Re: PF_MEMALLOC in wakeup_bdflush()

[Steve Lord <lord@xxxxxxx>]

> 
> Steve,
> 
> I'm looking into the 2.4.1 XFS tree and I noticed you're setting the
> PF_MEMALLOC bit in the task flags before calling flush_dirty_buffers(),
> inside wakeup_bdflush().
> 
> Have you actually seen any oom deadlock caused by flush_dirty_buffers() or
> you just added the bit there to be safe?
> 

I am? Not in the code I am looking at here:

void wakeup_bdflush(int block)
{
        if (current != bdflush_tsk) {
                wake_up_process(bdflush_tsk);

                if (block)
                        flush_dirty_buffers(0);
        }
}

The code in the cvs tree I have does not do that either, you either
have a modified tree, or I am misinterpreting where you mean in
the code. The only PF_MEMALLOC I messed with was in page_buf_io.c
out of the writepage method.

Steve