[RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Borislav Petkov]

Hi,

thanks to all reviewers of the previous submission, here is the second
version of this series.

Highlights are the addition of two helpers to read/write MSRs on several
CPUs, denoted by a cpumask and using an array of MSR values per-CPU, as
Peter suggested. Since IMHO they look generic enough I've added them to
arch/x86/lib/msr-on-cpu.c (now renamed to msr.c).

Moreover, I've addressed all the issues raised from the previous series.
Please let me know should there be anything else remaining.

Thanks,
Boris.

 arch/x86/include/asm/msr.h |   11 +
 arch/x86/lib/Makefile      |    2 +-
 arch/x86/lib/msr-on-cpu.c  |   97 -
 arch/x86/lib/msr.c         |  151 ++
 drivers/edac/Kconfig       |   26 +
 drivers/edac/Makefile      |    1 +
 drivers/edac/amd64_edac.c  | 5385 ++++++++++++++++++++++++++++++++++++++++++++
 7 files changed, 5575 insertions(+), 98 deletions(-)

[PATCH 01/21] x86: add methods for writing of an MSR on several CPUs

[Borislav Petkov]

- Rename msr-on-cpu.c to msr.c accordingly.

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 arch/x86/include/asm/msr.h |   11 +++
 arch/x86/lib/Makefile      |    2 +-
 arch/x86/lib/msr-on-cpu.c  |   97 ----------------------------
 arch/x86/lib/msr.c         |  151 ++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 163 insertions(+), 98 deletions(-)
 delete mode 100644 arch/x86/lib/msr-on-cpu.c
 create mode 100644 arch/x86/lib/msr.c

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index 638bf62..43d9537 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -2,6 +2,7 @@
 #define _ASM_X86_MSR_H
 
 #include <asm/msr-index.h>
+#include <asm/cpumask.h>
 
 #ifndef __ASSEMBLY__
 # include <linux/types.h>
@@ -216,6 +217,8 @@ do {                                                            \
 #ifdef CONFIG_SMP
 int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
+int rdmsr_on_cpus(const cpumask_t *mask, u32 msr_no, u32 *l, u32 *h);
+int wrmsr_on_cpus(const cpumask_t *mask, u32 msr_no, u32 *l, u32 *h);
 int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
 #else  /*  CONFIG_SMP  */
@@ -229,6 +232,14 @@ static inline int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 	wrmsr(msr_no, l, h);
 	return 0;
 }
+static inline int rdmsr_on_cpus(const cpumask_t *m, u32 msr_no, u32 *l, u32 *h)
+{
+	return rdmsr_on_cpu(msr_no, l, h);
+}
+static inline int wrmsr_on_cpus(const cpumask_t *m, u32 msr_no, u32 *l, u32 *h)
+{
+	return wrmsr_on_cpu(msr_no, l, h);
+}
 static inline int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no,
 				    u32 *l, u32 *h)
 {
diff --git a/arch/x86/lib/Makefile b/arch/x86/lib/Makefile
index 55e11aa..f9d3563 100644
--- a/arch/x86/lib/Makefile
+++ b/arch/x86/lib/Makefile
@@ -2,7 +2,7 @@
 # Makefile for x86 specific library files.
 #
 
-obj-$(CONFIG_SMP) := msr-on-cpu.o
+obj-$(CONFIG_SMP) := msr.o
 
 lib-y := delay.o
 lib-y += thunk_$(BITS).o
diff --git a/arch/x86/lib/msr-on-cpu.c b/arch/x86/lib/msr-on-cpu.c
deleted file mode 100644
index 321cf72..0000000
--- a/arch/x86/lib/msr-on-cpu.c
+++ /dev/null
@@ -1,97 +0,0 @@
-#include <linux/module.h>
-#include <linux/preempt.h>
-#include <linux/smp.h>
-#include <asm/msr.h>
-
-struct msr_info {
-	u32 msr_no;
-	u32 l, h;
-	int err;
-};
-
-static void __rdmsr_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rdmsr(rv->msr_no, rv->l, rv->h);
-}
-
-static void __wrmsr_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	wrmsr(rv->msr_no, rv->l, rv->h);
-}
-
-int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
-	*l = rv.l;
-	*h = rv.h;
-
-	return err;
-}
-
-int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	rv.l = l;
-	rv.h = h;
-	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
-
-	return err;
-}
-
-/* These "safe" variants are slower and should be used when the target MSR
-   may not actually exist. */
-static void __rdmsr_safe_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h);
-}
-
-static void __wrmsr_safe_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h);
-}
-
-int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
-	*l = rv.l;
-	*h = rv.h;
-
-	return err ? err : rv.err;
-}
-
-int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	rv.l = l;
-	rv.h = h;
-	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
-
-	return err ? err : rv.err;
-}
-
-EXPORT_SYMBOL(rdmsr_on_cpu);
-EXPORT_SYMBOL(wrmsr_on_cpu);
-EXPORT_SYMBOL(rdmsr_safe_on_cpu);
-EXPORT_SYMBOL(wrmsr_safe_on_cpu);
diff --git a/arch/x86/lib/msr.c b/arch/x86/lib/msr.c
new file mode 100644
index 0000000..4948de1
--- /dev/null
+++ b/arch/x86/lib/msr.c
@@ -0,0 +1,151 @@
+#include <linux/module.h>
+#include <linux/preempt.h>
+#include <linux/smp.h>
+#include <asm/msr.h>
+
+struct msr_info {
+	u32 msr_no;
+	u32 l, h;
+	int err;
+};
+
+static void __rdmsr_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rdmsr(rv->msr_no, rv->l, rv->h);
+}
+
+static void __wrmsr_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	wrmsr(rv->msr_no, rv->l, rv->h);
+}
+
+int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
+	*l = rv.l;
+	*h = rv.h;
+
+	return err;
+}
+
+int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	rv.l = l;
+	rv.h = h;
+	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
+
+	return err;
+}
+
+/* rdmsr on a bunch of cpus
+ *
+ * @mask:	which cpus
+ * @msr_no:	which MSR
+ * @l:		array of the lower 32-bit MSR values
+ * @h:		array of the higher 32-bit MSR values
+ *
+ * Returns:
+ * 0 - success
+ * <0 - read failed on at least one CPU (latter in the mask)
+ */
+int rdmsr_on_cpus(const cpumask_t *mask, u32 msr_no, u32 *l, u32 *h)
+{
+	int cpu, tmp, err = 0;
+
+	int off = cpumask_first(mask);
+
+	for_each_cpu(cpu, mask) {
+		tmp = rdmsr_on_cpu(cpu, msr_no, l + (cpu - off),
+						h + (cpu - off));
+		if (tmp)
+			err = tmp;
+	}
+	return err;
+}
+
+/*
+ * wrmsr of a bunch of cpus
+ *
+ * @msr_no:	which MSR
+ * @l:		array of the lower 32-bit MSR values
+ * @h:		array of the higher 32-bit MSR values
+ *
+ * Returns:
+ * 0 - success
+ * <0 - write failed on at least one CPU (latter in the mask)
+ */
+int wrmsr_on_cpus(const cpumask_t *mask, u32 msr_no, u32 *l, u32 *h)
+{
+	int cpu, tmp, err = 0;
+
+	int off = cpumask_first(mask);
+
+	for_each_cpu(cpu, mask) {
+		tmp = wrmsr_on_cpu(cpu, msr_no, *(l + (cpu - off)),
+						*(h + (cpu - off)));
+		if (tmp)
+			err = tmp;
+	}
+	return err;
+}
+
+/* These "safe" variants are slower and should be used when the target MSR
+   may not actually exist. */
+static void __rdmsr_safe_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h);
+}
+
+static void __wrmsr_safe_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h);
+}
+
+int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
+	*l = rv.l;
+	*h = rv.h;
+
+	return err ? err : rv.err;
+}
+
+int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	rv.l = l;
+	rv.h = h;
+	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
+
+	return err ? err : rv.err;
+}
+
+EXPORT_SYMBOL(rdmsr_on_cpu);
+EXPORT_SYMBOL(wrmsr_on_cpu);
+EXPORT_SYMBOL(rdmsr_on_cpus);
+EXPORT_SYMBOL(wrmsr_on_cpus);
+EXPORT_SYMBOL(rdmsr_safe_on_cpu);
+EXPORT_SYMBOL(wrmsr_safe_on_cpu);
-- 
1.6.2.4

[PATCH 02/21] amd64_edac: add PCI config register defines

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  738 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 738 insertions(+), 0 deletions(-)
 create mode 100644 drivers/edac/amd64_edac.c

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
new file mode 100644
index 0000000..d43be21
--- /dev/null
+++ b/drivers/edac/amd64_edac.c
@@ -0,0 +1,738 @@
+/*
+ * AMD64 class Memory Controller kernel module
+ *
+ * Copyright (c) 2009 SoftwareBitMaker.
+ * Copyright (c) 2009 Advanced Micro Devices, Inc.
+ *
+ * This file may be distributed under the terms of the
+ * GNU General Public License.
+ *
+ *	Originally Written by Thayne Harbaugh
+ *
+ *      Changes by Douglas "norsk" Thompson  <dougthompson@xmission.com>:
+ *      	- K8 CPU Revision D and greater support
+ *
+ *      Changes by Dave Peterson <dsp@llnl.gov> <dave_peterson@pobox.com>:
+ *		- Module largely rewritten, with new (and hopefully correct)
+ *		code for dealing with node and chip select interleaving,
+ *		various code cleanup, and bug fixes
+ *		- Added support for memory hoisting using DRAM hole address
+ *		register
+ *
+ *	Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
+ *		-K8 Rev (1207) revision support added, required Revision
+ *		specific mini-driver code to support Rev F as well as
+ *		prior revisions
+ *
+ *	Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
+ *		-Family 10h revision support added. New PCI Device IDs,
+ *		indicating new changes. Actual registers modified
+ *		were slight, less than the Rev E to Rev F transition
+ *		but changing the PCI Device ID was the proper thing to
+ *		do, as it provides for almost automactic family
+ *		detection. The mods to Rev F required more family
+ *		information detection.
+ *
+ *	Changes/Fixes by Borislav Petkov <borislav.petkov@amd.com>:
+ *		- misc fixes and code cleanups
+ *
+ * This module is based on the following documents
+ * (available from http://www.amd.com/):
+ *
+ *	Title:	BIOS and Kernel Developer's Guide for AMD Athlon 64 and AMD
+ *		Opteron Processors
+ *	AMD publication #: 26094
+ *`	Revision: 3.26
+ *
+ *	Title:	BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh
+ *		Processors
+ *	AMD publication #: 32559
+ *	Revision: 3.00
+ *	Issue Date: May 2006
+ *
+ *	Title:	BIOS and Kernel Developer's Guide (BKDG) For AMD Family 10h
+ *		Processors
+ *	AMD publication #: 31116
+ *	Revision: 3.00
+ *	Issue Date: September 07, 2007
+ *
+ * Sections in the first 2 documents are no longer in sync with each other.
+ * The Family 10h BKDG was totally re-written from scratch with a new
+ * presentation model.
+ * Therefore, comments that refer to a Document section might be off.
+ */
+
+#include <linux/module.h>
+#include <linux/ctype.h>
+#include <linux/init.h>
+#include <linux/pci.h>
+#include <linux/pci_ids.h>
+#include <linux/slab.h>
+#include <linux/mmzone.h>
+#include <linux/edac.h>
+#include "edac_core.h"
+
+#define amd64_printk(level, fmt, arg...) \
+	edac_printk(level, "amd64", fmt, ##arg)
+
+#define amd64_mc_printk(mci, level, fmt, arg...) \
+	edac_mc_chipset_printk(mci, level, "amd64", fmt, ##arg)
+
+/*
+ * Throughout the comments in this code, the following terms are used:
+ *
+ *	SysAddr, DramAddr, and InputAddr
+ *
+ *  These terms come directly from the amd64 documentation
+ * (AMD publication #26094).  They are defined as follows:
+ *
+ *     SysAddr:
+ *         This is a physical address generated by a CPU core or a device
+ *         doing DMA.  If generated by a CPU core, a SysAddr is the result of
+ *         a virtual to physical address translation by the CPU core's address
+ *         translation mechanism (MMU).
+ *
+ *     DramAddr:
+ *         A DramAddr is derived from a SysAddr by subtracting an offset that
+ *         depends on which node the SysAddr maps to and whether the SysAddr
+ *         is within a range affected by memory hoisting.  The DRAM Base
+ *         (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
+ *         determine which node a SysAddr maps to.
+ *
+ *         If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
+ *         is within the range of addresses specified by this register, then
+ *         a value x from the DHAR is subtracted from the SysAddr to produce a
+ *         DramAddr.  Here, x represents the base address for the node that
+ *         the SysAddr maps to plus an offset due to memory hoisting.  See
+ *         section 3.4.8 and the comments in amd64_get_dram_hole_info() and
+ *         sys_addr_to_dram_addr() below for more information.
+ *
+ *         If the SysAddr is not affected by the DHAR then a value y is
+ *         subtracted from the SysAddr to produce a DramAddr.  Here, y is the
+ *         base address for the node that the SysAddr maps to.  See section
+ *         3.4.4 and the comments in sys_addr_to_dram_addr() below for more
+ *         information.
+ *
+ *     InputAddr:
+ *         A DramAddr is translated to an InputAddr before being passed to the
+ *         memory controller for the node that the DramAddr is associated
+ *         with.  The memory controller then maps the InputAddr to a csrow.
+ *         If node interleaving is not in use, then the InputAddr has the same
+ *         value as the DramAddr.  Otherwise, the InputAddr is produced by
+ *         discarding the bits used for node interleaving from the DramAddr.
+ *         See section 3.4.4 for more information.
+ *
+ *         The memory controller for a given node uses its DRAM CS Base and
+ *         DRAM CS Mask registers to map an InputAddr to a csrow.  See
+ *         sections 3.5.4 and 3.5.5 for more information.
+ */
+
+/*
+ * Alter this version for the K8 module when modifications are made
+ */
+#define EDAC_AMD64_VERSION	" Ver: 3.2.0 " __DATE__
+#define EDAC_MOD_STR		"amd64_edac"
+
+/* Extended Model from CPUID, for CPU Revision numbers */
+#define OPTERON_CPU_LE_REV_C    0
+#define OPTERON_CPU_REV_D       1
+#define OPTERON_CPU_REV_E       2
+
+/* NPT processors have the following Extended Models */
+#define OPTERON_CPU_REV_F       4
+#define OPTERON_CPU_REV_FA	5
+
+/* Hardware limit on ChipSelect rows per MC and processors per system */
+#define CHIPSELECT_COUNT	8
+#define DRAM_REG_COUNT		8
+
+/*************************************************************/
+/* K8 register addresses - device 0 Function 1 - Address Map */
+/*************************************************************/
+#define K8_DRAM_BASE_LOW	0x40
+				/* Function 1: DRAM Base Register (8 x 32b
+				 * interlaced with K8_DRAM_LIMIT_LOW)
+				 *
+				 * 31:16 DRAM base address reg bits[39:24]
+				 * 15:11 reserved
+				 * 10:8  interleave enable
+				 *  7:2  reserved
+				 *  1    write enable
+				 *  0    read enable
+				 *
+				 */
+
+#define K8_DRAM_LIMIT_LOW	0x44
+				/* Function 1: DRAM Limit Register (8 x 32b
+				 * interlaced with K8_DRAM_BASE_LOW)
+				 *
+				 * 31:16 DRAM Limit addr 32:24
+				 * 15:11 reserved
+				 * 10:8  interleave select
+				 *  7:3  reserved
+				 *  2:0  destination node ID
+				 */
+
+#define K8_DHAR         0xf0
+				/* Function 1: DRAM Hole Address Register
+				 *
+				 * K8
+				 * 31:24 DramHoleBase
+				 * 23:16 reserved
+				 * 15:8  DramHoleOffset
+				 *  7:1  reserved
+				 *    0  DramHoleValid
+				 *
+				 * F10
+				 * 31:24 DramHoleBase
+				 * 23:16 reserved
+				 * 15:7  DramHoleOffset
+				 *  6:2  reserved
+				 *    1  DramMemHoistValid
+				 *    0  DramHoleValid
+				 */
+#define DHAR_VALID			BIT(0)
+#define F10_DRAM_MEM_HOIST_VALID	BIT(1)
+
+#define DHAR_BASE_MASK			0xff000000
+#define dhar_base(dhar)			(dhar & DHAR_BASE_MASK)
+
+#define K8_DHAR_OFFSET_MASK		0x0000ff00
+#define k8_dhar_offset(dhar)		((dhar & K8_DHAR_OFFSET_MASK) << 16)
+
+#define F10_DHAR_OFFSET_MASK		0x0000ff80
+					/* NOTE: Extra mask bit vs K8 */
+#define f10_dhar_offset(dhar)		((dhar & F10_DHAR_OFFSET_MASK) << 16)
+
+
+/* F10 High BASE/LIMIT registers */
+#define F10_DRAM_BASE_HIGH	0x140
+				/* Function 1: DRAM Base register HIGH
+				 *
+				 * 7:0  Drambase[47:40] DRAM base address reg
+				 *	bits[47:40]
+				 */
+
+#define F10_DRAM_LIMIT_HIGH	0x144
+				/* Function 1: DRAM Limit Register HIGH
+				 *
+				 * 7:0  DRAM limit address register bits[47:40]
+				 */
+
+/*****************************************************************/
+/* K8 register addresses - device 0 Function 2 - DRAM controller */
+/*************************************************************/
+#define K8_DCSB0		0x40
+#define F10_DCSB1		0x140
+				/* Function 2: DRAM Chip-Select Base (8 x 32b)
+				 *
+				 * For Rev E and prior
+				 * 31:21 Base addr high 35:25
+				 * 20:16 reserved
+				 * 15:9  Base addr low 19:13 (interlvd)
+				 *  8:1  reserved
+				 *  0    chip-select bank enable
+				 *
+				 * For Rev F (NPT) and later
+				 * 31:29 reserved
+				 * 28:19 Base address (36:27)
+				 * 18:14 reserved
+				 * 13:5  Base address (21:13)
+				 * 4:3   reserved
+				 * 2     TestFail
+				 * 1     Spare Rank
+				 * 0     CESenable
+				 *
+				 */
+#define K8_DCSB_CS_ENABLE	BIT(0)
+#define K8_DCSB_NPT_SPARE	BIT(1)
+#define K8_DCSB_NPT_TESTFAIL	BIT(2)
+
+/* REV E: selects bits 31-21 and 15-9 from DCSB
+ * and the shift amount to form address
+ */
+#define REV_E_DCSB_BASE_BITS	(0xFFE0FE00ULL)
+#define REV_E_DCS_SHIFT		4
+#define REV_E_DCSM_COUNT	8
+
+#define REV_F_F1Xh_DCSB_BASE_BITS    (0x1FF83FE0ULL)
+#define REV_F_F1Xh_DCS_SHIFT        8
+
+/* REV F and later : selects bits 28-19 and 13-5 from DCSB
+ * and the shift amount to form address
+ */
+#define REV_F_DCSB_BASE_BITS	(0x1FF83FE0ULL)
+#define REV_F_DCS_SHIFT		8
+#define REV_F_DCSM_COUNT        4
+#define F10_DCSM_COUNT		4
+#define F11_DCSM_COUNT		2
+
+/* DRAM CS Mask Registers */
+#define K8_DCSM0		0x60
+#define F10_DCSM1		0x160
+				/* Function 2: DRAM Chip-Select Mask (8 x 32b)
+				 *
+				 * 31:30 reserved
+				 * 29:21 addr mask high 33:25
+				 * 20:16 reserved
+				 * 15:9  addr mask low  19:13
+				 *  8:0  reserved
+				 */
+
+/* REV E: selects bits 29-21 and 15-9 from DCSM */
+#define REV_E_DCSM_MASK_BITS        0x3FE0FE00
+/*    represents unused bits [24-20] and [12-0] */
+#define REV_E_DCS_NOTUSED_BITS        0x01F01FFF
+
+/* REV F and later: selects bits 28-19 and 13-5 from DCSM */
+#define REV_F_F1Xh_DCSM_MASK_BITS    0x1FF83FE0
+/*    represents unused bits [26-22] and [12-0] */
+#define REV_F_F1Xh_DCS_NOTUSED_BITS    0x07C01FFF
+
+#define DBAM0			0x80
+#define DBAM1			0x180
+				/* Function 2: DRAM Base Addr Mapping (32b) */
+
+/* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
+#define DBAM_DIMM(i, reg)	((((reg) >> (4*i))) & 0xF)
+
+#define DBAM_MAX_VALUE	11
+
+
+#define F10_DCLR_0	0x90
+#define F10_DCLR_1	0x190
+				/* Function 2: DRAM configuration low reg (32b)
+				 *	One for each DCTx
+				 *
+				 * Rev E and earlier CPUS:
+				 *
+				 * 31:28 reserved
+				 * 27:25 Bypass Max: 000b=respect
+				 * 24    Dissable receivers - no sockets
+				 * 23:20 x4 DIMMS
+				 * 19    32byte chunks
+				 * 18    Unbuffered
+				 * 17    ECC enabled
+				 * 16    128/64 bit (dual/single chan)
+				 * 15:14 R/W Queue bypass count
+				 * 13    Self refresh
+				 * 12    exit self refresh
+				 * 11    mem clear status
+				 * 10    DRAM enable
+				 *  9    reserved
+				 *  8    DRAM init
+				 *  7:4  reserved
+				 *  3    dis DQS hysteresis
+				 *  2    QFC enabled
+				 *  1    DRAM drive strength
+				 *  0    Digital Locked Loop disable
+				 *
+				 * Rev F
+				 *
+				 * 31:20 reserved
+				 * 19    DIMM ECC Enable
+				 * 18:17 reserved
+				 * 16    Unbuffered DIMM
+				 * 15:12 x4 DIMMs
+				 * 11    Width128 bits
+				 * 10    burstLength32
+				 *  9    SelRefRateEn
+				 *  8    ParEn
+				 *  7    DramDrvWeak
+				 *  6    reserved
+				 * 5:4   DramTerm
+				 * 3:2   reserved
+				 *  1    ExitSelfRef
+				 *  0    InitDram
+				 *
+				 * Rev F10h
+				 *
+				 * 31:24 reserved
+				 * 23    FoceAutoPcg
+				 * 22:21 IdleCycLowLimit
+				 * 20    DynPageCloseEn
+				 * 19    DIMM ECC Enable
+				 * 18    PendRefPayback
+				 * 17    EnterSelRef
+				 * 16    Unbuffered DIMM
+				 * 15:12 x4 DIMMs
+				 * 11    Width128 bits
+				 * 10    burstLength32
+				 *  9    SelRefRateEn
+				 *  8    ParEn
+				 *  7    DramDrvWeak
+				 *  6    DisDqsBar
+				 * 5:4   DramTerm
+				 * 3:2   reserved
+				 *  1    ExitSelfRef
+				 *  0    InitDram
+				 */
+#define REVE_WIDTH_128	BIT(16)
+#define F10_WIDTH_128	BIT(11)
+
+
+#define F10_DCHR_0	0x94
+#define F10_DCHR_1	0x194
+				/* Function 2: DRAM Configuration High Reg */
+
+#define F10_DCHR_FOUR_RANK_DIMM	BIT(18)
+#define F10_DCHR_Ddr3Mode	BIT(8)
+#define F10_DCHR_MblMode	BIT(6)
+
+
+#define F10_DCTL_SEL_LOW	0x110
+				/* Function 2: DRAM Controller SELECT LOW */
+
+#define dct_sel_baseaddr(pvt)    \
+	((pvt->dram_ctl_select_low) & 0xFFFFF800)
+
+#define dct_sel_interleave_addr(pvt)    \
+	(((pvt->dram_ctl_select_low) >> 6) & 0x3)
+
+enum {
+	F10_DCTL_SEL_LOW_DctSelHiRngEn	= BIT(0),
+	F10_DCTL_SEL_LOW_DctSelIntLvEn	= BIT(2),
+	F10_DCTL_SEL_LOW_DctGangEn	= BIT(4),
+	F10_DCTL_SEL_LOW_DctDatIntLv	= BIT(5),
+	F10_DCTL_SEL_LOW_DramEnable	= BIT(8),
+	F10_DCTL_SEL_LOW_MemCleared	= BIT(10),
+};
+
+#define    dct_high_range_enabled(pvt)    \
+	(pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctSelHiRngEn)
+
+#define dct_interleave_enabled(pvt)    \
+	(pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctSelIntLvEn)
+
+#define dct_ganging_enabled(pvt)        \
+	(pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctGangEn)
+
+#define dct_data_interleave_enabled(pvt)    \
+	(pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctDatIntLv)
+
+#define dct_dram_enabled(pvt)    \
+	(pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DramEnable)
+
+#define dct_memory_cleared(pvt)    \
+	(pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_MemCleared)
+
+
+#define F10_DCTL_SEL_HIGH	0x114
+				/* device 0 Function 2 - DRAM
+				 * Controller SELECT HIGH
+				 */
+
+/**************************************************************/
+/* K8 register addresses - device 0 Function 3 - Misc Control */
+/**************************************************************/
+#define K8_NBCTL	0x40
+				/* Function 3: MCA NB Control (32b)
+				 *
+				 *  1    MCA UE Reporting
+				 *  0    MCA CE Reporting
+				 */
+/* Correctable ECC error reporting enable */
+#define K8_NBCTL_CECCEn		BIT(0)
+
+/* UnCorrectable ECC error reporting enable */
+#define K8_NBCTL_UECCEn		BIT(1)
+
+#define K8_NBCFG	0x44
+				/* Function 3: MCA NB Config (32b)
+				 *
+				 * 23    Chip-kill x4 ECC enable
+				 * 22    ECC enable
+				 */
+#define		K8_NBCFG_CHIPKILL	BIT(23)
+#define		K8_NBCFG_ECC_ENABLE	BIT(22)
+
+#define K8_NBSL		0x48
+				/* Function 3: MCA NB Status Low (32b)
+				 *
+				 * 31:24 Syndrome 15:8 chip-kill x4
+				 * 23:20 reserved
+				 * 19:16 Extended err code (F0fh and earlier)
+				 * 20:16 Extended err code (F10h and later)
+				 * 15:0  Err code
+				 */
+
+
+#define	EXTRACT_HIGH_SYNDROME(x)	(((x) >> 24) & 0xff)
+#define EXTRACT_EXT_ERROR_CODE(x)	(((x) >> 16) & 0x1f)
+
+/* Start Family F10h: Normalized Extended Error Codes */
+#define F10_NBSL_EXT_ERR_RES		(0x0)
+#define F10_NBSL_EXT_ERR_CRC		(0x1)
+#define F10_NBSL_EXT_ERR_SYNC		(0x2)
+#define F10_NBSL_EXT_ERR_MST		(0x3)
+#define F10_NBSL_EXT_ERR_TGT		(0x4)
+#define F10_NBSL_EXT_ERR_GART		(0x5)
+#define F10_NBSL_EXT_ERR_RMW		(0x6)
+#define F10_NBSL_EXT_ERR_WDT		(0x7)
+#define F10_NBSL_EXT_ERR_ECC		(0x8)
+#define F10_NBSL_EXT_ERR_DEV		(0x9)
+#define F10_NBSL_EXT_ERR_LINK_DATA	(0xA)
+
+/* Next two are overloaded values */
+#define F10_NBSL_EXT_ERR_LINK_PROTO	(0xB)
+#define F10_NBSL_EXT_ERR_L3_PROTO	(0xB)
+
+#define F10_NBSL_EXT_ERR_NB_ARRAY	(0xC)
+#define F10_NBSL_EXT_ERR_DRAM_PARITY	(0xD)
+#define F10_NBSL_EXT_ERR_LINK_RETRY	(0xE)
+
+/* Next two are overloaded values */
+#define F10_NBSL_EXT_ERR_GART_WALK	(0xF)
+#define F10_NBSL_EXT_ERR_DEV_WALK	(0xF)
+
+/* 0x10 to 0x1B: Reserved */
+
+#define F10_NBSL_EXT_ERR_L3_DATA	(0x1C)
+#define F10_NBSL_EXT_ERR_L3_TAG		(0x1D)
+#define F10_NBSL_EXT_ERR_L3_LRU		(0x1E)
+/* End Family F10h: Extended Error Codes */
+
+/* Start K8: Normalized Extended Error Codes */
+#define K8_NBSL_EXT_ERR_ECC		(0x0)
+#define K8_NBSL_EXT_ERR_CRC		(0x1)
+#define K8_NBSL_EXT_ERR_SYNC		(0x2)
+#define K8_NBSL_EXT_ERR_MST		(0x3)
+#define K8_NBSL_EXT_ERR_TGT		(0x4)
+#define K8_NBSL_EXT_ERR_GART		(0x5)
+#define K8_NBSL_EXT_ERR_RMW		(0x6)
+#define K8_NBSL_EXT_ERR_WDT		(0x7)
+#define K8_NBSL_EXT_ERR_CHIPKILL_ECC	(0x8)
+#define K8_NBSL_EXT_ERR_DRAM_PARITY	(0xD)
+/* End K8: Extended Error Codes */
+
+
+/* Error Code */
+#define EXTRACT_ERROR_CODE(x)	((x) & 0xffff)
+#define		TEST_TLB_ERROR(x)	(((x) & 0xFFF0) == 0x0010)
+#define		TEST_MEM_ERROR(x)	(((x) & 0xFF00) == 0x0100)
+#define		TEST_BUS_ERROR(x)	(((x) & 0xF800) == 0x0800)
+#define 	EXTRACT_TT_CODE(x)	(((x) >> 2) & 0x3)
+#define 	EXTRACT_II_CODE(x)	(((x) >> 2) & 0x3)
+#define 	EXTRACT_LL_CODE(x)	(((x) >> 0) & 0x3)
+#define		EXTRACT_RRRR_CODE(x)	(((x) >> 4) & 0xf)
+#define 	EXTRACT_TO_CODE(x)	(((x) >> 8) & 0x1)
+#define		EXTRACT_PP_CODE(x)	(((x) >> 9) & 0x3)
+
+/* The following are for BUS type errors AFTER values have been
+ * normalized by shifting right
+ */
+#define K8_NBSL_PP_SRC		(0x0)
+#define K8_NBSL_PP_RES		(0x1)
+#define K8_NBSL_PP_OBS		(0x2)
+#define K8_NBSL_PP_GENERIC	(0x3)
+
+
+#define K8_NBSH		0x4C
+				/* Function 3: MCA NB Status High (32b)
+				 *
+				 * 31    Err valid
+				 * 30    Err overflow
+				 * 29    Uncorrected err
+				 * 28    Err enable
+				 * 27    Misc err reg valid
+				 * 26    Err addr valid
+				 * 25    proc context corrupt
+				 * 24:23 reserved
+				 * 22:15 Syndrome bits 7:0
+				 * 14    CE
+				 * 13    UE
+				 * 12:9  reserved
+				 *  8    err found by scrubber
+				 *  7    reserved
+				 *  6:4  Hyper-transport link number
+				 *    3:2  reserved= Rev F/Family 10h=Quad Core
+				 *  3    Err CPU 3	(F10)
+				 *  2    Err CPU 2	(F10)
+				 *  1    Err CPU 1	(Dual Core)
+				 *  0    Err CPU 0
+				 */
+
+#define K8_NBSH_VALID_BIT		BIT(31)
+#define K8_NBSH_OVERFLOW		BIT(30)
+#define K8_NBSH_UNCORRECTED_ERR		BIT(29)
+#define K8_NBSH_ERR_ENABLE		BIT(28)
+#define K8_NBSH_MISC_ERR_VALID		BIT(27)
+#define K8_NBSH_VALID_ERROR_ADDR	BIT(26)
+#define K8_NBSH_PCC			BIT(25)
+#define K8_NBSH_CECC			BIT(14)
+#define K8_NBSH_UECC			BIT(13)
+#define K8_NBSH_ERR_SCRUBER		BIT(8)
+#define K8_NBSH_CORE3			BIT(3)
+#define K8_NBSH_CORE2			BIT(2)
+#define K8_NBSH_CORE1			BIT(1)
+#define K8_NBSH_CORE0			BIT(0)
+
+#define EXTRACT_LDT_LINK(x) 		(((x) >> 4) & 0x7)
+#define EXTRACT_ERR_CPU_MAP(x)		((x) & 0xF)
+#define EXTRACT_LOW_SYNDROME(x)		(((x) >> 15) & 0xff)
+
+
+#define K8_NBEAL	0x50
+				/* Function 3: MCA NB err addr low (32b)
+				 *
+				 * 31:3  Err addr low 31:3
+				 *  2:0  reserved
+				 */
+
+#define K8_NBEAH	0x54
+				/* Function 3: MCA NB err addr high (32b)
+				 *
+				 * 31:8  reserved
+				 *  7:0  Err addr high 39:32
+				 */
+
+#define K8_SCRCTRL      0x58
+				/* Function 3: Memory scrub control register.
+				 *
+				 * 30:21 reserved
+				 * 20:16 dcache scrub
+				 * 15:13 reserved
+				 * 12:8  L2Scrub
+				 * 7:5   reserved
+				 * 4:0   dramscrub
+				 *
+				 */
+
+
+
+#define F10_NB_CFG_LOW	0x88
+				/* Function 3: NB Configuration reg low */
+#define 	F10_NB_CFG_LOW_ENABLE_EXT_CFG	BIT(14)
+
+#define F10_NB_CFG_HIGH	0x8C
+				/* Function 3: NB Configuration reg high */
+
+
+#define F10_ONLINE_SPARE	0xB0
+				/* Function 3: On-Line Spare Control Register */
+#define F10_ONLINE_SPARE_SWAPDONE0(x)	((x) & BIT(1))
+#define F10_ONLINE_SPARE_SWAPDONE1(x)	((x) & BIT(3))
+#define F10_ONLINE_SPARE_BADDRAM_CS0(x)	(((x) >> 4) & 0x00000007)
+#define F10_ONLINE_SPARE_BADDRAM_CS1(x)	(((x) >> 8) & 0x00000007)
+
+
+#define F10_NB_ARRAY_ADDR	0xB8
+				/* Function 3:
+				 *	Error injection NB Array Addr Reg
+				 *
+				 * For a 64-byte cacheline, we can select
+				 * one of 4 16-byte sections (0,1,2,3) of
+				 * that cacheline.
+				 * Bits 2:1 provide that selection.
+				 *
+				 * which 16-bit word of the section is
+				 * selected by bitmap 28:20 of F10_NB_ARRAY_DATA
+				 *
+				 * which bit of the the selected word
+				 * is then masked by bits 15:0 of the same reg
+				 *
+				 * Thus we have a tuple of:
+				 *	section,word,bit
+				 */
+
+				/* DRAM ECC Array Select */
+#define F10_NB_ARRAY_DRAM_ECC	0x80000000
+
+				/* Bits 2:1 are used to select 16-byte
+				 * section within a 64-byte cacheline
+				 */
+#define SET_NB_ARRAY_ADDRESS(section)	(((section) & 0x3) << 1)
+
+#define F10_NB_ARRAY_DATA	0xBC
+				/* Function 3:
+				 *	Error injection NB Array Data Reg
+				 *
+				 * 28:20	ErrInjEn selects 16-bit word
+				 *		(bit map: 0-8)
+				 * 17		EccWrReq
+				 * 16		EccRdReq
+				 * 15:0		EccVector, select bits
+				 */
+#define SET_NB_DRAM_INJECTION_WRITE(word, bits)  \
+				(BIT(((word) & 0xF) + 20) | \
+				BIT(17) |  \
+				((bits) & 0xF))
+#define SET_NB_DRAM_INJECTION_READ(word, bits)  \
+				(BIT(((word) & 0xF) + 20) | \
+				BIT(16) |  \
+				((bits) & 0xF))
+
+
+
+#define K8_NBCAP	0xE8
+				/* Function 3: MCA NB capabilities (32b)
+				 *
+				 * 31:9  reserved
+				 *  4    ChipKill S4ECD4ED capable
+				 *  3    SECDED capable
+				 */
+#define K8_NBCAP_CORES		(BIT(12)|BIT(13))
+#define K8_NBCAP_CHIPKILL	BIT(4)
+#define K8_NBCAP_SECDED		BIT(3)
+#define K8_NBCAP_8_NODE		BIT(2)
+#define K8_NBCAP_DUAL_NODE	BIT(1)
+#define K8_NBCAP_DCT_DUAL	BIT(0)
+
+				/* MSR's */
+
+				/*
+				 * K8_MSR_MCxCTL (64b)
+				 * (0x400,404,408,40C,410)
+				 * 63    Enable reporting source 63
+				 *  .
+				 *  .
+				 *  .
+				 *  2    Enable error source 2
+				 *  1    Enable error source 1
+				 *  0    Enable error source 0
+				 */
+
+				/*
+				 * K8_MSR_MCxSTAT (64b)
+				 * (0x401,405,409,40D,411)
+				 * 63    Error valid
+				 * 62    Status overflow
+				 * 61    UE
+				 * 60    Enabled error condition
+				 * 59    Misc register valid (not used)
+				 * 58    Err addr register valid
+				 * 57    Processor context corrupt
+				 * 56:32 Other information
+				 * 31:16 Model specific error code
+				 * 15:0  MCA err code
+				 */
+
+				/*
+				 * K8_MSR_MCxADDR (64b)
+				 * (0x402,406,40A,40E,412)
+				 * 63:48 reserved
+				 * 47:0  Address
+				 */
+
+				/*
+				 * K8_MSR_MCxMISC (64b)
+				 * (0x403,407,40B,40F,413)
+				 * Unused on Athlon64 and K8
+				 */
+
+/* MSR Regs */
+#define K8_MSR_MCGCTL	0x017b
+				/* Machine Chk Global report ctl (64b)
+				 *
+				 * 31:5  reserved
+				 *  4    North Bridge
+				 *  3    Load/Store
+				 *  2    Bus Unit
+				 *  1    Instruction Cache
+				 *  0    Data Cache
+				 */
+#define K8_MSR_MCGCTL_NBE	BIT(4)
+
+#define K8_MSR_MC4CTL	0x0410	/* North Bridge Check report ctl (64b) */
+#define K8_MSR_MC4STAT	0x0411	/* North Bridge status (64b) */
+#define K8_MSR_MC4ADDR	0x0412	/* North Bridge Address (64b) */
-- 
1.6.2.4

[PATCH 03/21] amd64_edac: add driver structs

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  264 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 264 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index d43be21..532a059 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -736,3 +736,267 @@ enum {
 #define K8_MSR_MC4CTL	0x0410	/* North Bridge Check report ctl (64b) */
 #define K8_MSR_MC4STAT	0x0411	/* North Bridge status (64b) */
 #define K8_MSR_MC4ADDR	0x0412	/* North Bridge Address (64b) */
+
+/**
+ * popcnt - count the set bits in a bit vector.
+ * @vec - bit vector
+ *
+ * This instruction is supported only on F10h and later CPUs.
+ */
+#define popcnt(x)						\
+({								\
+	typeof(x) __ret;					\
+	__asm__("popcnt %1, %0" : "=r" (__ret) : "r" (x));	\
+	__ret;							\
+})
+
+
+static struct edac_pci_ctl_info *amd64_ctl_pci;
+
+static int report_gart_errors;
+module_param(report_gart_errors, int, 0644);
+
+/*
+ * Set by command line parameter. If BIOS has enabled the ECC, this override is
+ * cleared to prevent re-enabling the hardware by this driver.
+ */
+static int ecc_enable_override;
+module_param(ecc_enable_override, int, 0644);
+
+/* AMD sets the first MC device at device ID 0x18. */
+static inline int get_mc_node_id_from_pdev(struct pci_dev *pdev)
+{
+	return PCI_SLOT(pdev->devfn) - 0x18;
+}
+
+/* Lookup table for all possible MC control instances */
+struct amd64_pvt;
+static struct mem_ctl_info *mci_lookup[MAX_NUMNODES];
+static struct amd64_pvt *pvt_lookup[MAX_NUMNODES];
+
+enum amd64_chipset_families {
+	K8_CPUS = 0,
+	F10_CPUS,
+	F11_CPUS,
+};
+
+/*
+ * Structure to hold:
+ *	1) dynamicly read status and error address HW registers
+ *	2) sysfs entered values
+ *	3) MCE values
+ *
+ * Depends on entry into the modules
+ */
+struct amd64_error_info_regs {
+	u32 nbcfg;
+	u32 nbsh;
+	u32 nbsl;
+	u32 nbeah;
+	u32 nbeal;
+};
+
+/*
+ * Each of the PCI Device IDs types have their own set of hardware
+ * accessor function and per device encoding/decoding logic.
+ */
+struct low_ops {
+	int (*probe_valid_hardware)(struct amd64_pvt *pvt);
+	int (*early_channel_count)(struct amd64_pvt *pvt);
+
+	u64 (*get_error_address)(struct mem_ctl_info *mci,
+			struct amd64_error_info_regs *info);
+	void (*read_dram_base_limit)(struct amd64_pvt *pvt, int dram);
+	void (*read_dram_ctl_register)(struct amd64_pvt *pvt);
+	void (*map_sysaddr_to_csrow)(struct mem_ctl_info *mci,
+					struct amd64_error_info_regs *info,
+					u64 SystemAddr);
+	int (*dbam_map_to_pages)(struct amd64_pvt *pvt, int dram_map);
+};
+
+/*
+ * amd64 family unique informatoin
+ */
+struct amd64_family_type {
+	const char *ctl_name;
+	u16 addr_f1_ctl;
+	u16 misc_f3_ctl;
+	struct low_ops ops;
+};
+
+static struct amd64_family_type amd64_family_types[];
+
+static inline const char *get_amd_family_name(int index)
+{
+	return amd64_family_types[index].ctl_name;
+}
+
+static inline struct low_ops *get_amd_family_ops(int index)
+{
+	return &amd64_family_types[index].ops;
+}
+
+/*
+ * Error injection control structure
+ */
+struct error_injection {
+	u32	section;
+	u32	word;
+	u32	bit_map;
+};
+
+struct amd64_pvt {
+	/* pci_device handles which we utilize */
+	struct pci_dev *addr_f1_ctl;
+	struct pci_dev *dram_f2_ctl;
+	struct pci_dev *misc_f3_ctl;
+
+	int mc_node_id;		/* MC index of this MC node */
+	int ext_model;		/* extended model value of this node */
+
+	struct low_ops	*ops;	/* pointer to per PCI Device ID func table */
+
+	int channel_count;	/* Count of 'channels' */
+
+	/* Raw registers */
+	u32 dclr0;		/* DRAM Configuration Low DCT0 reg */
+	u32 dclr1;		/* DRAM Configuration Low DCT1 reg */
+	u32 dchr0;		/* DRAM Configuration High DCT0 reg */
+	u32 dchr1;		/* DRAM Configuration High DCT1 reg */
+	u32 nbcap;		/* North Bridge Capabilities */
+	u32 nbcfg;		/* F10 North Bridge Configuration */
+	u32 ext_nbcfg;		/* Extended F10 North Bridge Configuration */
+	u32 dhar;		/* DRAM Hoist reg */
+	u32 dbam0;		/* DRAM Base Address Mapping reg for DCT0 */
+	u32 dbam1;		/* DRAM Base Address Mapping reg for DCT1 */
+
+	/* DRAM CS Base Address Registers
+	 *   F2x[1,0][5C:40]
+	 */
+	u32 dcsb0[CHIPSELECT_COUNT];    /* DRAM CS Base Registers */
+	u32 dcsb1[CHIPSELECT_COUNT];    /* DRAM CS Base Registers */
+
+	/* DRAM CS Mask Registers
+	 *   F2x[1,0][6C:60]
+	 */
+	u32 dcsm0[CHIPSELECT_COUNT];    /* DRAM CS Mask Registers */
+	u32 dcsm1[CHIPSELECT_COUNT];    /* DRAM CS Mask Registers */
+
+	/* Decoded parts of DRAM BASE and LIMIT Registers
+	 * F1x[78,70,68,60,58,50,48,40]
+	 */
+	u64 dram_base[DRAM_REG_COUNT];/* DRAM Base Reg */
+	u64 dram_limit[DRAM_REG_COUNT];/* DRAM Limit Reg */
+	u8 dram_IntlvSel[DRAM_REG_COUNT];
+	u8 dram_IntlvEn[DRAM_REG_COUNT];
+	u8 dram_DstNode[DRAM_REG_COUNT];
+	u8 dram_rw_en[DRAM_REG_COUNT];
+
+	/* The following fields are set at (load) run time, after Revision has
+	 * been determined, since the dct_base and dct_mask registers vary
+	 * by CPU Revsion
+	 */
+	u32 dcsb_base;		/* DCSB base bits */
+	u32 dcsm_mask;		/* DCSM mask bits */
+	u32 num_dcsm;		/* Number of DCSM registers */
+	u32 dcs_mask_notused;	/* DCSM notused mask bits */
+	u32 dcs_shift;		/* DCSB and DCSM shift value */
+
+	u64 top_mem;		/* top of memory below 4GB */
+	u64 top_mem2;		/* top of memory above 4GB */
+
+	/* F10 registers */
+	u32 dram_ctl_select_low;	/* DRAM Controller Select Low Reg */
+	u32 dram_ctl_select_high;	/* DRAM Controller Select High Reg */
+	u32 online_spare;               /* On-Line spare Reg */
+
+	/* sysfs storage area: Temp storage for when input
+	 * is received from sysfs
+	 */
+	struct amd64_error_info_regs ctl_error_info;
+
+	/* Place to store error injection parameters prior to issue */
+	struct error_injection injection;
+
+	/* Save old hw registers' values before we modified them */
+	u32 nbctl_mcgctl_saved;		/* When true, following 2 are valid */
+	u32 old_nbctl;
+	u32 *old_mcgctl;		/* per core on this node */
+
+	/* MC Type Index value: socket F vs Family 10h */
+	u32 mc_type_index;
+
+	/* misc settings */
+	struct flags {
+		unsigned long cf8_extcfg:1;
+	} flags;
+};
+
+/*
+ * See F2x80 for K8 and F2x[1,0]80 for Fam10 and later. The table below is only
+ * for DDR2 DRAM mapping.
+ */
+static u32 revf_quad_ddr2_shift[] = {
+	0,	/* 0000b NULL DIMM (128mb) */
+	28,	/* 0001b 256mb */
+	29,	/* 0010b 512mb */
+	29,	/* 0011b 512mb */
+	29,	/* 0100b 512mb */
+	30,	/* 0101b 1gb */
+	30,	/* 0110b 1gb */
+	31,	/* 0111b 2gb */
+	31,	/* 1000b 2gb */
+	32,	/* 1001b 4gb */
+	32,	/* 1010b 4gb */
+	33,	/* 1011b 8gb */
+	0,	/* 1100b future */
+	0,	/* 1101b future */
+	0,	/* 1110b future */
+	0	/* 1111b future */
+};
+
+/* Valid scrub rates for the K8 hardware memory scrubber. We map
+ *	the scrubbing bandwidth to a valid bit pattern. The 'set'
+ *	operation finds the 'matching- or higher value'.
+ *
+ *FIXME: Produce a better mapping/linearisation.
+ */
+
+static struct scrubrate {
+	u32 scrubval;		/* bit pattern for scrub rate */
+	u32 bandwidth;		/* bandwidth consumed (bytes/sec) */
+} scrubrates[] = {
+	{ 0x01, 1600000000UL},
+	{ 0x02, 800000000UL},
+	{ 0x03, 400000000UL},
+	{ 0x04, 200000000UL},
+	{ 0x05, 100000000UL},
+	{ 0x06, 50000000UL},
+	{ 0x07, 25000000UL},
+	{ 0x08, 12284069UL},
+	{ 0x09, 6274509UL},
+	{ 0x0A, 3121951UL},
+	{ 0x0B, 1560975UL},
+	{ 0x0C, 781440UL},
+	{ 0x0D, 390720UL},
+	{ 0x0E, 195300UL},
+	{ 0x0F, 97650UL},
+	{ 0x10, 48854UL},
+	{ 0x11, 24427UL},
+	{ 0x12, 12213UL},
+	{ 0x13, 6101UL},
+	{ 0x14, 3051UL},
+	{ 0x15, 1523UL},
+	{ 0x16, 761UL},
+	{ 0x00, 0UL},        /* scrubbing off */
+
+};
+
+/*
+ * For future CPU versions, verify the following as new 'slow' rates appear and
+ * modify the necessary skip values for the supported CPU.
+ */
+#define K8_MIN_SCRUB_RATE_BITS		0
+#define F10_MIN_SCRUB_RATE_BITS	0x5
+#define F11_MIN_SCRUB_RATE_BITS	0x6
+
-- 
1.6.2.4

[PATCH 04/21] amd64_edac: add memory scrubber interface

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  116 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 116 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 532a059..a121785 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -1000,3 +1000,119 @@ static struct scrubrate {
 #define F10_MIN_SCRUB_RATE_BITS	0x5
 #define F11_MIN_SCRUB_RATE_BITS	0x6
 
+/*
+ * Memory scrubber control interface. For the K8, memory scrubbing is handled by
+ * hardware and can involve L2 cache, dcache as well as the main memory. With
+ * F10, this is extended to L3 cache scrubbing on CPU models sporting that
+ * functionality.
+ * This causes the "units" for the scrubbing speed to vary from 64 byte blocks
+ * (dram) over to cache lines. This is nasty, so we will use bandwidth in
+ * bytes/sec for the setting.
+ * Currently, we only do dram scrubbing. If the scrubbing is done in software on
+ * other archs, we might not have access to the caches directly.
+ */
+
+/*
+ * amd64_search_set_scrub_rate
+ *
+ * scan the scrub rate mapping table for a close or matching bandwidth value to
+ * issue. If requested is too big, then use last maximum value found.
+ */
+static int amd64_search_set_scrub_rate(struct pci_dev *ctl, u32 new_bw,
+				       u32 min_scrubrate)
+{
+	u32 scrubval;
+	int i;
+
+	/*
+	 * map the configured rate (new_bw) to a value specific to the AMD64
+	 * memory controller and apply to register. Search for the first
+	 * bandwidth entry that is greater or equal than the setting requested
+	 * and program that. If at last entry, accept that as the maximum. This
+	 * saves the user from determing the maximum empirically.
+	 */
+	for (i = 0; i < ARRAY_SIZE(scrubrates); i++) {
+		/*
+		 * skip scrub rates which aren't recommended
+		 * (see F10 BKDG, F3x58)
+		 */
+		if (scrubrates[i].scrubval < min_scrubrate)
+			continue;
+
+		if (scrubrates[i].bandwidth <= new_bw)
+			break;
+
+		/*
+		 * if no suitable bandwidth found, turn off DRAM scrubbing
+		 * entirely by falling back to the last element in the
+		 * scrubrates array.
+		 */
+	}
+
+	scrubval = scrubrates[i].scrubval;
+	if (scrubval)
+		edac_printk(KERN_DEBUG, EDAC_MC,
+			"Setting scrub rate bandwidth: %u\n",
+			scrubrates[i].bandwidth);
+	else
+		edac_printk(KERN_DEBUG, EDAC_MC,
+			"Turning scrubbing off.\n");
+
+	pci_write_bits32(ctl, K8_SCRCTRL, scrubval, 0x001F);
+
+	return 0;
+}
+
+static int amd64_set_scrub_rate(struct mem_ctl_info *mci, u32 *bandwidth)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u32 min_scrubrate = 0x0;
+
+	switch (boot_cpu_data.x86) {
+	case 0xf:
+		min_scrubrate = K8_MIN_SCRUB_RATE_BITS;
+		break;
+	case 0x10:
+		min_scrubrate = F10_MIN_SCRUB_RATE_BITS;
+		break;
+	case 0x11:
+		min_scrubrate = F11_MIN_SCRUB_RATE_BITS;
+		break;
+
+	default:
+		amd64_printk(KERN_ERR, "Unsupported family!\n");
+		break;
+	}
+	return amd64_search_set_scrub_rate(pvt->misc_f3_ctl, *bandwidth,
+			min_scrubrate);
+}
+
+static int amd64_get_scrub_rate(struct mem_ctl_info *mci, u32 *bw)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u32 scrubval = 0;
+	int status = -1;
+	int i;
+	int err;
+
+	err = pci_read_config_dword(pvt->misc_f3_ctl, K8_SCRCTRL, &scrubval);
+	if (err != 0)
+		debugf0("%s() Reading K8_SCRCTRL failed\n", __func__);
+
+	scrubval = scrubval & 0x001F;
+
+	edac_printk(KERN_DEBUG, EDAC_MC,
+		    "pci-read, sdram scrub control value: %d \n", scrubval);
+
+	for (i = 0; ARRAY_SIZE(scrubrates); i++) {
+		if (scrubrates[i].scrubval == scrubval) {
+			*bw = scrubrates[i].bandwidth;
+			status = 0;
+			break;
+		}
+	}
+
+	return status;
+}
+
+
-- 
1.6.2.4

[PATCH 05/21] amd64_edac: add sys addr to memory controller mapping helpers

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  163 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 163 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index a121785..49c931f 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -1115,4 +1115,167 @@ static int amd64_get_scrub_rate(struct mem_ctl_info *mci, u32 *bw)
 	return status;
 }
 
+/*
+ * amd64_map_to_dcs_mask
+ *
+ *	Map from a CSROW entry to the mask entry that operates on it
+ */
+static inline u32 amd64_map_to_dcs_mask(struct amd64_pvt *pvt, int csrow)
+{
+	return csrow >> (pvt->num_dcsm >> 3);
+}
+
+/*
+ * amd64_get_dct_base()
+ *
+ *	getter function to return the 'base' address the i'th CS entry
+ *	of the 'dct' DRAM controller
+ */
+static u32 amd64_get_dct_base(struct amd64_pvt *pvt, int dct, int csrow)
+{
+	if (dct == 0)
+		return pvt->dcsb0[csrow];
+	else
+		return pvt->dcsb1[csrow];
+}
+
+/*
+ * amd64_get_dct_mask()
+ *
+ *	getter function to return the 'mask' address the i'th CS entry.
+ *	This getter function is needed because there different number
+ *	of DCSM registers on Rev E and prior vs Rev F and later
+ */
+static u32 amd64_get_dct_mask(struct amd64_pvt *pvt, int dct, int csrow)
+{
+	if (dct == 0)
+		return pvt->dcsm0[amd64_map_to_dcs_mask(pvt, csrow)];
+	else
+		return pvt->dcsm1[amd64_map_to_dcs_mask(pvt, csrow)];
+}
+
+
+/*
+ * amd64_get_base_and_limit()
+ *
+ * In *base and *limit, pass back the full 40-bit base and limit physical
+ * addresses for the node given by node_id.  This information is obtained from
+ * DRAM Base (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers. The
+ * base and limit addresses are of type SysAddr, as defined at the start of
+ * section 3.4.4 (p. 70).  They are the lowest and highest physical addresses
+ * in the address range they represent.
+ */
+static void amd64_get_base_and_limit(struct amd64_pvt *pvt, int node_id,
+			       u64 *base, u64 *limit)
+{
+	*base = pvt->dram_base[node_id];
+	*limit = pvt->dram_limit[node_id];
+}
+
+/*
+ * Return 1 if the SysAddr given by sys_addr matches the base/limit associated
+ * with node_id
+ */
+static int amd64_base_limit_match(struct amd64_pvt *pvt,
+					u64 sys_addr, int node_id)
+{
+	u64 base, limit, addr;
 
+	amd64_get_base_and_limit(pvt, node_id, &base, &limit);
+
+	/* The K8 treats this as a 40-bit value.  However, bits 63-40 will be
+	 * all ones if the most significant implemented address bit is 1.
+	 * Here we discard bits 63-40.  See section 3.4.2 of AMD publication
+	 * 24592: AMD x86-64 Architecture Programmer's Manual Volume 1
+	 * Application Programming.
+	 */
+	addr = sys_addr & 0x000000ffffffffffull;
+
+	return (addr >= base) && (addr <= limit);
+}
+
+/* find_mc_by_sys_addr
+ *
+ *	Attempt to map a SysAddr to a node.
+ *
+ *	On success, return a pointer to the mem_ctl_info structure for
+ *	the node that the SysAddr maps to.
+ *
+ *	On failure, return NULL
+ */
+static struct mem_ctl_info *find_mc_by_sys_addr(struct mem_ctl_info *mci,
+						u64 sys_addr)
+{
+	struct amd64_pvt *pvt;
+	int node_id;
+	u32 intlv_en, bits;
+
+	/*
+	 * Here we use the DRAM Base (section 3.4.4.1) and DRAM Limit (section
+	 * 3.4.4.2) registers to map the SysAddr to a node ID.
+	 */
+	pvt = mci->pvt_info;
+
+	/* The value of this field should be the same for all DRAM Base
+	 * registers.  Therefore we arbitrarily choose to read it from the
+	 * register for node 0.
+	 */
+	intlv_en = pvt->dram_IntlvEn[0];
+
+	if (intlv_en == 0) {
+		debugf2("%s(): node interleaving disabled\n", __func__);
+		for (node_id = 0; ; ) {
+			if (amd64_base_limit_match(pvt, sys_addr, node_id))
+				break;
+
+			if (++node_id >= DRAM_REG_COUNT) {
+				debugf2("%s(): sys_addr 0x%lx "
+					"does not match any node\n", __func__,
+					(unsigned long)sys_addr);
+				return NULL;
+			}
+		}
+
+		goto found;
+	}
+
+	if (unlikely((intlv_en != (0x01 << 8)) &&
+		     (intlv_en != (0x03 << 8)) && (intlv_en != (0x07 << 8)))) {
+		amd64_printk(KERN_WARNING,
+			  "%s(): junk value of 0x%x extracted from IntlvEn "
+			  "field of DRAM Base Register for node 0: This "
+			  "probably indicates a BIOS bug.\n", __func__,
+			  intlv_en);
+		return NULL;
+	}
+
+	debugf2("%s(): node interleaving enabled\n", __func__);
+	bits = (((u32) sys_addr) >> 12) & intlv_en;
+
+	for (node_id = 0; ; ) {
+		if ((pvt->dram_limit[node_id] & intlv_en) == bits)
+			break;	/* intlv_sel field matches */
+
+		if (++node_id >= DRAM_REG_COUNT) {
+			debugf2("%s(): sys_addr 0x%lx does not match any "
+				"node\n", __func__, (unsigned long)sys_addr);
+			return NULL;
+		}
+	}
+
+	/* sanity test for sys_addr */
+	if (unlikely(!amd64_base_limit_match(pvt, sys_addr, node_id))) {
+		amd64_printk(KERN_WARNING,
+			  "%s(): sys_addr 0x%lx falls outside base/limit "
+			  "address range for node %d with node interleaving "
+			  "enabled.\n", __func__, (unsigned long)sys_addr,
+			  node_id);
+		return NULL;
+	}
+
+found:
+	debugf2("%s(): sys_addr 0x%lx matches node %d\n", __func__,
+		(unsigned long)sys_addr, node_id);
+
+	return edac_mc_find(node_id);
+}
-- 
1.6.2.4

[PATCH 06/21] amd64_edac: add functionality to compute the DRAM hole

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  172 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 172 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 49c931f..2033dd4 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -1279,3 +1279,175 @@ found:
 
 	return edac_mc_find(node_id);
 }
+
+/*
+ * base_from_dct_base
+ *
+ *	Extract the DRAM CS base address from selected csrow register
+ */
+static u64 base_from_dct_base(struct amd64_pvt *pvt, int csrow)
+{
+	return ((u64) (amd64_get_dct_base(pvt, 0, csrow) & pvt->dcsb_base)) <<
+				pvt->dcs_shift;
+}
+
+/*
+ * mask_from_dct_mask
+ *
+ *	Extract the Mask from the dcsb0[csrow] entry
+ *	Depends on CPU Revision on how to extract this information
+ */
+static u64 mask_from_dct_mask(struct amd64_pvt *pvt, int csrow)
+{
+	u64 dcsm_bits, other_bits;
+	u64 mask;
+
+	/* Extract bits bits 29-21 and 15-9 from DCSM (section 3.5.5). */
+	dcsm_bits = amd64_get_dct_mask(pvt, 0, csrow) & pvt->dcsm_mask;
+
+	/* Set all bits except bits 33-25 and 19-13. */
+	other_bits = pvt->dcsm_mask;
+	other_bits = ~(other_bits << pvt->dcs_shift);
+
+	/* The extracted bits from DCSM belong in the spaces represented by
+	 * the cleared bits in other_bits.
+	 */
+	mask = (dcsm_bits << pvt->dcs_shift) | other_bits;
+
+	return mask;
+}
+
+/*
+ * input_addr_to_csrow
+ *
+ * input_addr is an InputAddr associated with the node given by mci.  Return
+ * the csrow that input_addr maps to, or -1 on failure (no csrow claims
+ * input_addr).
+ */
+static int input_addr_to_csrow(struct mem_ctl_info *mci, u64 input_addr)
+{
+	struct amd64_pvt *pvt;
+	int csrow;
+	u64 base, mask;
+
+	pvt = mci->pvt_info;
+
+	/* Here we use the DRAM CS Base (section 3.5.4) and DRAM CS Mask
+	 * (section 3.5.5) registers.  For each CS base/mask register pair,
+	 * test the condition shown near the start of section 3.5.4 (p. 84).
+	 */
+	for (csrow = 0; csrow < CHIPSELECT_COUNT; csrow++) {
+
+		if ((pvt->dcsb0[csrow] & K8_DCSB_CS_ENABLE) == 0) {
+			debugf2("input_addr_to_csrow: CSBE bit is cleared "
+				"for csrow %d (node %d)\n",
+				csrow, pvt->mc_node_id);
+			continue;
+		}
+
+		base = base_from_dct_base(pvt, csrow);
+		mask = ~mask_from_dct_mask(pvt, csrow);
+
+		if ((input_addr & mask) == (base & mask)) {
+			debugf2("InputAddr 0x%lx matches csrow %d "
+				"(MC node %d)\n",
+				(unsigned long)input_addr,
+				csrow, pvt->mc_node_id);
+			return csrow;
+		}
+	}
+
+	debugf2("no matching csrow for InputAddr 0x%lx (MC node %d)\n",
+		(unsigned long)input_addr, pvt->mc_node_id);
+	return -1;
+}
+
+/* Return the base value defined by the DRAM Base register for the node
+ * represented by mci.  This function returns the full 40-bit value despite
+ * the fact that the register only stores bits 39-24 of the value.  See
+ * section 3.4.4.1.
+ */
+static inline u64 get_dram_base(struct mem_ctl_info *mci)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	return pvt->dram_base[pvt->mc_node_id];
+}
+
+/*
+ * Obtain info from the DRAM Hole Address Register (section 3.4.8, pub #26094)
+ * for the node represented by mci.  Info is passed back in *hole_base,
+ * *hole_offset, and *hole_size.  Function returns 0 if info is valid or 1 if
+ * info is invalid. Info may be invalid for either of the following reasons:
+ *
+ *     - The revision of the node is not E or greater.  In this case, the DRAM
+ *       Hole Address Register does not exist.
+ *     - The DramHoleValid bit is cleared in the DRAM Hole Address Register,
+ *       indicating that its contents are not valid.
+ *
+ * The values passed back in *hole_base, *hole_offset, and *hole_size are
+ * complete 32-bit values despite the fact that the bitfields in the DHAR
+ * only represent bits 31-24 of the base and offset values.
+ */
+static int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
+				    u64 *hole_offset, u64 *hole_size)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u64 base;
+
+	/* only revE and later have the DRAM Hole Address Register */
+	if (boot_cpu_data.x86 == 0xf && pvt->ext_model < OPTERON_CPU_REV_E) {
+		debugf1("  revision %d for node %d does not support DHAR\n",
+			pvt->ext_model, pvt->mc_node_id);
+		return 1;
+	}
+
+	/* only valid for Fam10h */
+	if (boot_cpu_data.x86 == 0x10 &&
+	    (pvt->dhar & F10_DRAM_MEM_HOIST_VALID) == 0) {
+		debugf1("  Dram Memory Hoisting is DISABLED on this system\n");
+		return 1;
+	}
+
+	if ((pvt->dhar & DHAR_VALID) == 0) {
+		debugf1("  Dram Memory Hoisting is DISABLED on this node %d\n",
+			pvt->mc_node_id);
+		return 1;
+	}
+
+	/* This node has Memory Hoisting */
+
+	/* +------------------+--------------------+--------------------+-----
+	 * | memory           | DRAM hole          | relocated          |
+	 * | [0, (x - 1)]     | [x, 0xffffffff]    | addresses from     |
+	 * |                  |                    | DRAM hole          |
+	 * |                  |                    | [0x100000000,      |
+	 * |                  |                    |  (0x100000000+     |
+	 * |                  |                    |   (0xffffffff-x))] |
+	 * +------------------+--------------------+--------------------+-----
+	 *
+	 * Above is a diagram of physical memory showing the DRAM hole and the
+	 * relocated addresses from the DRAM hole.  As shown, the DRAM hole
+	 * starts at address x (the base address) and extends through address
+	 * 0xffffffff.  The DRAM Hole Address Register (DHAR) relocates the
+	 * addresses in the hole so that they start at 0x100000000.
+	 */
+
+	base = dhar_base(pvt->dhar);
+
+	*hole_base = base;
+	*hole_size = (0x1ull << 32) - base;
+
+	if (boot_cpu_data.x86 > 0xf)
+		*hole_offset = f10_dhar_offset(pvt->dhar);
+	else
+		*hole_offset = k8_dhar_offset(pvt->dhar);
+
+	debugf1("  DHAR info for node %d base 0x%lx offset 0x%lx size 0x%lx\n",
+		pvt->mc_node_id, (unsigned long)*hole_base,
+		(unsigned long)*hole_offset, (unsigned long)*hole_size);
+
+	return 0;
+}
+
+
-- 
1.6.2.4

[PATCH 07/21] amd64_edac: add DRAM address type conversion facilities

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  300 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 300 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 2033dd4..7c2f5fc 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -1450,4 +1450,304 @@ static int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
 	return 0;
 }
 
+/* Return the DramAddr that the SysAddr given by sys_addr maps to.  It is
+ * assumed that sys_addr maps to the node given by mci.
+ */
+static u64 sys_addr_to_dram_addr(struct mem_ctl_info *mci, u64 sys_addr)
+{
+	u64 dram_base, hole_base, hole_offset, hole_size, dram_addr;
+	int rc;
+
+	/* The first part of section 3.4.4 (p. 70) shows how the DRAM Base
+	 * (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers are
+	 * used to translate a SysAddr to a DramAddr.  If the DRAM Hole
+	 * Address Register (DHAR) is enabled, then it is also involved in
+	 * translating a SysAddr to a DramAddr.  Sections 3.4.8 and 3.5.8.2
+	 * describe the DHAR and how it is used for memory hoisting.  These
+	 * parts of the documentation are unclear.  I interpret them as
+	 * follows:
+	 *
+	 *     When node n receives a SysAddr, it processes the SysAddr as
+	 *     follows:
+	 *
+	 *         1.  It extracts the DRAMBase and DRAMLimit values from the
+	 *             DRAM Base and DRAM Limit registers for node n.  If the
+	 *             SysAddr is not within the range specified by the base
+	 *             and limit values, then node n ignores the Sysaddr
+	 *             (since it does not map to node n).  Otherwise continue
+	 *             to step 2 below.
+	 *
+	 *         2.  If the DramHoleValid bit of the DHAR for node n is
+	 *             clear, the DHAR is disabled so skip to step 3 below.
+	 *             Otherwise see if the SysAddr is within the range of
+	 *             relocated addresses (starting at 0x100000000) from the
+	 *             DRAM hole.  If not, skip to step 3 below.  Else get the
+	 *             value of the DramHoleOffset field from the DHAR.  To
+	 *             obtain the DramAddr, subtract the offset defined by
+	 *             this value from the SysAddr.
+	 *
+	 *         3.  Obtain the base address for node n from the DRAMBase
+	 *             field of the DRAM Base register for node n.  To obtain
+	 *             the DramAddr, subtract the base address from the
+	 *             SysAddr, as shown near the start of section 3.4.4
+	 *             (p. 70).
+	 */
+
+	dram_base = get_dram_base(mci);
+
+	rc = amd64_get_dram_hole_info(mci, &hole_base, &hole_offset,
+				      &hole_size);
+	if (!rc) {
+		if ((sys_addr >= (1ull << 32)) &&
+		    (sys_addr < ((1ull << 32) + hole_size))) {
+			/* use DHAR to translate SysAddr to DramAddr */
+			dram_addr = sys_addr - hole_offset;
+			debugf2("using DHAR to translate SysAddr 0x%lx to "
+				"DramAddr 0x%lx\n",
+				(unsigned long)sys_addr,
+				(unsigned long)dram_addr);
+			return dram_addr;
+		}
+	}
+
+	/* Translate the SysAddr to a DramAddr as shown near the start of
+	 * section 3.4.4 (p. 70).  Although sys_addr is a 64-bit value, the k8
+	 * only deals with 40-bit values.  Therefore we discard bits 63-40 of
+	 * sys_addr below.  If bit 39 of sys_addr is 1 then the bits we
+	 * discard are all 1s.  Otherwise the bits we discard are all 0s.  See
+	 * section 3.4.2 of AMD publication 24592: AMD x86-64 Architecture
+	 * Programmer's Manual Volume 1 Application Programming.
+	 */
+	dram_addr = (sys_addr & 0xffffffffffull) - dram_base;
+
+	debugf2("using DRAM Base register to translate SysAddr 0x%lx to "
+		"DramAddr 0x%lx\n", (unsigned long)sys_addr,
+		(unsigned long)dram_addr);
+	return dram_addr;
+}
+
+/* Parameter intlv_en is the value of the IntlvEn field from a DRAM Base
+ * register (section 3.4.4.1).  Return the number of bits from a SysAddr that
+ * are used for node interleaving.
+ */
+static int num_node_interleave_bits(unsigned intlv_en)
+{
+	static const int intlv_shift_table[] = { 0, 1, 0, 2, 0, 0, 0, 3 };
+	int n;
+
+	BUG_ON(intlv_en > 7);
+	n = intlv_shift_table[intlv_en];
+	return n;
+}
+
+/* Translate the DramAddr given by dram_addr to an InputAddr and return the
+ * result.
+ */
+static u64 dram_addr_to_input_addr(struct mem_ctl_info *mci, u64 dram_addr)
+{
+	struct amd64_pvt *pvt;
+	int intlv_shift;
+	u64 input_addr;
+
+	pvt = mci->pvt_info;
+
+	/* See the start of section 3.4.4 (p. 70) in the k8 documentation
+	 * concerning translating a DramAddr to an InputAddr.
+	 */
+	intlv_shift = num_node_interleave_bits(pvt->dram_IntlvEn[0]);
+	input_addr = ((dram_addr >> intlv_shift) & 0xffffff000ull) +
+	    (dram_addr & 0xfff);
+
+	debugf2("  Intlv Shift=%d DramAddr=0x%lx maps to InputAddr=0x%lx\n",
+		intlv_shift,
+		(unsigned long)dram_addr, (unsigned long)input_addr);
+	return input_addr;
+}
+
+/* Translate the SysAddr represented by sys_addr to an InputAddr and return
+ * the result.  It is assumed that sys_addr maps to the node given by mci.
+ */
+static u64 sys_addr_to_input_addr(struct mem_ctl_info *mci, u64 sys_addr)
+{
+	u64 input_addr;
+
+	input_addr =
+	    dram_addr_to_input_addr(mci, sys_addr_to_dram_addr(mci, sys_addr));
+	debugf2("%s(): SysAdddr 0x%lx translates to InputAddr 0x%lx\n",
+		__func__, (unsigned long)sys_addr, (unsigned long)input_addr);
+	return input_addr;
+}
+
+
+/* input_addr is an InputAddr associated with the node represented by mci.
+ * Translate input_addr to a DramAddr and return the result.
+ */
+static u64 input_addr_to_dram_addr(struct mem_ctl_info *mci, u64 input_addr)
+{
+	struct amd64_pvt *pvt;
+	int node_id, intlv_shift;
+	u64 bits, dram_addr;
+	u32 intlv_sel;
+
+	/* Near the start of section 3.4.4 (p. 70), the k8 documentation shows
+	 * how to translate a DramAddr to an InputAddr.  Here we reverse this
+	 * procedure.  When translating from a DramAddr to an InputAddr, the
+	 * bits used for node interleaving are discarded.  Here we recover
+	 * these bits from the IntlvSel field of the DRAM Limit register
+	 * (section 3.4.4.2) for the node that input_addr is associated with.
+	 */
+
+	pvt = mci->pvt_info;
+	node_id = pvt->mc_node_id;
+	BUG_ON((node_id < 0) || (node_id > 7));
+
+	intlv_shift = num_node_interleave_bits(pvt->dram_IntlvEn[0]);
+
+	if (intlv_shift == 0) {
+		debugf1("  node interleaving disabled:\n");
+		debugf1("    InputAddr 0x%lx translates "
+			"to DramAddr of same value\n",
+			(unsigned long)input_addr);
+		return input_addr;
+	}
+
+	bits = ((input_addr & 0xffffff000ull) << intlv_shift) +
+	    (input_addr & 0xfff);
+
+	intlv_sel = pvt->dram_IntlvSel[node_id] & ((1 << intlv_shift) - 1);
+	dram_addr = bits + (intlv_sel << 12);
+
+	debugf1("InputAddr 0x%lx translates to DramAddr 0x%lx "
+		"(%d node interleave bits)\n", (unsigned long)input_addr,
+		(unsigned long)dram_addr, intlv_shift);
+	return dram_addr;
+}
+
+/* dram_addr is a DramAddr that maps to the node represented by mci.  Convert
+ * dram_addr to a SysAddr and return the result.
+ */
+static u64 dram_addr_to_sys_addr(struct mem_ctl_info *mci, u64 dram_addr)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u64 hole_base, hole_offset, hole_size, base, limit, sys_addr;
+	int rc;
+
+	rc = amd64_get_dram_hole_info(mci, &hole_base, &hole_offset,
+				      &hole_size);
+	if (!rc) {
+		if ((dram_addr >= hole_base) &&
+		    (dram_addr < (hole_base + hole_size))) {
+			/* use DHAR to translate DramAddr to SysAddr */
+			sys_addr = dram_addr + hole_offset;
+			debugf1("using DHAR to translate DramAddr 0x%lx to "
+				"SysAddr 0x%lx\n", (unsigned long)dram_addr,
+				(unsigned long)sys_addr);
+			return sys_addr;
+		}
+	}
+
+	amd64_get_base_and_limit(pvt, pvt->mc_node_id, &base, &limit);
+	sys_addr = dram_addr + base;
+
+	/* The sys_addr we have computed up to this point is a 40-bit value
+	 * because the k8 deals with 40-bit values.  However, the value we are
+	 * supposed to return is a full 64-bit physical address.  The AMD
+	 * x86-64 architecture specifies that the most significant implemented
+	 * address bit through bit 63 of a physical address must be either all
+	 * 0s or all 1s.  Therefore we sign-extend the 40-bit sys_addr to a
+	 * 64-bit value below.  See section 3.4.2 of AMD publication 24592:
+	 * AMD x86-64 Architecture Programmer's Manual Volume 1 Application
+	 * Programming.
+	 */
+	sys_addr |= ~((sys_addr & (1ull << 39)) - 1);
+
+	debugf1("  Using DRAM Base reg on node %d to translate\n",
+		pvt->mc_node_id);
+	debugf1("    DramAddr 0x%lx to SysAddr 0x%lx\n",
+		(unsigned long)dram_addr, (unsigned long)sys_addr);
+	return sys_addr;
+}
+
+/* input_addr is an InputAddr associated with the node given by mci.
+ * Translate input_addr to a SysAddr and return the result.
+ */
+static inline u64 input_addr_to_sys_addr(struct mem_ctl_info *mci,
+					 u64 input_addr)
+{
+	return dram_addr_to_sys_addr(mci,
+				     input_addr_to_dram_addr(mci, input_addr));
+}
+
+/*
+ * Find the minimum and maximum InputAddr values that map to the given csrow.
+ * Pass back these values in *input_addr_min and *input_addr_max.
+ */
+static void find_csrow_limits(struct mem_ctl_info *mci, int csrow,
+			      u64 *input_addr_min, u64 *input_addr_max)
+{
+	struct amd64_pvt *pvt;
+	u64 base, mask;
+
+	pvt = mci->pvt_info;
+	BUG_ON((csrow < 0) || (csrow >= CHIPSELECT_COUNT));
+
+	base = base_from_dct_base(pvt, csrow);
+	mask = mask_from_dct_mask(pvt, csrow);
+
+	*input_addr_min = base & ~mask;
+	*input_addr_max = base | mask | pvt->dcs_mask_notused;
+}
+
+
+
+/*
+ * static u64 extract_error_address
+ *	Extract error address from MCA NB Address Low (section 3.6.4.5) and
+ *	MCA NB Address High (section 3.6.4.6) register values and return the
+ *	result. Address is located in the info structure (nbeah and nbeal)
+ *	the encoding is device specific.
+ */
+static u64 extract_error_address(struct mem_ctl_info *mci,
+			struct amd64_error_info_regs *info)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	return pvt->ops->get_error_address(mci, info);
+}
+
+
+/*
+ * error_address_to_page_and_offset
+ *
+ *	Map the Error address to a PAGE and PAGE OFFSET
+ */
+static inline void error_address_to_page_and_offset(u64 error_address,
+						    u32 *page, u32 *offset)
+{
+	*page = (u32) (error_address >> PAGE_SHIFT);
+	*offset = ((u32) error_address) & ~PAGE_MASK;
+}
+
+/*
+ * sys_addr_to_csrow
+ *
+ * 'sys_addr' is an error address (a SysAddr) extracted from the MCA NB Address
+ * Low (section 3.6.4.5) and MCA NB Address High (section 3.6.4.6) registers
+ * of a node that detected an ECC memory error.  mci represents the node that
+ * the error address maps to (possibly different from the node that detected
+ * the error).  Return the number of the csrow that sys_addr maps to, or -1 on
+ * error.
+ */
+static int sys_addr_to_csrow(struct mem_ctl_info *mci, u64 sys_addr)
+{
+	int csrow;
+
+	csrow = input_addr_to_csrow(mci, sys_addr_to_input_addr(mci, sys_addr));
+
+	if (csrow == -1)
+		amd64_mc_printk(mci, KERN_ERR,
+			     "Failed to translate InputAddr to csrow for "
+			     "address 0x%lx\n", (unsigned long)sys_addr);
+	return csrow;
+}
 
-- 
1.6.2.4

[PATCH 08/21] amd64_edac: add helper to dump relevant registers

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  157 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 157 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 7c2f5fc..4e84ccf 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -1751,3 +1751,160 @@ static int sys_addr_to_csrow(struct mem_ctl_info *mci, u64 sys_addr)
 	return csrow;
 }
 
+static int get_channel_from_x4_syndrome(unsigned short syndrome);
+static int get_channel_from_x8_syndrome(unsigned short syndrome);
+
+
+static void amd64_cpu_display_info(struct amd64_pvt *pvt)
+{
+	if (boot_cpu_data.x86 == 0x11)
+		edac_printk(KERN_DEBUG, EDAC_MC, "F11h CPU detected\n");
+	else if (boot_cpu_data.x86 == 0x10)
+		edac_printk(KERN_DEBUG, EDAC_MC, "F10h CPU detected\n");
+	else if (boot_cpu_data.x86 == 0xf)
+		edac_printk(KERN_DEBUG, EDAC_MC, "%s detected\n",
+			(pvt->ext_model >= OPTERON_CPU_REV_F) ?
+			"Rev F or later" : "Rev E or earlier");
+	else
+		/* we'll hardly ever ever get here */
+		edac_printk(KERN_ERR, EDAC_MC, "Unknown cpu!\n");
+}
+
+/*
+ * amd64_determine_edac_cap
+ *
+ *	Determine if the DIMMs have ECC enabled. ECC is enabled ONLY if all the
+ *	DIMMs are ECC capable.
+ */
+static enum edac_type amd64_determine_edac_cap(struct amd64_pvt *pvt)
+{
+	int bit;
+	enum dev_type edac_cap = EDAC_NONE;
+
+	bit = (boot_cpu_data.x86 > 0xf || pvt->ext_model >= OPTERON_CPU_REV_F)
+		? 19
+		: 17;
+
+	if (pvt->dclr0 >> BIT(bit)) {
+		debugf1("  edac_type is: EDAC_FLAG_SECDED\n");
+		edac_cap = EDAC_FLAG_SECDED;
+	}
+
+	return edac_cap;
+}
+
+
+static void f10_debug_display_dimm_sizes(int ctrl,
+					struct amd64_pvt *pvt, int ganged);
+
+/*
+ * amd64_dump_misc_regs
+ *
+ *	for debug purposes, display and decode various NB registers that
+ *	are for the f10 family.
+ *
+ *	This function become a no-op when DEBUG is disabled
+ */
+static void amd64_dump_misc_regs(struct amd64_pvt *pvt)
+{
+	int ganged;
+
+	debugf1("  nbcap:0x%8.08x DctDualCap=%s DualNode=%s 8-Node=%s\n",
+		pvt->nbcap,
+		(pvt->nbcap & K8_NBCAP_DCT_DUAL) ? "True" : "False",
+		(pvt->nbcap & K8_NBCAP_DUAL_NODE) ? "True" : "False",
+		(pvt->nbcap & K8_NBCAP_8_NODE) ? "True" : "False");
+	debugf1("    ECC Capable=%s   ChipKill Capable=%s\n",
+		(pvt->nbcap & K8_NBCAP_SECDED) ? "True" : "False",
+		(pvt->nbcap & K8_NBCAP_CHIPKILL) ? "True" : "False");
+	debugf1("  DramCfg0-low=0x%08x DIMM-ECC=%s Parity=%s Width=%s\n",
+		pvt->dclr0,
+		(pvt->dclr0 & BIT(19)) ?  "Enabled" : "Disabled",
+		(pvt->dclr0 & BIT(8)) ?  "Enabled" : "Disabled",
+		(pvt->dclr0 & BIT(11)) ?  "128b" : "64b");
+	debugf1("    DIMM x4 Present: L0=%s L1=%s L2=%s L3=%s  DIMM Type=%s\n",
+		(pvt->dclr0 & BIT(12)) ?  "Y" : "N",
+		(pvt->dclr0 & BIT(13)) ?  "Y" : "N",
+		(pvt->dclr0 & BIT(14)) ?  "Y" : "N",
+		(pvt->dclr0 & BIT(15)) ?  "Y" : "N",
+		(pvt->dclr0 & BIT(16)) ?  "UN-Buffered" : "Buffered");
+
+
+	debugf1("  online-spare: 0x%8.08x\n", pvt->online_spare);
+
+	if (boot_cpu_data.x86 == 0xf) {
+		/* K8 DHAR regiseter */
+		debugf1("  dhar: 0x%8.08x Base=0x%08x Offset=0x%08x\n",
+			pvt->dhar, dhar_base(pvt->dhar),
+			k8_dhar_offset(pvt->dhar));
+		debugf1("      DramHoleValid=%s\n",
+			(pvt->dhar & DHAR_VALID) ?  "True" : "False");
+
+		debugf1("  dbam-dkt: 0x%8.08x\n", pvt->dbam0);
+
+		/* everything below this point is Fam10h and above */
+		return;
+
+	} else {
+		/* F10 DHAR register */
+		debugf1("  dhar: 0x%8.08x Base=0x%08x Offset=0x%08x\n",
+			pvt->dhar, dhar_base(pvt->dhar),
+			f10_dhar_offset(pvt->dhar));
+		debugf1("    DramMemHoistValid=%s DramHoleValid=%s\n",
+			(pvt->dhar & F10_DRAM_MEM_HOIST_VALID) ?
+			"True" : "False",
+			(pvt->dhar & DHAR_VALID) ?
+			"True" : "False");
+	}
+
+	/* Only if NOT ganged does dcl1 have valid info */
+	if (!dct_ganging_enabled(pvt)) {
+		debugf1("  DramCfg1-low=0x%08x DIMM-ECC=%s Parity=%s "
+			"Width=%s\n", pvt->dclr1,
+			(pvt->dclr1 & BIT(19)) ?  "Enabled" : "Disabled",
+			(pvt->dclr1 & BIT(8)) ?  "Enabled" : "Disabled",
+			(pvt->dclr1 & BIT(11)) ?  "128b" : "64b");
+		debugf1("    DIMM x4 Present: L0=%s L1=%s L2=%s L3=%s  "
+			"DIMM Type=%s\n",
+			(pvt->dclr1 & BIT(12)) ?  "Y" : "N",
+			(pvt->dclr1 & BIT(13)) ?  "Y" : "N",
+			(pvt->dclr1 & BIT(14)) ?  "Y" : "N",
+			(pvt->dclr1 & BIT(15)) ?  "Y" : "N",
+			(pvt->dclr1 & BIT(16)) ?  "UN-Buffered" : "Buffered");
+	}
+
+	/*
+	 * Determine if ganged and then dump memory sizes for first controller,
+	 * and if NOT ganged dump info for 2nd controller.
+	 */
+	ganged = dct_ganging_enabled(pvt);
+
+	f10_debug_display_dimm_sizes(0, pvt, ganged);
+
+	if (!ganged)
+		f10_debug_display_dimm_sizes(1, pvt, ganged);
+}
+
+/*
+ * amd64_read_dbam_reg
+ *
+ * Read in both of DBAM registers
+ */
+static void amd64_read_dbam_reg(struct amd64_pvt *pvt)
+{
+	int err;
+
+	err = pci_read_config_dword(pvt->dram_f2_ctl, DBAM0, &pvt->dbam0);
+	if (err != 0)
+		debugf0("%s() Reading DBAM0 failed\n", __func__);
+
+	if (boot_cpu_data.x86 >= 0x10) {
+		err = pci_read_config_dword(pvt->dram_f2_ctl, DBAM1,
+					&pvt->dbam1);
+
+		if (err != 0)
+			debugf0("%s() Reading DBAM1 failed\n", __func__);
+	}
+}
+
+
-- 
1.6.2.4

[PATCH 09/21] amd64_edac: assign DRAM chip select base and mask in a family-specific way

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  153 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 153 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 4e84ccf..937e1f5 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -1907,4 +1907,157 @@ static void amd64_read_dbam_reg(struct amd64_pvt *pvt)
 	}
 }
 
+/*
+ * amd64_set_dct_base_and_mask(pvt)
+ *
+ *	NOTE: CPU Revision Dependent code: Rev E and Rev F
+ *
+ *	Set the DCSB and DCSM mask values depending on the CPU revision value.
+ *	Also set the shift factor for the DCSB and DCSM values.
+ *
+ *	->dcs_mask_notused, REV E:
+ *
+ *	To find the max InputAddr for the csrow, start with the base
+ *	address and set all bits that are "don't care" bits in the test at
+ *	the start of section 3.5.4 (p. 84).
+ *
+ *	The "don't care" bits are all set bits in the mask and
+ *	all bits in the gaps between bit ranges [35-25] and [19-13].
+ *	The value REV_E_DCS_NOTUSED_BITS represents bits [24-20] and [12-0],
+ *	which are all bits in the above-mentioned gaps.
+ *
+ *	->dcs_mask_notused, REV F and later:
+ *
+ *	To find the max InputAddr for the csrow, start with the base
+ *	address and set all bits that are "don't care" bits in the test at
+ *	the start of NPT section 4.5.4 (p. 87).
+ *
+ *	The "don't care" bits are all set bits in the mask and
+ *	all bits in the gaps between bit ranges [36-27] and [21-13].
+ *
+ *	The value REV_F_F1Xh_DCS_NOTUSED_BITS represents bits [26-22] and
+ *	[12-0], which are all bits in the above-mentioned gaps.
+ */
+static void amd64_set_dct_base_and_mask(struct amd64_pvt *pvt)
+{
+	if (pvt->ext_model >= OPTERON_CPU_REV_F) {
+		pvt->dcsb_base        = REV_F_F1Xh_DCSB_BASE_BITS;
+		pvt->dcsm_mask        = REV_F_F1Xh_DCSM_MASK_BITS;
+		pvt->dcs_mask_notused    = REV_F_F1Xh_DCS_NOTUSED_BITS;
+		pvt->dcs_shift        = REV_F_F1Xh_DCS_SHIFT;
+
+		switch (boot_cpu_data.x86) {
+		case 0xf:
+			pvt->num_dcsm = REV_F_DCSM_COUNT;
+			break;
+
+		case 0x10:
+			pvt->num_dcsm = F10_DCSM_COUNT;
+			break;
+
+		case 0x11:
+			pvt->num_dcsm = F11_DCSM_COUNT;
+			break;
+
+		default:
+			amd64_printk(KERN_ERR, "Unsupported family!\n");
+			break;
+		}
+	} else {
+		pvt->dcsb_base        = REV_E_DCSB_BASE_BITS;
+		pvt->dcsm_mask        = REV_E_DCSM_MASK_BITS;
+		pvt->dcs_mask_notused    = REV_E_DCS_NOTUSED_BITS;
+		pvt->dcs_shift        = REV_E_DCS_SHIFT;
+		pvt->num_dcsm        = REV_E_DCSM_COUNT;
+	}
+}
+
+/*
+ * amd64_read_dct_base_mask
+ *
+ *	Function 2 Offset F10_DCSB0
+ *	Read in the DCS Base and DCS Mask hw registers
+ */
+static void amd64_read_dct_base_mask(struct amd64_pvt *pvt)
+{
+	int cs;
+	int err;
+	int reg;
+
+	debugf0("%s()\n", __func__);
+
+	amd64_set_dct_base_and_mask(pvt);
+
+	for (cs = 0; cs < CHIPSELECT_COUNT; cs++) {
+		reg = K8_DCSB0 + (cs * 4);
+		err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
+						&pvt->dcsb0[cs]);
+		if (err != 0)
+			debugf0("%s() Reading K8_DCSB0[%d] failed\n",
+				__func__, cs);
+
+		debugf0("  DCSB0[%d]=0x%08x reg: F2x%x\n",
+			cs, pvt->dcsb0[cs], reg);
+
+		/* If DCT are NOT ganged, then read in DCT1's base */
+		if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) {
+			reg = F10_DCSB1 + (cs * 4);
+			err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
+							&pvt->dcsb1[cs]);
+			if (err != 0)
+				debugf0("%s() Reading F10_DCSB1[%d] failed\n",
+					__func__, cs);
+			debugf0("  DCSB1[%d]=0x%08x reg: F2x%x\n",
+				cs, pvt->dcsb1[cs], reg);
+		} else {
+			pvt->dcsb1[cs] = 0;
+		}
+	}
+
+	for (cs = 0; cs < pvt->num_dcsm; cs++) {
+		reg = K8_DCSB0 + (cs * 4);
+		err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
+					&pvt->dcsm0[cs]);
+		if (err != 0)
+			debugf0("%s() Reading K8_DCSM0 failed\n", __func__);
+		else
+			debugf0("    DCSM0[%d]=0x%08x reg: F2x%x\n",
+				cs, pvt->dcsm0[cs], reg);
+
+		/* If DCT are NOT ganged, then read in DCT1's mask */
+		if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) {
+			reg = F10_DCSM1 + (cs * 4);
+			err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
+					&pvt->dcsm1[cs]);
+			if (err != 0)
+				debugf0("%s() Reading F10_DCSM1[%d] failed\n",
+					__func__, cs);
+			else
+				debugf0("    DCSM1[%d]=0x%08x reg: F2x%x\n",
+					cs, pvt->dcsm1[cs], reg);
+		} else
+			pvt->dcsm1[cs] = 0;
+	}
+}
+
+static enum mem_type amd64_determine_memory_type(struct amd64_pvt *pvt)
+{
+	enum mem_type type;
+
+	if (boot_cpu_data.x86 >= 0x10 || pvt->ext_model >= OPTERON_CPU_REV_F) {
+		/* Rev F and later */
+		type = (pvt->dclr0 & BIT(16)) ? MEM_DDR2 : MEM_RDDR2;
+	} else {
+		/* Rev E and earlier */
+		type = (pvt->dclr0 & BIT(18)) ? MEM_DDR : MEM_RDDR;
+	}
+
+	debugf1("  Memory type is: %s\n",
+		(type == MEM_DDR2) ? "MEM_DDR2" :
+		(type == MEM_RDDR2) ? "MEM_RDDR2" :
+		(type == MEM_DDR) ? "MEM_DDR" : "MEM_RDDR");
+
+	return type;
+}
+
 
-- 
1.6.2.4

[PATCH 10/21] amd64_edac: add k8-specific methods

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  200 ++++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 197 insertions(+), 3 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 937e1f5..813a1ab 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -1751,9 +1751,7 @@ static int sys_addr_to_csrow(struct mem_ctl_info *mci, u64 sys_addr)
 	return csrow;
 }
 
-static int get_channel_from_x4_syndrome(unsigned short syndrome);
-static int get_channel_from_x8_syndrome(unsigned short syndrome);
-
+static int get_channel_from_ecc_syndrome(unsigned short syndrome);
 
 static void amd64_cpu_display_info(struct amd64_pvt *pvt)
 {
@@ -2060,4 +2058,200 @@ static enum mem_type amd64_determine_memory_type(struct amd64_pvt *pvt)
 	return type;
 }
 
+/*
+ * k8_early_channel_count
+ *
+ *      NOTE: CPU Revision Dependent code
+ *
+ *      MUST read the hardware DCL register, and decode it, then return it
+ *
+ *      the DCL - DRAM Configuration Low Register contains various
+ *      configuration bits on memory.
+ *
+ *      BUT it is different between CG, D & E revs and the later
+ *      Rev F memory controllers (DDR vs DDR2)
+ *
+ * Return:
+ *      number of Memory Channels in operation
+ * Pass back:
+ *      contents of the DCL0_LOW register
+ */
+static int k8_early_channel_count(struct amd64_pvt *pvt)
+{
+	int flag;
+	int err;
+
+	err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0);
+	if (err != 0)
+		debugf0("%s() Reading F10_DCLR_0 failed\n", __func__);
+
+	if ((boot_cpu_data.x86_model >> 4) >= OPTERON_CPU_REV_F) {
+		/* Rev F (NPT) and later */
+		flag = pvt->dclr0 & F10_WIDTH_128;
+	} else {
+		/* Rev E and earlier */
+		flag = pvt->dclr0 & REVE_WIDTH_128;
+	}
+
+	/* not used */
+	pvt->dclr1 = 0;
+
+	return (flag) ? 2 : 1;
+}
+
+/*
+ * k8_get_error_address
+ *	extract from the hardware copies of the error register
+ *	the ERROR ADDRESS for the K8 and Family 0Fh CPUs
+ */
+static u64 k8_get_error_address(struct mem_ctl_info *mci,
+			struct amd64_error_info_regs *info)
+{
+	return (((u64) (info->nbeah & 0xff)) << 32) +
+			(info->nbeal & ~0x03);
+}
+
+/*
+ * k8_read_dram_base_limit
+ *
+ *	Read the Base and Limit registers for K8 based Memory controllers
+ *	Extract fields from the 'raw' reg into separate data fields
+ *
+ *		Isolates: BASE, LIMIT, IntlvEn, IntlvSel, RW_EN
+ */
+static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram)
+{
+	u32 low;
+	u32 off = dram << 3;	/* 8 bytes between DRAM entries */
+	int err;
+
+	err = pci_read_config_dword(pvt->addr_f1_ctl,
+					K8_DRAM_BASE_LOW + off, &low);
+	if (err != 0)
+		debugf0("%s() Reading K8_DRAM_BASE_LOW failed\n", __func__);
+
+	/* Extract parts into separate data entries */
+	pvt->dram_base[dram] = ((u64) low & 0xFFFF0000) << 8;
+	pvt->dram_IntlvEn[dram] = (low >> 8) & 0x7;
+	pvt->dram_rw_en[dram] = (low & 0x3);
+
+	err = pci_read_config_dword(pvt->addr_f1_ctl, K8_DRAM_LIMIT_LOW + off,
+		&low);
+	if (err != 0)
+		debugf0("%s() Reading K8_DRAM_LIMIT_LOW failed\n", __func__);
+
+	/* Extract parts into separate data entries
+	 * Limit is the HIGHEST memory location of the region, so lower
+	 * 24-bit needs to be all ones
+	 */
+	pvt->dram_limit[dram] = (((u64) low & 0xFFFF0000) << 8) | 0x00FFFFFF;
+	pvt->dram_IntlvSel[dram] = (low >> 8) & 0x7;
+	pvt->dram_DstNode[dram] = (low & 0x7);
+}
+
+
+/*
+ * k8_map_sysaddr_to_csrow
+ *
+ *	map a SystemAddress to NodeID, CSROW, Channel
+ */
+static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci,
+					struct amd64_error_info_regs *info,
+					u64 SystemAddress)
+{
+	struct mem_ctl_info *src_mci;
+	unsigned short syndrome;
+	int channel, csrow;
+	u32 page, offset;
+
+	/* Extract the syndrome parts and form a 16-bit syndrome */
+	syndrome = EXTRACT_HIGH_SYNDROME(info->nbsl) << 8;
+	syndrome |= EXTRACT_LOW_SYNDROME(info->nbsh);
+
+	/* CHIPKILL enabled */
+	if (info->nbcfg & K8_NBCFG_CHIPKILL) {
+		/* x4 chipkill ecc mode - determine channel */
+		channel = get_channel_from_ecc_syndrome(syndrome);
+		if (channel < 0) {
+			/* Syndrome didn't map, so we don't know which of
+			 * the 2 DIMMs is in error. So we need to ID 'both'
+			 * of them as suspect.
+			 */
+			amd64_mc_printk(mci, KERN_WARNING,
+				     "unknown syndrome 0x%x - possible error "
+				     "reporting race\n", syndrome);
+			edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
+			return;
+		}
+	} else {
+		/* non-chipkill ecc mode
+		 *
+		 * The k8 documentation is unclear about how to determine the
+		 * channel number when using non-chipkill memory.  This method
+		 * was obtained from email communication with someone at AMD.
+		 * (Wish the email was placed in this comment - norsk)
+		 */
+		channel = ((SystemAddress & BIT(3)) != 0);
+	}
+
+	/* Find out which node the error address belongs to.  This may be
+	 * different from the node that detected the error.
+	 */
+	src_mci = find_mc_by_sys_addr(mci, SystemAddress);
+	if (src_mci == NULL) {
+		amd64_mc_printk(mci, KERN_ERR,
+			     "failed to map error address 0x%lx to a node\n",
+			     (unsigned long)SystemAddress);
+		edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
+		return;
+	}
+
+	/* Now map the SystemAddress to a CSROW */
+	csrow = sys_addr_to_csrow(src_mci, SystemAddress);
+	if (csrow < 0) {
+		edac_mc_handle_ce_no_info(src_mci, EDAC_MOD_STR);
+	} else {
+		error_address_to_page_and_offset(SystemAddress, &page, &offset);
+
+		edac_mc_handle_ce(src_mci, page, offset, syndrome, csrow,
+				  channel, EDAC_MOD_STR);
+	}
+}
+
+/*
+ * k8_dbam_map_to_pages
+ *
+ *	determrine the number of PAGES in for this DIMM's size
+ *	based on its DRAM Address Mapping.
+ */
+static int k8_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map)
+{
+	int nr_pages;
+
+	/* First step is to calc the number of bits to shift a value of 1
+	 * left to indicate show many pages. Start with the DBAM value
+	 * as the starting bits, then proceed to adjust those shift
+	 * bits, based on CPU REV and the table. See BKDG on the DBAM
+	 */
+	if (pvt->ext_model >= OPTERON_CPU_REV_F) {
+
+		/* REV F and greater section */
+		nr_pages = 1 << (revf_quad_ddr2_shift[dram_map] - PAGE_SHIFT);
+	} else {
+		/* REV E and less section This line is tricky.
+		 * It collapses the table used by revision D and later to one
+		 * that matches revision CG and earlier
+		 */
+		dram_map -= (pvt->ext_model >= OPTERON_CPU_REV_D) ?
+				(dram_map > 8 ? 4 : (dram_map > 5 ?
+				3 : (dram_map > 2 ? 1 : 0))) : 0;
+
+		/* 25 shift, is 32MiB minimum DIMM size in REV E and prior
+		 */
+		nr_pages = 1 << (dram_map + 25 - PAGE_SHIFT);
+	}
+
+	return nr_pages;
+}
+
 
-- 
1.6.2.4

[PATCH 11/21] amd64_edac: add f10-and-later methods-p1

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  233 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 233 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 813a1ab..2658852 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -2254,4 +2254,237 @@ static int k8_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map)
 	return nr_pages;
 }
 
+/*
+ * f10_early_channel_count
+ *
+ * Get the number of DCT channels in use
+ *
+ * Return:
+ *	number of Memory Channels in operation
+ * Pass back:
+ *	contents of the DCL0_LOW register
+ */
+static int f10_early_channel_count(struct amd64_pvt *pvt)
+{
+	int err;
+	int channels = 0;
+	u32 dbam;
+
+	err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0);
+	if (err != 0)
+		debugf0("%s() Reading F10_DCLR_0 failed\n", __func__);
+
+	err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_1, &pvt->dclr1);
+	if (err != 0)
+		debugf0("%s() Reading F10_DCLR_0 failed\n", __func__);
+
+	/* If we are in 128 bit mode, then we are using 2 channels */
+	if (pvt->dclr0 & F10_WIDTH_128) {
+		debugf0("%s() Data WIDTH is 128 bits - 2 channels\n",
+			__func__);
+		channels = 2;
+		return channels;
+	}
+
+	/* Need to check if in UN-ganged mode: In such, there are 2
+	 * channels, but they are NOT in 128 bit mode and thus the above
+	 * 'dcl0' status bit will be OFF, but there still are 2 channels
+	 *
+	 * Need to check DCT0[0] and DCT1[0] to see if only one of them
+	 * has their CSEnable bit on. If so, then SINGLE DIMM case.
+	 */
+	debugf0("%s() Data WIDTH is NOT 128 bits - need more decoding\n",
+		__func__);
+
+	/* Check DRAM Bank Address Mapping values for each DIMM
+	 * to see if there is more than just one DIMM present in an
+	 * unganged mode. Need to check both controllers sinc DIMMs
+	 * can be placed in either one
+	 */
+	channels = 0;
+	err = pci_read_config_dword(pvt->dram_f2_ctl, DBAM0, &dbam);
+	if (err != 0)
+		debugf0("%s() Reading DBAM0 failed\n", __func__);
+
+	if (DBAM_DIMM(0, dbam) > 0)
+		channels++;
+	if (DBAM_DIMM(1, dbam) > 0)
+		channels++;
+	if (DBAM_DIMM(2, dbam) > 0)
+		channels++;
+	if (DBAM_DIMM(3, dbam) > 0)
+		channels++;
+
+	/* If more than 2 DIMMs are present, then we have 2 channels */
+	if (channels > 2)
+		channels = 2;
+	else if (channels == 0) {
+		/* No DIMMs on DCT0, so look at DCT1 */
+		err = pci_read_config_dword(pvt->dram_f2_ctl, DBAM0, &dbam);
+		if (err != 0)
+			debugf0("%s() Reading DBAM1 failed\n", __func__);
+
+		if (DBAM_DIMM(0, dbam) > 0)
+			channels++;
+		if (DBAM_DIMM(1, dbam) > 0)
+			channels++;
+		if (DBAM_DIMM(2, dbam) > 0)
+			channels++;
+		if (DBAM_DIMM(3, dbam) > 0)
+			channels++;
+
+		if (channels > 2)
+			channels = 2;
+	}
+
+	/* If we found ALL 0 values, then assume just ONE DIMM-ONE Channel */
+	if (channels == 0)
+		channels = 1;
+
+	debugf0("%s() DIMM count= %d\n", __func__, channels);
+
+	return channels;
+}
+
+/*
+ * f10_dbam_map_to_pages
+ */
+static int f10_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map)
+{
+	return 1 << (revf_quad_ddr2_shift[dram_map] - PAGE_SHIFT);
+}
+
+/*
+ * amd64_setup
+ *
+ *	Perform needed operations on this NB thare are needed
+ *	for the duration of time that the module is operating
+ *
+ *	1) Enable "extended configuration access via 0xCF8" feature
+ */
+static void amd64_setup(struct amd64_pvt *pvt)
+{
+	u32 reg;
+	int err;
+
+	err = pci_read_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, &reg);
+	if (err != 0)
+		debugf0("%s() Reading F10_NB_CFG_HIGH failed\n",
+			__func__);
+
+	pvt->flags.cf8_extcfg = !!(reg & F10_NB_CFG_LOW_ENABLE_EXT_CFG);
+	reg |= F10_NB_CFG_LOW_ENABLE_EXT_CFG;
+	pci_write_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, reg);
+}
+
+/*
+ * amd64_teardown
+ *
+ *	teardown any items that were 'setup' in the setup func
+ *
+ *	1) Restore the "extended configuration access via 0xCF8" feature
+ */
+static void amd64_teardown(struct amd64_pvt *pvt)
+{
+	u32 reg;
+	int err;
+
+	err = pci_read_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, &reg);
+	if (err != 0)
+		debugf0("%s() Reading F10_NB_CFG_HIGH failed\n",
+			__func__);
+
+	reg &= ~F10_NB_CFG_LOW_ENABLE_EXT_CFG;
+	if (pvt->flags.cf8_extcfg)
+		reg |= F10_NB_CFG_LOW_ENABLE_EXT_CFG;
+	pci_write_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, reg);
+}
+
+/*
+ * f10_get_error_address
+ *	extract from the hardware copies of the error register
+ *	the ERROR ADDRESS for Family 10h CPUs
+ */
+static u64 f10_get_error_address(struct mem_ctl_info *mci,
+			struct amd64_error_info_regs *info)
+{
+	return (((u64) (info->nbeah & 0xffff)) << 32) +
+			(info->nbeal & ~0x01);
+}
+
+/*
+ * f10_read_dram_base_limit
+ *
+ *	Read the Base and Limit registers for F10 based Memory controllers
+ *	Extract fields from the 'raw' reg into separate data fields
+ *
+ *		Isolates: BASE, LIMIT, IntlvEn, IntlvSel, RW_EN
+ */
+static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram)
+{
+	u32 high_offset;
+	u32 low_offset;
+	u32 high_base;
+	u32 low_base;
+	u32 high_limit;
+	u32 low_limit;
+	int err;
+
+	low_offset = K8_DRAM_BASE_LOW + (dram << 3);
+	high_offset = F10_DRAM_BASE_HIGH + (dram << 3);
+
+	/* read the 'raw' DRAM BASE Address register */
+	err = pci_read_config_dword(pvt->addr_f1_ctl, low_offset, &low_base);
+	if (err != 0)
+		debugf0("%s() Reading low_offset= 0x%x failed\n", __func__,
+			low_offset);
+
+	/* Read from the ECS data register */
+	err = pci_read_config_dword(pvt->addr_f1_ctl,
+					high_offset, &high_base);
+	if (err != 0)
+		debugf0("%s() Reading high_offset= 0x%x failed. Default=0\n",
+			__func__, high_offset);
+
+	/* Extract parts into separate data entries */
+	pvt->dram_rw_en[dram] = (low_base & 0x3);
+
+	if (pvt->dram_rw_en[dram] == 0)
+		return;
+
+	pvt->dram_IntlvEn[dram] = (low_base >> 8) & 0x7;
+
+	pvt->dram_base[dram] = (((((u64) high_base & 0x000000FF) << 32) |
+				((u64) low_base & 0xFFFF0000))) << 8;
 
+	low_offset = K8_DRAM_LIMIT_LOW + (dram << 3);
+	high_offset = F10_DRAM_LIMIT_HIGH + (dram << 3);
+
+	/* read the 'raw' LIMIT registers */
+	err = pci_read_config_dword(pvt->addr_f1_ctl, low_offset, &low_limit);
+	if (err != 0)
+		debugf0("%s() Reading low_offset= 0x%x failed\n", __func__,
+			low_offset);
+
+	/* Read from the ECS data register for the HIGH portion */
+	err = pci_read_config_dword(pvt->addr_f1_ctl,
+			high_offset, &high_limit);
+	if (err != 0)
+		debugf0("%s() Reading high_ofset= 0x%x failed\n", __func__,
+			high_offset);
+
+	debugf0("%s() on dram=%d\n", __func__, dram);
+	debugf0("  HW Regs: BASE=0x%08x-%08x      LIMIT=  0x%08x-%08x\n",
+		high_base, low_base, high_limit, low_limit);
+
+	pvt->dram_DstNode[dram] = (low_limit & 0x7);
+	pvt->dram_IntlvSel[dram] = (low_limit >> 8) & 0x7;
+
+	/* Extract address values and form a LIMIT address
+	 * Limit is the HIGHEST memory location of the region, so lower
+	 * 24-bit needs to be all ones
+	 */
+	low_limit |= 0x0000FFFF;
+	pvt->dram_limit[dram] =
+		((((u64) high_limit << 32) + (u64) low_limit) << 8) | (0xFF);
+}
-- 
1.6.2.4

[PATCH 12/21] amd64_edac: add f10-and-later methods-p2

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  239 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 239 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 2658852..fe2342c 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -2488,3 +2488,242 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram)
 	pvt->dram_limit[dram] =
 		((((u64) high_limit << 32) + (u64) low_limit) << 8) | (0xFF);
 }
+
+/*
+ * f10_read_dram_ctl_register
+ *	Read DRAM Controller Select registers for the F10 that are NOT
+ *	in the K8 series
+ */
+static void f10_read_dram_ctl_register(struct amd64_pvt *pvt)
+{
+	int err;
+
+	err = pci_read_config_dword(pvt->dram_f2_ctl,
+				F10_DCTL_SEL_LOW,
+				&pvt->dram_ctl_select_low);
+	if (err != 0) {
+		debugf0("%s() Reading F10_DCTL_SEL_LOW failed\n", __func__);
+	} else {
+		debugf0("%s() DRAM_DCTL_SEL_LOW=0x%x  DctSelBaseAddr=0x%x\n",
+				__func__,
+				pvt->dram_ctl_select_low,
+				dct_sel_baseaddr(pvt));
+
+		debugf0("  DRAM DCTs are=%s DRAM Is=%s DRAM-Ctl-"
+				"sel-hi-range=%s\n",
+				(dct_ganging_enabled(pvt) ? "GANGED "
+							: "NOT GANGED "),
+				(dct_dram_enabled(pvt) ? "Enabled "
+							: "Disabled "),
+				(dct_high_range_enabled(pvt) ? "Enabled "
+							: "Disabled "));
+
+		debugf0("  DctDatIntLv=%s  MemCleared=%s "
+				"DctSelIntLvAddr=0x%x\n",
+				(dct_data_interleave_enabled(pvt) ? "Enabled "
+							: "Disabled"),
+				(dct_memory_cleared(pvt) ? "True " : "False "),
+				dct_sel_interleave_addr(pvt));
+	}
+
+	err = pci_read_config_dword(pvt->dram_f2_ctl,
+				F10_DCTL_SEL_HIGH,
+				&pvt->dram_ctl_select_high);
+	if (err != 0)
+		debugf0("%s() Reading F10_DCTL_SEL_HIGH failed\n",
+			__func__);
+	debugf0("%s() DRAM_CTL_SELECT_HIGH=0x%x\n", __func__,
+			pvt->dram_ctl_select_high);
+}
+
+static u32 f10_determine_channel(struct amd64_pvt *pvt, u64 SystemAddr,
+				int HiRangeSelected, u32 IntlvEn)
+{
+	u32 ChannelSelect;
+	u32 temp = pvt->dram_ctl_select_low;
+	u32 DctSelIntLvAddr, interleave;
+	u32 DctSelHi;
+
+	interleave = dct_interleave_enabled(pvt);
+	DctSelIntLvAddr = dct_sel_interleave_addr(pvt);
+	DctSelHi = (temp >> 1) & 1;
+
+	if (dct_ganging_enabled(pvt))
+		ChannelSelect = 0;
+	else if (HiRangeSelected)
+		ChannelSelect = DctSelHi;
+	else if (interleave) {
+		if (DctSelIntLvAddr == 0)
+			ChannelSelect = SystemAddr >> 6 & 1;
+		else if ((DctSelIntLvAddr >> 1) & 1) {
+			temp = popcnt((u32) ((SystemAddr >> 16) & 0x1F)) % 2;
+
+			if (DctSelIntLvAddr & 1)
+				ChannelSelect = (SystemAddr >> 9 & 1) ^ temp;
+			else
+				ChannelSelect = (SystemAddr >> 6 & 1) ^ temp;
+		} else if (IntlvEn & 4)
+			ChannelSelect = SystemAddr >> 15 & 1;
+		else if (IntlvEn & 2)
+			ChannelSelect = SystemAddr >> 14 & 1;
+		else if (IntlvEn & 1)
+			ChannelSelect = SystemAddr >> 13 & 1;
+		else
+			ChannelSelect = SystemAddr >> 12 & 1;
+	} else if (dct_high_range_enabled(pvt) && !dct_ganging_enabled(pvt))
+		ChannelSelect = ~DctSelHi & 1;
+	else
+		ChannelSelect = 0;
+
+	return ChannelSelect;
+}
+
+static inline u32 f10_map_IntlvEn_to_shift(u32 IntlvEn)
+{
+	u32 shift;
+
+	if (IntlvEn == 1)
+		shift = 1;
+	else if (IntlvEn == 3)
+		shift = 2;
+	else if (IntlvEn == 7)
+		shift = 3;
+	else
+		shift = 0;
+
+	return shift;
+}
+
+static inline u64 f10_determine_base_addr_offset(
+						u64 SystemAddr,
+						int HiRangeSelected,
+						u32 DctSelBaseAddr,
+						u64 DctSelBaseOffsetLong,
+						u32 HoleEn,
+						u32 HoleOffset,
+						u64 DramBaseLong)
+{
+	u64 ChannelAddrLong;
+	u64 ChannelOffsetLong;
+
+	if (HiRangeSelected) {
+		if ((!DctSelBaseAddr & 0xFFFF0000) &&
+		   (HoleEn & 1) && (SystemAddr >= 0x100000000ULL))
+			ChannelOffsetLong = HoleOffset << 16;
+		else
+			ChannelOffsetLong = DctSelBaseOffsetLong;
+	} else {
+		if ((HoleEn & 1) && (SystemAddr >= 0x100000000ULL))
+			ChannelOffsetLong = HoleOffset << 16;
+		else
+			ChannelOffsetLong = DramBaseLong & 0xFFFFF8000000ULL;
+	}
+
+	ChannelAddrLong = (SystemAddr & 0x0000FFFFFFFFFFC0ULL) -
+			(ChannelOffsetLong & 0x0000FFFFFF800000ULL);
+
+	return ChannelAddrLong;
+}
+
+/* Hack for the time being - Can we get this from BIOS?? */
+#define	CH0SPARE_RANK	0
+#define	CH1SPARE_RANK	1
+
+/*
+ * f10_process_possible_spare
+ *
+ *	checks if the csrow passed in is marked as SPARED, if so
+ *	returns the new spare row
+ */
+static inline int f10_process_possible_spare(int csrow,
+				u32 ChannelSelect, struct amd64_pvt *pvt)
+{
+	u32 SwapDone;
+	u32 BadDramCs;
+	u32 OnLineSpareCTL;
+
+	OnLineSpareCTL = pvt->online_spare;
+
+	/* Depending on channel, isolate respective SPARING info */
+	if (ChannelSelect) {
+		SwapDone = F10_ONLINE_SPARE_SWAPDONE1(OnLineSpareCTL);
+		BadDramCs = F10_ONLINE_SPARE_BADDRAM_CS1(OnLineSpareCTL);
+		if (SwapDone && (csrow == BadDramCs))
+			csrow = CH1SPARE_RANK;
+	} else {
+		SwapDone = F10_ONLINE_SPARE_SWAPDONE0(OnLineSpareCTL);
+		BadDramCs = F10_ONLINE_SPARE_BADDRAM_CS0(OnLineSpareCTL);
+		if (SwapDone && (csrow == BadDramCs))
+			csrow = CH0SPARE_RANK;
+	}
+	return csrow;
+}
+
+/*
+ * f10_lookup_addr_in_dct
+ *
+ *	Iterate over the DRAM DCT "base" and "mask" register looking for
+ *	a SystemAddr match on the specified 'ChannelSelect' and 'NodeID'
+ *
+ * Return:
+ *	-1  NOT FOUND
+ *	0..csrow = Chip-Select Row
+ */
+static int f10_lookup_addr_in_dct(u32 InputAddr, u32 NodeID, u32 ChannelSelect)
+{
+	struct mem_ctl_info *mci;
+	struct amd64_pvt *pvt;
+	u32 CSBase, CSMask;
+	int CSFound = -1;
+	int csrow;
+
+	mci = mci_lookup[NodeID];
+	if (!mci)
+		return CSFound;
+
+	pvt = mci->pvt_info;
+
+	debugf1("%s() InputAddr=0x%x  channelselect=%d\n",
+			__func__, InputAddr, ChannelSelect);
+
+	for (csrow = 0; csrow < CHIPSELECT_COUNT; csrow++) {
+
+		CSBase = amd64_get_dct_base(pvt, ChannelSelect, csrow);
+		if (!(CSBase & K8_DCSB_CS_ENABLE))
+			continue;
+
+		/* We have an ENABLED CSROW, Isolate just the MASK
+		 * bits of the target: 28:19 and 13:5, which map to
+		 * 36:27 and 21:13 of the actual address
+		 */
+		CSBase &= REV_F_F1Xh_DCSB_BASE_BITS;
+
+		/* Get the DCT Mask, and ENABLE the reserved bits:
+		 * 18:16 and 4:0 to become ON. Then mask off bits
+		 * 28:0 (36:8)
+		 */
+		CSMask = amd64_get_dct_mask(pvt, ChannelSelect, csrow);
+
+		debugf1("    CSROW=%d CSBase=0x%x RAW CSMask=0x%x\n",
+				csrow, CSBase, CSMask);
+
+		CSMask = (CSMask | 0x0007C01F) & 0x1FFFFFFF;
+
+		debugf1("              Final CSMask=0x%x\n", CSMask);
+		debugf1("    (InputAddr & ~CSMask)=0x%x "
+				"(CSBase & ~CSMask)=0x%x\n",
+				(InputAddr & ~CSMask), (CSBase & ~CSMask));
+
+		/* Perform the lookup MATCH operation */
+		if ((InputAddr & ~CSMask) == (CSBase & ~CSMask)) {
+			CSFound = f10_process_possible_spare(csrow,
+							ChannelSelect, pvt);
+
+			debugf1(" MATCH csrow=%d\n", CSFound);
+			break;
+		}
+	}
+	return CSFound;
+}
+
+
-- 
1.6.2.4

[PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  318 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 318 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index fe2342c..84075c0 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -2726,4 +2726,322 @@ static int f10_lookup_addr_in_dct(u32 InputAddr, u32 NodeID, u32 ChannelSelect)
 	return CSFound;
 }
 
+/*
+ * f10_match_to_this_node
+ *
+ * For a given 'DramRange' value, check if 'SystemAddr' fall within this value
+ */
+static int f10_match_to_this_node(struct amd64_pvt *pvt, int DramRange,
+				u64 SystemAddr,
+				int *node_id,
+				int *channel_select)
+{
+	int CSFound = -1;
+	int NodeID;
+	int HiRangeSelected;
+	u32 IntlvEn, IntlvSel;
+	u32 DramEn;
+	u32 Ilog;
+	u32 HoleOffset, HoleEn;
+	u32 InputAddr, Temp;
+	u32 DctSelBaseAddr, DctSelIntLvAddr;
+	u32 DctSelHi;
+	u32 ChannelSelect;
+	u64 DramBaseLong, DramLimitLong;
+	u64 DctSelBaseOffsetLong, ChannelAddrLong;
+
+	/* DRAM Base value for this DRAM instance */
+	DramBaseLong = pvt->dram_base[DramRange];
+	DramEn = pvt->dram_rw_en[DramRange];
+	IntlvEn = pvt->dram_IntlvEn[DramRange];
+
+	/* DRAM Limit value for this DRAM instance */
+	DramLimitLong = pvt->dram_limit[DramRange];
+	NodeID = pvt->dram_DstNode[DramRange];
+	IntlvSel = pvt->dram_IntlvSel[DramRange];
+
+	debugf1("%s(dram=%d) Base=0x%llx SystemAddr= 0x%llx Limit=0x%llx\n",
+		__func__, DramRange, DramBaseLong, SystemAddr, DramLimitLong);
+
+	/* This assumes that one node's DHAR is the same as
+	 * all the other node's DHARs
+	 */
+	HoleEn = pvt->dhar;
+	HoleOffset = (HoleEn & 0x0000FF80);
+	HoleEn = (HoleEn & 0x00000003);
+
+	debugf1("   HoleOffset=0x%x  HoleEn=0x%x IntlvSel=0x%x\n",
+			HoleOffset, HoleEn, IntlvSel);
+
+	if ((IntlvEn == 0) || IntlvSel == ((SystemAddr >> 12) & IntlvEn)) {
+
+		Ilog = f10_map_IntlvEn_to_shift(IntlvEn);
+
+		Temp = pvt->dram_ctl_select_low;
+		DctSelBaseOffsetLong = pvt->dram_ctl_select_high << 16;
+
+		DctSelHi = (Temp >> 1) & 1;
+		DctSelIntLvAddr = dct_sel_interleave_addr(pvt);
+		DctSelBaseAddr = dct_sel_baseaddr(pvt);
+
+		if (dct_high_range_enabled(pvt) &&
+		   !dct_ganging_enabled(pvt) &&
+		   ((SystemAddr >> 27) >= (DctSelBaseAddr >> 11)))
+			HiRangeSelected = 1;
+		else
+			HiRangeSelected = 0;
+
+		ChannelSelect = f10_determine_channel(pvt, SystemAddr,
+						HiRangeSelected, IntlvEn);
+
+		ChannelAddrLong = f10_determine_base_addr_offset(
+						SystemAddr,
+						HiRangeSelected,
+						DctSelBaseAddr,
+						DctSelBaseOffsetLong,
+						HoleEn,
+						HoleOffset,
+						DramBaseLong);
+
+		/* Remove Node ID (in case of processor interleaving) */
+		Temp = ChannelAddrLong & 0xFC0;
+
+		ChannelAddrLong = ((ChannelAddrLong >> Ilog) &
+					0xFFFFFFFFF000ULL) | Temp;
+
+		/* Remove Channel interleave and hash */
+		if (dct_interleave_enabled(pvt) &&
+		   !dct_high_range_enabled(pvt) &&
+		   !dct_ganging_enabled(pvt)) {
+			if (DctSelIntLvAddr != 1)
+				ChannelAddrLong =
+					(ChannelAddrLong >> 1) &
+					0xFFFFFFFFFFFFFFC0ULL;
+			else {
+				Temp = ChannelAddrLong & 0xFC0;
+				ChannelAddrLong =
+					((ChannelAddrLong &
+					0xFFFFFFFFFFFFC000ULL)
+					>> 1) | Temp;
+			}
+		}
+
+		/* Form a normalize InputAddr (Move bits 36:8 down to 28:0
+		 * which will set it up to match the DCT Base register
+		 */
+		InputAddr = ChannelAddrLong >> 8;
+
+		debugf1("   (ChannelAddrLong=0x%llx) >> 8 becomes "
+			"InputAddr=0x%x\n", ChannelAddrLong, InputAddr);
+
+		/* Iterate over the DRAM DCTs looking for a
+		 * match for InputAddr on the selected NodeID
+		 */
+		CSFound = f10_lookup_addr_in_dct(InputAddr,
+						NodeID, ChannelSelect);
+
+		if (CSFound >= 0) {
+			*node_id = NodeID;
+			*channel_select = ChannelSelect;
+		}
+	}
+
+	return CSFound;
+}
+
+static int f10_translate_sysaddr_to_CS(struct amd64_pvt *pvt,
+			u64 SysAddr,
+			int *node,
+			int *chanSel)
+{
+	int DramRange;
+	int CSFound = -1;
+	u64 DramBaseLong, DramLimitLong;
+
+	for (DramRange = 0; DramRange < DRAM_REG_COUNT; DramRange++) {
+
+		if (!pvt->dram_rw_en[DramRange])
+			continue;
+
+		DramBaseLong = pvt->dram_base[DramRange];
+		DramLimitLong = pvt->dram_limit[DramRange];
+
+		if ((DramBaseLong <= SysAddr) && (SysAddr <= DramLimitLong)) {
+
+			CSFound = f10_match_to_this_node(pvt,
+							DramRange, SysAddr,
+							node, chanSel);
+			if (CSFound >= 0)
+				break;
+		}
+	}
+	return CSFound;
+}
+
+/*
+ * f10_map_sysaddr_to_csrow
+ *
+ *	This the F10 reference code from AMD to
+ *	map a SystemAddress to NodeID, CSROW, Channel
+ *
+ *	See the Family 10h BKDG (with bug fixes in the code from AMD)
+ *
+ *	The SystemAddress is usually an error address received from the
+ *	hardware error detector.
+ */
+static void f10_map_sysaddr_to_csrow(struct mem_ctl_info *mci,
+					struct amd64_error_info_regs *info,
+					u64 SystemAddress)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u32 page, offset;
+	unsigned short syndrome;
+	int chan = 0;
+	int node_id;
+	int csrow;
+
+	csrow = f10_translate_sysaddr_to_CS(pvt,
+					SystemAddress,
+					&node_id,
+					&chan);
+
+	if (csrow >= 0) {
+		error_address_to_page_and_offset(SystemAddress, &page, &offset);
+
+		syndrome = EXTRACT_HIGH_SYNDROME(info->nbsl) << 8;
+		syndrome |= EXTRACT_LOW_SYNDROME(info->nbsh);
+
+		/* Is CHIPKILL ON?
+		 * If so, then we can attempt to use the 'syndrome' to isolate
+		 * which channel the error was on
+		 */
+		if (pvt->nbcfg & K8_NBCFG_CHIPKILL)
+			chan = get_channel_from_ecc_syndrome(syndrome);
+
+		if (chan >= 0) {
+			edac_mc_handle_ce(mci, page, offset, syndrome,
+					csrow, chan, EDAC_MOD_STR);
+		} else {
+			/* Channel is not known,
+			 * report all channels on this CSROW as failed.
+			 */
+			for (chan = 0; chan < mci->csrows[csrow].nr_channels;
+								chan++) {
+					edac_mc_handle_ce(mci, page, offset,
+							syndrome,
+							csrow, chan,
+							EDAC_MOD_STR);
+			}
+		}
+
+	} else {
+		edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
+	}
+}
+
+/* map_dbam_to_csrow_size
+ *
+ * Input (index) is the DBAM DIMM value (1 of 4) used as an index into a
+ * shift table (revf_quad_ddr2_shift) which starts at 128MB DIMM size.
+ *
+ * Index of 0 indicates an empty DIMM slot, as reported by Hardware on
+ * empty slots.
+ *
+ * Normalize to 128MB by subracting 27 bit shift
+ */
+static int map_dbam_to_csrow_size(int index)
+{
+	int mega_bytes = 0;
+
+	if (index > 0 && index <= DBAM_MAX_VALUE)
+		mega_bytes = ((128 << (revf_quad_ddr2_shift[index]-27)));
+
+	return mega_bytes;
+}
+
+/*
+ * f10_debug_display_dimm_sizes
+ *
+ *	debug routine to display the memory sizes of a DIMM
+ *	(ganged or not) and it CSROWs as well
+ */
+static void f10_debug_display_dimm_sizes(int ctrl,
+					struct amd64_pvt *pvt, int ganged)
+{
+	int dimm;
+	int size0;
+	int size1;
+	u32 dbam;
+	u32 *dcsb;
+
+	debugf1("  dbam%d: 0x%8.08x  CSROW is %s\n", ctrl,
+			ctrl ? pvt->dbam1 : pvt->dbam0,
+			ganged ? "GANGED - dbam1 not used" : "NON-GANGED");
+
+	dbam = ctrl ? pvt->dbam1 : pvt->dbam0;
+	dcsb = ctrl ? pvt->dcsb1 : pvt->dcsb0;
+
+	/* Dump memory sizes for DIMM and its CSROWs */
+	for (dimm = 0; dimm < 4; dimm++) {
+
+		size0 = 0;
+		if (dcsb[dimm*2] & K8_DCSB_CS_ENABLE)
+			size0 = map_dbam_to_csrow_size(DBAM_DIMM(dimm, dbam));
+
+		size1 = 0;
+		if (dcsb[dimm*2 + 1] & K8_DCSB_CS_ENABLE)
+			size1 = map_dbam_to_csrow_size(DBAM_DIMM(dimm, dbam));
+
+		debugf1("     CTRL-%d DIMM-%d=%5dMB   CSROW-%d=%5dMB "
+				"CSROW-%d=%5dMB\n",
+				ctrl,
+				dimm,
+				size0 + size1,
+				dimm * 2,
+				size0,
+				dimm * 2 + 1,
+				size1);
+	}
+}
+
+/*
+ * f10_probe_valid_hardware
+ *
+ * Very early hardware probe on pci_probe thread to determine if this module can
+ * support the hardware.
+ *
+ * Return:
+ *      0 for OK
+ *      1 for error
+ */
+static int f10_probe_valid_hardware(struct amd64_pvt *pvt)
+{
+	int rc = 0;
+
+	/*
+	 * If we are on a DDR3 machine, we don't know yet if
+	 * we support that properly at this time
+	 */
+	if ((pvt->dchr0 & F10_DCHR_Ddr3Mode) ||
+	    (pvt->dchr1 & F10_DCHR_Ddr3Mode)) {
+
+		amd64_printk(KERN_WARNING,
+			"%s() This machine is running with DDR3 memory. "
+			"This is not currently supported. "
+			"DCHR0=0x%x DCHR1=0x%x\n",
+			__func__, pvt->dchr0, pvt->dchr1);
+
+		amd64_printk(KERN_WARNING,
+			"   Contact '%s' module MAINTAINER to help add"
+			" support.\n",
+			EDAC_MOD_STR);
+
+		rc = 1;
+
+	} else {
+		debugf0("%s() DDR2 Memory Installed\n", __func__);
+	}
+
+	return rc;
+}
 
-- 
1.6.2.4

[PATCH 14/21] amd64_edac: add per-family descriptors

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |   76 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 76 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 84075c0..afdb60d 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -3045,3 +3045,79 @@ static int f10_probe_valid_hardware(struct amd64_pvt *pvt)
 	return rc;
 }
 
+/*
+ * There currently are 3 types type of MC devices for AMD
+ * Athlon/Opterons (as per PCI DEVICE_IDs):
+ *
+ * Family K8: That is the Athlon64 and Opteron CPUs. They all have the
+ * same PCI DEVICE ID, even though there is differences between
+ * the different Revisions (CG,D,E,F).
+ *
+ * Family F10h.
+ *
+ * Family F11h.
+ *
+ */
+static struct amd64_family_type amd64_family_types[] = {
+	[K8_CPUS] = {
+		.ctl_name = "RevF",
+		.addr_f1_ctl = PCI_DEVICE_ID_AMD_K8_NB_ADDRMAP,
+		.misc_f3_ctl = PCI_DEVICE_ID_AMD_K8_NB_MISC,
+		.ops = {
+			.early_channel_count = k8_early_channel_count,
+			.get_error_address = k8_get_error_address,
+			.read_dram_base_limit = k8_read_dram_base_limit,
+			.map_sysaddr_to_csrow = k8_map_sysaddr_to_csrow,
+			.dbam_map_to_pages = k8_dbam_map_to_pages,
+		}
+	},
+	[F10_CPUS] = {
+		.ctl_name = "Family 10h",
+		.addr_f1_ctl = PCI_DEVICE_ID_AMD_10H_NB_MAP,
+		.misc_f3_ctl = PCI_DEVICE_ID_AMD_10H_NB_MISC,
+		.ops = {
+			.probe_valid_hardware = f10_probe_valid_hardware,
+			.early_channel_count = f10_early_channel_count,
+			.get_error_address = f10_get_error_address,
+			.read_dram_base_limit = f10_read_dram_base_limit,
+			.read_dram_ctl_register = f10_read_dram_ctl_register,
+			.map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow,
+			.dbam_map_to_pages = f10_dbam_map_to_pages,
+		}
+	},
+	[F11_CPUS] = {
+		.ctl_name = "Family 11h",
+		.addr_f1_ctl = PCI_DEVICE_ID_AMD_11H_NB_MAP,
+		.misc_f3_ctl = PCI_DEVICE_ID_AMD_11H_NB_MISC,
+		.ops = {
+			.probe_valid_hardware = f10_probe_valid_hardware,
+			.early_channel_count = f10_early_channel_count,
+			.get_error_address = f10_get_error_address,
+			.read_dram_base_limit = f10_read_dram_base_limit,
+			.read_dram_ctl_register = f10_read_dram_ctl_register,
+			.map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow,
+			.dbam_map_to_pages = f10_dbam_map_to_pages,
+		}
+	},
+};
+
+static struct pci_dev *pci_get_related_function(unsigned int vendor,
+						unsigned int device,
+						struct pci_dev *related)
+{
+	struct pci_dev *dev;
+
+	dev = NULL;
+
+	dev = pci_get_device(vendor, device, dev);
+	while (dev != NULL) {
+		if ((dev->bus->number == related->bus->number) &&
+		    (PCI_SLOT(dev->devfn) == PCI_SLOT(related->devfn)))
+			break;
+		dev = pci_get_device(vendor, device, dev);
+	}
+
+	return dev;
+}
+
+
-- 
1.6.2.4

[PATCH 15/21] amd64_edac: add ECC chipkill syndrome mapping table

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  141 ++++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 140 insertions(+), 1 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index afdb60d..88849a4 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -2170,7 +2170,6 @@ static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci,
 
 	/* CHIPKILL enabled */
 	if (info->nbcfg & K8_NBCFG_CHIPKILL) {
-		/* x4 chipkill ecc mode - determine channel */
 		channel = get_channel_from_ecc_syndrome(syndrome);
 		if (channel < 0) {
 			/* Syndrome didn't map, so we don't know which of
@@ -3120,4 +3119,144 @@ static struct pci_dev *pci_get_related_function(unsigned int vendor,
 	return dev;
 }
 
+/*
+ * syndrome mapping table for ECC ChipKill devices
+ *
+ * The comment in each row is the token (nibble) number that is in error.
+ * The least significant nibble of the syndrome is the mask for the bits
+ * that are in error (need to be toggled) for the particular nibble.
+ *
+ * Each row contains 16 entries.
+ * The first entry (0th) is the channel number for that row of syndromes.
+ * The remaining 15 entries are the syndromes for the respective Error
+ * bit mask index.
+ *
+ * 1st index entry is 0x0001 mask, indicating that the rightmost bit is the
+ * bit in error.
+ * The 2nd index entry is 0x0010 that the second bit is damaged.
+ * The 3rd index entry is 0x0011 indicating that the rightmost 2 bits
+ * are damaged.
+ * Thus so on until index 15, 0x1111, whose entry has the syndrome
+ * indicating that all 4 bits are damaged.
+ *
+ * A search is performed on this table looking for a given syndrome.
+ *
+ * See the AMD documentation for ECC syndromes. This ECC table is valid
+ * across all the versions of the AMD64 processors.
+ *
+ * A fast lookup is to use the LAST four bits of the 16-bit syndrome as a
+ * COLUMN index, then search all ROWS of that column, looking for a match
+ * with the input syndrome. The ROW value will be the token number.
+ *
+ * The 0'th entry on that row, can be returned as the CHANNEL (0 or 1) of this
+ * error.
+ */
+#define NUMBER_ECC_ROWS  36
+static const unsigned short ecc_chipkill_syndromes[NUMBER_ECC_ROWS][16] = {
+	/* Channel 0 syndromes */
+	{/*0*/  0, 0xe821, 0x7c32, 0x9413, 0xbb44, 0x5365, 0xc776, 0x2f57,
+	   0xdd88, 0x35a9, 0xa1ba, 0x499b, 0x66cc, 0x8eed, 0x1afe, 0xf2df },
+	{/*1*/  0, 0x5d31, 0xa612, 0xfb23, 0x9584, 0xc8b5, 0x3396, 0x6ea7,
+	   0xeac8, 0xb7f9, 0x4cda, 0x11eb, 0x7f4c, 0x227d, 0xd95e, 0x846f },
+	{/*2*/  0, 0x0001, 0x0002, 0x0003, 0x0004, 0x0005, 0x0006, 0x0007,
+	   0x0008, 0x0009, 0x000a, 0x000b, 0x000c, 0x000d, 0x000e, 0x000f },
+	{/*3*/  0, 0x2021, 0x3032, 0x1013, 0x4044, 0x6065, 0x7076, 0x5057,
+	   0x8088, 0xa0a9, 0xb0ba, 0x909b, 0xc0cc, 0xe0ed, 0xf0fe, 0xd0df },
+	{/*4*/  0, 0x5041, 0xa082, 0xf0c3, 0x9054, 0xc015, 0x30d6, 0x6097,
+	   0xe0a8, 0xb0e9, 0x402a, 0x106b, 0x70fc, 0x20bd, 0xd07e, 0x803f },
+	{/*5*/  0, 0xbe21, 0xd732, 0x6913, 0x2144, 0x9f65, 0xf676, 0x4857,
+	   0x3288, 0x8ca9, 0xe5ba, 0x5b9b, 0x13cc, 0xaded, 0xc4fe, 0x7adf },
+	{/*6*/  0, 0x4951, 0x8ea2, 0xc7f3, 0x5394, 0x1ac5, 0xdd36, 0x9467,
+	   0xa1e8, 0xe8b9, 0x2f4a, 0x661b, 0xf27c, 0xbb2d, 0x7cde, 0x358f },
+	{/*7*/  0, 0x74e1, 0x9872, 0xec93, 0xd6b4, 0xa255, 0x4ec6, 0x3a27,
+	   0x6bd8, 0x1f39, 0xf3aa, 0x874b, 0xbd6c, 0xc98d, 0x251e, 0x51ff },
+	{/*8*/  0, 0x15c1, 0x2a42, 0x3f83, 0xcef4, 0xdb35, 0xe4b6, 0xf177,
+	   0x4758, 0x5299, 0x6d1a, 0x78db, 0x89ac, 0x9c6d, 0xa3ee, 0xb62f },
+	{/*9*/  0, 0x3d01, 0x1602, 0x2b03, 0x8504, 0xb805, 0x9306, 0xae07,
+	   0xca08, 0xf709, 0xdc0a, 0xe10b, 0x4f0c, 0x720d, 0x590e, 0x640f },
+	{/*a*/  0, 0x9801, 0xec02, 0x7403, 0x6b04, 0xf305, 0x8706, 0x1f07,
+	   0xbd08, 0x2509, 0x510a, 0xc90b, 0xd60c, 0x4e0d, 0x3a0e, 0xa20f },
+	{/*b*/  0, 0xd131, 0x6212, 0xb323, 0x3884, 0xe9b5, 0x5a96, 0x8ba7,
+	   0x1cc8, 0xcdf9, 0x7eda, 0xafeb, 0x244c, 0xf57d, 0x465e, 0x976f },
+	{/*c*/  0, 0xe1d1, 0x7262, 0x93b3, 0xb834, 0x59e5, 0xca56, 0x2b87,
+	   0xdc18, 0x3dc9, 0xae7a, 0x4fab, 0x542c, 0x85fd, 0x164e, 0xf79f },
+	{/*d*/  0, 0x6051, 0xb0a2, 0xd0f3, 0x1094, 0x70c5, 0xa036, 0xc067,
+	   0x20e8, 0x40b9, 0x904a, 0x601b, 0x307c, 0x502d, 0x80de, 0xe08f },
+	{/*e*/  0, 0xa4c1, 0xf842, 0x5c83, 0xe6f4, 0x4235, 0x1eb6, 0xba77,
+	   0x7b58, 0xdf99, 0x831a, 0x27db, 0x9dac, 0x396d, 0x65ee, 0xc12f },
+	{/*f*/  0, 0x11c1, 0x2242, 0x3383, 0xc8f4, 0xd935, 0xeab6, 0xfb77,
+	   0x4c58, 0x5d99, 0x6e1a, 0x7fdb, 0x84ac, 0x956d, 0xa6ee, 0xb72f },
+
+	/* Channel 1 syndromes */
+	{/*10*/ 1, 0x45d1, 0x8a62, 0xcfb3, 0x5e34, 0x1be5, 0xd456, 0x9187,
+	   0xa718, 0xe2c9, 0x2d7a, 0x68ab, 0xf92c, 0xbcfd, 0x734e, 0x369f },
+	{/*11*/ 1, 0x63e1, 0xb172, 0xd293, 0x14b4, 0x7755, 0xa5c6, 0xc627,
+	   0x28d8, 0x4b39, 0x99aa, 0xfa4b, 0x3c6c, 0x5f8d, 0x8d1e, 0xeeff },
+	{/*12*/ 1, 0xb741, 0xd982, 0x6ec3, 0x2254, 0x9515, 0xfbd6, 0x4c97,
+	   0x33a8, 0x84e9, 0xea2a, 0x5d6b, 0x11fc, 0xa6bd, 0xc87e, 0x7f3f },
+	{/*13*/ 1, 0xdd41, 0x6682, 0xbbc3, 0x3554, 0xe815, 0x53d6, 0xce97,
+	   0x1aa8, 0xc7e9, 0x7c2a, 0xa1fb, 0x2ffc, 0xf2bd, 0x497e, 0x943f },
+	{/*14*/ 1, 0x2bd1, 0x3d62, 0x16b3, 0x4f34, 0x64e5, 0x7256, 0x5987,
+	   0x8518, 0xaec9, 0xb87a, 0x93ab, 0xca2c, 0xe1fd, 0xf74e, 0xdc9f },
+	{/*15*/ 1, 0x83c1, 0xc142, 0x4283, 0xa4f4, 0x2735, 0x65b6, 0xe677,
+	   0xf858, 0x7b99, 0x391a, 0xbadb, 0x5cac, 0xdf6d, 0x9dee, 0x1e2f },
+	{/*16*/ 1, 0x8fd1, 0xc562, 0x4ab3, 0xa934, 0x26e5, 0x6c56, 0xe387,
+	   0xfe18, 0x71c9, 0x3b7a, 0xb4ab, 0x572c, 0xd8fd, 0x924e, 0x1d9f },
+	{/*17*/ 1, 0x4791, 0x89e2, 0xce73, 0x5264, 0x15f5, 0xdb86, 0x9c17,
+	   0xa3b8, 0xe429, 0x2a5a, 0x6dcb, 0xf1dc, 0xb64d, 0x783e, 0x3faf },
+	{/*18*/ 1, 0x5781, 0xa9c2, 0xfe43, 0x92a4, 0xc525, 0x3b66, 0x6ce7,
+	   0xe3f8, 0xb479, 0x4a3a, 0x1dbb, 0x715c, 0x26dd, 0xd89e, 0x8f1f },
+	{/*19*/ 1, 0xbf41, 0xd582, 0x6ac3, 0x2954, 0x9615, 0xfcd6, 0x4397,
+	   0x3ea8, 0x81e9, 0xeb2a, 0x546b, 0x17fc, 0xa8bd, 0xc27e, 0x7d3f },
+	{/*1a*/ 1, 0x9891, 0xe1e2, 0x7273, 0x6464, 0xf7f5, 0x8586, 0x1617,
+	   0xb8b8, 0x2b29, 0x595a, 0xcacb, 0xdcdc, 0x4f4d, 0x3d3e, 0xaeaf },
+	{/*1b*/ 1, 0xcce1, 0x4472, 0x8893, 0xfdb4, 0x3f55, 0xb9c6, 0x7527,
+	   0x56d8, 0x9a39, 0x12aa, 0xde4b, 0xab6c, 0x678d, 0xef1e, 0x23ff },
+	{/*1c*/ 1, 0xa761, 0xf9b2, 0x5ed3, 0xe214, 0x4575, 0x1ba6, 0xbcc7,
+	   0x7328, 0xd449, 0x8a9a, 0x2dfb, 0x913c, 0x365d, 0x688e, 0xcfef },
+	{/*1d*/ 1, 0xff61, 0x55b2, 0xaad3, 0x7914, 0x8675, 0x2ca6, 0xd3c7,
+	   0x9e28, 0x6149, 0xcb9a, 0x34fb, 0xe73c, 0x185d, 0xb28e, 0x4def },
+	{/*1e*/ 1, 0x5451, 0xa8a2, 0xfcf3, 0x9694, 0xc2c5, 0x3e36, 0x6a67,
+	   0xebe8, 0xbfb9, 0x434a, 0x171b, 0x7d7c, 0x292d, 0xd5de, 0x818f },
+	{/*1f*/ 1, 0x6fc1, 0xb542, 0xda83, 0x19f4, 0x7635, 0xacb6, 0xc377,
+	   0x2e58, 0x4199, 0x9b1a, 0xf4db, 0x37ac, 0x586d, 0x82ee, 0xed2f },
+
+	/* ECC bits are also in the set of tokens and they too can go bad
+	 * first 2 cover channel 0, while the second 2 cover channel 1
+	 */
+	{/*20*/ 0, 0xbe01, 0xd702, 0x6903, 0x2104, 0x9f05, 0xf606, 0x4807,
+	   0x3208, 0x8c09, 0xe50a, 0x5b0b, 0x130c, 0xad0d, 0xc40e, 0x7a0f },
+	{/*21*/ 0, 0x4101, 0x8202, 0xc303, 0x5804, 0x1905, 0xda06, 0x9b07,
+	   0xac08, 0xed09, 0x2e0a, 0x6f0b, 0x640c, 0xb50d, 0x760e, 0x370f },
+	{/*22*/ 1, 0xc441, 0x4882, 0x8cc3, 0xf654, 0x3215, 0xbed6, 0x7a97,
+	   0x5ba8, 0x9fe9, 0x132a, 0xd76b, 0xadfc, 0x69bd, 0xe57e, 0x213f },
+	{/*23*/ 1, 0x7621, 0x9b32, 0xed13, 0xda44, 0xac65, 0x4176, 0x3757,
+	   0x6f88, 0x19a9, 0xf4ba, 0x829b, 0xb5cc, 0xc3ed, 0x2efe, 0x58df }
+};
 
+/*
+ * get_channel_from_ecc_syndrome
+ *
+ *	given the syndrome argument, scan each of the channel tables
+ *	looking for a syndrome match. Depending on which table it is
+ *	found, return the channel number
+ */
+static int get_channel_from_ecc_syndrome(unsigned short syndrome)
+{
+	int row;
+	int column;
+
+	debugf0("%s()\n", __func__);
+
+	/* Determine column to scan */
+	column = syndrome & 0xF;
+
+	/* Scan all rows, looking for syndrome, or end of table */
+	for (row = 0; row < NUMBER_ECC_ROWS; row++) {
+		if (ecc_chipkill_syndromes[row][column] == syndrome)
+			return ecc_chipkill_syndromes[row][0];
+	}
+
+	debugf0("%s(): syndrome(%x) not found\n", __func__, syndrome);
+	return -1;
+}
-- 
1.6.2.4

[PATCH 16/21] amd64_edac: add error decoding logic

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  569 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 569 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 88849a4..fee9ec4 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -3260,3 +3260,572 @@ static int get_channel_from_ecc_syndrome(unsigned short syndrome)
 	debugf0("%s(): syndrome(%x) not found\n", __func__, syndrome);
 	return -1;
 }
+
+/*
+ * toString tables for the various error decoding values
+ */
+static const char *tt_msgs[] = {        /* transaction type */
+	"instruction",
+	"data",
+	"generic",
+	"reserved"
+};
+
+static const char *ll_msgs[] = {	/* cache level */
+	"L0",
+	"L1",
+	"L2",
+	"L3/generic"
+};
+
+static const char *rrrr_msgs[] = {
+	"generic",
+	"generic read",
+	"generic write",
+	"data read",
+	"data write",
+	"inst fetch",
+	"prefetch",
+	"evict",
+	"snoop",
+	"reserved RRRR= 9",
+	"reserved RRRR= 10",
+	"reserved RRRR= 11",
+	"reserved RRRR= 12",
+	"reserved RRRR= 13",
+	"reserved RRRR= 14",
+	"reserved RRRR= 15"
+};
+
+static const char *pp_msgs[] = {	/* participating processor */
+	"local node originated (SRC)",
+	"local node responded to request (RES)",
+	"local node observed as 3rd party (OBS)",
+	"generic"
+};
+
+static const char *to_msgs[] = {
+	"no timeout",
+	"timed out"
+};
+
+static const char *ii_msgs[] = {	/* memory or i/o */
+	"mem access",
+	"reserved",
+	"i/o access",
+	"generic"
+};
+
+/* Map the 5 bits of Extended Error code toString table
+ *	enhanced to support the new F10 error codes
+ *	esp, with L3 cache errors
+ */
+static const char *ext_msgs[] = {	/* extended error */
+	"K8 ECC error/F10 reserved",	/* 0_0000b */
+	"CRC error",			/* 0_0001b */
+	"sync error",			/* 0_0010b */
+	"mst abort",			/* 0_0011b */
+	"tgt abort",			/* 0_0100b */
+	"GART error",			/* 0_0101b */
+	"RMW error",			/* 0_0110b */
+	"Wdog timer error",		/* 0_0111b */
+	"F10-ECC/K8-Chipkill error",	/* 0_1000b */
+	"DEV Error",			/* 0_1001b */
+	"Link Data error",		/* 0_1010b */
+	"Link or L3 Protocol error",	/* 0_1011b */
+	"NB Array error",		/* 0_1100b */
+	"DRAM Parity error",		/* 0_1101b */
+	"Link Retry/GART Table Walk/DEV Table Walk error", /* 0_1110b */
+	"Res 0x0ff error",		/* 0_1111b */
+	"Res 0x100 error",		/* 1_0000b */
+	"Res 0x101 error",		/* 1_0001b */
+	"Res 0x102 error",		/* 1_0010b */
+	"Res 0x103 error",		/* 1_0011b */
+	"Res 0x104 error",		/* 1_0100b */
+	"Res 0x105 error",		/* 1_0101b */
+	"Res 0x106 error",		/* 1_0110b */
+	"Res 0x107 error",		/* 1_0111b */
+	"Res 0x108 error",		/* 1_1000b */
+	"Res 0x109 error",		/* 1_1001b */
+	"Res 0x10A error",		/* 1_1010b */
+	"Res 0x10B error",		/* 1_1011b */
+	"L3 Cache Data error",		/* 1_1100b */
+	"L3 CacheTag error",		/* 1_1101b */
+	"L3 Cache LRU error",		/* 1_1110b */
+	"Res 0x1FF error"		/* 1_1111b */
+};
+
+static const char *htlink_msgs[] = {
+	"none",
+	"1",
+	"2",
+	"1 2",
+	"3",
+	"1 3",
+	"2 3",
+	"1 2 3"
+};
+
+/*
+ * amd64_error_info_valid
+ *
+ * Return 1 if hardware contains valid error information.
+ * Else return 0, no valid error information.
+ */
+static inline int amd64_error_info_valid(struct amd64_error_info_regs *regs)
+{
+	return ((regs->nbsh & K8_NBSH_VALID_BIT) != 0);
+}
+
+/*
+ * amd64_get_error_info_regs
+ *
+ *	read the North Bridge Status register HIGH and test the
+ *	VALID ERROR status bit in that register.
+ *
+ *	If set, proceed to read:
+ *		NBS Low
+ *		NBEA LOW
+ *		NBEA HIGH
+ *
+ *	and store data into error structure
+ *
+ * return 1: if hardware regs contains valid error info
+ * return 0: if no valid error is indicated
+ */
+static int amd64_get_error_info_regs(struct mem_ctl_info *mci,
+				  struct amd64_error_info_regs *regs)
+{
+	struct amd64_pvt *pvt;
+	struct pci_dev *misc_f3_ctl;
+	int err;
+
+	pvt = mci->pvt_info;
+	misc_f3_ctl = pvt->misc_f3_ctl;
+
+	/* Read the NB Status Register High to get ERROR VALID bit */
+	err = pci_read_config_dword(misc_f3_ctl, K8_NBSH, &regs->nbsh);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBSH failed\n", __func__);
+
+	if (!amd64_error_info_valid(regs))
+		return 0;
+
+	/* valid error, read remaining error information registers */
+	err = pci_read_config_dword(misc_f3_ctl, K8_NBSL, &regs->nbsl);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBSL failed\n", __func__);
+	err = pci_read_config_dword(misc_f3_ctl, K8_NBEAL, &regs->nbeal);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBEAL failed\n", __func__);
+	err = pci_read_config_dword(misc_f3_ctl, K8_NBEAH, &regs->nbeah);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBEAH failed\n", __func__);
+	err = pci_read_config_dword(misc_f3_ctl, K8_NBCFG, &regs->nbcfg);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBCFG failed\n", __func__);
+
+	return 1;
+}
+
+/*
+ * amd64_get_error_info
+ *
+ *	this function is called to retrieve the error data from hardware
+ *	and store it in the info structure.
+ *
+ *	return:
+ *		1 if a valid error is found
+ *		0 if no error is found
+ */
+static int amd64_get_error_info(struct mem_ctl_info *mci,
+			      struct amd64_error_info_regs *info)
+{
+	struct amd64_pvt *pvt;
+	struct amd64_error_info_regs regs;
+
+	pvt = mci->pvt_info;
+
+	if (!amd64_get_error_info_regs(mci, info))
+		return 0;
+
+	/*
+	 * Here's the problem with the K8's EDAC reporting:
+	 * There are four registers which report pieces of error
+	 * information.  These four registers are shared between
+	 * CEs and UEs.  Furthermore, contrary to what is stated in
+	 * the OBKG, the overflow bit is never used!  Every error
+	 * always updates the reporting registers.
+	 *
+	 * Can you see the race condition?  All four error reporting
+	 * registers must be read before a new error updates them!
+	 * There is no way to read all four registers atomically.  The
+	 * best than can be done is to detect that a race has occured
+	 * and then report the error without any kind of precision.
+	 *
+	 * What is still positive is that errors are
+	 * still reported and thus problems can still be detected -
+	 * just not localized because the syndrome and address are
+	 * spread out across registers.
+	 *
+	 * Grrrrr!!!!!  Here's hoping that AMD fixes this in some
+	 * future K8 rev. UEs and CEs should have separate
+	 * register sets with proper overflow bits that are used!
+	 * At very least the problem can be fixed by honoring the
+	 * ErrValid bit in 'nbsh' and not updating registers - just
+	 * set the overflow bit - unless the current error is CE
+	 * and the new error is UE which would be the only situation
+	 * for overwriting the current values.
+	 */
+
+	regs = *info;
+
+	/* Use info from the second read - most current */
+	if (unlikely(!amd64_get_error_info_regs(mci, info)))
+		return 0;
+
+	/* clear the error bits in hardware */
+	pci_write_bits32(pvt->misc_f3_ctl, K8_NBSH, 0, K8_NBSH_VALID_BIT);
+
+	/* Check for the possible race condition */
+	if ((regs.nbsh != info->nbsh) ||
+	     (regs.nbsl != info->nbsl) ||
+	     (regs.nbeah != info->nbeah) ||
+	     (regs.nbeal != info->nbeal)) {
+		amd64_mc_printk(mci, KERN_WARNING,
+			"hardware STATUS read access race "
+			"condition detected!\n");
+		return 0;
+	}
+
+	return 1;
+}
+
+static inline void amd64_decode_gart_tlb_error(struct mem_ctl_info *mci,
+					 struct amd64_error_info_regs *info)
+{
+	u32 err_code;
+	u32 ec_tt;		/* error code transaction type (2b) */
+	u32 ec_ll;		/* error code cache level (2b) */
+
+	err_code = EXTRACT_ERROR_CODE(info->nbsl);
+	ec_ll = EXTRACT_LL_CODE(err_code);
+	ec_tt = EXTRACT_TT_CODE(err_code);
+
+	amd64_mc_printk(mci, KERN_ERR,
+		     "GART TLB event: transaction type(%s), "
+		     "cache level(%s)\n", tt_msgs[ec_tt], ll_msgs[ec_ll]);
+}
+
+static inline void amd64_decode_mem_cache_error(struct mem_ctl_info *mci,
+				      struct amd64_error_info_regs *info)
+{
+	u32 err_code;
+	u32 ec_rrrr;		/* error code memory transaction (4b) */
+	u32 ec_tt;		/* error code transaction type (2b) */
+	u32 ec_ll;		/* error code cache level (2b) */
+
+	err_code = EXTRACT_ERROR_CODE(info->nbsl);
+	ec_ll = EXTRACT_LL_CODE(err_code);
+	ec_tt = EXTRACT_TT_CODE(err_code);
+	ec_rrrr = EXTRACT_RRRR_CODE(err_code);
+
+	amd64_mc_printk(mci, KERN_ERR,
+		     "cache hierarchy error: memory transaction type(%s), "
+		     "transaction type(%s), cache level(%s)\n",
+		     rrrr_msgs[ec_rrrr], tt_msgs[ec_tt], ll_msgs[ec_ll]);
+}
+
+
+/*
+ * amd64_handle_ce
+ *
+ *	this routine is called to handle any Correctable Errors (CEs)
+ *	that have occurred. Check for valid ERROR ADDRESS and process
+ */
+static void amd64_handle_ce(struct mem_ctl_info *mci,
+				struct amd64_error_info_regs *info)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u64 SystemAddress;
+
+	/* Ensure that the Error Address is VALID */
+	if ((info->nbsh & K8_NBSH_VALID_ERROR_ADDR) == 0) {
+		amd64_mc_printk(mci, KERN_ERR,
+			"HW has no ERROR_ADDRESS available\n");
+		edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
+		return;
+	}
+
+	SystemAddress = extract_error_address(mci, info);
+
+	amd64_mc_printk(mci, KERN_ERR,
+		"CE ERROR_ADDRESS= 0x%llx\n", SystemAddress);
+
+	pvt->ops->map_sysaddr_to_csrow(mci, info, SystemAddress);
+}
+
+/*
+ * amd64_handle_ue
+ *
+ *	this routine is called to handle any Un-correctable Errors (UEs)
+ */
+static void amd64_handle_ue(struct mem_ctl_info *mci,
+				struct amd64_error_info_regs *info)
+{
+	int csrow;
+	u64 SystemAddress;
+	u32 page, offset;
+	struct mem_ctl_info *log_mci, *src_mci;
+
+	log_mci = mci;
+
+	/* Ensure that the Error Address is VALID */
+	if ((info->nbsh & K8_NBSH_VALID_ERROR_ADDR) == 0) {
+		amd64_mc_printk(mci, KERN_CRIT,
+			"HW has no ERROR_ADDRESS available\n");
+		edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR);
+		return;
+	}
+
+	SystemAddress = extract_error_address(mci, info);
+
+	/* Find out which node the error address belongs to.  This may be
+	 * different from the node that detected the error.
+	 */
+	src_mci = find_mc_by_sys_addr(mci, SystemAddress);
+	if (src_mci == NULL) {
+		amd64_mc_printk(mci, KERN_CRIT,
+			"ERROR ADDRESS (0x%lx) value NOT mapped to a MC\n",
+			(unsigned long)SystemAddress);
+		edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR);
+		return;
+	}
+
+	log_mci = src_mci;
+
+	csrow = sys_addr_to_csrow(log_mci, SystemAddress);
+	if (csrow < 0) {
+		amd64_mc_printk(mci, KERN_CRIT,
+			"ERROR_ADDRESS (0x%lx) value NOT mapped to 'csrow'\n",
+			(unsigned long)SystemAddress);
+		edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR);
+	} else {
+		error_address_to_page_and_offset(SystemAddress, &page, &offset);
+		edac_mc_handle_ue(log_mci, page, offset, csrow, EDAC_MOD_STR);
+	}
+}
+
+/*
+ * amd64_decode_bus_error
+ *
+ *	Perform decode of the AMD64 "BUS" error code
+ */
+static void amd64_decode_bus_error(struct mem_ctl_info *mci,
+			     struct amd64_error_info_regs *info)
+{
+	u32 err_code, ext_ec;
+	u32 ec_pp;		/* error code participating processor (2p) */
+	u32 ec_to;		/* error code timed out (1b) */
+	u32 ec_rrrr;		/* error code memory transaction (4b) */
+	u32 ec_ii;		/* error code memory or I/O (2b) */
+	u32 ec_ll;		/* error code cache level (2b) */
+
+	debugf0("MC%d: %s()\n", mci->mc_idx, __func__);
+
+	ext_ec = EXTRACT_EXT_ERROR_CODE(info->nbsl);
+	err_code = EXTRACT_ERROR_CODE(info->nbsl);
+
+	ec_ll = EXTRACT_LL_CODE(err_code);
+	ec_ii = EXTRACT_II_CODE(err_code);
+	ec_rrrr = EXTRACT_RRRR_CODE(err_code);
+	ec_to = EXTRACT_TO_CODE(err_code);
+	ec_pp = EXTRACT_PP_CODE(err_code);
+
+	/* Log AMD Specific information */
+	amd64_mc_printk(mci, KERN_ERR,
+		"BUS ERROR:\n"
+		"  time-out(%s) mem or i/o(%s)\n"
+		"  participating processor(%s)\n"
+		"  memory transaction type(%s)\n"
+		"  cache level(%s) Error Found by: %s\n",
+		to_msgs[ec_to],
+		ii_msgs[ec_ii],
+		pp_msgs[ec_pp],
+		rrrr_msgs[ec_rrrr],
+		ll_msgs[ec_ll],
+		(info->nbsh & K8_NBSH_ERR_SCRUBER) ?
+			"Scrubber" : "Normal Operation");
+
+	/* If this was an 'observed' error, early out */
+	if (ec_pp == K8_NBSL_PP_OBS)
+		return;		/* We aren't the node involved */
+
+	/* Parse out the extended error code for ECC events */
+	switch (ext_ec) {
+	/* F10 changed to one Extended ECC error code */
+	case F10_NBSL_EXT_ERR_RES:		/* Reserved field */
+	case F10_NBSL_EXT_ERR_ECC:		/* F10 ECC ext err code */
+		break;
+
+	default:
+		amd64_mc_printk(mci, KERN_ERR,
+			"NOT ECC: no special error handling for this error\n");
+		return;
+	}
+
+	/* CE or UE event? */
+	if (info->nbsh & K8_NBSH_CECC)
+		amd64_handle_ce(mci, info);
+	else if (info->nbsh & K8_NBSH_UECC)
+		amd64_handle_ue(mci, info);
+
+	/* If main error is CE then overflow must be CE.  If main error is UE
+	 * then overflow is unknown.  We'll call the overflow a CE - if
+	 * panic_on_ue is set then we're already panic'ed and won't arrive
+	 * here.  If panic_on_ue is not set then apparently someone doesn't
+	 * think that UE's are catastrophic.
+	 */
+	if (info->nbsh & K8_NBSH_OVERFLOW)
+		edac_mc_handle_ce_no_info(mci,
+			EDAC_MOD_STR " Error Overflow set");
+}
+
+/*
+ * amd64_process_error_info
+ *
+ *	process the error information of the AMD64 chipset
+ *
+ *	return:
+ *		1 if error found or
+ *		0 if all ok
+ */
+static int amd64_process_error_info(struct mem_ctl_info *mci,
+					struct amd64_error_info_regs *info,
+					int handle_errors)
+{
+	struct amd64_pvt *pvt;
+	struct amd64_error_info_regs *regs;
+	u32 err_code, ext_ec;
+	int gart_tlb_error = 0;
+
+	pvt = mci->pvt_info;
+
+	/* If caller doesn't want us to process the error, return */
+	if (!handle_errors)
+		return 1;
+
+	regs = info;
+
+	debugf1("NorthBridge ERROR: mci(0x%p)\n", mci);
+	debugf1("  MC node(%d) Error-Address(0x%.8x-%.8x)\n",
+		pvt->mc_node_id, regs->nbeah, regs->nbeal);
+	debugf1("  nbsh(0x%.8x) nbsl(0x%.8x)\n",
+		regs->nbsh, regs->nbsl);
+	debugf1("  Valid Error=%s Overflow=%s\n",
+		(regs->nbsh & K8_NBSH_VALID_BIT) ? "True" : "False",
+		(regs->nbsh & K8_NBSH_OVERFLOW) ? "True" : "False");
+	debugf1("  Err Uncorrected=%s MCA Error Reporting=%s\n",
+		(regs->nbsh & K8_NBSH_UNCORRECTED_ERR) ?
+			"True" : "False",
+		(regs->nbsh & K8_NBSH_ERR_ENABLE) ?
+			"True" : "False");
+	debugf1("  MiscErr Valid=%s ErrAddr Valid=%s PCC=%s\n",
+		(regs->nbsh & K8_NBSH_MISC_ERR_VALID) ?
+			"True" : "False",
+		(regs->nbsh & K8_NBSH_VALID_ERROR_ADDR) ?
+			"True" : "False",
+		(regs->nbsh & K8_NBSH_PCC) ?
+			"True" : "False");
+	debugf1("  CECC=%s UECC=%s Found by Scruber=%s\n",
+		(regs->nbsh & K8_NBSH_CECC) ?
+			"True" : "False",
+		(regs->nbsh & K8_NBSH_UECC) ?
+			"True" : "False",
+		(regs->nbsh & K8_NBSH_ERR_SCRUBER) ?
+			"True" : "False");
+	debugf1("  CORE0=%s CORE1=%s CORE2=%s CORE3=%s\n",
+		(regs->nbsh & K8_NBSH_CORE0) ? "True" : "False",
+		(regs->nbsh & K8_NBSH_CORE1) ? "True" : "False",
+		(regs->nbsh & K8_NBSH_CORE2) ? "True" : "False",
+		(regs->nbsh & K8_NBSH_CORE3) ? "True" : "False");
+
+
+	err_code = EXTRACT_ERROR_CODE(regs->nbsl);
+
+	/* Determine which error type:
+	 *	1) GART errors - non-fatal, developmental events
+	 *	2) MEMORY errors
+	 *	3) BUS errors
+	 *	4) Unknown error
+	 */
+	if (TEST_TLB_ERROR(err_code)) {
+		/*
+		 * GART errors are intended to help graphics driver
+		 * developers to detect bad GART PTEs. It is recommended by
+		 * AMD to disable GART table walk error reporting by default[1]
+		 * (currently being disabled in mce_cpu_quirks()) and according
+		 * to the comment in mce_cpu_quirks(), such GART errors can be
+		 * incorrectly triggered. We may see these errors anyway and
+		 * unless requested by the user, they won't be reported.
+		 *
+		 * [1] section 13.10.1 on BIOS and Kernel Developers Guide for
+		 *     AMD NPT family 0Fh processors
+		 */
+		if (report_gart_errors == 0)
+			return 1;
+
+		/* Only if GART error reporting are requested should
+		 * we generate any logs. If not requested, we left above.
+		 */
+		gart_tlb_error = 1;
+
+		debugf1("GART TLB error\n");
+		amd64_decode_gart_tlb_error(mci, info);
+	} else if (TEST_MEM_ERROR(err_code)) {
+		debugf1("Memory/Cache error\n");
+		amd64_decode_mem_cache_error(mci, info);
+	} else if (TEST_BUS_ERROR(err_code)) {
+		debugf1("Bus (Link/DRAM) error\n");
+		amd64_decode_bus_error(mci, info);
+	} else {
+		/* shouldn't reach here! */
+		amd64_mc_printk(mci, KERN_WARNING,
+			     "%s(): unknown MCE error 0x%x\n", __func__,
+			     err_code);
+	}
+
+	ext_ec = EXTRACT_EXT_ERROR_CODE(regs->nbsl);
+	amd64_mc_printk(mci, KERN_ERR,
+		"ExtErr=(0x%x) %s\n", ext_ec, ext_msgs[ext_ec]);
+
+	if (((ext_ec >= F10_NBSL_EXT_ERR_CRC &&
+			ext_ec <= F10_NBSL_EXT_ERR_TGT) ||
+			(ext_ec == F10_NBSL_EXT_ERR_RMW)) &&
+			EXTRACT_LDT_LINK(info->nbsh)) {
+
+		amd64_mc_printk(mci, KERN_ERR,
+			"Error on hypertransport link: %s\n",
+			htlink_msgs[
+			EXTRACT_LDT_LINK(info->nbsh)]);
+	}
+
+	/* Check the UE bit of the NB status high register, if set
+	 * Generate some logs.
+	 * If NOT a GART error, then process the event as a NO-INFO event.
+	 * If it was a GART error, skip that process.
+	 */
+	if (regs->nbsh & K8_NBSH_UNCORRECTED_ERR) {
+		amd64_mc_printk(mci, KERN_CRIT, "uncorrected error\n");
+		if (!gart_tlb_error)
+			edac_mc_handle_ue_no_info(mci, "UE bit is set\n");
+	}
+
+	if (regs->nbsh & K8_NBSH_PCC)
+		amd64_mc_printk(mci, KERN_CRIT,
+			"PCC (processor context corrupt) set\n");
+
+	return 1;
+}
+
+
-- 
1.6.2.4

[PATCH 17/21] amd64_edac: add EDAC core-related initializers

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  336 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 336 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index fee9ec4..8cf8060 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -3828,4 +3828,340 @@ static int amd64_process_error_info(struct mem_ctl_info *mci,
 	return 1;
 }
 
+/*
+ * amd64_check
+ *
+ * The main polling 'check' function, called FROM the edac core to perform the
+ * error checking and if an error is encountered, error processing.
+ */
+static void amd64_check(struct mem_ctl_info *mci)
+{
+	struct amd64_error_info_regs info;
+
+	debugf3("%s()\n", __func__);
+	if (amd64_get_error_info(mci, &info))
+		amd64_process_error_info(mci, &info, 1);
+}
+
+/*
+ * amd64_reserve_mc_sibling_devices
+ *
+ * Input:
+ *	1) struct amd64_pvt which contains pvt->dram_f2_ctl pointer
+ *	2) AMD Family index value
+ *
+ * Ouput:
+ *	Upon return of 0, the following filled in:
+ *
+ *		struct pvt->addr_f1_ctl
+ *		struct pvt->misc_f3_ctl
+ *
+ *	Filled in with related device funcitions of 'dram_f2_ctl'
+ *	These devices are "reserved" via the pci_get_device()
+ *
+ *	Upon return of 1 (error status):
+ *
+ *		Nothing reserved
+ */
+static int amd64_reserve_mc_sibling_devices(struct amd64_pvt *pvt,
+					 int mc_type_index)
+{
+	const struct amd64_family_type
+		*amd64_dev = &amd64_family_types[mc_type_index];
+
+	/* Reserve the ADDRESS MAP Device */
+	pvt->addr_f1_ctl = pci_get_related_function(pvt->dram_f2_ctl->vendor,
+						amd64_dev->addr_f1_ctl,
+						pvt->dram_f2_ctl);
+	if (pvt->addr_f1_ctl == NULL) {
+		amd64_printk(KERN_ERR, "error address map device not found: "
+			  "vendor %x device 0x%x (broken BIOS?)\n",
+			  PCI_VENDOR_ID_AMD, amd64_dev->addr_f1_ctl);
+		return 1;
+	}
+
+	/* Reserve the MISC Device */
+	pvt->misc_f3_ctl = pci_get_related_function(pvt->dram_f2_ctl->vendor,
+						amd64_dev->misc_f3_ctl,
+						pvt->dram_f2_ctl);
+	if (pvt->misc_f3_ctl == NULL) {
+		pci_dev_put(pvt->addr_f1_ctl);
+		pvt->addr_f1_ctl = NULL;
+
+		amd64_printk(KERN_ERR, "error miscellaneous device not found: "
+			  "vendor %x device 0x%x (broken BIOS?)\n",
+			  PCI_VENDOR_ID_AMD, amd64_dev->misc_f3_ctl);
+		return 1;
+	}
+
+	debugf1("    Addr Map device PCI Bus ID:\t%s\n",
+		pci_name(pvt->addr_f1_ctl));
+	debugf1("    DRAM MEM-CTL PCI Bus ID:\t%s\n",
+		pci_name(pvt->dram_f2_ctl));
+	debugf1("    Misc device PCI Bus ID:\t\t%s\n",
+		pci_name(pvt->misc_f3_ctl));
+
+	return 0;
+}
+
+/*
+ * amd64_read_mc_registers
+ *
+ * Retrieve the hardware registers of the memory controller (this includes the
+ * 'Address Map' and 'Misc' device regs)
+ */
+static void amd64_read_mc_registers(struct amd64_pvt *pvt)
+{
+	u64 msr_val;
+	int dram;
+	int err;
+
+	debugf0("%s(MC node-id=%d): (ExtModel=%d)\n",
+		__func__, pvt->mc_node_id, pvt->ext_model);
+
+	/*
+	 * Retrieve TOP_MEM and TOP_MEM2; no masking off of reserved
+	 * bits since those are Read-As-Zero
+	 */
+	rdmsrl(MSR_K8_TOP_MEM1, msr_val);
+	pvt->top_mem = msr_val >> 23;
+	debugf0("  TOP_MEM=0x%08llx\n", pvt->top_mem);
+
+	/* check first whether TOP_MEM2 is enabled */
+	rdmsrl(MSR_K8_SYSCFG, msr_val);
+	if (msr_val & (1U << 21)) {
+		rdmsrl(MSR_K8_TOP_MEM2, msr_val);
+		pvt->top_mem2 = msr_val >> 23;
+		debugf0("  TOP_MEM2=0x%08llx\n", pvt->top_mem2);
+	} else
+		debugf0("  TOP_MEM2 disabled.\n");
+
+	amd64_cpu_display_info(pvt);
+
+	err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCAP, &pvt->nbcap);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBCAP failed\n", __func__);
+
+	if (pvt->ops->read_dram_ctl_register)
+		pvt->ops->read_dram_ctl_register(pvt);
+
+	for (dram = 0; dram < DRAM_REG_COUNT; dram++) {
+		/*
+		 * Call CPU specific READ function to get the DRAM Base and
+		 * Limit values from the DCT.
+		 */
+		pvt->ops->read_dram_base_limit(pvt, dram);
+
+		/* Only print out debug info on rows with both R and W
+		 * Enabled.  Normal processing, compiler should optimize
+		 * this whole 'if' debug output block away
+		 */
+		if (pvt->dram_rw_en[dram] != 0) {
+			debugf1("  DRAM_BASE[%d]: 0x%8.08x-%8.08x "
+				"DRAM_LIMIT:  0x%8.08x-%8.08x\n",
+				dram,
+				(u32)(pvt->dram_base[dram] >> 32),
+				(u32)(pvt->dram_base[dram] & 0xFFFFFFFF),
+				(u32)(pvt->dram_limit[dram] >> 32),
+				(u32)(pvt->dram_limit[dram] & 0xFFFFFFFF));
+			debugf1("        IntlvEn=%s %s %s "
+				"IntlvSel=%d DstNode=%d\n",
+				pvt->dram_IntlvEn[dram] ?
+					"Enabled" : "Disabled",
+				(pvt->dram_rw_en[dram] & 0x2) ? "W" : "!W",
+				(pvt->dram_rw_en[dram] & 0x1) ? "R" : "!R",
+				pvt->dram_IntlvSel[dram],
+				pvt->dram_DstNode[dram]);
+		}
+	}
+
+	amd64_read_dct_base_mask(pvt);
+
+	err = pci_read_config_dword(pvt->addr_f1_ctl, K8_DHAR, &pvt->dhar);
+	if (err != 0)
+		debugf0("%s() Reading K8_DHAR failed\n", __func__);
+
+	amd64_read_dbam_reg(pvt);
+
+	err = pci_read_config_dword(pvt->misc_f3_ctl,
+				F10_ONLINE_SPARE, &pvt->online_spare);
+	if (err != 0)
+		debugf0("%s() Reading ONLINE_SPARE failed\n", __func__);
+
+	err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0);
+	if (err != 0)
+		debugf0("%s() Reading F10_DCLR_0 failed\n", __func__);
+	err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCHR_0, &pvt->dchr0);
+	if (err != 0)
+		debugf0("%s() Reading F10_DCHR_0 failed\n", __func__);
+
+	if (!dct_ganging_enabled(pvt)) {
+		err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_1,
+						&pvt->dclr1);
+		if (err != 0)
+			debugf0("%s() Reading F10_DCLR_1 failed\n", __func__);
+		err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCHR_1,
+						&pvt->dchr1);
+		if (err != 0)
+			debugf0("%s() Reading F10_DCHR_1 failed\n", __func__);
+	}
+
+	amd64_dump_misc_regs(pvt);
+}
+
+/*
+ * amd64_csrow_nr_pages
+ *
+ *	NOTE: CPU Revision Dependent code
+ *
+ *	Input:
+ *		csrow_nr  ChipSelect Row Number (0..CHIPSELECT_COUNT-1)
+ *		k8 private pointer to -->
+ *			DRAM Bank Address mapping register
+ *			node_id
+ *			DCL register where dual_channel_active is
+ *
+ *	The DBAM register consists of 4 sets of 4 bits each definitions:
+ *
+ *	Bits:		CSROWs
+ *	0-3		CSROWs 0 and 1
+ *	4-7		CSROWs 2 and 3
+ *	8-11		CSROWs 4 and 5
+ *	12-15		CSROWs 6 and 7
+ *
+ *	Values range from: 0 to 15
+ *	The meanings of the values depends on CPU REV and dual-channel state
+ *
+ *	Various CPU revisions have different size definitions for this
+ *	4 bit values.
+ *	Therefore, we need to examine which CPU we are running on to
+ *	extract meaning for these bits:
+ *
+ *	REV CG and earlier have a given DIMM size definition
+ *	REV D & REV E share another set of size definitions
+ *
+ *	REV CG, D and E all have a common beginning set of size definitions,
+ *	but change the meanings midway into the table.
+ *
+ *	REV F has yet another set of size definitions, which has a totally
+ *	different starting point.
+ *
+ *	REV ? future revs? We won't know till we get the specs
+ *
+ *	The memory controller provides for total of only 8 CSROWs in its
+ *	current architecture. Each "pair" of CSROWs normally represents
+ *	just one (1) *	DIMM in single channel or just two (2) DIMMs
+ *	in dual channel mode.
+ *
+ *	The following code logic collapses the various tables for CSROW
+ *	based on CPU Rev number.
+ *	See the tables/algorithms in the respective BKDG manuals.
+ *
+ *	return:
+ *		The number of PAGE_SIZE pages on the specified CSROW number
+ *		it encompasses
+ *
+ */
+static u32 amd64_csrow_nr_pages(int csrow_nr, struct amd64_pvt *pvt)
+{
+	u32 dram_map;
+	u32 nr_pages;
+
+	/* The math on this doesn't look right on the surface because x/2*4
+	 * can be simplified to x*2 but this expression makes use of the fact
+	 * that it is integral math where 1/2=0. This intermediate value
+	 * becomes the number of bits to shift the DBAM register to extract
+	 * the proper CSROW field.
+	 */
+	dram_map = (pvt->dbam0 >> ((csrow_nr / 2) * 4)) & 0xF;	/* PG88 */
+
+	nr_pages = pvt->ops->dbam_map_to_pages(pvt, dram_map);
+
+	/* If dual channel then double the memory size of single channel */
+	/* channel count is 1 or 2 */
+	nr_pages <<= (pvt->channel_count - 1);
+
+	debugf0("  %s(csrow=%d) DBAM map index= %d\n", __func__,
+		csrow_nr, dram_map);
+	debugf0("    nr_pages= %u  channel-count = %d\n",
+		nr_pages, pvt->channel_count);
+
+	return nr_pages;
+}
+
+/*
+ * amd64_init_csrows
+ *
+ *	perform initialization on the array of csrow attribute instances,
+ *	based on the values from pci config hardware registers
+ */
+static int amd64_init_csrows(struct mem_ctl_info *mci)
+{
+	struct csrow_info *csrow;
+	struct amd64_pvt *pvt;
+	int i;
+	int empty = 1;
+	u64 input_addr_min, input_addr_max, sys_addr;
+	int err;
+
+	pvt = mci->pvt_info;
+
+	err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &pvt->nbcfg);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBCFG failed\n", __func__);
+
+	debugf0("%s() NBCFG= 0x%x  CHIPKILL= %s DRAM ECC= %s\n",
+		__func__, pvt->nbcfg,
+		(pvt->nbcfg & K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled",
+		(pvt->nbcfg & K8_NBCFG_ECC_ENABLE) ? "Enabled" : "Disabled"
+		);
+
+	for (i = 0; i < CHIPSELECT_COUNT; i++) {
+		csrow = &mci->csrows[i];
+
+		if ((pvt->dcsb0[i] & K8_DCSB_CS_ENABLE) == 0) {
+			debugf1("----CSROW %d EMPTY for node %d\n", i,
+				pvt->mc_node_id);
+			continue;
+		}
+
+		debugf1("----CSROW %d VALID for MC node %d\n",
+			i, pvt->mc_node_id);
+
+		empty = 0;
+		csrow->nr_pages = amd64_csrow_nr_pages(i, pvt);
+		find_csrow_limits(mci, i, &input_addr_min, &input_addr_max);
+		sys_addr = input_addr_to_sys_addr(mci, input_addr_min);
+		csrow->first_page = (u32) (sys_addr >> PAGE_SHIFT);
+		sys_addr = input_addr_to_sys_addr(mci, input_addr_max);
+		csrow->last_page = (u32) (sys_addr >> PAGE_SHIFT);
+		csrow->page_mask = ~mask_from_dct_mask(pvt, i);
+		/* 8 bytes of resolution */
+
+		csrow->mtype = amd64_determine_memory_type(pvt);
+
+		debugf1("  for MC node %d csrow %d:\n", pvt->mc_node_id, i);
+		debugf1("    input_addr_min: 0x%lx input_addr_max: 0x%lx\n",
+			(unsigned long)input_addr_min,
+			(unsigned long)input_addr_max);
+		debugf1("    sys_addr: 0x%lx  page_mask: 0x%lx\n",
+			(unsigned long)sys_addr, csrow->page_mask);
+		debugf1("    nr_pages: %u  first_page: 0x%lx "
+			"last_page: 0x%lx\n",
+			(unsigned)csrow->nr_pages,
+			csrow->first_page, csrow->last_page);
+
+		/*
+		 * determine whether CHIPKILL or JUST ECC or NO ECC is operating
+		 */
+		if (pvt->nbcfg & K8_NBCFG_ECC_ENABLE)
+			csrow->edac_mode =
+			    (pvt->nbcfg & K8_NBCFG_CHIPKILL) ?
+			    EDAC_S4ECD4ED : EDAC_SECDED;
+		else
+			csrow->edac_mode = EDAC_NONE;
+	}
+
+	return empty;
+}
 
-- 
1.6.2.4

[PATCH 18/21] amd64_edac: add ECC reporting initializers

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  242 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 242 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 8cf8060..43f236d 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -70,6 +70,7 @@
 #include <linux/slab.h>
 #include <linux/mmzone.h>
 #include <linux/edac.h>
+#include <asm/msr.h>
 #include "edac_core.h"
 
 #define amd64_printk(level, fmt, arg...) \
@@ -4165,3 +4166,244 @@ static int amd64_init_csrows(struct mem_ctl_info *mci)
 	return empty;
 }
 
+/*
+ * amd64_enable_ecc_error_reporting
+ *
+ *	Only if 'ecc_enable_override' is set AND BIOS had ECC disabled,
+ *	do "we" enable it.
+ *
+ *	On each NB we need to enable the hardware to
+ *	generate and detect error events
+ *
+ *	1) NB Control Register
+ *	2) Global MCE Reporting Control Reg (MCGCTL)
+ */
+static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	const cpumask_t *cpumask = cpumask_of_node(pvt->mc_node_id);
+	int idx = 0, cpu, err;
+	int cpus_on_node = cpumask_weight(cpumask);
+	u32 mcgctl_l[cpus_on_node], mcgctl_h[cpus_on_node];
+	u32 value;
+	u32 mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn;
+
+	if (!ecc_enable_override)
+		return;
+
+	amd64_printk(KERN_WARNING,
+		"'ecc_enable_override' parameter is active, "
+		"Enabling AMD ECC hardware now: CAUTION\n");
+
+	/* 1) read the NB Control register, and save old Enable bits */
+	err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCTL, &value);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBCTL failed\n", __func__);
+
+	/* save old value and then turn on UECCn and CECCEn bits
+	 * and write it back out, thus turning ON ECC for sure
+	 */
+	pvt->old_nbctl = value & mask;
+	pvt->nbctl_mcgctl_saved = 1;	/* Mark 'old' ECC values valid */
+
+	value |= mask;
+	pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCTL, value);
+
+	debugf0("%s() Old NBCTL 0x%x New NBCTL= 0x%x\n",
+		__func__, pvt->old_nbctl, value);
+
+	/* 2) Read and save the NB Enable bit at entry. Enable the bit
+	 * then write the enabled value back to hardware
+	 */
+	rdmsr_on_cpus(cpumask, K8_MSR_MCGCTL, mcgctl_l, mcgctl_h);
+
+	for_each_cpu(cpu, cpumask) {
+		pvt->old_mcgctl[idx] = mcgctl_l[idx] & K8_MSR_MCGCTL_NBE;
+		mcgctl_l[idx] |= K8_MSR_MCGCTL_NBE;
+
+		debugf0("%s(), cpu %d, Old MCGCTL[NBE] = 0x%x New MCGCTL=0x%x\n",
+			__func__, cpu, (unsigned int) pvt->old_mcgctl[idx],
+			(unsigned int) mcgctl_l[idx]);
+
+		idx++;
+	}
+	wrmsr_on_cpus(cpumask, K8_MSR_MCGCTL, mcgctl_l, mcgctl_h);
+
+	/* 3) Read the NB CFG to ensure DRAM ECC is on and then
+	 * keep a copy of the hw register in the control structure
+	 */
+	err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBCFG failed\n", __func__);
+
+	debugf0("%s() NBCFG(1)= 0x%x  CHIPKILL= %s ECC_ENABLE= %s\n",
+		__func__, value,
+		value & (K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled",
+		value & (K8_NBCFG_ECC_ENABLE) ? "Enabled" : "Disabled"
+		);
+
+	if (!(value & K8_NBCFG_ECC_ENABLE)) {
+		amd64_printk(KERN_WARNING,
+			"This node reports that DRAM ECC is "
+			"currently Disabled; ENABLING now\n");
+
+		/* Attempt to turn on DRAM ECC Enable */
+		value |= K8_NBCFG_ECC_ENABLE;
+		pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCFG, value);
+
+		err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value);
+		if (err != 0)
+			debugf0("%s() Reading K8_NBCFG failed\n", __func__);
+
+		if (!(value & K8_NBCFG_ECC_ENABLE)) {
+			amd64_printk(KERN_WARNING,
+				"Hardware rejects Enabling DRAM ECC checking\n"
+				"Check memory DIMM configuration\n");
+		} else {
+			amd64_printk(KERN_DEBUG,
+				"Hardware accepted DRAM ECC Enable\n");
+		}
+	}
+	debugf0("%s() NBCFG(2)= 0x%x  CHIPKILL= %s ECC_ENABLE= %s\n",
+		__func__, value,
+		(value & K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled",
+		(value & K8_NBCFG_ECC_ENABLE) ? "Enabled" : "Disabled"
+		);
+
+	pvt->ctl_error_info.nbcfg = value;
+}
+
+/*
+ * amd64_restore_ecc_error_reporting
+ *
+ *	restore the hardware registers to their initial condition
+ *	prior to when amd64_enable_ecc_error_reporting was called
+ */
+static void amd64_restore_ecc_error_reporting(struct amd64_pvt *pvt)
+{
+	const cpumask_t *cpumask = cpumask_of_node(pvt->mc_node_id);
+	int cpus_on_node = cpumask_weight(cpumask), idx = 0, cpu;
+	u32 mcgctl_l[cpus_on_node], mcgctl_h[cpus_on_node];
+	u32 value;
+	u32 mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn;
+	int err;
+
+	if (!pvt->nbctl_mcgctl_saved)
+		return;
+
+	err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCTL, &value);
+	if (err != 0)
+		debugf0("%s() Reading K8_NBCTL failed\n", __func__);
+	value &= ~mask;
+	value |= pvt->old_nbctl;
+
+	/* restore the NB Enable MCGCTL bit */
+	pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCTL, value);
+
+	rdmsr_on_cpus(cpumask, K8_MSR_MCGCTL, mcgctl_l, mcgctl_h);
+
+	for_each_cpu(cpu, cpumask) {
+		mcgctl_l[idx] &= ~K8_MSR_MCGCTL_NBE;
+		mcgctl_l[idx] |= pvt->old_mcgctl[idx];
+		idx++;
+	}
+
+	wrmsr_on_cpus(cpumask, K8_MSR_MCGCTL, mcgctl_l, mcgctl_h);
+}
+
+static void check_mcg_ctl(void *ret)
+{
+	u64 msr_val = 0;
+	u8 nbe;
+
+	rdmsrl(MSR_IA32_MCG_CTL, msr_val);
+	nbe = msr_val & K8_MSR_MCGCTL_NBE;
+
+	debugf0("%s: core: %u, MCG_CTL: 0x%llx, NB MSR is %s\n",
+		__func__, raw_smp_processor_id(), msr_val,
+		(nbe ? "enabled" : "disabled"));
+
+	if (!nbe)
+		*(int *)ret = 0;
+}
+
+static int amd64_mcg_ctl_enabled_on_cpus(const cpumask_t *mask)
+{
+	int rc = 1;
+	preempt_disable();
+	smp_call_function_many(mask, check_mcg_ctl, &rc, 1);
+	preempt_enable();
+
+	return rc;
+}
+
+/*
+ * amd64_check_ecc_enabled
+ *
+ *	EDAC requires that the BIOS have ECC enabled before taking over the
+ *	processing of ECC errors. This is because the BIOS can properly
+ *	initialize the memory system completely.
+ *
+ *	For development and other purposes, there is a command line option
+ *	which allows for overriding this contraint. If supplied on the kernrel
+ *	command line, hardware ECC is force-enabled later in
+ *	amd64_enable_ecc_error_reporting().
+ */
+static int amd64_check_ecc_enabled(struct amd64_pvt *pvt)
+{
+	u32 value;
+	int tmp;
+	int rc = 0;
+
+	tmp = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value);
+	if (tmp != 0)
+		debugf0("%s() Reading K8_NBCTL failed\n", __func__);
+
+	/* check MCG_CTL on all the cpus on this node */
+	rc = amd64_mcg_ctl_enabled_on_cpus(cpumask_of_node(pvt->mc_node_id));
+
+	debugf0("%s() K8_NBCFG=0x%x,  DRAM ECC is %s\n",
+		__func__, value, (value & K8_NBCFG_ECC_ENABLE ? "enabled"
+					: "disabled"));
+	if (!tmp || !rc) {
+		if (!tmp) {
+			amd64_printk(KERN_WARNING, "This node reports that "
+						   "Memory ECC is currently "
+						   "disabled.\n");
+
+			amd64_printk(KERN_WARNING, "bit 0x%lx in register "
+				"F3x%x of the MISC_CONTROL device (%s) "
+				"should be enabled\n", K8_NBCFG_ECC_ENABLE,
+				K8_NBCFG, pci_name(pvt->misc_f3_ctl));
+		}
+		if (!rc) {
+			amd64_printk(KERN_WARNING, "bit 0x%016lx in MSR 0x%08x "
+					"of node %d should be enabled\n",
+					K8_MSR_MCGCTL_NBE, MSR_IA32_MCG_CTL,
+					pvt->mc_node_id);
+		}
+		if (!ecc_enable_override) {
+			amd64_printk(KERN_WARNING, "WARNING: ECC is NOT "
+				"currently enabled by the BIOS. Module "
+				"will NOT be loaded.\n"
+				"    Either Enable ECC in the BIOS, "
+				"or use the 'ecc_enable_override' "
+				"parameter.\n"
+				"    Might be a BIOS bug, if BIOS says "
+				"ECC is enabled\n"
+				"    Use of the override can cause "
+				"unknown side effects.\n");
+			rc = -ENODEV;
+		}
+	} else {
+		amd64_printk(KERN_INFO,
+			"ECC is enabled by BIOS, Proceeding "
+			"with EDAC module initialization\n");
+
+		/* CLEAR the override, since BIOS controlled it */
+		ecc_enable_override = 0;
+	}
+
+	return rc;
+}
+
-- 
1.6.2.4

[PATCH 19/21] amd64_edac: add debugging/testing code

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

This is for dumping different registers and testing the address mapping
logic using the ECC syndromes.

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  214 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 214 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 43f236d..b1a7e8c 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -4407,3 +4407,217 @@ static int amd64_check_ecc_enabled(struct amd64_pvt *pvt)
 	return rc;
 }
 
+#ifdef	CONFIG_EDAC_DEBUG
+/*
+ * amd64_nbea_store
+ *
+ * 	Accept a hex value and store it into the virutal error
+ *	register file, field:	nbeal and nbeah
+ *
+ *	Assume virtual error values have already been set for:
+ *		NBSL, NBSH and NBCFG
+ *
+ *	Then proceed to map the error values to a:
+ *		MC, CSROW and CHANNEL
+ */
+static ssize_t amd64_nbea_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long long value;
+	int rc;
+
+	rc = strict_strtoull(data, 16, &value);
+	if (rc != -EINVAL) {
+		debugf0("%s() received NBEA= 0x%llx\n", __func__, value);
+
+		/* place the value into the virtual error packet */
+		pvt->ctl_error_info.nbeal = (u32) value;
+		value >>= 32;
+		pvt->ctl_error_info.nbeah = (u32) value;
+
+		/* Process the Mapping request */
+		/* TODO: Add race preventation */
+		amd64_process_error_info(mci, &pvt->ctl_error_info, 1);
+
+		return count;
+	}
+	return rc;
+}
+
+/*
+ * amd64_nbea_show
+ *
+ *	display back what the last NBEA address was written
+ */
+static ssize_t amd64_nbea_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u64 value;
+
+	value = pvt->ctl_error_info.nbeah;
+	value <<= 32;
+	value |= pvt->ctl_error_info.nbeal;
+
+	return sprintf(data, "%llx\n", value);
+}
+
+/*
+ * amd64_nbsl_store
+ *
+ *	accept and store the NBSL value user desires
+ */
+static ssize_t amd64_nbsl_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	int rc;
+
+	rc = strict_strtoul(data, 16, &value);
+	if (rc != -EINVAL) {
+		debugf0("%s() received NBSL= 0x%lx\n", __func__, value);
+
+		/* place the NBSL value into the virtual error packet */
+		pvt->ctl_error_info.nbsl = (u32) value;
+
+		return count;
+	}
+	return rc;
+}
+
+/*
+ * amd64_nbsl_show
+ *
+ *	display back what the last NBSL value written
+ */
+static ssize_t amd64_nbsl_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u32 value;
+
+	value = pvt->ctl_error_info.nbsl;
+
+	return sprintf(data, "%x\n", value);
+}
+
+/*
+ * amd64_nbsh_store
+ *
+ *	accept and store the NBSH value user desires
+ */
+static ssize_t amd64_nbsh_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	int rc;
+
+	rc = strict_strtoul(data, 16, &value);
+	if (rc != -EINVAL) {
+		debugf0("%s() received NBSL= 0x%lx\n", __func__, value);
+
+		/* place the NBSL value into the virtual error packet */
+		pvt->ctl_error_info.nbsh = (u32) value;
+
+		return count;
+	}
+	return rc;
+}
+
+/*
+ * amd64_nbsh_show
+ *
+ *	display back what the last NBSL value written
+ */
+static ssize_t amd64_nbsh_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	u32 value;
+
+	value = pvt->ctl_error_info.nbsh;
+
+	return sprintf(data, "%x\n", value);
+}
+
+/*
+ * amd64_nbcfg_store
+ *
+ *	accept and store the NBSL value user desires
+ */
+static ssize_t amd64_nbcfg_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	int rc;
+
+	rc = strict_strtoul(data, 16, &value);
+	if (rc != -EINVAL) {
+		debugf0("%s() received NBCFG= 0x%lx\n", __func__, value);
+
+		/* place the NBSL value into the virtual error packet */
+		pvt->ctl_error_info.nbcfg = (u32) value;
+
+		return count;
+	}
+	return rc;
+}
+
+/*
+ *	Various show routines for the controls of a MCI
+ */
+static ssize_t amd64_nbcfg_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	return sprintf(data, "%x\n",
+		pvt->ctl_error_info.nbcfg);
+}
+
+
+static ssize_t amd64_dhar_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	return sprintf(data, "%x\n", pvt->dhar);
+}
+
+
+static ssize_t amd64_dbam_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	return sprintf(data, "%x\n", pvt->dbam0);
+}
+
+
+static ssize_t amd64_topmem_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	return sprintf(data, "%llx\n", pvt->top_mem);
+}
+
+
+static ssize_t amd64_topmem2_show(struct mem_ctl_info *mci, char *data)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	return sprintf(data, "%llx\n", pvt->top_mem2);
+}
+
+static ssize_t amd64_hole_show(struct mem_ctl_info *mci, char *data)
+{
+	u64 hole_base = 0;
+	u64 hole_offset = 0;
+	u64 hole_size = 0;
+
+	amd64_get_dram_hole_info(mci, &hole_base, &hole_offset, &hole_size);
+
+	return sprintf(data, "%llx %llx %llx\n", hole_base, hole_offset,
+						 hole_size);
+}
+
+#endif	/* DEBUG */
+
-- 
1.6.2.4

[PATCH 20/21] amd64_edac: add DRAM error injection logic using sysfs

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/amd64_edac.c |  287 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 287 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index b1a7e8c..4d1076f 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -4621,3 +4621,290 @@ static ssize_t amd64_hole_show(struct mem_ctl_info *mci, char *data)
 
 #endif	/* DEBUG */
 
+#ifdef CONFIG_EDAC_AMD64_OPTERON_ERROR_INJECTION
+/*
+ * amd64_inject_section_store
+ *
+ *	accept and store error injection section value
+ *	range: 0..3
+ *		value refers to one of 4 16-byte sections
+ *		within a 64-byte cacheline
+ */
+static ssize_t amd64_inject_section_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	int rc;
+
+	rc = strict_strtoul(data, 10, &value);
+	if (rc != -EINVAL) {
+
+		/* save the 16-byte cache section */
+		pvt->injection.section = (u32) value;
+
+		return count;
+	}
+	return 0;
+}
+
+/*
+ * amd64_inject_word_store
+ *
+ *	accept and store error injection word value
+ *	range: 0..8
+ *		value refers to one of 9 16-bit word of the 16-byte section
+ *		128-bit + ECC bits
+ */
+static ssize_t amd64_inject_word_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	int rc;
+
+	rc = strict_strtoul(data, 10, &value);
+	if (rc != -EINVAL) {
+
+		/* save the 16-bit word */
+		value = (value <= 8) ? value : 0;
+		pvt->injection.word = (u32) value;
+
+		return count;
+	}
+	return 0;
+}
+
+/*
+ * amd64_inject_bit_store
+ *
+ *	accept and store error injection hexidecimal bit value
+ *	16-bits of a bit-vector marking which bits to error-out on
+ */
+static ssize_t amd64_inject_bit_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	int rc;
+
+	rc = strict_strtoul(data, 16, &value);
+	if (rc != -EINVAL) {
+
+		/* save the bit within the 16-bit word */
+		pvt->injection.bit_map = (u32) value & 0xFFFF;
+
+		return count;
+	}
+	return 0;
+}
+
+/*
+ * amd64_inject_read_store
+ *
+ *	READ action. When called, assemble staged values in the pvt
+ *	area and format into fields needed by the Injection hardware
+ *	Output to hardware and issue a READ operation
+ */
+static ssize_t amd64_inject_read_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	u32 section, word_bits;
+	int rc;
+
+	rc = strict_strtoul(data, 10, &value);
+	if (rc != -EINVAL) {
+
+		/* Form value to choose 16-byte section of cacheline */
+		section = F10_NB_ARRAY_DRAM_ECC |
+				SET_NB_ARRAY_ADDRESS(pvt->injection.section);
+		pci_write_config_dword(pvt->misc_f3_ctl,
+					F10_NB_ARRAY_ADDR, section);
+
+		word_bits = SET_NB_DRAM_INJECTION_READ(pvt->injection.word,
+						pvt->injection.bit_map);
+
+		/* Issue 'word' and 'bit' along with the READ request now */
+		pci_write_config_dword(pvt->misc_f3_ctl,
+					F10_NB_ARRAY_DATA, word_bits);
+
+		debugf0("%s() section=0x%x word_bits=0x%x\n", __func__,
+			section, word_bits);
+
+		return count;
+	}
+	return 0;
+}
+
+/*
+ * amd64_inject_write_store
+ *
+ *	WRITE action. When called, assemble staged values in the pvt
+ *	area and format into fields needed by the Injection hardware
+ *	Output to hardware and issue a WRITE operation
+ */
+static ssize_t amd64_inject_write_store(struct mem_ctl_info *mci,
+					const char *data, size_t count)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+	unsigned long value;
+	u32 section, word_bits;
+	int rc;
+
+	rc = strict_strtoul(data, 10, &value);
+	if (rc != -EINVAL) {
+
+		/* Form value to choose 16-byte section of cacheline */
+		section = F10_NB_ARRAY_DRAM_ECC |
+				SET_NB_ARRAY_ADDRESS(pvt->injection.section);
+		pci_write_config_dword(pvt->misc_f3_ctl,
+					F10_NB_ARRAY_ADDR, section);
+
+		word_bits = SET_NB_DRAM_INJECTION_WRITE(pvt->injection.word,
+						pvt->injection.bit_map);
+
+		/* Issue 'word' and 'bit' along with the READ request now */
+		pci_write_config_dword(pvt->misc_f3_ctl,
+					F10_NB_ARRAY_DATA, word_bits);
+
+		debugf0("%s() section=0x%x word_bits=0x%x\n", __func__,
+			section, word_bits);
+
+		return count;
+	}
+	return 0;
+}
+#endif
+
+/*
+ * Per MC instance Attribute/Control data control structure
+ *	Can add for debug or for normal use
+ */
+static struct mcidev_sysfs_attribute amd64_mc_sysfs_ctls_attrs[] = {
+
+#ifdef CONFIG_EDAC_AMD64_OPTERON_ERROR_INJECTION
+	/* Error injection methods */
+	{
+		.attr = {
+			.name = "z_inject_section",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = NULL,
+		.store = amd64_inject_section_store,
+	},
+	{
+		.attr = {
+			.name = "z_inject_word",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = NULL,
+		.store = amd64_inject_word_store,
+	},
+	{
+		.attr = {
+			.name = "z_inject_bit_map",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = NULL,
+		.store = amd64_inject_bit_store,
+	},
+	{
+		.attr = {
+			.name = "z_inject_write",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = NULL,
+		.store = amd64_inject_write_store,
+	},
+	{
+		.attr = {
+			.name = "z_inject_read",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = NULL,
+		.store = amd64_inject_read_store,
+	},
+#endif	/* CONFIG_EDAC_AMD64_OPTERON_ERROR_INJECTION */
+
+#ifdef	CONFIG_EDAC_DEBUG
+	/* RAW register accessors */
+	{
+		.attr = {
+			.name = "zctl_nbea",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = amd64_nbea_show,
+		.store = amd64_nbea_store,
+	},
+	{
+		.attr = {
+			.name = "zctl_nbsl",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = amd64_nbsl_show,
+		.store = amd64_nbsl_store,
+	},
+	{
+		.attr = {
+			.name = "zctl_nbsh",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = amd64_nbsh_show,
+		.store = amd64_nbsh_store,
+	},
+	{
+		.attr = {
+			.name = "zctl_nbcfg",
+			.mode = (S_IRUGO | S_IWUSR)
+		},
+		.show = amd64_nbcfg_show,
+		.store = amd64_nbcfg_store,
+	},
+	{
+		.attr = {
+			.name = "zhw_dhar",
+			.mode = (S_IRUGO)
+		},
+		.show = amd64_dhar_show,
+		.store = NULL,
+	},
+	{
+		.attr = {
+			.name = "zhw_dbam",
+			.mode = (S_IRUGO)
+		},
+		.show = amd64_dbam_show,
+		.store = NULL,
+	},
+	{
+		.attr = {
+			.name = "zhw_topmem",
+			.mode = (S_IRUGO)
+		},
+		.show = amd64_topmem_show,
+		.store = NULL,
+	},
+	{
+		.attr = {
+			.name = "zhw_topmem2",
+			.mode = (S_IRUGO)
+		},
+		.show = amd64_topmem2_show,
+		.store = NULL,
+	},
+	{
+		.attr = {
+			.name = "zhw_hole",
+			.mode = (S_IRUGO)
+		},
+		.show = amd64_hole_show,
+		.store = NULL,
+	},
+#endif
+	{
+		.attr = { .name = NULL}
+	}
+};
+
-- 
1.6.2.4

[PATCH 21/21] amd64_edac: add module registration routines

[Borislav Petkov]

From: Doug Thompson <dougthompson@xmission.com>

Also, link into Kbuild by adding Kconfig and Makefile entries.

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 drivers/edac/Kconfig      |   26 +++
 drivers/edac/Makefile     |    1 +
 drivers/edac/amd64_edac.c |  475 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 502 insertions(+), 0 deletions(-)

diff --git a/drivers/edac/Kconfig b/drivers/edac/Kconfig
index e5f5c5a..e854de1 100644
--- a/drivers/edac/Kconfig
+++ b/drivers/edac/Kconfig
@@ -58,6 +58,32 @@ config EDAC_MM_EDAC
 	  occurred so that a particular failing memory module can be
 	  replaced.  If unsure, select 'Y'.
 
+config EDAC_AMD64_OPTERON
+	tristate "AMD64 (Opteron, Athlon64) K8, F10h, F11h"
+	depends on EDAC_MM_EDAC && X86 && PCI && NUMA
+	help
+	Support for error detection and correction on the AMD 64
+	Families of Memory Controllers (K8, F10h and F11h)
+
+config EDAC_AMD64_OPTERON_ERROR_INJECTION
+	bool "/sys Error Injection access"
+	depends on EDAC_AMD64_OPTERON
+	help
+	  Recent Opterons (Family 10h and later) provide for Memory Error
+	  Injection into the ECC detection circuits. The amd64_edac module
+	  allows the operator/user to inject Uncorrectable and Correctable
+	  errors into DRAM.
+
+	  When enabled, in each of the respective memory controller directories
+	  (/sys/devices/system/edac/mc/mcX), there are 3 input files:
+
+	  - z_inject_section (0..3, 16-byte section of 64-byte cacheline),
+	  - z_inject_word (0..8, 16-bit word of 16-byte section),
+	  - z_inject_bit_map (hex bitmap vector: mask bits of 16 bit word to
+	    error-out)
+
+	  In addition, there are two control files, z_inject_read and
+	  z_inject_write, which trigger the Read and Write errors respectively.
 
 config EDAC_AMD76X
 	tristate "AMD 76x (760, 762, 768)"
diff --git a/drivers/edac/Makefile b/drivers/edac/Makefile
index a5fdcf0..262fee7 100644
--- a/drivers/edac/Makefile
+++ b/drivers/edac/Makefile
@@ -30,6 +30,7 @@ obj-$(CONFIG_EDAC_I3000)		+= i3000_edac.o
 obj-$(CONFIG_EDAC_X38)			+= x38_edac.o
 obj-$(CONFIG_EDAC_I82860)		+= i82860_edac.o
 obj-$(CONFIG_EDAC_R82600)		+= r82600_edac.o
+obj-$(CONFIG_EDAC_AMD64_OPTERON)	+= amd64_edac.o
 obj-$(CONFIG_EDAC_PASEMI)		+= pasemi_edac.o
 obj-$(CONFIG_EDAC_MPC85XX)		+= mpc85xx_edac.o
 obj-$(CONFIG_EDAC_MV64X60)		+= mv64x60_edac.o
diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 4d1076f..15f1fce 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -4908,3 +4908,478 @@ static struct mcidev_sysfs_attribute amd64_mc_sysfs_ctls_attrs[] = {
 	}
 };
 
+/*
+ * amd64_set_mc_sysfs_attributes
+ */
+static void amd64_set_mc_sysfs_attributes(struct mem_ctl_info *mci)
+{
+	mci->mc_driver_sysfs_attributes = amd64_mc_sysfs_ctls_attrs;
+}
+
+/*
+ * amd64_setup_mci_misc_attributes
+ *
+ *	initialize various attributes of the mci structure
+ */
+static void amd64_setup_mci_misc_attributes(struct mem_ctl_info *mci)
+{
+	struct amd64_pvt *pvt = mci->pvt_info;
+
+	/* Initialize various states */
+	mci->mtype_cap = MEM_FLAG_DDR2 | MEM_FLAG_RDDR2;
+	mci->edac_ctl_cap = EDAC_FLAG_NONE;
+	mci->edac_cap = EDAC_FLAG_NONE;
+
+	/* Exam the capabilities of the northbridge in order to reflect them
+	 * in the presentation via sysfs attributes, etc
+	 */
+	if (pvt->nbcap & K8_NBCAP_SECDED)
+		mci->edac_ctl_cap |= EDAC_FLAG_SECDED;
+
+	if (pvt->nbcap & K8_NBCAP_CHIPKILL)
+		mci->edac_ctl_cap |= EDAC_FLAG_S4ECD4ED;
+
+	/* What type of SECDED is there? */
+	mci->edac_cap = amd64_determine_edac_cap(pvt);
+
+	/* Misc attributes to set */
+	mci->mod_name = EDAC_MOD_STR;
+	mci->mod_ver = EDAC_AMD64_VERSION;
+	mci->ctl_name = get_amd_family_name(pvt->mc_type_index);
+	mci->dev_name = pci_name(pvt->dram_f2_ctl);
+	mci->ctl_page_to_phys = NULL;
+
+	/* IMPORTANT: Set the polling 'check' function in this module */
+	mci->edac_check = amd64_check;
+
+	/* memory scrubber interface */
+	mci->set_sdram_scrub_rate = amd64_set_scrub_rate;
+	mci->get_sdram_scrub_rate = amd64_get_scrub_rate;
+}
+
+/*
+ * amd64_probe_one_instance
+ *
+ *    probe function to determine if there is a DRAM Controller device is
+ *    present and to construct data tables for it.
+ *
+ *    Due to a hardware feature on Family 10H cpus, the Enable Extended
+ *    Configuration Space feature MUST be enabled on ALL Processors prior to
+ *    actually reading from the ECS registers. Since the loading of the module
+ *    can occur on any 'core', and cores don't 'see' all the other processors
+ *    ECS data when the others are NOT enabled. Our solution is to first
+ *    enable ECS access in this routine on all processors, gather some data in
+ *    a amd64_pvt structure and later come back in a 'finishup_setup' function
+ *    to perform that final initialization.
+ *
+ *    See also amd64_init_2nd_stage().
+ */
+static int amd64_probe_one_instance(struct pci_dev *dram_f2_ctl,
+					int mc_type_index)
+{
+	struct amd64_pvt *pvt;
+	int err, num_cpus;
+	int rc;
+
+	pvt = kzalloc(sizeof(struct amd64_pvt), GFP_KERNEL);
+	if (pvt == NULL) {
+		rc = -ENOMEM;
+		goto exit_now;
+	}
+
+	pvt->mc_node_id = get_mc_node_id_from_pdev(dram_f2_ctl);
+
+	debugf0("=========== %s(Instance= %d) ===========\n",
+		__func__, pvt->mc_node_id);
+
+	pvt->dram_f2_ctl = dram_f2_ctl;
+	pvt->ext_model = boot_cpu_data.x86_model >> 4;
+	pvt->mc_type_index = mc_type_index;
+	pvt->ops = get_amd_family_ops(mc_type_index);
+
+	num_cpus = cpumask_weight(cpumask_of_node(pvt->mc_node_id));
+	pvt->old_mcgctl = kzalloc(sizeof(u32) * num_cpus, GFP_KERNEL);
+	if (!pvt->old_mcgctl) {
+		rc = -ENOMEM;
+		goto free_pvt;
+	}
+
+	/*
+	 * We have the dram_f2_ctl device as an argument, now go reserved its
+	 * sibling devices from the PCI system.
+	 */
+	err = amd64_reserve_mc_sibling_devices(pvt, mc_type_index);
+	if (err) {
+		rc = -ENODEV;
+		goto exit_now;
+	}
+
+	rc = amd64_check_ecc_enabled(pvt);
+	if (rc)
+		goto exit_release_devices;
+
+	/*
+	 * Key operation here: setup of HW prior to performing ops on it. Some
+	 * setup is required to access ECS data. After this is performed, then
+	 * the 'teardown' function must be called upon error and normal exit
+	 * paths.
+	 */
+	if (boot_cpu_data.x86 > 0xf)
+		amd64_setup(pvt);
+
+	/*
+	 * Save the pointer to the private data for use in 2nd initialization
+	 * stage
+	 */
+	pvt_lookup[pvt->mc_node_id] = pvt;
+
+	debugf0("%s(): init 1st stage done pvt-%d\n", __func__,
+		pvt->mc_node_id);
+	return 0;
+
+
+exit_release_devices:
+	pci_dev_put(pvt->addr_f1_ctl);
+	pci_dev_put(pvt->misc_f3_ctl);
+
+free_pvt:
+	kfree(pvt);
+
+exit_now:
+	return rc;
+}
+
+/*
+ * amd64_init_2nd_stage
+ *
+ *	this is the "finishing" up initialization code
+ *	Needs to be performed after all MCs' Hardware have been
+ *	"prep'ed" for accessing extended config space.
+ */
+static int amd64_init_2nd_stage(struct amd64_pvt *pvt_temp)
+{
+	int node_id = pvt_temp->mc_node_id;
+	struct mem_ctl_info *mci;
+	struct amd64_pvt *pvt;
+	int rc;
+	int err;
+
+	debugf0("%s()\n", __func__);
+
+	amd64_read_mc_registers(pvt_temp);
+
+	/* Check hardware to see if this module can support HW at this time */
+	if (pvt_temp->ops->probe_valid_hardware) {
+		err = pvt_temp->ops->probe_valid_hardware(pvt_temp);
+		if (err) {
+			rc = -ENODEV;
+			goto exit_failure;
+		}
+	}
+
+	/*
+	 * We need to determine how many memory channels there are. Then use
+	 * that information for calculating the size of the dynamic instance
+	 * tables in the 'mci' structure
+	 */
+	pvt_temp->channel_count = pvt_temp->ops->early_channel_count(pvt_temp);
+
+	mci = edac_mc_alloc(sizeof(*pvt_temp),
+				CHIPSELECT_COUNT,
+				pvt_temp->channel_count,
+				node_id);
+	if (mci == NULL) {
+		rc = -ENOMEM;
+		goto exit_failure;
+	}
+
+	/*
+	 * transfer the info from the interium pvt area to the private area of
+	 * the MC instance structure
+	 */
+	pvt = mci->pvt_info;
+	*pvt = *pvt_temp;
+
+	mci->dev = &pvt_temp->dram_f2_ctl->dev;
+	amd64_setup_mci_misc_attributes(mci);
+
+	if (amd64_init_csrows(mci)) {
+		debugf1("Setting mci->edac_cap to EDAC_FLAG_NONE because\n");
+		debugf1("   amd64_init_csrows() returned NO csrows found\n");
+		mci->edac_cap = EDAC_FLAG_NONE;
+	}
+
+	amd64_enable_ecc_error_reporting(mci);
+	amd64_set_mc_sysfs_attributes(mci);
+
+	if (edac_mc_add_mc(mci)) {
+		debugf1("%s(): failed edac_mc_add_mc()\n", __func__);
+		rc = -ENODEV;
+		goto exit_add_mc_failure;
+	}
+
+	debugf0("%s(): init 2nd stage done mci%d\n", __func__,
+		pvt->mc_node_id);
+
+	mci_lookup[node_id] = mci;
+
+	kfree((pvt_lookup[pvt->mc_node_id])->old_mcgctl);
+	kfree(pvt_lookup[pvt->mc_node_id]);
+	pvt_lookup[node_id] = NULL;
+	return 0;
+
+exit_add_mc_failure:
+	edac_mc_free(mci);
+
+exit_failure:
+	debugf0("%s() failure init 2nd stage: rc=%d\n", __func__, rc);
+
+	amd64_restore_ecc_error_reporting(pvt);
+
+	if (boot_cpu_data.x86 > 0xf)
+		amd64_teardown(pvt);
+
+	pci_dev_put(pvt->addr_f1_ctl);
+	pci_dev_put(pvt->misc_f3_ctl);
+
+	kfree((pvt_lookup[pvt->mc_node_id])->old_mcgctl);
+	kfree(pvt_lookup[pvt->mc_node_id]);
+	pvt_lookup[node_id] = NULL;
+
+	return rc;
+}
+
+
+/*
+ * amd64_init_one_instance
+ *
+ *	initialize just one device
+ *
+ *	returns:
+ *		 count (>= 0), or
+ *		negative on error
+ */
+static int __devinit amd64_init_one_instance(struct pci_dev *pdev,
+				 const struct pci_device_id *mc_type)
+{
+	int rc;
+
+	debugf0("%s(MC node=%d,mc_type='%s')\n",
+		__func__,
+		get_mc_node_id_from_pdev(pdev),
+		get_amd_family_name(mc_type->driver_data));
+
+	/* wake up and enable device */
+	rc = pci_enable_device(pdev);
+	if (rc < 0)
+		rc = -EIO;
+	else
+		rc = amd64_probe_one_instance(pdev, mc_type->driver_data);
+
+	if (rc < 0)
+		debugf0("%s() rc=%d\n", __func__, rc);
+
+	return rc;
+}
+
+/*
+ * amd64_remove_one_instance
+ *
+ *	remove just one device instance upon driver unloading
+ */
+static void __devexit amd64_remove_one_instance(struct pci_dev *pdev)
+{
+	struct mem_ctl_info *mci;
+	struct amd64_pvt *pvt;
+
+	debugf0("%s()\n", __func__);
+
+	/* Remove from EDAC CORE tracking list */
+	mci = edac_mc_del_mc(&pdev->dev);
+	if (mci == NULL)
+		return;
+
+	pvt = mci->pvt_info;
+
+	amd64_restore_ecc_error_reporting(pvt);
+
+	if (boot_cpu_data.x86 > 0xf)
+		amd64_teardown(pvt);
+
+	pci_dev_put(pvt->addr_f1_ctl);
+	pci_dev_put(pvt->misc_f3_ctl);
+
+	mci_lookup[pvt->mc_node_id] = NULL;
+
+	/* Free the EDAC CORE resources */
+	edac_mc_free(mci);
+}
+
+/*
+ * The 'pci_device_id' table.
+ *
+ *	This table is part of the interface for loading drivers for PCI
+ *	devices. The PCI core identifies what devices are on a system
+ *	during boot, and then inquiry this table to see if this driver
+ *	is for a given device found.
+ *
+ *	The PCI helpper functions walk this table and call the
+ *	'.probe' function of the 'pci_driver' table, for each
+ *	instance in this table
+ */
+static const struct pci_device_id amd64_pci_table[] __devinitdata = {
+	{
+		/* Rev F and prior */
+		.vendor = PCI_VENDOR_ID_AMD,
+		.device = PCI_DEVICE_ID_AMD_K8_NB_MEMCTL,
+		.subvendor = PCI_ANY_ID,
+		.subdevice = PCI_ANY_ID,
+		.class = 0,
+		.class_mask = 0,
+		.driver_data = K8_CPUS
+	},
+	{
+		/* Family 10h */
+		.vendor = PCI_VENDOR_ID_AMD,
+		.device = PCI_DEVICE_ID_AMD_10H_NB_DRAM,
+		.subvendor = PCI_ANY_ID,
+		.subdevice = PCI_ANY_ID,
+		.class = 0,
+		.class_mask = 0,
+		.driver_data = F10_CPUS
+	},
+	{
+		/* Family 11h */
+		.vendor = PCI_VENDOR_ID_AMD,
+		.device = PCI_DEVICE_ID_AMD_11H_NB_DRAM,
+		.subvendor = PCI_ANY_ID,
+		.subdevice = PCI_ANY_ID,
+		.class = 0,
+		.class_mask = 0,
+		.driver_data = F11_CPUS
+	},
+	{0, }
+};
+MODULE_DEVICE_TABLE(pci, amd64_pci_table);
+
+/*
+ * The 'pci_driver' structure to define the name, probe and removal
+ * functions
+ */
+static struct pci_driver amd64_pci_driver = {
+	.name = EDAC_MOD_STR,
+	.probe = amd64_init_one_instance,
+	.remove = __devexit_p(amd64_remove_one_instance),
+	.id_table = amd64_pci_table,
+};
+
+
+/*
+ * amd64_setup_pci_device
+ *
+ *	setup the PCI Device Driver for monitoring PCI errors
+ */
+static void amd64_setup_pci_device(void)
+{
+	struct mem_ctl_info *mci;
+	struct amd64_pvt *pvt;
+
+	if (!amd64_ctl_pci) {
+
+		mci = mci_lookup[0];
+		if (mci) {
+			debugf1("%s(): Registering ONE PCI control\n",
+				__func__);
+
+			pvt = mci->pvt_info;
+			amd64_ctl_pci = edac_pci_create_generic_ctl(
+						&pvt->dram_f2_ctl->dev,
+						EDAC_MOD_STR);
+			if (!amd64_ctl_pci) {
+				printk(KERN_WARNING
+					"%s(): Unable to create PCI control\n",
+					__func__);
+				printk(KERN_WARNING
+					"%s(): PCI error report via EDAC "
+					"not setup\n",
+					__func__);
+			}
+		} else {
+			debugf1("%s(): ONE PCI control already registered\n",
+				__func__);
+		}
+	}
+}
+
+static int __init amd64_edac_init(void)
+{
+	int err;
+	int node;
+
+	edac_printk(KERN_INFO, EDAC_MOD_STR, EDAC_AMD64_VERSION "\n");
+
+	opstate_init();
+
+	debugf0("%s() ******************  ENTRY  **********************\n",
+		__func__);
+
+	/* Attempt to register drivers for instances
+	 * DUE to the failure of some 'cores' to access Extended Config Space
+	 * prior to all memory controllers having their ECS register enabled,
+	 * the initialization has been created into 2 stages. Here we
+	 * call for the 1st stage. After all have been enabled, then we
+	 * do the 2nd stage to finishup setup.
+	 */
+	err = pci_register_driver(&amd64_pci_driver);
+
+	/* At this point, the array 'pvt_lookup[]' contains pointers to
+	 * allocated struct amd64_pvt control structures. These will be used
+	 * in the 2nd stage init function to finish initialization of
+	 * the MC instances.
+	 */
+
+	/* if no error occurred on first pass init, then do 2nd pass init */
+	if (!err) {
+		for_each_online_node(node) {
+			if (!pvt_lookup[node])
+				continue;
+
+			/* If any failure then need to clean up */
+			err = amd64_init_2nd_stage(pvt_lookup[node]);
+			if (err) {
+				debugf0("%s() 'finish_setup' stage failed\n",
+					__func__);
+
+				/* undo prior instances' registrations
+				 * and leave as failed
+				 */
+				pci_unregister_driver(&amd64_pci_driver);
+				goto error_exit;
+			}
+		}
+		amd64_setup_pci_device();
+	}
+
+error_exit:
+	return err;
+}
+
+static void __exit amd64_edac_exit(void)
+{
+	if (amd64_ctl_pci)
+		edac_pci_release_generic_ctl(amd64_ctl_pci);
+
+	pci_unregister_driver(&amd64_pci_driver);
+}
+
+module_init(amd64_edac_init);
+module_exit(amd64_edac_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("SoftwareBitMaker: Doug Thompson, "
+		"Dave Peterson, Thayne Harbaugh");
+MODULE_DESCRIPTION("MC support for AMD64 memory controllers - "
+		EDAC_AMD64_VERSION);
+
+module_param(edac_op_state, int, 0444);
+MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
-- 
1.6.2.4

Re: [PATCH 01/21] x86: add methods for writing of an MSR on several CPUs

[H. Peter Anvin]

Borislav Petkov wrote:
> +}
> +
> +/* rdmsr on a bunch of cpus
> + *
> + * @mask:	which cpus
> + * @msr_no:	which MSR
> + * @l:		array of the lower 32-bit MSR values
> + * @h:		array of the higher 32-bit MSR values
> + *

Urk!

Please, for the love of God, no!

Make it an array of u64s or (equivalently!) an array of (l,h)
structures, not separate arrays for the halves of the register.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: [PATCH 20/21] amd64_edac: add DRAM error injection logic using sysfs

[Ingo Molnar]

* Borislav Petkov <borislav.petkov@amd.com> wrote:

> From: Doug Thompson <dougthompson@xmission.com>
> 
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>  drivers/edac/amd64_edac.c |  287 +++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 287 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index b1a7e8c..4d1076f 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -4621,3 +4621,290 @@ static ssize_t amd64_hole_show(struct mem_ctl_info *mci, char *data)
>  
>  #endif	/* DEBUG */
>  
> +#ifdef CONFIG_EDAC_AMD64_OPTERON_ERROR_INJECTION

this should be in a separate .c file under drivers/edac/amd64/. 

Introducing large #ifdef sections like that is not very clean. The 
amd64_edac.c file is _way_ too large at more than 5000 lines of 
code.

	Ingo

Re: [PATCH 19/21] amd64_edac: add debugging/testing code

[Ingo Molnar]

* Borislav Petkov <borislav.petkov@amd.com> wrote:

> From: Doug Thompson <dougthompson@xmission.com>
> 
> This is for dumping different registers and testing the address mapping
> logic using the ECC syndromes.
> 
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>  drivers/edac/amd64_edac.c |  214 +++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 214 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index 43f236d..b1a7e8c 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -4407,3 +4407,217 @@ static int amd64_check_ecc_enabled(struct amd64_pvt *pvt)
>  	return rc;
>  }
>  
> +#ifdef	CONFIG_EDAC_DEBUG

should be in a separate .c file too i suspect.

	Ingo

Re: [PATCH 16/21] amd64_edac: add error decoding logic

[Ingo Molnar]

* Borislav Petkov <borislav.petkov@amd.com> wrote:

> From: Doug Thompson <dougthompson@xmission.com>
> 
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>  drivers/edac/amd64_edac.c |  569 +++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 569 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index 88849a4..fee9ec4 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -3260,3 +3260,572 @@ static int get_channel_from_ecc_syndrome(unsigned short syndrome)
>  	debugf0("%s(): syndrome(%x) not found\n", __func__, syndrome);
>  	return -1;
>  }
> +
> +/*
> + * toString tables for the various error decoding values
> + */
> +static const char *tt_msgs[] = {        /* transaction type */
> +	"instruction",
> +	"data",
> +	"generic",
> +	"reserved"
> +};

this too should be in a separate file.

	Ingo

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Ingo Molnar]

* Borislav Petkov <borislav.petkov@amd.com> wrote:

> From: Doug Thompson <dougthompson@xmission.com>
> 
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>  drivers/edac/amd64_edac.c |  318 +++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 318 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index fe2342c..84075c0 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -2726,4 +2726,322 @@ static int f10_lookup_addr_in_dct(u32 InputAddr, u32 NodeID, u32 ChannelSelect)
>  	return CSFound;
>  }
>  
> +/*
> + * f10_match_to_this_node
> + *
> + * For a given 'DramRange' value, check if 'SystemAddr' fall within this value
> + */
> +static int f10_match_to_this_node(struct amd64_pvt *pvt, int DramRange,
> +				u64 SystemAddr,
> +				int *node_id,
> +				int *channel_select)
> +{
> +	int CSFound = -1;
> +	int NodeID;
> +	int HiRangeSelected;
> +	u32 IntlvEn, IntlvSel;
> +	u32 DramEn;
> +	u32 Ilog;
> +	u32 HoleOffset, HoleEn;
> +	u32 InputAddr, Temp;
> +	u32 DctSelBaseAddr, DctSelIntLvAddr;
> +	u32 DctSelHi;
> +	u32 ChannelSelect;
> +	u64 DramBaseLong, DramLimitLong;
> +	u64 DctSelBaseOffsetLong, ChannelAddrLong;
> +
> +	/* DRAM Base value for this DRAM instance */
> +	DramBaseLong = pvt->dram_base[DramRange];
> +	DramEn = pvt->dram_rw_en[DramRange];
> +	IntlvEn = pvt->dram_IntlvEn[DramRange];
> +
> +	/* DRAM Limit value for this DRAM instance */
> +	DramLimitLong = pvt->dram_limit[DramRange];
> +	NodeID = pvt->dram_DstNode[DramRange];
> +	IntlvSel = pvt->dram_IntlvSel[DramRange];
> +
> +	debugf1("%s(dram=%d) Base=0x%llx SystemAddr= 0x%llx Limit=0x%llx\n",
> +		__func__, DramRange, DramBaseLong, SystemAddr, DramLimitLong);
> +
> +	/* This assumes that one node's DHAR is the same as
> +	 * all the other node's DHARs
> +	 */
> +	HoleEn = pvt->dhar;
> +	HoleOffset = (HoleEn & 0x0000FF80);
> +	HoleEn = (HoleEn & 0x00000003);
> +
> +	debugf1("   HoleOffset=0x%x  HoleEn=0x%x IntlvSel=0x%x\n",
> +			HoleOffset, HoleEn, IntlvSel);
> +
> +	if ((IntlvEn == 0) || IntlvSel == ((SystemAddr >> 12) & IntlvEn)) {
> +
> +		Ilog = f10_map_IntlvEn_to_shift(IntlvEn);
> +
> +		Temp = pvt->dram_ctl_select_low;
> +		DctSelBaseOffsetLong = pvt->dram_ctl_select_high << 16;
> +
> +		DctSelHi = (Temp >> 1) & 1;
> +		DctSelIntLvAddr = dct_sel_interleave_addr(pvt);
> +		DctSelBaseAddr = dct_sel_baseaddr(pvt);
> +
> +		if (dct_high_range_enabled(pvt) &&
> +		   !dct_ganging_enabled(pvt) &&
> +		   ((SystemAddr >> 27) >= (DctSelBaseAddr >> 11)))
> +			HiRangeSelected = 1;
> +		else
> +			HiRangeSelected = 0;
> +
> +		ChannelSelect = f10_determine_channel(pvt, SystemAddr,
> +						HiRangeSelected, IntlvEn);
> +
> +		ChannelAddrLong = f10_determine_base_addr_offset(
> +						SystemAddr,
> +						HiRangeSelected,
> +						DctSelBaseAddr,
> +						DctSelBaseOffsetLong,
> +						HoleEn,
> +						HoleOffset,
> +						DramBaseLong);
> +
> +		/* Remove Node ID (in case of processor interleaving) */
> +		Temp = ChannelAddrLong & 0xFC0;
> +
> +		ChannelAddrLong = ((ChannelAddrLong >> Ilog) &
> +					0xFFFFFFFFF000ULL) | Temp;
> +
> +		/* Remove Channel interleave and hash */
> +		if (dct_interleave_enabled(pvt) &&
> +		   !dct_high_range_enabled(pvt) &&
> +		   !dct_ganging_enabled(pvt)) {
> +			if (DctSelIntLvAddr != 1)
> +				ChannelAddrLong =
> +					(ChannelAddrLong >> 1) &
> +					0xFFFFFFFFFFFFFFC0ULL;
> +			else {
> +				Temp = ChannelAddrLong & 0xFC0;
> +				ChannelAddrLong =
> +					((ChannelAddrLong &
> +					0xFFFFFFFFFFFFC000ULL)
> +					>> 1) | Temp;
> +			}
> +		}
> +
> +		/* Form a normalize InputAddr (Move bits 36:8 down to 28:0
> +		 * which will set it up to match the DCT Base register
> +		 */
> +		InputAddr = ChannelAddrLong >> 8;
> +
> +		debugf1("   (ChannelAddrLong=0x%llx) >> 8 becomes "
> +			"InputAddr=0x%x\n", ChannelAddrLong, InputAddr);
> +
> +		/* Iterate over the DRAM DCTs looking for a
> +		 * match for InputAddr on the selected NodeID
> +		 */
> +		CSFound = f10_lookup_addr_in_dct(InputAddr,
> +						NodeID, ChannelSelect);
> +
> +		if (CSFound >= 0) {
> +			*node_id = NodeID;
> +			*channel_select = ChannelSelect;
> +		}
> +	}
> +
> +	return CSFound;
> +}

this function is probably too large, and also it uses some weird 
hungarian notation coding style. Please dont do that! It's 
completely unacceptable.

this condition:

> +	if ((IntlvEn == 0) || IntlvSel == ((SystemAddr >> 12) & IntlvEn)) {

could be inverted and an early "return cs_found" could be done - 
saving an indentitation level for most of the above code.

etc. etc.

Please look at the function in a really large xterm, from far away. 
If the shape does not look 'good', and the structure is not an 
obvious pattern seen a hundred times elsewhere in the kernel, 
there's something weird going on with the function. It should be 
split up, cleaned up, simplified. Variable names could become 
shorter, etc. etc.

	Ingo

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Ingo Molnar]

* Ingo Molnar <mingo@elte.hu> wrote:

> 
> * Borislav Petkov <borislav.petkov@amd.com> wrote:
> 
> > From: Doug Thompson <dougthompson@xmission.com>
> > 
> > Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> > Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> > ---
> >  drivers/edac/amd64_edac.c |  318 +++++++++++++++++++++++++++++++++++++++++++++
> >  1 files changed, 318 insertions(+), 0 deletions(-)
> > 
> > diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> > index fe2342c..84075c0 100644
> > --- a/drivers/edac/amd64_edac.c
> > +++ b/drivers/edac/amd64_edac.c
> > @@ -2726,4 +2726,322 @@ static int f10_lookup_addr_in_dct(u32 InputAddr, u32 NodeID, u32 ChannelSelect)
> >  	return CSFound;
> >  }
> >  
> > +/*
> > + * f10_match_to_this_node
> > + *
> > + * For a given 'DramRange' value, check if 'SystemAddr' fall within this value
> > + */
> > +static int f10_match_to_this_node(struct amd64_pvt *pvt, int DramRange,
> > +				u64 SystemAddr,
> > +				int *node_id,
> > +				int *channel_select)
> > +{
> > +	int CSFound = -1;
> > +	int NodeID;
> > +	int HiRangeSelected;
> > +	u32 IntlvEn, IntlvSel;
> > +	u32 DramEn;
> > +	u32 Ilog;
> > +	u32 HoleOffset, HoleEn;
> > +	u32 InputAddr, Temp;
> > +	u32 DctSelBaseAddr, DctSelIntLvAddr;
> > +	u32 DctSelHi;
> > +	u32 ChannelSelect;
> > +	u64 DramBaseLong, DramLimitLong;
> > +	u64 DctSelBaseOffsetLong, ChannelAddrLong;
> > +
> > +	/* DRAM Base value for this DRAM instance */
> > +	DramBaseLong = pvt->dram_base[DramRange];
> > +	DramEn = pvt->dram_rw_en[DramRange];
> > +	IntlvEn = pvt->dram_IntlvEn[DramRange];
> > +
> > +	/* DRAM Limit value for this DRAM instance */
> > +	DramLimitLong = pvt->dram_limit[DramRange];
> > +	NodeID = pvt->dram_DstNode[DramRange];
> > +	IntlvSel = pvt->dram_IntlvSel[DramRange];
> > +
> > +	debugf1("%s(dram=%d) Base=0x%llx SystemAddr= 0x%llx Limit=0x%llx\n",
> > +		__func__, DramRange, DramBaseLong, SystemAddr, DramLimitLong);
> > +
> > +	/* This assumes that one node's DHAR is the same as
> > +	 * all the other node's DHARs
> > +	 */
> > +	HoleEn = pvt->dhar;
> > +	HoleOffset = (HoleEn & 0x0000FF80);
> > +	HoleEn = (HoleEn & 0x00000003);
> > +
> > +	debugf1("   HoleOffset=0x%x  HoleEn=0x%x IntlvSel=0x%x\n",
> > +			HoleOffset, HoleEn, IntlvSel);
> > +
> > +	if ((IntlvEn == 0) || IntlvSel == ((SystemAddr >> 12) & IntlvEn)) {
> > +
> > +		Ilog = f10_map_IntlvEn_to_shift(IntlvEn);
> > +
> > +		Temp = pvt->dram_ctl_select_low;
> > +		DctSelBaseOffsetLong = pvt->dram_ctl_select_high << 16;
> > +
> > +		DctSelHi = (Temp >> 1) & 1;
> > +		DctSelIntLvAddr = dct_sel_interleave_addr(pvt);
> > +		DctSelBaseAddr = dct_sel_baseaddr(pvt);
> > +
> > +		if (dct_high_range_enabled(pvt) &&
> > +		   !dct_ganging_enabled(pvt) &&
> > +		   ((SystemAddr >> 27) >= (DctSelBaseAddr >> 11)))
> > +			HiRangeSelected = 1;
> > +		else
> > +			HiRangeSelected = 0;
> > +
> > +		ChannelSelect = f10_determine_channel(pvt, SystemAddr,
> > +						HiRangeSelected, IntlvEn);
> > +
> > +		ChannelAddrLong = f10_determine_base_addr_offset(
> > +						SystemAddr,
> > +						HiRangeSelected,
> > +						DctSelBaseAddr,
> > +						DctSelBaseOffsetLong,
> > +						HoleEn,
> > +						HoleOffset,
> > +						DramBaseLong);
> > +
> > +		/* Remove Node ID (in case of processor interleaving) */
> > +		Temp = ChannelAddrLong & 0xFC0;
> > +
> > +		ChannelAddrLong = ((ChannelAddrLong >> Ilog) &
> > +					0xFFFFFFFFF000ULL) | Temp;
> > +
> > +		/* Remove Channel interleave and hash */
> > +		if (dct_interleave_enabled(pvt) &&
> > +		   !dct_high_range_enabled(pvt) &&
> > +		   !dct_ganging_enabled(pvt)) {
> > +			if (DctSelIntLvAddr != 1)
> > +				ChannelAddrLong =
> > +					(ChannelAddrLong >> 1) &
> > +					0xFFFFFFFFFFFFFFC0ULL;
> > +			else {
> > +				Temp = ChannelAddrLong & 0xFC0;
> > +				ChannelAddrLong =
> > +					((ChannelAddrLong &
> > +					0xFFFFFFFFFFFFC000ULL)
> > +					>> 1) | Temp;
> > +			}
> > +		}
> > +
> > +		/* Form a normalize InputAddr (Move bits 36:8 down to 28:0
> > +		 * which will set it up to match the DCT Base register
> > +		 */
> > +		InputAddr = ChannelAddrLong >> 8;
> > +
> > +		debugf1("   (ChannelAddrLong=0x%llx) >> 8 becomes "
> > +			"InputAddr=0x%x\n", ChannelAddrLong, InputAddr);
> > +
> > +		/* Iterate over the DRAM DCTs looking for a
> > +		 * match for InputAddr on the selected NodeID
> > +		 */
> > +		CSFound = f10_lookup_addr_in_dct(InputAddr,
> > +						NodeID, ChannelSelect);
> > +
> > +		if (CSFound >= 0) {
> > +			*node_id = NodeID;
> > +			*channel_select = ChannelSelect;
> > +		}
> > +	}
> > +
> > +	return CSFound;
> > +}
> 
> this function is probably too large, and also it uses some weird 
> hungarian notation coding style. Please dont do that! It's 
> completely unacceptable.
> 
> this condition:
> 
> > +	if ((IntlvEn == 0) || IntlvSel == ((SystemAddr >> 12) & IntlvEn)) {
> 
> could be inverted and an early "return cs_found" could be done - 
> saving an indentitation level for most of the above code.
> 
> etc. etc.
> 
> Please look at the function in a really large xterm, from far 
> away. If the shape does not look 'good', and the structure is not 
> an obvious pattern seen a hundred times elsewhere in the kernel, 
> there's something weird going on with the function. It should be 
> split up, cleaned up, simplified. Variable names could become 
> shorter, etc. etc.

... and this general observation about variable naming and general 
structure holds for the rest of the patches as well. This really 
needs to be sorted out before a more detailed review can be done.

	Ingo

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Andrew Morton]

On Wed, 29 Apr 2009 20:22:55 +0200
Ingo Molnar <mingo@elte.hu> wrote:

> > +		InputAddr = ChannelAddrLong >> 8;
> > +
> > +		debugf1("   (ChannelAddrLong=0x%llx) >> 8 becomes "
> > +			"InputAddr=0x%x\n", ChannelAddrLong, InputAddr);
> > +
> > +		/* Iterate over the DRAM DCTs looking for a
> > +		 * match for InputAddr on the selected NodeID
> > +		 */
> > +		CSFound = f10_lookup_addr_in_dct(InputAddr,
> > +						NodeID, ChannelSelect);
> > +
> > +		if (CSFound >= 0) {
> > +			*node_id = NodeID;
> > +			*channel_select = ChannelSelect;
> > +		}
> > +	}
> > +
> > +	return CSFound;
> > +}
> 
> this function is probably too large, and also it uses some weird 
> hungarian notation coding style. Please dont do that! It's 
> completely unacceptable.

These identifers (or at least, DctSelBaseOffsetLong, which is the only
one I googled for) come straight out of the AMD "BIOS and Kernel
Developer's Guide".

Sucky though they are, there's value in making the kernel code match up
with the documentation.

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Ingo Molnar]

* Andrew Morton <akpm@linux-foundation.org> wrote:

> On Wed, 29 Apr 2009 20:22:55 +0200
> Ingo Molnar <mingo@elte.hu> wrote:
> 
> > > +		InputAddr = ChannelAddrLong >> 8;
> > > +
> > > +		debugf1("   (ChannelAddrLong=0x%llx) >> 8 becomes "
> > > +			"InputAddr=0x%x\n", ChannelAddrLong, InputAddr);
> > > +
> > > +		/* Iterate over the DRAM DCTs looking for a
> > > +		 * match for InputAddr on the selected NodeID
> > > +		 */
> > > +		CSFound = f10_lookup_addr_in_dct(InputAddr,
> > > +						NodeID, ChannelSelect);
> > > +
> > > +		if (CSFound >= 0) {
> > > +			*node_id = NodeID;
> > > +			*channel_select = ChannelSelect;
> > > +		}
> > > +	}
> > > +
> > > +	return CSFound;
> > > +}
> > 
> > this function is probably too large, and also it uses some weird 
> > hungarian notation coding style. Please dont do that! It's 
> > completely unacceptable.
> 
> These identifers (or at least, DctSelBaseOffsetLong, which is the 
> only one I googled for) come straight out of the AMD "BIOS and 
> Kernel Developer's Guide".
> 
> Sucky though they are, there's value in making the kernel code 
> match up with the documentation.

I'm generally resisting patches that hungarinize arch/x86/ (and heck 
there's been many attempts ...) but there's some conflicting advice 
here. I've Cc:-ed Linus, maybe he has an opinion about this.

My gut reaction would be 'hell no'. There's other, structural 
problems with this code too, and doing some saner naming would 
mostly be a sed job and would take minimal amount of time. The 
naming can still be intuitive. The symbols from the documentation 
can perhaps be mentioned in a couple of comments to establish a 
mapping.

	Ingo

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Andi Kleen]

Borislav Petkov <borislav.petkov@amd.com> writes:

> Hi,
>
> thanks to all reviewers of the previous submission, here is the second
> version of this series.

The classic problem of the previous versions of these patches was that
they consume the same error registers (even if using pci config versus
msrs as access methods) as the kernel machine check poll/threshold
interrupt code.  And with two logging agents racing on the same
registers you will always get junk results. Typically with threshold 
enabled the mce code wins the race. I suspect this patchkit has
exactly the same fundamental design problem. EDAC really is not
particularly fitting for integrated memory controllers that report
their errors using standard machine check events.

-Andi (who thinks all of this decoding should be in user space anyways)

-- 
ak@linux.intel.com -- Speaking for myself only.

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Andrew Morton]

On Wed, 29 Apr 2009 21:23:26 +0200
Ingo Molnar <mingo@elte.hu> wrote:

> > > > +		if (CSFound >= 0) {
> > > > +			*node_id = NodeID;
> > > > +			*channel_select = ChannelSelect;
> > > > +		}
> > > > +	}
> > > > +
> > > > +	return CSFound;
> > > > +}
> > > 
> > > this function is probably too large, and also it uses some weird 
> > > hungarian notation coding style. Please dont do that! It's 
> > > completely unacceptable.
> > 
> > These identifers (or at least, DctSelBaseOffsetLong, which is the 
> > only one I googled for) come straight out of the AMD "BIOS and 
> > Kernel Developer's Guide".
> > 
> > Sucky though they are, there's value in making the kernel code 
> > match up with the documentation.
> 
> I'm generally resisting patches that hungarinize arch/x86/ (and heck 
> there's been many attempts ...) but there's some conflicting advice 
> here. I've Cc:-ed Linus, maybe he has an opinion about this.
> 
> My gut reaction would be 'hell no'. There's other, structural 
> problems with this code too, and doing some saner naming would 
> mostly be a sed job and would take minimal amount of time. The 
> naming can still be intuitive. The symbols from the documentation 
> can perhaps be mentioned in a couple of comments to establish a 
> mapping.

I think I disagree.  For those identifiers which map 1:1 with the
manufacturer's document, the ugliness involved in exactly copying the
manufacturer's chosen identifiers is outweighed by the benefit of
exactly copying the manufacturer's chosen identifiers.

Of course, we don't have to use StinkyIdentifiers anywhere else.  And
the nice thing about that is that when one reads the code and comes
across a StinkyIdentifier, one immeditely knows that it's an
AMD-provided thing rather than a Linux-provided thing.

Zillions of StinkyIdentifiers get merged via this logic.

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Ingo Molnar]

* Andrew Morton <akpm@linux-foundation.org> wrote:

> On Wed, 29 Apr 2009 21:23:26 +0200
> Ingo Molnar <mingo@elte.hu> wrote:
> 
> > > > > +		if (CSFound >= 0) {
> > > > > +			*node_id = NodeID;
> > > > > +			*channel_select = ChannelSelect;
> > > > > +		}
> > > > > +	}
> > > > > +
> > > > > +	return CSFound;
> > > > > +}
> > > > 
> > > > this function is probably too large, and also it uses some weird 
> > > > hungarian notation coding style. Please dont do that! It's 
> > > > completely unacceptable.
> > > 
> > > These identifers (or at least, DctSelBaseOffsetLong, which is the 
> > > only one I googled for) come straight out of the AMD "BIOS and 
> > > Kernel Developer's Guide".
> > > 
> > > Sucky though they are, there's value in making the kernel code 
> > > match up with the documentation.
> > 
> > I'm generally resisting patches that hungarinize arch/x86/ (and heck 
> > there's been many attempts ...) but there's some conflicting advice 
> > here. I've Cc:-ed Linus, maybe he has an opinion about this.
> > 
> > My gut reaction would be 'hell no'. There's other, structural 
> > problems with this code too, and doing some saner naming would 
> > mostly be a sed job and would take minimal amount of time. The 
> > naming can still be intuitive. The symbols from the documentation 
> > can perhaps be mentioned in a couple of comments to establish a 
> > mapping.
> 
> I think I disagree.  For those identifiers which map 1:1 with the 
> manufacturer's document, the ugliness involved in exactly copying 
> the manufacturer's chosen identifiers is outweighed by the benefit 
> of exactly copying the manufacturer's chosen identifiers.
> 
> Of course, we don't have to use StinkyIdentifiers anywhere else.  
> And the nice thing about that is that when one reads the code and 
> comes across a StinkyIdentifier, one immeditely knows that it's an 
> AMD-provided thing rather than a Linux-provided thing.
> 
> Zillions of StinkyIdentifiers get merged via this logic.

Andrew, for heaven's sake, please review the patchset - as i did.

The thing is, up to 12/21, the patches look like normal Linux 
patches. (there's problems with them too, but on a different level)

Then do the StinkyIdentifiers show up, in full force:

+static int f10_match_to_this_node(struct amd64_pvt *pvt, int DramRange,
+                               u64 SystemAddr,
+                               int *node_id,
+                               int *channel_select)
+{
+       int CSFound = -1;
+       int NodeID;
+       int HiRangeSelected;
+       u32 IntlvEn, IntlvSel;
+       u32 DramEn;
+       u32 Ilog;
+	u32 HoleOffset, HoleEn;
+       u32 InputAddr, Temp;
+       u32 DctSelBaseAddr, DctSelIntLvAddr;
+       u32 DctSelHi;
+       u32 ChannelSelect;
+       u64 DramBaseLong, DramLimitLong;
+	u64 DctSelBaseOffsetLong, ChannelAddrLong;

Tell me, how is 'SystemAddr' or 'Temp' or 'Ilog' an AMD document 
thing?

I have a much simpler explanation really: someone got really bored 
at converting some code written For Another OS, somewhere in the 
middle - and started plopping Other OS Code into a Linux driver ...

I dont mind the occasional _constant_ that tells us a hw API detail 
in whatever externally dictated style - but this thing stinks 
HeadToToe ... ;-)

	Ingo

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Ingo Molnar]

* Ingo Molnar <mingo@elte.hu> wrote:

> 
> * Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> > On Wed, 29 Apr 2009 21:23:26 +0200
> > Ingo Molnar <mingo@elte.hu> wrote:
> > 
> > > > > > +		if (CSFound >= 0) {
> > > > > > +			*node_id = NodeID;
> > > > > > +			*channel_select = ChannelSelect;
> > > > > > +		}
> > > > > > +	}
> > > > > > +
> > > > > > +	return CSFound;
> > > > > > +}
> > > > > 
> > > > > this function is probably too large, and also it uses some weird 
> > > > > hungarian notation coding style. Please dont do that! It's 
> > > > > completely unacceptable.
> > > > 
> > > > These identifers (or at least, DctSelBaseOffsetLong, which is the 
> > > > only one I googled for) come straight out of the AMD "BIOS and 
> > > > Kernel Developer's Guide".
> > > > 
> > > > Sucky though they are, there's value in making the kernel code 
> > > > match up with the documentation.
> > > 
> > > I'm generally resisting patches that hungarinize arch/x86/ (and heck 
> > > there's been many attempts ...) but there's some conflicting advice 
> > > here. I've Cc:-ed Linus, maybe he has an opinion about this.
> > > 
> > > My gut reaction would be 'hell no'. There's other, structural 
> > > problems with this code too, and doing some saner naming would 
> > > mostly be a sed job and would take minimal amount of time. The 
> > > naming can still be intuitive. The symbols from the documentation 
> > > can perhaps be mentioned in a couple of comments to establish a 
> > > mapping.
> > 
> > I think I disagree.  For those identifiers which map 1:1 with the 
> > manufacturer's document, the ugliness involved in exactly copying 
> > the manufacturer's chosen identifiers is outweighed by the benefit 
> > of exactly copying the manufacturer's chosen identifiers.
> > 
> > Of course, we don't have to use StinkyIdentifiers anywhere else.  
> > And the nice thing about that is that when one reads the code and 
> > comes across a StinkyIdentifier, one immeditely knows that it's an 
> > AMD-provided thing rather than a Linux-provided thing.
> > 
> > Zillions of StinkyIdentifiers get merged via this logic.
> 
> Andrew, for heaven's sake, please review the patchset - as i did.

Let me apologize for this rude reply ... it appears we do agree, i 
just didnt properly read your paragraphs above :-/

What i point out below is precisely what you say is ineligible 
under:

> > Of course, we don't have to use StinkyIdentifiers anywhere else.  

I'd extend that rule to say that StinkyIdentifiers should only be 
used for hw API definitions/constants - macros, enums - not really 
local variable names. The moment they are allowed into local 
variables the stuff below happens.

Thanks,

	Ingo

> 
> The thing is, up to 12/21, the patches look like normal Linux 
> patches. (there's problems with them too, but on a different level)
> 
> Then do the StinkyIdentifiers show up, in full force:
> 
> +static int f10_match_to_this_node(struct amd64_pvt *pvt, int DramRange,
> +                               u64 SystemAddr,
> +                               int *node_id,
> +                               int *channel_select)
> +{
> +       int CSFound = -1;
> +       int NodeID;
> +       int HiRangeSelected;
> +       u32 IntlvEn, IntlvSel;
> +       u32 DramEn;
> +       u32 Ilog;
> +	u32 HoleOffset, HoleEn;
> +       u32 InputAddr, Temp;
> +       u32 DctSelBaseAddr, DctSelIntLvAddr;
> +       u32 DctSelHi;
> +       u32 ChannelSelect;
> +       u64 DramBaseLong, DramLimitLong;
> +	u64 DctSelBaseOffsetLong, ChannelAddrLong;
> 
> Tell me, how is 'SystemAddr' or 'Temp' or 'Ilog' an AMD document 
> thing?
> 
> I have a much simpler explanation really: someone got really bored 
> at converting some code written For Another OS, somewhere in the 
> middle - and started plopping Other OS Code into a Linux driver ...
> 
> I dont mind the occasional _constant_ that tells us a hw API detail 
> in whatever externally dictated style - but this thing stinks 
> HeadToToe ... ;-)
> 
> 	Ingo

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Borislav Petkov]

Hi,

On Wed, Apr 29, 2009 at 10:47:30PM +0200, Ingo Molnar wrote:

[..]

> What i point out below is precisely what you say is ineligible 
> under:
> 
> > > Of course, we don't have to use StinkyIdentifiers anywhere else.  
> 
> I'd extend that rule to say that StinkyIdentifiers should only be 
> used for hw API definitions/constants - macros, enums - not really 
> local variable names. The moment they are allowed into local 
> variables the stuff below happens.

to agree with Andrew, at a certain point in time I thought that having
the same register bit names as in the docs would be preferential when
you look at the docs and what the code does. But Ingo's also quite
right: we can't have "normal kernel coding style" and StinkyIdentifiers
:) in the same source file.

/me locking himself back in the patch creation basement.

-- 
Regards/Gruss,
Boris.

Operating | Advanced Micro Devices GmbH
  System  | Karl-Hammerschmidt-Str. 34, 85609 Dornach b. München, Germany
 Research | Geschäftsführer: Jochen Polster, Thomas M. McCoy, Giuliano Meroni
  Center  | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
  (OSRC)  | Registergericht München, HRB Nr. 43632

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Ingo Molnar]

* Borislav Petkov <borislav.petkov@amd.com> wrote:

> Hi,
> 
> On Wed, Apr 29, 2009 at 10:47:30PM +0200, Ingo Molnar wrote:
> 
> [..]
> 
> > What i point out below is precisely what you say is ineligible 
> > under:
> > 
> > > > Of course, we don't have to use StinkyIdentifiers anywhere else.  
> > 
> > I'd extend that rule to say that StinkyIdentifiers should only be 
> > used for hw API definitions/constants - macros, enums - not really 
> > local variable names. The moment they are allowed into local 
> > variables the stuff below happens.
> 
> to agree with Andrew, at a certain point in time I thought that 
> having the same register bit names as in the docs would be 
> preferential when you look at the docs and what the code does. But 
> Ingo's also quite right: we can't have "normal kernel coding 
> style" and StinkyIdentifiers
> :) in the same source file.
> 
> /me locking himself back in the patch creation basement.

I think you can still cleanly use those identifiers for hardware 
constants, register offsets and similar. But if it shows up in a 
variable (or function) name, it has spread too far IMHO :-)

And it's not like we dont have our own historic mistakes in that 
area, right in the heart of Linux - just type:

  git grep Page mm/*.c

and cringe.

IIRC i might even have added a new method or two to that array of 
CrappyPageAPIs, many years ago. (back in the days when i wrote lot 
of crappy code myself ;-) Oh, PageHighMem() it is.

	Ingo

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Borislav Petkov]

Hi,

On Wed, Apr 29, 2009 at 09:30:31PM +0200, Andi Kleen wrote:
> Borislav Petkov <borislav.petkov@amd.com> writes:
> 
> > Hi,
> >
> > thanks to all reviewers of the previous submission, here is the second
> > version of this series.
> 
> The classic problem of the previous versions of these patches was that
> they consume the same error registers (even if using pci config versus
> msrs as access methods) as the kernel machine check poll/threshold
> interrupt code.  And with two logging agents racing on the same
> registers you will always get junk results. Typically with threshold 
> enabled the mce code wins the race. I suspect this patchkit has
> exactly the same fundamental design problem. EDAC really is not
> particularly fitting for integrated memory controllers that report
> their errors using standard machine check events.

ok, how about we remove tha MSR/PCI cfg space reading bits and leave
that task solely to the mce core. Then, iff you have edac turned on in
Kconfig, mce code delivers needed error info to edac which, in turn,
goes and decodes the error/does the mapping to DIMM blocks/supplies DRAM
error injection facility for testing purposes and similar things. That
way you have both and they don't overlap in functionality.

By the way, I think there's a similar attempt/proposal of letting mce
and edac talk to each other from Red Hat so I think this could be a
viable thing to try.

> -Andi (who thinks all of this decoding should be in user space anyways)

Think of a big data center with a thousands of 2,4,8 socket blades
and the admin collecting mce output and running around decoding the
errors on his workstation. Even worse, the blades have different DIMM
configurations due to hw upgrades/newer machines. I'd much rather have
the complete decoding done in kernel, where all the information needed
for proper decoding is present and with the error landing in syslog or
some other monitored buffer instead of reconstructing it in userspace.

Thanks.

-- 
Regards/Gruss,
Boris.

Operating | Advanced Micro Devices GmbH
  System  | Karl-Hammerschmidt-Str. 34, 85609 Dornach b. München, Germany
 Research | Geschäftsführer: Jochen Polster, Thomas M. McCoy, Giuliano Meroni
  Center  | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
  (OSRC)  | Registergericht München, HRB Nr. 43632

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Ingo Molnar]

* Borislav Petkov <borislav.petkov@amd.com> wrote:

> > -Andi (who thinks all of this decoding should be in user space 
> > anyways)
> 
> Think of a big data center with a thousands of 2,4,8 socket blades 
> and the admin collecting mce output and running around decoding 
> the errors on his workstation. Even worse, the blades have 
> different DIMM configurations due to hw upgrades/newer machines. 
> I'd much rather have the complete decoding done in kernel, where 
> all the information needed for proper decoding is present and with 
> the error landing in syslog or some other monitored buffer instead 
> of reconstructing it in userspace.

Yes, this aspect of the design is correct. The MCE code is seriously 
mis-designed that way, lets not repeat it for EDAC :)

	Ingo

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Andi Kleen]

> ok, how about we remove tha MSR/PCI cfg space reading bits and leave
> that task solely to the mce core. Then, iff you have edac turned on in

That's the minimum fix, but even then the patchkit does a lot of 
things, not necessarily all needs to be together.

> Kconfig, mce code delivers needed error info to edac which, in turn,
> goes and decodes the error/does the mapping to DIMM blocks/supplies DRAM
> error injection facility for testing purposes and similar things. That
> way you have both and they don't overlap in functionality.

You can do that, but it's redundant because mcelog can do this
this already. I had some conversations with existing EDAC users
recently and they seem to only care about the resulting output,
so just querying from mcelog is fine.
The only issue is that mcelog needs to get the DIMM data. In many
cases it can do so from SMBIOS output, if not a suitable interface
would need to be provided by the kernel.

> By the way, I think there's a similar attempt/proposal of letting mce
> and edac talk to each other from Red Hat so I think this could be a

There was a fairly dubious patch floating around I think, but it
had a couple of problems.

> > -Andi (who thinks all of this decoding should be in user space anyways)
> 
> Think of a big data center with a thousands of 2,4,8 socket blades
> and the admin collecting mce output and running around decoding the

Nobody said anything about admins decoding on their workstation.

Corrected events (which are the 90+% case) get decoded in user space on the 
same system.  Uncorrected events get decoded after the reboot. Both happens 
automatically and transparently.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only.

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Aristeu Rozanski]

(adding Ben Woodard, Mauro C. Chehab to Cc list)

> > Kconfig, mce code delivers needed error info to edac which, in turn,
> > goes and decodes the error/does the mapping to DIMM blocks/supplies DRAM
> > error injection facility for testing purposes and similar things. That
> > way you have both and they don't overlap in functionality.
> You can do that, but it's redundant because mcelog can do this
> this already. I had some conversations with existing EDAC users
> recently and they seem to only care about the resulting output,
> so just querying from mcelog is fine.
what about using the same EDAC interface? for lots of memory controllers, even
in other architectures than x86, EDAC interface is available. sounds
inconsistent to force users to have to handle special cases on their scripts
just because _optional_ sharing of error information from mce code is not
available.
how about SW/HW scrubbing?

> The only issue is that mcelog needs to get the DIMM data. In many
> cases it can do so from SMBIOS output, if not a suitable interface
> would need to be provided by the kernel.
that can be done already in EDAC and in this driver

> > By the way, I think there's a similar attempt/proposal of letting mce
> > and edac talk to each other from Red Hat so I think this could be a
> There was a fairly dubious patch floating around I think, but it
> had a couple of problems.
and what if those problems are solved? a patch like that would make possible
to have EDAC support for both AMD64 and Nehalem and wouldn't hurt the
performance of people who choose not to use EDAC.

-- 
Aristeu

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Mauro Carvalho Chehab]

On Thu, 30 Apr 2009, Andi Kleen wrote:

>> Kconfig, mce code delivers needed error info to edac which, in turn,
>> goes and decodes the error/does the mapping to DIMM blocks/supplies DRAM
>> error injection facility for testing purposes and similar things. That
>> way you have both and they don't overlap in functionality.
>
> You can do that, but it's redundant because mcelog can do this
> this already. I had some conversations with existing EDAC users
> recently and they seem to only care about the resulting output,
> so just querying from mcelog is fine.
> The only issue is that mcelog needs to get the DIMM data. In many
> cases it can do so from SMBIOS output, if not a suitable interface
> would need to be provided by the kernel.

>From what I've heard from the existing EDAC users, they have several 
concerns that mcelog could be viable replacement to their EDAC usage, due 
to performance issues, including the need of accessing SMBIOS in order to 
get such information.

Also, EDAC interface is already stablished, and, as pointed by Doug, it is 
very useful on cluster environments, where memory failures is a big issue 
and need to be solved as soon as possible.

EDAC solves this issue very well and works on a wider range of designs 
than mcelog. So, there's no reason to deprecate it or to reject patches 
adding EDAC interfaces to other chips.

On the other hand, mcelog is also useful on different scenarios. So, they 
are not competing technologies, but complementary ones.

So, assuming that both EDAC and mcelog are needed, the proper design for 
those chipsets where the memory controller is integrated with other log 
functions (like AMD64 and Nethalem) seem to build an unique kernel layer 
that retrieves the error logs from the harware and allows access to the 
same data via both mcelog and EDAC userspace API's.

Cheers,
Mauro

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Borislav Petkov]

Hi all,

I thought I should summarize some points on the design direction, which,
in my impression are what all the parties involved more or less would
agree on:

* mce interrupt delivers, if EDAC compiled in, the required batch of
dumped MSRs to chipset/CPU specific EDAC driver. Aristeu, I guess you
guys have some code on that, right?

* EDAC module decodes MCE info, computes the DIMM label out of
that. Additional hw setup stuff like memory hoisting/interleaving,
ganged/unganged MC mode, testing infrastructure like DRAM error
injection and all that pertaining to the specific hw is handled by the
EDAC module.

* If EDAC not enabled, mce operates as before.

Comments/suggestions?

Hit me! :)

-- 
Regards/Gruss,
    Boris.

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Ingo Molnar]

* Andi Kleen <andi@firstfloor.org> wrote:

> > Kconfig, mce code delivers needed error info to edac which, in 
> > turn, goes and decodes the error/does the mapping to DIMM 
> > blocks/supplies DRAM error injection facility for testing 
> > purposes and similar things. That way you have both and they 
> > don't overlap in functionality.
> 
> You can do that, but it's redundant because mcelog can do this 
> this already. [...]

The thing is, when we took up x86 maintenance i had a good look at 
the MCE situation, i checked both the kernel and the user-space 
side.

The kernel side MCE code was in pretty bad shape to begin with, but 
mcelog (the user-space tool) is a big stinking pile of poo on every 
level.

It's one of the worst piece of kernel related code i ever saw. I 
think you wrote all of it, and you should be ashamed of that code, 
and you should be ashamed of the design and you should be ashamed of 
the concept.

It even came with its own 'database' code: mcelog*/db.[ch] is 600+ 
lines of needless code instead of obvious library use. It's NIH and 
self-serving complexity all over.

And the thing is, mcelog/mcedecode never really _did_ anything real 
an useful, other than to:

 1) Confuse kernel users who see a fatal MCE panic, with cryptic, 
    quirky codes, who write that down on paper, then run it through 
    the user-space tool - just to see a piece of information the 
    kernel could have provided already. (if they didnt make any 
    mistakes while writing down the codes)

 2) Decode a quirky, binary MCE record and combine it with DMI data.
    (which the kernel can and should do just fine.)

Yes, i know about tolerant=3 and certain people/companies opting to 
ignore MCE fatality levels and live dangerously (and i also know 
about non-fatal reporting and correction extensions in hw) - but for 
99.999% of the Linux users the whole thing is just needless 
complexity today, that does not offer anything valuable.

And that is really what happens when code is misdesigned and the 
wrong pieces of code are pushed to user-space: a crappy, limited ABI 
and an under-maintained, big pile of junk user-space kit.

The obvious truth is that hardware faults have to be caught, decoded 
and optionally handled by the kernel.

The EDAC code at least has a sane design: it realizes that hardware 
faults _must_ be fully known, decoded and potentially handled in the 
kernel.

Piggyback-ing to user-space is plain idiotic and not defensible. So 
if a piece of hardware capability is handled by the EDAC code, the 
x86 MCE code will step aside and will stay the heck out of that 
business. At least until the two concepts are merged into some sane 
kernel hardware fault logging and handling framework.

And Andi, until you dont grasp such _basic_ design concepts, you 
have no business writing such code really. You should stay the heck 
away from it and you should stop 'advising' people who made the 
right calls while you messed up. It is mcelog that is crap, not the 
EDAC code.

	Ingo

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Aristeu Rozanski]

> I thought I should summarize some points on the design direction, which,
> in my impression are what all the parties involved more or less would
> agree on:
> 
> * mce interrupt delivers, if EDAC compiled in, the required batch of
> dumped MSRs to chipset/CPU specific EDAC driver. Aristeu, I guess you
> guys have some code on that, right?
it's in really early stage and like Andi said, it needs work.

> * EDAC module decodes MCE info, computes the DIMM label out of
> that. Additional hw setup stuff like memory hoisting/interleaving,
> ganged/unganged MC mode, testing infrastructure like DRAM error
> injection and all that pertaining to the specific hw is handled by the
> EDAC module.
> 
> * If EDAC not enabled, mce operates as before.
agreed.

-- 
Aristeu

Re: [PATCH 01/21] x86: add methods for writing of an MSR on several CPUs

[Borislav Petkov]

Hi,

(adding Mauro to CC)

On Wed, Apr 29, 2009 at 10:39:28AM -0700, H. Peter Anvin wrote:
> Borislav Petkov wrote:
> > +}
> > +
> > +/* rdmsr on a bunch of cpus
> > + *
> > + * @mask:	which cpus
> > + * @msr_no:	which MSR
> > + * @l:		array of the lower 32-bit MSR values
> > + * @h:		array of the higher 32-bit MSR values
> > + *
> 
> Urk!
> 
> Please, for the love of God, no!
> 
> Make it an array of u64s or (equivalently!) an array of (l,h)
> structures, not separate arrays for the halves of the register.

how about something like the following then? If there's agreement I could
convert all users to struct msr later.

@Mauro: here's 1/21.

--
From: Borislav Petkov <borislav.petkov@amd.com>
Date: Wed, 29 Apr 2009 15:20:11 +0200
Subject: [PATCH] x86: add methods for writing of an MSR on several CPUs

Add a struct representing a 64bit MSR pair consisting of a low and high
register part.

Also, rename msr-on-cpu.c to msr.c accordingly.

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 arch/x86/include/asm/msr.h |   16 +++++
 arch/x86/lib/Makefile      |    2 +-
 arch/x86/lib/msr-on-cpu.c  |   97 ----------------------------
 arch/x86/lib/msr.c         |  152 ++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 169 insertions(+), 98 deletions(-)
 delete mode 100644 arch/x86/lib/msr-on-cpu.c
 create mode 100644 arch/x86/lib/msr.c

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index 638bf62..7f117a6 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -2,6 +2,7 @@
 #define _ASM_X86_MSR_H
 
 #include <asm/msr-index.h>
+#include <asm/cpumask.h>
 
 #ifndef __ASSEMBLY__
 # include <linux/types.h>
@@ -13,6 +14,11 @@
 #include <asm/asm.h>
 #include <asm/errno.h>
 
+struct msr {
+	u32 l;
+	u32 h;
+};
+
 static inline unsigned long long native_read_tscp(unsigned int *aux)
 {
 	unsigned long low, high;
@@ -216,6 +222,8 @@ do {                                                            \
 #ifdef CONFIG_SMP
 int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
+int rdmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs);
+int wrmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs);
 int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
 #else  /*  CONFIG_SMP  */
@@ -229,6 +237,14 @@ static inline int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 	wrmsr(msr_no, l, h);
 	return 0;
 }
+static inline int rdmsr_on_cpus(const cpumask_t *m, u32 msr_no, struct msr **msrs)
+{
+	return rdmsr_on_cpu(msr_no, &(msrs[0].l), &(msrs[0].h));
+}
+static inline int wrmsr_on_cpus(const cpumask_t *m, u32 msr_no, struct msr **msrs)
+{
+	return wrmsr_on_cpu(msr_no, msrs[0].l, msrs[0].h);
+}
 static inline int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no,
 				    u32 *l, u32 *h)
 {
diff --git a/arch/x86/lib/Makefile b/arch/x86/lib/Makefile
index 55e11aa..f9d3563 100644
--- a/arch/x86/lib/Makefile
+++ b/arch/x86/lib/Makefile
@@ -2,7 +2,7 @@
 # Makefile for x86 specific library files.
 #
 
-obj-$(CONFIG_SMP) := msr-on-cpu.o
+obj-$(CONFIG_SMP) := msr.o
 
 lib-y := delay.o
 lib-y += thunk_$(BITS).o
diff --git a/arch/x86/lib/msr-on-cpu.c b/arch/x86/lib/msr-on-cpu.c
deleted file mode 100644
index 321cf72..0000000
--- a/arch/x86/lib/msr-on-cpu.c
+++ /dev/null
@@ -1,97 +0,0 @@
-#include <linux/module.h>
-#include <linux/preempt.h>
-#include <linux/smp.h>
-#include <asm/msr.h>
-
-struct msr_info {
-	u32 msr_no;
-	u32 l, h;
-	int err;
-};
-
-static void __rdmsr_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rdmsr(rv->msr_no, rv->l, rv->h);
-}
-
-static void __wrmsr_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	wrmsr(rv->msr_no, rv->l, rv->h);
-}
-
-int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
-	*l = rv.l;
-	*h = rv.h;
-
-	return err;
-}
-
-int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	rv.l = l;
-	rv.h = h;
-	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
-
-	return err;
-}
-
-/* These "safe" variants are slower and should be used when the target MSR
-   may not actually exist. */
-static void __rdmsr_safe_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h);
-}
-
-static void __wrmsr_safe_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h);
-}
-
-int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
-	*l = rv.l;
-	*h = rv.h;
-
-	return err ? err : rv.err;
-}
-
-int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	rv.l = l;
-	rv.h = h;
-	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
-
-	return err ? err : rv.err;
-}
-
-EXPORT_SYMBOL(rdmsr_on_cpu);
-EXPORT_SYMBOL(wrmsr_on_cpu);
-EXPORT_SYMBOL(rdmsr_safe_on_cpu);
-EXPORT_SYMBOL(wrmsr_safe_on_cpu);
diff --git a/arch/x86/lib/msr.c b/arch/x86/lib/msr.c
new file mode 100644
index 0000000..0ba5402
--- /dev/null
+++ b/arch/x86/lib/msr.c
@@ -0,0 +1,152 @@
+#include <linux/module.h>
+#include <linux/preempt.h>
+#include <linux/smp.h>
+#include <asm/msr.h>
+
+struct msr_info {
+	u32 msr_no;
+	u32 l, h;
+	int err;
+};
+
+static void __rdmsr_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rdmsr(rv->msr_no, rv->l, rv->h);
+}
+
+static void __wrmsr_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	wrmsr(rv->msr_no, rv->l, rv->h);
+}
+
+int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
+	*l = rv.l;
+	*h = rv.h;
+
+	return err;
+}
+
+int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	rv.l = l;
+	rv.h = h;
+	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
+
+	return err;
+}
+
+/* rdmsr on a bunch of CPUs
+ *
+ * @mask:	which CPUs
+ * @msr_no:	which MSR
+ * @msrs:	array of MSR values
+ *
+ * Returns:
+ * 0 - success
+ * <0 - read failed on at least one CPU (latter in the mask)
+ */
+int rdmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs)
+{
+	struct msr reg;
+	int cpu, tmp, err = 0;
+	int off = cpumask_first(mask);
+
+	for_each_cpu(cpu, mask) {
+		reg = msrs[cpu - off];
+
+		tmp = rdmsr_on_cpu(cpu, msr_no, &reg.l, &reg.h);
+		if (tmp)
+			err = tmp;
+	}
+	return err;
+}
+
+/*
+ * wrmsr of a bunch of CPUs
+ *
+ * @mask:	which CPUs
+ * @msr_no:	which MSR
+ * @msrs:	array of MSR values
+  *
+ * Returns:
+ * 0 - success
+ * <0 - write failed on at least one CPU (latter in the mask)
+ */
+int wrmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs)
+{
+	struct msr reg;
+	int cpu, tmp, err = 0;
+	int off = cpumask_first(mask);
+
+	for_each_cpu(cpu, mask) {
+		reg = msrs[cpu - off];
+
+		tmp = wrmsr_on_cpu(cpu, msr_no, reg.l, reg.h);
+		if (tmp)
+			err = tmp;
+	}
+	return err;
+}
+
+/* These "safe" variants are slower and should be used when the target MSR
+   may not actually exist. */
+static void __rdmsr_safe_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h);
+}
+
+static void __wrmsr_safe_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h);
+}
+
+int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
+	*l = rv.l;
+	*h = rv.h;
+
+	return err ? err : rv.err;
+}
+
+int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	rv.l = l;
+	rv.h = h;
+	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
+
+	return err ? err : rv.err;
+}
+
+EXPORT_SYMBOL(rdmsr_on_cpu);
+EXPORT_SYMBOL(wrmsr_on_cpu);
+EXPORT_SYMBOL(rdmsr_on_cpus);
+EXPORT_SYMBOL(wrmsr_on_cpus);
+EXPORT_SYMBOL(rdmsr_safe_on_cpu);
+EXPORT_SYMBOL(wrmsr_safe_on_cpu);
-- 
1.6.2.4


-- 
Regards/Gruss,
Boris.

Operating | Advanced Micro Devices GmbH
  System  | Karl-Hammerschmidt-Str. 34, 85609 Dornach b. München, Germany
 Research | Geschäftsführer: Thomas M. McCoy, Giuliano Meroni
  Center  | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
  (OSRC)  | Registergericht München, HRB Nr. 43632

Re: [PATCH 01/21] x86: add methods for writing of an MSR on several CPUs

[H. Peter Anvin]

Borislav Petkov wrote:
>>
>> Please, for the love of God, no!
>>
>> Make it an array of u64s or (equivalently!) an array of (l,h)
>> structures, not separate arrays for the halves of the register.
> 
> how about something like the following then? If there's agreement I could
> convert all users to struct msr later.
> 

I personally would prefer if you just used an array of u64s.  The whole
l/h split for MSRs was a mistake in the first place.  However, if not,
at least make it a union; using something like:

union msr {
	u32 l[2];
	u64 q;
};

Or fancier, using gcc's anonymous structs/unions:

struct msr {
	union {
		struct {
			u32 l, h;
		};
		u64 q;
	};
};

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: [PATCH 01/21] x86: add methods for writing of an MSR on several CPUs

[Borislav Petkov]

On Mon, May 04, 2009 at 10:25:12AM -0700, H. Peter Anvin wrote:
> Borislav Petkov wrote:
> >>
> >> Please, for the love of God, no!
> >>
> >> Make it an array of u64s or (equivalently!) an array of (l,h)
> >> structures, not separate arrays for the halves of the register.
> > 
> > how about something like the following then? If there's agreement I could
> > convert all users to struct msr later.
> > 
> 
> I personally would prefer if you just used an array of u64s.  The whole
> l/h split for MSRs was a mistake in the first place.

... on the other hand, the two u32s kinda resemble more the EDX:EAX
register pair of rdmsr/wrmsr, hm...

[..]

> Or fancier, using gcc's anonymous structs/unions:
> 
> struct msr {
> 	union {
> 		struct {
> 			u32 l, h;
> 		};
> 		u64 q;
> 	};
> };

yeah, that sounds good:

--
From: Borislav Petkov <borislav.petkov@amd.com>
Date: Wed, 29 Apr 2009 15:20:11 +0200
Subject: [PATCH 1/2] x86: add methods for writing of an MSR on several CPUs

Add a struct representing a 64bit MSR pair consisting of a low and high
register part.

Also, rename msr-on-cpu.c to msr.c accordingly.

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
---
 arch/x86/include/asm/msr.h |   21 ++++++
 arch/x86/lib/Makefile      |    2 +-
 arch/x86/lib/msr-on-cpu.c  |   97 ----------------------------
 arch/x86/lib/msr.c         |  152 ++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 174 insertions(+), 98 deletions(-)
 delete mode 100644 arch/x86/lib/msr-on-cpu.c
 create mode 100644 arch/x86/lib/msr.c

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index 638bf62..624c895 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -2,6 +2,7 @@
 #define _ASM_X86_MSR_H
 
 #include <asm/msr-index.h>
+#include <asm/cpumask.h>
 
 #ifndef __ASSEMBLY__
 # include <linux/types.h>
@@ -13,6 +14,16 @@
 #include <asm/asm.h>
 #include <asm/errno.h>
 
+struct msr {
+	union {
+		struct {
+			u32 l;
+			u32 h;
+		};
+		u64 q;
+	};
+};
+
 static inline unsigned long long native_read_tscp(unsigned int *aux)
 {
 	unsigned long low, high;
@@ -216,6 +227,8 @@ do {                                                            \
 #ifdef CONFIG_SMP
 int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
+int rdmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs);
+int wrmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs);
 int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
 int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
 #else  /*  CONFIG_SMP  */
@@ -229,6 +242,14 @@ static inline int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
 	wrmsr(msr_no, l, h);
 	return 0;
 }
+static inline int rdmsr_on_cpus(const cpumask_t *m, u32 msr_no, struct msr **msrs)
+{
+	return rdmsr_on_cpu(msr_no, &(msrs[0].l), &(msrs[0].h));
+}
+static inline int wrmsr_on_cpus(const cpumask_t *m, u32 msr_no, struct msr **msrs)
+{
+	return wrmsr_on_cpu(msr_no, msrs[0].l, msrs[0].h);
+}
 static inline int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no,
 				    u32 *l, u32 *h)
 {
diff --git a/arch/x86/lib/Makefile b/arch/x86/lib/Makefile
index 55e11aa..f9d3563 100644
--- a/arch/x86/lib/Makefile
+++ b/arch/x86/lib/Makefile
@@ -2,7 +2,7 @@
 # Makefile for x86 specific library files.
 #
 
-obj-$(CONFIG_SMP) := msr-on-cpu.o
+obj-$(CONFIG_SMP) := msr.o
 
 lib-y := delay.o
 lib-y += thunk_$(BITS).o
diff --git a/arch/x86/lib/msr-on-cpu.c b/arch/x86/lib/msr-on-cpu.c
deleted file mode 100644
index 321cf72..0000000
--- a/arch/x86/lib/msr-on-cpu.c
+++ /dev/null
@@ -1,97 +0,0 @@
-#include <linux/module.h>
-#include <linux/preempt.h>
-#include <linux/smp.h>
-#include <asm/msr.h>
-
-struct msr_info {
-	u32 msr_no;
-	u32 l, h;
-	int err;
-};
-
-static void __rdmsr_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rdmsr(rv->msr_no, rv->l, rv->h);
-}
-
-static void __wrmsr_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	wrmsr(rv->msr_no, rv->l, rv->h);
-}
-
-int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
-	*l = rv.l;
-	*h = rv.h;
-
-	return err;
-}
-
-int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	rv.l = l;
-	rv.h = h;
-	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
-
-	return err;
-}
-
-/* These "safe" variants are slower and should be used when the target MSR
-   may not actually exist. */
-static void __rdmsr_safe_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h);
-}
-
-static void __wrmsr_safe_on_cpu(void *info)
-{
-	struct msr_info *rv = info;
-
-	rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h);
-}
-
-int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
-	*l = rv.l;
-	*h = rv.h;
-
-	return err ? err : rv.err;
-}
-
-int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
-{
-	int err;
-	struct msr_info rv;
-
-	rv.msr_no = msr_no;
-	rv.l = l;
-	rv.h = h;
-	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
-
-	return err ? err : rv.err;
-}
-
-EXPORT_SYMBOL(rdmsr_on_cpu);
-EXPORT_SYMBOL(wrmsr_on_cpu);
-EXPORT_SYMBOL(rdmsr_safe_on_cpu);
-EXPORT_SYMBOL(wrmsr_safe_on_cpu);
diff --git a/arch/x86/lib/msr.c b/arch/x86/lib/msr.c
new file mode 100644
index 0000000..0ba5402
--- /dev/null
+++ b/arch/x86/lib/msr.c
@@ -0,0 +1,152 @@
+#include <linux/module.h>
+#include <linux/preempt.h>
+#include <linux/smp.h>
+#include <asm/msr.h>
+
+struct msr_info {
+	u32 msr_no;
+	u32 l, h;
+	int err;
+};
+
+static void __rdmsr_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rdmsr(rv->msr_no, rv->l, rv->h);
+}
+
+static void __wrmsr_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	wrmsr(rv->msr_no, rv->l, rv->h);
+}
+
+int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_on_cpu, &rv, 1);
+	*l = rv.l;
+	*h = rv.h;
+
+	return err;
+}
+
+int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	rv.l = l;
+	rv.h = h;
+	err = smp_call_function_single(cpu, __wrmsr_on_cpu, &rv, 1);
+
+	return err;
+}
+
+/* rdmsr on a bunch of CPUs
+ *
+ * @mask:	which CPUs
+ * @msr_no:	which MSR
+ * @msrs:	array of MSR values
+ *
+ * Returns:
+ * 0 - success
+ * <0 - read failed on at least one CPU (latter in the mask)
+ */
+int rdmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs)
+{
+	struct msr reg;
+	int cpu, tmp, err = 0;
+	int off = cpumask_first(mask);
+
+	for_each_cpu(cpu, mask) {
+		reg = msrs[cpu - off];
+
+		tmp = rdmsr_on_cpu(cpu, msr_no, &reg.l, &reg.h);
+		if (tmp)
+			err = tmp;
+	}
+	return err;
+}
+
+/*
+ * wrmsr of a bunch of CPUs
+ *
+ * @mask:	which CPUs
+ * @msr_no:	which MSR
+ * @msrs:	array of MSR values
+  *
+ * Returns:
+ * 0 - success
+ * <0 - write failed on at least one CPU (latter in the mask)
+ */
+int wrmsr_on_cpus(const cpumask_t *mask, u32 msr_no, struct msr *msrs)
+{
+	struct msr reg;
+	int cpu, tmp, err = 0;
+	int off = cpumask_first(mask);
+
+	for_each_cpu(cpu, mask) {
+		reg = msrs[cpu - off];
+
+		tmp = wrmsr_on_cpu(cpu, msr_no, reg.l, reg.h);
+		if (tmp)
+			err = tmp;
+	}
+	return err;
+}
+
+/* These "safe" variants are slower and should be used when the target MSR
+   may not actually exist. */
+static void __rdmsr_safe_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rv->err = rdmsr_safe(rv->msr_no, &rv->l, &rv->h);
+}
+
+static void __wrmsr_safe_on_cpu(void *info)
+{
+	struct msr_info *rv = info;
+
+	rv->err = wrmsr_safe(rv->msr_no, rv->l, rv->h);
+}
+
+int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	err = smp_call_function_single(cpu, __rdmsr_safe_on_cpu, &rv, 1);
+	*l = rv.l;
+	*h = rv.h;
+
+	return err ? err : rv.err;
+}
+
+int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h)
+{
+	int err;
+	struct msr_info rv;
+
+	rv.msr_no = msr_no;
+	rv.l = l;
+	rv.h = h;
+	err = smp_call_function_single(cpu, __wrmsr_safe_on_cpu, &rv, 1);
+
+	return err ? err : rv.err;
+}
+
+EXPORT_SYMBOL(rdmsr_on_cpu);
+EXPORT_SYMBOL(wrmsr_on_cpu);
+EXPORT_SYMBOL(rdmsr_on_cpus);
+EXPORT_SYMBOL(wrmsr_on_cpus);
+EXPORT_SYMBOL(rdmsr_safe_on_cpu);
+EXPORT_SYMBOL(wrmsr_safe_on_cpu);
-- 
1.6.2.4


-- 
Regards/Gruss,
Boris.

Operating | Advanced Micro Devices GmbH
  System  | Karl-Hammerschmidt-Str. 34, 85609 Dornach b. München, Germany
 Research | Geschäftsführer: Thomas M. McCoy, Giuliano Meroni
  Center  | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
  (OSRC)  | Registergericht München, HRB Nr. 43632

Re: [PATCH 03/21] amd64_edac: add driver structs

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>  drivers/edac/amd64_edac.c |  264 +++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 264 insertions(+), 0 deletions(-)
> +/*
> + * amd64 family unique informatoin
> + */
>   

There's a typo above:
    s/informatoin/information/

Except for that, the patch looks sane to my eyes.

Reviewed-by: Mauro Carvalho Chehab

Re: [PATCH 01/21] x86: add methods for writing of an MSR on several CPUs

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> On Mon, May 04, 2009 at 10:25:12AM -0700, H. Peter Anvin wrote:
>   
>> Borislav Petkov wrote:
>>     
>>>> Please, for the love of God, no!
>>>>
>>>> Make it an array of u64s or (equivalently!) an array of (l,h)
>>>> structures, not separate arrays for the halves of the register.
>>>>         
>>> how about something like the following then? If there's agreement I could
>>> convert all users to struct msr later.
>>>
>>>       
>> I personally would prefer if you just used an array of u64s.  The whole
>> l/h split for MSRs was a mistake in the first place.
>>     
>
> ... on the other hand, the two u32s kinda resemble more the EDX:EAX
> register pair of rdmsr/wrmsr, hm...
>
> [..]
>
>   
>> Or fancier, using gcc's anonymous structs/unions:
>>
>> struct msr {
>> 	union {
>> 		struct {
>> 			u32 l, h;
>> 		};
>> 		u64 q;
>> 	};
>> };
>>     
>
> yeah, that sounds good:
>
> --
> From: Borislav Petkov <borislav.petkov@amd.com>
> Date: Wed, 29 Apr 2009 15:20:11 +0200
> Subject: [PATCH 1/2] x86: add methods for writing of an MSR on several CPUs
>
> Add a struct representing a 64bit MSR pair consisting of a low and high
> register part.
>
> Also, rename msr-on-cpu.c to msr.c accordingly.
>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>
>   
That sounds good also to me.

Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 02/21] amd64_edac: add PCI config register defines

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>Borislav Petkov escreveu:
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 04/21] amd64_edac: add memory scrubber interface

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 05/21] amd64_edac: add sys addr to memory controller mapping helpers

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 06/21] amd64_edac: add functionality to compute the DRAM hole

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 07/21] amd64_edac: add DRAM address type conversion facilities

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 08/21] amd64_edac: add helper to dump relevant registers

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 09/21] amd64_edac: assign DRAM chip select base and mask in a family-specific way

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> +	for (cs = 0; cs < CHIPSELECT_COUNT; cs++) {
> +		reg = K8_DCSB0 + (cs * 4);
> +		err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
> +						&pvt->dcsb0[cs]);
> +		if (err != 0)
> +			debugf0("%s() Reading K8_DCSB0[%d] failed\n",
> +				__func__, cs);
> +
> +		debugf0("  DCSB0[%d]=0x%08x reg: F2x%x\n",
> +			cs, pvt->dcsb0[cs], reg);
>   

Hmm... I suspect that there's a missing else before the debugf0(). If 
you got an error while reading it, you shouldn't be showing the results. 
> +
> +		/* If DCT are NOT ganged, then read in DCT1's base */
> +		if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) {
> +			reg = F10_DCSB1 + (cs * 4);
> +			err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
> +							&pvt->dcsb1[cs]);
> +			if (err != 0)
> +				debugf0("%s() Reading F10_DCSB1[%d] failed\n",
> +					__func__, cs);
> +			debugf0("  DCSB1[%d]=0x%08x reg: F2x%x\n",
> +				cs, pvt->dcsb1[cs], reg);
>   
The same issue here: if you got an error while reading it, you shouldn't 
be showing the results. 

Cheers,
Mauro.

Re: [PATCH 10/21] amd64_edac: add k8-specific methods

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 11/21] amd64_edac: add f10-and-later methods-p1

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>   

Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Re: [PATCH 12/21] amd64_edac: add f10-and-later methods-p2

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> +/*
> + * f10_lookup_addr_in_dct
> + *
> + *	Iterate over the DRAM DCT "base" and "mask" register looking for
> + *	a SystemAddr match on the specified 'ChannelSelect' and 'NodeID'
> + *
> + * Return:
> + *	-1  NOT FOUND
> + *	0..csrow = Chip-Select Row
> + */
> +static int f10_lookup_addr_in_dct(u32 InputAddr, u32 NodeID, u32 ChannelSelect)
> +{
> +	struct mem_ctl_info *mci;
> +	struct amd64_pvt *pvt;
> +	u32 CSBase, CSMask;
> +	int CSFound = -1;
> +	int csrow;
> +
> +	mci = mci_lookup[NodeID];
> +	if (!mci)
> +		return CSFound;
>   

It is better to use a standard error code like -EINVAL instead of using 
-1 for error.


Cheers,
Mauro.

Re: [PATCH 13/21] amd64_edac: add f10-and-later methods-p3

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>  drivers/edac/amd64_edac.c |  318 +++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 318 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index fe2342c..84075c0 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -2726,4 +2726,322 @@ static int f10_lookup_addr_in_dct(u32 InputAddr, u32 NodeID, u32 ChannelSelect)
>  	return CSFound;
>  }
>  
> +/*
> + * f10_match_to_this_node
> + *
> + * For a given 'DramRange' value, check if 'SystemAddr' fall within this value
> + */
> +static int f10_match_to_this_node(struct amd64_pvt *pvt, int DramRange,
> +				u64 SystemAddr,
> +				int *node_id,
> +				int *channel_select)
> +{
> +	int CSFound = -1;
>   

As in the previous patch, please use a standard error code, instead of -1.

> +static int f10_translate_sysaddr_to_CS(struct amd64_pvt *pvt,
> +			u64 SysAddr,
> +			int *node,
> +			int *chanSel)
> +{
> +	int DramRange;
> +	int CSFound = -1;
>   
Same here.

Cheers,
Mauro

Re: [PATCH 14/21] amd64_edac: add per-family descriptors

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Cheers,
Mauro.

Re: [PATCH 15/21] amd64_edac: add ECC chipkill syndrome mapping table

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Cheers,
Mauro

Re: [PATCH 16/21] amd64_edac: add error decoding logic

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Cheers,
Mauro

Re: [PATCH 17/21] amd64_edac: add EDAC core-related initializers

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   
Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Cheers,
Mauro.

Re: [PATCH 18/21] amd64_edac: add ECC reporting initializers

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   

Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Cheers,
Mauro.

Re: [PATCH 20/21] amd64_edac: add DRAM error injection logic using sysfs

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
> ---
>  drivers/edac/amd64_edac.c |  287 +++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 287 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index b1a7e8c..4d1076f 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -4621,3 +4621,290 @@ static ssize_t amd64_hole_show(struct mem_ctl_info *mci, char *data)
>  
>  #endif	/* DEBUG */
>  
> +#ifdef CONFIG_EDAC_AMD64_OPTERON_ERROR_INJECTION
> +/*
> + * amd64_inject_section_store
> + *
> + *	accept and store error injection section value
> + *	range: 0..3
> + *		value refers to one of 4 16-byte sections
> + *		within a 64-byte cacheline
> + */
> +static ssize_t amd64_inject_section_store(struct mem_ctl_info *mci,
> +					const char *data, size_t count)
> +{
> +	struct amd64_pvt *pvt = mci->pvt_info;
> +	unsigned long value;
> +	int rc;
> +
> +	rc = strict_strtoul(data, 10, &value);
> +	if (rc != -EINVAL) {
> +
> +		/* save the 16-byte cache section */
> +		pvt->injection.section = (u32) value;
> +
> +		return count;
> +	}
> +	return 0;
> +}
> +
> +/*
> + * amd64_inject_word_store
> + *
> + *	accept and store error injection word value
> + *	range: 0..8
> + *		value refers to one of 9 16-bit word of the 16-byte section
> + *		128-bit + ECC bits
> + */
> +static ssize_t amd64_inject_word_store(struct mem_ctl_info *mci,
> +					const char *data, size_t count)
> +{
> +	struct amd64_pvt *pvt = mci->pvt_info;
> +	unsigned long value;
> +	int rc;
> +
> +	rc = strict_strtoul(data, 10, &value);
> +	if (rc != -EINVAL) {
> +
> +		/* save the 16-bit word */
> +		value = (value <= 8) ? value : 0;
> +		pvt->injection.word = (u32) value;
> +
> +		return count;
> +	}
> +	return 0;
> +}
> +
> +/*
> + * amd64_inject_bit_store
> + *
> + *	accept and store error injection hexidecimal bit value
> + *	16-bits of a bit-vector marking which bits to error-out on
> + */
> +static ssize_t amd64_inject_bit_store(struct mem_ctl_info *mci,
> +					const char *data, size_t count)
> +{
> +	struct amd64_pvt *pvt = mci->pvt_info;
> +	unsigned long value;
> +	int rc;
> +
> +	rc = strict_strtoul(data, 16, &value);
> +	if (rc != -EINVAL) {
> +
> +		/* save the bit within the 16-bit word */
> +		pvt->injection.bit_map = (u32) value & 0xFFFF;
> +
> +		return count;
> +	}
> +	return 0;
> +}
> +
> +/*
> + * amd64_inject_read_store
> + *
> + *	READ action. When called, assemble staged values in the pvt
> + *	area and format into fields needed by the Injection hardware
> + *	Output to hardware and issue a READ operation
> + */
> +static ssize_t amd64_inject_read_store(struct mem_ctl_info *mci,
> +					const char *data, size_t count)
> +{
> +	struct amd64_pvt *pvt = mci->pvt_info;
> +	unsigned long value;
> +	u32 section, word_bits;
> +	int rc;
> +
> +	rc = strict_strtoul(data, 10, &value);
> +	if (rc != -EINVAL) {
> +
> +		/* Form value to choose 16-byte section of cacheline */
> +		section = F10_NB_ARRAY_DRAM_ECC |
> +				SET_NB_ARRAY_ADDRESS(pvt->injection.section);
> +		pci_write_config_dword(pvt->misc_f3_ctl,
> +					F10_NB_ARRAY_ADDR, section);
> +
> +		word_bits = SET_NB_DRAM_INJECTION_READ(pvt->injection.word,
> +						pvt->injection.bit_map);
> +
> +		/* Issue 'word' and 'bit' along with the READ request now */
> +		pci_write_config_dword(pvt->misc_f3_ctl,
> +					F10_NB_ARRAY_DATA, word_bits);
> +
> +		debugf0("%s() section=0x%x word_bits=0x%x\n", __func__,
> +			section, word_bits);
> +
> +		return count;
> +	}
> +	return 0;
> +}
> +
> +/*
> + * amd64_inject_write_store
> + *
> + *	WRITE action. When called, assemble staged values in the pvt
> + *	area and format into fields needed by the Injection hardware
> + *	Output to hardware and issue a WRITE operation
> + */
> +static ssize_t amd64_inject_write_store(struct mem_ctl_info *mci,
> +					const char *data, size_t count)
> +{
> +	struct amd64_pvt *pvt = mci->pvt_info;
> +	unsigned long value;
> +	u32 section, word_bits;
> +	int rc;
> +
> +	rc = strict_strtoul(data, 10, &value);
> +	if (rc != -EINVAL) {
> +
> +		/* Form value to choose 16-byte section of cacheline */
> +		section = F10_NB_ARRAY_DRAM_ECC |
> +				SET_NB_ARRAY_ADDRESS(pvt->injection.section);
> +		pci_write_config_dword(pvt->misc_f3_ctl,
> +					F10_NB_ARRAY_ADDR, section);
> +
> +		word_bits = SET_NB_DRAM_INJECTION_WRITE(pvt->injection.word,
> +						pvt->injection.bit_map);
> +
> +		/* Issue 'word' and 'bit' along with the READ request now */
> +		pci_write_config_dword(pvt->misc_f3_ctl,
> +					F10_NB_ARRAY_DATA, word_bits);
> +
> +		debugf0("%s() section=0x%x word_bits=0x%x\n", __func__,
> +			section, word_bits);
> +
> +		return count;
> +	}
> +	return 0;
> +}
> +#endif
> +
> +/*
> + * Per MC instance Attribute/Control data control structure
> + *	Can add for debug or for normal use
> + */
> +static struct mcidev_sysfs_attribute amd64_mc_sysfs_ctls_attrs[] = {
> +
> +#ifdef CONFIG_EDAC_AMD64_OPTERON_ERROR_INJECTION
> +	/* Error injection methods */
> +	{
> +		.attr = {
> +			.name = "z_inject_section",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = NULL,
> +		.store = amd64_inject_section_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "z_inject_word",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = NULL,
> +		.store = amd64_inject_word_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "z_inject_bit_map",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = NULL,
> +		.store = amd64_inject_bit_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "z_inject_write",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = NULL,
> +		.store = amd64_inject_write_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "z_inject_read",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = NULL,
> +		.store = amd64_inject_read_store,
> +	},
> +#endif	/* CONFIG_EDAC_AMD64_OPTERON_ERROR_INJECTION */
> +
> +#ifdef	CONFIG_EDAC_DEBUG
> +	/* RAW register accessors */
> +	{
> +		.attr = {
> +			.name = "zctl_nbea",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = amd64_nbea_show,
> +		.store = amd64_nbea_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "zctl_nbsl",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = amd64_nbsl_show,
> +		.store = amd64_nbsl_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "zctl_nbsh",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = amd64_nbsh_show,
> +		.store = amd64_nbsh_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "zctl_nbcfg",
> +			.mode = (S_IRUGO | S_IWUSR)
> +		},
> +		.show = amd64_nbcfg_show,
> +		.store = amd64_nbcfg_store,
> +	},
> +	{
> +		.attr = {
> +			.name = "zhw_dhar",
> +			.mode = (S_IRUGO)
> +		},
> +		.show = amd64_dhar_show,
> +		.store = NULL,
> +	},
> +	{
> +		.attr = {
> +			.name = "zhw_dbam",
> +			.mode = (S_IRUGO)
> +		},
> +		.show = amd64_dbam_show,
> +		.store = NULL,
> +	},
> +	{
> +		.attr = {
> +			.name = "zhw_topmem",
> +			.mode = (S_IRUGO)
> +		},
> +		.show = amd64_topmem_show,
> +		.store = NULL,
> +	},
> +	{
> +		.attr = {
> +			.name = "zhw_topmem2",
> +			.mode = (S_IRUGO)
> +		},
> +		.show = amd64_topmem2_show,
> +		.store = NULL,
> +	},
> +	{
> +		.attr = {
> +			.name = "zhw_hole",
> +			.mode = (S_IRUGO)
> +		},
> +		.show = amd64_hole_show,
> +		.store = NULL,
> +	},
> +#endif
> +	{
> +		.attr = { .name = NULL}
> +	}
> +};
> +
>   
The code looks fine. I agree with Ingo: It is better to move such large 
#ifdefs present on patches 19 and 20 into a separate c file.

Cheers,
Mauro.

Re: [PATCH 21/21] amd64_edac: add module registration routines

[Mauro Carvalho Chehab]

Borislav Petkov escreveu:
> From: Doug Thompson <dougthompson@xmission.com>
>
> Also, link into Kbuild by adding Kconfig and Makefile entries.
>
> Signed-off-by: Doug Thompson <dougthompson@xmission.com>
> Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
>   

Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>

Cheers,
Mauro.

Re: [PATCH 09/21] amd64_edac: assign DRAM chip select base and mask in a family-specific way

[Borislav Petkov]

On Mon, May 04, 2009 at 06:59:14PM -0300, Mauro Carvalho Chehab wrote:
> Borislav Petkov escreveu:
>> +	for (cs = 0; cs < CHIPSELECT_COUNT; cs++) {
>> +		reg = K8_DCSB0 + (cs * 4);
>> +		err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
>> +						&pvt->dcsb0[cs]);
>> +		if (err != 0)
>> +			debugf0("%s() Reading K8_DCSB0[%d] failed\n",
>> +				__func__, cs);
>> +
>> +		debugf0("  DCSB0[%d]=0x%08x reg: F2x%x\n",
>> +			cs, pvt->dcsb0[cs], reg);
>>   
>
> Hmm... I suspect that there's a missing else before the debugf0(). If  
> you got an error while reading it, you shouldn't be showing the results. 
>> +
>> +		/* If DCT are NOT ganged, then read in DCT1's base */
>> +		if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) {
>> +			reg = F10_DCSB1 + (cs * 4);
>> +			err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
>> +							&pvt->dcsb1[cs]);
>> +			if (err != 0)
>> +				debugf0("%s() Reading F10_DCSB1[%d] failed\n",
>> +					__func__, cs);
>> +			debugf0("  DCSB1[%d]=0x%08x reg: F2x%x\n",
>> +				cs, pvt->dcsb1[cs], reg);
>>   
> The same issue here: if you got an error while reading it, you shouldn't  
> be showing the results. 

correct, thanks.

-- 
Regards/Gruss,
Boris.

Operating | Advanced Micro Devices GmbH
  System  | Karl-Hammerschmidt-Str. 34, 85609 Dornach b. München, Germany
 Research | Geschäftsführer: Thomas M. McCoy, Giuliano Meroni
  Center  | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
  (OSRC)  | Registergericht München, HRB Nr. 43632

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Doug Thompson]

--- On Thu, 4/30/09, Andi Kleen <andi@firstfloor.org> wrote:

> From: Andi Kleen <andi@firstfloor.org>
> Subject: Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64
> To: "Doug Thompson" <norsk5@yahoo.com>
> Cc: "Andi Kleen" <andi@firstfloor.org>
> Date: Thursday, April 30, 2009, 1:05 AM
> > The problem we have had is once
> > an Uncorrected Error fires and dumps the address, mapping it
> > to the DIMM silk screen label is difficult, especially in
> > user space, in gaining access to the registers of the
> > controller.  
> 
> You can just do it either after reboot or in the crash
> kernel. I don't
> think it's required to put it all in kernel. Also you don't
> really
> need access to the registers; 

Actually, according to AMD, their reference code for mapping from an error address to a memory slot does require access to the controller's registers. On page 67 of the BKDG for family F10 from their website is 2 and 1/2 pages of the code to perform that mapping. It takes into consideration interleaving of all kinds, etc. It is narly to say the least.

> SMBIOS provides this
> information and
> mcelog knows how to convert it. 

As I undestand SMBIOS it provides a linear assignment of basic memory starts and lengths but does not provide the memory controller context as AMD's reference code takes into consideration

> 
> Trying to add other consumers to mce.c will be likely very
> messy;
> there's really no generic way to do it. I hope you're not
> planning
> turning the nicely CPU independent code in mce.c into a
> mess
> of twisty CPU specific passages like the old 32bit code
> was.
> 
> -Andi

No, not at all. Keeping the "clean" code is paramount, but we are seeking for an interface to accept the MCE error register structure and  map that information to at least a DIMM label field, if not more.

The EDAC module would register for that interface upon loading and unregister upon module unload.

The MCE code would call a stub routine that either returns no mapping occurred OR call the EDAC mapper. MCE could then determine from that return code if a mapping occurred or not. If it did, then display the desired information, otherwise proceed as normal.

doug t

Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64

[Doug Thompson]

W1DUG


--- On Thu, 4/30/09, Borislav Petkov <borislav.petkov@amd.com> wrote:

> From: Borislav Petkov <borislav.petkov@amd.com>
> Subject: Re: [RFC PATCH 00/21 v2] amd64_edac: EDAC module for AMD64
> To: "Andi Kleen" <andi@firstfloor.org>
> Cc: akpm@linux-foundation.org, greg@kroah.com, mingo@elte.hu, tglx@linutronix.de, hpa@zytor.com, dougthompson@xmission.com, linux-kernel@vger.kernel.org
> Date: Thursday, April 30, 2009, 5:57 AM
> Hi,
> 
> On Wed, Apr 29, 2009 at 09:30:31PM +0200, Andi Kleen
> wrote:
> > Borislav Petkov <borislav.petkov@amd.com>
> writes:
> > 
> > > Hi,
> > >
> > > thanks to all reviewers of the previous
> submission, here is the second
> > > version of this series.
> > 
> > The classic problem of the previous versions of these
> patches was that
> > they consume the same error registers (even if using
> pci config versus
> > msrs as access methods) as the kernel machine check
> poll/threshold
> > interrupt code.

Even the recommendation of AMD of having a polling thread for CORRECTABLE ERROR has a race issue to the same error registers due to the fact that a MCE is an exception and cannot be deferred or blocked off. In the middle of any poll cycle a MCE could fire and touch the same registers. small but present.

> >  And with two logging agents
> racing on the same
> > registers you will always get junk results. Typically
> with threshold 
> > enabled the mce code wins the race. I suspect this
> patchkit has
> > exactly the same fundamental design problem. EDAC
> really is not
> > particularly fitting for integrated memory controllers
> that report
> > their errors using standard machine check events.
> 
> ok, how about we remove tha MSR/PCI cfg space reading bits
> and leave
> that task solely to the mce core. Then, iff you have edac
> turned on in
> Kconfig, mce code delivers needed error info to edac which,
> in turn,
> goes and decodes the error/does the mapping to DIMM
> blocks/supplies DRAM
> error injection facility for testing purposes and similar
> things. That
> way you have both and they don't overlap in functionality.

Adding the synchronization between the two is very doable. It is not yet in the current patch set, but a work in progress.

That is the solution we are pursuing, to have a mechanism to provide communication between MCE and EDAC providing the mapping operation to a DIMM label. The MCA exception fires retrieves the info and calls EDAC module for address mapping.

MCE polling handler calls the EDAC module for address mapping.

EDAC's basic model is a polling operation on the error registers at a 1 second (tunable) rate. 

AMD's manual describes the UNCORRECTABLE MEMORY error handling via the MCE handler.  It further recommends a polling thread to harvest CORRECTABLE MEMORY errors. Last time I checked the MCE poller was running on a 5 minute poll cycle.

That is where we have 2 different threads polling the same error registers without synchronization is problematic and where a "Listener" pattern can be created to provide callbacks for both or form into a single poller operation.

> 
> By the way, I think there's a similar attempt/proposal of
> letting mce
> and edac talk to each other from Red Hat so I think this
> could be a
> viable thing to try.

Exactly

> 
> > -Andi (who thinks all of this decoding should be in
> user space anyways)
> 
> Think of a big data center with a thousands of 2,4,8 socket
> blades
> and the admin collecting mce output and running around
> decoding the
> errors on his workstation. Even worse, the blades have
> different DIMM
> configurations due to hw upgrades/newer machines. I'd much
> rather have
> the complete decoding done in kernel, where all the
> information needed
> for proper decoding is present and with the error landing
> in syslog or
> some other monitored buffer instead of reconstructing it in
> userspace.
> 
> Thanks.
> 
> -- 
> Regards/Gruss,
> Boris.

This model of clusters with thousands of multi-core nodes (5,000 in one case I think of) is used many times. The system console is tie to a serial port via a BIOS switch. The serial port is then attached to "conman" and all the consoles are funneled to a cluster controller which parses for a "bad memory" event.

In sites with EDAC deployed now the parser finds the node number, the CPU number on the node and extracts the EDAC DIMM label provided and generates a Repair Ticket. The technician proceeds to find the proper rack, blade and DIMM and takes that node out of service (for MCEs that are intermittent the node is reboot earlier). Then the bad DIMM is replaced - the one identified from EDAC - and the node quickly brought back online.

Without the DIMM Label provided by EDAC - or with just mce 'bad address' information - ALL the DIMMS are swapped out for off-line testing  or all are return for warranty replacement.  Getting the node back on line is the priority and reducing the time the technician spends on the rack floor. 

Bare MCE information is logged on the cluster controller and no time is spent trying to retrieve the log and running a user space program. Cheaper (man hours) and faster to swapout out all the DIMMs. But that is frowned on, with EDAC solving the problem for themnow.

The requested feature from the customers is to provide the DIMM label WITH the MCE error information as well.

doug t