gcpu_amd.esc revision 20c794b39650d115e17a15983b6b82e46238cf45
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright 2007 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
/*
* Eversholt rules for generic AMD with on-chip memory-controller(s), as seen
* in AMD family 0xf and 0x10.
*
* In the absence of any model-specific support, any memory errors that
* are observed via MCA (typically through an on-chip memory-controller)
* will surface as ereport.cpu.generic-x86.bus_interconnect_memory[_uc]
* ereports and are diagnosed via generic rules in gcpu.esc.
*
* If full model-specific support is available, including full NorthBridge
* support, then memory ereports will surface in a more-specific subclass
* such as ereport.cpu.amd.mem_ce; these are diagnosed in amd64.esc.
*
* In the case where some "vendor generic" support is present, memory errors
* are reported as ereport.cpu.generic-x86.mem_{ce,ue} and include a
* syndrome and syndrome-type, and usually also a resource FMRI to identify
* the affected resource. In the AMD case a resource FMRI is included for
* those chip versions that include an Online Spare Control register; this
* register provides counts of ECC errors seen per channel and chip-select
* on a NorthBridge node. The resource FMRI has form
* in these cases.
*/
/*
* The number of pages that must be faulted on a chip-select for repeated
* correctable errors before we will consider one of the component dimms
* faulty.
*/
/*
* The maximum number of pages we will diagnose as faulty on any one
* chip-select (must be at least CS_PAGEFLT_THRESH). If a chip-select
* has a fault that will affect zillions of pages this limit stops us
* diagnosing excessive numbers of page faults.
*/
/*
* SERD paramters for individual page faults. When more than PAGE_SB_COUNT
* correctable ereports are experienced on a single chip-select within
* PAGE_SB_TIME the engine will fire and we will fault the most recent
* page.
*/
/*
* ADDR_VALID is true if the ereport payload includes IA32_MCi_ADDR.
*/
/*
* CONTAINS_CS is true if the resource nvlist array exists and one of its
* members matches the chip-select path. This is used to constrain
* propogations to those for which a resource element matches the
* chip-select path of the propogation. This is necessary because the
* detector element of memory ereports is a cpu and not the chip-select itself.
*/
/* Generic memory ereports. */
/*
* ========= Propogations for correctable page faults ============
* | |
* | Discard mem_ce with no resource in the ereport payload. |
* | Discard mem_ce with no address info - we can't fault the |
* | corresponding page without it. |
* | |
* | the payload detector info) whose resource payload member |
* | includes a chip/memory-controller/dram-channel/chip-select |
* | (CSPATH) for the same chip number, diagnose to an upset event |
* | associated with a per-CSPATH SERD engine as long as we are |
* | below the page fault limit for this CSPATH (defined below); |
* | if we are over that limit then discard the event since we |
* | will already have faulted a dimm and there is no point in |
* | continuing to diagnose endless page faults from a dimm with |
* | something like a pin failure. |
* | |
* | When the per-CSPATH SERD engine fires we fault the page |
* | containing the address included in the ereport that caused |
* | the trip, and increment a per-CSPATH counter to count page |
* | faults on that chip-select from repeated correctable errors. |
* | |
* | A mem_ue ereport produces an immediate page_ue fault. |
* |===============================================================|
*/
/* Counter for page faults diagnosed on a chip-select */
/* Page fault event for repeated correctable errors */
/* Upset to diagnose correctable ereports to */
/* Synthetic ereport generated when page_ce SERD engine trips */
/* SERD engine for each chip-select */
N=PAGE_SB_COUNT, T=PAGE_SB_TIME,
/* Upset to discard events to when we're over limit */
/*
* Discard ereports with no resource or no address info
*/
/*
* For as long as we are below the page fault limit diagnose correctable ereport
* observations as upsets to feed the SERD engine.
*/
/*
* Discard ereports if we are above the page fault limit on this chip-select,
*/
/* Diagnose a page fault when the pagefault SERD engine trips */
/* Include address info in the page fault diagnosed, for rewrite-ASRU */
/*
* ========= Propogations for correctable DIMM faults ============
* | |
* | A dimm_ce fault is diagnosed when we have faulted an |
* | excessive number of page_ce faults on a chip-select - more |
* | than CE_DIMMSB_THRESH. |
* | |
* | A dimm_ue fault is diagnosed on the first uncorrectable |
* | ereport from a chip-select. |
* |===============================================================|
*/
/* DIMM fault event for CE failures */
/*
* This upset is diagnosed in parallel with upset.memory.generic-x86.page_ce
* on the CSPATH, and the associated SERD engine has the same parameters
* as serd.memory.generic-x86.page_ce@CSPATH so they fire at the same time.
* When this one fires we check whether we have reached the diagnosis
* threshold for a dimm_ce.
*/
N=PAGE_SB_COUNT, T=PAGE_SB_TIME,
{ CS_DIMMSB_THRESH_REACHED } (0)->
{ !CS_DIMMSB_THRESH_REACHED } (0)->
/*
* ========= Propogations for uncorrectable page faults ==========
* | |
* | A UE produces an immediate page fault. But we also want a |
* | corresponding dimm fault and since we do not like multi-entry |
* | suspect lists we arrange two distinct fault management |
* | exercises by diagnosing a mem_ue to two upset events that |
* | feed instant-trip SERD engines. Yuck. |
* |===============================================================|
*/
/* Page fault event for uncorrectable errors */
/* Upset for page fault */
/* Synthetic erport generated when the page_ue SERD engine trips */
/* Instant-trip engine for page fault */
N=0, T=1h, /* trip on first upset */
/* Discard events with no address info */
/* Diagnose a page_ue upset on a mem_ue event */
/* On the immediate SERD trip diagnose a page fault */
/* Include address info in the page fault diagnosed, for rewrite-ASRU */
/*
* ========= Propogations for uncorrectable dimm faults ==========
* | |
* | A UE produces an immediate dimm fault. As explained in the |
* | page_ue block comment above we split the exercise in two in |
* | order to produce independent page_ue and dimm_ue diagnoses. |
* |===============================================================|
*/
/* Dimm fault for an uncorrectable error */
/* Upset for dimm fault */
/* Sythetic ereport generated when the dimm_ue SERD engine trips */
/* Instant-trip engine for dimm fault */
N=0, T=1h, /* trip on first upset */
/* Diagnose a dimm_ue upset on a mem_ue event (in addition to page_ue upset) */
{ CONTAINS_CS } (0)->
/* On the immediate SERD trip diagnose a dimm fault */
/*
* ========= Propogations for GART Table Walk Errors =============
* | |
* | These are usually due to software mis-programming of the GART |
* | TLB rather than from hardware errors. It would be incorrect |
* | to fault and potentially offline a cpu in response to these |
* | so they have their own fault class to facilitate us ignoring |
* | them. |
* |===============================================================|
*/