On Tue, Nov 06, 2012 at 12:12:20PM -0800, Andrew Morton wrote:
> On Tue, 6 Nov 2012 00:07:53 -0500
> Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx> wrote:
>
> > On Mon, Nov 05, 2012 at 02:01:54PM -0800, Andrew Morton wrote:
> > > On Fri, 2 Nov 2012 12:33:13 -0400
> > > Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx> wrote:
> > >
> > > > Printing out the information about which file can be affected by a
> > > > memory error in generic_error_remove_page() is helpful for user to
> > > > estimate the impact of the error.
> > > >
> > > > Changelog v2:
> > > > - dereference mapping->host after if (!mapping) check for robustness
> > > >
> > > > ...
> > > >
> > > > --- v3.7-rc3.orig/mm/truncate.c
> > > > +++ v3.7-rc3/mm/truncate.c
> > > > @@ -151,14 +151,20 @@ int truncate_inode_page(struct address_space *mapping, struct page *page)
> > > > */
> > > > int generic_error_remove_page(struct address_space *mapping, struct page *page)
> > > > {
> > > > + struct inode *inode;
> > > > +
> > > > if (!mapping)
> > > > return -EINVAL;
> > > > + inode = mapping->host;
> > > > /*
> > > > * Only punch for normal data pages for now.
> > > > * Handling other types like directories would need more auditing.
> > > > */
> > > > - if (!S_ISREG(mapping->host->i_mode))
> > > > + if (!S_ISREG(inode->i_mode))
> > > > return -EIO;
> > > > + pr_info("MCE %#lx: file info pgoff:%lu, inode:%lu, dev:%s\n",
> > > > + page_to_pfn(page), page_index(page),
> > > > + inode->i_ino, inode->i_sb->s_id);
> > > > return truncate_inode_page(mapping, page);
> > > > }
> > > > EXPORT_SYMBOL(generic_error_remove_page);
> > >
> > > A couple of things.
> > >
> > > - I worry that if a hardware error occurs, it might affect a large
> > > amount of memory all at the same time. For example, if a 4G memory
> > > block goes bad, this message will be printed a million times?
> >
> > If the error on 4G memory block triggered by SRAO MCE and these 1M pages
> > are all pagecache pages, the answer is yes.
>
> Well that's bad.
>
> > But I think that if it's a whole DIMM error, it should be reported by
> > another type of MCE than SRAO, so printing a million times seems to be
> > unlikely to happen.
>
> "should be" and "unlikely" aren't very reassuring things to hear!
> Emitting a million lines into syslog is pretty poor behaviour and
> should be reliably avoided.
So capping maximum lines of messages per some duration (a hour or a day)
is a possible option. BTW, even if we don't apply this patch, the kernel
can emit million lines of messages in the above-mentioned situation because
each memory error event emits a message like "MCE 0x3f57f4: dirty LRU page
recovery: Ignored" on syslog. If it's also bad, we need to do capping
also over existing printk()s, right?
> > > - hard-wiring "MCE" in here seems a bit of a layering violation?
> > > What right does the generic, core .error_remove_page() implementation
> > > have to assume that it was called because of an MCE?
> >
> > OK, we need not assume that. I change "MCE " prefix to more specific
> > one like "Memory error ".
> >
> > > Many CPU types don't eveh have such a thing?
> >
> > No. At least currently, only SRAO MCE triggers memory_failure() and
> > it's defined only on some newest highend models of Intel CPUs.
>
> Again, your reply is full of assumptions about one particualar
> implementation on one particular CPU. But this is generic,
> cross-architecture code!
>
> Now, it's pretty harmless to make these assumptions at this time. But
> this new code will need to redone if/when other CPU types come along,
> and because there's a printk in there, that rework will cause
> user-visible changes in kernel behaviour. It would be best if we can
> just avoid the problem on day one.
>
> Maybe move the printk into x86-specific code? And just one printk
> please - not a million!
OK, we need some cleanup for this at first.
Thanks,
Naoya
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>