lkml.org 
[lkml]   [2002]   [Mar]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: 2.4.19pre3aa2
On Fri, Mar 15, 2002 at 07:35:02PM +0200, Jari Ruusu wrote:
> Jens Axboe wrote:
> > On Fri, Mar 15 2002, Jari Ruusu wrote:
> > > - No more illegal sleeping in generic_make_request().
> >
> > I've told you this before -- sleeping in make_request is not illegal,
> > heck it happens _all the time_. Safely sleeping requires a reserved pool
> > of the units you wish to allocate, of course. In fact I think that would
> > be much nicer than the path you are following here by delaying
> > allocations to the loop thread (and still not using a reserved pool).
>
> Yes, I know you have told me that before, but I'm being overcareful. See:
>
> <quote> from device drivers book by Alessandro Rubini, chapter 12, page 331
> The request function has one very important constraint: it must be atomic.
> request is not usually called in direct response to user requests, and it is
> not running in the context of any particular process. It can be called at
> interrupt time, from tasklets, or from any number of other places. Thus, it
> must not sleep while carrying out its tasks.
> </quote>

loop isn't implement via ->request_fn anymore. Loop since 2.4 is only
driven by the ->make_request_fn, that for the other more normal devices
just means the old legacy __make_request. request_fn is subject to the
rules pointed out by Alessandro, but ->make_request_fn can sleep just
like ll_rw_block can sleep. ->make_request_fn and in turn
loop_make_request can only run in normal context with irq enabled and
they're both allowed to sleep just like ll_rw_block and submit_bh.

Nevertheless as Jens said the infinite-loop-allocation in the
->make_request_fn path are deadlock prone at the moment, not because
they sleeps but because they need a reserved mempool to guarantee
operations can go ahead slowly without deadlocks even if dynamic
allocation fails, but this is not a very pratical problem, it's very
unlikely to deadlock there (it's not worse than the other infinite loop
in getblk() that affects not just loop).

Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:25    [W:0.122 / U:2.692 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site