Huge memory fixes
v3 cover Huge memory fixes
Hi
The following changes since commit 47ecbdf07ed2c37bdfd2d77137d01bb319ce13da:
libcacard: build fixes (2012-06-21 20:04:24 +0000)
are available in the git repository at:
http://repo.or.cz/r/qemu/quintela.git huge_memory-v2
for you to fetch changes up to 565b6c96250305553c21714f3c482d7d408eb4b2:
Maintain the number of dirty pages (2012-06-22 15:24:14 +0200)
v2:
- Add a define MAX_DELAY (orit suggestion)
- fix handling of dirty pages number (now it is checked that is ok)
- one there, consolidate the handling of the bitmap, limiting access to the bitmap to only accessors.
Please review, Juan.
v1:
After a long, long time, this is v2.
This are basically the changes that we have for RHEL, due to the problems that we have with big memory machines. I just rebased the patches and fixed the easy parts:
- buffered_file_limit is gone: we just use 50ms and call it a day
- I let ram_addr_t as a valid type for a counter (no, I still don't agree with Anthony on this, but it is not important).
- Print total time of migration always. Notice that I also print it when migration is completed. Luiz, could you take a look to see if I did something worng (probably).
- Moved debug printfs to tracepointns. Thanks a lot to Stefan for helping with it. Once here, I had to put the traces in the middle of trace-events file, if I put them on the end of the file, when I enable them, I got generated the previous two tracepoints, instead of the ones I just defined. Stefan is looking on that. Workaround is defining them anywhere else.
- exit from cpu_physical_memory_reset_dirty(). Anthony wanted that I created an empty stub for kvm, and maintain the code for tcg. The problem is that we can have both kvm and tcg running from the same binary. Intead of exiting in the middle of the function, I just refactored the code out. Is there an struct where I could add a new function pointer for this behaviour?
- exit if we have been too long on ram_save_live() loop. Anthony didn't like this, I will sent a version based on the migration thread in the following days. But just need something working for other people to test.
Notice that I still got "lots" of more than 50ms printf's. (Yes, there is a debugging printf there).
- Bitmap handling. Still all code to count dirty pages, will try to get something saner based on bitmap optimizations.
Comments?
Later, Juan.
Juan Quintela (8):
Add spent time for migration
Add tracepoints for savevm section start/end No need to iterate if we already are over the limit Only TCG needs TLB handling
Only calculate expected_time for stage 2 Exit loop if we have been there too long dirty bitmap: abstract its use
Maintain the number of dirty pages
arch_init.c | 42 ++++++++++++++++++++++++------------------ cpu-all.h | 1 + exec-obsolete.h | 50 ++++++++++++++++++++++++++++++-------------------- exec.c | 34 ++++++++++++++++++++++------------ hmp.c | 2 ++ migration.c | 11 +++++++++++ migration.h | 1 + qapi-schema.json | 12 +++++++++--- savevm.c | 11 +++++++++++ trace-events | 5 +++++ 10 files changed, 116 insertions(+), 53 deletions(-)
Signature · 1 line
1.7.10.2
PATCH v3 1/8 Add spent time for migration
We add time spent for migration to the output of "info migrate" command. 'total_time' means time since the start fo migration if migration is 'active', and total time of migration if migration is completed. As we are also interested in transferred ram when migration completes, adding all ram statistics
--- hmp.c | 2 ++ migration.c | 11 +++++++++++ migration.h | 1 + qapi-schema.json | 12 +++++++++--- 4 files changed, 23 insertions(+), 3 deletions(-)
hmp.c +2 -0
@@ -145,6 +145,8 @@ void hmp_info_migrate(Monitor *mon) info->ram->remaining >> 10);monitor_printf (mon,"total ram: %" PRIu64" kbytes\n" , info->ram->total >>10 ); +monitor_printf (mon,"total time: %" PRIu64" milliseconds\n" , + info->ram->total_time); }if (info->has_disk) {
migration.c +11 -0
@@ -131,6 +131,8 @@ MigrationInfo *qmp_query_migrate(Error **errp) info->ram->transferred =ram_bytes_transferred (); info->ram->remaining =ram_bytes_remaining (); info->ram->total =ram_bytes_total (); + info->ram->total_time =qemu_get_clock_ms (rt_clock) + - s->total_time;if (blk_mig_active ()) { info->has_disk = true;
@@ -143,6 +145,13 @@ MigrationInfo *qmp_query_migrate(Error **errp)case MIG_STATE_COMPLETED : info->has_status = true; info->status =g_strdup ("completed" ); + + info->has_ram = true; + info->ram =g_malloc0 (sizeof (*info->ram)); + info->ram->transferred =ram_bytes_transferred (); + info->ram->remaining =0 ; + info->ram->total =ram_bytes_total (); + info->ram->total_time = s->total_time;break ;case MIG_STATE_ERROR : info->has_status = true;
@@ -260,6 +269,7 @@ static void migrate_fd_put_ready(void *opaque) }else {migrate_fd_completed (s); } + s->total_time =qemu_get_clock_ms (rt_clock) - s->total_time;if (s->state !=MIG_STATE_COMPLETED ) {if (old_vm_running) {vm_start ();
@@ -373,6 +383,7 @@ static MigrationState *migrate_init(int blk, int inc) s->bandwidth_limit = bandwidth_limit; s->state =MIG_STATE_SETUP ; + s->total_time =qemu_get_clock_ms (rt_clock);return s; }
migration.h +1 -0
@@ -33,6 +33,7 @@ struct MigrationStatevoid *opaque;int blk;int shared; +int64_t total_time; };void process_incoming_migration (QEMUFile *f);
qapi-schema.json +9 -3
@@ -260,10 +260,15 @@ # # @total: total amount of bytes involved in the migration process # +# @total_time: tota0l amount of ms since migration started. If +# migration has ended, it returns the total migration +# time. (since 1.2) +# # Since: 0.14.0. ## { 'type': 'MigrationStats', - 'data': {'transferred': 'int', 'remaining': 'int', 'total': 'int' } } + 'data': {'transferred': 'int', 'remaining': 'int', 'total': 'int' , + 'total_time': 'int' } } ## # @MigrationInfo
@@ -275,8 +280,9 @@ # 'cancelled'. If this field is not returned, no migration process # has been initiated # -# @ram: #optional @MigrationStats containing detailed migration status, -# only returned if status is 'active' +# @ram: #optional @MigrationStats containing detailed migration +# status, only returned if status is 'active' or +# 'completed'. 'comppleted' (since 1.2) # # @disk: #optional @MigrationStats containing detailed disk migration # status, only returned if status is 'active' and it is a block
Signature · 1 line
1.7.10.2
PATCH v3 2/8 Add tracepoints for savevm section start/end
In reply to Juan QuintelaThis allows to know how long each section takes to save.
An awk script like this tells us sections that takes more that 10ms
$1 ~ /savevm_state_iterate_end/ {
/* Print savevm_section_end line when > 10ms duration */
if ($2 > 10000) {
printf("%s times_missing=%u\n", $0, times_missing++);
}
}fix ws tracepoints
--- savevm.c | 8 ++++++++ trace-events | 5 +++++ 2 files changed, 13 insertions(+)
savevm.c +8 -0
@@ -85,6 +85,7 @@#include "cpus.h" #include "memory.h" #include "qmp-commands.h" +#include "trace.h" #define SELF_ANNOUNCE_ROUNDS 5
@@ -1624,11 +1625,14 @@ int qemu_savevm_state_iterate(QEMUFile *f)if (se->save_live_state ==NULL )continue ; +trace_savevm_section_start ();/* Section type */ qemu_put_byte (f,QEMU_VM_SECTION_PART );qemu_put_be32 (f, se->section_id); ret = se->save_live_state (f,QEMU_VM_SECTION_PART , se->opaque); +trace_savevm_section_end (se->section_id); +if (ret <=0 ) {/* Do not proceed to the next vmstate before this one reported completion of the current stage. This serializes the migration
@@ -1658,11 +1662,13 @@ int qemu_savevm_state_complete(QEMUFile *f)if (se->save_live_state ==NULL )continue ; +trace_savevm_section_start ();/* Section type */ qemu_put_byte (f,QEMU_VM_SECTION_END );qemu_put_be32 (f, se->section_id); ret = se->save_live_state (f,QEMU_VM_SECTION_END , se->opaque); +trace_savevm_section_end (se->section_id);if (ret <0 ) {return ret; }
@@ -1674,6 +1680,7 @@ int qemu_savevm_state_complete(QEMUFile *f)if (se->save_state ==NULL && se->vmsd ==NULL )continue ; +trace_savevm_section_start ();/* Section type */ qemu_put_byte (f,QEMU_VM_SECTION_FULL );qemu_put_be32 (f, se->section_id);
@@ -1687,6 +1694,7 @@ int qemu_savevm_state_complete(QEMUFile *f)qemu_put_be32 (f, se->version_id);vmstate_save (f, se); +trace_savevm_section_end (se->section_id); }qemu_put_byte (f,QEMU_VM_EOF );
trace-events +5 -0
@@ -782,6 +782,11 @@ displaysurface_resize(void *display_state, void *display_surface, int width, int # vga.c ppm_save(const char *filename, void *display_surface) "%s surface=%p" +# savevm.c + +savevm_section_start(void) "" +savevm_section_end(unsigned int section_id) "section_id %u" + # hw/qxl.c disable qxl_interface_set_mm_time(int qid, uint32_t mm_time) "%d %d" disable qxl_io_write_vga(int qid, const char *mode, uint32_t addr, uint32_t val) "%d %s addr=%u val=%u"
Signature · 1 line
1.7.10.2
PATCH v3 3/8 No need to iterate if we already are over the limit
In reply to Juan QuintelaIf buffers are full, don't iterate, just exit.
--- savevm.c | 3 +++ 1 file changed, 3 insertions(+)
savevm.c +3 -0
@@ -1625,6 +1625,9 @@ int qemu_savevm_state_iterate(QEMUFile *f)if (se->save_live_state ==NULL )continue ; +if (qemu_file_rate_limit (f)) { +return 0 ; + }trace_savevm_section_start ();/* Section type */ qemu_put_byte (f,QEMU_VM_SECTION_PART );
Signature · 1 line
1.7.10.2
PATCH v3 4/8 Only TCG needs TLB handling
In reply to Juan QuintelaRefactor the code that is only needed for tcg to an static function. Call that only when tcg is enabled. We can't refactor to a dummy function in the kvm case, as qemu can be compiled at the same time with tcg and kvm.
--- exec.c | 31 +++++++++++++++++++++---------- 1 file changed, 21 insertions(+), 10 deletions(-)
exec.c +21 -10
@@ -1824,11 +1824,29 @@ void tb_flush_jmp_cache(CPUArchState *env, target_ulong addr) TB_JMP_PAGE_SIZE * sizeof(TranslationBlock *)); } +static void tlb_reset_dirty_range_all (ram_addr_t start,ram_addr_t end, +uintptr_t length) +{ +uintptr_t start1; + +/* we modify the TLB cache so that the dirty bit will be set again +when accessing the range */ + start1 = (uintptr_t )qemu_safe_ram_ptr (start); +/* Check that we don't span multiple blocks - this breaks the +address comparisons below. */ +if ((uintptr_t )qemu_safe_ram_ptr (end -1 ) - start1 + != (end -1 ) - start) { +abort (); + } +cpu_tlb_reset_dirty_all (start1, length); + +} +/* Note: start and end must be within the same ram block. */ void cpu_physical_memory_reset_dirty (ram_addr_t start,ram_addr_t end,int dirty_flags) { -uintptr_t length, start1; +uintptr_t length; start &=TARGET_PAGE_MASK ; end =TARGET_PAGE_ALIGN (end);
@@ -1838,16 +1856,9 @@ void cpu_physical_memory_reset_dirty(ram_addr_t start, ram_addr_t end,return ;cpu_physical_memory_mask_dirty_range (start, length, dirty_flags); -/* we modify the TLB cache so that the dirty bit will be set again -when accessing the range */ - start1 = (uintptr_t )qemu_safe_ram_ptr (start); -/* Check that we don't span multiple blocks - this breaks the -address comparisons below. */ -if ((uintptr_t )qemu_safe_ram_ptr (end -1 ) - start1 - != (end -1 ) - start) { -abort (); +if (tcg_enabled ()) { +tlb_reset_dirty_range_all (start, end, length); } -cpu_tlb_reset_dirty_all (start1, length); }int cpu_physical_memory_set_dirty_tracking (int enable)
Signature · 1 line
1.7.10.2
PATCH v3 5/8 Only calculate expected_time for stage 2
In reply to Juan Quintelaram_save_remaining() is an expensive operation when there is a lot of memory. So we only call the function when we need it.
--- arch_init.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-)
arch_init.c +6 -4
@@ -299,7 +299,6 @@ int ram_save_live(QEMUFile *f, int stage, void *opaque)ram_addr_t addr;uint64_t bytes_transferred_last;double bwidth =0 ; -uint64_t expected_time =0 ;int ret;if (stage <0 ) {
@@ -376,9 +375,12 @@ int ram_save_live(QEMUFile *f, int stage, void *opaque)qemu_put_be64 (f,RAM_SAVE_FLAG_EOS ); - expected_time =ram_save_remaining () *TARGET_PAGE_SIZE / bwidth; - -return (stage ==2 ) && (expected_time <=migrate_max_downtime ()); +if (stage ==2 ) { +uint64_t expected_time; + expected_time =ram_save_remaining () *TARGET_PAGE_SIZE / bwidth; +return expected_time <=migrate_max_downtime (); + } +return 0 ; } static inline void *host_from_stream_offset(QEMUFile *f,
Signature · 1 line
1.7.10.2
PATCH v3 6/8 Exit loop if we have been there too long
In reply to Juan QuintelaChecking each 64 pages is a random magic number as good as any other. We don't want to test too many times, but on the other hand, qemu_get_clock_ns() is not so expensive either. We want to be sure that we spent less than 50ms (half of buffered_file timer), if we spent more than 100ms, all the accounting got wrong.
--- arch_init.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)
arch_init.c +17 -0
@@ -294,12 +294,15 @@ static void sort_ram_list(void)g_free (blocks); } +#define MAX_WAIT 50/* ms, half buffered_file limit */ +int ram_save_live (QEMUFile *f,int stage,void *opaque) {ram_addr_t addr;uint64_t bytes_transferred_last;double bwidth =0 ;int ret; +int i;if (stage <0 ) {memory_global_dirty_log_stop ();
@@ -339,6 +342,7 @@ int ram_save_live(QEMUFile *f, int stage, void *opaque) bytes_transferred_last = bytes_transferred; bwidth =qemu_get_clock_ns (rt_clock); + i =0 ;while ((ret =qemu_file_rate_limit (f)) ==0 ) {int bytes_sent;
@@ -347,6 +351,19 @@ int ram_save_live(QEMUFile *f, int stage, void *opaque)if (bytes_sent ==0 ) {/* no more blocks */ break ; } +/* we want to check in the 1st loop, just in case it was the 1st time +and we had to sync the dirty bitmap. +qemu_get_clock_ns() is a bit expensive, so we only check each some +iterations +*/ +if ((i &63 ) ==0 ) { +uint64_t t1 = (qemu_get_clock_ns (rt_clock) - bwidth) /1000000 ; +if (t1 >MAX_WAIT ) { +DPRINTF ("big wait: %ld milliseconds, %d iterations\n" , t1, i); +break ; + } + } + i++; }if (ret <0 ) {
Signature · 1 line
1.7.10.2
PATCH v3 7/8 dirty bitmap: abstract its use
In reply to Juan QuintelaAlways use accessors to read/set the dirty bitmap.
--- exec-obsolete.h | 40 ++++++++++++++++++++-------------------- exec.c | 3 +-- 2 files changed, 21 insertions(+), 22 deletions(-)
exec-obsolete.h +20 -20
@@ -45,15 +45,15 @@ int cpu_physical_memory_set_dirty_tracking(int enable);#define CODE_DIRTY_FLAG 0x02#define MIGRATION_DIRTY_FLAG 0x08 -/* read dirty bit (return 0 or 1) */ -static inline int cpu_physical_memory_is_dirty (ram_addr_t addr) +static inline int cpu_physical_memory_get_dirty_flags (ram_addr_t addr) { -return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ] ==0xff ; +return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ]; } -static inline int cpu_physical_memory_get_dirty_flags (ram_addr_t addr) +/* read dirty bit (return 0 or 1) */ +static inline int cpu_physical_memory_is_dirty (ram_addr_t addr) { -return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ]; +return cpu_physical_memory_get_dirty_flags (addr) ==0xff ; } static inline int cpu_physical_memory_get_dirty(ram_addr_t start,
@@ -61,41 +61,45 @@ static inline int cpu_physical_memory_get_dirty(ram_addr_t start, int dirty_flags) {int ret =0 ; -uint8_t *p;ram_addr_t addr, end; end =TARGET_PAGE_ALIGN (start + length); start &=TARGET_PAGE_MASK ; - p = ram_list.phys_dirty + (start >>TARGET_PAGE_BITS );for (addr = start; addr < end; addr +=TARGET_PAGE_SIZE ) { - ret |= *p++ & dirty_flags; + ret |=cpu_physical_memory_get_dirty_flags (addr) & dirty_flags; }return ret; } +static inline int cpu_physical_memory_set_dirty_flags (ram_addr_t addr, +int dirty_flags) +{ +return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ] |= dirty_flags; +} +static inline void cpu_physical_memory_set_dirty (ram_addr_t addr) { - ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ] =0xff ; +cpu_physical_memory_set_dirty_flags (addr,0xff ); } -static inline int cpu_physical_memory_set_dirty_flags (ram_addr_t addr, -int dirty_flags) +static inline int cpu_physical_memory_clear_dirty_flags (ram_addr_t addr, +int dirty_flags) { -return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ] |= dirty_flags; +int mask = ~dirty_flags; + +return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ] &= mask; }static inline void cpu_physical_memory_set_dirty_range (ram_addr_t start,ram_addr_t length,int dirty_flags) { -uint8_t *p;ram_addr_t addr, end; end =TARGET_PAGE_ALIGN (start + length); start &=TARGET_PAGE_MASK ; - p = ram_list.phys_dirty + (start >>TARGET_PAGE_BITS );for (addr = start; addr < end; addr +=TARGET_PAGE_SIZE ) { - *p++ |= dirty_flags; +cpu_physical_memory_set_dirty_flags (addr, dirty_flags); } }
@@ -103,16 +107,12 @@ static inline void cpu_physical_memory_mask_dirty_range(ram_addr_t start, ram_addr_t length, int dirty_flags) { - int mask; -uint8_t *p;ram_addr_t addr, end; end =TARGET_PAGE_ALIGN (start + length); start &=TARGET_PAGE_MASK ; - mask = ~dirty_flags; - p = ram_list.phys_dirty + (start >>TARGET_PAGE_BITS );for (addr = start; addr < end; addr +=TARGET_PAGE_SIZE ) { - *p++ &= mask; +cpu_physical_memory_clear_dirty_flags (addr, dirty_flags); } }
exec.c +1 -2
@@ -2565,8 +2565,7 @@ ram_addr_t qemu_ram_alloc_from_ptr(ram_addr_t size, void *host, ram_list.phys_dirty =g_realloc (ram_list.phys_dirty,last_ram_offset () >>TARGET_PAGE_BITS ); -memset (ram_list.phys_dirty + (new_block->offset >>TARGET_PAGE_BITS ), -0xff , size >>TARGET_PAGE_BITS ); +cpu_physical_memory_set_dirty_range (new_block->offset, size,0xff );if (kvm_enabled ())kvm_setup_guest_memory (new_block->host, size);
Signature · 1 line
1.7.10.2
PATCH v3 8/8 Maintain the number of dirty pages
In reply to Juan QuintelaCalculate the number of dirty pages takes a lot on hosts with lots of memory. Just maintain how many pages are dirty.
--- arch_init.c | 15 +-------------- cpu-all.h | 1 + exec-obsolete.h | 10 ++++++++++ 3 files changed, 12 insertions(+), 14 deletions(-)
arch_init.c +1 -14
@@ -228,20 +228,7 @@ static uint64_t bytes_transferred;static ram_addr_t ram_save_remaining (void ) { -RAMBlock *block; -ram_addr_t count =0 ; - - QLIST_FOREACH(block, &ram_list.blocks, next) { -ram_addr_t addr; -for (addr =0 ; addr < block->length; addr +=TARGET_PAGE_SIZE ) { -if (memory_region_get_dirty (block->mr, addr,TARGET_PAGE_SIZE , -DIRTY_MEMORY_MIGRATION )) { - count++; - } - } - } - -return count; +return ram_list.dirty_pages; }uint64_t ram_bytes_remaining (void )
cpu-all.h +1 -0
@@ -477,6 +477,7 @@ typedef struct RAMBlock {typedef struct RAMList {uint8_t *phys_dirty; QLIST_HEAD(, RAMBlock) blocks; +uint64_t dirty_pages; }RAMList ;extern RAMList ram_list;
exec-obsolete.h +10 -0
@@ -74,6 +74,11 @@ static inline int cpu_physical_memory_get_dirty(ram_addr_t start,static inline int cpu_physical_memory_set_dirty_flags (ram_addr_t addr,int dirty_flags) { +if ((dirty_flags &MIGRATION_DIRTY_FLAG ) && + !cpu_physical_memory_get_dirty (addr,TARGET_PAGE_SIZE , +MIGRATION_DIRTY_FLAG )) { + ram_list.dirty_pages++; + }return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ] |= dirty_flags; }
@@ -87,6 +92,11 @@ static inline int cpu_physical_memory_clear_dirty_flags(ram_addr_t addr, {int mask = ~dirty_flags; +if ((dirty_flags &MIGRATION_DIRTY_FLAG ) && +cpu_physical_memory_get_dirty (addr,TARGET_PAGE_SIZE , +MIGRATION_DIRTY_FLAG )) { + ram_list.dirty_pages--; + }return ram_list.phys_dirty[addr >>TARGET_PAGE_BITS ] &= mask; }
Signature · 1 line
1.7.10.2