[PATCH v5 1/2] migration: skip poisoned memory pages on "ram saving" phase

["“William Roche" <william.roche@oracle.com>]

From: William Roche <william.roche@oracle.com>

A memory page poisoned from the hypervisor level is no longer readable.
Thus, it is now treated as a zero-page for the ram saving migration phase.

The migration of a VM will crash Qemu when it tries to read the
memory address space and stumbles on the poisoned page with a similar
stack trace:

Program terminated with signal SIGBUS, Bus error.
#0  _mm256_loadu_si256
#1  buffer_zero_avx2
#2  select_accel_fn
#3  buffer_is_zero
#4  save_zero_page
#5  ram_save_target_page_legacy
#6  ram_save_host_page
#7  ram_find_and_save_block
#8  ram_save_iterate
#9  qemu_savevm_state_iterate
#10 migration_iteration_run
#11 migration_thread
#12 qemu_thread_start

Fix it by considering poisoned pages as if they were zero-pages for
the migration copy. This fix also works with underlying large pages,
taking into account the RAMBlock segment "page-size".

Standard migration and compressed transfers are handled by this code.
RDMA transfer isn't touched.

Reviewed-by: Peter Xu <peterx@redhat.com>
Tested-by: Li Zhijian <lizhijian@fujitsu.com> # RDMA
Signed-off-by: William Roche <william.roche@oracle.com>
---
 accel/kvm/kvm-all.c      | 14 ++++++++++++++
 accel/stubs/kvm-stub.c   |  5 +++++
 include/sysemu/kvm.h     | 10 ++++++++++
 migration/ram-compress.c |  3 ++-
 migration/ram.c          | 24 ++++++++++++++++++++++--
 migration/ram.h          |  2 ++
 6 files changed, 55 insertions(+), 3 deletions(-)

diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
index e39a810a4e..64c0b37823 100644
--- a/accel/kvm/kvm-all.c
+++ b/accel/kvm/kvm-all.c
@@ -1149,6 +1149,20 @@ static void kvm_unpoison_all(void *param)
     }
 }
 
+bool kvm_hwpoisoned_page(RAMBlock *block, void *offset)
+{
+    HWPoisonPage *pg;
+    ram_addr_t ram_addr = (ram_addr_t) offset;
+
+    QLIST_FOREACH(pg, &hwpoison_page_list, list) {
+        if ((ram_addr >= pg->ram_addr) &&
+            (ram_addr - pg->ram_addr < block->page_size)) {
+            return true;
+        }
+    }
+    return false;
+}
+
 void kvm_hwpoison_page_add(ram_addr_t ram_addr)
 {
     HWPoisonPage *page;
diff --git a/accel/stubs/kvm-stub.c b/accel/stubs/kvm-stub.c
index 1b37d9a302..17774fa5ef 100644
--- a/accel/stubs/kvm-stub.c
+++ b/accel/stubs/kvm-stub.c
@@ -124,3 +124,8 @@ uint32_t kvm_dirty_ring_size(void)
 {
     return 0;
 }
+
+bool kvm_hwpoisoned_page(RAMBlock *block, void *ram_addr)
+{
+    return false;
+}
diff --git a/include/sysemu/kvm.h b/include/sysemu/kvm.h
index 80b69d88f6..66937f9dfe 100644
--- a/include/sysemu/kvm.h
+++ b/include/sysemu/kvm.h
@@ -539,4 +539,14 @@ bool kvm_arch_cpu_check_are_resettable(void);
 bool kvm_dirty_ring_enabled(void);
 
 uint32_t kvm_dirty_ring_size(void);
+
+/**
+ * kvm_hwpoisoned_page - indicate if the given page is poisoned
+ * @block: memory block of the given page
+ * @ram_addr: offset of the page
+ *
+ * Returns: true: page is poisoned
+ *          false: page not yet poisoned
+ */
+bool kvm_hwpoisoned_page(RAMBlock *block, void *ram_addr);
 #endif
diff --git a/migration/ram-compress.c b/migration/ram-compress.c
index fa4388f6a6..a7772a08a2 100644
--- a/migration/ram-compress.c
+++ b/migration/ram-compress.c
@@ -35,6 +35,7 @@
 #include "qemu/stats64.h"
 #include "migration.h"
 #include "options.h"
+#include "ram.h"
 #include "io/channel-null.h"
 #include "exec/target_page.h"
 #include "exec/ramblock.h"
@@ -214,7 +215,7 @@ static CompressResult do_compress_ram_page(QEMUFile *f, z_stream *stream,
 
     assert(qemu_file_buffer_empty(f));
 
-    if (buffer_is_zero(p, page_size)) {
+    if (migration_buffer_is_zero(block, offset, page_size)) {
         return RES_ZEROPAGE;
     }
 
diff --git a/migration/ram.c b/migration/ram.c
index 8c7886ab79..5fd4d27854 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -1107,6 +1107,26 @@ void ram_release_page(const char *rbname, uint64_t offset)
     ram_discard_range(rbname, offset, TARGET_PAGE_SIZE);
 }
 
+/**
+ * migration_buffer_is_zero: indicate if the page at the given
+ * location is entirely filled with zero, or is a poisoned page.
+ *
+ * @block: block that contains the page
+ * @offset: offset inside the block for the page
+ * @len: size to consider
+ */
+bool migration_buffer_is_zero(RAMBlock *block, ram_addr_t offset,
+                                     size_t len)
+{
+    uint8_t *p = block->host + offset;
+
+    if (kvm_enabled() && kvm_hwpoisoned_page(block, (void *)offset)) {
+        return true;
+    }
+
+    return buffer_is_zero(p, len);
+}
+
 /**
  * save_zero_page: send the zero page to the stream
  *
@@ -1119,11 +1139,10 @@ void ram_release_page(const char *rbname, uint64_t offset)
 static int save_zero_page(RAMState *rs, PageSearchStatus *pss,
                           ram_addr_t offset)
 {
-    uint8_t *p = pss->block->host + offset;
     QEMUFile *file = pss->pss_channel;
     int len = 0;
 
-    if (!buffer_is_zero(p, TARGET_PAGE_SIZE)) {
+    if (!migration_buffer_is_zero(pss->block, offset, TARGET_PAGE_SIZE)) {
         return 0;
     }
 
@@ -1154,6 +1173,7 @@ static int save_zero_page(RAMState *rs, PageSearchStatus *pss,
  *        > 0 - number of pages written
  *
  * Return true if the pages has been saved, otherwise false is returned.
+ * TODO: hwpoison pages fail RDMA migration, should be handled.
  */
 static bool control_save_page(PageSearchStatus *pss,
                               ram_addr_t offset, int *pages)
diff --git a/migration/ram.h b/migration/ram.h
index 9b937a446b..d34ba79d36 100644
--- a/migration/ram.h
+++ b/migration/ram.h
@@ -65,6 +65,8 @@ void ram_handle_zero(void *host, uint64_t size);
 void ram_transferred_add(uint64_t bytes);
 void ram_release_page(const char *rbname, uint64_t offset);
 
+bool migration_buffer_is_zero(RAMBlock *block, ram_addr_t offset, size_t len);
+
 int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr);
 bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t byte_offset);
 void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr);
-- 
2.39.3