diff options
author | Hendrik Brueckner <brueckner@linux.vnet.ibm.com> | 2012-03-23 11:13:06 +0100 |
---|---|---|
committer | Martin Schwidefsky <schwidefsky@de.ibm.com> | 2012-03-23 11:13:25 +0100 |
commit | 212188a596d17d519842ef2173150315735b54e1 (patch) | |
tree | 22f4f391fc9979a231028e45fd11b8a6a6c99232 /arch/s390/kernel/perf_event.c | |
parent | [S390] oprofile: Allow multiple users of the measurement alert interrupt (diff) | |
download | linux-212188a596d17d519842ef2173150315735b54e1.tar.xz linux-212188a596d17d519842ef2173150315735b54e1.zip |
[S390] perf: add support for s390x CPU counters
Add a perf PMU to access the CPU-measurement counter facility CPUM CF.
CPUM CF provides multiple counter sets for measuring generic,
problem-state, and crypto activaties. Also an extended counter set for
the IBM System z10 and IBM z196 mainframes is available.
Counters from the basic and problem-state counter set are mapped to
generic perf hardware events. Other counters are accessible through
raw events.
For a list of available counter sets and counters, see:
- The Load-Program-Parameter and the CPU-Measurement Facilities (SA23-2260)
- The CPU-Measurement Facility Extended Counters Definition for
z10 and z196 (SA23-2261)
Reviewed-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Diffstat (limited to 'arch/s390/kernel/perf_event.c')
-rw-r--r-- | arch/s390/kernel/perf_event.c | 125 |
1 files changed, 125 insertions, 0 deletions
diff --git a/arch/s390/kernel/perf_event.c b/arch/s390/kernel/perf_event.c new file mode 100644 index 000000000000..609f985198cf --- /dev/null +++ b/arch/s390/kernel/perf_event.c @@ -0,0 +1,125 @@ +/* + * Performance event support for s390x + * + * Copyright IBM Corp. 2012 + * Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License (version 2 only) + * as published by the Free Software Foundation. + */ +#define KMSG_COMPONENT "perf" +#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt + +#include <linux/kernel.h> +#include <linux/perf_event.h> +#include <linux/percpu.h> +#include <linux/export.h> +#include <asm/system.h> +#include <asm/irq.h> +#include <asm/cpu_mf.h> +#include <asm/lowcore.h> +#include <asm/processor.h> + +const char *perf_pmu_name(void) +{ + if (cpum_cf_avail() || cpum_sf_avail()) + return "CPU-measurement facilities (CPUMF)"; + return "pmu"; +} +EXPORT_SYMBOL(perf_pmu_name); + +int perf_num_counters(void) +{ + int num = 0; + + if (cpum_cf_avail()) + num += PERF_CPUM_CF_MAX_CTR; + + return num; +} +EXPORT_SYMBOL(perf_num_counters); + +void perf_event_print_debug(void) +{ + struct cpumf_ctr_info cf_info; + unsigned long flags; + int cpu; + + if (!cpum_cf_avail()) + return; + + local_irq_save(flags); + + cpu = smp_processor_id(); + memset(&cf_info, 0, sizeof(cf_info)); + if (!qctri(&cf_info)) { + pr_info("CPU[%i] CPUM_CF: ver=%u.%u A=%04x E=%04x C=%04x\n", + cpu, cf_info.cfvn, cf_info.csvn, + cf_info.auth_ctl, cf_info.enable_ctl, cf_info.act_ctl); + print_hex_dump_bytes("CPUMF Query: ", DUMP_PREFIX_OFFSET, + &cf_info, sizeof(cf_info)); + } + + local_irq_restore(flags); +} + +/* See also arch/s390/kernel/traps.c */ +static unsigned long __store_trace(struct perf_callchain_entry *entry, + unsigned long sp, + unsigned long low, unsigned long high) +{ + struct stack_frame *sf; + struct pt_regs *regs; + + while (1) { + sp = sp & PSW_ADDR_INSN; + if (sp < low || sp > high - sizeof(*sf)) + return sp; + sf = (struct stack_frame *) sp; + perf_callchain_store(entry, sf->gprs[8] & PSW_ADDR_INSN); + /* Follow the backchain. */ + while (1) { + low = sp; + sp = sf->back_chain & PSW_ADDR_INSN; + if (!sp) + break; + if (sp <= low || sp > high - sizeof(*sf)) + return sp; + sf = (struct stack_frame *) sp; + perf_callchain_store(entry, + sf->gprs[8] & PSW_ADDR_INSN); + } + /* Zero backchain detected, check for interrupt frame. */ + sp = (unsigned long) (sf + 1); + if (sp <= low || sp > high - sizeof(*regs)) + return sp; + regs = (struct pt_regs *) sp; + perf_callchain_store(entry, sf->gprs[8] & PSW_ADDR_INSN); + low = sp; + sp = regs->gprs[15]; + } +} + +void perf_callchain_kernel(struct perf_callchain_entry *entry, + struct pt_regs *regs) +{ + unsigned long head; + struct stack_frame *head_sf; + + if (user_mode(regs)) + return; + + head = regs->gprs[15]; + head_sf = (struct stack_frame *) head; + + if (!head_sf || !head_sf->back_chain) + return; + + head = head_sf->back_chain; + head = __store_trace(entry, head, S390_lowcore.async_stack - ASYNC_SIZE, + S390_lowcore.async_stack); + + __store_trace(entry, head, S390_lowcore.thread_info, + S390_lowcore.thread_info + THREAD_SIZE); +} |