[Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Peter Xu]

Trying to get memory region size of an uninitialized memory region is
probably not a good idea. Let's just do the alloc no matter what.

Signed-off-by: Peter Xu <peterx@redhat.com>
---
 backends/hostmem-file.c | 17 +++++++----------
 1 file changed, 7 insertions(+), 10 deletions(-)

diff --git a/backends/hostmem-file.c b/backends/hostmem-file.c
index 42efb2f..a61d1bd 100644
--- a/backends/hostmem-file.c
+++ b/backends/hostmem-file.c
@@ -39,6 +39,7 @@ static void
 file_backend_memory_alloc(HostMemoryBackend *backend, Error **errp)
 {
     HostMemoryBackendFile *fb = MEMORY_BACKEND_FILE(backend);
+    gchar *path;
 
     if (!backend->size) {
         error_setg(errp, "can't create backend with size 0");
@@ -51,16 +52,12 @@ file_backend_memory_alloc(HostMemoryBackend *backend, Error **errp)
 #ifndef CONFIG_LINUX
     error_setg(errp, "-mem-path not supported on this host");
 #else
-    if (!memory_region_size(&backend->mr)) {
-        gchar *path;
-        backend->force_prealloc = mem_prealloc;
-        path = object_get_canonical_path(OBJECT(backend));
-        memory_region_init_ram_from_file(&backend->mr, OBJECT(backend),
-                                 path,
-                                 backend->size, fb->share,
-                                 fb->mem_path, errp);
-        g_free(path);
-    }
+    backend->force_prealloc = mem_prealloc;
+    path = object_get_canonical_path(OBJECT(backend));
+    memory_region_init_ram_from_file(&backend->mr, OBJECT(backend),
+                                     path, backend->size, fb->share,
+                                     fb->mem_path, errp);
+    g_free(path);
 #endif
 }
 
-- 
2.7.4

Re: [Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Paolo Bonzini]

On 10/03/2017 05:13, Peter Xu wrote:
> Trying to get memory region size of an uninitialized memory region is
> probably not a good idea. Let's just do the alloc no matter what.
> 
> Signed-off-by: Peter Xu <peterx@redhat.com>

What is the effect of the bug?  The idea was to do the initialization
once only (memory_region_size ought to be 0 when the MR is
uninitialized; now it is ugly but it made more sense when MemoryRegion
was just a C struct and not a QOM object).

Paolo

> ---
>  backends/hostmem-file.c | 17 +++++++----------
>  1 file changed, 7 insertions(+), 10 deletions(-)
> 
> diff --git a/backends/hostmem-file.c b/backends/hostmem-file.c
> index 42efb2f..a61d1bd 100644
> --- a/backends/hostmem-file.c
> +++ b/backends/hostmem-file.c
> @@ -39,6 +39,7 @@ static void
>  file_backend_memory_alloc(HostMemoryBackend *backend, Error **errp)
>  {
>      HostMemoryBackendFile *fb = MEMORY_BACKEND_FILE(backend);
> +    gchar *path;
>  
>      if (!backend->size) {
>          error_setg(errp, "can't create backend with size 0");
> @@ -51,16 +52,12 @@ file_backend_memory_alloc(HostMemoryBackend *backend, Error **errp)
>  #ifndef CONFIG_LINUX
>      error_setg(errp, "-mem-path not supported on this host");
>  #else
> -    if (!memory_region_size(&backend->mr)) {
> -        gchar *path;
> -        backend->force_prealloc = mem_prealloc;
> -        path = object_get_canonical_path(OBJECT(backend));
> -        memory_region_init_ram_from_file(&backend->mr, OBJECT(backend),
> -                                 path,
> -                                 backend->size, fb->share,
> -                                 fb->mem_path, errp);
> -        g_free(path);
> -    }
> +    backend->force_prealloc = mem_prealloc;
> +    path = object_get_canonical_path(OBJECT(backend));
> +    memory_region_init_ram_from_file(&backend->mr, OBJECT(backend),
> +                                     path, backend->size, fb->share,
> +                                     fb->mem_path, errp);
> +    g_free(path);
>  #endif
>  }
>  
> 

Re: [Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Peter Xu]

On Fri, Mar 10, 2017 at 09:33:57AM +0100, Paolo Bonzini wrote:
> 
> 
> On 10/03/2017 05:13, Peter Xu wrote:
> > Trying to get memory region size of an uninitialized memory region is
> > probably not a good idea. Let's just do the alloc no matter what.
> > 
> > Signed-off-by: Peter Xu <peterx@redhat.com>
> 
> What is the effect of the bug? The idea was to do the initialization
> once only (memory_region_size ought to be 0 when the MR is
> uninitialized; now it is ugly but it made more sense when MemoryRegion
> was just a C struct and not a QOM object).

It's not really a bug. I just saw it, thought it was something not
quite right, so posted a patch. Since it's intended, then please
ignore this patch, and sorry for the noise... :)

(Considering it's a QOM object, it'll better explain itself if we were
 using something like object_initialized() here for the check, but
 since we don't have that, I think fetching size is okay as well)

Thanks,

-- peterx

Re: [Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Peter Maydell]

On 10 March 2017 at 09:59, Peter Xu <peterx@redhat.com> wrote:
> On Fri, Mar 10, 2017 at 09:33:57AM +0100, Paolo Bonzini wrote:
>>
>>
>> On 10/03/2017 05:13, Peter Xu wrote:
>> > Trying to get memory region size of an uninitialized memory region is
>> > probably not a good idea. Let's just do the alloc no matter what.
>> >
>> > Signed-off-by: Peter Xu <peterx@redhat.com>
>>
>> What is the effect of the bug? The idea was to do the initialization
>> once only (memory_region_size ought to be 0 when the MR is
>> uninitialized; now it is ugly but it made more sense when MemoryRegion
>> was just a C struct and not a QOM object).
>
> It's not really a bug. I just saw it, thought it was something not
> quite right, so posted a patch.

We could reasonably abstract out the test into a function
bool backend_mr_initialized(HostMemoryBackend *backend)
{
    /* We forbid zero-length in file_backend_memory_alloc,
     * so zero always means "we haven't allocated the backend
     * MR yet".
     */
    return memory_region_size(&backend->mr) != 0;
}

and use it in file_backend_memory_alloc(), set_mem_path()
and file_memory_backend_set_share(). That would make the intent
clearer here I think.

thanks
-- PMM

Re: [Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Peter Xu]

On Fri, Mar 10, 2017 at 10:17:30AM +0100, Peter Maydell wrote:
> On 10 March 2017 at 09:59, Peter Xu <peterx@redhat.com> wrote:
> > On Fri, Mar 10, 2017 at 09:33:57AM +0100, Paolo Bonzini wrote:
> >>
> >>
> >> On 10/03/2017 05:13, Peter Xu wrote:
> >> > Trying to get memory region size of an uninitialized memory region is
> >> > probably not a good idea. Let's just do the alloc no matter what.
> >> >
> >> > Signed-off-by: Peter Xu <peterx@redhat.com>
> >>
> >> What is the effect of the bug? The idea was to do the initialization
> >> once only (memory_region_size ought to be 0 when the MR is
> >> uninitialized; now it is ugly but it made more sense when MemoryRegion
> >> was just a C struct and not a QOM object).
> >
> > It's not really a bug. I just saw it, thought it was something not
> > quite right, so posted a patch.
> 
> We could reasonably abstract out the test into a function
> bool backend_mr_initialized(HostMemoryBackend *backend)
> {
>     /* We forbid zero-length in file_backend_memory_alloc,
>      * so zero always means "we haven't allocated the backend
>      * MR yet".
>      */
>     return memory_region_size(&backend->mr) != 0;
> }
> 
> and use it in file_backend_memory_alloc(), set_mem_path()
> and file_memory_backend_set_share(). That would make the intent
> clearer here I think.

Agree, maybe use it in hostmem.c as well where capable?

(Paolo: I wasn't planning to add anything there, but if any of you
 like me to do this cleanup, I would be glad to :)

-- peterx

Re: [Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Paolo Bonzini]

On 10/03/2017 10:36, Peter Xu wrote:
> On Fri, Mar 10, 2017 at 10:17:30AM +0100, Peter Maydell wrote:
>> On 10 March 2017 at 09:59, Peter Xu <peterx@redhat.com> wrote:
>>> On Fri, Mar 10, 2017 at 09:33:57AM +0100, Paolo Bonzini wrote:
>>>>
>>>>
>>>> On 10/03/2017 05:13, Peter Xu wrote:
>>>>> Trying to get memory region size of an uninitialized memory region is
>>>>> probably not a good idea. Let's just do the alloc no matter what.
>>>>>
>>>>> Signed-off-by: Peter Xu <peterx@redhat.com>
>>>>
>>>> What is the effect of the bug? The idea was to do the initialization
>>>> once only (memory_region_size ought to be 0 when the MR is
>>>> uninitialized; now it is ugly but it made more sense when MemoryRegion
>>>> was just a C struct and not a QOM object).
>>>
>>> It's not really a bug. I just saw it, thought it was something not
>>> quite right, so posted a patch.
>>
>> We could reasonably abstract out the test into a function
>> bool backend_mr_initialized(HostMemoryBackend *backend)
>> {
>>     /* We forbid zero-length in file_backend_memory_alloc,
>>      * so zero always means "we haven't allocated the backend
>>      * MR yet".
>>      */
>>     return memory_region_size(&backend->mr) != 0;
>> }
>>
>> and use it in file_backend_memory_alloc(), set_mem_path()
>> and file_memory_backend_set_share(). That would make the intent
>> clearer here I think.
> 
> Agree, maybe use it in hostmem.c as well where capable?
> 
> (Paolo: I wasn't planning to add anything there, but if any of you
>  like me to do this cleanup, I would be glad to :)

Yes, please (to name things more consistently it should be
host_memory_backend_initialized).

Paolo

Re: [Qemu-devel] [PATCH] hostmem: fix reference to uninit mr

[Paolo Bonzini]

On 10/03/2017 09:59, Peter Xu wrote:
>> What is the effect of the bug? The idea was to do the initialization
>> once only (memory_region_size ought to be 0 when the MR is
>> uninitialized; now it is ugly but it made more sense when MemoryRegion
>> was just a C struct and not a QOM object).
> It's not really a bug. I just saw it, thought it was something not
> quite right, so posted a patch. Since it's intended, then please
> ignore this patch, and sorry for the noise... :)

It's intended but ugly. :)  Peter's idea is a good one if you want to
implement something along those lines.

Paolo