PCI Error Recovery API Proposal. (WAS:: [PATCH/RFC] PCI ErrorRecovery)

[Benjamin Herrenschmidt]

On Wed, 2005-03-16 at 15:01 -0800, Nguyen, Tom L wrote:
> On Wednesday, March 16, 2005 1:40 PM Paul Mackerras wrote:
> >> This API does not support PCI Express AER precise errors. I prefer to
> >> have param int message being replaced by union error_src structure as
> >> below to include PCI Express AER precise errors.
> >
> >I think you are misunderstanding the purpose of the "message"
> >parameter.  It is not there to give you details of the error that
> >occurred, it is there to tell the driver what stage of the recovery
> >process we are up to.  The details of the error would be reported
> >through an io_check_error() or similar interface.
> 
> How does an io_check_error() support PCI Express comprehensive error
> information? Would you please explain it to me? Do you think there is an
> overlap between error_handler and io_check_error usages when dealing
> with notifying the driver of an error occurred?

Those are two different things. One is when you have a bunch of IOs, to
be able to check wether an error occurred in there. Especially useful on
a driver that is sort-of waiting for something or looping around
something and is suddenly getting ff's. Also read closely my proposed
API for the case where a driver tries to recover. THat part must be
"protected", that is an IO error done in that part must not lead to a
new error message sent to the driver but should be reported
synchronously to the caller.

Wether the error information is "comprehensive" or not is unrelated to
the discussion :) The idea, as I explained, is to provide an opaque
error token with functions to eventually extract details.

Ben.