Skip to contentSkip to conversation

migration/ram: Merge zero page handling

PATCH Superseded Fabiano Rosas 2 participants 13 messages Aug 15, 2023

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover migration/ram: Merge zero page handling

Hi,

This is another small series that I extracted from my fixed-ram series and that could be already considered for merging.

This is just code movement, no functional change. The objective is to consolidate the zero page handling in the same routine that saves the page header and does accounting. Then in the future I'll be able to just return early because fixed-ram ignores zero pages.

CI run: https://gitlab.com/farosas/qemu/-/pipelines/968300062

Fabiano Rosas (5):
migration/ram: Remove RAMState from xbzrle_cache_zero_page migration/ram: Stop passing QEMUFile around in save_zero_page migration/ram: Move xbzrle zero page handling into save_zero_page migration/ram: Return early from save_zero_page migration/ram: Merge save_zero_page functions

 migration/ram.c | 75 ++++++++++++++++++++-----------------------------
 1 file changed, 30 insertions(+), 45 deletions(-)
Signature · 1 line
2.35.3
#1 lore raw

PATCH 1/5 migration/ram: Remove RAMState from xbzrle_cache_zero_page

'rs' is not used in that function. It's a leftover from commit 9360447d34 ("ram: Use MigrationStats for statistics").

---
 migration/ram.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
migration/ram.c +2 -3
index 9040d66e61..87efab72e8 100644
@@ -561,7 +561,6 @@ void mig_throttle_counter_reset(void)
 /**
  * xbzrle_cache_zero_page: insert a zero page in the XBZRLE cache
  *
- * @rs: current RAM state
  * @current_addr: address for the zero page
  *
  * Update the xbzrle cache to reflect a page that's been sent as all 0.
@@ -570,7 +569,7 @@ void mig_throttle_counter_reset(void)
  * As a bonus, if the page wasn't in the cache it gets added so that
  * when a small write is made into the 0'd page it gets XBZRLE sent.
  */
-static void xbzrle_cache_zero_page(RAMState *rs, ram_addr_t current_addr)
+static void xbzrle_cache_zero_page(ram_addr_t current_addr)
 {
     /* We don't care if this fails to allocate a new cache page
      * as long as it updated an old one */
@@ -2148,7 +2147,7 @@ static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss)
          */
         if (rs->xbzrle_started) {
             XBZRLE_cache_lock();
-            xbzrle_cache_zero_page(rs, block->offset + offset);
+            xbzrle_cache_zero_page(block->offset + offset);
             XBZRLE_cache_unlock();
         }
         return res;
Signature · 1 line
2.35.3
Reviewed-by on v1 1/5
#2 lore raw

Re: [PATCH 1/5] migration/ram: Remove RAMState from xbzrle_cache_zero_page

Quoted 4 lines · Fabiano Rosas

'rs' is not used in that function. It's a leftover from commit 9360447d34 ("ram: Use MigrationStats for statistics").

Signature · 1 line
Peter Xu
#3 lore raw

PATCH 2/5 migration/ram: Stop passing QEMUFile around in save_zero_page

In reply to Fabiano Rosas

We don't need the QEMUFile when we're already passing the PageSearchStatus.

---
 migration/ram.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
migration/ram.c +6 -5
index 87efab72e8..761f43dc34 100644
@@ -1138,10 +1138,11 @@ void ram_release_page(const char *rbname, uint64_t offset)
  * @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(PageSearchStatus *pss, QEMUFile *file,
-                                  RAMBlock *block, ram_addr_t offset)
+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)) {
@@ -1162,10 +1163,10 @@ static int save_zero_page_to_file(PageSearchStatus *pss, 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(PageSearchStatus *pss, QEMUFile *f, RAMBlock *block,
+static int save_zero_page(PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(pss, f, block, offset);
+    int len = save_zero_page_to_file(pss, block, offset);
 
     if (len) {
         stat64_add(&mig_stats.zero_pages, 1);
@@ -2140,7 +2141,7 @@ static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss)
         return 1;
     }
 
-    res = save_zero_page(pss, 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
Signature · 1 line
2.35.3
#5 lore raw

PATCH 3/5 migration/ram: Move xbzrle zero page handling into save_zero_page

In reply to Fabiano Rosas

It makes a bit more sense to have the zero page handling of xbzrle right where we save the zero page.

This also makes save_zero_page() follow the same format as save_compress_page() at the top level of ram_save_target_page_legacy().

---
 migration/ram.c | 27 +++++++++++++++------------
 1 file changed, 15 insertions(+), 12 deletions(-)
migration/ram.c +15 -12
index 761f43dc34..a10410a1a5 100644
@@ -1159,11 +1159,12 @@ static int save_zero_page_to_file(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
  */
-static int save_zero_page(PageSearchStatus *pss, RAMBlock *block,
+static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
     int len = save_zero_page_to_file(pss, block, offset);
@@ -1171,6 +1172,17 @@ static int save_zero_page(PageSearchStatus *pss, RAMBlock *block,
     if (len) {
         stat64_add(&mig_stats.zero_pages, 1);
         ram_transferred_add(len);
+
+        /*
+         * Must let xbzrle know, otherwise a previous (now 0'd) cached
+         * page would be stale.
+         */
+        if (rs->xbzrle_started) {
+            XBZRLE_cache_lock();
+            xbzrle_cache_zero_page(block->offset + offset);
+            XBZRLE_cache_unlock();
+        }
+
         return 1;
     }
     return -1;
@@ -2141,17 +2153,8 @@ static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss)
         return 1;
     }
 
-    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
-         */
-        if (rs->xbzrle_started) {
-            XBZRLE_cache_lock();
-            xbzrle_cache_zero_page(block->offset + offset);
-            XBZRLE_cache_unlock();
-        }
-        return res;
+    if (save_zero_page(rs, pss, block, offset)) {
+        return 1;
     }
 
     /*
Signature · 1 line
2.35.3
Reviewed-by on v1 3/5
#6 lore raw

Re: [PATCH 3/5] migration/ram: Move xbzrle zero page handling into save_zero_page

Quoted 8 lines · Fabiano Rosas

It makes a bit more sense to have the zero page handling of xbzrle right where we save the zero page.

This also makes save_zero_page() follow the same format as save_compress_page() at the top level of ram_save_target_page_legacy().

Signature · 1 line
Peter Xu
#7 lore raw

Re: [PATCH 3/5] migration/ram: Move xbzrle zero page handling into save_zero_page

In reply to Fabiano Rosas
Quoted 49 lines · Fabiano Rosas

It makes a bit more sense to have the zero page handling of xbzrle right where we save the zero page.

This also makes save_zero_page() follow the same format as save_compress_page() at the top level of ram_save_target_page_legacy().

---
 migration/ram.c | 27 +++++++++++++++------------
 1 file changed, 15 insertions(+), 12 deletions(-)
migration/ram.c +13 -1
index 761f43dc34..a10410a1a5 100644
@@ -1159,11 +1159,12 @@ static int save_zero_page_to_file(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
  */
-static int save_zero_page(PageSearchStatus *pss, RAMBlock *block,
+static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
     int len = save_zero_page_to_file(pss, block, offset);
@@ -1171,6 +1172,17 @@ static int save_zero_page(PageSearchStatus *pss, RAMBlock *block,
     if (len) {
         stat64_add(&mig_stats.zero_pages, 1);
         ram_transferred_add(len);
+
+        /*
+         * Must let xbzrle know, otherwise a previous (now 0'd) cached
+         * page would be stale.
+         */
+        if (rs->xbzrle_started) {
+            XBZRLE_cache_lock();
+            xbzrle_cache_zero_page(block->offset + offset);
+            XBZRLE_cache_unlock();
+        }
+
         return 1;
     }
     return -1;

migration/ram.c static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss) · line 2145

Quoted 5 lines
migration/ram.c +0 -2
@@ -2141,17 +2153,8 @@ static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss)
         return 1;
     }
 
-    res = save_zero_page(pss, block, offset);
-    if (res > 0) {

These two subtractions should be in patch 5.

migration/ram.c static int ram_save_target_page_legacy(RAMState *rs, PageSearchStatus *pss) · line 2157

Quoted 11 lines
migration/ram.c +2 -9
-        /* Must let xbzrle know, otherwise a previous (now 0'd) cached
-         * page would be stale
-         */
-        if (rs->xbzrle_started) {
-            XBZRLE_cache_lock();
-            xbzrle_cache_zero_page(block->offset + offset);
-            XBZRLE_cache_unlock();
-        }
-        return res;
+    if (save_zero_page(rs, pss, block, offset)) {
+        return 1;

These two additions should be in patch 5.

Quoted 3 lines
migration/ram.c +0 -0
     }
 
     /*
#8 lore raw

PATCH 4/5 migration/ram: Return early from save_zero_page

In reply to Fabiano Rosas

Invert the first conditional so we return early when len == 0. This is merely to make the next patch easier to read.

---
 migration/ram.c | 29 +++++++++++++++--------------
 1 file changed, 15 insertions(+), 14 deletions(-)
migration/ram.c +15 -14
index a10410a1a5..8ec38f69e8 100644
@@ -1169,23 +1169,24 @@ static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
 {
     int len = save_zero_page_to_file(pss, block, offset);
 
-    if (len) {
-        stat64_add(&mig_stats.zero_pages, 1);
-        ram_transferred_add(len);
+    if (!len) {
+        return -1;
+    }
 
-        /*
-         * Must let xbzrle know, otherwise a previous (now 0'd) cached
-         * page would be stale.
-         */
-        if (rs->xbzrle_started) {
-            XBZRLE_cache_lock();
-            xbzrle_cache_zero_page(block->offset + offset);
-            XBZRLE_cache_unlock();
-        }
+    stat64_add(&mig_stats.zero_pages, 1);
+    ram_transferred_add(len);
 
-        return 1;
+    /*
+     * Must let xbzrle know, otherwise a previous (now 0'd) cached
+     * page would be stale.
+     */
+    if (rs->xbzrle_started) {
+        XBZRLE_cache_lock();
+        xbzrle_cache_zero_page(block->offset + offset);
+        XBZRLE_cache_unlock();
     }
-    return -1;
+
+    return 1;
 }
 
 /*
Signature · 1 line
2.35.3
#9 lore raw

Re: [PATCH 4/5] migration/ram: Return early from save_zero_page

Quoted 39 lines · Fabiano Rosas

Invert the first conditional so we return early when len == 0. This is merely to make the next patch easier to read.

---
 migration/ram.c | 29 +++++++++++++++--------------
 1 file changed, 15 insertions(+), 14 deletions(-)
migration/ram.c +8 -13
index a10410a1a5..8ec38f69e8 100644
@@ -1169,23 +1169,24 @@ static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
 {
     int len = save_zero_page_to_file(pss, block, offset);
 
-    if (len) {
-        stat64_add(&mig_stats.zero_pages, 1);
-        ram_transferred_add(len);
+    if (!len) {
+        return -1;
+    }
 
-        /*
-         * Must let xbzrle know, otherwise a previous (now 0'd) cached
-         * page would be stale.
-         */
-        if (rs->xbzrle_started) {
-            XBZRLE_cache_lock();
-            xbzrle_cache_zero_page(block->offset + offset);
-            XBZRLE_cache_unlock();
-        }
+    stat64_add(&mig_stats.zero_pages, 1);
+    ram_transferred_add(len);
 
-        return 1;
+    /*
+     * Must let xbzrle know, otherwise a previous (now 0'd) cached
+     * page would be stale.

migration/ram.c static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block, · line 1190

Quoted 10 lines
migration/ram.c +7 -1
+     */
+    if (rs->xbzrle_started) {
+        XBZRLE_cache_lock();
+        xbzrle_cache_zero_page(block->offset + offset);
+        XBZRLE_cache_unlock();
     }
-    return -1;
+
+    return 1;
 }

Can we squash this into previous patch?

Signature · 1 line
Peter Xu
#10 lore raw

PATCH 5/5 migration/ram: Merge save_zero_page functions

In reply to Fabiano Rosas

We don't need to do this in two pieces. One single function makes it easier to grasp, specially since it removes the indirection on the return value handling.

---
 migration/ram.c | 41 +++++++++++------------------------------
 1 file changed, 11 insertions(+), 30 deletions(-)
migration/ram.c +11 -30
index 8ec38f69e8..13935ead1c 100644
@@ -1128,32 +1128,6 @@ void ram_release_page(const char *rbname, uint64_t offset)
     ram_discard_range(rbname, offset, TARGET_PAGE_SIZE);
 }
 
-/**
- * save_zero_page_to_file: send the zero page to the file
- *
- * Returns the size of data written to the file, 0 means the page is not
- * a zero page
- *
- * @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(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(pss, file, block, offset | RAM_SAVE_FLAG_ZERO);
-        qemu_put_byte(file, 0);
-        len += 1;
-        ram_release_page(block->idstr, offset);
-    }
-    return len;
-}
-
 /**
  * save_zero_page: send the zero page to the stream
  *
@@ -1167,12 +1141,19 @@ static int save_zero_page_to_file(PageSearchStatus *pss, RAMBlock *block,
 static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(pss, block, offset);
+    uint8_t *p = block->host + offset;
+    QEMUFile *file = pss->pss_channel;
+    int len = 0;
 
-    if (!len) {
-        return -1;
+    if (!buffer_is_zero(p, TARGET_PAGE_SIZE)) {
+        return 0;
     }
 
+    len += save_page_header(pss, file, block, offset | RAM_SAVE_FLAG_ZERO);
+    qemu_put_byte(file, 0);
+    len += 1;
+    ram_release_page(block->idstr, offset);
+
     stat64_add(&mig_stats.zero_pages, 1);
     ram_transferred_add(len);
 
@@ -1186,7 +1167,7 @@ static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
         XBZRLE_cache_unlock();
     }
 
-    return 1;
+    return len;
 }
 
 /*
Signature · 1 line
2.35.3
#11 lore raw

Re: [PATCH 5/5] migration/ram: Merge save_zero_page functions

Quoted 70 lines · Fabiano Rosas

We don't need to do this in two pieces. One single function makes it easier to grasp, specially since it removes the indirection on the return value handling.

---
 migration/ram.c | 41 +++++++++++------------------------------
 1 file changed, 11 insertions(+), 30 deletions(-)
migration/ram.c +10 -29
index 8ec38f69e8..13935ead1c 100644
@@ -1128,32 +1128,6 @@ void ram_release_page(const char *rbname, uint64_t offset)
     ram_discard_range(rbname, offset, TARGET_PAGE_SIZE);
 }
 
-/**
- * save_zero_page_to_file: send the zero page to the file
- *
- * Returns the size of data written to the file, 0 means the page is not
- * a zero page
- *
- * @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(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(pss, file, block, offset | RAM_SAVE_FLAG_ZERO);
-        qemu_put_byte(file, 0);
-        len += 1;
-        ram_release_page(block->idstr, offset);
-    }
-    return len;
-}
-
 /**
  * save_zero_page: send the zero page to the stream
  *
@@ -1167,12 +1141,19 @@ static int save_zero_page_to_file(PageSearchStatus *pss, RAMBlock *block,
 static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(pss, block, offset);
+    uint8_t *p = block->host + offset;
+    QEMUFile *file = pss->pss_channel;
+    int len = 0;
 
-    if (!len) {
-        return -1;
+    if (!buffer_is_zero(p, TARGET_PAGE_SIZE)) {
+        return 0;
     }
 
+    len += save_page_header(pss, file, block, offset | RAM_SAVE_FLAG_ZERO);
+    qemu_put_byte(file, 0);
+    len += 1;
+    ram_release_page(block->idstr, offset);
+
     stat64_add(&mig_stats.zero_pages, 1);
     ram_transferred_add(len);
 

migration/ram.c static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block, · line 1170

Quoted 5 lines
migration/ram.c +1 -1
@@ -1186,7 +1167,7 @@ static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
         XBZRLE_cache_unlock();
     }
 
-    return 1;
+    return len;

I don't think it's correct.. We need to keep the retval definition (how many pages were sent) rather than returning num of bytes, I think.

I'm curious how did this pass any form of test.. because I think we did assert that:

            /* Be strict to return code; it must be 1, or what else? */
            if (migration_ops->ram_save_target_page(rs, pss) != 1) {
                error_report_once("%s: ram_save_target_page failed", __func__);
                ret = -1;
                goto out;
            }

Did I miss something?

Signature · 1 line
Peter Xu
#12 lore raw

Re: [PATCH 5/5] migration/ram: Merge save_zero_page functions

Quoted 83 lines · Peter Xu
Quoted 70 lines · Fabiano Rosas

We don't need to do this in two pieces. One single function makes it easier to grasp, specially since it removes the indirection on the return value handling.

---
 migration/ram.c | 41 +++++++++++------------------------------
 1 file changed, 11 insertions(+), 30 deletions(-)
migration/ram.c +10 -29
index 8ec38f69e8..13935ead1c 100644
@@ -1128,32 +1128,6 @@ void ram_release_page(const char *rbname, uint64_t offset)
     ram_discard_range(rbname, offset, TARGET_PAGE_SIZE);
 }
 
-/**
- * save_zero_page_to_file: send the zero page to the file
- *
- * Returns the size of data written to the file, 0 means the page is not
- * a zero page
- *
- * @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(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(pss, file, block, offset | RAM_SAVE_FLAG_ZERO);
-        qemu_put_byte(file, 0);
-        len += 1;
-        ram_release_page(block->idstr, offset);
-    }
-    return len;
-}
-
 /**
  * save_zero_page: send the zero page to the stream
  *
@@ -1167,12 +1141,19 @@ static int save_zero_page_to_file(PageSearchStatus *pss, RAMBlock *block,
 static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
                           ram_addr_t offset)
 {
-    int len = save_zero_page_to_file(pss, block, offset);
+    uint8_t *p = block->host + offset;
+    QEMUFile *file = pss->pss_channel;
+    int len = 0;
 
-    if (!len) {
-        return -1;
+    if (!buffer_is_zero(p, TARGET_PAGE_SIZE)) {
+        return 0;
     }
 
+    len += save_page_header(pss, file, block, offset | RAM_SAVE_FLAG_ZERO);
+    qemu_put_byte(file, 0);
+    len += 1;
+    ram_release_page(block->idstr, offset);
+
     stat64_add(&mig_stats.zero_pages, 1);
     ram_transferred_add(len);
 

migration/ram.c static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block, · line 1170

Quoted 5 lines
migration/ram.c +1 -1
@@ -1186,7 +1167,7 @@ static int save_zero_page(RAMState *rs, PageSearchStatus *pss, RAMBlock *block,
         XBZRLE_cache_unlock();
     }
 
-    return 1;
+    return len;

I don't think it's correct.. We need to keep the retval definition (how many pages were sent) rather than returning num of bytes, I think.

I'm curious how did this pass any form of test.. because I think we did assert that:

Quoted 8 lines
            /* Be strict to return code; it must be 1, or what else? */
            if (migration_ops->ram_save_target_page(rs, pss) != 1) {
                error_report_once("%s: ram_save_target_page failed", __func__);
                ret = -1;
                goto out;
            }

Did I miss something?

Kind of, this code is correct. It's just that I made save_zero_page() return bytes like save_zero_page_to_file() used to do and made ram_save_target_page() return 1 instead of passing the value from save_zero_page() along.

But there's a bug in patch 3 because what I described above should only happen in this patch 5.