[uClinux-dev] [PATCH] : Avoid filename < TASK_SIZE test in do_getname() when no MMU

Philippe De Muyter phdm at macqel.be
Wed May 26 04:21:31 EDT 2010


Hi Greg,

On Wed, May 26, 2010 at 12:54:49PM +1000, Greg Ungerer wrote:
>
> Hi Philippe,
>
> Philippe De Muyter wrote:
>> Hi Greg,
>> On Tue, May 25, 2010 at 11:19:43AM +1000, Greg Ungerer wrote:
>>> Hi Philippe,
>>>
>>> Philippe De Muyter wrote:
>>>> On Mon, May 24, 2010 at 11:29:50AM +1000, Greg Ungerer wrote:
>>>> [...]
>>>>> +#else
>>>>> +#define TASK_SIZE	(0xFFFFFFFFUL)
>>>>> +#endif
>>>> Because of do_getname() :
>>>> 	len = TASK_SIZE - (unsigned long) filename;
>>>> we should rather have
>>>> 	#define TASK_SIZE (0x100000000ull)
>>> I see what you mean. But in practice here I don't think it matters.
>> Can no process have its stack allocated in the last block, and hence have 
>> some
>> argv[i] put in the last addresses, with the terminating '\0' at 0xffffffff 
>> ?
>
> I think that is possible. But the "in practice" part is that
> I don't know of any m68knommu platforms that actually map the RAM
> right at the very top of the 32bit address space - so that it ends
> at 0xffffffff.

OK then (I had missed/forgotten that your patch was for m68knommu only)

Best regards

Philippe

PS: I have already been bitten (in another part of the kernel) by a bug
that refused a zone containing 0xffffffff (see http://lkml.org/lkml/2009/12/6/204)

-- 
Philippe De Muyter  phdm at macqel dot be  Tel +32 27029044
Macq Electronique SA  rue de l'Aeronef 2  B-1140 Bruxelles  Fax +32 27029077



More information about the uClinux-dev mailing list