Re: floppy driver 2.6.3 question

[Jens Axboe <axboe@suse.de>]

On Thu, Mar 18 2004, Peter T. Breuer wrote:
> > > know.  128 requests have just previously been errored due to readahead
> > > and the check_media_changed result setting the driver request function
> > > to error out requests.  Perhaps we have run out of requests (they're
> > > all put_ as far as I can see). Maybe the block layers get tired of
> > > talking to a device that errors requests. I'm just feeling my way!
> > > Any wild ideas are welcome!
> > 
> > If the 129th request fails, then that's a pretty good clue that you
> > aren't getting the requests freed. Perhaps you are overwriting something
> > in the request after allocating it? Always mask change ->flags (don't
> > just set it), and don't overwrite ->rl.
> 
> Good idea. rl is inviolate, but I set at least |=REQ_NOMERGE sometimes
> on flags. And I pass ioctl information in fake requests by setting

May I ask on what commands you set that bit?

> the bit just beyond the edge of those currently used (__REQ_BITS) to
> indicate its an ioctl and treating it specially in end request. Maybe
> on error I forgot to remove the extra bit before doing put_blk_request

Ugh, that sounds like very bad practice... The 'standard' way of doing
something like that is to flag REQ_SPECIAL and put whatever structure
you want in ->special.

-- 
Jens Axboe