Re: [CHECKER] repetitive/contradictory comparison bugs for 2.4.7

[Alex Bligh - linux-kernel <linux-kernel@alex.org.uk>]

>> > other 10 are questionable.  Those 10 are all simple variations on the
>> > following code:
>> >
>> > Start --->
>> > 	if (!tmp_buf) {
>> > 		page = get_free_page(GFP_KERNEL);
>> >
>> > Error --->
>> > 		if (tmp_buf)
>> > 			free_page(page);
>> > 		else
>> > 			tmp_buf = (unsigned char *) page;
>> > 	}
>>
>> That one is not a bug. The serial drivers do this to handle a race.
>> Really it should be

May be I'm being dumb here, and without wishing to open the 'volatile'
can of worms elsewhere, but:

   static char * tmp_buf;

How will this be guaranteed to help handle a race, when gcc is
likely either to have tmp_buf in a register (not declared
volatile), or perhaps even optimize out the second reference.
Seems to me (and I may well be wrong), either there is a
race thread (tmp_buf being assigned between the first
test and grabbing the page), in which case as tmp_buf may
be in a register, it doesn't avoid the race (and potentially
stomps on the existing buffer), or there is not a race, in
which case the second check is unnecessary. IE the checker
found a real bug.

--
Alex Bligh