ClearPageReserved (page); +set_page_count (page,1 ); +__free_page (page); + } + } }else { i+=BITS_PER_LONG ; - page+=BITS_PER_LONG ; - } - } + page +=BITS_PER_LONG ; + } + } total += count;/*
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 5 lines · Christoph Hellwig
This patch is from the IA64 tree, with some minor cleanups by me. David described it as:
This is a performance speed up and some minor indendation fixups.
OK, thanks - I'll queue this up for a bit of testing.
Martin, can you please also test this?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 8 lines
Quoted 4 lines
This patch is from the IA64 tree, with some minor cleanups by me. David described it as:
This is a performance speed up and some minor indendation fixups.
OK, thanks - I'll queue this up for a bit of testing.
Martin, can you please also test this?
Sure. Will do both of those together.
M.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 8 lines
Quoted 4 lines
This patch is from the IA64 tree, with some minor cleanups by me. David described it as:
This is a performance speed up and some minor indendation fixups.
OK, thanks - I'll queue this up for a bit of testing.
Martin, can you please also test this?
Compile-tested against every config I had lying around, and run tested on my big wierdo-box. Works fine.
M.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
mm/bootmem.c line 171
Quoted 7 lines · Christoph Hellwig
mm/bootmem.c +4
-1if (goal && (goal >= bdata->node_boot_start) && - ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { + ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { preferred = goal - bdata->node_boot_start; + +if (last_success >= preferred) + preferred = last_success;
I suspect you need a range check on last_success here for machines which have multiple nodes of memory, or else store it in bdata.
-ben
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 12 lines · Benjamin LaHaise
Review
Quoted 7 lines · Christoph Hellwig
(quoted diff) +4
-1if (goal && (goal >= bdata->node_boot_start) && - ((goal >> PAGE_SHIFT) < bdata->node_low_pfn)) { + ((goal >> PAGE_SHIFT) < bdata->node_low_pfn)) { preferred = goal - bdata->node_boot_start; + + if (last_success >= preferred) + preferred = last_success;I suspect you need a range check on
last_successhere for machines which have multiple nodes of memory, or else store it in bdata.
Agreed. I've updated the patch thusly.
Bootmem igornamus says:
Do we have a problem with using an `unsigned long' byte address in there on ia32 PAE? Or are we guaranteed that this will only ever be used in the lower 4G of physical memory?
Does the last_success cache ever need to be updated if someone frees some previously-allocated memory?
From: Christoph Hellwig <[email protected]>
This patch is from the IA64 tree, with some minor cleanups by me. David described it as:
This is a performance speed up and some minor indendation fixups.
The problem is that the bootmem code is (a) hugely slow and (b) has execution that grow quadratically with the size of the bootmap bitmap. This causes noticable slowdowns, especially on machines with (relatively) large holes in the physical memory map. Issue (b) is addressed by maintaining the "last_success" cache, so that we start the next search from the place where we last found some memory (this part of the patch could stand additional reviewing/testing). Issue (a) is addressed by using find_next_zero_bit() instead of the slow bit-by-bit testing.
25-akpm/include/linux/bootmem.h | 2 + 25-akpm/mm/bootmem.c | 75 +++++++++++++++++++++------------------- 2 files changed, 42 insertions(+), 35 deletions(-)
mm/bootmem.c renamed from mm/bootmem.c~bootmem-speedup +40 -35
@@ -135,26 +135,23 @@ static void __init free_bootmem_core(boo* is not a problem. * * On low memory boxes we get it right in 100% of the cases. -*/ - -/* +* * alignment has to be a power of 2 value. +* +* NOTE: This function is _not_ reenetrant. */ -static void *__init __alloc_bootmem_core (bootmem_data_t *bdata, -unsigned long size,unsigned long align,unsigned long goal) +static void *__init +__alloc_bootmem_core (struct bootmem_data *bdata,unsigned long size, +unsigned long align,unsigned long goal) { -unsigned long i, start =0 ; +unsigned long offset, remaining_size, areasize, preferred; +unsigned long i, start =0 , incr, eidx;void *ret; -unsigned long offset, remaining_size; -unsigned long areasize, preferred, incr; -unsigned long eidx = bdata->node_low_pfn - (bdata->node_boot_start >> -PAGE_SHIFT ); -if (!size)BUG (); - -if (align & (align-1 )) -BUG (); +BUG_ON (!size); +BUG_ON (align & (align-1 )); + eidx = bdata->node_low_pfn - (bdata->node_boot_start >>PAGE_SHIFT ); offset =0 ;if (align && (bdata->node_boot_start & (align -1UL )) !=0 )
@@ -166,8 +163,11 @@ static void * __init __alloc_bootmem_cor* first, then we try to allocate lower pages. */ if (goal && (goal >= bdata->node_boot_start) && - ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { + ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { preferred = goal - bdata->node_boot_start; + +if (bdata->last_success >= preferred) + preferred = bdata->last_success; }else preferred =0 ;
@@ -179,6 +179,8 @@ static void * __init __alloc_bootmem_cor restart_scan:for (i = preferred; i < eidx; i += incr) {unsigned long j; + i =find_next_zero_bit (bdata->node_bootmem_map, eidx, i); + i = (i + incr -1 ) & -incr;if (test_bit (i, bdata->node_bootmem_map))continue ;for (j = i +1 ; j < i + areasize; ++j) {
@@ -189,31 +191,33 @@ restart_scan: } start = i; goto found; - fail_block:; + fail_block: + ; } +if (preferred) { preferred = offset; goto restart_scan; }return NULL ; + found: -if (start >= eidx) -BUG (); + bdata->last_success = start <<PAGE_SHIFT ; +BUG_ON (start >= eidx);/* * Is the next page of the previous allocation-end the start * of this allocation's buffer? If yes then we can 'merge' * the previous partial page with this allocation. */ -if (align <PAGE_SIZE - && bdata->last_offset && bdata->last_pos+1 == start) { +if (align <PAGE_SIZE && + bdata->last_offset && bdata->last_pos+1 == start) { offset = (bdata->last_offset+align-1 ) & ~(align-1 ); -if (offset >PAGE_SIZE ) -BUG (); +BUG_ON (offset >PAGE_SIZE ); remaining_size =PAGE_SIZE -offset;if (size < remaining_size) { areasize =0 ; -// last_pos unchanged +/* last_pos unchanged */ bdata->last_offset = offset+size; ret =phys_to_virt (bdata->last_pos*PAGE_SIZE + offset + bdata->node_boot_start);
@@ -231,11 +235,12 @@ found: bdata->last_offset = size & ~PAGE_MASK ; ret =phys_to_virt (start *PAGE_SIZE + bdata->node_boot_start); } +/* * Reserve the area now: */ for (i = start; i < start+areasize; i++) -if (test_and_set_bit (i, bdata->node_bootmem_map)) +if (unlikely (test_and_set_bit (i, bdata->node_bootmem_map)))BUG ();memset (ret,0 , size);return ret;
@@ -256,21 +261,21 @@ static unsigned long __init free_all_boo map = bdata->node_bootmem_map;for (i =0 ; i < idx; ) {unsigned long v = ~map[i /BITS_PER_LONG ]; -if (v) { +if (v) {unsigned long m; -for (m =1 ; m && i < idx; m<<=1 , page++, i++) { +for (m =1 ; m && i < idx; m<<=1 , page++, i++) {if (v & m) { - count++; -ClearPageReserved (page); -set_page_count (page,1 ); -__free_page (page); - } - } + count++; +ClearPageReserved (page); +set_page_count (page,1 ); +__free_page (page); + } + } }else { i+=BITS_PER_LONG ; - page+=BITS_PER_LONG ; - } - } + page +=BITS_PER_LONG ; + } + } total += count;/*
include/linux/bootmem.h renamed from include/linux/bootmem.h~bootmem-speedup +2 -0
@@ -32,6 +32,8 @@ typedef struct bootmem_data {void *node_bootmem_map;unsigned long last_offset;unsigned long last_pos; +unsigned long last_success;/* Previous allocation point. To speed +* up searching */ } bootmem_data_t;extern unsigned long __init bootmem_bootmap_pages (unsigned long );
_
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 5 lines
Bootmem igornamus says:
Do we have a problem with using an `unsigned long' byte address in there on ia32 PAE? Or are we guaranteed that this will only ever be used in the lower 4G of physical memory?
IIRC, only ZONE_NORMAL goes into bootmem, so we should be OK.
M.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 9 lines · Martin J. Bligh
Quoted 6 lines
Bootmem igornamus says:
Do we have a problem with using an `unsigned long' byte address in there on ia32 PAE? Or are we guaranteed that this will only ever be used in the lower 4G of physical memory?
IIRC, only
ZONE_NORMALgoes into bootmem, so we should be OK.
OK, thanks. You didn't answer my other question ;)
I stuck
if (addr < bdata->last_success)
bdata->last_success = addr;in free_bootmem_core().
mm/bootmem.c renamed from mm/bootmem.c~bootmem-speedup +43 -35
@@ -115,6 +115,9 @@ static void __init free_bootmem_core(booif (end > bdata->node_low_pfn)BUG (); +if (addr < bdata->last_success) + bdata->last_success = addr; +/* * Round up the beginning of the address. */
@@ -135,26 +138,23 @@ static void __init free_bootmem_core(boo* is not a problem. * * On low memory boxes we get it right in 100% of the cases. -*/ - -/* +* * alignment has to be a power of 2 value. +* +* NOTE: This function is _not_ reenetrant. */ -static void *__init __alloc_bootmem_core (bootmem_data_t *bdata, -unsigned long size,unsigned long align,unsigned long goal) +static void *__init +__alloc_bootmem_core (struct bootmem_data *bdata,unsigned long size, +unsigned long align,unsigned long goal) { -unsigned long i, start =0 ; +unsigned long offset, remaining_size, areasize, preferred; +unsigned long i, start =0 , incr, eidx;void *ret; -unsigned long offset, remaining_size; -unsigned long areasize, preferred, incr; -unsigned long eidx = bdata->node_low_pfn - (bdata->node_boot_start >> -PAGE_SHIFT ); - -if (!size)BUG (); -if (align & (align-1 )) -BUG (); +BUG_ON (!size); +BUG_ON (align & (align-1 )); + eidx = bdata->node_low_pfn - (bdata->node_boot_start >>PAGE_SHIFT ); offset =0 ;if (align && (bdata->node_boot_start & (align -1UL )) !=0 )
@@ -166,8 +166,11 @@ static void * __init __alloc_bootmem_cor* first, then we try to allocate lower pages. */ if (goal && (goal >= bdata->node_boot_start) && - ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { + ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { preferred = goal - bdata->node_boot_start; + +if (bdata->last_success >= preferred) + preferred = bdata->last_success; }else preferred =0 ;
@@ -179,6 +182,8 @@ static void * __init __alloc_bootmem_cor restart_scan:for (i = preferred; i < eidx; i += incr) {unsigned long j; + i =find_next_zero_bit (bdata->node_bootmem_map, eidx, i); + i = (i + incr -1 ) & -incr;if (test_bit (i, bdata->node_bootmem_map))continue ;for (j = i +1 ; j < i + areasize; ++j) {
@@ -189,31 +194,33 @@ restart_scan: } start = i; goto found; - fail_block:; + fail_block: + ; } +if (preferred) { preferred = offset; goto restart_scan; }return NULL ; + found: -if (start >= eidx) -BUG (); + bdata->last_success = start <<PAGE_SHIFT ; +BUG_ON (start >= eidx);/* * Is the next page of the previous allocation-end the start * of this allocation's buffer? If yes then we can 'merge' * the previous partial page with this allocation. */ -if (align <PAGE_SIZE - && bdata->last_offset && bdata->last_pos+1 == start) { +if (align <PAGE_SIZE && + bdata->last_offset && bdata->last_pos+1 == start) { offset = (bdata->last_offset+align-1 ) & ~(align-1 ); -if (offset >PAGE_SIZE ) -BUG (); +BUG_ON (offset >PAGE_SIZE ); remaining_size =PAGE_SIZE -offset;if (size < remaining_size) { areasize =0 ; -// last_pos unchanged +/* last_pos unchanged */ bdata->last_offset = offset+size; ret =phys_to_virt (bdata->last_pos*PAGE_SIZE + offset + bdata->node_boot_start);
@@ -231,11 +238,12 @@ found: bdata->last_offset = size & ~PAGE_MASK ; ret =phys_to_virt (start *PAGE_SIZE + bdata->node_boot_start); } +/* * Reserve the area now: */ for (i = start; i < start+areasize; i++) -if (test_and_set_bit (i, bdata->node_bootmem_map)) +if (unlikely (test_and_set_bit (i, bdata->node_bootmem_map)))BUG ();memset (ret,0 , size);return ret;
@@ -256,21 +264,21 @@ static unsigned long __init free_all_boo map = bdata->node_bootmem_map;for (i =0 ; i < idx; ) {unsigned long v = ~map[i /BITS_PER_LONG ]; -if (v) { +if (v) {unsigned long m; -for (m =1 ; m && i < idx; m<<=1 , page++, i++) { +for (m =1 ; m && i < idx; m<<=1 , page++, i++) {if (v & m) { - count++; -ClearPageReserved (page); -set_page_count (page,1 ); -__free_page (page); - } - } + count++; +ClearPageReserved (page); +set_page_count (page,1 ); +__free_page (page); + } + } }else { i+=BITS_PER_LONG ; - page+=BITS_PER_LONG ; - } - } + page +=BITS_PER_LONG ; + } + } total += count;/*
include/linux/bootmem.h renamed from include/linux/bootmem.h~bootmem-speedup +2 -0
@@ -32,6 +32,8 @@ typedef struct bootmem_data {void *node_bootmem_map;unsigned long last_offset;unsigned long last_pos; +unsigned long last_success;/* Previous allocation point. To speed +* up searching */ } bootmem_data_t;extern unsigned long __init bootmem_bootmap_pages (unsigned long );
_
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 5 lines · Andrew Morton
Agreed. I've updated the patch thusly.
Bootmem igornamus says:
Do we have a problem with using an `unsigned long' byte address in there on ia32 PAE? Or are we guaranteed that this will only ever be used in the lower 4G of physical memory?
It's only ever used for lowmem on ia32, which is even below 1GB.
Quoted 2 lines · Andrew Morton
Does the
last_successcache ever need to be updated if someone frees some previously-allocated memory?
Setting preferred only puts a finger on where to begin a search. The search (and validity checking) are still carried out as usual. It could be suboptimal to set it to somewhere that's not as good as possible after a free, but it's only advice as to where to start a search and so doesn't affect correctness so long as it's in-bounds.
I'm just going to grab a barfbag and run now.
-- wli
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 2 lines · Andrew Morton
Does the
last_successcache ever need to be updated if someone frees some previously-allocated memory?
I've heard rumours that some IA64 trees can't boot without this "optimisation", suggesting that they use bootmem after freeing it.
Doesn't make the optimisation any less valid, though ...
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 9 lines · Rik van Riel
Quoted 2 lines · Andrew Morton
Does the
last_successcache ever need to be updated if someone frees some previously-allocated memory?I've heard rumours that some IA64 trees can't boot without this "optimisation", suggesting that they use bootmem after freeing it.
hm. Well I assume there's only one functional ia64 2.5 tree at present, and that's David.
David, could you please test this?
mm/bootmem.c renamed from mm/bootmem.c~bootmem-speedup +43 -35
@@ -115,6 +115,9 @@ static void __init free_bootmem_core(booif (end > bdata->node_low_pfn)BUG (); +if (addr < bdata->last_success) + bdata->last_success = addr; +/* * Round up the beginning of the address. */
@@ -135,26 +138,23 @@ static void __init free_bootmem_core(boo* is not a problem. * * On low memory boxes we get it right in 100% of the cases. -*/ - -/* +* * alignment has to be a power of 2 value. +* +* NOTE: This function is _not_ reenetrant. */ -static void *__init __alloc_bootmem_core (bootmem_data_t *bdata, -unsigned long size,unsigned long align,unsigned long goal) +static void *__init +__alloc_bootmem_core (struct bootmem_data *bdata,unsigned long size, +unsigned long align,unsigned long goal) { -unsigned long i, start =0 ; +unsigned long offset, remaining_size, areasize, preferred; +unsigned long i, start =0 , incr, eidx;void *ret; -unsigned long offset, remaining_size; -unsigned long areasize, preferred, incr; -unsigned long eidx = bdata->node_low_pfn - (bdata->node_boot_start >> -PAGE_SHIFT ); - -if (!size)BUG (); -if (align & (align-1 )) -BUG (); +BUG_ON (!size); +BUG_ON (align & (align-1 )); + eidx = bdata->node_low_pfn - (bdata->node_boot_start >>PAGE_SHIFT ); offset =0 ;if (align && (bdata->node_boot_start & (align -1UL )) !=0 )
@@ -166,8 +166,11 @@ static void * __init __alloc_bootmem_cor* first, then we try to allocate lower pages. */ if (goal && (goal >= bdata->node_boot_start) && - ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { + ((goal >>PAGE_SHIFT ) < bdata->node_low_pfn)) { preferred = goal - bdata->node_boot_start; + +if (bdata->last_success >= preferred) + preferred = bdata->last_success; }else preferred =0 ;
@@ -179,6 +182,8 @@ static void * __init __alloc_bootmem_cor restart_scan:for (i = preferred; i < eidx; i += incr) {unsigned long j; + i =find_next_zero_bit (bdata->node_bootmem_map, eidx, i); + i = (i + incr -1 ) & -incr;if (test_bit (i, bdata->node_bootmem_map))continue ;for (j = i +1 ; j < i + areasize; ++j) {
@@ -189,31 +194,33 @@ restart_scan: } start = i; goto found; - fail_block:; + fail_block: + ; } +if (preferred) { preferred = offset; goto restart_scan; }return NULL ; + found: -if (start >= eidx) -BUG (); + bdata->last_success = start <<PAGE_SHIFT ; +BUG_ON (start >= eidx);/* * Is the next page of the previous allocation-end the start * of this allocation's buffer? If yes then we can 'merge' * the previous partial page with this allocation. */ -if (align <PAGE_SIZE - && bdata->last_offset && bdata->last_pos+1 == start) { +if (align <PAGE_SIZE && + bdata->last_offset && bdata->last_pos+1 == start) { offset = (bdata->last_offset+align-1 ) & ~(align-1 ); -if (offset >PAGE_SIZE ) -BUG (); +BUG_ON (offset >PAGE_SIZE ); remaining_size =PAGE_SIZE -offset;if (size < remaining_size) { areasize =0 ; -// last_pos unchanged +/* last_pos unchanged */ bdata->last_offset = offset+size; ret =phys_to_virt (bdata->last_pos*PAGE_SIZE + offset + bdata->node_boot_start);
@@ -231,11 +238,12 @@ found: bdata->last_offset = size & ~PAGE_MASK ; ret =phys_to_virt (start *PAGE_SIZE + bdata->node_boot_start); } +/* * Reserve the area now: */ for (i = start; i < start+areasize; i++) -if (test_and_set_bit (i, bdata->node_bootmem_map)) +if (unlikely (test_and_set_bit (i, bdata->node_bootmem_map)))BUG ();memset (ret,0 , size);return ret;
@@ -256,21 +264,21 @@ static unsigned long __init free_all_boo map = bdata->node_bootmem_map;for (i =0 ; i < idx; ) {unsigned long v = ~map[i /BITS_PER_LONG ]; -if (v) { +if (v) {unsigned long m; -for (m =1 ; m && i < idx; m<<=1 , page++, i++) { +for (m =1 ; m && i < idx; m<<=1 , page++, i++) {if (v & m) { - count++; -ClearPageReserved (page); -set_page_count (page,1 ); -__free_page (page); - } - } + count++; +ClearPageReserved (page); +set_page_count (page,1 ); +__free_page (page); + } + } }else { i+=BITS_PER_LONG ; - page+=BITS_PER_LONG ; - } - } + page +=BITS_PER_LONG ; + } + } total += count;/*
include/linux/bootmem.h renamed from include/linux/bootmem.h~bootmem-speedup +2 -0
@@ -32,6 +32,8 @@ typedef struct bootmem_data {void *node_bootmem_map;unsigned long last_offset;unsigned long last_pos; +unsigned long last_success;/* Previous allocation point. To speed +* up searching */ } bootmem_data_t;extern unsigned long __init bootmem_bootmap_pages (unsigned long );
_
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Andrew Morton
Quoted 8 lines · Andrew> Rik van Riel <[email protected]> wrote:
Quoted 8 lines · Andrew Morton
Quoted 3 lines
Does the
last_successcache ever need to be updated if someone frees some previously-allocated memory?I've heard rumours that some IA64 trees can't boot without this "optimisation", suggesting that they use bootmem after freeing it.
Andrew> hm. Well I assume there's only one functional ia64 2.5 tree at present, and Andrew> that's David.
Andrew> David, could you please test this?
I tried the patch with the Ski simulator (simulating a 4GB hole) and it booted as fast as ever. Looks great to me.
The new code is in my tree now so it will be exposed to real hardware today and over the next couple of days. I don't anticipate any problems, but something unexpected crops up, I'll let you know.
Thanks,
--david
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Rik van Riel
Quoted 2 lines · Andrew Morton
Quoted 2 lines
Does the
last_successcache ever need to be updated if someone frees some previously-allocated memory?
Rik> I've heard rumours that some IA64 trees can't boot without Rik> this "optimisation", suggesting that they use bootmem after Rik> freeing it.
Huh? Where do you hear such rumors?
--david
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>
Quoted 10 lines
This is a performance speed up and some minor indendation fixups.
The problem is that the bootmem code is (a) hugely slow and (b) has execution that grow quadratically with the size of the bootmap bitmap. This causes noticable slowdowns, especially on machines with (relatively) large holes in the physical memory map. Issue (b) is addressed by maintaining the "
last_success" cache, so that we start the next search from the place where we last found some memory (this part of the patch could stand additional reviewing/testing). Issue (a) is addressed by usingfind_next_zero_bit()instead of the slow bit-by-bit testing.
FYI I have some ppc64 machines with a memory layout of
1GB MEM
3GB IO
63GB MEM
And see the same problem.
Anton
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]">[email protected]</a>