Re: [Xen-devel] [PATCH] tools/libxencall: enforce proper alignment of hypercall buffers

[Juergen Gross <jgross@xxxxxxxx>]

On 08/06/18 12:09, Andrew Cooper wrote:
> On 08/06/18 10:51, Juergen Gross wrote:
>> xencall_alloc_buffer() is used throughout Xen tools for allocating
>> hypercall buffers. Allocation is done at page granularity. For simple
>> administration each allocated set of pages contains a small header
>> holding the number of pages of that set. The hypercall buffer is
>> located directly after the 4 byte sized header, leading to a wrong
>> alignment for e.g. pointers.
>>
>> Repair that by using a 16 byte sized header enforcing the same
>> alignment as malloc().
>>
>> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
>> ---
>> Not sure whether this wants to be backported. In case the answer is
>> "yes" I think it should be part of 4.11.
>> ---
>>  tools/libs/call/buffer.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/tools/libs/call/buffer.c b/tools/libs/call/buffer.c
>> index 2d8fc29ac6..0b6af2db60 100644
>> --- a/tools/libs/call/buffer.c
>> +++ b/tools/libs/call/buffer.c
>> @@ -151,6 +151,7 @@ void xencall_free_buffer_pages(xencall_handle *xcall, 
>> void *p, size_t nr_pages)
>>  
>>  struct allocation_header {
>>      int nr_pages;
> 
> This hunk reveals another bug...
> 
> There are a rather large number of size_t => int truncations in the
> code, which is surely going to catch people by surprise.  (Again - I was
> certain I commented on this during the original review of this library.)

On the practical side I don't think this really matters here. Passing
more than 2GB as a hypercall buffer is nothing we want to support...

> 
>> +    int pad[3];
> 
> This brings it to 16 byte alignment, but is that enough?  An xsave state
> block in x86 for example has 64 byte alignment.  I suppose we mostly
> care about ARM here, as its memcpy() will be most impacted by this
> misalignment.

Special purpose buffers can still be directly allocated via
xencall_alloc_buffer_pages(). I don't think we'll need alignment better
than malloc().

> Irrespective, might be easier to accomplish with an
> __attribute__((__aligned__($N))) rather than explicit padding?

I don't mind which way to go here. I'll leave the decision for the
maintainers. :-)


Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel