[RFE] Add stacked registers to SAL records

[RFE] Add stacked registers to SAL records

[Keith Owens]

SAL error records contain most registers in sal_processor_static_info,
but not the stacked registers.  What is the process for adding another
GUID to the SAL specification to contain the stacked registers?

Layout:

  number of stacked registers in this record
  cfm for frame at the time of failure
  stacked registers, ending with the cfm.sof registers

This GUID would be optional, if a particular SAL implementation did not
produce it then that platform would not have stacked registers printed
via salinfo.

[RFE] Add stacked registers to SAL records (resend)

[Keith Owens]

Originally sent November 13 with no response, try again.

SAL error records contain most registers in sal_processor_static_info,
but not the stacked registers.  What is the process for adding another
GUID to the SAL specification to contain the stacked registers?

Layout:

  number of stacked registers in this record
  cfm for frame at the time of failure
  stacked registers, ending with the cfm.sof registers

This GUID would be optional, if a particular SAL implementation did not
produce it then that platform would not have stacked registers printed
via salinfo.

Re: [RFE] Add stacked registers to SAL records (resend)

[Bjorn Helgaas]

On Thursday 04 December 2003 5:59 pm, Keith Owens wrote:
> SAL error records contain most registers in sal_processor_static_info,
> but not the stacked registers.  What is the process for adding another
> GUID to the SAL specification to contain the stacked registers?

I've always been puzzled about why the stacked registers aren't
available after an INIT, so this proposal sounds incredibly useful
to me.  Much of the useful information, i.e., function arguments,
is in the stacked registers and hence currently unavailable.

As for getting it added to the spec, I don't really know the process.
But I'll forward this to our people who deal with the spec.  I think
it would help to have an actual draft of a proposal.

Tony Luck might have a good Intel contact, too.

Bjorn

RE: [RFE] Add stacked registers to SAL records (resend)

[Luck, Tony]

> -----Original Message-----
> From: linux-ia64-owner@vger.kernel.org
> [mailto:linux-ia64-owner@vger.kernel.org]On Behalf Of Bjorn Helgaas
> Sent: Friday, December 05, 2003 8:29 AM
> To: Keith Owens; linux-ia64@vger.kernel.org
> Subject: Re: [RFE] Add stacked registers to SAL records (resend)
> 
> 
> On Thursday 04 December 2003 5:59 pm, Keith Owens wrote:
> > SAL error records contain most registers in 
> sal_processor_static_info,
> > but not the stacked registers.  What is the process for 
> adding another
> > GUID to the SAL specification to contain the stacked registers?
> 
> I've always been puzzled about why the stacked registers aren't
> available after an INIT, so this proposal sounds incredibly useful
> to me.  Much of the useful information, i.e., function arguments,
> is in the stacked registers and hence currently unavailable.
> 
> As for getting it added to the spec, I don't really know the process.
> But I'll forward this to our people who deal with the spec.  I think
> it would help to have an actual draft of a proposal.
> 
> Tony Luck might have a good Intel contact, too.

I'm looking around for the right person to pass this to.

-Tony

Re: [RFE] Add stacked registers to SAL records (resend)

[Bjorn Helgaas]

On Thursday 04 December 2003 5:59 pm, Keith Owens wrote:
> SAL error records contain most registers in sal_processor_static_info,
> but not the stacked registers.  What is the process for adding another
> GUID to the SAL specification to contain the stacked registers?

One of our architecture guys pointed out that the stacked registers
should still be intact at the OS_MCA and OS_INIT handoffs.  They
are unchanged in the PALE_CHECK and PALE_INIT exit state (I finally
found this buried in the RSE description), and the OS_MCA and OS_INIT
handoff state is defined to be the same for stacked registers.

So I assume your proposal is to address the fact that we don't
actually save the error records until after a reboot, and the
stacked register contents are gone by then.  Right?

Bjorn

Re: [RFE] Add stacked registers to SAL records (resend)

[Keith Owens]

On Mon, 8 Dec 2003 11:41:32 -0700, 
Bjorn Helgaas <bjorn.helgaas@hp.com> wrote:
>On Thursday 04 December 2003 5:59 pm, Keith Owens wrote:
>> SAL error records contain most registers in sal_processor_static_info,
>> but not the stacked registers.  What is the process for adding another
>> GUID to the SAL specification to contain the stacked registers?
>
>One of our architecture guys pointed out that the stacked registers
>should still be intact at the OS_MCA and OS_INIT handoffs.  They
>are unchanged in the PALE_CHECK and PALE_INIT exit state (I finally
>found this buried in the RSE description), and the OS_MCA and OS_INIT
>handoff state is defined to be the same for stacked registers.
>
>So I assume your proposal is to address the fact that we don't
>actually save the error records until after a reboot, and the
>stacked register contents are gone by then.  Right?

Partly.  Bu mainly because we need a new SAL error record section with
its own GUID and an agreed format.

You mentioned in another mail "Someone also raised the technical
question of whether the new format included enough information to
relate it to a particular processor, i.e., maybe it should include the
LID".  We already have the lid, in sal_log_processor_info.