Date:	Mon, 30 Jun 2008 06:14:42 -0700 (PDT)
From:	eranian@...glemail.com
To:	linux-kernel@...r.kernel.org
Subject: [patch 18/19] perfmon2 minimal v3:  Intel architectural PMU support

This patch adds Intel architectural PMU support (version 1 and 2).

Signed-off-by: Stephane Eranian <eranian@...il.com>
--

Index: o/arch/x86/perfmon/Makefile
===================================================================
--- o.orig/arch/x86/perfmon/Makefile	2008-06-13 16:40:03.000000000 +0200
+++ o/arch/x86/perfmon/Makefile	2008-06-13 16:42:00.000000000 +0200
@@ -3,3 +3,4 @@
 # Contributed by Stephane Eranian <eranian@....hp.com>
 #
 obj-$(CONFIG_PERFMON)			+= perfmon.o
+obj-$(CONFIG_X86_PERFMON_INTEL_ARCH)	+= perfmon_intel_arch.o
Index: o/arch/x86/perfmon/perfmon_intel_arch.c
===================================================================
--- /dev/null	1970-01-01 00:00:00.000000000 +0000
+++ o/arch/x86/perfmon/perfmon_intel_arch.c	2008-06-13 16:40:04.000000000 +0200
@@ -0,0 +1,492 @@
+/*
+ * This file contains the Intel architectural perfmon v1 or v2
+ * description tables.
+ *
+ * Architectural perfmon was introduced with Intel Core Solo/Duo
+ * processors.
+ *
+ * Copyright (c) 2006-2007 Hewlett-Packard Development Company, L.P.
+ * Contributed by Stephane Eranian <eranian@....hp.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+ * 02111-1307 USA
+ */
+#include <linux/kprobes.h>
+#include <linux/perfmon_kern.h>
+#include <asm/msr.h>
+#include <asm/apic.h>
+
+static u64 enable_mask[PFM_MAX_PMCS];
+static u16 max_enable;
+
+/*
+ * - upper 32 bits are reserved
+ * - INT: APIC enable bit is reserved (forced to 1)
+ * - bit 21 is reserved
+ *
+ * RSVD: reserved bits are 1
+ */
+#define PFM_IA_PMC_RSVD	((~((1ULL<<32)-1)) \
+			| (1ULL<<20) \
+			| (1ULL<<21))
+
+/*
+ * force Local APIC interrupt on overflow
+ * disable with NO_EMUL64
+ */
+#define PFM_IA_PMC_VAL	(1ULL<<20)
+#define PFM_IA_NO64	(1ULL<<20)
+
+/*
+ * architectuture specifies that:
+ * IA32_PMCx MSR        : starts at 0x0c1 & occupy a contiguous block of MSR
+ * IA32_PERFEVTSELx MSR : starts at 0x186 & occupy a contiguous block of MSR
+ * MSR_GEN_FIXED_CTR0   : starts at 0x309 & occupy a contiguous block of MSR
+ */
+#define MSR_GEN_SEL_BASE	MSR_P6_EVNTSEL0
+#define MSR_GEN_PMC_BASE	MSR_P6_PERFCTR0
+#define MSR_GEN_FIXED_PMC_BASE	MSR_CORE_PERF_FIXED_CTR0
+
+#define PFM_IA_SEL(n)	{ 			\
+	.addrs[0] = MSR_GEN_SEL_BASE+(n),	\
+	.ctr = n,				\
+	.reg_type = PFM_REGT_EN}
+
+#define PFM_IA_CTR(n) {				\
+	.addrs[0] = MSR_GEN_PMC_BASE+(n),	\
+	.ctr = n,				\
+	.reg_type = PFM_REGT_CTR}
+
+#define PFM_IA_FCTR(n) {			\
+	.addrs[0] = MSR_GEN_FIXED_PMC_BASE+(n),	\
+	.ctr = n,				\
+	.reg_type = PFM_REGT_CTR}
+
+/*
+ * layout of EAX for CPUID.0xa leaf function
+ */
+struct pmu_eax {
+	unsigned int version:8;		/* architectural perfmon version */
+	unsigned int num_cnt:8; 	/* number of generic counters */
+	unsigned int cnt_width:8;	/* width of generic counters */
+	unsigned int ebx_length:8;	/* number of architected events */
+};
+
+/*
+ * layout of EDX for CPUID.0xa leaf function when perfmon v2 is detected
+ */
+struct pmu_edx {
+	unsigned int num_cnt:5;		/* number of fixed counters */
+	unsigned int cnt_width:8;	/* width of fixed counters */
+	unsigned int reserved:19;
+};
+
+static int pfm_intel_arch_stop_save(struct pfm_context *ctx,
+				    struct pfm_event_set *set);
+static int pfm_intel_arch_has_ovfls(struct pfm_context *ctx);
+static void __kprobes pfm_intel_arch_quiesce(void);
+
+/*
+ * physical addresses of MSR controlling the perfevtsel and counter registers
+ */
+struct pfm_arch_pmu_info pfm_intel_arch_pmu_info = {
+	.stop_save = pfm_intel_arch_stop_save,
+	.has_ovfls = pfm_intel_arch_has_ovfls,
+	.quiesce = pfm_intel_arch_quiesce,
+};
+
+#define PFM_IA_C(n) {                   \
+	.type = PFM_REG_I64,            \
+	.desc = "PERFEVTSEL"#n,         \
+	.dfl_val = PFM_IA_PMC_VAL,      \
+	.rsvd_msk = PFM_IA_PMC_RSVD,    \
+	.no_emul64_msk = PFM_IA_NO64,   \
+	.hw_addr = MSR_GEN_SEL_BASE+(n) \
+	}
+
+#define PFM_IA_D(n) PMD_D(PFM_REG_C, "PMC"#n, MSR_P6_PERFCTR0+n)
+#define PFM_IA_FD(n) PMD_D(PFM_REG_C, "FIXED_CTR"#n, MSR_CORE_PERF_FIXED_CTR0+n)
+
+static struct pfm_regmap_desc pfm_intel_arch_pmc_desc[] = {
+/* pmc0  */ PFM_IA_C(0),  PFM_IA_C(1),   PFM_IA_C(2),  PFM_IA_C(3),
+/* pmc4  */ PFM_IA_C(4),  PFM_IA_C(5),   PFM_IA_C(6),  PFM_IA_C(7),
+/* pmc8  */ PFM_IA_C(8),  PFM_IA_C(9),  PFM_IA_C(10), PFM_IA_C(11),
+/* pmc12 */ PFM_IA_C(12), PFM_IA_C(13), PFM_IA_C(14), PFM_IA_C(15),
+
+/* pmc16 */ { .type = PFM_REG_I,
+	      .desc = "FIXED_CTRL",
+	      .dfl_val = 0x8888888888888888ULL,
+	      .rsvd_msk = 0xccccccccccccccccULL,
+	      .no_emul64_msk = 0,
+	      .hw_addr = MSR_CORE_PERF_FIXED_CTR_CTRL
+	    },
+};
+#define PFM_IA_MAX_PMCS	ARRAY_SIZE(pfm_intel_arch_pmc_desc)
+
+static struct pfm_regmap_desc pfm_intel_arch_pmd_desc[] = {
+/* pmd0  */  PFM_IA_D(0),  PFM_IA_D(1),  PFM_IA_D(2),  PFM_IA_D(3),
+/* pmd4  */  PFM_IA_D(4),  PFM_IA_D(5),  PFM_IA_D(6),  PFM_IA_D(7),
+/* pmd8  */  PFM_IA_D(8),  PFM_IA_D(9), PFM_IA_D(10), PFM_IA_D(11),
+/* pmd12 */ PFM_IA_D(12), PFM_IA_D(13), PFM_IA_D(14), PFM_IA_D(15),
+
+/* pmd16 */ PFM_IA_FD(0), PFM_IA_FD(1), PFM_IA_FD(2), PFM_IA_FD(3),
+/* pmd20 */ PFM_IA_FD(4), PFM_IA_FD(5), PFM_IA_FD(6), PFM_IA_FD(7),
+/* pmd24 */ PFM_IA_FD(8), PFM_IA_FD(9), PFM_IA_FD(10), PFM_IA_FD(11),
+/* pmd28 */ PFM_IA_FD(16), PFM_IA_FD(17), PFM_IA_FD(18), PFM_IA_FD(19)
+};
+#define PFM_IA_MAX_PMDS	ARRAY_SIZE(pfm_intel_arch_pmd_desc)
+
+#define PFM_IA_MAX_CNT		16 /* # generic counters in mapping table */
+#define PFM_IA_MAX_FCNT		16 /* # of fixed counters in mapping table */
+#define PFM_IA_FCNT_BASE	16 /* base index of fixed counters PMD */
+
+static struct pfm_pmu_config pfm_intel_arch_pmu_conf;
+
+static void pfm_intel_arch_check_errata(void)
+{
+	/*
+	 * Core Duo errata AE49 (no fix). Both counters share a single
+	 * enable bit in PERFEVTSEL0
+	 */
+	if (current_cpu_data.x86 == 6 && current_cpu_data.x86_model == 14)
+		pfm_intel_arch_pmu_info.flags |= PFM_X86_FL_NO_SHARING;
+}
+
+static int pfm_intel_arch_probe_pmu(void)
+{
+	union {
+		unsigned int val;
+		struct pmu_eax eax;
+		struct pmu_edx edx;
+	} eax, edx;
+	unsigned int ebx, ecx;
+	unsigned int num_cnt, i;
+	u64 dfl, rsvd;
+
+	edx.val = 0;
+
+	if (!cpu_has_arch_perfmon) {
+		PFM_INFO("no support for Intel architectural PMU");
+		return -1;
+	}
+
+	if (!cpu_has_apic) {
+		PFM_INFO("no Local APIC, try rebooting with lapic option");
+		return -1;
+	}
+
+	/* cpuid() call protected by cpu_has_arch_perfmon */
+	cpuid(0xa, &eax.val, &ebx, &ecx, &edx.val);
+
+#if 0
+	/*
+	 * XXX: this switch statement is needed when we have
+	 * sepcialized PMU support for processors which also support
+	 * architected PMU. Otherwise, the link order may cause
+	 * this generic support to take over the more speciailized
+	 * module and therefore prvent access to some advanced features.
+	 */
+
+	/*
+	 * reject processors supported by perfmon_intel_core
+	 *
+	 * We need to do this explicitely to avoid depending
+	 * on the link order in case, the modules are compiled as
+	 * builtin.
+	 *
+	 * non Intel processors are rejected by cpu_has_arch_perfmon
+	 */
+	if (current_cpu_data.x86 == 6) {
+		switch (current_cpu_data.x86_model) {
+		case 15: /* Merom: use perfmon_intel_core  */
+		case 23: /* Penryn: use perfmon_intel_core */
+			return -1;
+		default:
+			break;
+		}
+	}
+#endif
+
+	/*
+	 * some 6/15 models have buggy BIOS
+	 */
+	if (eax.eax.version == 0
+	    && current_cpu_data.x86 == 6 && current_cpu_data.x86_model == 15) {
+		PFM_INFO("buggy v2 BIOS, adjusting for 2 generic counters");
+		eax.eax.version = 2;
+		eax.eax.num_cnt = 2;
+		eax.eax.cnt_width = 40;
+	}
+
+	/*
+	 * some v2 BIOSes are incomplete
+	 */
+	if (eax.eax.version == 2 && !edx.edx.num_cnt) {
+		PFM_INFO("buggy v2 BIOS, adjusting for 3 fixed counters");
+		edx.edx.num_cnt = 3;
+		edx.edx.cnt_width = 40;
+	}
+
+	/*
+	 * no fixed counters on earlier versions
+	 */
+	if (eax.eax.version < 2)
+		edx.val = 0;
+
+	PFM_INFO("detected architecural perfmon v%d", eax.eax.version);
+	PFM_INFO("num_gen=%d width=%d num_fixed=%d width=%d",
+		  eax.eax.num_cnt,
+		  eax.eax.cnt_width,
+		  edx.edx.num_cnt,
+		  edx.edx.cnt_width);
+
+	/* number of generic counters */
+	num_cnt = eax.eax.num_cnt;
+
+	if (num_cnt >= PFM_IA_MAX_CNT) {
+		printk(KERN_INFO "perfmon: Limiting number of generic counters"
+				 " to %zu, HW supports %u",
+				 PFM_IA_MAX_PMCS, num_cnt);
+		num_cnt = PFM_IA_MAX_CNT;
+
+	}
+
+	/*
+	 * adjust rsvd_msk for generic counters based on actual width
+	 * set enable_mask, we can use the PMD loop, because there is
+	 * one pmc per pmd and all pmds are counters
+	 */
+	for (i = 0; i < num_cnt; i++) {
+		pfm_intel_arch_pmd_desc[i].rsvd_msk =
+			~((1ULL<<eax.eax.cnt_width)-1);
+		__set_bit(i, cast_ulp(enable_mask));
+	}
+	max_enable = i;
+
+	/*
+	 * mark unused generic counters as not available
+	 */
+	for (i = num_cnt; i < PFM_IA_MAX_CNT; i++) {
+		pfm_intel_arch_pmd_desc[i].type = PFM_REG_NA;
+		pfm_intel_arch_pmc_desc[i].type = PFM_REG_NA;
+	}
+
+	/*
+	 * now process fixed counters (if any)
+	 */
+	num_cnt = edx.edx.num_cnt;
+
+	/*
+	 * adjust rsvd_msk for fixed counters based on actual width
+	 */
+	for (i = 0; i < num_cnt; i++)
+		pfm_intel_arch_pmd_desc[PFM_IA_FCNT_BASE+i].rsvd_msk =
+			~((1ULL<<edx.edx.cnt_width)-1);
+
+	/*
+	 * mark unused fixed counters as
+	 * unavailable.
+	 * update the rsvd_msk, dfl_val for
+	 * FIXED_CTRL:
+	 * 	rsvd_msk: set all 4 bits
+	 *	dfl_val : clear all 4 bits
+	 */
+	dfl = pfm_intel_arch_pmc_desc[16].dfl_val;
+	rsvd = pfm_intel_arch_pmc_desc[16].rsvd_msk;
+
+	for (i = num_cnt; i < PFM_IA_MAX_FCNT; i++) {
+		pfm_intel_arch_pmd_desc[PFM_IA_FCNT_BASE+i].type = PFM_REG_NA;
+		rsvd |= 0xfULL << (i<<2);
+		dfl &= ~(0xfULL << (i<<2));
+	}
+
+	/*
+	 * FIXED_CTR_CTRL unavailable when no fixed counters are defined
+	 */
+	if (!num_cnt) {
+		pfm_intel_arch_pmc_desc[16].type = PFM_REG_NA;
+	} else {
+		pfm_intel_arch_pmc_desc[16].rsvd_msk = rsvd;
+		pfm_intel_arch_pmc_desc[16].dfl_val = dfl;
+	}
+
+	/*
+	 * Maximum number of entries in both tables (some maybe NA)
+	 */
+	pfm_intel_arch_pmu_conf.num_pmc_entries = PFM_IA_MAX_PMCS;
+	pfm_intel_arch_pmu_conf.num_pmd_entries = PFM_IA_MAX_PMDS;
+
+	pfm_intel_arch_check_errata();
+
+	return 0;
+}
+
+/**
+ * pfm_intel_arch_has_ovfls - check for pending overflow condition
+ * @ctx: context to work on
+ *
+ * detect if counters have overflowed.
+ * return:
+ * 	0 : no overflow
+ * 	1 : at least one overflow
+ */
+static int __kprobes pfm_intel_arch_has_ovfls(struct pfm_context *ctx)
+{
+	u64 *cnt_mask;
+	u64 wmask, val;
+	u16 i, num;
+
+	cnt_mask = pfm_pmu_conf->regs.cnt_pmds;
+	num = pfm_pmu_conf->regs.num_counters;
+	wmask = 1ULL << pfm_pmu_conf->counter_width;
+
+	/*
+	 * we can leverage the fact that we know the mapping
+	 * to hardcode the MSR address and avoid accessing
+	 * more cachelines
+	 *
+	 * We need to check cnt_mask because not all registers
+	 * may be available.
+	 */
+	for (i = 0; num; i++) {
+		if (test_bit(i, cast_ulp(cnt_mask))) {
+			rdmsrl(MSR_P6_PERFCTR0+i, val);
+			if (!(val & wmask))
+				return 1;
+			num--;
+		}
+	}
+	return 0;
+}
+
+static int pfm_intel_arch_stop_save(struct pfm_context *ctx,
+				    struct pfm_event_set *set)
+{
+	u64 used_mask[PFM_PMC_BV];
+	u64 *cnt_pmds;
+	u64 val, wmask, ovfl_mask;
+	u32 i, count;
+
+	wmask = 1ULL << pfm_pmu_conf->counter_width;
+
+	bitmap_and(cast_ulp(used_mask),
+		   cast_ulp(set->used_pmcs),
+		   cast_ulp(enable_mask),
+		   max_enable);
+
+	count = bitmap_weight(cast_ulp(used_mask), max_enable);
+
+	/*
+	 * stop monitoring
+	 * Unfortunately, this is very expensive!
+	 * wrmsrl() is serializing.
+	 */
+	for (i = 0; count; i++) {
+		if (test_bit(i, cast_ulp(used_mask))) {
+			wrmsrl(pfm_pmu_conf->pmc_desc[i].hw_addr, 0);
+			count--;
+		}
+	}
+
+	/*
+	 * if we already having a pending overflow condition, we simply
+	 * return to take care of this first.
+	 */
+	if (set->npend_ovfls)
+		return 1;
+
+	ovfl_mask = pfm_pmu_conf->ovfl_mask;
+	cnt_pmds = pfm_pmu_conf->regs.cnt_pmds;
+
+	/*
+	 * check for pending overflows and save PMDs (combo)
+	 * we employ used_pmds because we also need to save
+	 * and not just check for pending interrupts.
+	 *
+	 * Must check for counting PMDs because of virtual PMDs
+	 */
+	count = set->nused_pmds;
+	for (i = 0; count; i++) {
+		if (test_bit(i, cast_ulp(set->used_pmds))) {
+			val = pfm_arch_read_pmd(ctx, i);
+			if (likely(test_bit(i, cast_ulp(cnt_pmds)))) {
+				if (!(val & wmask)) {
+					__set_bit(i, cast_ulp(set->povfl_pmds));
+					set->npend_ovfls++;
+				}
+				val = (set->pmds[i].value & ~ovfl_mask)
+					| (val & ovfl_mask);
+			}
+			set->pmds[i].value = val;
+			count--;
+		}
+	}
+	/* 0 means: no need to save PMDs at upper level */
+	return 0;
+}
+
+/**
+ * pfm_intel_arch_quiesce - stop monitoring without grabbing any lock
+ *
+ * called from NMI interrupt handler to immediately stop monitoring
+ * cannot grab any lock, including perfmon related locks
+ */
+static void __kprobes pfm_intel_arch_quiesce(void)
+{
+	u16 i;
+
+	/*
+	 * PMC16 is the fixed control control register so it has a
+	 * distinct MSR address
+	 *
+	 * We do not use the hw_addr field in the table to avoid touching
+	 * too many cachelines
+ 	 */
+	for (i = 0; i < pfm_pmu_conf->regs.max_pmc; i++) {
+		if (test_bit(i, cast_ulp(pfm_pmu_conf->regs.pmcs))) {
+			if (i == 16)
+				wrmsrl(MSR_CORE_PERF_FIXED_CTR_CTRL, 0);
+			else
+				wrmsrl(MSR_P6_EVNTSEL0+i, 0);
+		}
+	}
+}
+
+/*
+ * Counters may have model-specific width. Yet the documentation says
+ * that only the lower 32 bits can be written to due to the specification
+ * of wrmsr. bits [32-(w-1)] are sign extensions of bit 31. Bits [w-63] must
+ * not be set (see rsvd_msk for PMDs). As such the effective width of a
+ * counter is 31 bits only regardless of what CPUID.0xa returns.
+ *
+ * See IA-32 Intel Architecture Software developer manual Vol 3B chapter 18
+ */
+static struct pfm_pmu_config pfm_intel_arch_pmu_conf = {
+	.pmu_name = "Intel architectural",
+	.pmd_desc = pfm_intel_arch_pmd_desc,
+	.counter_width   = 31,
+	.pmc_desc = pfm_intel_arch_pmc_desc,
+	.version = "1.0",
+	.pmu_info = &pfm_intel_arch_pmu_info
+};
+
+static int __init pfm_intel_arch_pmu_init_module(void)
+{
+	if (pfm_intel_arch_probe_pmu())
+		return -ENOSYS;
+
+	return pfm_pmu_register(&pfm_intel_arch_pmu_conf);
+}
+
+device_initcall(pfm_intel_arch_pmu_init_module);
Index: o/arch/x86/perfmon/Kconfig
===================================================================
--- o.orig/arch/x86/perfmon/Kconfig	2008-06-13 16:40:03.000000000 +0200
+++ o/arch/x86/perfmon/Kconfig	2008-06-13 16:42:25.000000000 +0200
@@ -15,4 +15,11 @@
 	help
 	Enables perfmon debugging support
 
+config  X86_PERFMON_INTEL_ARCH
+	bool "Support for Intel architectural perfmon v1/v2"
+	depends on PERFMON
+	default n
+	help
+	Enables support for Intel architectural performance counters.
+	This feature was introduced with Intel Core Solo/Core Duo processors.
 endmenu

-- 

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists