Re: [PATCH][1/2]page_fault retry with NOPAGE_RETRY

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, 10 Apr 2009, Wu Fengguang wrote:

> On Fri, Apr 10, 2009 at 02:02:05PM +0800, Andrew Morton wrote:
> 
> > Can we please redo this as:
> > 
> > 
> > 	int write;
> > 	unsigned int flags;
> > 
> > 	/*
> > 	 * Big fat comment explaining the next three lines goes here
> > 	 */
> 
> Basically it's doing a
>         (is_write_access  | FAULT_FLAG_RETRY) =>
>         (FAULT_FLAG_WRITE | FAULT_FLAG_RETRY)
> by extracting the bool part:
> > 	write = write_access & ~FAULT_FLAG_RETRY;
> convert bool to a bit flag:
> > 	unsigned int flags = (write ? FAULT_FLAG_WRITE : 0);

The point is, we shouldn't do that. 

Your code is confused, because it uses "write_access" as if it had the old 
behaviour (boolean to say "write") _plus_ the new behavior (bitmask to say 
"retry"), and that's just wrong.

Just get rid of "write_access" entirely, and switch it over to something 
that is a pure bitmask.

Yes, it means a couple of new preliminary patches that switch all callers 
of handle_mm_fault() over to using the VM_FLAGS, but that's not a big 
deal.

I'm following up this email with two _example_ patches. They are untested, 
but they look sane. I'd like the series to _start_ with these, and then 
you can pass FAULT_FLAGS_WRITE | FAULT_FLAGS_RETRY down to 
handle_mm_fault() cleanly.

Hmm? Note the _untested_ part on the patches to follow. It was done very 
mechanically, and the patches look sane, but .. !!!

			Linus

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>