Re: [Qemu-devel] e1000 memory corruption in guest OS

[Alexey Kardashevskiy <aik@ozlabs.ru>]

On 03/03/2014 07:33 PM, Paolo Bonzini wrote:
> Il 03/03/2014 02:58, Alexey Kardashevskiy ha scritto:
>> On 03/02/2014 08:31 AM, Paolo Bonzini wrote:
>>> Il 01/03/2014 13:30, Alexey Kardashevskiy ha scritto:
>>>>>>
>>>>>> Corrupted DMA buffer is 0x e00000 -- 0x7f15c313f000.
>>>>>> The e1000 packet is at  0x12ffac2 -- 0x7f15c313eac2.
>>>>>>
>>>>>> (0x7f15c313f000 - 0x7f15c313eac2) = 0x53e which is less than 0x5aa and
>>>>>> (0x5aa - 0x53e) = 0x6c bytes get corrupted.
>>>>>>
>>>>>> I see here buffer overrun from e1000 and I suspect that your patch just
>>>>>> hides this problem. What did I miss?
>>>>
>>>> Ping, anyone?
>>>
>>> You missed that this is a Xen-specific problem.  Xen maps things a page at
>>> a time, so address_space_map/unmap/rw can operate only on a small part of
>>> the requested [address, address+length) range.
>>
>> Sorry, I am not following you here. Does KVM map things not page-aligned?
> 
> Look in exec.c for xen_enabled().  Xen's implementation of
> address_space_map/unmap is completely different.

Honestly cannot see much difference in the current QEMU...

> 
>>> So there is no overrun in e1000.  The patch is incomplete, because it fixes
>>> only address_space_rw, but the problem is indeed in exec.c.
>>
>> So you know what the problem is? We have a bunch of bugreports against
>> e1000 breaking things...
> 
> This one had been reported and fixed already.

Any hint? :) Could not spot any related to overrun except:
a0ae17a 8 months ago Andrew Jones e1000: cleanup process_tx_desc
but commit message says "complaints are false positives"...


-- 
Alexey