Re: [PATCH] [13/16] HWPOISON: The high level memory error handlerin the VM v5

[Wu Fengguang]

On Wed, Jun 10, 2009 at 07:03:05PM +0800, Nick Piggin wrote:
> On Wed, Jun 10, 2009 at 05:20:11PM +0800, Wu Fengguang wrote:
> > On Wed, Jun 10, 2009 at 04:59:39PM +0800, Nick Piggin wrote:
> > > On Wed, Jun 10, 2009 at 04:38:03PM +0800, Wu Fengguang wrote:
> > > > On Wed, Jun 10, 2009 at 12:05:53AM +0800, Hugh Dickins wrote:
> > > > > I think a much more sensible approach would be to follow the page
> > > > > migration technique of replacing the page's ptes by a special swap-like
> > > > > entry, then do the killing from do_swap_page() if a process actually
> > > > > tries to access the page.
> > > > 
> > > > We call that "late kill" and will be enabled when
> > > > sysctl_memory_failure_early_kill=0. Its default value is 1.
> > > 
> > > What's the use of this? What are the tradeoffs, in what situations
> > > should an admin set this sysctl one way or the other?
> > 
> > Good questions.
> > 
> > My understanding is, when an application is generating data A, B, C in
> > sequence, and A is found to be corrupted by the kernel. Does it make
> > sense for the application to continue generate B and C? Or, are there
> > data dependencies between them? With late kill, it becomes more likely
> > that the disk contain new versions of B/C and old version of A, so
> > will more likely create data inconsistency.
> > 
> > So early kill is more safe.
> 
> Hmm, I think that's pretty speculative, and doesn't seem possible for
> an admin (or even kernel programmer) to choose the "right" value.
> 

Agreed. It's not easy to choose if I'm myself an admin ;)

> The application equally may not need to touch the data again, so
> killing it might cause some inconsistency in whatever it is currently
> doing.

Yes, early kill can also be evil. What I can do now is to document the
early kill parameter more carefully.

Thanks,
Fengguang

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/