RE: 1.3.93, loop better but still broke
Ray_Van_Tassle-CRV004@email.mot.com
24 Apr 96 07:57:24 -0500
> From: gadio@netvision.net.il@INTERNET on Tue, Apr 23, 1996 9:36 PM
> On 23 Apr 1996 Ray_Van_Tassle-CRV004@email.mot.com wrote:
>
> > In 1.3.93, loop (use a file for a filesystem) works better, but still
locks
> > up. Before 91, it would lock up in less than a minute.
>
> The basic problem seems to be that we have a dependency between two block
> devices, yet some resources are shared between all drivers (the unplug
task
> queue and the empty requests pool).
Dependency--yes. But it's not those two things. In my experimentation, I've
eliminated them, but it still locks up.
>
> Does the following tiny patch help ? (it is not really mine - it was
> introduced around 1.3.78 for MD, when it was still a block device driver).
>
> Gadi Oxman
> gadio@netvision.net.il
>
> --- ll_rw_blk.c.old Wed Apr 24 05:08:07 1996
> +++ ll_rw_blk.c Wed Apr 24 05:09:18 1996
> @@ -335,6 +335,11 @@
> return;
> }
>
> +#ifdef CONFIG_BLK_DEV_LOOP
> + if (major == LOOP_MAJOR)
> + max_req /= 2;
> +#endif /* CONFIG_BLK_DEV_LOOP */
> +
> /* look for a free request. */
I (independently) did this myself. And, no, it does not solve the lockup.
-30- Ray
=========================
I hope I solve this before _all_ my hair gets pulled out!