can generic_make_request in 2.4 kernel sleep?

can generic_make_request in 2.4 kernel sleep?

[Ming Zhang]

Hi, I have a quick question.

In 2.4 kernel, can generic_make_request sleep?

I read the 2.4.24 kernel and find that there is printk in it. so i
assume it can sleep. if so, why the per queue make_request_fn can not
sleep base on the LDD book? and i do not see any place that the
io_request_lock is held. is this lock removed from system? or we do not
need this lock at this place any more?

and in md code, it use blk_queue_make_request to use its own request_fn
instead a queue, for example, in raid1_make_request(), it calls
raid1_alloc_r1ch() which also call schedule() if need, then this own
request_fn can sleep?

i think i am little confused about the LDD (linux device driver 2nd)
book and the new kernel code. can anybody point out some latest
reference for me?

thanks a bunch.


ming

Re: can generic_make_request in 2.4 kernel sleep?

[Jens Axboe]

On Sat, Feb 28 2004, Ming Zhang wrote:
> Hi, I have a quick question.
> 
> In 2.4 kernel, can generic_make_request sleep?

yes

> I read the 2.4.24 kernel and find that there is printk in it. so i
> assume it can sleep. if so, why the per queue make_request_fn can not
> sleep base on the LDD book? and i do not see any place that the
> io_request_lock is held. is this lock removed from system? or we do not
> need this lock at this place any more?
> 
> and in md code, it use blk_queue_make_request to use its own request_fn
> instead a queue, for example, in raid1_make_request(), it calls
> raid1_alloc_r1ch() which also call schedule() if need, then this own
> request_fn can sleep?
> 
> i think i am little confused about the LDD (linux device driver 2nd)
> book and the new kernel code. can anybody point out some latest
> reference for me?

->make_request_fn() is not entered with the queue/io_request_lock. It
can sleep, it's not recommended since on writeout you are blocking
kswapd/bdflush.

-- 
Jens Axboe