Skip to contentSkip to conversation

migration: Postcopy Preempt-Full

PATCH Superseded Peter Xu 2 participants 44 messages Sep 20, 2022 – Oct 6, 2022

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Peter Xu

author
#0 lore raw

cover migration: Postcopy Preempt-Full

Based-on: <[email protected]>
[PATCH 0/5] migration: Bug fixes (prepare for preempt-full)

Tree is here:
https://github.com/xzpeter/qemu/tree/preempt-full

RFC:
https://lore.kernel.org/qemu-devel/[email protected]

This patchset is the v1 formal version of preempt-full series. The RFC tag is removed as more tests was done, and all items mentioned in the TODO items previously in the RFC cover letters are implemented in this patchset.

A few patches are added. Most of the patches are still the same as the RFC ones but may have some trivial touch ups here and there. E.g., comment touch-ups as Dave used to suggest on bitmap_mutex. If to have a look on the diff stat we'll see mostly "+" is the same as "-" this time though, because with the rp-return thread change we can logically drop a lot of complicated preempt logic previously maintained in migration thread:

  3 files changed, 371 insertions(+), 399 deletions(-)

Feel free to have a look at patch "migration: Remove old preempt code around state maintainance" where we dropped a lot of old code on preempt state maintainance of migration thread (but the major part of the preempt old code still needs to be there, e.g. channel managements) along with the break-huge parameter (as we never need to break-huge anymore.. because we already run in parallel).

Comparing to the recently merged preempt mode I called it "preempt-full" because it threadifies the postcopy channels so now urgent pages can be fully handled separately outside of the ram save loop.

The existing preempt code has reduced ramdom page req latency over 10Gbps network from ~12ms to ~500us which has already landed.

This preempt-full series can further reduces that ~500us to ~230us per my initial test. More to share below.

Note that no new capability is needed, IOW it's fully compatible with the existing preempt mode. So the naming is actually not important but just to identify the difference on the binaries.

The logic of the series is simple: send urgent pages in rp-return thread rather than migration thread. It also means rp-thread will take over the ownership of the newly created preempt channel. It can slow down rp-return thread on receiving page requests, but I don't really see a major issue with it so far but only benefits.

For detailed performance numbers, please refer to the rfc cover letter.

Please have a look, thanks.

Peter Xu (14):
migration: Add postcopy_preempt_active()
migration: Cleanup xbzrle zero page cache update logic
migration: Trivial cleanup save_page_header() on same block check
migration: Remove RAMState.f references in compression code
migration: Yield bitmap_mutex properly when sending/sleeping
migration: Use atomic ops properly for page accountings
migration: Teach PSS about host page
migration: Introduce pss_channel
migration: Add pss_init()
migration: Make PageSearchStatus part of RAMState
migration: Move last_sent_block into PageSearchStatus
migration: Send requested page directly in rp-return thread
migration: Remove old preempt code around state maintainance
migration: Drop rs->f

 migration/migration.c |  47 +--
 migration/multifd.c   |   2 +-
 migration/ram.c       | 721 ++++++++++++++++++++----------------------
 3 files changed, 371 insertions(+), 399 deletions(-)
Signature · 1 line
2.32.0
author
#1 lore raw

PATCH 1/14 migration: Add postcopy_preempt_active()

Add the helper to show that postcopy preempt enabled, meanwhile active.

---
 migration/ram.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
migration/ram.c +7 -2
index 1d42414ecc..d8cf7cc901 100644
@@ -162,6 +162,11 @@ out:
     return ret;
 }
 
+static bool postcopy_preempt_active(void)
+{
+    return migrate_postcopy_preempt() && migration_in_postcopy();
+}
+
 bool ramblock_is_ignored(RAMBlock *block)
 {
     return !qemu_ram_is_migratable(block) ||
@@ -2434,7 +2439,7 @@ static void postcopy_preempt_choose_channel(RAMState *rs, PageSearchStatus *pss)
 /* We need to make sure rs->f always points to the default channel elsewhere */
 static void postcopy_preempt_reset_channel(RAMState *rs)
 {
-    if (migrate_postcopy_preempt() && migration_in_postcopy()) {
+    if (postcopy_preempt_active()) {
         rs->postcopy_channel = RAM_CHANNEL_PRECOPY;
         rs->f = migrate_get_current()->to_dst_file;
         trace_postcopy_preempt_reset_channel();
@@ -2472,7 +2477,7 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
         return 0;
     }
 
-    if (migrate_postcopy_preempt() && migration_in_postcopy()) {
+    if (postcopy_preempt_active()) {
         postcopy_preempt_choose_channel(rs, pss);
     }
 
Signature · 1 line
2.32.0
author
#2 lore raw

PATCH 2/14 migration: Cleanup xbzrle zero page cache update logic

In reply to Peter Xu

The major change is to replace "!save_page_use_compression()" with "xbzrle_enabled" to make it clear.

Reasonings:

(1) When compression enabled, "!save_page_use_compression()" is exactly the same as checking "xbzrle_enabled".

(2) When compression disabled, "!save_page_use_compression()" always return true. We used to try calling the xbzrle code, but after this change we won't, and we shouldn't need to.

Since at it, drop the xbzrle_enabled check in xbzrle_cache_zero_page() because with this change it's not needed anymore.

---
 migration/ram.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
migration/ram.c +1 -5
index d8cf7cc901..fc59c052cf 100644
@@ -741,10 +741,6 @@ void mig_throttle_counter_reset(void)
  */
 static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr)
 {
-    if (!rs->xbzrle_enabled) {
-        return;
-    }
-
     /* We don't care if this fails to allocate a new cache page
      * as long as it updated an old one */
     cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page,
@@ -2301,7 +2297,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
         /* Must let xbzrle know, otherwise a previous (now 0'd) cached
          * page would be stale
          */
-        if (!save_page_use_compression(rs)) {
+        if (rs->xbzrle_enabled) {
             XBZRLE_cache_lock();
             xbzrle_cache_zero_page(rs, block->offset + offset);
             XBZRLE_cache_unlock();
Signature · 1 line
2.32.0
Reviewed-by on v1 2/14
#3 lore raw

Re: [PATCH 02/14] migration: Cleanup xbzrle zero page cache update logic

Quoted 8 lines · * Peter Xu ([email protected]) wrote:

The major change is to replace "!save_page_use_compression()" with "xbzrle_enabled" to make it clear.

Reasonings:

(1) When compression enabled, "!save_page_use_compression()" is exactly the same as checking "xbzrle_enabled".

Quoted 8 lines

(2) When compression disabled, "!save_page_use_compression()" always return true. We used to try calling the xbzrle code, but after this change we won't, and we shouldn't need to.

Since at it, drop the xbzrle_enabled check in xbzrle_cache_zero_page() because with this change it's not needed anymore.

Yes, I think that's right - it took me a bit of thinking to check it. The important thing is that once xbzrle is enaled then we must always take in the zero pages to squash old data in the cache.

Quoted 31 lines
---
 migration/ram.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
migration/ram.c +1 -5
index d8cf7cc901..fc59c052cf 100644
@@ -741,10 +741,6 @@ void mig_throttle_counter_reset(void)
  */
 static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr)
 {
-    if (!rs->xbzrle_enabled) {
-        return;
-    }
-
     /* We don't care if this fails to allocate a new cache page
      * as long as it updated an old one */
     cache_insert(XBZRLE.cache, current_addr, XBZRLE.zero_target_page,
@@ -2301,7 +2297,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
         /* Must let xbzrle know, otherwise a previous (now 0'd) cached
          * page would be stale
          */
-        if (!save_page_use_compression(rs)) {
+        if (rs->xbzrle_enabled) {
             XBZRLE_cache_lock();
             xbzrle_cache_zero_page(rs, block->offset + offset);
             XBZRLE_cache_unlock();
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#4 lore raw

PATCH 3/14 migration: Trivial cleanup save_page_header() on same block check

In reply to Peter Xu

The 2nd check on RAM_SAVE_FLAG_CONTINUE is a bit redundant. Use a boolean to be clearer.

---
 migration/ram.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
migration/ram.c +3 -2
index fc59c052cf..62ff2c1469 100644
@@ -661,14 +661,15 @@ static size_t save_page_header(RAMState *rs, QEMUFile *f,  RAMBlock *block,
                                ram_addr_t offset)
 {
     size_t size, len;
+    bool same_block = (block == rs->last_sent_block);
 
-    if (block == rs->last_sent_block) {
+    if (same_block) {
         offset |= RAM_SAVE_FLAG_CONTINUE;
     }
     qemu_put_be64(f, offset);
     size = 8;
 
-    if (!(offset & RAM_SAVE_FLAG_CONTINUE)) {
+    if (!same_block) {
         len = strlen(block->idstr);
         qemu_put_byte(f, len);
         qemu_put_buffer(f, (uint8_t *)block->idstr, len);
Signature · 1 line
2.32.0
Reviewed-by on v1 3/14
#5 lore raw

Re: [PATCH 03/14] migration: Trivial cleanup save_page_header() on same block check

Quoted 4 lines · * Peter Xu ([email protected]) wrote:

The 2nd check on RAM_SAVE_FLAG_CONTINUE is a bit redundant. Use a boolean to be clearer.

Quoted 29 lines
---
 migration/ram.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
migration/ram.c +3 -2
index fc59c052cf..62ff2c1469 100644
@@ -661,14 +661,15 @@ static size_t save_page_header(RAMState *rs, QEMUFile *f,  RAMBlock *block,
                                ram_addr_t offset)
 {
     size_t size, len;
+    bool same_block = (block == rs->last_sent_block);
 
-    if (block == rs->last_sent_block) {
+    if (same_block) {
         offset |= RAM_SAVE_FLAG_CONTINUE;
     }
     qemu_put_be64(f, offset);
     size = 8;
 
-    if (!(offset & RAM_SAVE_FLAG_CONTINUE)) {
+    if (!same_block) {
         len = strlen(block->idstr);
         qemu_put_byte(f, len);
         qemu_put_buffer(f, (uint8_t *)block->idstr, len);
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#6 lore raw

PATCH 4/14 migration: Remove RAMState.f references in compression code

In reply to Peter Xu

Removing referencing to RAMState.f in compress_page_with_multi_thread() and flush_compressed_data().

Compression code by default isn't compatible with having >1 channels (or it won't currently know which channel to flush the compressed data), so to make it simple we always flush on the default to_dst_file port until someone wants to add >1 ports support, as rs->f right now can really change (after postcopy preempt is introduced).

There should be no functional change at all after patch applied, since as long as rs->f referenced in compression code, it must be to_dst_file.

---
 migration/ram.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)
migration/ram.c +7 -5
index 62ff2c1469..8303252b6d 100644
@@ -1461,6 +1461,7 @@ static bool save_page_use_compression(RAMState *rs);
 
 static void flush_compressed_data(RAMState *rs)
 {
+    MigrationState *ms = migrate_get_current();
     int idx, len, thread_count;
 
     if (!save_page_use_compression(rs)) {
@@ -1479,7 +1480,7 @@ static void flush_compressed_data(RAMState *rs)
     for (idx = 0; idx < thread_count; idx++) {
         qemu_mutex_lock(&comp_param[idx].mutex);
         if (!comp_param[idx].quit) {
-            len = qemu_put_qemu_file(rs->f, comp_param[idx].file);
+            len = qemu_put_qemu_file(ms->to_dst_file, comp_param[idx].file);
             /*
              * it's safe to fetch zero_page without holding comp_done_lock
              * as there is no further request submitted to the thread,
@@ -1498,11 +1499,11 @@ static inline void set_compress_params(CompressParam *param, RAMBlock *block,
     param->offset = offset;
 }
 
-static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block,
-                                           ram_addr_t offset)
+static int compress_page_with_multi_thread(RAMBlock *block, ram_addr_t offset)
 {
     int idx, thread_count, bytes_xmit = -1, pages = -1;
     bool wait = migrate_compress_wait_thread();
+    MigrationState *ms = migrate_get_current();
 
     thread_count = migrate_compress_threads();
     qemu_mutex_lock(&comp_done_lock);
@@ -1510,7 +1511,8 @@ retry:
     for (idx = 0; idx < thread_count; idx++) {
         if (comp_param[idx].done) {
             comp_param[idx].done = false;
-            bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file);
+            bytes_xmit = qemu_put_qemu_file(ms->to_dst_file,
+                                            comp_param[idx].file);
             qemu_mutex_lock(&comp_param[idx].mutex);
             set_compress_params(&comp_param[idx], block, offset);
             qemu_cond_signal(&comp_param[idx].cond);
@@ -2263,7 +2265,7 @@ static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
         return false;
     }
 
-    if (compress_page_with_multi_thread(rs, block, offset) > 0) {
+    if (compress_page_with_multi_thread(block, offset) > 0) {
         return true;
     }
 
Signature · 1 line
2.32.0
Reviewed-by on v1 4/14
#7 lore raw

Re: [PATCH 04/14] migration: Remove RAMState.f references in compression code

Quoted 9 lines · * Peter Xu ([email protected]) wrote:

Removing referencing to RAMState.f in compress_page_with_multi_thread() and flush_compressed_data().

Compression code by default isn't compatible with having >1 channels (or it won't currently know which channel to flush the compressed data), so to make it simple we always flush on the default to_dst_file port until someone wants to add >1 ports support, as rs->f right now can really change (after postcopy preempt is introduced).

Quoted 4 lines

There should be no functional change at all after patch applied, since as long as rs->f referenced in compression code, it must be to_dst_file.

Yes, I think that's true - although I think we need to add checks to stop someone trying to enable compression+multifd?

Quoted 61 lines
---
 migration/ram.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)
migration/ram.c +7 -5
index 62ff2c1469..8303252b6d 100644
@@ -1461,6 +1461,7 @@ static bool save_page_use_compression(RAMState *rs);
 
 static void flush_compressed_data(RAMState *rs)
 {
+    MigrationState *ms = migrate_get_current();
     int idx, len, thread_count;
 
     if (!save_page_use_compression(rs)) {
@@ -1479,7 +1480,7 @@ static void flush_compressed_data(RAMState *rs)
     for (idx = 0; idx < thread_count; idx++) {
         qemu_mutex_lock(&comp_param[idx].mutex);
         if (!comp_param[idx].quit) {
-            len = qemu_put_qemu_file(rs->f, comp_param[idx].file);
+            len = qemu_put_qemu_file(ms->to_dst_file, comp_param[idx].file);
             /*
              * it's safe to fetch zero_page without holding comp_done_lock
              * as there is no further request submitted to the thread,
@@ -1498,11 +1499,11 @@ static inline void set_compress_params(CompressParam *param, RAMBlock *block,
     param->offset = offset;
 }
 
-static int compress_page_with_multi_thread(RAMState *rs, RAMBlock *block,
-                                           ram_addr_t offset)
+static int compress_page_with_multi_thread(RAMBlock *block, ram_addr_t offset)
 {
     int idx, thread_count, bytes_xmit = -1, pages = -1;
     bool wait = migrate_compress_wait_thread();
+    MigrationState *ms = migrate_get_current();
 
     thread_count = migrate_compress_threads();
     qemu_mutex_lock(&comp_done_lock);
@@ -1510,7 +1511,8 @@ retry:
     for (idx = 0; idx < thread_count; idx++) {
         if (comp_param[idx].done) {
             comp_param[idx].done = false;
-            bytes_xmit = qemu_put_qemu_file(rs->f, comp_param[idx].file);
+            bytes_xmit = qemu_put_qemu_file(ms->to_dst_file,
+                                            comp_param[idx].file);
             qemu_mutex_lock(&comp_param[idx].mutex);
             set_compress_params(&comp_param[idx], block, offset);
             qemu_cond_signal(&comp_param[idx].cond);
@@ -2263,7 +2265,7 @@ static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
         return false;
     }
 
-    if (compress_page_with_multi_thread(rs, block, offset) > 0) {
+    if (compress_page_with_multi_thread(block, offset) > 0) {
         return true;
     }
 
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#8 lore raw

Re: [PATCH 04/14] migration: Remove RAMState.f references in compression code

Quoted 10 lines · Dr. David Alan Gilbert
Quoted 9 lines · * Peter Xu ([email protected]) wrote:

Removing referencing to RAMState.f in compress_page_with_multi_thread() and flush_compressed_data().

Compression code by default isn't compatible with having >1 channels (or it won't currently know which channel to flush the compressed data), so to make it simple we always flush on the default to_dst_file port until someone wants to add >1 ports support, as rs->f right now can really change (after postcopy preempt is introduced).

Quoted 7 lines
Quoted 4 lines

There should be no functional change at all after patch applied, since as long as rs->f referenced in compression code, it must be to_dst_file.

Yes, I think that's true - although I think we need to add checks to stop someone trying to enable compression+multifd?

We could. I think they can be enabled and migration will just ignore the multifd feature as ram_save_target_page() handles compress earlier. Even if save_compress_page() fails with -EBUSY we'll still skip multifd:

    if (!save_page_use_compression(rs) && migrate_use_multifd()
        && !migration_in_postcopy()) {
        return ram_save_multifd_page(rs, block, offset);
    }

Explicitly disable compression would be still cleaner, then we can even drop above check on save_page_use_compression(). Slight risk of breaking someone's config, but I don't think it should be majority.

If that looks good, I can add one patch for it (probably in the other patchset, though, which I'll repost today).

Quoted 2 lines

Thanks!

Signature · 1 line
Peter Xu
author
#9 lore raw

PATCH 5/14 migration: Yield bitmap_mutex properly when sending/sleeping

In reply to Peter Xu

Don't take the bitmap mutex when sending pages, or when being throttled by migration_rate_limit() (which is a bit tricky to call it here in ram code, but seems still helpful).

It prepares for the possibility of concurrently sending pages in >1 threads using the function ram_save_host_page() because all threads may need the bitmap_mutex to operate on bitmaps, so that either sendmsg() or any kind of qemu_sem_wait() blocking for one thread will not block the other from progressing.

---
 migration/ram.c | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)
migration/ram.c +31 -11
index 8303252b6d..6e7de6087a 100644
@@ -2463,6 +2463,7 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
  */
 static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
 {
+    bool page_dirty, release_lock = postcopy_preempt_active();
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
@@ -2486,22 +2487,41 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
             break;
         }
 
+        page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page);
+        /*
+         * Properly yield the lock only in postcopy preempt mode because
+         * both migration thread and rp-return thread can operate on the
+         * bitmaps.
+         */
+        if (release_lock) {
+            qemu_mutex_unlock(&rs->bitmap_mutex);
+        }
+
         /* Check the pages is dirty and if it is send it */
-        if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) {
+        if (page_dirty) {
             tmppages = ram_save_target_page(rs, pss);
-            if (tmppages < 0) {
-                return tmppages;
+            if (tmppages >= 0) {
+                pages += tmppages;
+                /*
+                 * Allow rate limiting to happen in the middle of huge pages if
+                 * something is sent in the current iteration.
+                 */
+                if (pagesize_bits > 1 && tmppages > 0) {
+                    migration_rate_limit();
+                }
             }
+        } else {
+            tmppages = 0;
+        }
 
-            pages += tmppages;
-            /*
-             * Allow rate limiting to happen in the middle of huge pages if
-             * something is sent in the current iteration.
-             */
-            if (pagesize_bits > 1 && tmppages > 0) {
-                migration_rate_limit();
-            }
+        if (release_lock) {
+            qemu_mutex_lock(&rs->bitmap_mutex);
         }
+
+        if (tmppages < 0) {
+            return tmppages;
+        }
+
         pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page);
     } while ((pss->page < hostpage_boundary) &&
              offset_in_ramblock(pss->block,
Signature · 1 line
2.32.0
#10 lore raw

Re: [PATCH 05/14] migration: Yield bitmap_mutex properly when sending/sleeping

Quoted 11 lines · * Peter Xu ([email protected]) wrote:

Don't take the bitmap mutex when sending pages, or when being throttled by migration_rate_limit() (which is a bit tricky to call it here in ram code, but seems still helpful).

It prepares for the possibility of concurrently sending pages in >1 threads using the function ram_save_host_page() because all threads may need the bitmap_mutex to operate on bitmaps, so that either sendmsg() or any kind of qemu_sem_wait() blocking for one thread will not block the other from progressing.

I generally dont like taking locks conditionally; but this kind of looks OK; I think it needs a big comment on the start of the function saying that it's called and left with the lock held but that it might drop it temporarily.

migration/ram.c static void postcopy_preempt_reset_channel(RAMState *rs) · line 2466

Quoted 13 lines
---
 migration/ram.c | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)
migration/ram.c +1 -0
index 8303252b6d..6e7de6087a 100644
@@ -2463,6 +2463,7 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
  */
 static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
 {
+    bool page_dirty, release_lock = postcopy_preempt_active();

Could you rename that to something like 'drop_lock' - you are taking the lock at the end even when you have 'release_lock' set - which is a bit strange naming.

Quoted 6 lines
migration/ram.c +0 -0
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
@@ -2486,22 +2487,41 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
             break;
         }

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2498

Quoted 10 lines
migration/ram.c +9 -0
 
+        page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page);
+        /*
+         * Properly yield the lock only in postcopy preempt mode because
+         * both migration thread and rp-return thread can operate on the
+         * bitmaps.
+         */
+        if (release_lock) {
+            qemu_mutex_unlock(&rs->bitmap_mutex);
+        }

Shouldn't the unlock/lock move inside the 'if (page_dirty) {' ?

Quoted 4 lines
migration/ram.c +1 -1
         /* Check the pages is dirty and if it is send it */
-        if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) {
+        if (page_dirty) {
             tmppages = ram_save_target_page(rs, pss);

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2510

Quoted 10 lines
migration/ram.c +8 -2
-            if (tmppages < 0) {
-                return tmppages;
+            if (tmppages >= 0) {
+                pages += tmppages;
+                /*
+                 * Allow rate limiting to happen in the middle of huge pages if
+                 * something is sent in the current iteration.
+                 */
+                if (pagesize_bits > 1 && tmppages > 0) {
+                    migration_rate_limit();

This feels interesting, I know it's no change from before, and it's difficult to do here, but it seems odd to hold the lock around the sleeping in the rate limit.

Dave

Quoted 28 lines
migration/ram.c +11 -8
+                }
             }
+        } else {
+            tmppages = 0;
+        }
 
-            pages += tmppages;
-            /*
-             * Allow rate limiting to happen in the middle of huge pages if
-             * something is sent in the current iteration.
-             */
-            if (pagesize_bits > 1 && tmppages > 0) {
-                migration_rate_limit();
-            }
+        if (release_lock) {
+            qemu_mutex_lock(&rs->bitmap_mutex);
         }
+
+        if (tmppages < 0) {
+            return tmppages;
+        }
+
         pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page);
     } while ((pss->page < hostpage_boundary) &&
              offset_in_ramblock(pss->block,
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#11 lore raw

Re: [PATCH 05/14] migration: Yield bitmap_mutex properly when sending/sleeping

Quoted 11 lines · Dr. David Alan Gilbert
Quoted 10 lines · * Peter Xu ([email protected]) wrote:

Don't take the bitmap mutex when sending pages, or when being throttled by migration_rate_limit() (which is a bit tricky to call it here in ram code, but seems still helpful).

It prepares for the possibility of concurrently sending pages in >1 threads using the function ram_save_host_page() because all threads may need the bitmap_mutex to operate on bitmaps, so that either sendmsg() or any kind of qemu_sem_wait() blocking for one thread will not block the other from progressing.

Quoted 6 lines
Quoted 1 line

I generally dont like taking locks conditionally; but this kind of looks OK; I think it needs a big comment on the start of the function saying that it's called and left with the lock held but that it might drop it temporarily.

Right, the code is slightly hard to read, I just didn't yet see a good and easy solution for it yet. It's just that we may still want to keep the lock as long as possible for precopy in one shot.

Quoted 5 lines
Quoted 5 lines
---
 migration/ram.c | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)
Quoted 13 lines

migration/ram.c static void postcopy_preempt_reset_channel(RAMState *rs) · line 2466

Quoted 9 lines
migration/ram.c +1 -0
index 8303252b6d..6e7de6087a 100644
@@ -2463,6 +2463,7 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
  */
 static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
 {
+    bool page_dirty, release_lock = postcopy_preempt_active();

Could you rename that to something like 'drop_lock' - you are taking the lock at the end even when you have 'release_lock' set - which is a bit strange naming.

Is there any difference on "drop" or "release"? I'll change the name anyway since I definitely trust you on any English comments, but please still let me know - I love to learn more on those! :)

Quoted 8 lines
Quoted 8 lines
migration/ram.c +0 -0
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
@@ -2486,22 +2487,41 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
             break;
         }
 
Quoted 11 lines

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2498

Quoted 9 lines
migration/ram.c +9 -0
+        page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page);
+        /*
+         * Properly yield the lock only in postcopy preempt mode because
+         * both migration thread and rp-return thread can operate on the
+         * bitmaps.
+         */
+        if (release_lock) {
+            qemu_mutex_unlock(&rs->bitmap_mutex);
+        }

Shouldn't the unlock/lock move inside the 'if (page_dirty) {' ?

I think we can move into it, but it may not be as optimal as keeping it as-is.

Consider a case where we've got the bitmap with continous zero bits. During postcopy, the migration thread could be spinning here with the lock held even if it doesn't send a thing. It could still block the other return path thread on sending urgent pages which may be outside the zero zones.

Quoted 17 lines
Quoted 6 lines
migration/ram.c +1 -1
         /* Check the pages is dirty and if it is send it */
-        if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) {
+        if (page_dirty) {
             tmppages = ram_save_target_page(rs, pss);

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2510

Quoted 10 lines
migration/ram.c +8 -2
-            if (tmppages < 0) {
-                return tmppages;
+            if (tmppages >= 0) {
+                pages += tmppages;
+                /*
+                 * Allow rate limiting to happen in the middle of huge pages if
+                 * something is sent in the current iteration.
+                 */
+                if (pagesize_bits > 1 && tmppages > 0) {
+                    migration_rate_limit();
Quoted 3 lines

This feels interesting, I know it's no change from before, and it's difficult to do here, but it seems odd to hold the lock around the sleeping in the rate limit.

Good point.. I think I'll leave it there for this patch because it's totally irrelevant, but seems proper in the future to do unlocking too for normal precopy.

Maybe I'll just attach a patch at the end of this series when I repost. That'll be easier before things got forgotten again.

Signature · 1 line
Peter Xu
#12 lore raw

Re: [PATCH 05/14] migration: Yield bitmap_mutex properly when sending/sleeping

Quoted 42 lines · * Peter Xu ([email protected]) wrote:
Quoted 11 lines · Dr. David Alan Gilbert
Quoted 10 lines · * Peter Xu ([email protected]) wrote:

Don't take the bitmap mutex when sending pages, or when being throttled by migration_rate_limit() (which is a bit tricky to call it here in ram code, but seems still helpful).

It prepares for the possibility of concurrently sending pages in >1 threads using the function ram_save_host_page() because all threads may need the bitmap_mutex to operate on bitmaps, so that either sendmsg() or any kind of qemu_sem_wait() blocking for one thread will not block the other from progressing.

Quoted 6 lines
Quoted 1 line

I generally dont like taking locks conditionally; but this kind of looks OK; I think it needs a big comment on the start of the function saying that it's called and left with the lock held but that it might drop it temporarily.

Right, the code is slightly hard to read, I just didn't yet see a good and easy solution for it yet. It's just that we may still want to keep the lock as long as possible for precopy in one shot.

Quoted 5 lines
Quoted 5 lines
---
 migration/ram.c | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)
Quoted 13 lines

migration/ram.c static void postcopy_preempt_reset_channel(RAMState *rs) · line 2466

Quoted 9 lines
migration/ram.c +1 -0
index 8303252b6d..6e7de6087a 100644
@@ -2463,6 +2463,7 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
  */
 static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
 {
+    bool page_dirty, release_lock = postcopy_preempt_active();

Could you rename that to something like 'drop_lock' - you are taking the lock at the end even when you have 'release_lock' set - which is a bit strange naming.

Quoted 3 lines

Is there any difference on "drop" or "release"? I'll change the name anyway since I definitely trust you on any English comments, but please still let me know - I love to learn more on those! :)

I'm not sure 'drop' is much better either; I was struggling to find a good nam.

Quoted 20 lines
Quoted 8 lines
Quoted 8 lines
migration/ram.c +0 -0
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
@@ -2486,22 +2487,41 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
             break;
         }
 
Quoted 11 lines

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2498

Quoted 9 lines
migration/ram.c +9 -0
+        page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page);
+        /*
+         * Properly yield the lock only in postcopy preempt mode because
+         * both migration thread and rp-return thread can operate on the
+         * bitmaps.
+         */
+        if (release_lock) {
+            qemu_mutex_unlock(&rs->bitmap_mutex);
+        }

Shouldn't the unlock/lock move inside the 'if (page_dirty) {' ?

Quoted 8 lines

I think we can move into it, but it may not be as optimal as keeping it as-is.

Consider a case where we've got the bitmap with continous zero bits. During postcopy, the migration thread could be spinning here with the lock held even if it doesn't send a thing. It could still block the other return path thread on sending urgent pages which may be outside the zero zones.

OK, that reason needs commenting then - you're going to do a lot of release/take pairs in that case which is going to show up as very hot; so hmm, if ti was just for that type of 'yield' behaviour you wouldn't normally do it for each bit.

Quoted 21 lines
Quoted 17 lines
Quoted 6 lines
migration/ram.c +1 -1
         /* Check the pages is dirty and if it is send it */
-        if (migration_bitmap_clear_dirty(rs, pss->block, pss->page)) {
+        if (page_dirty) {
             tmppages = ram_save_target_page(rs, pss);

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2510

Quoted 10 lines
migration/ram.c +8 -2
-            if (tmppages < 0) {
-                return tmppages;
+            if (tmppages >= 0) {
+                pages += tmppages;
+                /*
+                 * Allow rate limiting to happen in the middle of huge pages if
+                 * something is sent in the current iteration.
+                 */
+                if (pagesize_bits > 1 && tmppages > 0) {
+                    migration_rate_limit();
Quoted 3 lines

This feels interesting, I know it's no change from before, and it's difficult to do here, but it seems odd to hold the lock around the sleeping in the rate limit.

Quoted 6 lines

Good point.. I think I'll leave it there for this patch because it's totally irrelevant, but seems proper in the future to do unlocking too for normal precopy.

Maybe I'll just attach a patch at the end of this series when I repost. That'll be easier before things got forgotten again.

Dave

Quoted 3 lines
Signature · 1 line
Peter Xu
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#13 lore raw

Re: [PATCH 05/14] migration: Yield bitmap_mutex properly when sending/sleeping

Quoted 43 lines · Dr. David Alan Gilbert
Quoted 42 lines · * Peter Xu ([email protected]) wrote:
Quoted 11 lines · Dr. David Alan Gilbert
Quoted 10 lines · * Peter Xu ([email protected]) wrote:

Don't take the bitmap mutex when sending pages, or when being throttled by migration_rate_limit() (which is a bit tricky to call it here in ram code, but seems still helpful).

It prepares for the possibility of concurrently sending pages in >1 threads using the function ram_save_host_page() because all threads may need the bitmap_mutex to operate on bitmaps, so that either sendmsg() or any kind of qemu_sem_wait() blocking for one thread will not block the other from progressing.

Quoted 6 lines
Quoted 1 line

I generally dont like taking locks conditionally; but this kind of looks OK; I think it needs a big comment on the start of the function saying that it's called and left with the lock held but that it might drop it temporarily.

Right, the code is slightly hard to read, I just didn't yet see a good and easy solution for it yet. It's just that we may still want to keep the lock as long as possible for precopy in one shot.

Quoted 18 lines
Quoted 10 lines
---
 migration/ram.c | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)
migration/ram.c +0 -0
index 8303252b6d..6e7de6087a 100644

migration/ram.c static void postcopy_preempt_reset_channel(RAMState *rs) · line 2466

Quoted 4 lines
migration/ram.c +1 -0
@@ -2463,6 +2463,7 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
  */
 static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
 {
+    bool page_dirty, release_lock = postcopy_preempt_active();

Could you rename that to something like 'drop_lock' - you are taking the lock at the end even when you have 'release_lock' set - which is a bit strange naming.

Quoted 6 lines
Quoted 3 lines

Is there any difference on "drop" or "release"? I'll change the name anyway since I definitely trust you on any English comments, but please still let me know - I love to learn more on those! :)

I'm not sure 'drop' is much better either; I was struggling to find a good nam.

I can also call it "preempt_enabled".

Actually I can directly replace it with calling postcopy_preempt_active() always but I just want to make it crystal clear that the value is not changing and lock & unlock are always paired - in our case I think it is not changing, but the var helps to be 100% sure there'll be no possible bug on e.g. deadlock caused by state changing.

Quoted 30 lines
Quoted 21 lines
Quoted 9 lines
Quoted 9 lines
migration/ram.c +0 -0
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
@@ -2486,22 +2487,41 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
             break;
         }
 
Quoted 11 lines

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2498

Quoted 9 lines
migration/ram.c +9 -0
+        page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page);
+        /*
+         * Properly yield the lock only in postcopy preempt mode because
+         * both migration thread and rp-return thread can operate on the
+         * bitmaps.
+         */
+        if (release_lock) {
+            qemu_mutex_unlock(&rs->bitmap_mutex);
+        }

Shouldn't the unlock/lock move inside the 'if (page_dirty) {' ?

Quoted 8 lines

I think we can move into it, but it may not be as optimal as keeping it as-is.

Consider a case where we've got the bitmap with continous zero bits. During postcopy, the migration thread could be spinning here with the lock held even if it doesn't send a thing. It could still block the other return path thread on sending urgent pages which may be outside the zero zones.

Quoted 4 lines

OK, that reason needs commenting then - you're going to do a lot of release/take pairs in that case which is going to show up as very hot; so hmm, if ti was just for that type of 'yield' behaviour you wouldn't normally do it for each bit.

Hold on.. I think my assumption won't easily trigger, because at the end of the loop we'll try to look for the next "dirty" page. So continuously clean pages are unlikely, or I even think it's impossible because we're holding the mutex during scanning and clear-dirty, so no one will be able to flip the bit.

So yeah I think it's okay to move it into "page_dirty", but since we'll mostly always go into dirty maybe it's just that it won't help a lot either, because it'll be mostly the same as keeping it outside?

Signature · 1 line
Peter Xu
author
#14 lore raw

Re: [PATCH 05/14] migration: Yield bitmap_mutex properly when sending/sleeping

Quoted 100 lines · Peter Xu
Quoted 43 lines · Dr. David Alan Gilbert
Quoted 42 lines · * Peter Xu ([email protected]) wrote:
Quoted 11 lines · Dr. David Alan Gilbert
Quoted 10 lines · * Peter Xu ([email protected]) wrote:

Don't take the bitmap mutex when sending pages, or when being throttled by migration_rate_limit() (which is a bit tricky to call it here in ram code, but seems still helpful).

It prepares for the possibility of concurrently sending pages in >1 threads using the function ram_save_host_page() because all threads may need the bitmap_mutex to operate on bitmaps, so that either sendmsg() or any kind of qemu_sem_wait() blocking for one thread will not block the other from progressing.

Quoted 6 lines
Quoted 1 line

I generally dont like taking locks conditionally; but this kind of looks OK; I think it needs a big comment on the start of the function saying that it's called and left with the lock held but that it might drop it temporarily.

Right, the code is slightly hard to read, I just didn't yet see a good and easy solution for it yet. It's just that we may still want to keep the lock as long as possible for precopy in one shot.

Quoted 18 lines
Quoted 10 lines
---
 migration/ram.c | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)
migration/ram.c +0 -0
index 8303252b6d..6e7de6087a 100644

migration/ram.c static void postcopy_preempt_reset_channel(RAMState *rs) · line 2466

Quoted 4 lines
migration/ram.c +1 -0
@@ -2463,6 +2463,7 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
  */
 static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
 {
+    bool page_dirty, release_lock = postcopy_preempt_active();

Could you rename that to something like 'drop_lock' - you are taking the lock at the end even when you have 'release_lock' set - which is a bit strange naming.

Quoted 6 lines
Quoted 3 lines

Is there any difference on "drop" or "release"? I'll change the name anyway since I definitely trust you on any English comments, but please still let me know - I love to learn more on those! :)

I'm not sure 'drop' is much better either; I was struggling to find a good nam.

I can also call it "preempt_enabled".

Actually I can directly replace it with calling postcopy_preempt_active() always but I just want to make it crystal clear that the value is not changing and lock & unlock are always paired - in our case I think it is not changing, but the var helps to be 100% sure there'll be no possible bug on e.g. deadlock caused by state changing.

Quoted 30 lines
Quoted 21 lines
Quoted 9 lines
Quoted 9 lines
migration/ram.c +0 -0
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
@@ -2486,22 +2487,41 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
             break;
         }
 
Quoted 11 lines

migration/ram.c static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss) · line 2498

Quoted 9 lines
migration/ram.c +9 -0
+        page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page);
+        /*
+         * Properly yield the lock only in postcopy preempt mode because
+         * both migration thread and rp-return thread can operate on the
+         * bitmaps.
+         */
+        if (release_lock) {
+            qemu_mutex_unlock(&rs->bitmap_mutex);
+        }

Shouldn't the unlock/lock move inside the 'if (page_dirty) {' ?

Quoted 8 lines

I think we can move into it, but it may not be as optimal as keeping it as-is.

Consider a case where we've got the bitmap with continous zero bits. During postcopy, the migration thread could be spinning here with the lock held even if it doesn't send a thing. It could still block the other return path thread on sending urgent pages which may be outside the zero zones.

Quoted 4 lines

OK, that reason needs commenting then - you're going to do a lot of release/take pairs in that case which is going to show up as very hot; so hmm, if ti was just for that type of 'yield' behaviour you wouldn't normally do it for each bit.

Hold on.. I think my assumption won't easily trigger, because at the end of the loop we'll try to look for the next "dirty" page. So continuously clean pages are unlikely, or I even think it's impossible because we're holding the mutex during scanning and clear-dirty, so no one will be able to flip the bit.

Quoted 3 lines

So yeah I think it's okay to move it into "page_dirty", but since we'll mostly always go into dirty maybe it's just that it won't help a lot either, because it'll be mostly the same as keeping it outside?

IOW, maybe I should drop page_dirty directly and replace it with a check, failing migration if migration_bitmap_clear_dirty() returned false?

Signature · 1 line
Peter Xu
author
#15 lore raw

PATCH 6/14 migration: Use atomic ops properly for page accountings

In reply to Peter Xu

To prepare for thread-safety on page accountings, at least below counters need to be accessed only atomically, they are:

ram_counters.transferred
ram_counters.duplicate
ram_counters.normal
ram_counters.postcopy_bytes

There are a lot of other counters but they won't be accessed outside migration thread, then they're still safe to be accessed without atomic ops.

---
 migration/migration.c | 10 +++++-----
 migration/multifd.c   |  2 +-
 migration/ram.c       | 29 +++++++++++++++--------------
 3 files changed, 21 insertions(+), 20 deletions(-)
migration/migration.c +5 -5
index 07c74a79a2..0eacc0c99b 100644
@@ -1048,13 +1048,13 @@ static void populate_ram_info(MigrationInfo *info, MigrationState *s)
 
     info->has_ram = true;
     info->ram = g_malloc0(sizeof(*info->ram));
-    info->ram->transferred = ram_counters.transferred;
+    info->ram->transferred = qatomic_read(&ram_counters.transferred);
     info->ram->total = ram_bytes_total();
-    info->ram->duplicate = ram_counters.duplicate;
+    info->ram->duplicate = qatomic_read(&ram_counters.duplicate);
     /* legacy value.  It is not used anymore */
     info->ram->skipped = 0;
-    info->ram->normal = ram_counters.normal;
-    info->ram->normal_bytes = ram_counters.normal * page_size;
+    info->ram->normal = qatomic_read(&ram_counters.normal);
+    info->ram->normal_bytes = info->ram->normal * page_size;
     info->ram->mbps = s->mbps;
     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
     info->ram->dirty_sync_missed_zero_copy =
@@ -1065,7 +1065,7 @@ static void populate_ram_info(MigrationInfo *info, MigrationState *s)
     info->ram->pages_per_second = s->pages_per_second;
     info->ram->precopy_bytes = ram_counters.precopy_bytes;
     info->ram->downtime_bytes = ram_counters.downtime_bytes;
-    info->ram->postcopy_bytes = ram_counters.postcopy_bytes;
+    info->ram->postcopy_bytes = qatomic_read(&ram_counters.postcopy_bytes);
 
     if (migrate_use_xbzrle()) {
         info->has_xbzrle_cache = true;
migration/multifd.c +1 -1
index 586ddc9d65..460326acd4 100644
@@ -437,7 +437,7 @@ static int multifd_send_pages(QEMUFile *f)
                 + p->packet_len;
     qemu_file_acct_rate_limit(f, transferred);
     ram_counters.multifd_bytes += transferred;
-    ram_counters.transferred += transferred;
+    qatomic_add(&ram_counters.transferred, transferred);
     qemu_mutex_unlock(&p->mutex);
     qemu_sem_post(&p->sem);
 
migration/ram.c +15 -14
index 6e7de6087a..5bd3d76bf0 100644
@@ -432,11 +432,11 @@ static void ram_transferred_add(uint64_t bytes)
     if (runstate_is_running()) {
         ram_counters.precopy_bytes += bytes;
     } else if (migration_in_postcopy()) {
-        ram_counters.postcopy_bytes += bytes;
+        qatomic_add(&ram_counters.postcopy_bytes, bytes);
     } else {
         ram_counters.downtime_bytes += bytes;
     }
-    ram_counters.transferred += bytes;
+    qatomic_add(&ram_counters.transferred, bytes);
 }
 
 void dirty_sync_missed_zero_copy(void)
@@ -725,7 +725,7 @@ void mig_throttle_counter_reset(void)
 
     rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
     rs->num_dirty_pages_period = 0;
-    rs->bytes_xfer_prev = ram_counters.transferred;
+    rs->bytes_xfer_prev = qatomic_read(&ram_counters.transferred);
 }
 
 /**
@@ -1085,8 +1085,9 @@ uint64_t ram_pagesize_summary(void)
 
 uint64_t ram_get_total_transferred_pages(void)
 {
-    return  ram_counters.normal + ram_counters.duplicate +
-                compression_counters.pages + xbzrle_counters.pages;
+    return  qatomic_read(&ram_counters.normal) +
+        qatomic_read(&ram_counters.duplicate) +
+        compression_counters.pages + xbzrle_counters.pages;
 }
 
 static void migration_update_rates(RAMState *rs, int64_t end_time)
@@ -1145,8 +1146,8 @@ static void migration_trigger_throttle(RAMState *rs)
 {
     MigrationState *s = migrate_get_current();
     uint64_t threshold = s->parameters.throttle_trigger_threshold;
-
-    uint64_t bytes_xfer_period = ram_counters.transferred - rs->bytes_xfer_prev;
+    uint64_t bytes_xfer_period =
+        qatomic_read(&ram_counters.transferred) - rs->bytes_xfer_prev;
     uint64_t bytes_dirty_period = rs->num_dirty_pages_period * TARGET_PAGE_SIZE;
     uint64_t bytes_dirty_threshold = bytes_xfer_period * threshold / 100;
 
@@ -1285,7 +1286,7 @@ static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
     int len = save_zero_page_to_file(rs, rs->f, block, offset);
 
     if (len) {
-        ram_counters.duplicate++;
+        qatomic_inc(&ram_counters.duplicate);
         ram_transferred_add(len);
         return 1;
     }
@@ -1322,9 +1323,9 @@ static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
     }
 
     if (bytes_xmit > 0) {
-        ram_counters.normal++;
+        qatomic_inc(&ram_counters.normal);
     } else if (bytes_xmit == 0) {
-        ram_counters.duplicate++;
+        qatomic_inc(&ram_counters.duplicate);
     }
 
     return true;
@@ -1354,7 +1355,7 @@ static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
         qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE);
     }
     ram_transferred_add(TARGET_PAGE_SIZE);
-    ram_counters.normal++;
+    qatomic_inc(&ram_counters.normal);
     return 1;
 }
 
@@ -1448,7 +1449,7 @@ update_compress_thread_counts(const CompressParam *param, int bytes_xmit)
     ram_transferred_add(bytes_xmit);
 
     if (param->zero_page) {
-        ram_counters.duplicate++;
+        qatomic_inc(&ram_counters.duplicate);
         return;
     }
 
@@ -2620,9 +2621,9 @@ void acct_update_position(QEMUFile *f, size_t size, bool zero)
     uint64_t pages = size / TARGET_PAGE_SIZE;
 
     if (zero) {
-        ram_counters.duplicate += pages;
+        qatomic_add(&ram_counters.duplicate, pages);
     } else {
-        ram_counters.normal += pages;
+        qatomic_add(&ram_counters.normal, pages);
         ram_transferred_add(size);
         qemu_file_credit_transfer(f, size);
     }
Signature · 1 line
2.32.0
Reviewed-by on v1 6/14
#16 lore raw

Re: [PATCH 06/14] migration: Use atomic ops properly for page accountings

Quoted 8 lines · * Peter Xu ([email protected]) wrote:

To prepare for thread-safety on page accountings, at least below counters need to be accessed only atomically, they are:

ram_counters.transferred
ram_counters.duplicate
ram_counters.normal
ram_counters.postcopy_bytes

Quoted 5 lines

There are a lot of other counters but they won't be accessed outside migration thread, then they're still safe to be accessed without atomic ops.

I think this is OK; I'm not sure whether the memset 0's of ram_counters technically need changing.
I'd love to put a comment somewhere saying these fields need to be atomically read, but their qapi defined so I don't think we can.

Finally, we probably need to check these are happy on 32 bit builds, sometimes it's a bit funny with atomic adds.

Quoted 154 lines
---
 migration/migration.c | 10 +++++-----
 migration/multifd.c   |  2 +-
 migration/ram.c       | 29 +++++++++++++++--------------
 3 files changed, 21 insertions(+), 20 deletions(-)
migration/migration.c +5 -5
index 07c74a79a2..0eacc0c99b 100644
@@ -1048,13 +1048,13 @@ static void populate_ram_info(MigrationInfo *info, MigrationState *s)
 
     info->has_ram = true;
     info->ram = g_malloc0(sizeof(*info->ram));
-    info->ram->transferred = ram_counters.transferred;
+    info->ram->transferred = qatomic_read(&ram_counters.transferred);
     info->ram->total = ram_bytes_total();
-    info->ram->duplicate = ram_counters.duplicate;
+    info->ram->duplicate = qatomic_read(&ram_counters.duplicate);
     /* legacy value.  It is not used anymore */
     info->ram->skipped = 0;
-    info->ram->normal = ram_counters.normal;
-    info->ram->normal_bytes = ram_counters.normal * page_size;
+    info->ram->normal = qatomic_read(&ram_counters.normal);
+    info->ram->normal_bytes = info->ram->normal * page_size;
     info->ram->mbps = s->mbps;
     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
     info->ram->dirty_sync_missed_zero_copy =
@@ -1065,7 +1065,7 @@ static void populate_ram_info(MigrationInfo *info, MigrationState *s)
     info->ram->pages_per_second = s->pages_per_second;
     info->ram->precopy_bytes = ram_counters.precopy_bytes;
     info->ram->downtime_bytes = ram_counters.downtime_bytes;
-    info->ram->postcopy_bytes = ram_counters.postcopy_bytes;
+    info->ram->postcopy_bytes = qatomic_read(&ram_counters.postcopy_bytes);
 
     if (migrate_use_xbzrle()) {
         info->has_xbzrle_cache = true;
migration/multifd.c +1 -1
index 586ddc9d65..460326acd4 100644
@@ -437,7 +437,7 @@ static int multifd_send_pages(QEMUFile *f)
                 + p->packet_len;
     qemu_file_acct_rate_limit(f, transferred);
     ram_counters.multifd_bytes += transferred;
-    ram_counters.transferred += transferred;
+    qatomic_add(&ram_counters.transferred, transferred);
     qemu_mutex_unlock(&p->mutex);
     qemu_sem_post(&p->sem);
 
migration/ram.c +15 -14
index 6e7de6087a..5bd3d76bf0 100644
@@ -432,11 +432,11 @@ static void ram_transferred_add(uint64_t bytes)
     if (runstate_is_running()) {
         ram_counters.precopy_bytes += bytes;
     } else if (migration_in_postcopy()) {
-        ram_counters.postcopy_bytes += bytes;
+        qatomic_add(&ram_counters.postcopy_bytes, bytes);
     } else {
         ram_counters.downtime_bytes += bytes;
     }
-    ram_counters.transferred += bytes;
+    qatomic_add(&ram_counters.transferred, bytes);
 }
 
 void dirty_sync_missed_zero_copy(void)
@@ -725,7 +725,7 @@ void mig_throttle_counter_reset(void)
 
     rs->time_last_bitmap_sync = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
     rs->num_dirty_pages_period = 0;
-    rs->bytes_xfer_prev = ram_counters.transferred;
+    rs->bytes_xfer_prev = qatomic_read(&ram_counters.transferred);
 }
 
 /**
@@ -1085,8 +1085,9 @@ uint64_t ram_pagesize_summary(void)
 
 uint64_t ram_get_total_transferred_pages(void)
 {
-    return  ram_counters.normal + ram_counters.duplicate +
-                compression_counters.pages + xbzrle_counters.pages;
+    return  qatomic_read(&ram_counters.normal) +
+        qatomic_read(&ram_counters.duplicate) +
+        compression_counters.pages + xbzrle_counters.pages;
 }
 
 static void migration_update_rates(RAMState *rs, int64_t end_time)
@@ -1145,8 +1146,8 @@ static void migration_trigger_throttle(RAMState *rs)
 {
     MigrationState *s = migrate_get_current();
     uint64_t threshold = s->parameters.throttle_trigger_threshold;
-
-    uint64_t bytes_xfer_period = ram_counters.transferred - rs->bytes_xfer_prev;
+    uint64_t bytes_xfer_period =
+        qatomic_read(&ram_counters.transferred) - rs->bytes_xfer_prev;
     uint64_t bytes_dirty_period = rs->num_dirty_pages_period * TARGET_PAGE_SIZE;
     uint64_t bytes_dirty_threshold = bytes_xfer_period * threshold / 100;
 
@@ -1285,7 +1286,7 @@ static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
     int len = save_zero_page_to_file(rs, rs->f, block, offset);
 
     if (len) {
-        ram_counters.duplicate++;
+        qatomic_inc(&ram_counters.duplicate);
         ram_transferred_add(len);
         return 1;
     }
@@ -1322,9 +1323,9 @@ static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
     }
 
     if (bytes_xmit > 0) {
-        ram_counters.normal++;
+        qatomic_inc(&ram_counters.normal);
     } else if (bytes_xmit == 0) {
-        ram_counters.duplicate++;
+        qatomic_inc(&ram_counters.duplicate);
     }
 
     return true;
@@ -1354,7 +1355,7 @@ static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
         qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE);
     }
     ram_transferred_add(TARGET_PAGE_SIZE);
-    ram_counters.normal++;
+    qatomic_inc(&ram_counters.normal);
     return 1;
 }
 
@@ -1448,7 +1449,7 @@ update_compress_thread_counts(const CompressParam *param, int bytes_xmit)
     ram_transferred_add(bytes_xmit);
 
     if (param->zero_page) {
-        ram_counters.duplicate++;
+        qatomic_inc(&ram_counters.duplicate);
         return;
     }
 
@@ -2620,9 +2621,9 @@ void acct_update_position(QEMUFile *f, size_t size, bool zero)
     uint64_t pages = size / TARGET_PAGE_SIZE;
 
     if (zero) {
-        ram_counters.duplicate += pages;
+        qatomic_add(&ram_counters.duplicate, pages);
     } else {
-        ram_counters.normal += pages;
+        qatomic_add(&ram_counters.normal, pages);
         ram_transferred_add(size);
         qemu_file_credit_transfer(f, size);
     }
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#17 lore raw

Re: [PATCH 06/14] migration: Use atomic ops properly for page accountings

Quoted 9 lines · Dr. David Alan Gilbert
Quoted 8 lines · * Peter Xu ([email protected]) wrote:

To prepare for thread-safety on page accountings, at least below counters need to be accessed only atomically, they are:

ram_counters.transferred
ram_counters.duplicate
ram_counters.normal
ram_counters.postcopy_bytes

Quoted 8 lines
Quoted 5 lines

There are a lot of other counters but they won't be accessed outside migration thread, then they're still safe to be accessed without atomic ops.

I think this is OK; I'm not sure whether the memset 0's of ram_counters technically need changing.

IMHO they're fine - what we need there should be thing like WRITE_ONCE() just to make sure no register caches (actually atomic_write() is normally implemented with WRITE_ONCE afaik). But I think that's already guaranteed by memset() as the function call does, so we should be 100% safe.

Quoted 2 lines

I'd love to put a comment somewhere saying these fields need to be atomically read, but their qapi defined so I don't think we can.

How about I add a comment above ram_counters declarations in ram.c?

Quoted 3 lines

Finally, we probably need to check these are happy on 32 bit builds, sometimes it's a bit funny with atomic adds.

Yeah.. I hope using qatomic_*() APIs can help me avoid any issues. Or anything concerning? I'd be happy to test on specific things if there are.

Quoted 3 lines

Thanks!

Signature · 1 line
Peter Xu
#18 lore raw

Re: [PATCH 06/14] migration: Use atomic ops properly for page accountings

Quoted 16 lines · * Peter Xu ([email protected]) wrote:
Quoted 15 lines · Dr. David Alan Gilbert
Quoted 13 lines · * Peter Xu ([email protected]) wrote:

To prepare for thread-safety on page accountings, at least below counters need to be accessed only atomically, they are:

ram_counters.transferred
ram_counters.duplicate
ram_counters.normal
ram_counters.postcopy_bytes

There are a lot of other counters but they won't be accessed outside migration thread, then they're still safe to be accessed without atomic ops.

Quoted 7 lines
Quoted 2 lines

I think this is OK; I'm not sure whether the memset 0's of ram_counters technically need changing.

IMHO they're fine - what we need there should be thing like WRITE_ONCE() just to make sure no register caches (actually atomic_write() is normally implemented with WRITE_ONCE afaik). But I think that's already guaranteed by memset() as the function call does, so we should be 100% safe.

I agree you're probably OK.

Quoted 4 lines
Quoted 2 lines

I'd love to put a comment somewhere saying these fields need to be atomically read, but their qapi defined so I don't think we can.

How about I add a comment above ram_counters declarations in ram.c?

Yeh.

Quoted 6 lines
Quoted 3 lines

Finally, we probably need to check these are happy on 32 bit builds, sometimes it's a bit funny with atomic adds.

Yeah.. I hope using qatomic_*() APIs can help me avoid any issues. Or anything concerning? I'd be happy to test on specific things if there are.

I just remember hitting problems in the past; especially if we end up with trying to do a 64 bit atomic on a platofmr that can only do 32???

Dave

Quoted 9 lines
Quoted 3 lines

Thanks!

Signature · 1 line
Peter Xu
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#19 lore raw

Re: [PATCH 06/14] migration: Use atomic ops properly for page accountings

Quoted 35 lines · Dr. David Alan Gilbert
Quoted 16 lines · * Peter Xu ([email protected]) wrote:
Quoted 15 lines · Dr. David Alan Gilbert
Quoted 13 lines · * Peter Xu ([email protected]) wrote:

To prepare for thread-safety on page accountings, at least below counters need to be accessed only atomically, they are:

ram_counters.transferred
ram_counters.duplicate
ram_counters.normal
ram_counters.postcopy_bytes

There are a lot of other counters but they won't be accessed outside migration thread, then they're still safe to be accessed without atomic ops.

Quoted 7 lines
Quoted 2 lines

I think this is OK; I'm not sure whether the memset 0's of ram_counters technically need changing.

IMHO they're fine - what we need there should be thing like WRITE_ONCE() just to make sure no register caches (actually atomic_write() is normally implemented with WRITE_ONCE afaik). But I think that's already guaranteed by memset() as the function call does, so we should be 100% safe.

I agree you're probably OK.

Quoted 4 lines
Quoted 2 lines

I'd love to put a comment somewhere saying these fields need to be atomically read, but their qapi defined so I don't think we can.

How about I add a comment above ram_counters declarations in ram.c?

Yeh.

Quoted 1 line
Quoted 1 line
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

Finally, we probably need to check these are happy on 32 bit builds, sometimes it's a bit funny with atomic adds.

Yeah.. I hope using qatomic_*() APIs can help me avoid any issues. Or anything concerning? I'd be happy to test on specific things if there are.

I just remember hitting problems in the past; especially if we end up with trying to do a 64 bit atomic on a platofmr that can only do 32???

I see what you meant... when I was looking in the existing callers of qatomic_add(), I do find that we seem to have Stat64 just for that !CONFIG_ATOMIC64 problem.

I'll dig a bit on whether and how we can do that; the thing is these counters are in the qapi so I need to make sure it can support Stat64 somehow. Hmm..

Signature · 1 line
Peter Xu
author
#20 lore raw

Re: [PATCH 06/14] migration: Use atomic ops properly for page accountings

Quoted 45 lines · Peter Xu
Quoted 35 lines · Dr. David Alan Gilbert
Quoted 16 lines · * Peter Xu ([email protected]) wrote:
Quoted 15 lines · Dr. David Alan Gilbert
Quoted 13 lines · * Peter Xu ([email protected]) wrote:

To prepare for thread-safety on page accountings, at least below counters need to be accessed only atomically, they are:

ram_counters.transferred
ram_counters.duplicate
ram_counters.normal
ram_counters.postcopy_bytes

There are a lot of other counters but they won't be accessed outside migration thread, then they're still safe to be accessed without atomic ops.

Quoted 7 lines
Quoted 2 lines

I think this is OK; I'm not sure whether the memset 0's of ram_counters technically need changing.

IMHO they're fine - what we need there should be thing like WRITE_ONCE() just to make sure no register caches (actually atomic_write() is normally implemented with WRITE_ONCE afaik). But I think that's already guaranteed by memset() as the function call does, so we should be 100% safe.

I agree you're probably OK.

Quoted 4 lines
Quoted 2 lines

I'd love to put a comment somewhere saying these fields need to be atomically read, but their qapi defined so I don't think we can.

How about I add a comment above ram_counters declarations in ram.c?

Yeh.

Quoted 1 line
Quoted 1 line
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

Finally, we probably need to check these are happy on 32 bit builds, sometimes it's a bit funny with atomic adds.

Yeah.. I hope using qatomic_*() APIs can help me avoid any issues. Or anything concerning? I'd be happy to test on specific things if there are.

I just remember hitting problems in the past; especially if we end up with trying to do a 64 bit atomic on a platofmr that can only do 32???

Quoted 7 lines

I see what you meant... when I was looking in the existing callers of qatomic_add(), I do find that we seem to have Stat64 just for that !CONFIG_ATOMIC64 problem.

I'll dig a bit on whether and how we can do that; the thing is these counters are in the qapi so I need to make sure it can support Stat64 somehow. Hmm..

I think I can't directly change the qapi MigrationStats to make some of them to Stat64 since for !ATOMIC_64 systems Stat64 actually takes more than 64 bits space (since we'll need to do the locking with Stat64.lock), so it'll definitely break the ABI no matter what..

I don't have a better option but introduce another ram_counters_internal to maintain the fields that need atomic access, declaring as a Stat64 array. Then we only mirror those values to MigrationStats in QMP queries when needed. The mirror will not contain the lock itself so it'll keep the ABI.

Let me know if there's early comment for that, or I'll go with it. I'll definitely add some comment for ram_counters to explain the mirror counters in that case.

Thanks,

Signature · 1 line
Peter Xu
author
#21 lore raw

PATCH 7/14 migration: Teach PSS about host page

In reply to Peter Xu

Migration code has a lot to do with host pages. Teaching PSS core about the idea of host page helps a lot and makes the code clean. Meanwhile, this prepares for the future changes that can leverage the new PSS helpers that this patch introduces to send host page in another thread.

Three more fields are introduced for this:

(1) host_page_sending: this is set to true when QEMU is sending a host page, false otherwise.

(2) host_page_{start|end}: these point to the start/end of host page

      we're sending, and it's only valid when host_page_sending==true.

For example, when we look up the next dirty page on the ramblock, with host_page_sending==true, we'll not try to look for anything beyond the current host page boundary. This can be slightly efficient than current code because currently we'll set pss->page to next dirty bit (which can be over current host page boundary) and reset it to host page boundary if we found it goes beyond that.

With above, we can easily make migration_bitmap_find_dirty() self contained by updating pss->page properly. rs* parameter is removed because it's not even used in old code.

When sending a host page, we should use the pss helpers like this:

- pss_host_page_prepare(pss): called before sending host page
- pss_within_range(pss): whether we're still working on the cur host page?
- pss_host_page_finish(pss): called after sending a host page

Then we can use ram_save_target_page() to save one small page.

Currently ram_save_host_page() is still the only user. If there'll be another function to send host page (e.g. in return path thread) in the future, it should follow the same style.

---
 migration/ram.c | 95 +++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 76 insertions(+), 19 deletions(-)
migration/ram.c +76 -19
index 5bd3d76bf0..3f720b6de2 100644
@@ -474,6 +474,11 @@ struct PageSearchStatus {
      * postcopy pages via postcopy preempt channel.
      */
     bool         postcopy_target_channel;
+    /* Whether we're sending a host page */
+    bool          host_page_sending;
+    /* The start/end of current host page.  Only valid if host_page_sending==true */
+    unsigned long host_page_start;
+    unsigned long host_page_end;
 };
 typedef struct PageSearchStatus PageSearchStatus;
 
@@ -851,26 +856,38 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
 }
 
 /**
- * migration_bitmap_find_dirty: find the next dirty page from start
+ * pss_find_next_dirty: find the next dirty page of current ramblock
  *
- * Returns the page offset within memory region of the start of a dirty page
+ * This function updates pss->page to point to the next dirty page index
+ * within the ramblock to migrate, or the end of ramblock when nothing
+ * found.  Note that when pss->host_page_sending==true it means we're
+ * during sending a host page, so we won't look for dirty page that is
+ * outside the host page boundary.
  *
- * @rs: current RAM state
- * @rb: RAMBlock where to search for dirty pages
- * @start: page where we start the search
+ * @pss: the current page search status
  */
-static inline
-unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb,
-                                          unsigned long start)
+static void pss_find_next_dirty(PageSearchStatus *pss)
 {
+    RAMBlock *rb = pss->block;
     unsigned long size = rb->used_length >> TARGET_PAGE_BITS;
     unsigned long *bitmap = rb->bmap;
 
     if (ramblock_is_ignored(rb)) {
-        return size;
+        /* Points directly to the end, so we know no dirty page */
+        pss->page = size;
+        return;
+    }
+
+    /*
+     * If during sending a host page, only look for dirty pages within the
+     * current host page being send.
+     */
+    if (pss->host_page_sending) {
+        assert(pss->host_page_end);
+        size = MIN(size, pss->host_page_end);
     }
 
-    return find_next_bit(bitmap, size, start);
+    pss->page = find_next_bit(bitmap, size, pss->page);
 }
 
 static void migration_clear_memory_region_dirty_bitmap(RAMBlock *rb,
@@ -1556,7 +1573,9 @@ static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again)
     pss->postcopy_requested = false;
     pss->postcopy_target_channel = RAM_CHANNEL_PRECOPY;
 
-    pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page);
+    /* Update pss->page for the next dirty bit in ramblock */
+    pss_find_next_dirty(pss);
+
     if (pss->complete_round && pss->block == rs->last_seen_block &&
         pss->page >= rs->last_page) {
         /*
@@ -2446,6 +2465,44 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
     }
 }
 
+/* Should be called before sending a host page */
+static void pss_host_page_prepare(PageSearchStatus *pss)
+{
+    /* How many guest pages are there in one host page? */
+    size_t guest_pfns = qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
+
+    pss->host_page_sending = true;
+    pss->host_page_start = ROUND_DOWN(pss->page, guest_pfns);
+    pss->host_page_end = ROUND_UP(pss->page + 1, guest_pfns);
+}
+
+/*
+ * Whether the page pointed by PSS is within the host page being sent.
+ * Must be called after a previous pss_host_page_prepare().
+ */
+static bool pss_within_range(PageSearchStatus *pss)
+{
+    ram_addr_t ram_addr;
+
+    assert(pss->host_page_sending);
+
+    /* Over host-page boundary? */
+    if (pss->page >= pss->host_page_end) {
+        return false;
+    }
+
+    ram_addr = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS;
+
+    return offset_in_ramblock(pss->block, ram_addr);
+}
+
+static void pss_host_page_finish(PageSearchStatus *pss)
+{
+    pss->host_page_sending = false;
+    /* This is not needed, but just to reset it */
+    pss->host_page_start = pss->host_page_end = 0;
+}
+
 /**
  * ram_save_host_page: save a whole host page
  *
@@ -2468,8 +2525,6 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
-    unsigned long hostpage_boundary =
-        QEMU_ALIGN_UP(pss->page + 1, pagesize_bits);
     unsigned long start_page = pss->page;
     int res;
 
@@ -2482,6 +2537,9 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
         postcopy_preempt_choose_channel(rs, pss);
     }
 
+    /* Update host page boundary information */
+    pss_host_page_prepare(pss);
+
     do {
         if (postcopy_needs_preempt(rs, pss)) {
             postcopy_do_preempt(rs, pss);
@@ -2520,15 +2578,14 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
         }
 
         if (tmppages < 0) {
+            pss_host_page_finish(pss);
             return tmppages;
         }
 
-        pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page);
-    } while ((pss->page < hostpage_boundary) &&
-             offset_in_ramblock(pss->block,
-                                ((ram_addr_t)pss->page) << TARGET_PAGE_BITS));
-    /* The offset we leave with is the min boundary of host page and block */
-    pss->page = MIN(pss->page, hostpage_boundary);
+        pss_find_next_dirty(pss);
+    } while (pss_within_range(pss));
+
+    pss_host_page_finish(pss);
 
     /*
      * When with postcopy preempt mode, flush the data as soon as possible for
Signature · 1 line
2.32.0
Reviewed-by on v1 7/14
#22 lore raw

Re: [PATCH 07/14] migration: Teach PSS about host page

Quoted 30 lines · * Peter Xu ([email protected]) wrote:

Migration code has a lot to do with host pages. Teaching PSS core about the idea of host page helps a lot and makes the code clean. Meanwhile, this prepares for the future changes that can leverage the new PSS helpers that this patch introduces to send host page in another thread.

Three more fields are introduced for this:

(1) host_page_sending: this is set to true when QEMU is sending a host page, false otherwise.

(2) host_page_{start|end}: these point to the start/end of host page

      we're sending, and it's only valid when host_page_sending==true.

For example, when we look up the next dirty page on the ramblock, with host_page_sending==true, we'll not try to look for anything beyond the current host page boundary. This can be slightly efficient than current code because currently we'll set pss->page to next dirty bit (which can be over current host page boundary) and reset it to host page boundary if we found it goes beyond that.

With above, we can easily make migration_bitmap_find_dirty() self contained by updating pss->page properly. rs* parameter is removed because it's not even used in old code.

When sending a host page, we should use the pss helpers like this:

- pss_host_page_prepare(pss): called before sending host page
- pss_within_range(pss): whether we're still working on the cur host page?
- pss_host_page_finish(pss): called after sending a host page

Quoted 7 lines

Then we can use ram_save_target_page() to save one small page.

Currently ram_save_host_page() is still the only user. If there'll be another function to send host page (e.g. in return path thread) in the future, it should follow the same style.

Quoted 168 lines
---
 migration/ram.c | 95 +++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 76 insertions(+), 19 deletions(-)
migration/ram.c +76 -19
index 5bd3d76bf0..3f720b6de2 100644
@@ -474,6 +474,11 @@ struct PageSearchStatus {
      * postcopy pages via postcopy preempt channel.
      */
     bool         postcopy_target_channel;
+    /* Whether we're sending a host page */
+    bool          host_page_sending;
+    /* The start/end of current host page.  Only valid if host_page_sending==true */
+    unsigned long host_page_start;
+    unsigned long host_page_end;
 };
 typedef struct PageSearchStatus PageSearchStatus;
 
@@ -851,26 +856,38 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
 }
 
 /**
- * migration_bitmap_find_dirty: find the next dirty page from start
+ * pss_find_next_dirty: find the next dirty page of current ramblock
  *
- * Returns the page offset within memory region of the start of a dirty page
+ * This function updates pss->page to point to the next dirty page index
+ * within the ramblock to migrate, or the end of ramblock when nothing
+ * found.  Note that when pss->host_page_sending==true it means we're
+ * during sending a host page, so we won't look for dirty page that is
+ * outside the host page boundary.
  *
- * @rs: current RAM state
- * @rb: RAMBlock where to search for dirty pages
- * @start: page where we start the search
+ * @pss: the current page search status
  */
-static inline
-unsigned long migration_bitmap_find_dirty(RAMState *rs, RAMBlock *rb,
-                                          unsigned long start)
+static void pss_find_next_dirty(PageSearchStatus *pss)
 {
+    RAMBlock *rb = pss->block;
     unsigned long size = rb->used_length >> TARGET_PAGE_BITS;
     unsigned long *bitmap = rb->bmap;
 
     if (ramblock_is_ignored(rb)) {
-        return size;
+        /* Points directly to the end, so we know no dirty page */
+        pss->page = size;
+        return;
+    }
+
+    /*
+     * If during sending a host page, only look for dirty pages within the
+     * current host page being send.
+     */
+    if (pss->host_page_sending) {
+        assert(pss->host_page_end);
+        size = MIN(size, pss->host_page_end);
     }
 
-    return find_next_bit(bitmap, size, start);
+    pss->page = find_next_bit(bitmap, size, pss->page);
 }
 
 static void migration_clear_memory_region_dirty_bitmap(RAMBlock *rb,
@@ -1556,7 +1573,9 @@ static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again)
     pss->postcopy_requested = false;
     pss->postcopy_target_channel = RAM_CHANNEL_PRECOPY;
 
-    pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page);
+    /* Update pss->page for the next dirty bit in ramblock */
+    pss_find_next_dirty(pss);
+
     if (pss->complete_round && pss->block == rs->last_seen_block &&
         pss->page >= rs->last_page) {
         /*
@@ -2446,6 +2465,44 @@ static void postcopy_preempt_reset_channel(RAMState *rs)
     }
 }
 
+/* Should be called before sending a host page */
+static void pss_host_page_prepare(PageSearchStatus *pss)
+{
+    /* How many guest pages are there in one host page? */
+    size_t guest_pfns = qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
+
+    pss->host_page_sending = true;
+    pss->host_page_start = ROUND_DOWN(pss->page, guest_pfns);
+    pss->host_page_end = ROUND_UP(pss->page + 1, guest_pfns);
+}
+
+/*
+ * Whether the page pointed by PSS is within the host page being sent.
+ * Must be called after a previous pss_host_page_prepare().
+ */
+static bool pss_within_range(PageSearchStatus *pss)
+{
+    ram_addr_t ram_addr;
+
+    assert(pss->host_page_sending);
+
+    /* Over host-page boundary? */
+    if (pss->page >= pss->host_page_end) {
+        return false;
+    }
+
+    ram_addr = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS;
+
+    return offset_in_ramblock(pss->block, ram_addr);
+}
+
+static void pss_host_page_finish(PageSearchStatus *pss)
+{
+    pss->host_page_sending = false;
+    /* This is not needed, but just to reset it */
+    pss->host_page_start = pss->host_page_end = 0;
+}
+
 /**
  * ram_save_host_page: save a whole host page
  *
@@ -2468,8 +2525,6 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
     int tmppages, pages = 0;
     size_t pagesize_bits =
         qemu_ram_pagesize(pss->block) >> TARGET_PAGE_BITS;
-    unsigned long hostpage_boundary =
-        QEMU_ALIGN_UP(pss->page + 1, pagesize_bits);
     unsigned long start_page = pss->page;
     int res;
 
@@ -2482,6 +2537,9 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
         postcopy_preempt_choose_channel(rs, pss);
     }
 
+    /* Update host page boundary information */
+    pss_host_page_prepare(pss);
+
     do {
         if (postcopy_needs_preempt(rs, pss)) {
             postcopy_do_preempt(rs, pss);
@@ -2520,15 +2578,14 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
         }
 
         if (tmppages < 0) {
+            pss_host_page_finish(pss);
             return tmppages;
         }
 
-        pss->page = migration_bitmap_find_dirty(rs, pss->block, pss->page);
-    } while ((pss->page < hostpage_boundary) &&
-             offset_in_ramblock(pss->block,
-                                ((ram_addr_t)pss->page) << TARGET_PAGE_BITS));
-    /* The offset we leave with is the min boundary of host page and block */
-    pss->page = MIN(pss->page, hostpage_boundary);
+        pss_find_next_dirty(pss);
+    } while (pss_within_range(pss));
+
+    pss_host_page_finish(pss);
 
     /*
      * When with postcopy preempt mode, flush the data as soon as possible for
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#23 lore raw

PATCH 8/14 migration: Introduce pss_channel

In reply to Peter Xu

Introduce pss_channel for PageSearchStatus, define it as "the migration channel to be used to transfer this host page".

We used to have rs->f, which is a mirror to MigrationState.to_dst_file.

After postcopy preempt initial version, rs->f can be dynamically changed depending on which channel we want to use.

But that later work still doesn't grant full concurrency of sending pages in e.g. different threads, because rs->f can either be the PRECOPY channel or POSTCOPY channel. This needs to be per-thread too.

PageSearchStatus is actually a good piece of struct which we can leverage if we want to have multiple threads sending pages. Sending a single guest page may not make sense, so we make the granule to be "host page", and in the PSS structure we allow specify a QEMUFile* to migrate a specific host page. Then we open the possibility to specify different channels in different threads with different PSS structures.

The PSS prefix can be slightly misleading here because e.g. for the upcoming usage of postcopy channel/thread it's not "searching" (or, scanning) at all but sending the explicit page that was requested. However since PSS existed for some years keep it as-is until someone complains.

This patch mostly (simply) replace rs->f with pss->pss_channel only. No functional change intended for this patch yet. But it does prepare to finally drop rs->f, and make ram_save_guest_page() thread safe.

---
 migration/ram.c | 70 +++++++++++++++++++++++++++----------------------
 1 file changed, 38 insertions(+), 32 deletions(-)
migration/ram.c +38 -32
index 3f720b6de2..40ff5dc49f 100644
@@ -446,6 +446,8 @@ void dirty_sync_missed_zero_copy(void)
 
 /* used by the search for pages to send */
 struct PageSearchStatus {
+    /* The migration channel used for a specific host page */
+    QEMUFile    *pss_channel;
     /* Current block being searched */
     RAMBlock    *block;
     /* Current page to search from */
@@ -768,9 +770,9 @@ static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr)
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
-                            ram_addr_t current_addr, RAMBlock *block,
-                            ram_addr_t offset)
+static int save_xbzrle_page(RAMState *rs, QEMUFile *file,
+                            uint8_t **current_data, ram_addr_t current_addr,
+                            RAMBlock *block, ram_addr_t offset)
 {
     int encoded_len = 0, bytes_xbzrle;
     uint8_t *prev_cached_page;
@@ -838,11 +840,11 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
     }
 
     /* Send XBZRLE based compressed page */
-    bytes_xbzrle = save_page_header(rs, rs->f, block,
+    bytes_xbzrle = save_page_header(rs, file, block,
                                     offset | RAM_SAVE_FLAG_XBZRLE);
-    qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE);
-    qemu_put_be16(rs->f, encoded_len);
-    qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len);
+    qemu_put_byte(file, ENCODING_FLAG_XBZRLE);
+    qemu_put_be16(file, encoded_len);
+    qemu_put_buffer(file, XBZRLE.encoded_buf, encoded_len);
     bytes_xbzrle += encoded_len + 1 + 2;
     /*
      * Like compressed_size (please see update_compress_thread_counts),
@@ -1298,9 +1300,10 @@ static int save_zero_page_to_file(RAMState *rs, QEMUFile *file,
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
+static int save_zero_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+                          ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(rs, rs->f, block, offset);
+    int len = save_zero_page_to_file(rs, file, block, offset);
 
     if (len) {
         qatomic_inc(&ram_counters.duplicate);
@@ -1317,15 +1320,15 @@ static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
  *
  * Return true if the pages has been saved, otherwise false is returned.
  */
-static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
-                              int *pages)
+static bool control_save_page(PageSearchStatus *pss, RAMBlock *block,
+                              ram_addr_t offset, int *pages)
 {
     uint64_t bytes_xmit = 0;
     int ret;
 
     *pages = -1;
-    ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE,
-                                &bytes_xmit);
+    ret = ram_control_save_page(pss->pss_channel, block->offset, offset,
+                                TARGET_PAGE_SIZE, &bytes_xmit);
     if (ret == RAM_SAVE_CONTROL_NOT_SUPP) {
         return false;
     }
@@ -1359,17 +1362,17 @@ static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
  * @buf: the page to be sent
  * @async: send to page asyncly
  */
-static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
-                            uint8_t *buf, bool async)
+static int save_normal_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+                            ram_addr_t offset, uint8_t *buf, bool async)
 {
-    ram_transferred_add(save_page_header(rs, rs->f, block,
+    ram_transferred_add(save_page_header(rs, file, block,
                                          offset | RAM_SAVE_FLAG_PAGE));
     if (async) {
-        qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE,
+        qemu_put_buffer_async(file, buf, TARGET_PAGE_SIZE,
                               migrate_release_ram() &&
                               migration_in_postcopy());
     } else {
-        qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE);
+        qemu_put_buffer(file, buf, TARGET_PAGE_SIZE);
     }
     ram_transferred_add(TARGET_PAGE_SIZE);
     qatomic_inc(&ram_counters.normal);
@@ -1402,8 +1405,8 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     XBZRLE_cache_lock();
     if (rs->xbzrle_enabled && !migration_in_postcopy()) {
-        pages = save_xbzrle_page(rs, &p, current_addr, block,
-                                 offset);
+        pages = save_xbzrle_page(rs, pss->pss_channel, &p, current_addr,
+                                 block, offset);
         if (!rs->last_stage) {
             /* Can't send this cached data async, since the cache page
              * might get updated before it gets to the wire
@@ -1414,7 +1417,8 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     /* XBZRLE overflow or normal page */
     if (pages == -1) {
-        pages = save_normal_page(rs, block, offset, p, send_async);
+        pages = save_normal_page(rs, pss->pss_channel, block, offset,
+                                 p, send_async);
     }
 
     XBZRLE_cache_unlock();
@@ -1422,10 +1426,10 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
     return pages;
 }
 
-static int ram_save_multifd_page(RAMState *rs, RAMBlock *block,
+static int ram_save_multifd_page(QEMUFile *file, RAMBlock *block,
                                  ram_addr_t offset)
 {
-    if (multifd_queue_page(rs->f, block, offset) < 0) {
+    if (multifd_queue_page(file, block, offset) < 0) {
         return -1;
     }
     ram_counters.normal++;
@@ -1720,7 +1724,7 @@ static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss,
         uint64_t run_length = (pss->page - start_page) << TARGET_PAGE_BITS;
 
         /* Flush async buffers before un-protect. */
-        qemu_fflush(rs->f);
+        qemu_fflush(pss->pss_channel);
         /* Un-protect memory range. */
         res = uffd_change_protection(rs->uffdio_fd, page_address, run_length,
                 false, false);
@@ -2307,7 +2311,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
     ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS;
     int res;
 
-    if (control_save_page(rs, block, offset, &res)) {
+    if (control_save_page(pss, block, offset, &res)) {
         return res;
     }
 
@@ -2315,7 +2319,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
         return 1;
     }
 
-    res = save_zero_page(rs, block, offset);
+    res = save_zero_page(rs, pss->pss_channel, block, offset);
     if (res > 0) {
         /* Must let xbzrle know, otherwise a previous (now 0'd) cached
          * page would be stale
@@ -2336,7 +2340,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
      */
     if (!save_page_use_compression(rs) && migrate_use_multifd()
         && !migration_in_postcopy()) {
-        return ram_save_multifd_page(rs, block, offset);
+        return ram_save_multifd_page(pss->pss_channel, block, offset);
     }
 
     return ram_save_page(rs, pss);
@@ -2533,10 +2537,6 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
         return 0;
     }
 
-    if (postcopy_preempt_active()) {
-        postcopy_preempt_choose_channel(rs, pss);
-    }
-
     /* Update host page boundary information */
     pss_host_page_prepare(pss);
 
@@ -2597,7 +2597,7 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
      * explicit flush or it won't flush until the buffer is full.
      */
     if (migrate_postcopy_preempt() && pss->postcopy_requested) {
-        qemu_fflush(rs->f);
+        qemu_fflush(pss->pss_channel);
     }
 
     res = ram_save_release_protection(rs, pss, start_page);
@@ -2663,6 +2663,12 @@ static int ram_find_and_save_block(RAMState *rs)
         }
 
         if (found) {
+            /* Update rs->f with correct channel */
+            if (postcopy_preempt_active()) {
+                postcopy_preempt_choose_channel(rs, &pss);
+            }
+            /* Cache rs->f in pss_channel (TODO: remove rs->f) */
+            pss.pss_channel = rs->f;
             pages = ram_save_host_page(rs, &pss);
         }
     } while (!pages && again);
Signature · 1 line
2.32.0
Reviewed-by on v1 8/14
#24 lore raw

Re: [PATCH 08/14] migration: Introduce pss_channel

Quoted 24 lines · * Peter Xu ([email protected]) wrote:

Introduce pss_channel for PageSearchStatus, define it as "the migration channel to be used to transfer this host page".

We used to have rs->f, which is a mirror to MigrationState.to_dst_file.

After postcopy preempt initial version, rs->f can be dynamically changed depending on which channel we want to use.

But that later work still doesn't grant full concurrency of sending pages in e.g. different threads, because rs->f can either be the PRECOPY channel or POSTCOPY channel. This needs to be per-thread too.

PageSearchStatus is actually a good piece of struct which we can leverage if we want to have multiple threads sending pages. Sending a single guest page may not make sense, so we make the granule to be "host page", and in the PSS structure we allow specify a QEMUFile* to migrate a specific host page. Then we open the possibility to specify different channels in different threads with different PSS structures.

The PSS prefix can be slightly misleading here because e.g. for the upcoming usage of postcopy channel/thread it's not "searching" (or, scanning) at all but sending the explicit page that was requested. However since PSS existed for some years keep it as-is until someone complains.

Quoted 5 lines

This patch mostly (simply) replace rs->f with pss->pss_channel only. No functional change intended for this patch yet. But it does prepare to finally drop rs->f, and make ram_save_guest_page() thread safe.

Quoted 208 lines
---
 migration/ram.c | 70 +++++++++++++++++++++++++++----------------------
 1 file changed, 38 insertions(+), 32 deletions(-)
migration/ram.c +38 -32
index 3f720b6de2..40ff5dc49f 100644
@@ -446,6 +446,8 @@ void dirty_sync_missed_zero_copy(void)
 
 /* used by the search for pages to send */
 struct PageSearchStatus {
+    /* The migration channel used for a specific host page */
+    QEMUFile    *pss_channel;
     /* Current block being searched */
     RAMBlock    *block;
     /* Current page to search from */
@@ -768,9 +770,9 @@ static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr)
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
-                            ram_addr_t current_addr, RAMBlock *block,
-                            ram_addr_t offset)
+static int save_xbzrle_page(RAMState *rs, QEMUFile *file,
+                            uint8_t **current_data, ram_addr_t current_addr,
+                            RAMBlock *block, ram_addr_t offset)
 {
     int encoded_len = 0, bytes_xbzrle;
     uint8_t *prev_cached_page;
@@ -838,11 +840,11 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
     }
 
     /* Send XBZRLE based compressed page */
-    bytes_xbzrle = save_page_header(rs, rs->f, block,
+    bytes_xbzrle = save_page_header(rs, file, block,
                                     offset | RAM_SAVE_FLAG_XBZRLE);
-    qemu_put_byte(rs->f, ENCODING_FLAG_XBZRLE);
-    qemu_put_be16(rs->f, encoded_len);
-    qemu_put_buffer(rs->f, XBZRLE.encoded_buf, encoded_len);
+    qemu_put_byte(file, ENCODING_FLAG_XBZRLE);
+    qemu_put_be16(file, encoded_len);
+    qemu_put_buffer(file, XBZRLE.encoded_buf, encoded_len);
     bytes_xbzrle += encoded_len + 1 + 2;
     /*
      * Like compressed_size (please see update_compress_thread_counts),
@@ -1298,9 +1300,10 @@ static int save_zero_page_to_file(RAMState *rs, QEMUFile *file,
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
+static int save_zero_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+                          ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(rs, rs->f, block, offset);
+    int len = save_zero_page_to_file(rs, file, block, offset);
 
     if (len) {
         qatomic_inc(&ram_counters.duplicate);
@@ -1317,15 +1320,15 @@ static int save_zero_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
  *
  * Return true if the pages has been saved, otherwise false is returned.
  */
-static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
-                              int *pages)
+static bool control_save_page(PageSearchStatus *pss, RAMBlock *block,
+                              ram_addr_t offset, int *pages)
 {
     uint64_t bytes_xmit = 0;
     int ret;
 
     *pages = -1;
-    ret = ram_control_save_page(rs->f, block->offset, offset, TARGET_PAGE_SIZE,
-                                &bytes_xmit);
+    ret = ram_control_save_page(pss->pss_channel, block->offset, offset,
+                                TARGET_PAGE_SIZE, &bytes_xmit);
     if (ret == RAM_SAVE_CONTROL_NOT_SUPP) {
         return false;
     }
@@ -1359,17 +1362,17 @@ static bool control_save_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
  * @buf: the page to be sent
  * @async: send to page asyncly
  */
-static int save_normal_page(RAMState *rs, RAMBlock *block, ram_addr_t offset,
-                            uint8_t *buf, bool async)
+static int save_normal_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+                            ram_addr_t offset, uint8_t *buf, bool async)
 {
-    ram_transferred_add(save_page_header(rs, rs->f, block,
+    ram_transferred_add(save_page_header(rs, file, block,
                                          offset | RAM_SAVE_FLAG_PAGE));
     if (async) {
-        qemu_put_buffer_async(rs->f, buf, TARGET_PAGE_SIZE,
+        qemu_put_buffer_async(file, buf, TARGET_PAGE_SIZE,
                               migrate_release_ram() &&
                               migration_in_postcopy());
     } else {
-        qemu_put_buffer(rs->f, buf, TARGET_PAGE_SIZE);
+        qemu_put_buffer(file, buf, TARGET_PAGE_SIZE);
     }
     ram_transferred_add(TARGET_PAGE_SIZE);
     qatomic_inc(&ram_counters.normal);
@@ -1402,8 +1405,8 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     XBZRLE_cache_lock();
     if (rs->xbzrle_enabled && !migration_in_postcopy()) {
-        pages = save_xbzrle_page(rs, &p, current_addr, block,
-                                 offset);
+        pages = save_xbzrle_page(rs, pss->pss_channel, &p, current_addr,
+                                 block, offset);
         if (!rs->last_stage) {
             /* Can't send this cached data async, since the cache page
              * might get updated before it gets to the wire
@@ -1414,7 +1417,8 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     /* XBZRLE overflow or normal page */
     if (pages == -1) {
-        pages = save_normal_page(rs, block, offset, p, send_async);
+        pages = save_normal_page(rs, pss->pss_channel, block, offset,
+                                 p, send_async);
     }
 
     XBZRLE_cache_unlock();
@@ -1422,10 +1426,10 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
     return pages;
 }
 
-static int ram_save_multifd_page(RAMState *rs, RAMBlock *block,
+static int ram_save_multifd_page(QEMUFile *file, RAMBlock *block,
                                  ram_addr_t offset)
 {
-    if (multifd_queue_page(rs->f, block, offset) < 0) {
+    if (multifd_queue_page(file, block, offset) < 0) {
         return -1;
     }
     ram_counters.normal++;
@@ -1720,7 +1724,7 @@ static int ram_save_release_protection(RAMState *rs, PageSearchStatus *pss,
         uint64_t run_length = (pss->page - start_page) << TARGET_PAGE_BITS;
 
         /* Flush async buffers before un-protect. */
-        qemu_fflush(rs->f);
+        qemu_fflush(pss->pss_channel);
         /* Un-protect memory range. */
         res = uffd_change_protection(rs->uffdio_fd, page_address, run_length,
                 false, false);
@@ -2307,7 +2311,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
     ram_addr_t offset = ((ram_addr_t)pss->page) << TARGET_PAGE_BITS;
     int res;
 
-    if (control_save_page(rs, block, offset, &res)) {
+    if (control_save_page(pss, block, offset, &res)) {
         return res;
     }
 
@@ -2315,7 +2319,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
         return 1;
     }
 
-    res = save_zero_page(rs, block, offset);
+    res = save_zero_page(rs, pss->pss_channel, block, offset);
     if (res > 0) {
         /* Must let xbzrle know, otherwise a previous (now 0'd) cached
          * page would be stale
@@ -2336,7 +2340,7 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
      */
     if (!save_page_use_compression(rs) && migrate_use_multifd()
         && !migration_in_postcopy()) {
-        return ram_save_multifd_page(rs, block, offset);
+        return ram_save_multifd_page(pss->pss_channel, block, offset);
     }
 
     return ram_save_page(rs, pss);
@@ -2533,10 +2537,6 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
         return 0;
     }
 
-    if (postcopy_preempt_active()) {
-        postcopy_preempt_choose_channel(rs, pss);
-    }
-
     /* Update host page boundary information */
     pss_host_page_prepare(pss);
 
@@ -2597,7 +2597,7 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
      * explicit flush or it won't flush until the buffer is full.
      */
     if (migrate_postcopy_preempt() && pss->postcopy_requested) {
-        qemu_fflush(rs->f);
+        qemu_fflush(pss->pss_channel);
     }
 
     res = ram_save_release_protection(rs, pss, start_page);
@@ -2663,6 +2663,12 @@ static int ram_find_and_save_block(RAMState *rs)
         }
 
         if (found) {
+            /* Update rs->f with correct channel */
+            if (postcopy_preempt_active()) {
+                postcopy_preempt_choose_channel(rs, &pss);
+            }
+            /* Cache rs->f in pss_channel (TODO: remove rs->f) */
+            pss.pss_channel = rs->f;
             pages = ram_save_host_page(rs, &pss);
         }
     } while (!pages && again);
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#25 lore raw

PATCH 9/14 migration: Add pss_init()

In reply to Peter Xu

Helper to init PSS structures.

---
 migration/ram.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
migration/ram.c +9 -3
index 40ff5dc49f..b4b36ca59e 100644
@@ -535,6 +535,14 @@ static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block,
 static void postcopy_preempt_restore(RAMState *rs, PageSearchStatus *pss,
                                      bool postcopy_requested);
 
+/* NOTE: page is the PFN not real ram_addr_t. */
+static void pss_init(PageSearchStatus *pss, RAMBlock *rb, ram_addr_t page)
+{
+    pss->block = rb;
+    pss->page = page;
+    pss->complete_round = false;
+}
+
 static void *do_data_compress(void *opaque)
 {
     CompressParam *param = opaque;
@@ -2640,9 +2648,7 @@ static int ram_find_and_save_block(RAMState *rs)
         rs->last_page = 0;
     }
 
-    pss.block = rs->last_seen_block;
-    pss.page = rs->last_page;
-    pss.complete_round = false;
+    pss_init(&pss, rs->last_seen_block, rs->last_page);
 
     do {
         again = true;
Signature · 1 line
2.32.0
Reviewed-by on v1 9/14
#26 lore raw

Re: [PATCH 09/14] migration: Add pss_init()

Quoted 3 lines · * Peter Xu ([email protected]) wrote:

Helper to init PSS structures.

Quoted 37 lines
---
 migration/ram.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
migration/ram.c +9 -3
index 40ff5dc49f..b4b36ca59e 100644
@@ -535,6 +535,14 @@ static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block,
 static void postcopy_preempt_restore(RAMState *rs, PageSearchStatus *pss,
                                      bool postcopy_requested);
 
+/* NOTE: page is the PFN not real ram_addr_t. */
+static void pss_init(PageSearchStatus *pss, RAMBlock *rb, ram_addr_t page)
+{
+    pss->block = rb;
+    pss->page = page;
+    pss->complete_round = false;
+}
+
 static void *do_data_compress(void *opaque)
 {
     CompressParam *param = opaque;
@@ -2640,9 +2648,7 @@ static int ram_find_and_save_block(RAMState *rs)
         rs->last_page = 0;
     }
 
-    pss.block = rs->last_seen_block;
-    pss.page = rs->last_page;
-    pss.complete_round = false;
+    pss_init(&pss, rs->last_seen_block, rs->last_page);
 
     do {
         again = true;
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#27 lore raw

PATCH 10/14 migration: Make PageSearchStatus part of RAMState

In reply to Peter Xu

We used to allocate PSS structure on the stack for precopy when sending pages. Make it static, so as to describe per-channel ram migration status.

Here we declared RAM_CHANNEL_MAX instances, preparing for postcopy to use it, even though this patch has not yet to start using the 2nd instance.

This should not have any functional change per se, but it already starts to export PSS information via the RAMState, so that e.g. one PSS channel can start to reference the other PSS channel.

Always protect PSS access using the same RAMState.bitmap_mutex. We already do so, so no code change needed, just some comment update. Maybe we should consider renaming bitmap_mutex some day as it's going to be a more commonly and big mutex we use for ram states, but just leave it for later.

---
 migration/ram.c | 112 ++++++++++++++++++++++++++----------------------
 1 file changed, 61 insertions(+), 51 deletions(-)
migration/ram.c +61 -51
index b4b36ca59e..dbe11e1ace 100644
@@ -85,6 +85,46 @@
 
 XBZRLECacheStats xbzrle_counters;
 
+/* used by the search for pages to send */
+struct PageSearchStatus {
+    /* The migration channel used for a specific host page */
+    QEMUFile    *pss_channel;
+    /* Current block being searched */
+    RAMBlock    *block;
+    /* Current page to search from */
+    unsigned long page;
+    /* Set once we wrap around */
+    bool         complete_round;
+    /*
+     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
+     * postcopy.  When set, the request is "urgent" because the dest QEMU
+     * threads are waiting for us.
+     */
+    bool         postcopy_requested;
+    /*
+     * [POSTCOPY-ONLY] The target channel to use to send current page.
+     *
+     * Note: This may _not_ match with the value in postcopy_requested
+     * above. Let's imagine the case where the postcopy request is exactly
+     * the page that we're sending in progress during precopy. In this case
+     * we'll have postcopy_requested set to true but the target channel
+     * will be the precopy channel (so that we don't split brain on that
+     * specific page since the precopy channel already contains partial of
+     * that page data).
+     *
+     * Besides that specific use case, postcopy_target_channel should
+     * always be equal to postcopy_requested, because by default we send
+     * postcopy pages via postcopy preempt channel.
+     */
+    bool         postcopy_target_channel;
+    /* Whether we're sending a host page */
+    bool          host_page_sending;
+    /* The start/end of current host page.  Invalid if host_page_sending==false */
+    unsigned long host_page_start;
+    unsigned long host_page_end;
+};
+typedef struct PageSearchStatus PageSearchStatus;
+
 /* struct contains XBZRLE cache and a static page
    used by the compression */
 static struct {
@@ -319,6 +359,11 @@ typedef struct {
 struct RAMState {
     /* QEMUFile used for this migration */
     QEMUFile *f;
+    /*
+     * PageSearchStatus structures for the channels when send pages.
+     * Protected by the bitmap_mutex.
+     */
+    PageSearchStatus pss[RAM_CHANNEL_MAX];
     /* UFFD file descriptor, used in 'write-tracking' migration */
     int uffdio_fd;
     /* Last block that we have visited searching for dirty pages */
@@ -362,7 +407,12 @@ struct RAMState {
     uint64_t target_page_count;
     /* number of dirty bits in the bitmap */
     uint64_t migration_dirty_pages;
-    /* Protects modification of the bitmap and migration dirty pages */
+    /*
+     * Protects:
+     * - dirty/clear bitmap
+     * - migration_dirty_pages
+     * - pss structures
+     */
     QemuMutex bitmap_mutex;
     /* The RAMBlock used in the last src_page_requests */
     RAMBlock *last_req_rb;
@@ -444,46 +494,6 @@ void dirty_sync_missed_zero_copy(void)
     ram_counters.dirty_sync_missed_zero_copy++;
 }
 
-/* used by the search for pages to send */
-struct PageSearchStatus {
-    /* The migration channel used for a specific host page */
-    QEMUFile    *pss_channel;
-    /* Current block being searched */
-    RAMBlock    *block;
-    /* Current page to search from */
-    unsigned long page;
-    /* Set once we wrap around */
-    bool         complete_round;
-    /*
-     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
-     * postcopy.  When set, the request is "urgent" because the dest QEMU
-     * threads are waiting for us.
-     */
-    bool         postcopy_requested;
-    /*
-     * [POSTCOPY-ONLY] The target channel to use to send current page.
-     *
-     * Note: This may _not_ match with the value in postcopy_requested
-     * above. Let's imagine the case where the postcopy request is exactly
-     * the page that we're sending in progress during precopy. In this case
-     * we'll have postcopy_requested set to true but the target channel
-     * will be the precopy channel (so that we don't split brain on that
-     * specific page since the precopy channel already contains partial of
-     * that page data).
-     *
-     * Besides that specific use case, postcopy_target_channel should
-     * always be equal to postcopy_requested, because by default we send
-     * postcopy pages via postcopy preempt channel.
-     */
-    bool         postcopy_target_channel;
-    /* Whether we're sending a host page */
-    bool          host_page_sending;
-    /* The start/end of current host page.  Only valid if host_page_sending==true */
-    unsigned long host_page_start;
-    unsigned long host_page_end;
-};
-typedef struct PageSearchStatus PageSearchStatus;
-
 CompressionStats compression_counters;
 
 struct CompressParam {
@@ -2627,7 +2637,7 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
  */
 static int ram_find_and_save_block(RAMState *rs)
 {
-    PageSearchStatus pss;
+    PageSearchStatus *pss = &rs->pss[RAM_CHANNEL_PRECOPY];
     int pages = 0;
     bool again, found;
 
@@ -2648,11 +2658,11 @@ static int ram_find_and_save_block(RAMState *rs)
         rs->last_page = 0;
     }
 
-    pss_init(&pss, rs->last_seen_block, rs->last_page);
+    pss_init(pss, rs->last_seen_block, rs->last_page);
 
     do {
         again = true;
-        found = get_queued_page(rs, &pss);
+        found = get_queued_page(rs, pss);
 
         if (!found) {
             /*
@@ -2660,27 +2670,27 @@ static int ram_find_and_save_block(RAMState *rs)
              * preempted precopy.  Otherwise find the next dirty bit.
              */
             if (postcopy_preempt_triggered(rs)) {
-                postcopy_preempt_restore(rs, &pss, false);
+                postcopy_preempt_restore(rs, pss, false);
                 found = true;
             } else {
                 /* priority queue empty, so just search for something dirty */
-                found = find_dirty_block(rs, &pss, &again);
+                found = find_dirty_block(rs, pss, &again);
             }
         }
 
         if (found) {
             /* Update rs->f with correct channel */
             if (postcopy_preempt_active()) {
-                postcopy_preempt_choose_channel(rs, &pss);
+                postcopy_preempt_choose_channel(rs, pss);
             }
             /* Cache rs->f in pss_channel (TODO: remove rs->f) */
-            pss.pss_channel = rs->f;
-            pages = ram_save_host_page(rs, &pss);
+            pss->pss_channel = rs->f;
+            pages = ram_save_host_page(rs, pss);
         }
     } while (!pages && again);
 
-    rs->last_seen_block = pss.block;
-    rs->last_page = pss.page;
+    rs->last_seen_block = pss->block;
+    rs->last_page = pss->page;
 
     return pages;
 }
Signature · 1 line
2.32.0
#28 lore raw

Re: [PATCH 10/14] migration: Make PageSearchStatus part of RAMState

Quoted 72 lines · * Peter Xu ([email protected]) wrote:

We used to allocate PSS structure on the stack for precopy when sending pages. Make it static, so as to describe per-channel ram migration status.

Here we declared RAM_CHANNEL_MAX instances, preparing for postcopy to use it, even though this patch has not yet to start using the 2nd instance.

This should not have any functional change per se, but it already starts to export PSS information via the RAMState, so that e.g. one PSS channel can start to reference the other PSS channel.

Always protect PSS access using the same RAMState.bitmap_mutex. We already do so, so no code change needed, just some comment update. Maybe we should consider renaming bitmap_mutex some day as it's going to be a more commonly and big mutex we use for ram states, but just leave it for later.

---
 migration/ram.c | 112 ++++++++++++++++++++++++++----------------------
 1 file changed, 61 insertions(+), 51 deletions(-)
migration/ram.c +40 -0
index b4b36ca59e..dbe11e1ace 100644
@@ -85,6 +85,46 @@
 
 XBZRLECacheStats xbzrle_counters;
 
+/* used by the search for pages to send */
+struct PageSearchStatus {
+    /* The migration channel used for a specific host page */
+    QEMUFile    *pss_channel;
+    /* Current block being searched */
+    RAMBlock    *block;
+    /* Current page to search from */
+    unsigned long page;
+    /* Set once we wrap around */
+    bool         complete_round;
+    /*
+     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
+     * postcopy.  When set, the request is "urgent" because the dest QEMU
+     * threads are waiting for us.
+     */
+    bool         postcopy_requested;
+    /*
+     * [POSTCOPY-ONLY] The target channel to use to send current page.
+     *
+     * Note: This may _not_ match with the value in postcopy_requested
+     * above. Let's imagine the case where the postcopy request is exactly
+     * the page that we're sending in progress during precopy. In this case
+     * we'll have postcopy_requested set to true but the target channel
+     * will be the precopy channel (so that we don't split brain on that
+     * specific page since the precopy channel already contains partial of
+     * that page data).
+     *
+     * Besides that specific use case, postcopy_target_channel should
+     * always be equal to postcopy_requested, because by default we send
+     * postcopy pages via postcopy preempt channel.
+     */
+    bool         postcopy_target_channel;
+    /* Whether we're sending a host page */
+    bool          host_page_sending;
+    /* The start/end of current host page.  Invalid if host_page_sending==false */
+    unsigned long host_page_start;
+    unsigned long host_page_end;
+};
+typedef struct PageSearchStatus PageSearchStatus;
+
 /* struct contains XBZRLE cache and a static page
    used by the compression */
 static struct {

migration/ram.c typedef struct { · line 366

Quoted 8 lines
migration/ram.c +5 -0
@@ -319,6 +359,11 @@ typedef struct {
 struct RAMState {
     /* QEMUFile used for this migration */
     QEMUFile *f;
+    /*
+     * PageSearchStatus structures for the channels when send pages.
+     * Protected by the bitmap_mutex.
+     */
+    PageSearchStatus pss[RAM_CHANNEL_MAX];

Why statically size this rather than allocate it in ram_state_init ?

Dave

Quoted 125 lines
migration/ram.c +16 -51
     /* UFFD file descriptor, used in 'write-tracking' migration */
     int uffdio_fd;
     /* Last block that we have visited searching for dirty pages */
@@ -362,7 +407,12 @@ struct RAMState {
     uint64_t target_page_count;
     /* number of dirty bits in the bitmap */
     uint64_t migration_dirty_pages;
-    /* Protects modification of the bitmap and migration dirty pages */
+    /*
+     * Protects:
+     * - dirty/clear bitmap
+     * - migration_dirty_pages
+     * - pss structures
+     */
     QemuMutex bitmap_mutex;
     /* The RAMBlock used in the last src_page_requests */
     RAMBlock *last_req_rb;
@@ -444,46 +494,6 @@ void dirty_sync_missed_zero_copy(void)
     ram_counters.dirty_sync_missed_zero_copy++;
 }
 
-/* used by the search for pages to send */
-struct PageSearchStatus {
-    /* The migration channel used for a specific host page */
-    QEMUFile    *pss_channel;
-    /* Current block being searched */
-    RAMBlock    *block;
-    /* Current page to search from */
-    unsigned long page;
-    /* Set once we wrap around */
-    bool         complete_round;
-    /*
-     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
-     * postcopy.  When set, the request is "urgent" because the dest QEMU
-     * threads are waiting for us.
-     */
-    bool         postcopy_requested;
-    /*
-     * [POSTCOPY-ONLY] The target channel to use to send current page.
-     *
-     * Note: This may _not_ match with the value in postcopy_requested
-     * above. Let's imagine the case where the postcopy request is exactly
-     * the page that we're sending in progress during precopy. In this case
-     * we'll have postcopy_requested set to true but the target channel
-     * will be the precopy channel (so that we don't split brain on that
-     * specific page since the precopy channel already contains partial of
-     * that page data).
-     *
-     * Besides that specific use case, postcopy_target_channel should
-     * always be equal to postcopy_requested, because by default we send
-     * postcopy pages via postcopy preempt channel.
-     */
-    bool         postcopy_target_channel;
-    /* Whether we're sending a host page */
-    bool          host_page_sending;
-    /* The start/end of current host page.  Only valid if host_page_sending==true */
-    unsigned long host_page_start;
-    unsigned long host_page_end;
-};
-typedef struct PageSearchStatus PageSearchStatus;
-
 CompressionStats compression_counters;
 
 struct CompressParam {
@@ -2627,7 +2637,7 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
  */
 static int ram_find_and_save_block(RAMState *rs)
 {
-    PageSearchStatus pss;
+    PageSearchStatus *pss = &rs->pss[RAM_CHANNEL_PRECOPY];
     int pages = 0;
     bool again, found;
 
@@ -2648,11 +2658,11 @@ static int ram_find_and_save_block(RAMState *rs)
         rs->last_page = 0;
     }
 
-    pss_init(&pss, rs->last_seen_block, rs->last_page);
+    pss_init(pss, rs->last_seen_block, rs->last_page);
 
     do {
         again = true;
-        found = get_queued_page(rs, &pss);
+        found = get_queued_page(rs, pss);
 
         if (!found) {
             /*
@@ -2660,27 +2670,27 @@ static int ram_find_and_save_block(RAMState *rs)
              * preempted precopy.  Otherwise find the next dirty bit.
              */
             if (postcopy_preempt_triggered(rs)) {
-                postcopy_preempt_restore(rs, &pss, false);
+                postcopy_preempt_restore(rs, pss, false);
                 found = true;
             } else {
                 /* priority queue empty, so just search for something dirty */
-                found = find_dirty_block(rs, &pss, &again);
+                found = find_dirty_block(rs, pss, &again);
             }
         }
 
         if (found) {
             /* Update rs->f with correct channel */
             if (postcopy_preempt_active()) {
-                postcopy_preempt_choose_channel(rs, &pss);
+                postcopy_preempt_choose_channel(rs, pss);
             }
             /* Cache rs->f in pss_channel (TODO: remove rs->f) */
-            pss.pss_channel = rs->f;
-            pages = ram_save_host_page(rs, &pss);
+            pss->pss_channel = rs->f;
+            pages = ram_save_host_page(rs, pss);
         }
     } while (!pages && again);
 
-    rs->last_seen_block = pss.block;
-    rs->last_page = pss.page;
+    rs->last_seen_block = pss->block;
+    rs->last_page = pss->page;
 
     return pages;
 }
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#29 lore raw

Re: [PATCH 10/14] migration: Make PageSearchStatus part of RAMState

Quoted 6 lines · Dr. David Alan Gilbert
Quoted 6 lines

/* struct contains XBZRLE cache and a static page
used by the compression */
static struct {

migration/ram.c +0 -0
@@ -319,6 +359,11 @@ typedef struct {
 struct RAMState {
     /* QEMUFile used for this migration */
Quoted 8 lines

migration/ram.c typedef struct { · line 366

Quoted 6 lines
migration/ram.c +5 -0
     QEMUFile *f;
+    /*
+     * PageSearchStatus structures for the channels when send pages.
+     * Protected by the bitmap_mutex.
+     */
+    PageSearchStatus pss[RAM_CHANNEL_MAX];

Why statically size this rather than allocate it in ram_state_init ?

I don't strongly feel like it needs the complexity? As there're only at most 2 channels anyway, so the best chance is we save ~56 bytes on src qemu but only during migration (RAMState allocated only in ram setup).

If you think we should still do the dynamic allcation, definitely doable on my side too.

Signature · 1 line
Peter Xu
#30 lore raw

Re: [PATCH 10/14] migration: Make PageSearchStatus part of RAMState

Quoted 16 lines · * Peter Xu ([email protected]) wrote:
Quoted 6 lines · Dr. David Alan Gilbert
Quoted 6 lines

/* struct contains XBZRLE cache and a static page
used by the compression */
static struct {

migration/ram.c +0 -0
@@ -319,6 +359,11 @@ typedef struct {
 struct RAMState {
     /* QEMUFile used for this migration */
Quoted 8 lines

migration/ram.c typedef struct { · line 366

Quoted 6 lines
migration/ram.c +5 -0
     QEMUFile *f;
+    /*
+     * PageSearchStatus structures for the channels when send pages.
+     * Protected by the bitmap_mutex.
+     */
+    PageSearchStatus pss[RAM_CHANNEL_MAX];

Why statically size this rather than allocate it in ram_state_init ?

Quoted 6 lines

I don't strongly feel like it needs the complexity? As there're only at most 2 channels anyway, so the best chance is we save ~56 bytes on src qemu but only during migration (RAMState allocated only in ram setup).

If you think we should still do the dynamic allcation, definitely doable on my side too.

Oh for 2 channels, yes that's fine - what confused me here was 'RAM_CHANNEL_...' - I forgot that was PRE vs POST - I was getting confused with multifd channel numbering; too many concepts of 'channel'.

Dave

Quoted 3 lines
Signature · 1 line
Peter Xu
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Reviewed-by on v1 10/14
#31 lore raw

Re: [PATCH 10/14] migration: Make PageSearchStatus part of RAMState

In reply to Peter Xu
Quoted 10 lines · * Peter Xu ([email protected]) wrote:

We used to allocate PSS structure on the stack for precopy when sending pages. Make it static, so as to describe per-channel ram migration status.

Here we declared RAM_CHANNEL_MAX instances, preparing for postcopy to use it, even though this patch has not yet to start using the 2nd instance.

This should not have any functional change per se, but it already starts to export PSS information via the RAMState, so that e.g. one PSS channel can start to reference the other PSS channel.

Quoted 6 lines

Always protect PSS access using the same RAMState.bitmap_mutex. We already do so, so no code change needed, just some comment update. Maybe we should consider renaming bitmap_mutex some day as it's going to be a more commonly and big mutex we use for ram states, but just leave it for later.

Quoted 189 lines
---
 migration/ram.c | 112 ++++++++++++++++++++++++++----------------------
 1 file changed, 61 insertions(+), 51 deletions(-)
migration/ram.c +61 -51
index b4b36ca59e..dbe11e1ace 100644
@@ -85,6 +85,46 @@
 
 XBZRLECacheStats xbzrle_counters;
 
+/* used by the search for pages to send */
+struct PageSearchStatus {
+    /* The migration channel used for a specific host page */
+    QEMUFile    *pss_channel;
+    /* Current block being searched */
+    RAMBlock    *block;
+    /* Current page to search from */
+    unsigned long page;
+    /* Set once we wrap around */
+    bool         complete_round;
+    /*
+     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
+     * postcopy.  When set, the request is "urgent" because the dest QEMU
+     * threads are waiting for us.
+     */
+    bool         postcopy_requested;
+    /*
+     * [POSTCOPY-ONLY] The target channel to use to send current page.
+     *
+     * Note: This may _not_ match with the value in postcopy_requested
+     * above. Let's imagine the case where the postcopy request is exactly
+     * the page that we're sending in progress during precopy. In this case
+     * we'll have postcopy_requested set to true but the target channel
+     * will be the precopy channel (so that we don't split brain on that
+     * specific page since the precopy channel already contains partial of
+     * that page data).
+     *
+     * Besides that specific use case, postcopy_target_channel should
+     * always be equal to postcopy_requested, because by default we send
+     * postcopy pages via postcopy preempt channel.
+     */
+    bool         postcopy_target_channel;
+    /* Whether we're sending a host page */
+    bool          host_page_sending;
+    /* The start/end of current host page.  Invalid if host_page_sending==false */
+    unsigned long host_page_start;
+    unsigned long host_page_end;
+};
+typedef struct PageSearchStatus PageSearchStatus;
+
 /* struct contains XBZRLE cache and a static page
    used by the compression */
 static struct {
@@ -319,6 +359,11 @@ typedef struct {
 struct RAMState {
     /* QEMUFile used for this migration */
     QEMUFile *f;
+    /*
+     * PageSearchStatus structures for the channels when send pages.
+     * Protected by the bitmap_mutex.
+     */
+    PageSearchStatus pss[RAM_CHANNEL_MAX];
     /* UFFD file descriptor, used in 'write-tracking' migration */
     int uffdio_fd;
     /* Last block that we have visited searching for dirty pages */
@@ -362,7 +407,12 @@ struct RAMState {
     uint64_t target_page_count;
     /* number of dirty bits in the bitmap */
     uint64_t migration_dirty_pages;
-    /* Protects modification of the bitmap and migration dirty pages */
+    /*
+     * Protects:
+     * - dirty/clear bitmap
+     * - migration_dirty_pages
+     * - pss structures
+     */
     QemuMutex bitmap_mutex;
     /* The RAMBlock used in the last src_page_requests */
     RAMBlock *last_req_rb;
@@ -444,46 +494,6 @@ void dirty_sync_missed_zero_copy(void)
     ram_counters.dirty_sync_missed_zero_copy++;
 }
 
-/* used by the search for pages to send */
-struct PageSearchStatus {
-    /* The migration channel used for a specific host page */
-    QEMUFile    *pss_channel;
-    /* Current block being searched */
-    RAMBlock    *block;
-    /* Current page to search from */
-    unsigned long page;
-    /* Set once we wrap around */
-    bool         complete_round;
-    /*
-     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
-     * postcopy.  When set, the request is "urgent" because the dest QEMU
-     * threads are waiting for us.
-     */
-    bool         postcopy_requested;
-    /*
-     * [POSTCOPY-ONLY] The target channel to use to send current page.
-     *
-     * Note: This may _not_ match with the value in postcopy_requested
-     * above. Let's imagine the case where the postcopy request is exactly
-     * the page that we're sending in progress during precopy. In this case
-     * we'll have postcopy_requested set to true but the target channel
-     * will be the precopy channel (so that we don't split brain on that
-     * specific page since the precopy channel already contains partial of
-     * that page data).
-     *
-     * Besides that specific use case, postcopy_target_channel should
-     * always be equal to postcopy_requested, because by default we send
-     * postcopy pages via postcopy preempt channel.
-     */
-    bool         postcopy_target_channel;
-    /* Whether we're sending a host page */
-    bool          host_page_sending;
-    /* The start/end of current host page.  Only valid if host_page_sending==true */
-    unsigned long host_page_start;
-    unsigned long host_page_end;
-};
-typedef struct PageSearchStatus PageSearchStatus;
-
 CompressionStats compression_counters;
 
 struct CompressParam {
@@ -2627,7 +2637,7 @@ static int ram_save_host_page(RAMState *rs, PageSearchStatus *pss)
  */
 static int ram_find_and_save_block(RAMState *rs)
 {
-    PageSearchStatus pss;
+    PageSearchStatus *pss = &rs->pss[RAM_CHANNEL_PRECOPY];
     int pages = 0;
     bool again, found;
 
@@ -2648,11 +2658,11 @@ static int ram_find_and_save_block(RAMState *rs)
         rs->last_page = 0;
     }
 
-    pss_init(&pss, rs->last_seen_block, rs->last_page);
+    pss_init(pss, rs->last_seen_block, rs->last_page);
 
     do {
         again = true;
-        found = get_queued_page(rs, &pss);
+        found = get_queued_page(rs, pss);
 
         if (!found) {
             /*
@@ -2660,27 +2670,27 @@ static int ram_find_and_save_block(RAMState *rs)
              * preempted precopy.  Otherwise find the next dirty bit.
              */
             if (postcopy_preempt_triggered(rs)) {
-                postcopy_preempt_restore(rs, &pss, false);
+                postcopy_preempt_restore(rs, pss, false);
                 found = true;
             } else {
                 /* priority queue empty, so just search for something dirty */
-                found = find_dirty_block(rs, &pss, &again);
+                found = find_dirty_block(rs, pss, &again);
             }
         }
 
         if (found) {
             /* Update rs->f with correct channel */
             if (postcopy_preempt_active()) {
-                postcopy_preempt_choose_channel(rs, &pss);
+                postcopy_preempt_choose_channel(rs, pss);
             }
             /* Cache rs->f in pss_channel (TODO: remove rs->f) */
-            pss.pss_channel = rs->f;
-            pages = ram_save_host_page(rs, &pss);
+            pss->pss_channel = rs->f;
+            pages = ram_save_host_page(rs, pss);
         }
     } while (!pages && again);
 
-    rs->last_seen_block = pss.block;
-    rs->last_page = pss.page;
+    rs->last_seen_block = pss->block;
+    rs->last_page = pss->page;
 
     return pages;
 }
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#32 lore raw

PATCH 11/14 migration: Move last_sent_block into PageSearchStatus

In reply to Peter Xu

Since we use PageSearchStatus to represent a channel, it makes perfect sense to keep last_sent_block (aka, leverage RAM_SAVE_FLAG_CONTINUE) to be per-channel rather than global because each channel can be sending different pages on ramblocks.

Hence move it from RAMState into PageSearchStatus.

---
 migration/ram.c | 71 ++++++++++++++++++++++++++++---------------------
 1 file changed, 41 insertions(+), 30 deletions(-)
migration/ram.c +41 -30
index dbe11e1ace..fdcb61a2c8 100644
@@ -89,6 +89,8 @@ XBZRLECacheStats xbzrle_counters;
 struct PageSearchStatus {
     /* The migration channel used for a specific host page */
     QEMUFile    *pss_channel;
+    /* Last block from where we have sent data */
+    RAMBlock *last_sent_block;
     /* Current block being searched */
     RAMBlock    *block;
     /* Current page to search from */
@@ -368,8 +370,6 @@ struct RAMState {
     int uffdio_fd;
     /* Last block that we have visited searching for dirty pages */
     RAMBlock *last_seen_block;
-    /* Last block from where we have sent data */
-    RAMBlock *last_sent_block;
     /* Last dirty target page we have sent */
     ram_addr_t last_page;
     /* last ram version we have seen */
@@ -677,16 +677,17 @@ exit:
  *
  * Returns the number of bytes written
  *
- * @f: QEMUFile where to send the data
+ * @pss: current PSS channel status
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  *          in the lower bits, it contains flags
  */
-static size_t save_page_header(RAMState *rs, QEMUFile *f,  RAMBlock *block,
+static size_t save_page_header(PageSearchStatus *pss, RAMBlock *block,
                                ram_addr_t offset)
 {
     size_t size, len;
-    bool same_block = (block == rs->last_sent_block);
+    bool same_block = (block == pss->last_sent_block);
+    QEMUFile *f = pss->pss_channel;
 
     if (same_block) {
         offset |= RAM_SAVE_FLAG_CONTINUE;
@@ -699,7 +700,7 @@ static size_t save_page_header(RAMState *rs, QEMUFile *f,  RAMBlock *block,
         qemu_put_byte(f, len);
         qemu_put_buffer(f, (uint8_t *)block->idstr, len);
         size += 1 + len;
-        rs->last_sent_block = block;
+        pss->last_sent_block = block;
     }
     return size;
 }
@@ -783,17 +784,19 @@ static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr)
  *          -1 means that xbzrle would be longer than normal
  *
  * @rs: current RAM state
+ * @pss: current PSS channel
  * @current_data: pointer to the address of the page contents
  * @current_addr: addr of the page
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_xbzrle_page(RAMState *rs, QEMUFile *file,
+static int save_xbzrle_page(RAMState *rs, PageSearchStatus *pss,
                             uint8_t **current_data, ram_addr_t current_addr,
                             RAMBlock *block, ram_addr_t offset)
 {
     int encoded_len = 0, bytes_xbzrle;
     uint8_t *prev_cached_page;
+    QEMUFile *file = pss->pss_channel;
 
     if (!cache_is_cached(XBZRLE.cache, current_addr,
                          ram_counters.dirty_sync_count)) {
@@ -858,7 +861,7 @@ static int save_xbzrle_page(RAMState *rs, QEMUFile *file,
     }
 
     /* Send XBZRLE based compressed page */
-    bytes_xbzrle = save_page_header(rs, file, block,
+    bytes_xbzrle = save_page_header(pss, block,
                                     offset | RAM_SAVE_FLAG_XBZRLE);
     qemu_put_byte(file, ENCODING_FLAG_XBZRLE);
     qemu_put_be16(file, encoded_len);
@@ -1289,19 +1292,19 @@ static void ram_release_page(const char *rbname, uint64_t offset)
  * Returns the size of data written to the file, 0 means the page is not
  * a zero page
  *
- * @rs: current RAM state
- * @file: the file where the data is saved
+ * @pss: current PSS channel
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_zero_page_to_file(RAMState *rs, QEMUFile *file,
+static int save_zero_page_to_file(PageSearchStatus *pss,
                                   RAMBlock *block, ram_addr_t offset)
 {
     uint8_t *p = block->host + offset;
+    QEMUFile *file = pss->pss_channel;
     int len = 0;
 
     if (buffer_is_zero(p, TARGET_PAGE_SIZE)) {
-        len += save_page_header(rs, file, block, offset | RAM_SAVE_FLAG_ZERO);
+        len += save_page_header(pss, block, offset | RAM_SAVE_FLAG_ZERO);
         qemu_put_byte(file, 0);
         len += 1;
         ram_release_page(block->idstr, offset);
@@ -1314,14 +1317,14 @@ static int save_zero_page_to_file(RAMState *rs, QEMUFile *file,
  *
  * Returns the number of pages written.
  *
- * @rs: current RAM state
+ * @pss: current PSS channel
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_zero_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+static int save_zero_page(PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(rs, file, block, offset);
+    int len = save_zero_page_to_file(pss, block, offset);
 
     if (len) {
         qatomic_inc(&ram_counters.duplicate);
@@ -1374,16 +1377,18 @@ static bool control_save_page(PageSearchStatus *pss, RAMBlock *block,
  *
  * Returns the number of pages written.
  *
- * @rs: current RAM state
+ * @pss: current PSS channel
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  * @buf: the page to be sent
  * @async: send to page asyncly
  */
-static int save_normal_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+static int save_normal_page(PageSearchStatus *pss, RAMBlock *block,
                             ram_addr_t offset, uint8_t *buf, bool async)
 {
-    ram_transferred_add(save_page_header(rs, file, block,
+    QEMUFile *file = pss->pss_channel;
+
+    ram_transferred_add(save_page_header(pss, block,
                                          offset | RAM_SAVE_FLAG_PAGE));
     if (async) {
         qemu_put_buffer_async(file, buf, TARGET_PAGE_SIZE,
@@ -1423,7 +1428,7 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     XBZRLE_cache_lock();
     if (rs->xbzrle_enabled && !migration_in_postcopy()) {
-        pages = save_xbzrle_page(rs, pss->pss_channel, &p, current_addr,
+        pages = save_xbzrle_page(rs, pss, &p, current_addr,
                                  block, offset);
         if (!rs->last_stage) {
             /* Can't send this cached data async, since the cache page
@@ -1435,8 +1440,7 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     /* XBZRLE overflow or normal page */
     if (pages == -1) {
-        pages = save_normal_page(rs, pss->pss_channel, block, offset,
-                                 p, send_async);
+        pages = save_normal_page(pss, block, offset, p, send_async);
     }
 
     XBZRLE_cache_unlock();
@@ -1459,14 +1463,15 @@ static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block,
                                  ram_addr_t offset, uint8_t *source_buf)
 {
     RAMState *rs = ram_state;
+    PageSearchStatus *pss = &rs->pss[RAM_CHANNEL_PRECOPY];
     uint8_t *p = block->host + offset;
     int ret;
 
-    if (save_zero_page_to_file(rs, f, block, offset)) {
+    if (save_zero_page_to_file(pss, block, offset)) {
         return true;
     }
 
-    save_page_header(rs, f, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE);
+    save_page_header(pss, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE);
 
     /*
      * copy it to a internal buffer to avoid it being modified by VM
@@ -2286,7 +2291,8 @@ static bool save_page_use_compression(RAMState *rs)
  * has been properly handled by compression, otherwise needs other
  * paths to handle it
  */
-static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
+static bool save_compress_page(RAMState *rs, PageSearchStatus *pss,
+                               RAMBlock *block, ram_addr_t offset)
 {
     if (!save_page_use_compression(rs)) {
         return false;
@@ -2302,7 +2308,7 @@ static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
      * We post the fist page as normal page as compression will take
      * much CPU resource.
      */
-    if (block != rs->last_sent_block) {
+    if (block != pss->last_sent_block) {
         flush_compressed_data(rs);
         return false;
     }
@@ -2333,11 +2339,11 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
         return res;
     }
 
-    if (save_compress_page(rs, block, offset)) {
+    if (save_compress_page(rs, pss, block, offset)) {
         return 1;
     }
 
-    res = save_zero_page(rs, pss->pss_channel, block, offset);
+    res = save_zero_page(pss, block, offset);
     if (res > 0) {
         /* Must let xbzrle know, otherwise a previous (now 0'd) cached
          * page would be stale
@@ -2469,7 +2475,7 @@ static void postcopy_preempt_choose_channel(RAMState *rs, PageSearchStatus *pss)
          * If channel switched, reset last_sent_block since the old sent block
          * may not be on the same channel.
          */
-        rs->last_sent_block = NULL;
+        pss->last_sent_block = NULL;
 
         trace_postcopy_preempt_switch_channel(channel);
     }
@@ -2804,8 +2810,13 @@ static void ram_save_cleanup(void *opaque)
 
 static void ram_state_reset(RAMState *rs)
 {
+    int i;
+
+    for (i = 0; i < RAM_CHANNEL_MAX; i++) {
+        rs->pss[i].last_sent_block = NULL;
+    }
+
     rs->last_seen_block = NULL;
-    rs->last_sent_block = NULL;
     rs->last_page = 0;
     rs->last_version = ram_list.version;
     rs->xbzrle_enabled = false;
@@ -2999,8 +3010,8 @@ void ram_postcopy_send_discard_bitmap(MigrationState *ms)
     migration_bitmap_sync(rs);
 
     /* Easiest way to make sure we don't resume in the middle of a host-page */
+    rs->pss[RAM_CHANNEL_PRECOPY].last_sent_block = NULL;
     rs->last_seen_block = NULL;
-    rs->last_sent_block = NULL;
     rs->last_page = 0;
 
     postcopy_each_ram_send_discard(ms);
Signature · 1 line
2.32.0
#33 lore raw

Re: [PATCH 11/14] migration: Move last_sent_block into PageSearchStatus

Quoted 253 lines · * Peter Xu ([email protected]) wrote:

Since we use PageSearchStatus to represent a channel, it makes perfect sense to keep last_sent_block (aka, leverage RAM_SAVE_FLAG_CONTINUE) to be per-channel rather than global because each channel can be sending different pages on ramblocks.

Hence move it from RAMState into PageSearchStatus.

---
 migration/ram.c | 71 ++++++++++++++++++++++++++++---------------------
 1 file changed, 41 insertions(+), 30 deletions(-)
migration/ram.c +40 -29
index dbe11e1ace..fdcb61a2c8 100644
@@ -89,6 +89,8 @@ XBZRLECacheStats xbzrle_counters;
 struct PageSearchStatus {
     /* The migration channel used for a specific host page */
     QEMUFile    *pss_channel;
+    /* Last block from where we have sent data */
+    RAMBlock *last_sent_block;
     /* Current block being searched */
     RAMBlock    *block;
     /* Current page to search from */
@@ -368,8 +370,6 @@ struct RAMState {
     int uffdio_fd;
     /* Last block that we have visited searching for dirty pages */
     RAMBlock *last_seen_block;
-    /* Last block from where we have sent data */
-    RAMBlock *last_sent_block;
     /* Last dirty target page we have sent */
     ram_addr_t last_page;
     /* last ram version we have seen */
@@ -677,16 +677,17 @@ exit:
  *
  * Returns the number of bytes written
  *
- * @f: QEMUFile where to send the data
+ * @pss: current PSS channel status
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  *          in the lower bits, it contains flags
  */
-static size_t save_page_header(RAMState *rs, QEMUFile *f,  RAMBlock *block,
+static size_t save_page_header(PageSearchStatus *pss, RAMBlock *block,
                                ram_addr_t offset)
 {
     size_t size, len;
-    bool same_block = (block == rs->last_sent_block);
+    bool same_block = (block == pss->last_sent_block);
+    QEMUFile *f = pss->pss_channel;
 
     if (same_block) {
         offset |= RAM_SAVE_FLAG_CONTINUE;
@@ -699,7 +700,7 @@ static size_t save_page_header(RAMState *rs, QEMUFile *f,  RAMBlock *block,
         qemu_put_byte(f, len);
         qemu_put_buffer(f, (uint8_t *)block->idstr, len);
         size += 1 + len;
-        rs->last_sent_block = block;
+        pss->last_sent_block = block;
     }
     return size;
 }
@@ -783,17 +784,19 @@ static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr)
  *          -1 means that xbzrle would be longer than normal
  *
  * @rs: current RAM state
+ * @pss: current PSS channel
  * @current_data: pointer to the address of the page contents
  * @current_addr: addr of the page
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_xbzrle_page(RAMState *rs, QEMUFile *file,
+static int save_xbzrle_page(RAMState *rs, PageSearchStatus *pss,
                             uint8_t **current_data, ram_addr_t current_addr,
                             RAMBlock *block, ram_addr_t offset)
 {
     int encoded_len = 0, bytes_xbzrle;
     uint8_t *prev_cached_page;
+    QEMUFile *file = pss->pss_channel;
 
     if (!cache_is_cached(XBZRLE.cache, current_addr,
                          ram_counters.dirty_sync_count)) {
@@ -858,7 +861,7 @@ static int save_xbzrle_page(RAMState *rs, QEMUFile *file,
     }
 
     /* Send XBZRLE based compressed page */
-    bytes_xbzrle = save_page_header(rs, file, block,
+    bytes_xbzrle = save_page_header(pss, block,
                                     offset | RAM_SAVE_FLAG_XBZRLE);
     qemu_put_byte(file, ENCODING_FLAG_XBZRLE);
     qemu_put_be16(file, encoded_len);
@@ -1289,19 +1292,19 @@ static void ram_release_page(const char *rbname, uint64_t offset)
  * Returns the size of data written to the file, 0 means the page is not
  * a zero page
  *
- * @rs: current RAM state
- * @file: the file where the data is saved
+ * @pss: current PSS channel
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_zero_page_to_file(RAMState *rs, QEMUFile *file,
+static int save_zero_page_to_file(PageSearchStatus *pss,
                                   RAMBlock *block, ram_addr_t offset)
 {
     uint8_t *p = block->host + offset;
+    QEMUFile *file = pss->pss_channel;
     int len = 0;
 
     if (buffer_is_zero(p, TARGET_PAGE_SIZE)) {
-        len += save_page_header(rs, file, block, offset | RAM_SAVE_FLAG_ZERO);
+        len += save_page_header(pss, block, offset | RAM_SAVE_FLAG_ZERO);
         qemu_put_byte(file, 0);
         len += 1;
         ram_release_page(block->idstr, offset);
@@ -1314,14 +1317,14 @@ static int save_zero_page_to_file(RAMState *rs, QEMUFile *file,
  *
  * Returns the number of pages written.
  *
- * @rs: current RAM state
+ * @pss: current PSS channel
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  */
-static int save_zero_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+static int save_zero_page(PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(rs, file, block, offset);
+    int len = save_zero_page_to_file(pss, block, offset);
 
     if (len) {
         qatomic_inc(&ram_counters.duplicate);
@@ -1374,16 +1377,18 @@ static bool control_save_page(PageSearchStatus *pss, RAMBlock *block,
  *
  * Returns the number of pages written.
  *
- * @rs: current RAM state
+ * @pss: current PSS channel
  * @block: block that contains the page we want to send
  * @offset: offset inside the block for the page
  * @buf: the page to be sent
  * @async: send to page asyncly
  */
-static int save_normal_page(RAMState *rs, QEMUFile *file, RAMBlock *block,
+static int save_normal_page(PageSearchStatus *pss, RAMBlock *block,
                             ram_addr_t offset, uint8_t *buf, bool async)
 {
-    ram_transferred_add(save_page_header(rs, file, block,
+    QEMUFile *file = pss->pss_channel;
+
+    ram_transferred_add(save_page_header(pss, block,
                                          offset | RAM_SAVE_FLAG_PAGE));
     if (async) {
         qemu_put_buffer_async(file, buf, TARGET_PAGE_SIZE,
@@ -1423,7 +1428,7 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     XBZRLE_cache_lock();
     if (rs->xbzrle_enabled && !migration_in_postcopy()) {
-        pages = save_xbzrle_page(rs, pss->pss_channel, &p, current_addr,
+        pages = save_xbzrle_page(rs, pss, &p, current_addr,
                                  block, offset);
         if (!rs->last_stage) {
             /* Can't send this cached data async, since the cache page
@@ -1435,8 +1440,7 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss)
 
     /* XBZRLE overflow or normal page */
     if (pages == -1) {
-        pages = save_normal_page(rs, pss->pss_channel, block, offset,
-                                 p, send_async);
+        pages = save_normal_page(pss, block, offset, p, send_async);
     }
 
     XBZRLE_cache_unlock();
@@ -1459,14 +1463,15 @@ static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block,
                                  ram_addr_t offset, uint8_t *source_buf)
 {
     RAMState *rs = ram_state;
+    PageSearchStatus *pss = &rs->pss[RAM_CHANNEL_PRECOPY];
     uint8_t *p = block->host + offset;
     int ret;
 
-    if (save_zero_page_to_file(rs, f, block, offset)) {
+    if (save_zero_page_to_file(pss, block, offset)) {
         return true;
     }
 
-    save_page_header(rs, f, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE);
+    save_page_header(pss, block, offset | RAM_SAVE_FLAG_COMPRESS_PAGE);
 
     /*
      * copy it to a internal buffer to avoid it being modified by VM
@@ -2286,7 +2291,8 @@ static bool save_page_use_compression(RAMState *rs)
  * has been properly handled by compression, otherwise needs other
  * paths to handle it
  */
-static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
+static bool save_compress_page(RAMState *rs, PageSearchStatus *pss,
+                               RAMBlock *block, ram_addr_t offset)
 {
     if (!save_page_use_compression(rs)) {
         return false;
@@ -2302,7 +2308,7 @@ static bool save_compress_page(RAMState *rs, RAMBlock *block, ram_addr_t offset)
      * We post the fist page as normal page as compression will take
      * much CPU resource.
      */
-    if (block != rs->last_sent_block) {
+    if (block != pss->last_sent_block) {
         flush_compressed_data(rs);
         return false;
     }
@@ -2333,11 +2339,11 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss)
         return res;
     }
 
-    if (save_compress_page(rs, block, offset)) {
+    if (save_compress_page(rs, pss, block, offset)) {
         return 1;
     }
 
-    res = save_zero_page(rs, pss->pss_channel, block, offset);
+    res = save_zero_page(pss, block, offset);
     if (res > 0) {
         /* Must let xbzrle know, otherwise a previous (now 0'd) cached
          * page would be stale
@@ -2469,7 +2475,7 @@ static void postcopy_preempt_choose_channel(RAMState *rs, PageSearchStatus *pss)
          * If channel switched, reset last_sent_block since the old sent block
          * may not be on the same channel.
          */
-        rs->last_sent_block = NULL;
+        pss->last_sent_block = NULL;
 
         trace_postcopy_preempt_switch_channel(channel);
     }
@@ -2804,8 +2810,13 @@ static void ram_save_cleanup(void *opaque)
 
 static void ram_state_reset(RAMState *rs)
 {
+    int i;
+
+    for (i = 0; i < RAM_CHANNEL_MAX; i++) {
+        rs->pss[i].last_sent_block = NULL;
+    }
+
     rs->last_seen_block = NULL;
-    rs->last_sent_block = NULL;
     rs->last_page = 0;
     rs->last_version = ram_list.version;
     rs->xbzrle_enabled = false;

migration/ram.c void ram_postcopy_send_discard_bitmap(MigrationState *ms) · line 3013

Quoted 4 lines
migration/ram.c +1 -0
@@ -2999,8 +3010,8 @@ void ram_postcopy_send_discard_bitmap(MigrationState *ms)
     migration_bitmap_sync(rs);
 
     /* Easiest way to make sure we don't resume in the middle of a host-page */
+    rs->pss[RAM_CHANNEL_PRECOPY].last_sent_block = NULL;

Why don't we reset the postcopy one here as well?

Dave

Quoted 8 lines
migration/ram.c +0 -1
     rs->last_seen_block = NULL;
-    rs->last_sent_block = NULL;
     rs->last_page = 0;
 
     postcopy_each_ram_send_discard(ms);
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#34 lore raw

Re: [PATCH 11/14] migration: Move last_sent_block into PageSearchStatus

Quoted 7 lines · Dr. David Alan Gilbert

migration/ram.c void ram_postcopy_send_discard_bitmap(MigrationState *ms) · line 3013

Quoted 5 lines
migration/ram.c +1 -0
@@ -2999,8 +3010,8 @@ void ram_postcopy_send_discard_bitmap(MigrationState *ms)
     migration_bitmap_sync(rs);
 
     /* Easiest way to make sure we don't resume in the middle of a host-page */
+    rs->pss[RAM_CHANNEL_PRECOPY].last_sent_block = NULL;

Why don't we reset the postcopy one here as well?

Because ram_postcopy_send_discard_bitmap() is only called before postcopy starts, so the other field should be NULL already. Thanks,

Signature · 1 line
Peter Xu
#35 lore raw

Re: [PATCH 11/14] migration: Move last_sent_block into PageSearchStatus

Quoted 11 lines · * Peter Xu ([email protected]) wrote:
Quoted 7 lines · Dr. David Alan Gilbert

migration/ram.c void ram_postcopy_send_discard_bitmap(MigrationState *ms) · line 3013

Quoted 5 lines
migration/ram.c +1 -0
@@ -2999,8 +3010,8 @@ void ram_postcopy_send_discard_bitmap(MigrationState *ms)
     migration_bitmap_sync(rs);
 
     /* Easiest way to make sure we don't resume in the middle of a host-page */
+    rs->pss[RAM_CHANNEL_PRECOPY].last_sent_block = NULL;

Why don't we reset the postcopy one here as well?

Because ram_postcopy_send_discard_bitmap() is only called before postcopy starts, so the other field should be NULL already. Thanks,

Ah, yes, OK

Quoted 3 lines
Signature · 1 line
Peter Xu
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#36 lore raw

PATCH 12/14 migration: Send requested page directly in rp-return thread

In reply to Peter Xu

With all the facilities ready, send the requested page directly in the rp-return thread rather than queuing it in the request queue, if and only if postcopy preempt is enabled. It can achieve so because it uses separate channel for sending urgent pages. The only shared data is bitmap and it's protected by the bitmap_mutex.

Note that since we're moving the ownership of the urgent channel from the migration thread to rp thread it also means the rp thread is responsible for managing the qemufile, e.g. properly close it when pausing migration happens. For this, let migration_release_from_dst_file to cover shutdown of the urgent channel too, renaming it as migration_release_dst_files() to better show what it does.

---
 migration/migration.c |  35 +++++++------
 migration/ram.c       | 112 ++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 131 insertions(+), 16 deletions(-)
migration/migration.c +19 -16
index 0eacc0c99b..fae8fd378b 100644
@@ -2845,8 +2845,11 @@ static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
     return 0;
 }
 
-/* Release ms->rp_state.from_dst_file in a safe way */
-static void migration_release_from_dst_file(MigrationState *ms)
+/*
+ * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
+ * existed) in a safe way.
+ */
+static void migration_release_dst_files(MigrationState *ms)
 {
     QEMUFile *file;
 
@@ -2859,6 +2862,18 @@ static void migration_release_from_dst_file(MigrationState *ms)
         ms->rp_state.from_dst_file = NULL;
     }
 
+    /*
+     * Do the same to postcopy fast path socket too if there is.  No
+     * locking needed because this qemufile should only be managed by
+     * return path thread.
+     */
+    if (ms->postcopy_qemufile_src) {
+        migration_ioc_unregister_yank_from_file(ms->postcopy_qemufile_src);
+        qemu_file_shutdown(ms->postcopy_qemufile_src);
+        qemu_fclose(ms->postcopy_qemufile_src);
+        ms->postcopy_qemufile_src = NULL;
+    }
+
     qemu_fclose(file);
 }
 
@@ -3003,7 +3018,7 @@ out:
              * Maybe there is something we can do: it looks like a
              * network down issue, and we pause for a recovery.
              */
-            migration_release_from_dst_file(ms);
+            migration_release_dst_files(ms);
             rp = NULL;
             if (postcopy_pause_return_path_thread(ms)) {
                 /*
@@ -3021,7 +3036,7 @@ out:
     }
 
     trace_source_return_path_thread_end();
-    migration_release_from_dst_file(ms);
+    migration_release_dst_files(ms);
     rcu_unregister_thread();
     return NULL;
 }
@@ -3544,18 +3559,6 @@ static MigThrError postcopy_pause(MigrationState *s)
         qemu_file_shutdown(file);
         qemu_fclose(file);
 
-        /*
-         * Do the same to postcopy fast path socket too if there is.  No
-         * locking needed because no racer as long as we do this before setting
-         * status to paused.
-         */
-        if (s->postcopy_qemufile_src) {
-            migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src);
-            qemu_file_shutdown(s->postcopy_qemufile_src);
-            qemu_fclose(s->postcopy_qemufile_src);
-            s->postcopy_qemufile_src = NULL;
-        }
-
         migrate_set_state(&s->state, s->state,
                           MIGRATION_STATUS_POSTCOPY_PAUSED);
 
migration/ram.c +112 -0

136 lines not on this page · raw message

Signature · 1 line
2.32.0
Reviewed-by on v1 12/14
#37 lore raw

Re: [PATCH 12/14] migration: Send requested page directly in rp-return thread

Quoted 6 lines · * Peter Xu ([email protected]) wrote:

With all the facilities ready, send the requested page directly in the rp-return thread rather than queuing it in the request queue, if and only if postcopy preempt is enabled. It can achieve so because it uses separate channel for sending urgent pages. The only shared data is bitmap and it's protected by the bitmap_mutex.

Quoted 8 lines

Note that since we're moving the ownership of the urgent channel from the migration thread to rp thread it also means the rp thread is responsible for managing the qemufile, e.g. properly close it when pausing migration happens. For this, let migration_release_from_dst_file to cover shutdown of the urgent channel too, renaming it as migration_release_dst_files() to better show what it does.

Yes, getting a bit complex, but I think OK.

Quoted 226 lines
---
 migration/migration.c |  35 +++++++------
 migration/ram.c       | 112 ++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 131 insertions(+), 16 deletions(-)
migration/migration.c +19 -16
index 0eacc0c99b..fae8fd378b 100644
@@ -2845,8 +2845,11 @@ static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
     return 0;
 }
 
-/* Release ms->rp_state.from_dst_file in a safe way */
-static void migration_release_from_dst_file(MigrationState *ms)
+/*
+ * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
+ * existed) in a safe way.
+ */
+static void migration_release_dst_files(MigrationState *ms)
 {
     QEMUFile *file;
 
@@ -2859,6 +2862,18 @@ static void migration_release_from_dst_file(MigrationState *ms)
         ms->rp_state.from_dst_file = NULL;
     }
 
+    /*
+     * Do the same to postcopy fast path socket too if there is.  No
+     * locking needed because this qemufile should only be managed by
+     * return path thread.
+     */
+    if (ms->postcopy_qemufile_src) {
+        migration_ioc_unregister_yank_from_file(ms->postcopy_qemufile_src);
+        qemu_file_shutdown(ms->postcopy_qemufile_src);
+        qemu_fclose(ms->postcopy_qemufile_src);
+        ms->postcopy_qemufile_src = NULL;
+    }
+
     qemu_fclose(file);
 }
 
@@ -3003,7 +3018,7 @@ out:
              * Maybe there is something we can do: it looks like a
              * network down issue, and we pause for a recovery.
              */
-            migration_release_from_dst_file(ms);
+            migration_release_dst_files(ms);
             rp = NULL;
             if (postcopy_pause_return_path_thread(ms)) {
                 /*
@@ -3021,7 +3036,7 @@ out:
     }
 
     trace_source_return_path_thread_end();
-    migration_release_from_dst_file(ms);
+    migration_release_dst_files(ms);
     rcu_unregister_thread();
     return NULL;
 }
@@ -3544,18 +3559,6 @@ static MigThrError postcopy_pause(MigrationState *s)
         qemu_file_shutdown(file);
         qemu_fclose(file);
 
-        /*
-         * Do the same to postcopy fast path socket too if there is.  No
-         * locking needed because no racer as long as we do this before setting
-         * status to paused.
-         */
-        if (s->postcopy_qemufile_src) {
-            migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src);
-            qemu_file_shutdown(s->postcopy_qemufile_src);
-            qemu_fclose(s->postcopy_qemufile_src);
-            s->postcopy_qemufile_src = NULL;
-        }
-
         migrate_set_state(&s->state, s->state,
                           MIGRATION_STATUS_POSTCOPY_PAUSED);
 
migration/ram.c +112 -0
index fdcb61a2c8..fd301d793c 100644
@@ -539,6 +539,8 @@ static QemuThread *decompress_threads;
 static QemuMutex decomp_done_lock;
 static QemuCond decomp_done_cond;
 
+static int ram_save_host_page_urgent(PageSearchStatus *pss);
+
 static bool do_compress_ram_page(QEMUFile *f, z_stream *stream, RAMBlock *block,
                                  ram_addr_t offset, uint8_t *source_buf);
 
@@ -553,6 +555,16 @@ static void pss_init(PageSearchStatus *pss, RAMBlock *rb, ram_addr_t page)
     pss->complete_round = false;
 }
 
+/*
+ * Check whether two PSSs are actively sending the same page.  Return true
+ * if it is, false otherwise.
+ */
+static bool pss_overlap(PageSearchStatus *pss1, PageSearchStatus *pss2)
+{
+    return pss1->host_page_sending && pss2->host_page_sending &&
+        (pss1->host_page_start == pss2->host_page_start);
+}
+
 static void *do_data_compress(void *opaque)
 {
     CompressParam *param = opaque;
@@ -2253,6 +2265,57 @@ int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len)
         return -1;
     }
 
+    /*
+     * When with postcopy preempt, we send back the page directly in the
+     * rp-return thread.
+     */
+    if (postcopy_preempt_active()) {
+        ram_addr_t page_start = start >> TARGET_PAGE_BITS;
+        size_t page_size = qemu_ram_pagesize(ramblock);
+        PageSearchStatus *pss = &ram_state->pss[RAM_CHANNEL_POSTCOPY];
+        int ret = 0;
+
+        qemu_mutex_lock(&rs->bitmap_mutex);
+
+        pss_init(pss, ramblock, page_start);
+        /*
+         * Always use the preempt channel, and make sure it's there.  It's
+         * safe to access without lock, because when rp-thread is running
+         * we should be the only one who operates on the qemufile
+         */
+        pss->pss_channel = migrate_get_current()->postcopy_qemufile_src;
+        pss->postcopy_requested = true;
+        assert(pss->pss_channel);
+
+        /*
+         * It must be either one or multiple of host page size.  Just
+         * assert; if something wrong we're mostly split brain anyway.
+         */
+        assert(len % page_size == 0);
+        while (len) {
+            if (ram_save_host_page_urgent(pss)) {
+                error_report("%s: ram_save_host_page_urgent() failed: "
+                             "ramblock=%s, start_addr=0x"RAM_ADDR_FMT,
+                             __func__, ramblock->idstr, start);
+                ret = -1;
+                break;
+            }
+            /*
+             * NOTE: after ram_save_host_page_urgent() succeeded, pss->page
+             * will automatically be moved and point to the next host page
+             * we're going to send, so no need to update here.
+             *
+             * Normally QEMU never sends >1 host page in requests, so
+             * logically we don't even need that as the loop should only
+             * run once, but just to be consistent.
+             */
+            len -= page_size;
+        };
+        qemu_mutex_unlock(&rs->bitmap_mutex);
+
+        return ret;
+    }
+
     struct RAMSrcPageRequest *new_entry =
         g_new0(struct RAMSrcPageRequest, 1);
     new_entry->rb = ramblock;
@@ -2531,6 +2594,55 @@ static void pss_host_page_finish(PageSearchStatus *pss)
     pss->host_page_start = pss->host_page_end = 0;
 }
 
+/*
+ * Send an urgent host page specified by `pss'.  Need to be called with
+ * bitmap_mutex held.
+ *
+ * Returns 0 if save host page succeeded, false otherwise.
+ */
+static int ram_save_host_page_urgent(PageSearchStatus *pss)
+{
+    bool page_dirty, sent = false;
+    RAMState *rs = ram_state;
+    int ret = 0;
+
+    trace_postcopy_preempt_send_host_page(pss->block->idstr, pss->page);
+    pss_host_page_prepare(pss);
+
+    /*
+     * If precopy is sending the same page, let it be done in precopy, or
+     * we could send the same page in two channels and none of them will
+     * receive the whole page.
+     */
+    if (pss_overlap(pss, &ram_state->pss[RAM_CHANNEL_PRECOPY])) {
+        trace_postcopy_preempt_hit(pss->block->idstr,
+                                   pss->page << TARGET_PAGE_BITS);
+        return 0;
+    }
+
+    do {
+        page_dirty = migration_bitmap_clear_dirty(rs, pss->block, pss->page);
+
+        if (page_dirty) {
+            /* Be strict to return code; it must be 1, or what else? */
+            if (ram_save_target_page(rs, pss) != 1) {
+                error_report_once("%s: ram_save_target_page failed", __func__);
+                ret = -1;
+                goto out;
+            }
+            sent = true;
+        }
+        pss_find_next_dirty(pss);
+    } while (pss_within_range(pss));
+out:
+    pss_host_page_finish(pss);
+    /* For urgent requests, flush immediately if sent */
+    if (sent) {
+        qemu_fflush(pss->pss_channel);
+    }
+    return ret;
+}
+
 /**
  * ram_save_host_page: save a whole host page
  *
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#38 lore raw

PATCH 13/14 migration: Remove old preempt code around state maintainance

In reply to Peter Xu

With the new code to send pages in rp-return thread, there's little help to keep lots of the old code on maintaining the preempt state in migration thread, because the new way should always be faster..

Then if we'll always send pages in the rp-return thread anyway, we don't need those logic to maintain preempt state anymore because now we serialize things using the mutex directly instead of using those fields.

It's very unfortunate to have those code for a short period, but that's still one intermediate step that we noticed the next bottleneck on the migration thread. Now what we can do best is to drop unnecessary code as long as the new code is stable to reduce the burden. It's actually a good thing because the new "sending page in rp-return thread" model is (IMHO) even cleaner and with better performance.

Remove the old code that was responsible for maintaining preempt states, at the meantime also remove x-postcopy-preempt-break-huge parameter because with concurrent sender threads we don't really need to break-huge anymore.

---
 migration/migration.c |   2 -
 migration/ram.c       | 258 +-----------------------------------------
 2 files changed, 3 insertions(+), 257 deletions(-)
migration/migration.c +0 -2
index fae8fd378b..698fd94591 100644
@@ -4399,8 +4399,6 @@ static Property migration_properties[] = {
     DEFINE_PROP_SIZE("announce-step", MigrationState,
                       parameters.announce_step,
                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
-    DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState,
-                      postcopy_preempt_break_huge, true),
     DEFINE_PROP_STRING("tls-creds", MigrationState, parameters.tls_creds),
     DEFINE_PROP_STRING("tls-hostname", MigrationState, parameters.tls_hostname),
     DEFINE_PROP_STRING("tls-authz", MigrationState, parameters.tls_authz),
migration/ram.c +3 -255

339 lines not on this page · raw message

Signature · 1 line
2.32.0
author
#39 lore raw

Re: [PATCH 13/14] migration: Remove old preempt code around state maintainance

Quoted 30 lines · Peter Xu

With the new code to send pages in rp-return thread, there's little help to keep lots of the old code on maintaining the preempt state in migration thread, because the new way should always be faster..

Then if we'll always send pages in the rp-return thread anyway, we don't need those logic to maintain preempt state anymore because now we serialize things using the mutex directly instead of using those fields.

It's very unfortunate to have those code for a short period, but that's still one intermediate step that we noticed the next bottleneck on the migration thread. Now what we can do best is to drop unnecessary code as long as the new code is stable to reduce the burden. It's actually a good thing because the new "sending page in rp-return thread" model is (IMHO) even cleaner and with better performance.

Remove the old code that was responsible for maintaining preempt states, at the meantime also remove x-postcopy-preempt-break-huge parameter because with concurrent sender threads we don't really need to break-huge anymore.

---
 migration/migration.c |   2 -
 migration/ram.c       | 258 +-----------------------------------------
 2 files changed, 3 insertions(+), 257 deletions(-)
migration/migration.c +0 -0
index fae8fd378b..698fd94591 100644

migration/migration.c static Property migration_properties[] = { · line 4403

Quoted 5 lines
migration/migration.c +0 -2
@@ -4399,8 +4399,6 @@ static Property migration_properties[] = {
     DEFINE_PROP_SIZE("announce-step", MigrationState,
                       parameters.announce_step,
                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
-    DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState,
-                      postcopy_preempt_break_huge, true),

Forgot to drop the variable altogether:

migration/migration.h +0 -7
index cdad8aceaa..ae4ffd3454 100644
@@ -340,13 +340,6 @@ struct MigrationState {
     bool send_configuration;
     /* Whether we send section footer during migration */
     bool send_section_footer;
-    /*
-     * Whether we allow break sending huge pages when postcopy preempt is
-     * enabled.  When disabled, we won't interrupt precopy within sending a
-     * host huge page, which is the old behavior of vanilla postcopy.
-     * NOTE: this parameter is ignored if postcopy preempt is not enabled.
-     */
-    bool postcopy_preempt_break_huge;
 
     /* Needed by postcopy-pause state */
     QemuSemaphore postcopy_pause_sem;

Will squash this in in next version.

Signature · 1 line
Peter Xu
author
#40 lore raw

Re: [PATCH 13/14] migration: Remove old preempt code around state maintainance

Quoted 58 lines · Peter Xu
Quoted 30 lines · Peter Xu

With the new code to send pages in rp-return thread, there's little help to keep lots of the old code on maintaining the preempt state in migration thread, because the new way should always be faster..

Then if we'll always send pages in the rp-return thread anyway, we don't need those logic to maintain preempt state anymore because now we serialize things using the mutex directly instead of using those fields.

It's very unfortunate to have those code for a short period, but that's still one intermediate step that we noticed the next bottleneck on the migration thread. Now what we can do best is to drop unnecessary code as long as the new code is stable to reduce the burden. It's actually a good thing because the new "sending page in rp-return thread" model is (IMHO) even cleaner and with better performance.

Remove the old code that was responsible for maintaining preempt states, at the meantime also remove x-postcopy-preempt-break-huge parameter because with concurrent sender threads we don't really need to break-huge anymore.

---
 migration/migration.c |   2 -
 migration/ram.c       | 258 +-----------------------------------------
 2 files changed, 3 insertions(+), 257 deletions(-)
migration/migration.c +0 -0
index fae8fd378b..698fd94591 100644

migration/migration.c static Property migration_properties[] = { · line 4403

Quoted 5 lines
migration/migration.c +0 -2
@@ -4399,8 +4399,6 @@ static Property migration_properties[] = {
     DEFINE_PROP_SIZE("announce-step", MigrationState,
                       parameters.announce_step,
                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
-    DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState,
-                      postcopy_preempt_break_huge, true),

Forgot to drop the variable altogether:

migration/migration.h +0 -7
index cdad8aceaa..ae4ffd3454 100644
@@ -340,13 +340,6 @@ struct MigrationState {
     bool send_configuration;
     /* Whether we send section footer during migration */
     bool send_section_footer;
-    /*
-     * Whether we allow break sending huge pages when postcopy preempt is
-     * enabled.  When disabled, we won't interrupt precopy within sending a
-     * host huge page, which is the old behavior of vanilla postcopy.
-     * NOTE: this parameter is ignored if postcopy preempt is not enabled.
-     */
-    bool postcopy_preempt_break_huge;
 
     /* Needed by postcopy-pause state */
     QemuSemaphore postcopy_pause_sem;
Quoted 1 line

Will squash this in in next version.

Two more varialbes to drop, as attached..

Signature · 1 line
Peter Xu

From b3308e34398e21c19bd36ec21aae9c7f9f623d75 Mon Sep 17 00:00:00 2001
From: Peter Xu <[email protected]>
Date: Wed, 21 Sep 2022 09:51:55 -0400
Subject: [PATCH] fixup! migration: Remove old preempt code around state
maintainance
Content-type: text/plain

---
 migration/ram.c | 33 ---------------------------------
 1 file changed, 33 deletions(-)
migration/ram.c +0 -33
index 03bf2324ab..2599eee070 100644
@@ -97,28 +97,6 @@ struct PageSearchStatus {
     unsigned long page;
     /* Set once we wrap around */
     bool         complete_round;
-    /*
-     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
-     * postcopy.  When set, the request is "urgent" because the dest QEMU
-     * threads are waiting for us.
-     */
-    bool         postcopy_requested;
-    /*
-     * [POSTCOPY-ONLY] The target channel to use to send current page.
-     *
-     * Note: This may _not_ match with the value in postcopy_requested
-     * above. Let's imagine the case where the postcopy request is exactly
-     * the page that we're sending in progress during precopy. In this case
-     * we'll have postcopy_requested set to true but the target channel
-     * will be the precopy channel (so that we don't split brain on that
-     * specific page since the precopy channel already contains partial of
-     * that page data).
-     *
-     * Besides that specific use case, postcopy_target_channel should
-     * always be equal to postcopy_requested, because by default we send
-     * postcopy pages via postcopy preempt channel.
-     */
-    bool         postcopy_target_channel;
     /* Whether we're sending a host page */
     bool          host_page_sending;
     /* The start/end of current host page.  Invalid if host_page_sending==false */
@@ -1573,13 +1551,6 @@ retry:
  */
 static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again)
 {
-    /*
-     * This is not a postcopy requested page, mark it "not urgent", and use
-     * precopy channel to send it.
-     */
-    pss->postcopy_requested = false;
-    pss->postcopy_target_channel = RAM_CHANNEL_PRECOPY;
-
     /* Update pss->page for the next dirty bit in ramblock */
     pss_find_next_dirty(pss);
 
@@ -2091,9 +2062,6 @@ static bool get_queued_page(RAMState *rs, PageSearchStatus *pss)
          * really rare.
          */
         pss->complete_round = false;
-        /* Mark it an urgent request, meanwhile using POSTCOPY channel */
-        pss->postcopy_requested = true;
-        pss->postcopy_target_channel = RAM_CHANNEL_POSTCOPY;
     }
 
     return !!block;
@@ -2190,7 +2158,6 @@ int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len)
          * we should be the only one who operates on the qemufile
          */
         pss->pss_channel = migrate_get_current()->postcopy_qemufile_src;
-        pss->postcopy_requested = true;
         assert(pss->pss_channel);
 
         /*
Signature · 1 line
2.32.0
#41 lore raw

Re: [PATCH 13/14] migration: Remove old preempt code around state maintainance

For the set of 3:

Quoted 147 lines · * Peter Xu ([email protected]) wrote:
Quoted 58 lines · Peter Xu
Quoted 30 lines · Peter Xu

With the new code to send pages in rp-return thread, there's little help to keep lots of the old code on maintaining the preempt state in migration thread, because the new way should always be faster..

Then if we'll always send pages in the rp-return thread anyway, we don't need those logic to maintain preempt state anymore because now we serialize things using the mutex directly instead of using those fields.

It's very unfortunate to have those code for a short period, but that's still one intermediate step that we noticed the next bottleneck on the migration thread. Now what we can do best is to drop unnecessary code as long as the new code is stable to reduce the burden. It's actually a good thing because the new "sending page in rp-return thread" model is (IMHO) even cleaner and with better performance.

Remove the old code that was responsible for maintaining preempt states, at the meantime also remove x-postcopy-preempt-break-huge parameter because with concurrent sender threads we don't really need to break-huge anymore.

---
 migration/migration.c |   2 -
 migration/ram.c       | 258 +-----------------------------------------
 2 files changed, 3 insertions(+), 257 deletions(-)
migration/migration.c +0 -0
index fae8fd378b..698fd94591 100644

migration/migration.c static Property migration_properties[] = { · line 4403

Quoted 5 lines
migration/migration.c +0 -2
@@ -4399,8 +4399,6 @@ static Property migration_properties[] = {
     DEFINE_PROP_SIZE("announce-step", MigrationState,
                       parameters.announce_step,
                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
-    DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState,
-                      postcopy_preempt_break_huge, true),

Forgot to drop the variable altogether:

migration/migration.h +0 -7
index cdad8aceaa..ae4ffd3454 100644
@@ -340,13 +340,6 @@ struct MigrationState {
     bool send_configuration;
     /* Whether we send section footer during migration */
     bool send_section_footer;
-    /*
-     * Whether we allow break sending huge pages when postcopy preempt is
-     * enabled.  When disabled, we won't interrupt precopy within sending a
-     * host huge page, which is the old behavior of vanilla postcopy.
-     * NOTE: this parameter is ignored if postcopy preempt is not enabled.
-     */
-    bool postcopy_preempt_break_huge;
 
     /* Needed by postcopy-pause state */
     QemuSemaphore postcopy_pause_sem;
Quoted 1 line

Will squash this in in next version.

Two more varialbes to drop, as attached..

Signature · 13 lines
Peter Xu

From b3308e34398e21c19bd36ec21aae9c7f9f623d75 Mon Sep 17 00:00:00 2001
From: Peter Xu <[email protected]>
Date: Wed, 21 Sep 2022 09:51:55 -0400
Subject: [PATCH] fixup! migration: Remove old preempt code around state
 maintainance
Content-type: text/plain

Signed-off-by: Peter Xu <[email protected]>
---
 migration/ram.c | 33 ---------------------------------
 1 file changed, 33 deletions(-)
migration/ram.c +0 -33
index 03bf2324ab..2599eee070 100644
@@ -97,28 +97,6 @@ struct PageSearchStatus {
     unsigned long page;
     /* Set once we wrap around */
     bool         complete_round;
-    /*
-     * [POSTCOPY-ONLY] Whether current page is explicitly requested by
-     * postcopy.  When set, the request is "urgent" because the dest QEMU
-     * threads are waiting for us.
-     */
-    bool         postcopy_requested;
-    /*
-     * [POSTCOPY-ONLY] The target channel to use to send current page.
-     *
-     * Note: This may _not_ match with the value in postcopy_requested
-     * above. Let's imagine the case where the postcopy request is exactly
-     * the page that we're sending in progress during precopy. In this case
-     * we'll have postcopy_requested set to true but the target channel
-     * will be the precopy channel (so that we don't split brain on that
-     * specific page since the precopy channel already contains partial of
-     * that page data).
-     *
-     * Besides that specific use case, postcopy_target_channel should
-     * always be equal to postcopy_requested, because by default we send
-     * postcopy pages via postcopy preempt channel.
-     */
-    bool         postcopy_target_channel;
     /* Whether we're sending a host page */
     bool          host_page_sending;
     /* The start/end of current host page.  Invalid if host_page_sending==false */
@@ -1573,13 +1551,6 @@ retry:
  */
 static bool find_dirty_block(RAMState *rs, PageSearchStatus *pss, bool *again)
 {
-    /*
-     * This is not a postcopy requested page, mark it "not urgent", and use
-     * precopy channel to send it.
-     */
-    pss->postcopy_requested = false;
-    pss->postcopy_target_channel = RAM_CHANNEL_PRECOPY;
-
     /* Update pss->page for the next dirty bit in ramblock */
     pss_find_next_dirty(pss);
 
@@ -2091,9 +2062,6 @@ static bool get_queued_page(RAMState *rs, PageSearchStatus *pss)
          * really rare.
          */
         pss->complete_round = false;
-        /* Mark it an urgent request, meanwhile using POSTCOPY channel */
-        pss->postcopy_requested = true;
-        pss->postcopy_target_channel = RAM_CHANNEL_POSTCOPY;
     }
 
     return !!block;
@@ -2190,7 +2158,6 @@ int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len)
          * we should be the only one who operates on the qemufile
          */
         pss->pss_channel = migrate_get_current()->postcopy_qemufile_src;
-        pss->postcopy_requested = true;
         assert(pss->pss_channel);
 
         /*
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK
author
#42 lore raw

PATCH 14/14 migration: Drop rs->f

In reply to Peter Xu

Now with rs->pss we can already cache channels in pss->pss_channels. That pss_channel contains more infromation than rs->f because it's per-channel. So rs->f could be replaced by rss->pss[RAM_CHANNEL_PRECOPY].pss_channel, while rs->f itself is a bit vague now.

Note that vanilla postcopy still send pages via pss[RAM_CHANNEL_PRECOPY], that's slightly confusing but it reflects the reality.

Then, after the replacement we can safely drop rs->f.

---
 migration/ram.c | 12 ++++--------
 1 file changed, 4 insertions(+), 8 deletions(-)
migration/ram.c +4 -8
index f42efe02fc..03bf2324ab 100644
@@ -345,8 +345,6 @@ struct RAMSrcPageRequest {
 
 /* State of RAM for migration */
 struct RAMState {
-    /* QEMUFile used for this migration */
-    QEMUFile *f;
     /*
      * PageSearchStatus structures for the channels when send pages.
      * Protected by the bitmap_mutex.
@@ -2555,8 +2553,6 @@ static int ram_find_and_save_block(RAMState *rs)
         }
 
         if (found) {
-            /* Cache rs->f in pss_channel (TODO: remove rs->f) */
-            pss->pss_channel = rs->f;
             pages = ram_save_host_page(rs, pss);
         }
     } while (!pages && again);
@@ -3112,7 +3108,7 @@ static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out)
     ram_state_reset(rs);
 
     /* Update RAMState cache of output QEMUFile */
-    rs->f = out;
+    rs->pss[RAM_CHANNEL_PRECOPY].pss_channel = out;
 
     trace_ram_state_resume_prepare(pages);
 }
@@ -3203,7 +3199,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
             return -1;
         }
     }
-    (*rsp)->f = f;
+    (*rsp)->pss[RAM_CHANNEL_PRECOPY].pss_channel = f;
 
     WITH_RCU_READ_LOCK_GUARD() {
         qemu_put_be64(f, ram_bytes_total_common(true) | RAM_SAVE_FLAG_MEM_SIZE);
@@ -3338,7 +3334,7 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
 out:
     if (ret >= 0
         && migration_is_setup_or_active(migrate_get_current()->state)) {
-        ret = multifd_send_sync_main(rs->f);
+        ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel);
         if (ret < 0) {
             return ret;
         }
@@ -3406,7 +3402,7 @@ static int ram_save_complete(QEMUFile *f, void *opaque)
         return ret;
     }
 
-    ret = multifd_send_sync_main(rs->f);
+    ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel);
     if (ret < 0) {
         return ret;
     }
Signature · 1 line
2.32.0
Reviewed-by on v1 14/14
#43 lore raw

Re: [PATCH 14/14] migration: Drop rs->f

Quoted 11 lines · * Peter Xu ([email protected]) wrote:

Now with rs->pss we can already cache channels in pss->pss_channels. That pss_channel contains more infromation than rs->f because it's per-channel. So rs->f could be replaced by rss->pss[RAM_CHANNEL_PRECOPY].pss_channel, while rs->f itself is a bit vague now.

Note that vanilla postcopy still send pages via pss[RAM_CHANNEL_PRECOPY], that's slightly confusing but it reflects the reality.

Then, after the replacement we can safely drop rs->f.

Quoted 65 lines
---
 migration/ram.c | 12 ++++--------
 1 file changed, 4 insertions(+), 8 deletions(-)
migration/ram.c +4 -8
index f42efe02fc..03bf2324ab 100644
@@ -345,8 +345,6 @@ struct RAMSrcPageRequest {
 
 /* State of RAM for migration */
 struct RAMState {
-    /* QEMUFile used for this migration */
-    QEMUFile *f;
     /*
      * PageSearchStatus structures for the channels when send pages.
      * Protected by the bitmap_mutex.
@@ -2555,8 +2553,6 @@ static int ram_find_and_save_block(RAMState *rs)
         }
 
         if (found) {
-            /* Cache rs->f in pss_channel (TODO: remove rs->f) */
-            pss->pss_channel = rs->f;
             pages = ram_save_host_page(rs, pss);
         }
     } while (!pages && again);
@@ -3112,7 +3108,7 @@ static void ram_state_resume_prepare(RAMState *rs, QEMUFile *out)
     ram_state_reset(rs);
 
     /* Update RAMState cache of output QEMUFile */
-    rs->f = out;
+    rs->pss[RAM_CHANNEL_PRECOPY].pss_channel = out;
 
     trace_ram_state_resume_prepare(pages);
 }
@@ -3203,7 +3199,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
             return -1;
         }
     }
-    (*rsp)->f = f;
+    (*rsp)->pss[RAM_CHANNEL_PRECOPY].pss_channel = f;
 
     WITH_RCU_READ_LOCK_GUARD() {
         qemu_put_be64(f, ram_bytes_total_common(true) | RAM_SAVE_FLAG_MEM_SIZE);
@@ -3338,7 +3334,7 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
 out:
     if (ret >= 0
         && migration_is_setup_or_active(migrate_get_current()->state)) {
-        ret = multifd_send_sync_main(rs->f);
+        ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel);
         if (ret < 0) {
             return ret;
         }
@@ -3406,7 +3402,7 @@ static int ram_save_complete(QEMUFile *f, void *opaque)
         return ret;
     }
 
-    ret = multifd_send_sync_main(rs->f);
+    ret = multifd_send_sync_main(rs->pss[RAM_CHANNEL_PRECOPY].pss_channel);
     if (ret < 0) {
         return ret;
     }
Signature · 1 line
2.32.0
Signature · 1 line
Dr. David Alan Gilbert / [email protected] / Manchester, UK