Re: [PATCH 1/3] x86, ras: Add new infrastructure for machine check fixup tables

["Luck, Tony" <tony.luck@intel.com>]

On Thu, Nov 12, 2015 at 12:04:36PM -0800, Andy Lutomirski wrote:
> > We already have code to recover from machine checks encountered
> > while the processor is executing ring3 code.
> 
> I meant failures during copy_from_user, copy_to_user, etc.

Yes.  copy_from_user() will be pretty interesting from a coverage point
of view.  We can recover by sending a SIGBUS to the process just like
we would have if the process had accessed the data directly rather than
passing the address to the kernel to acccess it.

copy_to_user() is a lot harder. The machine check is on the kernel side
of the copy. If we are copying from page cache as part of a read(2)
syscall from a regular file we can probably nuke the page from the cache
and return -EIO to the user.  Other cases may be possible, but I don't
immediately see any way to do it as a general case.

-Tony