[Qemu-devel] [PATCH] kvm: Avoid COW if KVM MMU is asynchronous

[Qemu-devel] [PATCH] kvm: Avoid COW if KVM MMU is asynchronous

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1203 bytes --]

If the KVM MMU is asynchronous (kernel does not support MMU_NOTIFIER),
we have to avoid COW for the guest memory. Otherwise we risk serious
breakage when guest pages change there physical locations due to COW
after fork. Seen when forking smbd during runtime via -smb.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---

 exec.c |   14 ++++++++++++++
 1 files changed, 14 insertions(+), 0 deletions(-)

diff --git a/exec.c b/exec.c
index 23013fb..ce0603e 100644
--- a/exec.c
+++ b/exec.c
@@ -2477,6 +2477,20 @@ ram_addr_t qemu_ram_alloc(ram_addr_t size)
     new_block->next = ram_blocks;
     ram_blocks = new_block;
 
+    if (kvm_enabled() && !kvm_has_sync_mmu()) {
+#ifdef MADV_DONTFORK
+        int ret = madvise(new_block->host, size, MADV_DONTFORK);
+        if (ret) {
+            perror("madvice");
+            exit(1);
+        }
+#else
+        fprintf(stderr,
+                "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
+        exit(1);
+#endif
+    }
+
     phys_ram_dirty = qemu_realloc(phys_ram_dirty,
         (last_ram_offset + size) >> TARGET_PAGE_BITS);
     memset(phys_ram_dirty + (last_ram_offset >> TARGET_PAGE_BITS),


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]

Re: [Qemu-devel] [PATCH] kvm: Avoid COW if KVM MMU is asynchronous

[Avi Kivity]

Jan Kiszka wrote:
> If the KVM MMU is asynchronous (kernel does not support MMU_NOTIFIER),
> we have to avoid COW for the guest memory. Otherwise we risk serious
> breakage when guest pages change there physical locations due to COW
> after fork. Seen when forking smbd during runtime via -smb.
>
> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> ---
>
>  exec.c |   14 ++++++++++++++
>  1 files changed, 14 insertions(+), 0 deletions(-)
>
> diff --git a/exec.c b/exec.c
> index 23013fb..ce0603e 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -2477,6 +2477,20 @@ ram_addr_t qemu_ram_alloc(ram_addr_t size)
>      new_block->next = ram_blocks;
>      ram_blocks = new_block;
>  
> +    if (kvm_enabled() && !kvm_has_sync_mmu()) {
> +#ifdef MADV_DONTFORK
> +        int ret = madvise(new_block->host, size, MADV_DONTFORK);
> +        if (ret) {
> +            perror("madvice");
> +            exit(1);
> +        }
> +#else
> +        fprintf(stderr,
> +                "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
> +        exit(1);
> +#endif
> +    }
> +

Suggest wrapping in a function and hiding it deep inside kvm-all.c.

-- 
error compiling committee.c: too many arguments to function

[Qemu-devel] [PATCH v2] kvm: Avoid COW if KVM MMU is asynchronous

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 2094 bytes --]

Avi Kivity wrote:
> Suggest wrapping in a function and hiding it deep inside kvm-all.c.
> 

Done in v2:

---------->

If the KVM MMU is asynchronous (kernel does not support MMU_NOTIFIER),
we have to avoid COW for the guest memory. Otherwise we risk serious
breakage when guest pages change there physical locations due to COW
after fork. Seen when forking smbd during runtime via -smb.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---

 exec.c    |    3 +++
 kvm-all.c |   18 ++++++++++++++++++
 kvm.h     |    2 ++
 3 files changed, 23 insertions(+), 0 deletions(-)

diff --git a/exec.c b/exec.c
index 23013fb..81c6a29 100644
--- a/exec.c
+++ b/exec.c
@@ -2484,6 +2484,9 @@ ram_addr_t qemu_ram_alloc(ram_addr_t size)
 
     last_ram_offset += size;
 
+    if (kvm_enabled())
+        kvm_setup_guest_memory(new_block->host, size);
+
     return new_block->offset;
 }
 
diff --git a/kvm-all.c b/kvm-all.c
index 1128bee..f36b39b 100644
--- a/kvm-all.c
+++ b/kvm-all.c
@@ -783,6 +783,24 @@ int kvm_has_sync_mmu(void)
     return 0;
 }
 
+void kvm_setup_guest_memory(void *start, size_t size)
+{
+    if (!kvm_has_sync_mmu()) {
+#ifdef MADV_DONTFORK
+        int ret = madvise(start, size, MADV_DONTFORK);
+
+        if (ret) {
+            perror("madvice");
+            exit(1);
+        }
+#else
+        fprintf(stderr,
+                "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
+        exit(1);
+#endif
+    }
+}
+
 #ifdef KVM_CAP_SET_GUEST_DEBUG
 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
                                                  target_ulong pc)
diff --git a/kvm.h b/kvm.h
index 803a874..0ea2426 100644
--- a/kvm.h
+++ b/kvm.h
@@ -48,6 +48,8 @@ int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size);
 
 int kvm_has_sync_mmu(void);
 
+void kvm_setup_guest_memory(void *start, size_t size);
+
 int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size);
 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size);
 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]

Re: [Qemu-devel] [PATCH v2] kvm: Avoid COW if KVM MMU is asynchronous

[Avi Kivity]

Jan Kiszka wrote:
> If the KVM MMU is asynchronous (kernel does not support MMU_NOTIFIER),
> we have to avoid COW for the guest memory. Otherwise we risk serious
> breakage when guest pages change there physical locations due to COW
> after fork. Seen when forking smbd during runtime via -smb.
>
>  
>      last_ram_offset += size;
>  
> +    if (kvm_enabled())
> +        kvm_setup_guest_memory(new_block->host, size);
> +
>      return new_block->offset;
>  }
>   

I feel a little bad about picking such small nits, but the if statement 
requires braces.



-- 
error compiling committee.c: too many arguments to function

[Qemu-devel] [PATCH v3] kvm: Avoid COW if KVM MMU is asynchronous

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 2686 bytes --]

Avi Kivity wrote:
> Jan Kiszka wrote:
>> If the KVM MMU is asynchronous (kernel does not support MMU_NOTIFIER),
>> we have to avoid COW for the guest memory. Otherwise we risk serious
>> breakage when guest pages change there physical locations due to COW
>> after fork. Seen when forking smbd during runtime via -smb.
>>
>>  
>>      last_ram_offset += size;
>>  
>> +    if (kvm_enabled())
>> +        kvm_setup_guest_memory(new_block->host, size);
>> +
>>      return new_block->offset;
>>  }
>>   
> 
> I feel a little bad about picking such small nits, but the if statement
> requires braces.
> 

No problem. I'm in "trivial editing mode" ATM anyway.

Thanks,
Jan

---------->

If the KVM MMU is asynchronous (kernel does not support MMU_NOTIFIER),
we have to avoid COW for the guest memory. Otherwise we risk serious
breakage when guest pages change there physical locations due to COW
after fork. Seen when forking smbd during runtime via -smb.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---

 exec.c    |    3 +++
 kvm-all.c |   18 ++++++++++++++++++
 kvm.h     |    2 ++
 3 files changed, 23 insertions(+), 0 deletions(-)

diff --git a/exec.c b/exec.c
index 23013fb..9281bd7 100644
--- a/exec.c
+++ b/exec.c
@@ -2484,6 +2484,9 @@ ram_addr_t qemu_ram_alloc(ram_addr_t size)
 
     last_ram_offset += size;
 
+    if (kvm_enabled()) {
+        kvm_setup_guest_memory(new_block->host, size);
+    }
     return new_block->offset;
 }
 
diff --git a/kvm-all.c b/kvm-all.c
index 1128bee..f36b39b 100644
--- a/kvm-all.c
+++ b/kvm-all.c
@@ -783,6 +783,24 @@ int kvm_has_sync_mmu(void)
     return 0;
 }
 
+void kvm_setup_guest_memory(void *start, size_t size)
+{
+    if (!kvm_has_sync_mmu()) {
+#ifdef MADV_DONTFORK
+        int ret = madvise(start, size, MADV_DONTFORK);
+
+        if (ret) {
+            perror("madvice");
+            exit(1);
+        }
+#else
+        fprintf(stderr,
+                "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
+        exit(1);
+#endif
+    }
+}
+
 #ifdef KVM_CAP_SET_GUEST_DEBUG
 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *env,
                                                  target_ulong pc)
diff --git a/kvm.h b/kvm.h
index 803a874..0ea2426 100644
--- a/kvm.h
+++ b/kvm.h
@@ -48,6 +48,8 @@ int kvm_log_stop(target_phys_addr_t phys_addr, ram_addr_t size);
 
 int kvm_has_sync_mmu(void);
 
+void kvm_setup_guest_memory(void *start, size_t size);
+
 int kvm_coalesce_mmio_region(target_phys_addr_t start, ram_addr_t size);
 int kvm_uncoalesce_mmio_region(target_phys_addr_t start, ram_addr_t size);
 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]