Re: Asynch I/O broken in 2.2.15

[Jeff V. Merkey]

I found part of the problem after putting about 200 printk() calls
throughout the kernel with ll_rw_blk.  I am surprised you guys did not
see it when I posted the code.  init_waitqueue_head() MUST be called
when creating your own buffer heads -- if it's not inited, in crashes on
2.2.15 with locally allocated buffer heads, and on 2.3.99-Pre5, when the
DEBUG_SPINLOCKS flag is set, you get "bad magic number" in spinlock.h
line 80.

As for the getblk() problem on 2.2.15, this still seems to have some
problems when sending more than 4000 IO requests to it at once.

Problem fixed.  Next....

Jeff

"Jeff V. Merkey" wrote:
> 
> "Jeff V. Merkey" wrote:
> >
> >
> 
> Stephen,
> 
> I tried locking the pages with PG_locked, and guess what?  It still
> crashes when I use the linux buffer cache routines with getblk().  I
> think I am hitting the maximum limit for outstanding IO requests?
> 
> The Ooops is in the swapper process again.
> 
> Bottom line, Asynch IO is busted in 2.2.15 and does not appear to be
> useable unless I limit the number of requests posted at one time to
> under 5.
> 
> Jeff