On Thu, 15 Mar 2001, Marcelo Tosatti wrote:
> The old create_bounce code used to set PF_MEMALLOC on the task flags
> and call wakeup_bdflush(1) in case GFP_BUFFER page allocation failed.
> That was broken because flush_dirty_buffers() could try to flush a
> buffer pointing to highmem page, which would end up in create_bounce
> again, but with PF_MEMALLOC.
>
> Have you tried to make flush_dirty_buffers() only flush buffers
> pointing to lowmem pages in case the caller wants it to do so?
this makes sense too - although an emergency pool of some sort never
hurts, given that highmem buffers cannot be written out without allocating
bounce buffers. (this makes them more volatile wrt. resource shortages
than lowmem buffers.) Also, there is no guarantee that flushing lowmem
buffers yields any free pages.
> This way you can call flush_dirty_buffers() with the guarantee you're
> going to free useful (lowmem) memory. This also throttles high mem
> writes giving priority to low mem ones.
yep, i think we should do this in addition to the emergency pool thing, it
should improve balance.
Ingo
-
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/
This archive was generated by hypermail 2b29 : Fri Mar 23 2001 - 21:00:08 EST