Skip to contentSkip to conversation

CCIX Protocol error reporting.

PATCH In review Jonathan Cameron 3 participants 11 messages Nov 14, 2019 – Dec 13, 2019

Subsystems (5)
ACPI EFI EDAC-core ACPI APEI +1 more subsystems
lore
Share & tools
lore
#0 lore raw

v5 cover CCIX Protocol error reporting.

Changes since V4:

Reponses to Mauro's review.
* Use explicitly sized buffers for functions using the rcd_decode_str global array.
* Fix buffer overflow issue with not reducing the length as we move the start pointer in snprintf calls.
* Tidy up some first snprintf calls to take advantage of known 0s.
* Patch description cleanup to relect current state of the series.

A few minor formatting fixups and comment clarifications of things noticed whilst making the above changes.

Changes since V3:

Added #if defined(CONFIG_ACPI_APEI_CCIX) build protection to the tracepoints to avoid using undefined functions.
Reported-by: 0-day.

Changes since V2:

Dropped the legal boiler plate from the cover letter. The CCIX consortium have agreed that a simple tradmark statement is sufficient which I have put in the cper-ccix.c file and here.

The CCIX® trademark and CCIX trade name are owned solely by CCIX CONSORTIUM, INC. and all rights are reserved therein.

Changes since V1:

Addressed comments from James Morse
- Dropped kernel logging of vendor data. We just push it to the tracepoints.
- Tidied up this cover letter and added information to address questions raised. Includes removing questions where James and I agreed ;)

Note, this initial series attempts no 'handling' of errors. That will follow later.

EFI 2.8 defines a new CPER record Appendix N for CCIX Protocol Error Records (PER). www.uefi.org

These include Protocol Error Record logs which are defined in the CCIX 1.0 Base Specification www.ccixconsortium.com. A public evaluation version is now available.

Handling of coherency protocol errors is complex and how Linux does this will take some time to evolve. For now, fatal errors are handled via the usual means and everything else is reported.

There are 6 types of error defined, covering:
* Memory errors
* Cache errors
* Address translation unit errors
* CCIX port errors
* CCIX link errors
* Agent internal errors.

These errors are concerned (mostly) wth things happening in the CCIX protocol layer. They are parallel to AER errors which should be only concerned with the PCIe layer (which is underneath CCIX). The ATS errors break this rule slightly. You may get an error occurring that results in problems at both layers of the protocol stack and hence have to handle AER and PER errors simultaneously.

Some of these errors can 'almost' be mapped onto standard existing error types but only at the loss of information specific to CCIX such as where in the topology they occurred.

The set includes tracepoints to report the errors to RAS Daemon and a patch set for RAS Daemon will follow shortly.

Several design decisions that people may disagree with.
1. Reporting of vendor data. We have little choice but to do this via a dynamic array as these blocks can take arbitrary size. I had hoped no one would actually use these given the odd mismatch between a standard error structure and non standard element, but there are already designs out there that do use it. James suggested that it made sense to put these in the tracepoints, but we shouldn't spam the kernel log with them (done in V2).
2. The trade off between explicit tracepoint fields, on which we might want to filter in kernel, and the simplicity of a blob.
I have gone for having the whole of the block specific to the PER error type in an opaque blob.
The key elements that may be filtered on are the physical address and the source and component fields which allow you to identify faulty devices. Note that you have to know how the devices were enumerated to be able to do so.
3. Defined 6 new tracepoints rather than cramming everything into one.
* They are all defined by the CCIX specification as independent error classes.
* Many of them can only be generated by particular types of agent.
* The handling required will vary widely depending on types. In the kernel some map cleanly onto existing handling. Keeping the whole flow separate will aide this. They vary by a similar amount in scope to the RAS errors found on an existing system which have independent tracepoints.
* Separating them out allows for filtering on the tracepoints by elements that are not shared between them.
* Muxing the lot into one record type can lead to ugly code both in kernel and in userspace.

Rasdaemon patches posted.
https://patchwork.kernel.org/cover/11116735/

Jonathan Cameron (6):
efi / ras: CCIX Memory error reporting
efi / ras: CCIX Cache error reporting
efi / ras: CCIX Address Translation Cache error reporting efi / ras: CCIX Port error reporting
efi / ras: CCIX Link error reporting
efi / ras: CCIX Agent internal error reporting

 drivers/acpi/apei/Kconfig        |   8 +
 drivers/acpi/apei/ghes.c         |  59 ++
 drivers/firmware/efi/Kconfig     |   5 +
 drivers/firmware/efi/Makefile    |   1 +
 drivers/firmware/efi/cper-ccix.c | 923 +++++++++++++++++++++++++++++++
 drivers/firmware/efi/cper.c      |   6 +
 include/linux/cper.h             | 333 +++++++++++
 include/ras/ras_event.h          | 407 ++++++++++++++
 8 files changed, 1742 insertions(+)
 create mode 100644 drivers/firmware/efi/cper-ccix.c
Signature · 1 line
2.20.1
#1 lore raw

PATCH v5 1/6 efi / ras: CCIX Memory error reporting

CCIX defines a number of different error types
(See CCIX spec 1.0) and UEFI 2.8 defines a CPER record to allow for them to be reported when firmware first handling is in use. The last part of that record is a copy of the CCIX protocol error record which can provide very detailed information.

This patch introduces infrastructure and support for one of those error types, CCIX Memory Errors. Later patches will supply equivalent support for the other error types.

The variable length and content of the different messages makes a single tracepoint impractical. As such the RAS tracepoint introduced in this patch only covers the memory error. Additional trace points will be introduced for other error types along with their cper handling in the rest of this series.

Updated RAS daemon support to be posted shortly.
qemu injection patches also available but not currently planing to upstream those.

---

Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.

 drivers/acpi/apei/Kconfig        |   8 +
 drivers/acpi/apei/ghes.c         |  39 ++++
 drivers/firmware/efi/Kconfig     |   5 +
 drivers/firmware/efi/Makefile    |   1 +
 drivers/firmware/efi/cper-ccix.c | 362 +++++++++++++++++++++++++++++++
 drivers/firmware/efi/cper.c      |   6 +
 include/linux/cper.h             | 118 ++++++++++
 include/ras/ras_event.h          |  79 +++++++
 8 files changed, 618 insertions(+)
drivers/acpi/apei/Kconfig +8 -0
index 6b18f8bc7be3..e687b18dee34 100644
@@ -68,3 +68,11 @@ config ACPI_APEI_ERST_DEBUG
 	  error information to and from a persistent store. Enable this
 	  if you want to debugging and testing the ERST kernel support
 	  and firmware implementation.
+
+config ACPI_APEI_CCIX
+       bool "APEI CCIX error recovery support"
+       depends on ACPI_APEI && MEMORY_FAILURE
+       help
+	 CCIX has a number of defined error types. This option enables
+	 the handling of CPER records generated by a firmware performing
+	 firmware first error handling of these CCIX errors.
drivers/acpi/apei/ghes.c +39 -0
index 777f6f7122b4..75a177ae9de3 100644
@@ -490,6 +490,42 @@ static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
 #endif
 }
 
+static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
+{
+#ifdef CONFIG_ACPI_APEI_CCIX
+	struct cper_sec_ccix_header *header = acpi_hest_get_payload(gdata);
+	__u32 *dw;
+	enum ccix_per_type per_type;
+	static u32 err_seq;
+	void *payload;
+
+	/* Check if space for CCIX CPER header and 8 DW of a PER log header */
+	if (gdata->error_data_length <
+	    sizeof(*header) + CCIX_PER_LOG_HEADER_DWS * sizeof(__u32))
+		return;
+
+	if ((header->validation_bits & CPER_CCIX_VALID_PER_LOG) == 0)
+		return;
+
+	dw = (__u32 *)(header + 1);
+
+	per_type = FIELD_GET(CCIX_PER_LOG_DW1_PER_TYPE_M, dw[1]);
+	payload = acpi_hest_get_payload(gdata);
+
+	switch (per_type) {
+	case CCIX_MEMORY_ERROR:
+		trace_ccix_memory_error_event(payload, err_seq, sev,
+					      ccix_mem_err_ven_len_get(payload));
+		break;
+	default:
+		/* Unknown error type */
+		pr_info("CCIX error of unknown or vendor defined type\n");
+		break;
+	}
+	err_seq++;
+#endif
+}
+
 static void ghes_do_proc(struct ghes *ghes,
 			 const struct acpi_hest_generic_status *estatus)
 {
@@ -520,6 +556,9 @@ static void ghes_do_proc(struct ghes *ghes,
 		else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
 			ghes_handle_aer(gdata);
 		}
+		else if (guid_equal(sec_type, &CPER_SEC_CCIX)) {
+			ghes_handle_ccix_per(gdata, estatus->error_severity);
+		}
 		else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
 			struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
 
drivers/firmware/efi/Kconfig +5 -0
index b248870a9806..096e693a9522 100644
@@ -209,6 +209,11 @@ config UEFI_CPER_X86
 	depends on UEFI_CPER && X86
 	default y
 
+config UEFI_CPER_CCIX
+       bool
+       depends on UEFI_CPER
+       default y
+
 config EFI_DEV_PATH_PARSER
 	bool
 	depends on ACPI
drivers/firmware/efi/Makefile +1 -0
index 4ac2de4dfa72..9a52c7d01e94 100644
@@ -34,3 +34,4 @@ obj-$(CONFIG_EFI_CAPSULE_LOADER)	+= capsule-loader.o
 obj-$(CONFIG_EFI_EARLYCON)		+= earlycon.o
 obj-$(CONFIG_UEFI_CPER_ARM)		+= cper-arm.o
 obj-$(CONFIG_UEFI_CPER_X86)		+= cper-x86.o
+obj-$(CONFIG_UEFI_CPER_CCIX)		+= cper-ccix.o
drivers/firmware/efi/cper-ccix.c new file +362 -0

362 lines not on this page · raw message

drivers/firmware/efi/cper.c +6 -0
index b1af0de2e100..03bb27db2e87 100644
@@ -474,6 +474,12 @@ cper_estatus_print_section(const char *pfx, struct acpi_hest_generic_data *gdata
 			cper_print_pcie(newpfx, pcie, gdata);
 		else
 			goto err_section_too_small;
+	} else if (guid_equal(sec_type, &CPER_SEC_CCIX)) {
+		int ret;
+		/* CCIX CPER entries are variable length */
+		ret = cper_print_ccix_per(newpfx, gdata);
+		if (ret)
+			goto err_section_too_small;
 #if defined(CONFIG_ARM64) || defined(CONFIG_ARM)
 	} else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
 		struct cper_sec_proc_arm *arm_err = acpi_hest_get_payload(gdata);
include/linux/cper.h +118 -0

142 lines not on this page · raw message

include/ras/ras_event.h +79 -0

91 lines not on this page · raw message

Signature · 1 line
2.20.1
#2 lore raw

Re: [PATCH v5 1/6] efi / ras: CCIX Memory error reporting

Em Thu, 14 Nov 2019 21:39:14 +0800
Jonathan Cameron <[email protected]> escreveu:

Quoted 82 lines

CCIX defines a number of different error types
(See CCIX spec 1.0) and UEFI 2.8 defines a CPER record to allow for them to be reported when firmware first handling is in use. The last part of that record is a copy of the CCIX protocol error record which can provide very detailed information.

This patch introduces infrastructure and support for one of those error types, CCIX Memory Errors. Later patches will supply equivalent support for the other error types.

The variable length and content of the different messages makes a single tracepoint impractical. As such the RAS tracepoint introduced in this patch only covers the memory error. Additional trace points will be introduced for other error types along with their cper handling in the rest of this series.

Updated RAS daemon support to be posted shortly.
qemu injection patches also available but not currently planing to upstream those.

---

Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.

 drivers/acpi/apei/Kconfig        |   8 +
 drivers/acpi/apei/ghes.c         |  39 ++++
 drivers/firmware/efi/Kconfig     |   5 +
 drivers/firmware/efi/Makefile    |   1 +
 drivers/firmware/efi/cper-ccix.c | 362 +++++++++++++++++++++++++++++++
 drivers/firmware/efi/cper.c      |   6 +
 include/linux/cper.h             | 118 ++++++++++
 include/ras/ras_event.h          |  79 +++++++
 8 files changed, 618 insertions(+)
drivers/acpi/apei/Kconfig +8 -0
index 6b18f8bc7be3..e687b18dee34 100644
@@ -68,3 +68,11 @@ config ACPI_APEI_ERST_DEBUG
 	  error information to and from a persistent store. Enable this
 	  if you want to debugging and testing the ERST kernel support
 	  and firmware implementation.
+
+config ACPI_APEI_CCIX
+       bool "APEI CCIX error recovery support"
+       depends on ACPI_APEI && MEMORY_FAILURE
+       help
+	 CCIX has a number of defined error types. This option enables
+	 the handling of CPER records generated by a firmware performing
+	 firmware first error handling of these CCIX errors.
drivers/acpi/apei/ghes.c +20 -0
index 777f6f7122b4..75a177ae9de3 100644
@@ -490,6 +490,42 @@ static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
 #endif
 }
 
+static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
+{
+#ifdef CONFIG_ACPI_APEI_CCIX
+	struct cper_sec_ccix_header *header = acpi_hest_get_payload(gdata);
+	__u32 *dw;
+	enum ccix_per_type per_type;
+	static u32 err_seq;
+	void *payload;
+
+	/* Check if space for CCIX CPER header and 8 DW of a PER log header */
+	if (gdata->error_data_length <
+	    sizeof(*header) + CCIX_PER_LOG_HEADER_DWS * sizeof(__u32))
+		return;
+
+	if ((header->validation_bits & CPER_CCIX_VALID_PER_LOG) == 0)
+		return;
+
+	dw = (__u32 *)(header + 1);
+
+	per_type = FIELD_GET(CCIX_PER_LOG_DW1_PER_TYPE_M, dw[1]);

drivers/acpi/apei/ghes.c static void ghes_handle_aer(struct acpi_hest_generic_data *gdata) · line 522

Quoted 10 lines
drivers/acpi/apei/ghes.c +10 -0
+	payload = acpi_hest_get_payload(gdata);
+
+	switch (per_type) {
+	case CCIX_MEMORY_ERROR:
+		trace_ccix_memory_error_event(payload, err_seq, sev,
+					      ccix_mem_err_ven_len_get(payload));
+		break;
+	default:
+		/* Unknown error type */
+		pr_info("CCIX error of unknown or vendor defined type\n");

Hmm... maybe this could generate a trace too.

On another matter, IMHO, it would make sense to add the value of "per_type" at the error message, as this can be a lot helpful if this part of the code is ever reached.

Quoted 206 lines
drivers/acpi/apei/ghes.c +9 -0
+		break;
+	}
+	err_seq++;
+#endif
+}
+
 static void ghes_do_proc(struct ghes *ghes,
 			 const struct acpi_hest_generic_status *estatus)
 {
@@ -520,6 +556,9 @@ static void ghes_do_proc(struct ghes *ghes,
 		else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
 			ghes_handle_aer(gdata);
 		}
+		else if (guid_equal(sec_type, &CPER_SEC_CCIX)) {
+			ghes_handle_ccix_per(gdata, estatus->error_severity);
+		}
 		else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
 			struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
 
drivers/firmware/efi/Kconfig +5 -0
index b248870a9806..096e693a9522 100644
@@ -209,6 +209,11 @@ config UEFI_CPER_X86
 	depends on UEFI_CPER && X86
 	default y
 
+config UEFI_CPER_CCIX
+       bool
+       depends on UEFI_CPER
+       default y
+
 config EFI_DEV_PATH_PARSER
 	bool
 	depends on ACPI
drivers/firmware/efi/Makefile +1 -0
index 4ac2de4dfa72..9a52c7d01e94 100644
@@ -34,3 +34,4 @@ obj-$(CONFIG_EFI_CAPSULE_LOADER)	+= capsule-loader.o
 obj-$(CONFIG_EFI_EARLYCON)		+= earlycon.o
 obj-$(CONFIG_UEFI_CPER_ARM)		+= cper-arm.o
 obj-$(CONFIG_UEFI_CPER_X86)		+= cper-x86.o
+obj-$(CONFIG_UEFI_CPER_CCIX)		+= cper-ccix.o
drivers/firmware/efi/cper-ccix.c new file +156 -0
new file mode 100644
index 000000000000..588a7aaa6a72
@@ -0,0 +1,362 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * UEFI Common Platform Error Record (CPER) support for CCIX
+ * protocol errors.
+ *
+ * Copyright (C) 2019, Huawei
+ *	Author: Jonathan Cameron <[email protected]>
+ *
+ * The CCIX® trademark and CCIX trade name are owned solely by
+ * CCIX CONSORTIUM, INC. and all rights are reserved therein.
+ *
+ * CPER is the format used to describe platform hardware error by
+ * various tables, such as ERST, BERT and HEST etc.
+ *
+ * For more information about CPER, please refer to Appendix N of UEFI
+ * Specification version 2.9.
+ *
+ * CCIX defines a number of Protocol Error Messages which for the
+ * main body of the CCIX CPER records.  These are defined in the
+ * CCIX Specification 1.0.
+ */
+
+#include <acpi/ghes.h>
+#include <linux/acpi.h>
+#include <linux/bitfield.h>
+#include <linux/cper.h>
+#include <linux/kernel.h>
+#include <linux/printk.h>
+#include <ras/ras_event.h>
+
+static char rcd_decode_str[CPER_REC_LEN];
+
+static const char * const ccix_comp_type_strs[] = {
+	"Request Agent",
+	"Home Agent",
+	"Slave Agent",
+	"Port",
+	"Link",
+};
+
+const char *cper_ccix_comp_type_str(u8 comp_type)
+{
+	return comp_type < ARRAY_SIZE(ccix_comp_type_strs) ?
+		ccix_comp_type_strs[comp_type] : "Reserved";
+}
+
+static const char * const ccix_per_type_strs[] = {
+	"Memory Error",
+	"Cache Error",
+	"ATC Error",
+	"Port Error",
+	"Link Error",
+	"Agent Internal",
+};
+
+static const char * const ccix_mem_pool_gen_type_strs[] = {
+	"Other, Non-specified",
+	"ROM",
+	"Volatile",
+	"Non-volatile",
+	"Device",
+};
+
+static const char *cper_ccix_mem_err_generic_type_str(u16 type)
+{
+	const char *gen_type_str;
+
+	if (type < ARRAY_SIZE(ccix_mem_pool_gen_type_strs))
+		gen_type_str = ccix_mem_pool_gen_type_strs[type];
+	else if (type >= 0x80)
+		gen_type_str = "Vendor";
+	else
+		gen_type_str = "Reserved";
+
+	return gen_type_str;
+}
+
+static const char * const ccix_mem_op_type_strs[] = {
+	"Generic",
+	"Read",
+	"Write",
+	"Reserved",
+	"Scrub",
+};
+
+static const char *cper_ccix_mem_err_op_str(u8 op_type)
+{
+	return op_type < ARRAY_SIZE(ccix_mem_op_type_strs) ?
+		ccix_mem_op_type_strs[op_type] :
+		"Reserved";
+}
+
+/* Sightly different from the generic version */
+static const char * const ccix_mem_err_type_strs[] = {
+	"Unknown",
+	"No Error",
+	"Single-bit ECC",
+	"Multi-bit ECC",
+	"Single-symbol ChipKill ECC",
+	"Multi-symbol ChipKill ECC",
+	"Master Abort",
+	"Target Abort",
+	"Parity Error",
+	"Watchdog Timeout",
+	"Invalid Address",
+	"Mirror Broken",
+	"Memory Sparing",
+	"Scrub",
+	"Physical Memory Map-out Event",
+};
+
+const char *cper_ccix_mem_err_type_str(unsigned int error_type)
+{
+	return error_type < ARRAY_SIZE(ccix_mem_err_type_strs) ?
+		ccix_mem_err_type_strs[error_type] : "Reserved";
+}
+
+static const char * const ccix_mem_spec_type_strs[] = {
+	"Other, Not-specified",
+	"SRAM",
+	"DDR",
+	"NVDIMM-F",
+	"NVDIMM-N",
+	"HBM",
+	"Flash"
+};
+
+static const char *cper_ccix_mem_err_spec_type_str(u8 specific_type)
+{
+	if (specific_type < ARRAY_SIZE(ccix_mem_spec_type_strs))
+		return ccix_mem_spec_type_strs[specific_type];
+	else if (specific_type >= 0x80)
+		return "Vendor";
+	else
+		return "Reserved";
+}
+
+/*
+ * We pack up everything except those that are needed for software handling:
+ * - error_type, physical_addr
+ * and header values that would require additional validation bits:
+ * - source, component, severity,
+ * implicit: protocol error type (mem)
+ */
+void cper_ccix_mem_err_pack(const struct cper_sec_ccix_mem_error *mem_record,
+			    struct cper_ccix_mem_err_compact *cmem_err,
+			    const u16 vendor_data_len,
+			    u8 *vendor_data)
+{
+	cmem_err->validation_bits = mem_record->validation_bits;
+	cmem_err->mem_err_type = mem_record->memory_error_type;
+	cmem_err->pool_generic_type = mem_record->pool_generic_type;
+	cmem_err->op_type = mem_record->op_type;
+	cmem_err->card = mem_record->card;
+	cmem_err->module = mem_record->module;
+	cmem_err->bank = mem_record->bank;

drivers/firmware/efi/cper-ccix.c line 165

Quoted 10 lines
drivers/firmware/efi/cper-ccix.c new file +9 -0
+	cmem_err->device = mem_record->device;
+	cmem_err->row = mem_record->row;
+	cmem_err->column = mem_record->column;
+	cmem_err->rank = mem_record->rank;
+	cmem_err->bit_pos = mem_record->bit_pos;
+	cmem_err->chip_id = mem_record->chip_id;
+	cmem_err->pool_specific_type = mem_record->pool_specific_type;
+	cmem_err->fru = mem_record->fru;
 
+	memcpy(vendor_data, &mem_record->vendor_data[1], vendor_data_len);

Hmm... I was unable to identify where you allocated mem_record, but on a quicker view, I suspect you need to have some tests here, in order to check if the length of the mem_record matches vendor_data_len.

Quoted 61 lines
drivers/firmware/efi/cper-ccix.c new file +61 -0
+}
+
+static int cper_ccix_err_location(struct cper_ccix_mem_err_compact *cmem_err,
+				  char msg[CPER_REC_LEN])
+{
+	u32 len = CPER_REC_LEN - 1;
+	u32 n = 0;
+
+	if (!msg)
+		return 0;
+
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_GENERIC_MEM_VALID)
+		n = snprintf(msg, len, "Pool Generic Type: %s ",
+			     cper_ccix_mem_err_generic_type_str(cmem_err->pool_generic_type));
+
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_MEM_ERR_TYPE_VALID)
+		n += snprintf(msg + n, len - n, "Err Type: %s ",
+			      cper_ccix_mem_err_type_str(cmem_err->mem_err_type));
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_OP_VALID)
+		n += snprintf(msg + n, len - n, "Operation: %s ",
+			      cper_ccix_mem_err_op_str(cmem_err->op_type));
+
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_CARD_VALID)
+		n += snprintf(msg + n, len - n, "Card: %d ", cmem_err->card);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_MOD_VALID)
+		n += snprintf(msg + n, len - n, "Mod: %d ", cmem_err->module);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_BANK_VALID)
+		n += snprintf(msg + n, len - n, "Bank: %d ", cmem_err->bank);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_DEVICE_VALID)
+		n += snprintf(msg + n, len - n, "Device: %d ",
+			      cmem_err->device);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_ROW_VALID)
+		n += snprintf(msg + n, len - n, "Row: %d ", cmem_err->row);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_COL_VALID)
+		n += snprintf(msg + n, len - n, "Col: %d ", cmem_err->column);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_RANK_VALID)
+		n += snprintf(msg + n, len - n, "Rank: %d ", cmem_err->rank);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_BIT_POS_VALID)
+		n += snprintf(msg + n, len - n, "BitPos: %d ",
+			      cmem_err->bit_pos);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_CHIP_ID_VALID)
+		n += snprintf(msg + n, len - n, "ChipID: %d ",
+			      cmem_err->chip_id);
+	if (cmem_err->validation_bits & CCIX_MEM_ERR_SPEC_TYPE_VALID)
+		n += snprintf(msg + n, len - n, "Pool Specific Type: %s ",
+			      cper_ccix_mem_err_spec_type_str(cmem_err->pool_specific_type));
+	n += snprintf(msg + n, len - n, "FRU: %d ", cmem_err->fru);
+
+	return n;
+}
+
+const char *cper_ccix_mem_err_unpack(struct trace_seq *p,
+				     struct cper_ccix_mem_err_compact *cmem_err)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+
+	if (cper_ccix_err_location(cmem_err, rcd_decode_str))
+		trace_seq_printf(p, "%s", rcd_decode_str);
+	trace_seq_putc(p, '\0');
+
+	return ret;

drivers/firmware/efi/cper-ccix.c line 236

Quoted 10 lines
drivers/firmware/efi/cper-ccix.c new file +10 -0
+}
+
+static int cper_ccix_mem_err_details(const char *pfx,
+				     struct acpi_hest_generic_data *gdata)
+{
+	struct cper_ccix_mem_error *full_mem_err;
+	struct cper_sec_ccix_mem_error *mem_err;
+	u16 vendor_data_len;
+	int i;
+

Are you just printing here the same message that would be sent to trace?

If so, I would add here something like:

if (ras_userspace_consumers())
return; /* do not printk */

drivers/firmware/efi/cper-ccix.c line 247

Quoted 13 lines
drivers/firmware/efi/cper-ccix.c new file +11 -0
+	if (gdata->error_data_length < sizeof(*full_mem_err))
+		return -ENOSPC;
+
+	full_mem_err = acpi_hest_get_payload(gdata);
+
+	mem_err = &full_mem_err->mem_record;
 
 
+	printk("%s""FRU ID: %u, Length: %u\n", pfx,
+	       mem_err->fru, mem_err->length);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_GENERIC_MEM_VALID)
+		printk("%s""Pool Generic Type: %s\n", pfx,
+		       cper_ccix_mem_err_generic_type_str(mem_err->pool_generic_type));

Please don't call printk() directly. Instead, use the higher level debug macros,like pr_error().

The same two notes apply to all other similar code on this patch and at the next ones.

Quoted 369 lines
drivers/firmware/efi/cper-ccix.c new file +115 -0
+
+	if (mem_err->validation_bits & CCIX_MEM_ERR_OP_VALID)
+		printk("%s""Operation: %s\n", pfx,
+		       cper_ccix_mem_err_op_str(mem_err->op_type));
+
+	if (mem_err->validation_bits & CCIX_MEM_ERR_MEM_ERR_TYPE_VALID) {
+		printk("%s""Mem Error Type: %s\n", pfx,
+		       cper_ccix_mem_err_type_str(mem_err->memory_error_type));
+	}
+	if (mem_err->validation_bits & CCIX_MEM_ERR_CARD_VALID)
+		printk("%s""Card: %u\n", pfx, mem_err->card);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_MOD_VALID)
+		printk("%s""Module: %u\n", pfx, mem_err->module);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_BANK_VALID)
+		printk("%s""Bank: %u\n", pfx, mem_err->bank);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_DEVICE_VALID)
+		printk("%s""Device: %u\n", pfx, mem_err->device);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_ROW_VALID)
+		printk("%s""Row: %u\n", pfx, mem_err->row);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_COL_VALID)
+		printk("%s""Column: %u\n", pfx, mem_err->column);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_RANK_VALID)
+		printk("%s""Rank: %u\n", pfx, mem_err->rank);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_BIT_POS_VALID)
+		printk("%s""Bit Pos: %u\n", pfx, mem_err->bit_pos);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_CHIP_ID_VALID)
+		printk("%s""Chip ID: %u\n", pfx, mem_err->chip_id);
+	if (mem_err->validation_bits & CCIX_MEM_ERR_SPEC_TYPE_VALID)
+		printk("%s""Specific Type: %s\n", pfx,
+		       cper_ccix_mem_err_spec_type_str(mem_err->pool_specific_type));
+
+	if (mem_err->validation_bits & CCIX_MEM_ERR_VENDOR_DATA_VALID) {
+		if (gdata->error_data_length < sizeof(*full_mem_err) + 4)
+			return -ENOSPC;
+
+		vendor_data_len = mem_err->vendor_data[0] & GENMASK(15, 0);
+		if (gdata->error_data_length <
+		    sizeof(*full_mem_err) + vendor_data_len)
+			return -ENOSPC;
+
+		for (i = 0; i < vendor_data_len / 4 - 1; i++)
+			printk("%s""Vendor%d: 0x%08x\n", pfx, i,
+			       mem_err->vendor_data[i + 1]);
+	}
+
+	return 0;
+}
+
+int cper_print_ccix_per(const char *pfx, struct acpi_hest_generic_data *gdata)
+{
+	struct cper_sec_ccix_header *header = acpi_hest_get_payload(gdata);
+	__u32 *dw;
+	__u32 comp_type;
+	enum ccix_per_type per_type;
+	bool vendor_per;
+
+	if (gdata->error_data_length < sizeof(*header))
+		return -ENOSPC;
+
+	printk("%s""CPER Length: %u\n", pfx, header->length);
+	if (header->validation_bits & CPER_CCIX_VALID_SOURCE_ID)
+		printk("%s""Source: %u\n", pfx, header->source_id);
+	if (header->validation_bits & CPER_CCIX_VALID_PORT_ID)
+		printk("%s""Port: %u\n", pfx, header->port_id);
+	/* Not much use if we don't have the per log, in theory it's optional */
+	if ((header->validation_bits & CPER_CCIX_VALID_PER_LOG) == 0)
+		return 0;
+
+	/* The per log header is a packed structure so needs breaking up */
+	if (gdata->error_data_length < sizeof(*header) + 8 * 4)
+		return -ENOSPC;
+
+	dw = (__u32 *)(header + 1);
+
+	printk("%s""PER Rev: %lu, Log Length: %lu\n", pfx,
+	       FIELD_GET(CCIX_PER_LOG_DW0_REV_M, dw[0]),
+	       FIELD_GET(CCIX_PER_LOG_DW0_LEN_M, dw[0]));
+	comp_type = FIELD_GET(CCIX_PER_LOG_DW1_COMP_TYPE_M, dw[1]);
+	printk("%s""Component: %s\n", pfx, cper_ccix_comp_type_str(comp_type));
+	printk("%s""ME: %lu, SevUE: %lu, SevNoComm: %lu, SevDegraded: %lu, SevDeferred %lu\n",
+	       pfx,
+	       FIELD_GET(CCIX_PER_LOG_DW0_ME_M, dw[0]),
+	       FIELD_GET(CCIX_PER_LOG_DW1_SEV_UE_M, dw[1]),
+	       FIELD_GET(CCIX_PER_LOG_DW1_SEV_NO_COMM_M, dw[1]),
+	       FIELD_GET(CCIX_PER_LOG_DW1_SEV_DEGRADED_M, dw[1]),
+	       FIELD_GET(CCIX_PER_LOG_DW1_SEV_DEFFERABLE_M, dw[1]));
+
+	/* per_type is vendor defined if VEN is set */
+	vendor_per = FIELD_GET(CCIX_PER_LOG_DW1_VEN_VAL_M, dw[1]) ?
+		true : false;
+	per_type = FIELD_GET(CCIX_PER_LOG_DW1_PER_TYPE_M, dw[1]);
+	if (vendor_per)
+		printk("%s""Protocol Error Type: Vendor%u", pfx, per_type);
+	else
+		printk("%s""Protocol Error Type: %s\n", pfx,
+		       per_type < ARRAY_SIZE(ccix_per_type_strs) ?
+		       ccix_per_type_strs[per_type] : "Reserved");
+
+	if (FIELD_GET(CCIX_PER_LOG_DW1_ADDR_VAL_M, dw[1]))
+		printk("%s""Address: 0x%llx\n", pfx,
+		       (((__u64)dw[2]) << 32) | (dw[3] & 0xFFFFFFFC));
+
+	/* Vendor defined PER message, perhaps we could print it out */
+	if (vendor_per)
+		return 0;
+
+	switch (per_type) {
+	case CCIX_MEMORY_ERROR:
+		return cper_ccix_mem_err_details(pfx, gdata);
+	default:
+		/* Vendor defined so no formatting be done */
+		break;
+	}
+	return 0;
+}
drivers/firmware/efi/cper.c +6 -0
index b1af0de2e100..03bb27db2e87 100644
@@ -474,6 +474,12 @@ cper_estatus_print_section(const char *pfx, struct acpi_hest_generic_data *gdata
 			cper_print_pcie(newpfx, pcie, gdata);
 		else
 			goto err_section_too_small;
+	} else if (guid_equal(sec_type, &CPER_SEC_CCIX)) {
+		int ret;
+		/* CCIX CPER entries are variable length */
+		ret = cper_print_ccix_per(newpfx, gdata);
+		if (ret)
+			goto err_section_too_small;
 #if defined(CONFIG_ARM64) || defined(CONFIG_ARM)
 	} else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
 		struct cper_sec_proc_arm *arm_err = acpi_hest_get_payload(gdata);
include/linux/cper.h +118 -0
index 4f005d95ce88..df7a34c3ba4f 100644
@@ -174,6 +174,9 @@ enum {
 #define CPER_SEC_PCIE							\
 	GUID_INIT(0xD995E954, 0xBBC1, 0x430F, 0xAD, 0x91, 0xB4, 0x4D,	\
 		  0xCB, 0x3C, 0x6F, 0x35)
+#define CPER_SEC_CCIX							\
+	GUID_INIT(0x91335EF6, 0xEBFB, 0x4478, 0xA6, 0xA6, 0x88, 0xB7,	\
+		  0x28, 0xCF, 0x75, 0xD7)
 /* Firmware Error Record Reference */
 #define CPER_SEC_FW_ERR_REC_REF						\
 	GUID_INIT(0x81212A96, 0x09ED, 0x4996, 0x94, 0x71, 0x8D, 0x72,	\
@@ -242,6 +245,10 @@ enum {
 
 #define CPER_PCIE_SLOT_SHIFT			3
 
+#define CPER_CCIX_VALID_SOURCE_ID		BIT(0)
+#define CPER_CCIX_VALID_PORT_ID			BIT(1)
+#define CPER_CCIX_VALID_PER_LOG			BIT(2)
+
 #define CPER_ARM_VALID_MPIDR			BIT(0)
 #define CPER_ARM_VALID_AFFINITY_LEVEL		BIT(1)
 #define CPER_ARM_VALID_RUNNING_STATE		BIT(2)
@@ -521,6 +528,105 @@ struct cper_sec_pcie {
 	u8	aer_info[96];
 };
 
+struct cper_sec_ccix_header {
+	__u32	length;
+	__u64	validation_bits;
+	__u8	source_id;
+	__u8	port_id;
+	__u8	reserved[2];
+};
+
+#define CCIX_PER_LOG_DW0_REV_M			GENMASK(7, 0)
+#define CCIX_PER_LOG_DW0_LEN_M			GENMASK(14, 8)
+#define CCIX_PER_LOG_DW0_ME_M			BIT(15)
+#define CCIX_PER_LOG_DW1_COMP_TYPE_M		GENMASK(15, 12)
+#define CCIX_PER_LOG_DW1_SEV_UE_M		BIT(16)
+#define CCIX_PER_LOG_DW1_SEV_NO_COMM_M		BIT(17)
+#define CCIX_PER_LOG_DW1_SEV_DEGRADED_M		BIT(18)
+#define CCIX_PER_LOG_DW1_SEV_DEFFERABLE_M	BIT(19)
+#define CCIX_PER_LOG_DW1_PER_TYPE_M		GENMASK(27, 24)
+#define CCIX_PER_LOG_DW1_ADDR_VAL_M		BIT(30)
+#define CCIX_PER_LOG_DW1_VEN_VAL_M		BIT(31)
+enum ccix_per_type {
+	CCIX_MEMORY_ERROR = 0,
+	CCIX_CACHE_ERROR = 1,
+	CCIX_ATC_ERROR = 2,
+	CCIX_PORT_ERROR = 3,
+	CCIX_LINK_ERROR = 4,
+	CCIX_AGENT_INTERNAL_ERROR = 5,
+};
+
+#define CCIX_PER_LOG_HEADER_DWS 8
+
+struct cper_sec_ccix_mem_error {
+	__u32	validation_bits;
+#define CCIX_MEM_ERR_GENERIC_MEM_VALID		BIT(0)
+#define CCIX_MEM_ERR_OP_VALID			BIT(1)
+#define CCIX_MEM_ERR_MEM_ERR_TYPE_VALID		BIT(2)
+#define CCIX_MEM_ERR_CARD_VALID			BIT(3)
+#define CCIX_MEM_ERR_BANK_VALID			BIT(4)
+#define CCIX_MEM_ERR_DEVICE_VALID		BIT(5)
+#define CCIX_MEM_ERR_ROW_VALID			BIT(6)
+#define CCIX_MEM_ERR_COL_VALID			BIT(7)
+#define CCIX_MEM_ERR_RANK_VALID			BIT(8)
+#define CCIX_MEM_ERR_BIT_POS_VALID		BIT(9)
+#define CCIX_MEM_ERR_CHIP_ID_VALID		BIT(10)
+#define CCIX_MEM_ERR_VENDOR_DATA_VALID		BIT(11)
+#define CCIX_MEM_ERR_MOD_VALID			BIT(12)
+#define CCIX_MEM_ERR_SPEC_TYPE_VALID		BIT(13)
+
+	__u8	fru;
+	__u8	reserved;
+	__u16	length; /* Includes vendor specific log info */
+	__u8	pool_generic_type;
+	__u8	op_type;
+	__u8	memory_error_type;
+	__u8	card;
+	__u16	module;
+	__u16	bank;
+	__u32	device;
+	__u32	row;
+	__u32	column;
+	__u32	rank;
+	__u8	bit_pos;
+	__u8	chip_id;
+	__u8	pool_specific_type;
+	__u32	vendor_data[];
+};
+
+struct cper_ccix_mem_error {
+	struct cper_sec_ccix_header header;
+	__u32 ccix_header[CCIX_PER_LOG_HEADER_DWS];
+	struct cper_sec_ccix_mem_error mem_record;
+};
+
+static inline u16 ccix_mem_err_ven_len_get(struct cper_ccix_mem_error *mem_err)
+{
+	if (mem_err->mem_record.validation_bits &
+	    CCIX_MEM_ERR_VENDOR_DATA_VALID)
+		return mem_err->mem_record.vendor_data[0] & 0xFFFF;
+	else
+		return 0;
+}
+
+struct cper_ccix_mem_err_compact {
+	__u32	validation_bits;
+	__u8	mem_err_type;
+	__u8	pool_generic_type;
+	__u8	pool_specific_type;
+	__u8	op_type;
+	__u8	card;
+	__u16	module;
+	__u16	bank;
+	__u32	device;
+	__u32	row;
+	__u32	column;
+	__u32	rank;
+	__u8	bit_pos;
+	__u8	chip_id;
+	__u8	fru;
+};
+
 /* Reset to default packing */
 #pragma pack()
 
@@ -535,6 +641,18 @@ void cper_mem_err_pack(const struct cper_sec_mem_err *,
 		       struct cper_mem_err_compact *);
 const char *cper_mem_err_unpack(struct trace_seq *,
 				struct cper_mem_err_compact *);
+void cper_ccix_mem_err_pack(const struct cper_sec_ccix_mem_error *mem_record,
+			    struct cper_ccix_mem_err_compact *cmem_err,
+			    const u16 vendor_data_len,
+			    u8 *vendor_data);
+const char *cper_ccix_mem_err_unpack(struct trace_seq *p,
+				     struct cper_ccix_mem_err_compact *cmem_err);
+const char *cper_ccix_mem_err_type_str(unsigned int error_type);
+const char *cper_ccix_comp_type_str(u8 comp_type);
+struct acpi_hest_generic_data;
+int cper_print_ccix_per(const char *pfx,
+			struct acpi_hest_generic_data *gdata);
+
 void cper_print_proc_arm(const char *pfx,
 			 const struct cper_sec_proc_arm *proc);
 void cper_print_proc_ia(const char *pfx,
include/ras/ras_event.h +72 -0
index 36c5c5e38c1d..560e55958561 100644
@@ -14,6 +14,7 @@
 #include <linux/cper.h>
 #include <linux/mm.h>
 
+#include <linux/bitfield.h>
 /*
  * MCE Extended Error Log trace event
  *
@@ -338,6 +339,84 @@ TRACE_EVENT(aer_event,
 			"Not available")
 );
 
+#if defined(CONFIG_ACPI_APEI_CCIX)
+/*
+ * CCIX PER log memory error trace event
+ *
+ * These events are generated when hardware detects a corrected or
+ * uncorrected event.
+ *
+ * Some elements of the record are not included
+ * - PER version (tracepoint should remain compatible across versions)
+ * - Multiple Error
+ */
+TRACE_EVENT(ccix_memory_error_event,
+	TP_PROTO(struct cper_ccix_mem_error *mem,
+		 u32 err_seq,
+		 u8 sev,
+		 u16 ven_len),
+
+	TP_ARGS(mem, err_seq, sev, ven_len),
+
+	TP_STRUCT__entry(
+		__field(u32, err_seq)
+		__field(u8, sev)
+		__field(u8, sevdetail)
+		__field(u8, source)
+		__field(u8, component)
+		__field(u64, pa)
+		__field(u8, pa_mask_lsb)
+		__field_struct(struct cper_ccix_mem_err_compact, data)
+		__field(u16, vendor_data_length)
+		__dynamic_array(u8, vendor_data, ven_len)
+	),
+
+	TP_fast_assign(
+		__entry->err_seq = err_seq;
+		__entry->sev = sev;
+		__entry->sevdetail =
+			FIELD_GET(CCIX_PER_LOG_DW1_SEV_UE_M |
+				  CCIX_PER_LOG_DW1_SEV_NO_COMM_M |
+				  CCIX_PER_LOG_DW1_SEV_DEGRADED_M |
+				  CCIX_PER_LOG_DW1_SEV_DEFFERABLE_M,
+						   mem->ccix_header[1]);
+		if (mem->header.validation_bits & 0x1)
+			__entry->source = mem->header.source_id;
+		else
+			__entry->source = ~0;
+		__entry->component = FIELD_GET(CCIX_PER_LOG_DW1_COMP_TYPE_M,
+					       mem->ccix_header[1]);
+		if (mem->ccix_header[1] & CCIX_PER_LOG_DW1_ADDR_VAL_M) {
+			__entry->pa = (u64)mem->ccix_header[2] << 32 |
+				(mem->ccix_header[3] & 0xfffffffc);
+			__entry->pa_mask_lsb = mem->ccix_header[4] & 0xff;
+		} else {
+			__entry->pa = ~0ull;
+			__entry->pa_mask_lsb = ~0;
+		}
+		__entry->vendor_data_length = ven_len ? ven_len - 4 : 0;
+		cper_ccix_mem_err_pack(&mem->mem_record, &__entry->data,
+				       __entry->vendor_data_length,
+				       __get_dynamic_array(vendor_data));
+	),
+
+	TP_printk("{%d} %s CCIX PER Memory Error in %s SevUE:%d SevNoComm:%d SevDegraded:%d SevDeferred:%d physical addr: %016llx (mask: %x) %s vendor:%s",
+		__entry->err_seq,
+		cper_severity_str(__entry->sev),
+		cper_ccix_comp_type_str(__entry->component),
+		  __entry->sevdetail & BIT(0) ? 1 : 0,
+		  __entry->sevdetail & BIT(1) ? 1 : 0,
+		  __entry->sevdetail & BIT(2) ? 1 : 0,
+		  __entry->sevdetail & BIT(3) ? 1 : 0,
+		__entry->pa,
+		__entry->pa_mask_lsb,

include/ras/ras_event.h TRACE_EVENT(aer_event, · line 422

Quoted 10 lines
include/ras/ras_event.h +7 -0
+		cper_ccix_mem_err_unpack(p, &__entry->data),
+		__print_hex(__get_dynamic_array(vendor_data),
+			    __entry->vendor_data_length)
+	)
+);
+#endif
+
 /*
  * memory-failure recovery action result event
  *

Cheers,
Mauro

maintainer
#3 lore raw

Re: [PATCH v5 1/6] efi / ras: CCIX Memory error reporting

In reply to Jonathan Cameron
Quoted 43 lines · Jonathan Cameron

CCIX defines a number of different error types
(See CCIX spec 1.0) and UEFI 2.8 defines a CPER record to allow for them to be reported when firmware first handling is in use. The last part of that record is a copy of the CCIX protocol error record which can provide very detailed information.

This patch introduces infrastructure and support for one of those error types, CCIX Memory Errors. Later patches will supply equivalent support for the other error types.

The variable length and content of the different messages makes a single tracepoint impractical. As such the RAS tracepoint introduced in this patch only covers the memory error. Additional trace points will be introduced for other error types along with their cper handling in the rest of this series.

Updated RAS daemon support to be posted shortly.
qemu injection patches also available but not currently planing to upstream those.

---

Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.

 drivers/acpi/apei/Kconfig        |   8 +
 drivers/acpi/apei/ghes.c         |  39 ++++
 drivers/firmware/efi/Kconfig     |   5 +
 drivers/firmware/efi/Makefile    |   1 +
 drivers/firmware/efi/cper-ccix.c | 362 +++++++++++++++++++++++++++++++
 drivers/firmware/efi/cper.c      |   6 +
 include/linux/cper.h             | 118 ++++++++++
 include/ras/ras_event.h          |  79 +++++++
 8 files changed, 618 insertions(+)
drivers/acpi/apei/Kconfig +0 -0
index 6b18f8bc7be3..e687b18dee34 100644

drivers/acpi/apei/Kconfig config ACPI_APEI_ERST_DEBUG · line 76

Quoted 9 lines
drivers/acpi/apei/Kconfig +6 -0
@@ -68,3 +68,11 @@ config ACPI_APEI_ERST_DEBUG
 	  error information to and from a persistent store. Enable this
 	  if you want to debugging and testing the ERST kernel support
 	  and firmware implementation.
+
+config ACPI_APEI_CCIX
+       bool "APEI CCIX error recovery support"
+       depends on ACPI_APEI && MEMORY_FAILURE
+       help
+	 CCIX has a number of defined error types. This option enables

Write out that "CCIX" acronym at least in the help here. I had to go visit wikipedia to find out what it is.

Quoted 8 lines
drivers/acpi/apei/Kconfig +2 -0
+	 the handling of CPER records generated by a firmware performing
+	 firmware first error handling of these CCIX errors.
drivers/acpi/apei/ghes.c +0 -0
index 777f6f7122b4..75a177ae9de3 100644
@@ -490,6 +490,42 @@ static void ghes_handle_aer(struct acpi_hest_generic_data *gdata)
 #endif

drivers/acpi/apei/ghes.c static void ghes_handle_aer(struct acpi_hest_generic_data *gdata) · line 500

Quoted 10 lines
drivers/acpi/apei/ghes.c +8 -0
 }
 
+static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
+{
+#ifdef CONFIG_ACPI_APEI_CCIX
+	struct cper_sec_ccix_header *header = acpi_hest_get_payload(gdata);
+	__u32 *dw;
+	enum ccix_per_type per_type;
+	static u32 err_seq;
+	void *payload;

Please sort function local variables declaration in a reverse christmas tree order:

	<type A> longest_variable_name;
	<type B> shorter_var_name;
	<type C> even_shorter;
	<type D> i;

and check that for all your functions.

drivers/acpi/apei/ghes.c static void ghes_handle_aer(struct acpi_hest_generic_data *gdata) · line 504

Quoted 3 lines
drivers/acpi/apei/ghes.c +3 -0
+	/* Check if space for CCIX CPER header and 8 DW of a PER log header */
+	if (gdata->error_data_length <
+	    sizeof(*header) + CCIX_PER_LOG_HEADER_DWS * sizeof(__u32))

Put that size in a local variable so that the if-statement is more readable.

drivers/acpi/apei/ghes.c static void ghes_handle_aer(struct acpi_hest_generic_data *gdata) · line 507

Quoted 3 lines
drivers/acpi/apei/ghes.c +3 -0
+		return;
+
+	if ((header->validation_bits & CPER_CCIX_VALID_PER_LOG) == 0)
	if (!( ...
Quoted 39 lines
drivers/acpi/apei/ghes.c +24 -0
+		return;
+
+	dw = (__u32 *)(header + 1);
+
+	per_type = FIELD_GET(CCIX_PER_LOG_DW1_PER_TYPE_M, dw[1]);
+	payload = acpi_hest_get_payload(gdata);
+
+	switch (per_type) {
+	case CCIX_MEMORY_ERROR:
+		trace_ccix_memory_error_event(payload, err_seq, sev,
+					      ccix_mem_err_ven_len_get(payload));
+		break;
+	default:
+		/* Unknown error type */
+		pr_info("CCIX error of unknown or vendor defined type\n");
+		break;
+	}
+	err_seq++;
+#endif
+}
+
 static void ghes_do_proc(struct ghes *ghes,
 			 const struct acpi_hest_generic_status *estatus)
 {
@@ -520,6 +556,9 @@ static void ghes_do_proc(struct ghes *ghes,
 		else if (guid_equal(sec_type, &CPER_SEC_PCIE)) {
 			ghes_handle_aer(gdata);
 		}
+		else if (guid_equal(sec_type, &CPER_SEC_CCIX)) {
+			ghes_handle_ccix_per(gdata, estatus->error_severity);
+		}
 		else if (guid_equal(sec_type, &CPER_SEC_PROC_ARM)) {
 			struct cper_sec_proc_arm *err = acpi_hest_get_payload(gdata);
 
drivers/firmware/efi/Kconfig +0 -0
index b248870a9806..096e693a9522 100644

drivers/firmware/efi/Kconfig config UEFI_CPER_X86 · line 215

Quoted 7 lines
drivers/firmware/efi/Kconfig +4 -0
@@ -209,6 +209,11 @@ config UEFI_CPER_X86
 	depends on UEFI_CPER && X86
 	default y
 
+config UEFI_CPER_CCIX
+       bool
+       depends on UEFI_CPER
+       default y

Err, this whole CCIX gunk will get built on all systems now which wanna enable UEFI_CPER but how many of those do *actually* have those accelerators and use CCIX?

If not the majority, then this needs to be user-configurable so that most people who don't have it, do not have to enable it and bloat their kernels unnecessarily.

And all that code should be surrounded with ifdefs so that it is not even there on the majority of the systems.

...

Quoted 13 lines
include/ras/ras_event.h +1 -0
index 36c5c5e38c1d..560e55958561 100644
@@ -14,6 +14,7 @@
 #include <linux/cper.h>
 #include <linux/mm.h>
 
+#include <linux/bitfield.h>
 /*
  * MCE Extended Error Log trace event
  *

include/ras/ras_event.h TRACE_EVENT(aer_event, · line 342

Quoted 4 lines
include/ras/ras_event.h +1 -0
@@ -338,6 +339,84 @@ TRACE_EVENT(aer_event,
 			"Not available")
 );
 
+#if defined(CONFIG_ACPI_APEI_CCIX)

#ifdef doesn't work here?

Quoted 5 lines
include/ras/ras_event.h +5 -0
+/*
+ * CCIX PER log memory error trace event
+ *
+ * These events are generated when hardware detects a corrected or
+ * uncorrected event.

include/ras/ras_event.h TRACE_EVENT(aer_event, · line 357

Quoted 10 lines
include/ras/ras_event.h +10 -0
+ *
+ * Some elements of the record are not included
+ * - PER version (tracepoint should remain compatible across versions)
+ * - Multiple Error
+ */
+TRACE_EVENT(ccix_memory_error_event,
+	TP_PROTO(struct cper_ccix_mem_error *mem,
+		 u32 err_seq,
+		 u8 sev,
+		 u16 ven_len),

In any case, you have a lot of duplication between all those tracepoints and I'm wondering why can't you have a *single* TRACE_EVENT() definition which can deal with all the possible CCIX events?

Or have a TRACE_EVENT_CLASS()?

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#4 lore raw

PATCH v5 2/6 efi / ras: CCIX Cache error reporting

In reply to Jonathan Cameron

As CCIX Request Agents have fully cache coherent caches, the CCIX 1.0 Base Specification defines detailed error reporting for these caches.

A CCIX cache error is reported via a CPER record as defined in the UEFI 2.8 specification. The PER log section is defined in the CCIX 1.0 Base Specification.

---
Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.

drivers/acpi/apei/ghes.c | 4 +

 drivers/firmware/efi/cper-ccix.c | 170 +++++++++++++++++++++++++++++++
 include/linux/cper.h             |  57 +++++++++++
 include/ras/ras_event.h          |  67 ++++++++++++
 4 files changed, 298 insertions(+)
drivers/acpi/apei/ghes.c +4 -0
index 75a177ae9de3..c99a4216b67d 100644
@@ -517,6 +517,10 @@ static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
 		trace_ccix_memory_error_event(payload, err_seq, sev,
 					      ccix_mem_err_ven_len_get(payload));
 		break;
+	case CCIX_CACHE_ERROR:
+		trace_ccix_cache_error_event(payload, err_seq, sev,
+					     ccix_cache_err_ven_len_get(payload));
+		break;
 	default:
 		/* Unknown error type */
 		pr_info("CCIX error of unknown or vendor defined type\n");
drivers/firmware/efi/cper-ccix.c +170 -0

185 lines not on this page · raw message

include/linux/cper.h +57 -0

69 lines not on this page · raw message

include/ras/ras_event.h +67 -0

73 lines not on this page · raw message

Signature · 1 line
2.20.1
#5 lore raw

Re: [PATCH v5 2/6] efi / ras: CCIX Cache error reporting

Em Thu, 14 Nov 2019 21:39:15 +0800
Jonathan Cameron <[email protected]> escreveu:

Quoted 168 lines

As CCIX Request Agents have fully cache coherent caches, the CCIX 1.0 Base Specification defines detailed error reporting for these caches.

A CCIX cache error is reported via a CPER record as defined in the UEFI 2.8 specification. The PER log section is defined in the CCIX 1.0 Base Specification.

---
Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.

drivers/acpi/apei/ghes.c | 4 +

 drivers/firmware/efi/cper-ccix.c | 170 +++++++++++++++++++++++++++++++
 include/linux/cper.h             |  57 +++++++++++
 include/ras/ras_event.h          |  67 ++++++++++++
 4 files changed, 298 insertions(+)
drivers/acpi/apei/ghes.c +4 -0
index 75a177ae9de3..c99a4216b67d 100644
@@ -517,6 +517,10 @@ static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
 		trace_ccix_memory_error_event(payload, err_seq, sev,
 					      ccix_mem_err_ven_len_get(payload));
 		break;
+	case CCIX_CACHE_ERROR:
+		trace_ccix_cache_error_event(payload, err_seq, sev,
+					     ccix_cache_err_ven_len_get(payload));
+		break;
 	default:
 		/* Unknown error type */
 		pr_info("CCIX error of unknown or vendor defined type\n");
drivers/firmware/efi/cper-ccix.c +111 -0
index 588a7aaa6a72..7d29ae691cf5 100644
@@ -293,6 +293,110 @@ static int cper_ccix_mem_err_details(const char *pfx,
 	return 0;
 }
 
+static const char * const ccix_cache_type_strs[] = {
+	"Instruction Cache",
+	"Data Cache",
+	"Generic / Unified Cache",
+	"Snoop Filter Directory",
+};
+
+static const char *cper_ccix_cache_type_str(__u8 type)
+{
+	return type < ARRAY_SIZE(ccix_cache_type_strs) ?
+		ccix_cache_type_strs[type] : "Reserved";
+}
+
+static const char * const ccix_cache_err_type_strs[] = {
+	"Data",
+	"Tag",
+	"Timeout",
+	"Hang",
+	"Data Lost",
+	"Invalid Address",
+};
+
+const char *cper_ccix_cache_err_type_str(__u8 error_type)
+{
+	return error_type < ARRAY_SIZE(ccix_cache_err_type_strs) ?
+		ccix_cache_err_type_strs[error_type] : "Reserved";
+}
+
+static const char * const ccix_cache_err_op_strs[] = {
+	"Generic",
+	"Generic Read",
+	"Generic Write",
+	"Data Read",
+	"Data Write",
+	"Instruction Fetch",
+	"Prefetch",
+	"Eviction",
+	"Snooping",
+	"Snooped",
+	"Management / Command Error",
+};
+
+static const char *cper_ccix_cache_err_op_str(__u8 op)
+{
+	return op < ARRAY_SIZE(ccix_cache_err_op_strs) ?
+		ccix_cache_err_op_strs[op] : "Reserved";
+}
+
+static int cper_ccix_cache_err_details(const char *pfx,
+				       struct acpi_hest_generic_data *gdata)
+{
+	struct cper_ccix_cache_error *full_cache_err;
+	struct cper_sec_ccix_cache_error *cache_err;
+	u16 vendor_data_len;
+	int i;
+
+	if (gdata->error_data_length < sizeof(*full_cache_err))
+		return -ENOSPC;
+
+	full_cache_err = acpi_hest_get_payload(gdata);
+
+	cache_err = &full_cache_err->cache_record;
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_TYPE_VALID)
+		printk("%s""Cache Type: %s\n", pfx,
+		       cper_ccix_cache_type_str(cache_err->cache_type));
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_OP_VALID)
+		printk("%s""Operation: %s\n", pfx,
+		       cper_ccix_cache_err_op_str(cache_err->op_type));
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_CACHE_ERR_TYPE_VALID)
+		printk("%s""Cache Error Type: %s\n", pfx,
+		       cper_ccix_cache_err_type_str(cache_err->cache_error_type));
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_LEVEL_VALID)
+		printk("%s""Level: %d\n", pfx, cache_err->cache_level);
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_SET_VALID)
+		printk("%s""Set: %d\n", pfx, cache_err->set);
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_WAY_VALID)
+		printk("%s""Way: %d\n", pfx, cache_err->way);
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_INSTANCE_ID_VALID)
+		printk("%s""Instance ID: %d\n", pfx, cache_err->instance);
+
+	if (cache_err->validation_bits & CCIX_CACHE_ERR_VENDOR_DATA_VALID) {
+		if (gdata->error_data_length < sizeof(*full_cache_err) + 4)
+			return -ENOSPC;
+
+		vendor_data_len = cache_err->vendor_data[0] & GENMASK(15, 0);
+		if (gdata->error_data_length <
+		    sizeof(*full_cache_err) + vendor_data_len)
+			return -ENOSPC;
+
+		for (i = 0; i < vendor_data_len / 4 - 1; i++)
+			printk("%s""Vendor%d: 0x%08x\n", pfx, i,
+			       cache_err->vendor_data[i + 1]);
+	}
+
+	return 0;
+}
+
 int cper_print_ccix_per(const char *pfx, struct acpi_hest_generic_data *gdata)
 {
 	struct cper_sec_ccix_header *header = acpi_hest_get_payload(gdata);
@@ -354,9 +458,75 @@ int cper_print_ccix_per(const char *pfx, struct acpi_hest_generic_data *gdata)
 	switch (per_type) {
 	case CCIX_MEMORY_ERROR:
 		return cper_ccix_mem_err_details(pfx, gdata);
+	case CCIX_CACHE_ERROR:
+		return cper_ccix_cache_err_details(pfx, gdata);
 	default:
 		/* Vendor defined so no formatting be done */
 		break;
 	}
 	return 0;
 }
+
+void cper_ccix_cache_err_pack(const struct cper_sec_ccix_cache_error *cache_record,
+			      struct cper_ccix_cache_err_compact *ccache_err,
+			      const u16 vendor_data_len,
+			      u8 *vendor_data)

drivers/firmware/efi/cper-ccix.c int cper_print_ccix_per(const char *pfx, struct acpi_hest_generic_data *gdata) · line 483

Quoted 10 lines
drivers/firmware/efi/cper-ccix.c +10 -0
+{
+	ccache_err->validation_bits = cache_record->validation_bits;
+	ccache_err->set = cache_record->set;
+	ccache_err->way = cache_record->way;
+	ccache_err->cache_type = cache_record->cache_type;
+	ccache_err->op_type = cache_record->op_type;
+	ccache_err->cache_error_type = cache_record->cache_error_type;
+	ccache_err->cache_level = cache_record->cache_level;
+	ccache_err->instance = cache_record->instance;
+	memcpy(vendor_data, &cache_record->vendor_data[1], vendor_data_len);

Same comments as the ones I made on the previous patch.

Quoted 192 lines
drivers/firmware/efi/cper-ccix.c +49 -0
+}
+
+static int cper_ccix_err_cache_location(struct cper_ccix_cache_err_compact *ccache_err,
+					char msg[CPER_REC_LEN])
+{
+	u32 len = CPER_REC_LEN - 1;
+	u32 n = 0;
+
+	if (!msg)
+		return 0;
+
+	if (ccache_err->validation_bits & CCIX_CACHE_ERR_CACHE_ERR_TYPE_VALID)
+		n = snprintf(msg, len, "Error: %s ",
+			     cper_ccix_cache_err_type_str(ccache_err->cache_error_type));
+
+	if (ccache_err->validation_bits & CCIX_CACHE_ERR_TYPE_VALID)
+		n += snprintf(msg + n, len - n, "Type: %s ",
+			      cper_ccix_cache_type_str(ccache_err->cache_type));
+
+	if (ccache_err->validation_bits & CCIX_CACHE_ERR_OP_VALID)
+		n += snprintf(msg + n, len - n, "Op: %s ",
+			      cper_ccix_cache_err_op_str(ccache_err->op_type));
+
+	if (ccache_err->validation_bits & CCIX_CACHE_ERR_LEVEL_VALID)
+		n += snprintf(msg + n, len - n, "Level: %d ",
+			      ccache_err->cache_level);
+	if (ccache_err->validation_bits & CCIX_CACHE_ERR_SET_VALID)
+		n += snprintf(msg + n, len - n, "Set: %d ", ccache_err->set);
+	if (ccache_err->validation_bits & CCIX_CACHE_ERR_WAY_VALID)
+		n += snprintf(msg + n, len - n, "Way: %d ", ccache_err->way);
+	if (ccache_err->validation_bits & CCIX_CACHE_ERR_INSTANCE_ID_VALID)
+		n += snprintf(msg + n, len - n, "Instance: %d ",
+			      ccache_err->instance);
+
+	return n;
+}
+
+const char *cper_ccix_cache_err_unpack(struct trace_seq *p,
+				       struct cper_ccix_cache_err_compact *ccache_err)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+
+	if (cper_ccix_err_cache_location(ccache_err, rcd_decode_str))
+		trace_seq_printf(p, "%s", rcd_decode_str);
+
+	trace_seq_putc(p, '\0');
+
+	return ret;
+}
include/linux/cper.h +57 -0
index df7a34c3ba4f..eef254b8b8b7 100644
@@ -627,6 +627,54 @@ struct cper_ccix_mem_err_compact {
 	__u8	fru;
 };
 
+struct cper_sec_ccix_cache_error {
+	__u32	validation_bits;
+#define CCIX_CACHE_ERR_TYPE_VALID		BIT(0)
+#define CCIX_CACHE_ERR_OP_VALID			BIT(1)
+#define CCIX_CACHE_ERR_CACHE_ERR_TYPE_VALID	BIT(2)
+#define CCIX_CACHE_ERR_LEVEL_VALID		BIT(3)
+#define CCIX_CACHE_ERR_SET_VALID		BIT(4)
+#define CCIX_CACHE_ERR_WAY_VALID		BIT(5)
+#define CCIX_CACHE_ERR_INSTANCE_ID_VALID	BIT(6)
+#define CCIX_CACHE_ERR_VENDOR_DATA_VALID	BIT(7)
+	__u16	length; /* Includes vendor specific log info */
+	__u8	cache_type;
+	__u8	op_type;
+	__u8	cache_error_type;
+	__u8	cache_level;
+	__u32	set;
+	__u32	way;
+	__u8	instance;
+	__u8	reserved;
+	__u32	vendor_data[];
+};
+
+struct cper_ccix_cache_error {
+	struct cper_sec_ccix_header header;
+	__u32 ccix_header[CCIX_PER_LOG_HEADER_DWS];
+	struct cper_sec_ccix_cache_error cache_record;
+};
+
+static inline u16 ccix_cache_err_ven_len_get(struct cper_ccix_cache_error *cache_err)
+{
+	if (cache_err->cache_record.validation_bits &
+	    CCIX_CACHE_ERR_VENDOR_DATA_VALID)
+		return cache_err->cache_record.vendor_data[0] & 0xFFFF;
+	else
+		return 0;
+}
+
+struct cper_ccix_cache_err_compact {
+	__u32	validation_bits;
+	__u32	set;
+	__u32	way;
+	__u8	cache_type;
+	__u8	op_type;
+	__u8	cache_error_type;
+	__u8	cache_level;
+	__u8	instance;
+};
+
 /* Reset to default packing */
 #pragma pack()
 
@@ -649,6 +697,15 @@ const char *cper_ccix_mem_err_unpack(struct trace_seq *p,
 				     struct cper_ccix_mem_err_compact *cmem_err);
 const char *cper_ccix_mem_err_type_str(unsigned int error_type);
 const char *cper_ccix_comp_type_str(u8 comp_type);
+
+void cper_ccix_cache_err_pack(const struct cper_sec_ccix_cache_error *cache_record,
+			      struct cper_ccix_cache_err_compact *ccache_err,
+			      const u16 vendor_data_len,
+			      u8 *vendor_data);
+const char *cper_ccix_cache_err_unpack(struct trace_seq *p,
+				       struct cper_ccix_cache_err_compact *ccache_err);
+const char *cper_ccix_cache_err_type_str(__u8 error_type);
+
 struct acpi_hest_generic_data;
 int cper_print_ccix_per(const char *pfx,
 			struct acpi_hest_generic_data *gdata);
include/ras/ras_event.h +60 -0
index 560e55958561..4a158820074c 100644
@@ -415,6 +415,73 @@ TRACE_EVENT(ccix_memory_error_event,
 			    __entry->vendor_data_length)
 	)
 );
+
+TRACE_EVENT(ccix_cache_error_event,
+	TP_PROTO(struct cper_ccix_cache_error *err,
+		 u32 err_seq,
+		 u8 sev,
+		 u16 ven_len),
+
+	TP_ARGS(err, err_seq, sev, ven_len),
+
+	TP_STRUCT__entry(
+		__field(u32, err_seq)
+		__field(u8, sev)
+		__field(u8, sevdetail)
+		__field(u8, source)
+		__field(u8, component)
+		__field(u64, pa)
+		__field(u8, pa_mask_lsb)
+		__field_struct(struct cper_ccix_cache_err_compact, data)
+		__field(u16, vendor_data_length)
+		__dynamic_array(u8, vendor_data, ven_len)
+	),
+
+	TP_fast_assign(
+		__entry->err_seq = err_seq;
+
+		__entry->sev = sev;
+		__entry->sevdetail = FIELD_GET(CCIX_PER_LOG_DW1_SEV_UE_M |
+			CCIX_PER_LOG_DW1_SEV_NO_COMM_M |
+			CCIX_PER_LOG_DW1_SEV_DEGRADED_M |
+			CCIX_PER_LOG_DW1_SEV_DEFFERABLE_M,
+			err->ccix_header[1]);
+		if (err->header.validation_bits & 0x1)
+			__entry->source = err->header.source_id;
+		else
+			__entry->source = ~0;
+		__entry->component = FIELD_GET(CCIX_PER_LOG_DW1_COMP_TYPE_M,
+					       err->ccix_header[1]);
+		if (err->ccix_header[1] & CCIX_PER_LOG_DW1_ADDR_VAL_M) {
+			__entry->pa = (u64)err->ccix_header[2] << 32 |
+				(err->ccix_header[3] & 0xfffffffc);
+			__entry->pa_mask_lsb = err->ccix_header[4] & 0xff;
+		} else {
+			__entry->pa = ~0ull;
+			__entry->pa_mask_lsb = ~0;
+		}
+
+		__entry->vendor_data_length = ven_len ? ven_len - 4 : 0;
+		cper_ccix_cache_err_pack(&err->cache_record, &__entry->data,
+					 __entry->vendor_data_length,
+					 __get_dynamic_array(vendor_data));
+	),
+
+	TP_printk("{%d} %s CCIX PER Cache Error in %s SevUE:%d SevNoComm:%d SevDegraded:%d SevDeferred:%d physical addr: %016llx (mask: %x) %s vendor:%s",
+		__entry->err_seq,
+		cper_severity_str(__entry->sev),
+		cper_ccix_comp_type_str(__entry->component),
+		__entry->sevdetail & BIT(0) ? 1 : 0,
+		__entry->sevdetail & BIT(1) ? 1 : 0,
+		__entry->sevdetail & BIT(2) ? 1 : 0,
+		__entry->sevdetail & BIT(3) ? 1 : 0,

include/ras/ras_event.h TRACE_EVENT(ccix_memory_error_event, · line 487

Quoted 10 lines
include/ras/ras_event.h +7 -0
+		__entry->pa,
+		__entry->pa_mask_lsb,
+		cper_ccix_cache_err_unpack(p, &__entry->data),
+		__print_hex(__get_dynamic_array(vendor_data),
+			    __entry->vendor_data_length)
+	)
+);
 #endif
 
 /*

Cheers,
Mauro

#6 lore raw

PATCH v5 3/6 efi / ras: CCIX Address Translation Cache error reporting

In reply to Jonathan Cameron

CCIX devices tend to make heavy use of ATCs. The CCIX base specification defines a protocol error message (PER) that describes errors reported by such caches. The UEFI 2.8 specification includes a CCIX CPER record for firmware first handling to report these errors to the operating system.

This patch is very similar to the support previously added for CCIX Memory Errors and provides both logging and RAS tracepoint for this error class.

---
Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.

 drivers/acpi/apei/ghes.c         |  4 ++
 drivers/firmware/efi/cper-ccix.c | 84 ++++++++++++++++++++++++++++++++
 include/linux/cper.h             | 39 +++++++++++++++
 include/ras/ras_event.h          | 66 +++++++++++++++++++++++++
 4 files changed, 193 insertions(+)
drivers/acpi/apei/ghes.c +4 -0
index c99a4216b67d..ba73d3a5d564 100644
@@ -521,6 +521,10 @@ static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
 		trace_ccix_cache_error_event(payload, err_seq, sev,
 					     ccix_cache_err_ven_len_get(payload));
 		break;
+	case CCIX_ATC_ERROR:
+		trace_ccix_atc_error_event(payload, err_seq, sev,
+					   ccix_atc_err_ven_len_get(payload));
+		break;
 	default:
 		/* Unknown error type */
 		pr_info("CCIX error of unknown or vendor defined type\n");
drivers/firmware/efi/cper-ccix.c +84 -0

99 lines not on this page · raw message

include/linux/cper.h +39 -0
index eef254b8b8b7..6bb603e9a97a 100644
@@ -675,6 +675,38 @@ struct cper_ccix_cache_err_compact {
 	__u8	instance;
 };
 
+struct cper_sec_ccix_atc_error {
+	__u32	validation_bits;
+#define CCIX_ATC_ERR_OP_VALID			BIT(0)
+#define CCIX_ATC_ERR_INSTANCE_ID_VALID		BIT(1)
+#define CCIX_ATC_ERR_VENDOR_DATA_VALID		BIT(2)
+	__u16	length; /* Includes vendor specific log info */
+	__u8	op_type;
+	__u8	instance;
+	__u32	reserved;
+	__u32	vendor_data[];
+};
+
+struct cper_ccix_atc_error {
+	struct cper_sec_ccix_header header;
+	__u32 ccix_header[CCIX_PER_LOG_HEADER_DWS];
+	struct cper_sec_ccix_atc_error atc_record;
+};
+
+static inline u16 ccix_atc_err_ven_len_get(struct cper_ccix_atc_error *atc_err)
+{
+	if (atc_err->atc_record.validation_bits & CCIX_ATC_ERR_VENDOR_DATA_VALID)
+		return atc_err->atc_record.vendor_data[0] & 0xFFFF;
+	else
+		return 0;
+}
+
+struct cper_ccix_atc_err_compact {
+	__u32	validation_bits;
+	__u8	op_type;
+	__u8	instance;
+};
+
 /* Reset to default packing */
 #pragma pack()
 
@@ -706,6 +738,13 @@ const char *cper_ccix_cache_err_unpack(struct trace_seq *p,
 				       struct cper_ccix_cache_err_compact *ccache_err);
 const char *cper_ccix_cache_err_type_str(__u8 error_type);
 
+void cper_ccix_atc_err_pack(const struct cper_sec_ccix_atc_error *atc_record,
+			    struct cper_ccix_atc_err_compact *catc_err,
+			    const u16 vendor_data_len,
+			    u8 *vendor_data);
+const char *cper_ccix_atc_err_unpack(struct trace_seq *p,
+				     struct cper_ccix_atc_err_compact *catc_err);
+
 struct acpi_hest_generic_data;
 int cper_print_ccix_per(const char *pfx,
 			struct acpi_hest_generic_data *gdata);
include/ras/ras_event.h +66 -0

72 lines not on this page · raw message

Signature · 1 line
2.20.1
#7 lore raw

PATCH v5 4/6 efi / ras: CCIX Port error reporting

In reply to Jonathan Cameron

The CCIX 1.0 base specification defines a CCIX protocol layer port. The specification provides a mechanism for detailed error logging for these ports. The UEFI 2.8 specification includes a CCIX CPER record for firmware first handling to report these errors to the operating system.

This patch is very similar to the support previously added for for CCIX Memory Errors and provides both logging and RAS tracepoint for this error class.

Note that we do not currently print the raw message in the trace point print functions.

---
Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.
* Clarified comment on not printing the CCIX raw message when pretty printing the tracepoint.

 drivers/acpi/apei/ghes.c         |   4 +
 drivers/firmware/efi/cper-ccix.c | 123 +++++++++++++++++++++++++++++++
 include/linux/cper.h             |  42 +++++++++++
 include/ras/ras_event.h          |  66 +++++++++++++++++
 4 files changed, 235 insertions(+)
drivers/acpi/apei/ghes.c +4 -0
index ba73d3a5d564..7455db97319c 100644
@@ -525,6 +525,10 @@ static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
 		trace_ccix_atc_error_event(payload, err_seq, sev,
 					   ccix_atc_err_ven_len_get(payload));
 		break;
+	case CCIX_PORT_ERROR:
+		trace_ccix_port_error_event(payload, err_seq, sev,
+					    ccix_port_err_ven_len_get(payload));
+		break;
 	default:
 		/* Unknown error type */
 		pr_info("CCIX error of unknown or vendor defined type\n");
drivers/firmware/efi/cper-ccix.c +123 -0

138 lines not on this page · raw message

include/linux/cper.h +42 -0

54 lines not on this page · raw message

include/ras/ras_event.h +66 -0

72 lines not on this page · raw message

Signature · 1 line
2.20.1
#8 lore raw

PATCH v5 5/6 efi / ras: CCIX Link error reporting

In reply to Jonathan Cameron

The CCIX 1.0 Base Specification defines a protocol layer CCIX link and related error reporting mechanism. The UEFI 2.8 specification includes a CCIX CPER record for firmware first handling to report these errors to the operating system.

This patch is very similar to the support previously added for CCIX Memory Errors and provides both logging and RAS tracepoint for this error class.

Note that we currently don't print the raw message in the trace point print functions.

---
Changes since v4
* Fixed length array parameters
* Take advantage of being first print to string and simplify code.
* Fix buffer overflow risk in snprintf calls.
* Clarify comments on not printing the raw message for tracepoints.

drivers/acpi/apei/ghes.c | 4 +

 drivers/firmware/efi/cper-ccix.c | 141 +++++++++++++++++++++++++++++++
 include/linux/cper.h             |  48 +++++++++++
 include/ras/ras_event.h          |  65 ++++++++++++++
 4 files changed, 258 insertions(+)
drivers/acpi/apei/ghes.c +4 -0
index 7455db97319c..22df8c14ec13 100644
@@ -529,6 +529,10 @@ static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
 		trace_ccix_port_error_event(payload, err_seq, sev,
 					    ccix_port_err_ven_len_get(payload));
 		break;
+	case CCIX_LINK_ERROR:
+		trace_ccix_link_error_event(payload, err_seq, sev,
+					    ccix_link_err_ven_len_get(payload));
+		break;
 	default:
 		/* Unknown error type */
 		pr_info("CCIX error of unknown or vendor defined type\n");
drivers/firmware/efi/cper-ccix.c +141 -0

156 lines not on this page · raw message

include/linux/cper.h +48 -0

60 lines not on this page · raw message

include/ras/ras_event.h +65 -0

71 lines not on this page · raw message

Signature · 1 line
2.20.1
#9 lore raw

PATCH v5 6/6 efi / ras: CCIX Agent internal error reporting

In reply to Jonathan Cameron

The CCIX 1.0 Base specification defines an internal agent error, for which the specific data present afte the header is vendor defined.

---
Changes since v4
* none

 drivers/acpi/apei/ghes.c         |  4 ++
 drivers/firmware/efi/cper-ccix.c | 43 +++++++++++++++++++++
 include/linux/cper.h             | 29 +++++++++++++++
 include/ras/ras_event.h          | 64 ++++++++++++++++++++++++++++++++
 4 files changed, 140 insertions(+)
drivers/acpi/apei/ghes.c +4 -0
index 22df8c14ec13..5a75fb0374dd 100644
@@ -533,6 +533,10 @@ static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
 		trace_ccix_link_error_event(payload, err_seq, sev,
 					    ccix_link_err_ven_len_get(payload));
 		break;
+	case CCIX_AGENT_INTERNAL_ERROR:
+		trace_ccix_agent_error_event(payload, err_seq, sev,
+					     ccix_agent_err_ven_len_get(payload));
+		break;
 	default:
 		/* Unknown error type */
 		pr_info("CCIX error of unknown or vendor defined type\n");
drivers/firmware/efi/cper-ccix.c +43 -0

58 lines not on this page · raw message

include/linux/cper.h +29 -0
index d35be55351e3..373c1d387a70 100644
@@ -783,6 +783,30 @@ struct cper_ccix_link_err_compact {
 	__u8	credit_type;
 };
 
+struct cper_sec_ccix_agent_err {
+	__u32	validation_bits;
+#define CCIX_AGENT_INTERNAL_ERR_VENDOR_DATA_VALID	BIT(0)
+	__u32	vendor_data[];
+};
+
+struct cper_ccix_agent_err {
+	struct cper_sec_ccix_header header;
+	__u32 ccix_header[CCIX_PER_LOG_HEADER_DWS];
+	struct cper_sec_ccix_agent_err agent_record;
+};
+
+static inline u16 ccix_agent_err_ven_len_get(struct cper_ccix_agent_err *agent_err)
+{
+	if (agent_err->agent_record.validation_bits & CCIX_AGENT_INTERNAL_ERR_VENDOR_DATA_VALID)
+		return agent_err->agent_record.vendor_data[0] & 0xFFFF;
+	else
+		return 0;
+}
+
+struct cper_ccix_agent_err_compact {
+	__u32	validation_bits;
+};
+
 /* Reset to default packing */
 #pragma pack()
 
@@ -835,6 +859,11 @@ void cper_ccix_link_err_pack(const struct cper_sec_ccix_link_error *link_record,
 const char *cper_ccix_link_err_unpack(struct trace_seq *p,
 				      struct cper_ccix_link_err_compact *clink_err);
 
+void cper_ccix_agent_err_pack(const struct cper_sec_ccix_agent_err *agent_record,
+			      struct cper_ccix_agent_err_compact *cagent_err,
+			      const u16 vendor_data_len,
+			      u8 *vendor_data);
+
 struct acpi_hest_generic_data;
 int cper_print_ccix_per(const char *pfx,
 			struct acpi_hest_generic_data *gdata);
include/ras/ras_event.h +64 -0

70 lines not on this page · raw message

Signature · 1 line
2.20.1
#10 lore raw

Re: [PATCH v5 6/6] efi / ras: CCIX Agent internal error reporting

Em Thu, 14 Nov 2019 21:39:19 +0800
Jonathan Cameron <[email protected]> escreveu:

Quoted 4 lines

The CCIX 1.0 Base specification defines an internal agent error, for which the specific data present afte the header is vendor defined.

Patches 3 to 6 just repeats the same things I mentioned already on patch 1.

Except for that, the series look ready for merging on my PoV.

Regards,
Mauro

Quoted 203 lines

---
Changes since v4
* none

 drivers/acpi/apei/ghes.c         |  4 ++
 drivers/firmware/efi/cper-ccix.c | 43 +++++++++++++++++++++
 include/linux/cper.h             | 29 +++++++++++++++
 include/ras/ras_event.h          | 64 ++++++++++++++++++++++++++++++++
 4 files changed, 140 insertions(+)
drivers/acpi/apei/ghes.c +4 -0
index 22df8c14ec13..5a75fb0374dd 100644
@@ -533,6 +533,10 @@ static void ghes_handle_ccix_per(struct acpi_hest_generic_data *gdata, int sev)
 		trace_ccix_link_error_event(payload, err_seq, sev,
 					    ccix_link_err_ven_len_get(payload));
 		break;
+	case CCIX_AGENT_INTERNAL_ERROR:
+		trace_ccix_agent_error_event(payload, err_seq, sev,
+					     ccix_agent_err_ven_len_get(payload));
+		break;
 	default:
 		/* Unknown error type */
 		pr_info("CCIX error of unknown or vendor defined type\n");
drivers/firmware/efi/cper-ccix.c +43 -0
index ea17fb9140f9..d95fddf17bd3 100644
@@ -587,6 +587,38 @@ static int cper_ccix_link_err_details(const char *pfx,
 	return 0;
 }
 
+static int cper_ccix_agent_err_details(const char *pfx,
+				       struct acpi_hest_generic_data *gdata)
+{
+	struct cper_ccix_agent_err *full_agent_err;
+	struct cper_sec_ccix_agent_err *agent_err;
+	u16 vendor_data_len;
+	int i;
+
+	if (gdata->error_data_length < sizeof(*full_agent_err))
+		return -ENOSPC;
+
+	full_agent_err = acpi_hest_get_payload(gdata);
+
+	agent_err = &full_agent_err->agent_record;
+
+	if (agent_err->validation_bits & CCIX_AGENT_INTERNAL_ERR_VENDOR_DATA_VALID) {
+		if (gdata->error_data_length < sizeof(*full_agent_err) + 4)
+			return -ENOSPC;
+
+		vendor_data_len = agent_err->vendor_data[0] & GENMASK(15, 0);
+		if (gdata->error_data_length <
+		    sizeof(*full_agent_err) + vendor_data_len)
+			return -ENOSPC;
+
+		for (i = 0; i < vendor_data_len/4 - 1; i++)
+			printk("%s""Vendor%d: 0x%08x\n", pfx, i,
+			       agent_err->vendor_data[i + 1]);
+	}
+
+	return 0;
+}
+
 int cper_print_ccix_per(const char *pfx, struct acpi_hest_generic_data *gdata)
 {
 	struct cper_sec_ccix_header *header = acpi_hest_get_payload(gdata);
@@ -656,6 +688,8 @@ int cper_print_ccix_per(const char *pfx, struct acpi_hest_generic_data *gdata)
 		return cper_ccix_port_err_details(pfx, gdata);
 	case CCIX_LINK_ERROR:
 		return cper_ccix_link_err_details(pfx, gdata);
+	case CCIX_AGENT_INTERNAL_ERROR:
+		return cper_ccix_agent_err_details(pfx, gdata);
 	default:
 		/* Vendor defined so no formatting be done */
 		break;
@@ -878,3 +912,12 @@ const char *cper_ccix_link_err_unpack(struct trace_seq *p,
 
 	return ret;
 }
+
+void cper_ccix_agent_err_pack(const struct cper_sec_ccix_agent_err *agent_record,
+			      struct cper_ccix_agent_err_compact *cagent_err,
+			      const u16 vendor_data_len,
+			      u8 *vendor_data)
+{
+	cagent_err->validation_bits = agent_record->validation_bits;
+	memcpy(vendor_data, &agent_record->vendor_data[1], vendor_data_len);
+}
include/linux/cper.h +29 -0
index d35be55351e3..373c1d387a70 100644
@@ -783,6 +783,30 @@ struct cper_ccix_link_err_compact {
 	__u8	credit_type;
 };
 
+struct cper_sec_ccix_agent_err {
+	__u32	validation_bits;
+#define CCIX_AGENT_INTERNAL_ERR_VENDOR_DATA_VALID	BIT(0)
+	__u32	vendor_data[];
+};
+
+struct cper_ccix_agent_err {
+	struct cper_sec_ccix_header header;
+	__u32 ccix_header[CCIX_PER_LOG_HEADER_DWS];
+	struct cper_sec_ccix_agent_err agent_record;
+};
+
+static inline u16 ccix_agent_err_ven_len_get(struct cper_ccix_agent_err *agent_err)
+{
+	if (agent_err->agent_record.validation_bits & CCIX_AGENT_INTERNAL_ERR_VENDOR_DATA_VALID)
+		return agent_err->agent_record.vendor_data[0] & 0xFFFF;
+	else
+		return 0;
+}
+
+struct cper_ccix_agent_err_compact {
+	__u32	validation_bits;
+};
+
 /* Reset to default packing */
 #pragma pack()
 
@@ -835,6 +859,11 @@ void cper_ccix_link_err_pack(const struct cper_sec_ccix_link_error *link_record,
 const char *cper_ccix_link_err_unpack(struct trace_seq *p,
 				      struct cper_ccix_link_err_compact *clink_err);
 
+void cper_ccix_agent_err_pack(const struct cper_sec_ccix_agent_err *agent_record,
+			      struct cper_ccix_agent_err_compact *cagent_err,
+			      const u16 vendor_data_len,
+			      u8 *vendor_data);
+
 struct acpi_hest_generic_data;
 int cper_print_ccix_per(const char *pfx,
 			struct acpi_hest_generic_data *gdata);
include/ras/ras_event.h +57 -0
index 7cecfadb0b15..59b62d5cd8cf 100644
@@ -679,6 +679,70 @@ TRACE_EVENT(ccix_link_error_event,
 		__print_hex(__get_dynamic_array(vendor_data), __entry->vendor_data_length)
 	)
 );
+
+TRACE_EVENT(ccix_agent_error_event,
+	TP_PROTO(struct cper_ccix_agent_err *err,
+		 u32 err_seq,
+		 u8 sev, u16 ven_len),
+
+	TP_ARGS(err, err_seq, sev, ven_len),
+
+	TP_STRUCT__entry(
+		__field(u32, err_seq)
+		__field(u8, sev)
+		__field(u8, sevdetail)
+		__field(u8, source)
+		__field(u8, component)
+		__field(u64, pa)
+		__field(u8, pa_mask_lsb)
+		__field(u16, vendor_data_length)
+		__field_struct(struct cper_ccix_agent_err_compact, data)
+		__dynamic_array(u8, vendor_data, ven_len)
+	),
+
+	TP_fast_assign(
+		__entry->err_seq = err_seq;
+
+		__entry->sev = sev;
+		__entry->sevdetail = FIELD_GET(CCIX_PER_LOG_DW1_SEV_UE_M |
+			CCIX_PER_LOG_DW1_SEV_NO_COMM_M |
+			CCIX_PER_LOG_DW1_SEV_DEGRADED_M |
+			CCIX_PER_LOG_DW1_SEV_DEFFERABLE_M,
+			err->ccix_header[1]);
+		if (err->header.validation_bits & 0x1)
+			__entry->source = err->header.source_id;
+		else
+			__entry->source = ~0;
+		__entry->component = FIELD_GET(CCIX_PER_LOG_DW1_COMP_TYPE_M,
+						   err->ccix_header[1]);
+		if (err->ccix_header[1] & CCIX_PER_LOG_DW1_ADDR_VAL_M) {
+			__entry->pa = (u64)err->ccix_header[2] << 32 |
+				(err->ccix_header[3] & 0xfffffffc);
+			__entry->pa_mask_lsb = err->ccix_header[4] & 0xff;
+		} else {
+			__entry->pa = ~0ull;
+			__entry->pa_mask_lsb = ~0;
+		}
+		/* Do not store the vendor data header length */
+		__entry->vendor_data_length = ven_len ? ven_len - 4 : 0;
+		cper_ccix_agent_err_pack(&err->agent_record, &__entry->data,
+					__entry->vendor_data_length,
+					__get_dynamic_array(vendor_data));
+	),
+
+	TP_printk("{%d} %s CCIX PER Agent Internal Error in %s SevUE:%d SevNoComm:%d SevDegraded:%d SevDeferred:%d physical addr: %016llx (mask: %x) vendor:%s",
+		__entry->err_seq,
+		cper_severity_str(__entry->sev),
+		cper_ccix_comp_type_str(__entry->component),
+		__entry->sevdetail & BIT(0) ? 1 : 0,
+		__entry->sevdetail & BIT(1) ? 1 : 0,

include/ras/ras_event.h TRACE_EVENT(ccix_link_error_event, · line 748

Quoted 10 lines
include/ras/ras_event.h +7 -0
+		__entry->sevdetail & BIT(2) ? 1 : 0,
+		__entry->sevdetail & BIT(3) ? 1 : 0,
+		__entry->pa,
+		__entry->pa_mask_lsb,
+		__print_hex(__get_dynamic_array(vendor_data), __entry->vendor_data_length)
+	)
+);
 #endif
 
 /*

Cheers,
Mauro