Re: [PATCH v3 4/9] ACPI, x86: Extended error log driver for x86 platform

["Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com>]

On 10/18/2013 01:53 PM, Chen, Gong wrote:
> This H/W error log driver (a.k.a eMCA driver) is implemented based on
> http://www.intel.com/content/www/us/en/architecture-and-technology/enhanced-mca-logging-xeon-paper.html
>
> After errors are captured, more valuable information can be
> got via this new enhanced H/W error log driver.
>
> v3 -> v2: fix a MACRO definition error and some cleanup
> v2 -> v1: eliminate spin_lock & minor fixes suggested by Boris
>
> Signed-off-by: Chen, Gong <gong.chen@linux.intel.com>
> ---
>   arch/x86/include/asm/mce.h       |   5 +
>   arch/x86/kernel/cpu/mcheck/mce.c |  20 +++
>   drivers/acpi/Kconfig             |  20 +++
>   drivers/acpi/Makefile            |   2 +
>   drivers/acpi/acpi_extlog.c       | 319 +++++++++++++++++++++++++++++++++++++++
>   drivers/acpi/bus.c               |   3 +-
>   include/linux/acpi.h             |   1 +
>   7 files changed, 369 insertions(+), 1 deletion(-)
>   create mode 100644 drivers/acpi/acpi_extlog.c
>
> diff --git a/arch/x86/include/asm/mce.h b/arch/x86/include/asm/mce.h
> index cbe6b9e..072b2f8 100644
> --- a/arch/x86/include/asm/mce.h
> +++ b/arch/x86/include/asm/mce.h
> @@ -16,6 +16,7 @@
>   #define MCG_EXT_CNT_SHIFT	16
>   #define MCG_EXT_CNT(c)		(((c) & MCG_EXT_CNT_MASK) >> MCG_EXT_CNT_SHIFT)
>   #define MCG_SER_P		(1ULL<<24)   /* MCA recovery/new status bits */
> +#define MCG_ELOG_P		(1ULL<<26)   /* Extended error log supported */
>
>   /* MCG_STATUS register defines */
>   #define MCG_STATUS_RIPV  (1ULL<<0)   /* restart ip valid */
> @@ -186,6 +187,10 @@ enum mcp_flags {
>   	MCP_UC = (1 << 1),		/* log uncorrected errors */
>   	MCP_DONTLOG = (1 << 2),		/* only clear, don't log */
>   };
> +
> +void register_elog_handler(int (*f)(const char *, int, int));
> +void unregister_elog_handler(int (*f)(const char *, int, int));
> +
>   void machine_check_poll(enum mcp_flags flags, mce_banks_t *b);
>
>   int mce_notify_irq(void);
> diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
> index b3218cd..981e0d3 100644
> --- a/arch/x86/kernel/cpu/mcheck/mce.c
> +++ b/arch/x86/kernel/cpu/mcheck/mce.c
> @@ -48,6 +48,8 @@
>
>   #include "mce-internal.h"
>
> +static int (*mce_ext_err_print)(const char *, int, int);
> +
>   static DEFINE_MUTEX(mce_chrdev_read_mutex);
>
>   #define rcu_dereference_check_mce(p) \
> @@ -576,6 +578,21 @@ static void mce_read_aux(struct mce *m, int i)
>
>   DEFINE_PER_CPU(unsigned, mce_poll_count);
>
> +void register_elog_handler(int (*f)(const char *, int, int))
> +{
> +	mce_ext_err_print = f;
> +}
> +EXPORT_SYMBOL_GPL(register_elog_handler);
> +
> +void unregister_elog_handler(int (*f)(const char *, int, int))
> +{
> +	if (f) {
> +		WARN_ON(mce_ext_err_print != f);
> +		mce_ext_err_print = NULL;
> +	}
> +}
> +EXPORT_SYMBOL_GPL(unregister_elog_handler);
> +
>   /*
>    * Poll for corrected events or events that happened before reset.
>    * Those are just logged through /dev/mcelog.
> @@ -624,6 +641,9 @@ void machine_check_poll(enum mcp_flags flags, mce_banks_t *b)
>   		    (m.status & (mca_cfg.ser ? MCI_STATUS_S : MCI_STATUS_UC)))
>   			continue;
>
> +		if (mce_ext_err_print)
> +			mce_ext_err_print(NULL, m.extcpu, i);
> +

Can we use the notifier chain we already have: 
mce_register_decode_chain()? EDAC uses this and I'm wondering if it is a 
good fit here. As an added bonus, it seems to honor dont_log_ce option 
as well.

>   		mce_read_aux(&m, i);
>
>   		if (!(flags & MCP_TIMESTAMP))
> diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
> index 22327e6..c67ec61 100644
> --- a/drivers/acpi/Kconfig
> +++ b/drivers/acpi/Kconfig
> @@ -372,4 +372,24 @@ config ACPI_BGRT
>
>   source "drivers/acpi/apei/Kconfig"
>
> +config ACPI_EXTLOG
> +	tristate "Extended Error Log support"
> +	depends on X86_MCE

I think you also have a dependancy on ACPI_APEI for apei_estatus_print()

> +	default n
> +	help
> +	  Certain usages such as Predictive Failure Analysis (PFA) require
> +	  more information about the error than what can be described in
> +	  processor machine check banks. Most server processors log
> +	  additional information about the error in processor uncore
> +	  registers. Since the addresses and layout of these registers vary
> +	  widely from one processor to another, system software cannot
> +	  readily make use of them. To complicate matters further, some of
> +	  the additional error information cannot be constructed space
> +	  between "additional" and "error" without detailed knowledge

Oops... looks like copy+paste went wrong ;)

> +	  about platform topology.
> +
> +	  Enhanced MCA Logging allows firmware to provide additional error
> +	  information to system software, synchronous with MCE or CMCI. This
> +	  driver adds support for that functionality.
> +
>   endif	# ACPI
> diff --git a/drivers/acpi/Makefile b/drivers/acpi/Makefile
> index cdaf68b..bce34af 100644
> --- a/drivers/acpi/Makefile
> +++ b/drivers/acpi/Makefile
> @@ -82,3 +82,5 @@ processor-$(CONFIG_CPU_FREQ)	+= processor_perflib.o
>   obj-$(CONFIG_ACPI_PROCESSOR_AGGREGATOR) += acpi_pad.o
>
>   obj-$(CONFIG_ACPI_APEI)		+= apei/
> +
> +obj-$(CONFIG_ACPI_EXTLOG)	+= acpi_extlog.o
> diff --git a/drivers/acpi/acpi_extlog.c b/drivers/acpi/acpi_extlog.c
> new file mode 100644
> index 0000000..afeab59
> --- /dev/null
> +++ b/drivers/acpi/acpi_extlog.c
> @@ -0,0 +1,319 @@
> +/*
> + * Extended Error Log driver
> + *
> + * Copyright (C) 2013 Intel Corp.
> + * Author: Chen, Gong <gong.chen@intel.com>
> + *
> + * This file is licensed under GPLv2.
> + */
> +
> +#include <linux/module.h>
> +#include <linux/acpi.h>
> +#include <acpi/acpi_bus.h>
> +#include <linux/cper.h>
> +#include <linux/ratelimit.h>
> +#include <asm/mce.h>
> +
> +#include "apei/apei-internal.h"
> +
> +#define EXT_ELOG_ENTRY_MASK	GENMASK_ULL(51, 0) /* elog entry address mask */
> +
> +#define EXTLOG_DSM_REV		0x0
> +#define	EXTLOG_FN_QUERY		0x0
> +#define	EXTLOG_FN_ADDR		0x1
> +
> +#define FLAG_OS_OPTIN		BIT(0)
> +#define EXTLOG_QUERY_L1_EXIST	BIT(1)
> +#define ELOG_ENTRY_VALID	(1ULL<<63)
> +#define ELOG_ENTRY_LEN		0x1000
> +
> +#define EMCA_BUG \
> +	"Can not request iomem region <0x%016llx-0x%016llx> - eMCA disabled\n"
> +
> +struct extlog_l1_head {
> +	u32 ver;	/* Header Version */
> +	u32 hdr_len;	/* Header Length */
> +	u64 total_len;	/* entire L1 Directory length including this header */
> +	u64 elog_base;	/* MCA Error Log Directory base address */
> +	u64 elog_len;	/* MCA Error Log Directory length */
> +	u32 flags;	/* bit 0 - OS/VMM Opt-in */
> +	u8  rev0[12];
> +	u32 entries;	/* Valid L1 Directory entries per logical processor */
> +	u8  rev1[12];
> +};
> +
> +static u8 extlog_dsm_uuid[] = "663E35AF-CC10-41A4-88EA-5470AF055295";
> +
> +/* L1 table related physical address */
> +static u64 elog_base;
> +static size_t elog_size;
> +static u64 l1_dirbase;
> +static size_t l1_size;
> +
> +/* L1 table related virtual address */
> +static void __iomem *extlog_l1_addr;
> +static void __iomem *elog_addr;
> +
> +static void *elog_buf;
> +
> +static u64 *l1_entry_base;
> +static u32 l1_percpu_entry;
> +
> +#define ELOG_IDX(cpu, bank) \
> +	(cpu_physical_id(cpu) * l1_percpu_entry + (bank))
> +
> +#define ELOG_ENTRY_DATA(idx) \
> +	(*(l1_entry_base + (idx)))
> +
> +#define ELOG_ENTRY_ADDR(phyaddr) \
> +	(phyaddr - elog_base + (u8 *)elog_addr)
> +
> +static struct acpi_generic_status *extlog_elog_entry_check(int cpu, int bank)
> +{
> +	int idx;
> +	u64 data;
> +	struct acpi_generic_status *estatus;
> +
> +	WARN_ON(cpu < 0);
> +	idx = ELOG_IDX(cpu, bank);
> +	data = ELOG_ENTRY_DATA(idx);
> +	if ((data & ELOG_ENTRY_VALID) == 0)
> +		return NULL;
> +
> +	data &= EXT_ELOG_ENTRY_MASK;
> +	estatus = (struct acpi_generic_status *)ELOG_ENTRY_ADDR(data);
> +
> +	/* if no valid data in elog entry, just return */
> +	if (estatus->block_status == 0)
> +		return NULL;
> +
> +	return estatus;
> +}
> +
> +static void __print_extlog_rcd(const char *pfx,
> +			       struct acpi_generic_status *estatus, int cpu)
> +{
> +	static atomic_t seqno;
> +	unsigned int curr_seqno;
> +	char pfx_seq[64];
> +
> +	if (!pfx) {
> +		if (estatus->error_severity <= CPER_SEV_CORRECTED)
> +			pfx = KERN_INFO;
> +		else
> +			pfx = KERN_ERR;
> +	}
> +	curr_seqno = atomic_inc_return(&seqno);
> +	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}", pfx, curr_seqno);
> +	printk("%s""Hardware error detected on CPU%d\n", pfx_seq, cpu);
> +	cper_estatus_print(pfx_seq, estatus);
> +}
> +
> +static int print_extlog_rcd(const char *pfx,
> +			    struct acpi_generic_status *estatus, int cpu)
> +{
> +	/* Not more than 2 messages every 5 seconds */
> +	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
> +	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
> +	struct ratelimit_state *ratelimit;
> +
> +	if (estatus->error_severity == CPER_SEV_CORRECTED ||
> +	    (estatus->error_severity == CPER_SEV_INFORMATIONAL))
> +		ratelimit = &ratelimit_corrected;
> +	else
> +		ratelimit = &ratelimit_uncorrected;
> +	if (__ratelimit(ratelimit)) {
> +		__print_extlog_rcd(pfx, estatus, cpu);
> +		return 0;
> +	}
> +
> +	return 1;
> +}
> +
> +static int extlog_print(const char *pfx, int cpu, int bank)
> +{
> +	struct acpi_generic_status *estatus;
> +	int rc;
> +
> +	estatus = extlog_elog_entry_check(cpu, bank);
> +	if (estatus == NULL)
> +		return -EINVAL;
> +
> +	memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
> +	/* clear record status to enable BIOS to update it again */
> +	estatus->block_status = 0;
> +
> +	rc = print_extlog_rcd(pfx, (struct acpi_generic_status *)elog_buf, cpu);
> +
> +	return rc;
> +}
> +
> +static int extlog_get_dsm(acpi_handle handle, int rev, int func, u64 *ret)
> +{
> +	struct acpi_buffer buf = {ACPI_ALLOCATE_BUFFER, NULL};
> +	struct acpi_object_list input;
> +	union acpi_object params[4], *obj;
> +	u8 uuid[16];
> +	int i;
> +
> +	acpi_str_to_uuid(extlog_dsm_uuid, uuid);
> +	input.count = 4;
> +	input.pointer = params;
> +	params[0].type = ACPI_TYPE_BUFFER;
> +	params[0].buffer.length = 16;
> +	params[0].buffer.pointer = uuid;
> +	params[1].type = ACPI_TYPE_INTEGER;
> +	params[1].integer.value = rev;
> +	params[2].type = ACPI_TYPE_INTEGER;
> +	params[2].integer.value = func;
> +	params[3].type = ACPI_TYPE_PACKAGE;
> +	params[3].package.count = 0;
> +	params[3].package.elements = NULL;
> +
> +	if (ACPI_FAILURE(acpi_evaluate_object(handle, "_DSM", &input, &buf)))
> +		return -1;
> +
> +	*ret = 0;
> +	obj = (union acpi_object *)buf.pointer;
> +	if (obj->type == ACPI_TYPE_INTEGER) {
> +		*ret = obj->integer.value;
> +	} else if (obj->type == ACPI_TYPE_BUFFER) {
> +		if (obj->buffer.length <= 8) {
> +			for (i = 0; i < obj->buffer.length; i++)
> +				*ret |= (obj->buffer.pointer[i] << (i * 8));
> +		}
> +	}
> +	kfree(buf.pointer);
> +
> +	return 0;
> +}
> +
> +static bool extlog_get_l1addr(void)
> +{
> +	acpi_handle handle;
> +	u64 ret;
> +
> +	if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
> +		return false;
> +
> +	if (extlog_get_dsm(handle, EXTLOG_DSM_REV, EXTLOG_FN_QUERY, &ret) ||
> +	    !(ret & EXTLOG_QUERY_L1_EXIST))
> +		return false;
> +
> +	if (extlog_get_dsm(handle, EXTLOG_DSM_REV, EXTLOG_FN_ADDR, &ret))
> +		return false;
> +
> +	l1_dirbase = ret;
> +	/* Spec says L1 directory must be 4K aligned, bail out if it isn't */
> +	if (l1_dirbase & ((1 << 12) - 1)) {
> +		pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
> +			l1_dirbase);
> +		return false;
> +	}
> +
> +	return true;
> +}
> +
> +static int __init extlog_init(void)
> +{
> +	struct extlog_l1_head *l1_head;
> +	void __iomem *extlog_l1_hdr;
> +	size_t l1_hdr_size;
> +	struct resource *r;
> +	u64 cap;
> +	int rc;
> +
> +	rc = -ENODEV;
> +
> +	rdmsrl(MSR_IA32_MCG_CAP, cap);
> +	if (!(cap & MCG_ELOG_P))
> +		return rc;
> +
> +	if (!extlog_get_l1addr())
> +		return rc;
> +
> +	rc = -EINVAL;
> +	/* get L1 header to fetch necessary information */
> +	l1_hdr_size = sizeof(struct extlog_l1_head);
> +	r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR");
> +	if (!r) {
> +		pr_warn(FW_BUG EMCA_BUG,
> +			(unsigned long long)l1_dirbase,
> +			(unsigned long long)l1_dirbase + l1_hdr_size);
> +		goto err;
> +	}
> +
> +	extlog_l1_hdr = acpi_os_map_memory(l1_dirbase, l1_hdr_size);
> +	l1_head = (struct extlog_l1_head *)extlog_l1_hdr;
> +	l1_size = l1_head->total_len;
> +	l1_percpu_entry = l1_head->entries;
> +	elog_base = l1_head->elog_base;
> +	elog_size = l1_head->elog_len;
> +	acpi_os_unmap_memory(extlog_l1_hdr, l1_hdr_size);
> +	release_mem_region(l1_dirbase, l1_hdr_size);
> +
> +	/* remap L1 header again based on completed information */
> +	r = request_mem_region(l1_dirbase, l1_size, "L1 Table");
> +	if (!r) {
> +		pr_warn(FW_BUG EMCA_BUG,
> +			(unsigned long long)l1_dirbase,
> +			(unsigned long long)l1_dirbase + l1_size);
> +		goto err;
> +	}
> +	extlog_l1_addr = acpi_os_map_memory(l1_dirbase, l1_size);
> +	l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size);
> +
> +	/* remap elog table */
> +	r = request_mem_region(elog_base, elog_size, "Elog Table");
> +	if (!r) {
> +		pr_warn(FW_BUG EMCA_BUG,
> +			(unsigned long long)elog_base,
> +			(unsigned long long)elog_base + elog_size);
> +		goto err_release_l1_dir;
> +	}
> +	elog_addr = acpi_os_map_memory(elog_base, elog_size);
> +
> +	rc = -ENOMEM;
> +	/* allocate buffer to save elog record */
> +	elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL);
> +	if (elog_buf == NULL)
> +		goto err_release_elog;
> +
> +	register_elog_handler(extlog_print);
> +	/* enable OS to be involved to take over management from BIOS */
> +	((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN;
> +
> +	return 0;
> +
> +err_release_elog:
> +	if (elog_addr)
> +		acpi_os_unmap_memory(elog_addr, elog_size);
> +	release_mem_region(elog_base, elog_size);
> +err_release_l1_dir:
> +	if (extlog_l1_addr)
> +		acpi_os_unmap_memory(extlog_l1_addr, l1_size);
> +	release_mem_region(l1_dirbase, l1_size);
> +err:
> +	pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n");
> +	return rc;
> +}
> +
> +static void __exit extlog_exit(void)
> +{
> +	unregister_elog_handler(extlog_print);
> +	((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
> +	if (extlog_l1_addr)
> +		acpi_os_unmap_memory(extlog_l1_addr, l1_size);
> +	if (elog_addr)
> +		acpi_os_unmap_memory(elog_addr, elog_size);
> +	release_mem_region(elog_base, elog_size);
> +	release_mem_region(l1_dirbase, l1_size);
> +	kfree(elog_buf);
> +}
> +
> +module_init(extlog_init);
> +module_exit(extlog_exit);
> +
> +MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
> +MODULE_DESCRIPTION("Extended Error Log Driver");

"Extended MCA Error Log Driver"?

Regards,
Naveen

> +MODULE_LICENSE("GPL");
> diff --git a/drivers/acpi/bus.c b/drivers/acpi/bus.c
> index b587ec8..e1bd9a1 100644
> --- a/drivers/acpi/bus.c
> +++ b/drivers/acpi/bus.c
> @@ -174,7 +174,7 @@ static void acpi_print_osc_error(acpi_handle handle,
>   	printk("\n");
>   }
>
> -static acpi_status acpi_str_to_uuid(char *str, u8 *uuid)
> +acpi_status acpi_str_to_uuid(char *str, u8 *uuid)
>   {
>   	int i;
>   	static int opc_map_to_uuid[16] = {6, 4, 2, 0, 11, 9, 16, 14, 19, 21,
> @@ -195,6 +195,7 @@ static acpi_status acpi_str_to_uuid(char *str, u8 *uuid)
>   	}
>   	return AE_OK;
>   }
> +EXPORT_SYMBOL_GPL(acpi_str_to_uuid);
>
>   acpi_status acpi_run_osc(acpi_handle handle, struct acpi_osc_context *context)
>   {
> diff --git a/include/linux/acpi.h b/include/linux/acpi.h
> index a5db4ae..c30bac8 100644
> --- a/include/linux/acpi.h
> +++ b/include/linux/acpi.h
> @@ -311,6 +311,7 @@ struct acpi_osc_context {
>   #define OSC_INVALID_REVISION_ERROR	8
>   #define OSC_CAPABILITIES_MASK_ERROR	16
>
> +acpi_status acpi_str_to_uuid(char *str, u8 *uuid);
>   acpi_status acpi_run_osc(acpi_handle handle, struct acpi_osc_context *context);
>
>   /* platform-wide _OSC bits */
>