| Message ID | 20200910104650.1182807-2-dereksu@qnap.com |
|---|---|
| State | New |
| Headers | show |
| Series | COLO: only flush dirty ram pages from colo cache | expand |
Hello, all Ping... Regards, Derek Su Derek Su <dereksu@qnap.com> 於 2020年9月10日 週四 下午6:47寫道： > In secondary side, the colo_flush_ram_cache() calls > migration_bitmap_find_dirty() to finding the dirty pages and > flush them to host. But ram_state's ram_bulk_stage flag is always > enabled in secondary side, it leads to the whole ram pages copy > instead of only dirty pages. > > Here, the ram_bulk_stage in secondary side is disabled in the > preparation of COLO incoming process to avoid the whole dirty > ram pages flush. > > Signed-off-by: Derek Su <dereksu@qnap.com> > --- > migration/colo.c | 6 +++++- > migration/ram.c | 10 ++++++++++ > migration/ram.h | 3 +++ > 3 files changed, 18 insertions(+), 1 deletion(-) > > diff --git a/migration/colo.c b/migration/colo.c > index ea7d1e9d4e..6e644db306 100644 > --- a/migration/colo.c > +++ b/migration/colo.c > @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) > return NULL; > } > > + colo_disable_ram_bulk_stage(); > + > failover_init_state(); > > mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); > @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) > goto out; > } > #else > - abort(); > + abort(); > #endif > vm_start(); > trace_colo_vm_state_change("stop", "run"); > @@ -924,6 +926,8 @@ out: > qemu_fclose(fb); > } > > + colo_enable_ram_bulk_stage(); > + > /* Hope this not to be too long to loop here */ > qemu_sem_wait(&mis->colo_incoming_sem); > qemu_sem_destroy(&mis->colo_incoming_sem); > diff --git a/migration/ram.c b/migration/ram.c > index 76d4fee5d5..65e9b12058 100644 > --- a/migration/ram.c > +++ b/migration/ram.c > @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) > return ps >= POSTCOPY_INCOMING_LISTENING && ps < > POSTCOPY_INCOMING_END; > } > > +void colo_enable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = true; > +} > + > +void colo_disable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = false; > +} > + > /* > * Flush content of RAM cache into SVM's memory. > * Only flush the pages that be dirtied by PVM or SVM or both. > diff --git a/migration/ram.h b/migration/ram.h > index 2eeaacfa13..c1c0ebbe0f 100644 > --- a/migration/ram.h > +++ b/migration/ram.h > @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); > void colo_release_ram_cache(void); > void colo_incoming_start_dirty_log(void); > > +void colo_enable_ram_bulk_stage(void); > +void colo_disable_ram_bulk_stage(void); > + > #endif > -- > 2.25.1 > >
Hi Derek and Lei, It looks same with Lei’ patch: [PATCH 2/3] Reduce the time of checkpoint for COLO Can you discuss to merge it into one patch? Thanks Zhang Chen From: Derek Su <dereksu@qnap.com> Sent: Tuesday, September 22, 2020 11:31 AM To: qemu-devel <qemu-devel@nongnu.org> Cc: zhang.zhanghailiang@huawei.com; quintela@redhat.com; dgilbert@redhat.com; Zhang, Chen <chen.zhang@intel.com> Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo cache Hello, all Ping... Regards, Derek Su Derek Su <dereksu@qnap.com<mailto:dereksu@qnap.com>> 於 2020年9月10日 週四 下午6:47寫道： In secondary side, the colo_flush_ram_cache() calls migration_bitmap_find_dirty() to finding the dirty pages and flush them to host. But ram_state's ram_bulk_stage flag is always enabled in secondary side, it leads to the whole ram pages copy instead of only dirty pages. Here, the ram_bulk_stage in secondary side is disabled in the preparation of COLO incoming process to avoid the whole dirty ram pages flush. Signed-off-by: Derek Su <dereksu@qnap.com<mailto:dereksu@qnap.com>> --- migration/colo.c | 6 +++++- migration/ram.c | 10 ++++++++++ migration/ram.h | 3 +++ 3 files changed, 18 insertions(+), 1 deletion(-) diff --git a/migration/colo.c b/migration/colo.c index ea7d1e9d4e..6e644db306 100644 --- a/migration/colo.c +++ b/migration/colo.c @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) return NULL; } + colo_disable_ram_bulk_stage(); + failover_init_state(); mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) goto out; } #else - abort(); + abort(); #endif vm_start(); trace_colo_vm_state_change("stop", "run"); @@ -924,6 +926,8 @@ out: qemu_fclose(fb); } + colo_enable_ram_bulk_stage(); + /* Hope this not to be too long to loop here */ qemu_sem_wait(&mis->colo_incoming_sem); qemu_sem_destroy(&mis->colo_incoming_sem); diff --git a/migration/ram.c b/migration/ram.c index 76d4fee5d5..65e9b12058 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; } +void colo_enable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = true; +} + +void colo_disable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = false; +} + /* * Flush content of RAM cache into SVM's memory. * Only flush the pages that be dirtied by PVM or SVM or both. diff --git a/migration/ram.h b/migration/ram.h index 2eeaacfa13..c1c0ebbe0f 100644 --- a/migration/ram.h +++ b/migration/ram.h @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); void colo_release_ram_cache(void); void colo_incoming_start_dirty_log(void); +void colo_enable_ram_bulk_stage(void); +void colo_disable_ram_bulk_stage(void); + #endif -- 2.25.1
Hi, Chen Sure. BTW, I just went through Lei's patch. ram_bulk_stage() might need to reset to true after stopping COLO service as my patch. How about your opinion? Thanks. Best regards, Derek Zhang, Chen <chen.zhang@intel.com> 於 2020年9月22日 週二 上午11:41寫道： > Hi Derek and Lei, > > > > It looks same with Lei’ patch: > > [PATCH 2/3] Reduce the time of checkpoint for COLO > > Can you discuss to merge it into one patch? > > > > Thanks > > Zhang Chen > > > > *From:* Derek Su <dereksu@qnap.com> > *Sent:* Tuesday, September 22, 2020 11:31 AM > *To:* qemu-devel <qemu-devel@nongnu.org> > *Cc:* zhang.zhanghailiang@huawei.com; quintela@redhat.com; > dgilbert@redhat.com; Zhang, Chen <chen.zhang@intel.com> > *Subject:* Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo > cache > > > > Hello, all > > > > Ping... > > > > Regards, > > Derek Su > > > > Derek Su <dereksu@qnap.com> 於 2020年9月10日 週四 下午6:47寫道： > > In secondary side, the colo_flush_ram_cache() calls > migration_bitmap_find_dirty() to finding the dirty pages and > flush them to host. But ram_state's ram_bulk_stage flag is always > enabled in secondary side, it leads to the whole ram pages copy > instead of only dirty pages. > > Here, the ram_bulk_stage in secondary side is disabled in the > preparation of COLO incoming process to avoid the whole dirty > ram pages flush. > > Signed-off-by: Derek Su <dereksu@qnap.com> > --- > migration/colo.c | 6 +++++- > migration/ram.c | 10 ++++++++++ > migration/ram.h | 3 +++ > 3 files changed, 18 insertions(+), 1 deletion(-) > > diff --git a/migration/colo.c b/migration/colo.c > index ea7d1e9d4e..6e644db306 100644 > --- a/migration/colo.c > +++ b/migration/colo.c > @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) > return NULL; > } > > + colo_disable_ram_bulk_stage(); > + > failover_init_state(); > > mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); > @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) > goto out; > } > #else > - abort(); > + abort(); > #endif > vm_start(); > trace_colo_vm_state_change("stop", "run"); > @@ -924,6 +926,8 @@ out: > qemu_fclose(fb); > } > > + colo_enable_ram_bulk_stage(); > + > /* Hope this not to be too long to loop here */ > qemu_sem_wait(&mis->colo_incoming_sem); > qemu_sem_destroy(&mis->colo_incoming_sem); > diff --git a/migration/ram.c b/migration/ram.c > index 76d4fee5d5..65e9b12058 100644 > --- a/migration/ram.c > +++ b/migration/ram.c > @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) > return ps >= POSTCOPY_INCOMING_LISTENING && ps < > POSTCOPY_INCOMING_END; > } > > +void colo_enable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = true; > +} > + > +void colo_disable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = false; > +} > + > /* > * Flush content of RAM cache into SVM's memory. > * Only flush the pages that be dirtied by PVM or SVM or both. > diff --git a/migration/ram.h b/migration/ram.h > index 2eeaacfa13..c1c0ebbe0f 100644 > --- a/migration/ram.h > +++ b/migration/ram.h > @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); > void colo_release_ram_cache(void); > void colo_incoming_start_dirty_log(void); > > +void colo_enable_ram_bulk_stage(void); > +void colo_disable_ram_bulk_stage(void); > + > #endif > -- > 2.25.1 > >
Hi, Derek and Chen ram_bulk_stage is false by default before Hailiang's patch. For COLO, it does not seem to be used, so I think there is no need to reset it to true. Thanks, Lei. From: Derek Su <dereksu@qnap.com> Sent: Tuesday, September 22, 2020 11:48 AM To: Zhang, Chen <chen.zhang@intel.com> Cc: qemu-devel <qemu-devel@nongnu.org>; Rao, Lei <lei.rao@intel.com>; zhang.zhanghailiang@huawei.com; quintela@redhat.com; dgilbert@redhat.com Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo cache Hi, Chen Sure. BTW, I just went through Lei's patch. ram_bulk_stage() might need to reset to true after stopping COLO service as my patch. How about your opinion? Thanks. Best regards, Derek Zhang, Chen <mailto:chen.zhang@intel.com> 於 2020年9月22日 週二 上午11:41寫道： Hi Derek and Lei, It looks same with Lei’ patch: [PATCH 2/3] Reduce the time of checkpoint for COLO Can you discuss to merge it into one patch? Thanks Zhang Chen From: Derek Su <mailto:dereksu@qnap.com> Sent: Tuesday, September 22, 2020 11:31 AM To: qemu-devel <mailto:qemu-devel@nongnu.org> Cc: mailto:zhang.zhanghailiang@huawei.com; mailto:quintela@redhat.com; mailto:dgilbert@redhat.com; Zhang, Chen <mailto:chen.zhang@intel.com> Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo cache Hello, all Ping... Regards, Derek Su Derek Su <mailto:dereksu@qnap.com> 於 2020年9月10日 週四 下午6:47寫道： In secondary side, the colo_flush_ram_cache() calls migration_bitmap_find_dirty() to finding the dirty pages and flush them to host. But ram_state's ram_bulk_stage flag is always enabled in secondary side, it leads to the whole ram pages copy instead of only dirty pages. Here, the ram_bulk_stage in secondary side is disabled in the preparation of COLO incoming process to avoid the whole dirty ram pages flush. Signed-off-by: Derek Su <mailto:dereksu@qnap.com> --- migration/colo.c | 6 +++++- migration/ram.c | 10 ++++++++++ migration/ram.h | 3 +++ 3 files changed, 18 insertions(+), 1 deletion(-) diff --git a/migration/colo.c b/migration/colo.c index ea7d1e9d4e..6e644db306 100644 --- a/migration/colo.c +++ b/migration/colo.c @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) return NULL; } + colo_disable_ram_bulk_stage(); + failover_init_state(); mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) goto out; } #else - abort(); + abort(); #endif vm_start(); trace_colo_vm_state_change("stop", "run"); @@ -924,6 +926,8 @@ out: qemu_fclose(fb); } + colo_enable_ram_bulk_stage(); + /* Hope this not to be too long to loop here */ qemu_sem_wait(&mis->colo_incoming_sem); qemu_sem_destroy(&mis->colo_incoming_sem); diff --git a/migration/ram.c b/migration/ram.c index 76d4fee5d5..65e9b12058 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; } +void colo_enable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = true; +} + +void colo_disable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = false; +} + /* * Flush content of RAM cache into SVM's memory. * Only flush the pages that be dirtied by PVM or SVM or both. diff --git a/migration/ram.h b/migration/ram.h index 2eeaacfa13..c1c0ebbe0f 100644 --- a/migration/ram.h +++ b/migration/ram.h @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); void colo_release_ram_cache(void); void colo_incoming_start_dirty_log(void); +void colo_enable_ram_bulk_stage(void); +void colo_disable_ram_bulk_stage(void); + #endif
Hi, Lei Got it. Thanks. Regards, Derek Rao, Lei <lei.rao@intel.com> 於 2020年9月22日 週二 下午1:04寫道： > Hi, Derek and Chen > > ram_bulk_stage is false by default before Hailiang's patch. > For COLO, it does not seem to be used, so I think there is no need to > reset it to true. > > Thanks, > Lei. > > From: Derek Su <dereksu@qnap.com> > Sent: Tuesday, September 22, 2020 11:48 AM > To: Zhang, Chen <chen.zhang@intel.com> > Cc: qemu-devel <qemu-devel@nongnu.org>; Rao, Lei <lei.rao@intel.com>; > zhang.zhanghailiang@huawei.com; quintela@redhat.com; dgilbert@redhat.com > Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo > cache > > Hi, Chen > > Sure. > > BTW, I just went through Lei's patch. > ram_bulk_stage() might need to reset to true after stopping COLO service > as my patch. > How about your opinion? > > Thanks. > > > Best regards, > Derek > > > Zhang, Chen <mailto:chen.zhang@intel.com> 於 2020年9月22日 週二 上午11:41寫道： > Hi Derek and Lei, > > It looks same with Lei’ patch: > [PATCH 2/3] Reduce the time of checkpoint for COLO > Can you discuss to merge it into one patch? > > Thanks > Zhang Chen > > From: Derek Su <mailto:dereksu@qnap.com> > Sent: Tuesday, September 22, 2020 11:31 AM > To: qemu-devel <mailto:qemu-devel@nongnu.org> > Cc: mailto:zhang.zhanghailiang@huawei.com; mailto:quintela@redhat.com; > mailto:dgilbert@redhat.com; Zhang, Chen <mailto:chen.zhang@intel.com> > Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo > cache > > Hello, all > > Ping... > > Regards, > Derek Su > > Derek Su <mailto:dereksu@qnap.com> 於 2020年9月10日 週四 下午6:47寫道： > In secondary side, the colo_flush_ram_cache() calls > migration_bitmap_find_dirty() to finding the dirty pages and > flush them to host. But ram_state's ram_bulk_stage flag is always > enabled in secondary side, it leads to the whole ram pages copy > instead of only dirty pages. > > Here, the ram_bulk_stage in secondary side is disabled in the > preparation of COLO incoming process to avoid the whole dirty > ram pages flush. > > Signed-off-by: Derek Su <mailto:dereksu@qnap.com> > --- > migration/colo.c | 6 +++++- > migration/ram.c | 10 ++++++++++ > migration/ram.h | 3 +++ > 3 files changed, 18 insertions(+), 1 deletion(-) > > diff --git a/migration/colo.c b/migration/colo.c > index ea7d1e9d4e..6e644db306 100644 > --- a/migration/colo.c > +++ b/migration/colo.c > @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) > return NULL; > } > > + colo_disable_ram_bulk_stage(); > + > failover_init_state(); > > mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); > @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) > goto out; > } > #else > - abort(); > + abort(); > #endif > vm_start(); > trace_colo_vm_state_change("stop", "run"); > @@ -924,6 +926,8 @@ out: > qemu_fclose(fb); > } > > + colo_enable_ram_bulk_stage(); > + > /* Hope this not to be too long to loop here */ > qemu_sem_wait(&mis->colo_incoming_sem); > qemu_sem_destroy(&mis->colo_incoming_sem); > diff --git a/migration/ram.c b/migration/ram.c > index 76d4fee5d5..65e9b12058 100644 > --- a/migration/ram.c > +++ b/migration/ram.c > @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) > return ps >= POSTCOPY_INCOMING_LISTENING && ps < > POSTCOPY_INCOMING_END; > } > > +void colo_enable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = true; > +} > + > +void colo_disable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = false; > +} > + > /* > * Flush content of RAM cache into SVM's memory. > * Only flush the pages that be dirtied by PVM or SVM or both. > diff --git a/migration/ram.h b/migration/ram.h > index 2eeaacfa13..c1c0ebbe0f 100644 > --- a/migration/ram.h > +++ b/migration/ram.h > @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); > void colo_release_ram_cache(void); > void colo_incoming_start_dirty_log(void); > > +void colo_enable_ram_bulk_stage(void); > +void colo_disable_ram_bulk_stage(void); > + > #endif > -- > 2.25.1 >
So, Derek, you will send new version patch? Thanks Zhang Chen From: Derek Su <dereksu@qnap.com> Sent: Tuesday, September 22, 2020 1:18 PM To: Rao, Lei <lei.rao@intel.com> Cc: Zhang, Chen <chen.zhang@intel.com>; qemu-devel <qemu-devel@nongnu.org>; zhang.zhanghailiang@huawei.com; quintela@redhat.com; dgilbert@redhat.com Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo cache Hi, Lei Got it. Thanks. Regards, Derek Rao, Lei <lei.rao@intel.com<mailto:lei.rao@intel.com>> 於 2020年9月22日 週二 下午1:04寫道： Hi, Derek and Chen ram_bulk_stage is false by default before Hailiang's patch. For COLO, it does not seem to be used, so I think there is no need to reset it to true. Thanks, Lei. From: Derek Su <dereksu@qnap.com<mailto:dereksu@qnap.com>> Sent: Tuesday, September 22, 2020 11:48 AM To: Zhang, Chen <chen.zhang@intel.com<mailto:chen.zhang@intel.com>> Cc: qemu-devel <qemu-devel@nongnu.org<mailto:qemu-devel@nongnu.org>>; Rao, Lei <lei.rao@intel.com<mailto:lei.rao@intel.com>>; zhang.zhanghailiang@huawei.com<mailto:zhang.zhanghailiang@huawei.com>; quintela@redhat.com<mailto:quintela@redhat.com>; dgilbert@redhat.com<mailto:dgilbert@redhat.com> Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo cache Hi, Chen Sure. BTW, I just went through Lei's patch. ram_bulk_stage() might need to reset to true after stopping COLO service as my patch. How about your opinion? Thanks. Best regards, Derek Zhang, Chen <mailto:chen.zhang@intel.com<mailto:chen.zhang@intel.com>> 於 2020年9月22日 週二 上午11:41寫道： Hi Derek and Lei, It looks same with Lei’ patch: [PATCH 2/3] Reduce the time of checkpoint for COLO Can you discuss to merge it into one patch? Thanks Zhang Chen From: Derek Su <mailto:dereksu@qnap.com<mailto:dereksu@qnap.com>> Sent: Tuesday, September 22, 2020 11:31 AM To: qemu-devel <mailto:qemu-devel@nongnu.org<mailto:qemu-devel@nongnu.org>> Cc: mailto:zhang.zhanghailiang@huawei.com<mailto:zhang.zhanghailiang@huawei.com>; mailto:quintela@redhat.com<mailto:quintela@redhat.com>; mailto:dgilbert@redhat.com<mailto:dgilbert@redhat.com>; Zhang, Chen <mailto:chen.zhang@intel.com<mailto:chen.zhang@intel.com>> Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo cache Hello, all Ping... Regards, Derek Su Derek Su <mailto:dereksu@qnap.com<mailto:dereksu@qnap.com>> 於 2020年9月10日 週四 下午6:47寫道： In secondary side, the colo_flush_ram_cache() calls migration_bitmap_find_dirty() to finding the dirty pages and flush them to host. But ram_state's ram_bulk_stage flag is always enabled in secondary side, it leads to the whole ram pages copy instead of only dirty pages. Here, the ram_bulk_stage in secondary side is disabled in the preparation of COLO incoming process to avoid the whole dirty ram pages flush. Signed-off-by: Derek Su <mailto:dereksu@qnap.com<mailto:dereksu@qnap.com>> --- migration/colo.c | 6 +++++- migration/ram.c | 10 ++++++++++ migration/ram.h | 3 +++ 3 files changed, 18 insertions(+), 1 deletion(-) diff --git a/migration/colo.c b/migration/colo.c index ea7d1e9d4e..6e644db306 100644 --- a/migration/colo.c +++ b/migration/colo.c @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) return NULL; } + colo_disable_ram_bulk_stage(); + failover_init_state(); mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) goto out; } #else - abort(); + abort(); #endif vm_start(); trace_colo_vm_state_change("stop", "run"); @@ -924,6 +926,8 @@ out: qemu_fclose(fb); } + colo_enable_ram_bulk_stage(); + /* Hope this not to be too long to loop here */ qemu_sem_wait(&mis->colo_incoming_sem); qemu_sem_destroy(&mis->colo_incoming_sem); diff --git a/migration/ram.c b/migration/ram.c index 76d4fee5d5..65e9b12058 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; } +void colo_enable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = true; +} + +void colo_disable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = false; +} + /* * Flush content of RAM cache into SVM's memory. * Only flush the pages that be dirtied by PVM or SVM or both. diff --git a/migration/ram.h b/migration/ram.h index 2eeaacfa13..c1c0ebbe0f 100644 --- a/migration/ram.h +++ b/migration/ram.h @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); void colo_release_ram_cache(void); void colo_incoming_start_dirty_log(void); +void colo_enable_ram_bulk_stage(void); +void colo_disable_ram_bulk_stage(void); + #endif -- 2.25.1
Hi, Chen and Lei Using Lei's patch is OK to me. Please help to add "Signed-off-by: Derek Su <dereksu@qnap.com>" for merging it. Thank you. :) Regards Derek Zhang, Chen <chen.zhang@intel.com> 於 2020年9月22日 週二 下午1:37寫道： > So, Derek, you will send new version patch? > > > > Thanks > > Zhang Chen > > > > *From:* Derek Su <dereksu@qnap.com> > *Sent:* Tuesday, September 22, 2020 1:18 PM > *To:* Rao, Lei <lei.rao@intel.com> > *Cc:* Zhang, Chen <chen.zhang@intel.com>; qemu-devel < > qemu-devel@nongnu.org>; zhang.zhanghailiang@huawei.com; > quintela@redhat.com; dgilbert@redhat.com > *Subject:* Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo > cache > > > > Hi, Lei > > > > Got it. Thanks. > > > > Regards, > > Derek > > > > Rao, Lei <lei.rao@intel.com> 於 2020年9月22日 週二 下午1:04寫道： > > Hi, Derek and Chen > > ram_bulk_stage is false by default before Hailiang's patch. > For COLO, it does not seem to be used, so I think there is no need to > reset it to true. > > Thanks, > Lei. > > From: Derek Su <dereksu@qnap.com> > Sent: Tuesday, September 22, 2020 11:48 AM > To: Zhang, Chen <chen.zhang@intel.com> > Cc: qemu-devel <qemu-devel@nongnu.org>; Rao, Lei <lei.rao@intel.com>; > zhang.zhanghailiang@huawei.com; quintela@redhat.com; dgilbert@redhat.com > Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo > cache > > Hi, Chen > > Sure. > > BTW, I just went through Lei's patch. > ram_bulk_stage() might need to reset to true after stopping COLO service > as my patch. > How about your opinion? > > Thanks. > > > Best regards, > Derek > > > Zhang, Chen <mailto:chen.zhang@intel.com> 於 2020年9月22日 週二 上午11:41寫道： > Hi Derek and Lei, > > It looks same with Lei’ patch: > [PATCH 2/3] Reduce the time of checkpoint for COLO > Can you discuss to merge it into one patch? > > Thanks > Zhang Chen > > From: Derek Su <mailto:dereksu@qnap.com> > Sent: Tuesday, September 22, 2020 11:31 AM > To: qemu-devel <mailto:qemu-devel@nongnu.org> > Cc: mailto:zhang.zhanghailiang@huawei.com; mailto:quintela@redhat.com; > mailto:dgilbert@redhat.com; Zhang, Chen <mailto:chen.zhang@intel.com> > Subject: Re: [PATCH v1 1/1] COLO: only flush dirty ram pages from colo > cache > > Hello, all > > Ping... > > Regards, > Derek Su > > Derek Su <mailto:dereksu@qnap.com> 於 2020年9月10日 週四 下午6:47寫道： > In secondary side, the colo_flush_ram_cache() calls > migration_bitmap_find_dirty() to finding the dirty pages and > flush them to host. But ram_state's ram_bulk_stage flag is always > enabled in secondary side, it leads to the whole ram pages copy > instead of only dirty pages. > > Here, the ram_bulk_stage in secondary side is disabled in the > preparation of COLO incoming process to avoid the whole dirty > ram pages flush. > > Signed-off-by: Derek Su <mailto:dereksu@qnap.com> > --- > migration/colo.c | 6 +++++- > migration/ram.c | 10 ++++++++++ > migration/ram.h | 3 +++ > 3 files changed, 18 insertions(+), 1 deletion(-) > > diff --git a/migration/colo.c b/migration/colo.c > index ea7d1e9d4e..6e644db306 100644 > --- a/migration/colo.c > +++ b/migration/colo.c > @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) > return NULL; > } > > + colo_disable_ram_bulk_stage(); > + > failover_init_state(); > > mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); > @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) > goto out; > } > #else > - abort(); > + abort(); > #endif > vm_start(); > trace_colo_vm_state_change("stop", "run"); > @@ -924,6 +926,8 @@ out: > qemu_fclose(fb); > } > > + colo_enable_ram_bulk_stage(); > + > /* Hope this not to be too long to loop here */ > qemu_sem_wait(&mis->colo_incoming_sem); > qemu_sem_destroy(&mis->colo_incoming_sem); > diff --git a/migration/ram.c b/migration/ram.c > index 76d4fee5d5..65e9b12058 100644 > --- a/migration/ram.c > +++ b/migration/ram.c > @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) > return ps >= POSTCOPY_INCOMING_LISTENING && ps < > POSTCOPY_INCOMING_END; > } > > +void colo_enable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = true; > +} > + > +void colo_disable_ram_bulk_stage(void) > +{ > + ram_state->ram_bulk_stage = false; > +} > + > /* > * Flush content of RAM cache into SVM's memory. > * Only flush the pages that be dirtied by PVM or SVM or both. > diff --git a/migration/ram.h b/migration/ram.h > index 2eeaacfa13..c1c0ebbe0f 100644 > --- a/migration/ram.h > +++ b/migration/ram.h > @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); > void colo_release_ram_cache(void); > void colo_incoming_start_dirty_log(void); > > +void colo_enable_ram_bulk_stage(void); > +void colo_disable_ram_bulk_stage(void); > + > #endif > -- > 2.25.1 > >
diff --git a/migration/colo.c b/migration/colo.c index ea7d1e9d4e..6e644db306 100644 --- a/migration/colo.c +++ b/migration/colo.c @@ -844,6 +844,8 @@ void *colo_process_incoming_thread(void *opaque) return NULL; } + colo_disable_ram_bulk_stage(); + failover_init_state(); mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); @@ -873,7 +875,7 @@ void *colo_process_incoming_thread(void *opaque) goto out; } #else - abort(); + abort(); #endif vm_start(); trace_colo_vm_state_change("stop", "run"); @@ -924,6 +926,8 @@ out: qemu_fclose(fb); } + colo_enable_ram_bulk_stage(); + /* Hope this not to be too long to loop here */ qemu_sem_wait(&mis->colo_incoming_sem); qemu_sem_destroy(&mis->colo_incoming_sem); diff --git a/migration/ram.c b/migration/ram.c index 76d4fee5d5..65e9b12058 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -3357,6 +3357,16 @@ static bool postcopy_is_running(void) return ps >= POSTCOPY_INCOMING_LISTENING && ps < POSTCOPY_INCOMING_END; } +void colo_enable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = true; +} + +void colo_disable_ram_bulk_stage(void) +{ + ram_state->ram_bulk_stage = false; +} + /* * Flush content of RAM cache into SVM's memory. * Only flush the pages that be dirtied by PVM or SVM or both. diff --git a/migration/ram.h b/migration/ram.h index 2eeaacfa13..c1c0ebbe0f 100644 --- a/migration/ram.h +++ b/migration/ram.h @@ -69,4 +69,7 @@ void colo_flush_ram_cache(void); void colo_release_ram_cache(void); void colo_incoming_start_dirty_log(void); +void colo_enable_ram_bulk_stage(void); +void colo_disable_ram_bulk_stage(void); + #endif
In secondary side, the colo_flush_ram_cache() calls migration_bitmap_find_dirty() to finding the dirty pages and flush them to host. But ram_state's ram_bulk_stage flag is always enabled in secondary side, it leads to the whole ram pages copy instead of only dirty pages. Here, the ram_bulk_stage in secondary side is disabled in the preparation of COLO incoming process to avoid the whole dirty ram pages flush. Signed-off-by: Derek Su <dereksu@qnap.com> --- migration/colo.c | 6 +++++- migration/ram.c | 10 ++++++++++ migration/ram.h | 3 +++ 3 files changed, 18 insertions(+), 1 deletion(-)