mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Chen Yu <yu.c.chen@intel.com>
To: Len Brown <lenb@kernel.org>
Cc: linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org,
	Zhang Rui <rui.zhang@intel.com>, Chen Yu <yu.c.chen@intel.com>
Subject: [PATCH 2/2][RFC] tools/power turbostat: Introduce reliable RAPL display
Date: Wed,  1 Apr 2020 02:44:37 +0800	[thread overview]
Message-ID: <3e5aa6e7b34827c0245e10c14ca9457512ae8586.1585679838.git.yu.c.chen@intel.com> (raw)
In-Reply-To: <cover.1585679838.git.yu.c.chen@intel.com>

Since the RAPL Joule Counter is 32 bit, turbostat would
only print a *star* instead of printing the actual energy
consumed to indicate the overflow due to long duration.
This does not meet the requirement on servers as the
sampling time of turbostat is usually very long on servers.

So maintain a set of MSR buffer, and updates them
periodically before the 32bit msr register wrapped round.

The idea is similar to the implementation of ktime_get():
get_msr_sum() is used rather than get_msr() to get the
accumulated MSR.

This can be illustrated below:

MSR timer:
total_rapl_msr += (current_rapl_msr - last_rapl_msr);

get_msr_sum():
return (current_rapl_msr - last_rapl_msr) + total_rapl_msr;

Originally-by: Aaron Lu <aaron.lwe@gmail.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
---
 tools/power/x86/turbostat/Makefile    |   2 +-
 tools/power/x86/turbostat/turbostat.c | 292 ++++++++++++++++++++++++--
 2 files changed, 274 insertions(+), 20 deletions(-)

diff --git a/tools/power/x86/turbostat/Makefile b/tools/power/x86/turbostat/Makefile
index 2b6551269e43..d08765531bcb 100644
--- a/tools/power/x86/turbostat/Makefile
+++ b/tools/power/x86/turbostat/Makefile
@@ -16,7 +16,7 @@ override CFLAGS +=	-D_FORTIFY_SOURCE=2
 
 %: %.c
 	@mkdir -p $(BUILD_OUTPUT)
-	$(CC) $(CFLAGS) $< -o $(BUILD_OUTPUT)/$@ $(LDFLAGS) -lcap
+	$(CC) $(CFLAGS) $< -o $(BUILD_OUTPUT)/$@ $(LDFLAGS) -lcap -lrt
 
 .PHONY : clean
 clean :
diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
index 95f3047e94ae..a8979bec97e4 100644
--- a/tools/power/x86/turbostat/turbostat.c
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -47,6 +47,7 @@ unsigned int sums_need_wide_columns;
 unsigned int rapl_joules;
 unsigned int summary_only;
 unsigned int list_header_only;
+unsigned int longtime;
 unsigned int dump_only;
 unsigned int do_snb_cstates;
 unsigned int do_knl_cstates;
@@ -259,6 +260,113 @@ struct msr_counter {
 #define	SYSFS_PERCPU	(1 << 1)
 };
 
+/*
+ * The accumulated sum of MSR is defined as a monotonic
+ * increasing MSR, it will be accumulated periodically,
+ * despite its register's bit width.
+ */
+enum {
+	IDX_PKG_ENERGY,
+	IDX_DRAM_ENERGY,
+	IDX_PP0_ENERGY,
+	IDX_PP1_ENERGY,
+	IDX_PKG_PERF,
+	IDX_DRAM_PERF,
+	IDX_COUNT,
+};
+
+int get_msr_sum(int cpu, off_t offset, unsigned long long *msr);
+
+struct msr_sum_array {
+	/* get_msr_sum() = sum + (get_msr() - last) */
+	struct {
+		/*The accumulated MSR value is updated by the timer*/
+		unsigned long long sum;
+		/*The MSR footprint recorded in last timer*/
+		unsigned long long last;
+	} entries[IDX_COUNT];
+};
+
+/* The percpu MSR sum array.*/
+struct msr_sum_array *per_cpu_msr_sum;
+
+int idx_to_offset(int idx)
+{
+	int offset;
+
+	switch (idx) {
+	case IDX_PKG_ENERGY:
+		offset = MSR_PKG_ENERGY_STATUS;
+		break;
+	case IDX_DRAM_ENERGY:
+		offset = MSR_DRAM_ENERGY_STATUS;
+		break;
+	case IDX_PP0_ENERGY:
+		offset = MSR_PP0_ENERGY_STATUS;
+		break;
+	case IDX_PP1_ENERGY:
+		offset = MSR_PP1_ENERGY_STATUS;
+		break;
+	case IDX_PKG_PERF:
+		offset = MSR_PKG_PERF_STATUS;
+		break;
+	case IDX_DRAM_PERF:
+		offset = MSR_DRAM_PERF_STATUS;
+		break;
+	default:
+		offset = -1;
+	}
+	return offset;
+}
+
+int offset_to_idx(int offset)
+{
+	int idx;
+
+	switch (offset) {
+	case MSR_PKG_ENERGY_STATUS:
+		idx = IDX_PKG_ENERGY;
+		break;
+	case MSR_DRAM_ENERGY_STATUS:
+		idx = IDX_DRAM_ENERGY;
+		break;
+	case MSR_PP0_ENERGY_STATUS:
+		idx = IDX_PP0_ENERGY;
+		break;
+	case MSR_PP1_ENERGY_STATUS:
+		idx = IDX_PP1_ENERGY;
+		break;
+	case MSR_PKG_PERF_STATUS:
+		idx = IDX_PKG_PERF;
+		break;
+	case MSR_DRAM_PERF_STATUS:
+		idx = IDX_DRAM_PERF;
+		break;
+	default:
+		idx = -1;
+	}
+	return idx;
+}
+
+int idx_valid(int idx)
+{
+	switch (idx) {
+	case IDX_PKG_ENERGY:
+		return do_rapl & RAPL_PKG;
+	case IDX_DRAM_ENERGY:
+		return do_rapl & RAPL_DRAM;
+	case IDX_PP0_ENERGY:
+		return do_rapl & RAPL_CORES_ENERGY_STATUS;
+	case IDX_PP1_ENERGY:
+		return do_rapl & RAPL_GFX;
+	case IDX_PKG_PERF:
+		return do_rapl & RAPL_PKG_PERF_STATUS;
+	case IDX_DRAM_PERF:
+		return do_rapl & RAPL_DRAM_PERF_STATUS;
+	default:
+		return 0;
+	}
+}
 struct sys_counters {
 	unsigned int added_thread_counters;
 	unsigned int added_core_counters;
@@ -551,6 +659,7 @@ void help(void)
 	"		Override default 5-second measurement interval\n"
 	"  -J, --Joules	displays energy in Joules instead of Watts\n"
 	"  -l, --list	list column headers only\n"
+	"  -L, --Longtime	long time duration support\n"
 	"  -n, --num_iterations num\n"
 	"		number of the measurement iterations\n"
 	"  -o, --out file\n"
@@ -1962,34 +2071,70 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
 		p->sys_lpi = cpuidle_cur_sys_lpi_us;
 
 	if (do_rapl & RAPL_PKG) {
-		if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
-			return -13;
-		p->energy_pkg = msr & 0xFFFFFFFF;
+		if (longtime) {
+			if (get_msr_sum(cpu, MSR_PKG_ENERGY_STATUS, &msr))
+				return -13;
+			p->energy_pkg = msr;
+		} else {
+			if (get_msr(cpu, MSR_PKG_ENERGY_STATUS, &msr))
+				return -13;
+			p->energy_pkg = msr & 0xFFFFFFFF;
+		}
 	}
 	if (do_rapl & RAPL_CORES_ENERGY_STATUS) {
-		if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
-			return -14;
-		p->energy_cores = msr & 0xFFFFFFFF;
+		if (longtime) {
+			if (get_msr_sum(cpu, MSR_PP0_ENERGY_STATUS, &msr))
+				return -14;
+			p->energy_cores = msr;
+		} else {
+			if (get_msr(cpu, MSR_PP0_ENERGY_STATUS, &msr))
+				return -14;
+			p->energy_cores = msr & 0xFFFFFFFF;
+		}
 	}
 	if (do_rapl & RAPL_DRAM) {
-		if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
-			return -15;
-		p->energy_dram = msr & 0xFFFFFFFF;
+		if (longtime) {
+			if (get_msr_sum(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
+				return -15;
+			p->energy_dram = msr;
+		} else {
+			if (get_msr(cpu, MSR_DRAM_ENERGY_STATUS, &msr))
+				return -15;
+			p->energy_dram = msr & 0xFFFFFFFF;
+		}
 	}
 	if (do_rapl & RAPL_GFX) {
-		if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
-			return -16;
-		p->energy_gfx = msr & 0xFFFFFFFF;
+		if (longtime) {
+			if (get_msr_sum(cpu, MSR_PP1_ENERGY_STATUS, &msr))
+				return -16;
+			p->energy_gfx = msr;
+		} else {
+			if (get_msr(cpu, MSR_PP1_ENERGY_STATUS, &msr))
+				return -16;
+			p->energy_gfx = msr & 0xFFFFFFFF;
+		}
 	}
 	if (do_rapl & RAPL_PKG_PERF_STATUS) {
-		if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
-			return -16;
-		p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
+		if (longtime) {
+			if (get_msr_sum(cpu, MSR_DRAM_PERF_STATUS, &msr))
+				return -16;
+			p->rapl_dram_perf_status = msr;
+		} else {
+			if (get_msr(cpu, MSR_PKG_PERF_STATUS, &msr))
+				return -16;
+			p->rapl_pkg_perf_status = msr & 0xFFFFFFFF;
+		}
 	}
 	if (do_rapl & RAPL_DRAM_PERF_STATUS) {
-		if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
-			return -16;
-		p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
+		if (longtime) {
+			if (get_msr_sum(cpu, MSR_DRAM_PERF_STATUS, &msr))
+				return -16;
+			p->rapl_dram_perf_status = msr;
+		} else {
+			if (get_msr(cpu, MSR_DRAM_PERF_STATUS, &msr))
+				return -16;
+			p->rapl_dram_perf_status = msr & 0xFFFFFFFF;
+		}
 	}
 	if (do_rapl & RAPL_AMD_F17H) {
 		if (get_msr(cpu, MSR_PKG_ENERGY_STAT, &msr))
@@ -3053,6 +3198,109 @@ void do_sleep(void)
 	}
 }
 
+int get_msr_sum(int cpu, off_t offset, unsigned long long *msr)
+{
+	int ret, idx;
+	unsigned long long msr_cur, msr_last;
+
+	if (!per_cpu_msr_sum)
+		return 1;
+
+	idx = offset_to_idx(offset);
+	if (idx < 0)
+		return idx;
+	/* get_msr_sum() = sum + (get_msr() - last) */
+	ret = get_msr(cpu, offset, &msr_cur);
+	if (ret)
+		return ret;
+	msr_last = per_cpu_msr_sum[cpu].entries[idx].last;
+	DELTA_WRAP32(msr_cur, msr_last);
+	*msr = msr_last + per_cpu_msr_sum[cpu].entries[idx].sum;
+
+	return 0;
+}
+
+timer_t timerid;
+
+/* Timer callback, update the sum of MSRs periodically. */
+static int update_msr_sum(struct thread_data *t, struct core_data *c, struct pkg_data *p)
+{
+	int i, ret;
+	int cpu = t->cpu_id;
+
+	for (i = IDX_PKG_ENERGY; i < IDX_COUNT; i++) {
+		unsigned long long msr_cur, msr_last;
+		int offset;
+
+		if (!idx_valid(i))
+			continue;
+		offset = idx_to_offset(i);
+		if (offset < 0)
+			continue;
+		ret = get_msr(cpu, offset, &msr_cur);
+		if (ret) {
+			fprintf(outf, "Can not update msr(0x%x)\n", offset);
+			continue;
+		}
+
+		msr_last = per_cpu_msr_sum[cpu].entries[i].last;
+		per_cpu_msr_sum[cpu].entries[i].last = msr_cur & 0xffffffff;
+
+		DELTA_WRAP32(msr_cur, msr_last);
+		per_cpu_msr_sum[cpu].entries[i].sum += msr_last;
+	}
+	return 0;
+}
+
+static void
+msr_record_handler(union sigval v)
+{
+	for_all_cpus(update_msr_sum, EVEN_COUNTERS);
+}
+
+void msr_longtime_record(void)
+{
+	struct itimerspec its;
+	struct sigevent sev;
+
+	per_cpu_msr_sum = calloc(topo.max_cpu_num + 1, sizeof(struct msr_sum_array));
+	if (!per_cpu_msr_sum) {
+		fprintf(outf, "Can not allocate memory for long time MSR.\n");
+		return;
+	}
+	/*
+	 * Signal handler might be restricted, so use thread notifier instead.
+	 */
+	memset(&sev, 0, sizeof(struct sigevent));
+	sev.sigev_notify = SIGEV_THREAD;
+	sev.sigev_notify_function = msr_record_handler;
+
+	sev.sigev_value.sival_ptr = &timerid;
+	if (timer_create(CLOCK_REALTIME, &sev, &timerid) == -1) {
+		fprintf(outf, "Can not create timer.\n");
+		goto release_msr;
+	}
+
+	its.it_value.tv_sec = 0;
+	its.it_value.tv_nsec = 1;
+	/*
+	 * A wraparound time of around 60 secs when power consumption
+	 * is high, use 50 secs.
+	 */
+	its.it_interval.tv_sec = 50;
+	its.it_interval.tv_nsec = 0;
+
+	if (timer_settime(timerid, 0, &its, NULL) == -1) {
+		fprintf(outf, "Can not set timer.\n");
+		goto release_timer;
+	}
+	return;
+
+ release_timer:
+	timer_delete(timerid);
+ release_msr:
+	free(per_cpu_msr_sum);
+}
 
 void turbostat_loop()
 {
@@ -5735,6 +5983,7 @@ void cmdline(int argc, char **argv)
 		{"hide",	required_argument,	0, 'H'},	// meh, -h taken by --help
 		{"Joules",	no_argument,		0, 'J'},
 		{"list",	no_argument,		0, 'l'},
+		{"Longtime",	no_argument,		0, 'L'},
 		{"out",		required_argument,	0, 'o'},
 		{"quiet",	no_argument,		0, 'q'},
 		{"show",	required_argument,	0, 's'},
@@ -5746,7 +5995,7 @@ void cmdline(int argc, char **argv)
 
 	progname = argv[0];
 
-	while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v",
+	while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:JLn:o:qST:v",
 				long_options, &option_index)) != -1) {
 		switch (opt) {
 		case 'a':
@@ -5800,6 +6049,9 @@ void cmdline(int argc, char **argv)
 			list_header_only++;
 			quiet++;
 			break;
+		case 'L':
+			longtime = 1;
+			break;
 		case 'o':
 			outf = fopen_or_die(optarg, "w");
 			break;
@@ -5864,6 +6116,8 @@ int main(int argc, char **argv)
 		return 0;
 	}
 
+	if (longtime)
+		msr_longtime_record();
 	/*
 	 * if any params left, it must be a command to fork
 	 */
-- 
2.17.1


  parent reply	other threads:[~2020-03-31 18:45 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-31 18:44 [PATCH 0/2][RFC] Add long time sampling time support Chen Yu
2020-03-31 18:44 ` [PATCH 1/2][RFC] tools/power turbostat: Make the energy variable to be 64 bit Chen Yu
2020-03-31 18:44 ` Chen Yu [this message]
2020-08-13 21:50   ` [PATCH 2/2][RFC] tools/power turbostat: Introduce reliable RAPL display Len Brown
2020-08-14 14:34     ` Chen, Yu C

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=3e5aa6e7b34827c0245e10c14ca9457512ae8586.1585679838.git.yu.c.chen@intel.com \
    --to=yu.c.chen@intel.com \
    --cc=lenb@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pm@vger.kernel.org \
    --cc=rui.zhang@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®