From: "Alex Bennée" <alex.bennee@linaro.org>
To: qemu-devel@nongnu.org
Cc: fam@euphon.net, berrange@redhat.com, robert.foley@linaro.org,
"Alex Bennée" <alex.bennee@linaro.org>,
richard.henderson@linaro.org, f4bug@amsat.org,
robhenry@microsoft.com,
"Philippe Mathieu-Daudé" <philmd@redhat.com>,
aaron@os.amperecomputing.com, cota@braap.org,
kuhn.chenqun@huawei.com, peter.puhov@linaro.org,
aurelien@aurel32.net
Subject: [PATCH v2 11/11] plugins: new hwprofile plugin
Date: Mon, 13 Jul 2020 21:04:15 +0100 [thread overview]
Message-ID: <20200713200415.26214-12-alex.bennee@linaro.org> (raw)
In-Reply-To: <20200713200415.26214-1-alex.bennee@linaro.org>
This is a plugin intended to help with profiling access to various
bits of system hardware. It only really makes sense for system
emulation.
It takes advantage of the recently exposed helper API that allows us
to see the device name (memory region name) associated with a device.
You can specify arg=read or arg=write to limit the tracking to just
reads or writes (by default it does both).
The pattern option:
-plugin ./tests/plugin/libhwprofile.so,arg=pattern
will allow you to see the access pattern to devices, eg:
gic_cpu @ 0xffffffc010040000
off:00000000, 8, 1, 8, 1
off:00000000, 4, 1, 4, 1
off:00000000, 2, 1, 2, 1
off:00000000, 1, 1, 1, 1
The source option:
-plugin ./tests/plugin/libhwprofile.so,arg=source
will track the virtual source address of the instruction making the
access:
pl011 @ 0xffffffc010031000
pc:ffffffc0104c785c, 1, 4, 0, 0
pc:ffffffc0104c7898, 1, 4, 0, 0
pc:ffffffc010512bcc, 2, 1867, 0, 0
You cannot mix source and pattern.
Finally the match option allow you to limit the tracking to just the
devices you care about.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Robert Foley <robert.foley@linaro.org>
Tested-by: Robert Foley <robert.foley@linaro.org>
---
v2
- use 64 bit cpu masks
- warn if we will exceed cpu mask capability
- don't run in linux-user mode
- skip in check-tcg for linux-user test
v3
- update device name API
- re-factor and clean-up
- add source tracking
- use direct hash for source/pattern
- add match function
- expand the commit message
- checkpatch cleanups
---
tests/plugin/hwprofile.c | 305 ++++++++++++++++++++++++++++++++++++++
tests/plugin/Makefile | 1 +
tests/tcg/Makefile.target | 12 +-
3 files changed, 316 insertions(+), 2 deletions(-)
create mode 100644 tests/plugin/hwprofile.c
diff --git a/tests/plugin/hwprofile.c b/tests/plugin/hwprofile.c
new file mode 100644
index 0000000000..6dac1d5f85
--- /dev/null
+++ b/tests/plugin/hwprofile.c
@@ -0,0 +1,305 @@
+/*
+ * Copyright (C) 2020, Alex Bennée <alex.bennee@linaro.org>
+ *
+ * HW Profile - breakdown access patterns for IO to devices
+ *
+ * License: GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include <inttypes.h>
+#include <assert.h>
+#include <stdlib.h>
+#include <inttypes.h>
+#include <string.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <glib.h>
+
+#include <qemu-plugin.h>
+
+QEMU_PLUGIN_EXPORT int qemu_plugin_version = QEMU_PLUGIN_VERSION;
+
+#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
+
+typedef struct {
+ uint64_t cpu_read;
+ uint64_t cpu_write;
+ uint64_t reads;
+ uint64_t writes;
+} IOCounts;
+
+typedef struct {
+ uint64_t off_or_pc;
+ IOCounts counts;
+} IOLocationCounts;
+
+typedef struct {
+ const char *name;
+ uint64_t base;
+ IOCounts totals;
+ GHashTable *detail;
+} DeviceCounts;
+
+static GMutex lock;
+static GHashTable *devices;
+
+/* track the access pattern to a piece of HW */
+static bool pattern;
+/* track the source address of access to HW */
+static bool source;
+/* track only matched regions of HW */
+static bool check_match;
+static gchar **matches;
+
+static enum qemu_plugin_mem_rw rw = QEMU_PLUGIN_MEM_RW;
+
+static inline bool track_reads(void)
+{
+ return rw == QEMU_PLUGIN_MEM_RW || rw == QEMU_PLUGIN_MEM_R;
+}
+
+static inline bool track_writes(void)
+{
+ return rw == QEMU_PLUGIN_MEM_RW || rw == QEMU_PLUGIN_MEM_W;
+}
+
+static void plugin_init(void)
+{
+ devices = g_hash_table_new(NULL, NULL);
+}
+
+static gint sort_cmp(gconstpointer a, gconstpointer b)
+{
+ DeviceCounts *ea = (DeviceCounts *) a;
+ DeviceCounts *eb = (DeviceCounts *) b;
+ return ea->totals.reads + ea->totals.writes >
+ eb->totals.reads + eb->totals.writes ? -1 : 1;
+}
+
+static gint sort_loc(gconstpointer a, gconstpointer b)
+{
+ IOLocationCounts *ea = (IOLocationCounts *) a;
+ IOLocationCounts *eb = (IOLocationCounts *) b;
+ return ea->off_or_pc > eb->off_or_pc;
+}
+
+static void fmt_iocount_record(GString *s, IOCounts *rec)
+{
+ if (track_reads()) {
+ g_string_append_printf(s, ", %"PRIx64", %"PRId64,
+ rec->cpu_read, rec->reads);
+ }
+ if (track_writes()) {
+ g_string_append_printf(s, ", %"PRIx64", %"PRId64,
+ rec->cpu_write, rec->writes);
+ }
+}
+
+static void fmt_dev_record(GString *s, DeviceCounts *rec)
+{
+ g_string_append_printf(s, "%s, 0x%"PRIx64,
+ rec->name, rec->base);
+ fmt_iocount_record(s, &rec->totals);
+ g_string_append_c(s, '\n');
+}
+
+static void plugin_exit(qemu_plugin_id_t id, void *p)
+{
+ g_autoptr(GString) report = g_string_new("");
+ GList *counts;
+
+ if (!(pattern || source)) {
+ g_string_printf(report, "Device, Address");
+ if (track_reads()) {
+ g_string_append_printf(report, ", RCPUs, Reads");
+ }
+ if (track_writes()) {
+ g_string_append_printf(report, ", WCPUs, Writes");
+ }
+ g_string_append_c(report, '\n');
+ }
+
+ counts = g_hash_table_get_values(devices);
+ if (counts && g_list_next(counts)) {
+ GList *it;
+
+ it = g_list_sort(counts, sort_cmp);
+
+ while (it) {
+ DeviceCounts *rec = (DeviceCounts *) it->data;
+ if (rec->detail) {
+ GList *accesses = g_hash_table_get_values(rec->detail);
+ GList *io_it = g_list_sort(accesses, sort_loc);
+ const char *prefix = pattern ? "off" : "pc";
+ g_string_append_printf(report, "%s @ 0x%"PRIx64"\n",
+ rec->name, rec->base);
+ while (io_it) {
+ IOLocationCounts *loc = (IOLocationCounts *) io_it->data;
+ g_string_append_printf(report, " %s:%08"PRIx64,
+ prefix, loc->off_or_pc);
+ fmt_iocount_record(report, &loc->counts);
+ g_string_append_c(report, '\n');
+ io_it = io_it->next;
+ }
+ } else {
+ fmt_dev_record(report, rec);
+ }
+ it = it->next;
+ };
+ g_list_free(it);
+ }
+
+ qemu_plugin_outs(report->str);
+}
+
+static DeviceCounts *new_count(const char *name, uint64_t base)
+{
+ DeviceCounts *count = g_new0(DeviceCounts, 1);
+ count->name = name;
+ count->base = base;
+ if (pattern || source) {
+ count->detail = g_hash_table_new(NULL, NULL);
+ }
+ g_hash_table_insert(devices, (gpointer) name, count);
+ return count;
+}
+
+static IOLocationCounts *new_location(GHashTable *table, uint64_t off_or_pc)
+{
+ IOLocationCounts *loc = g_new0(IOLocationCounts, 1);
+ loc->off_or_pc = off_or_pc;
+ g_hash_table_insert(table, (gpointer) off_or_pc, loc);
+ return loc;
+}
+
+static void hwprofile_match_hit(DeviceCounts *rec, uint64_t off)
+{
+ g_autoptr(GString) report = g_string_new("hwprofile: match @ offset");
+ g_string_append_printf(report, "%"PRIx64", previous hits\n", off);
+ fmt_dev_record(report, rec);
+ qemu_plugin_outs(report->str);
+}
+
+static void inc_count(IOCounts *count, bool is_write, unsigned int cpu_index)
+{
+ if (is_write) {
+ count->writes++;
+ count->cpu_write |= (1 << cpu_index);
+ } else {
+ count->reads++;
+ count->cpu_read |= (1 << cpu_index);
+ }
+}
+
+static void vcpu_haddr(unsigned int cpu_index, qemu_plugin_meminfo_t meminfo,
+ uint64_t vaddr, void *udata)
+{
+ struct qemu_plugin_hwaddr *hwaddr = qemu_plugin_get_hwaddr(meminfo, vaddr);
+
+ if (!hwaddr || !qemu_plugin_hwaddr_is_io(hwaddr)) {
+ return;
+ } else {
+ const char *name = qemu_plugin_hwaddr_device_name(hwaddr);
+ uint64_t off = qemu_plugin_hwaddr_device_offset(hwaddr);
+ bool is_write = qemu_plugin_mem_is_store(meminfo);
+ DeviceCounts *counts;
+
+ g_mutex_lock(&lock);
+ counts = (DeviceCounts *) g_hash_table_lookup(devices, name);
+
+ if (!counts) {
+ uint64_t base = vaddr - off;
+ counts = new_count(name, base);
+ }
+
+ if (check_match) {
+ if (g_strv_contains((const char * const *)matches, counts->name)) {
+ hwprofile_match_hit(counts, off);
+ inc_count(&counts->totals, is_write, cpu_index);
+ }
+ } else {
+ inc_count(&counts->totals, is_write, cpu_index);
+ }
+
+ /* either track offsets or source of access */
+ if (source) {
+ off = (uint64_t) udata;
+ }
+
+ if (pattern || source) {
+ IOLocationCounts *io_count = g_hash_table_lookup(counts->detail,
+ (gpointer) off);
+ if (!io_count) {
+ io_count = new_location(counts->detail, off);
+ }
+ inc_count(&io_count->counts, is_write, cpu_index);
+ }
+
+ g_mutex_unlock(&lock);
+ }
+}
+
+static void vcpu_tb_trans(qemu_plugin_id_t id, struct qemu_plugin_tb *tb)
+{
+ size_t n = qemu_plugin_tb_n_insns(tb);
+ size_t i;
+
+ for (i = 0; i < n; i++) {
+ struct qemu_plugin_insn *insn = qemu_plugin_tb_get_insn(tb, i);
+ gpointer udata = (gpointer) (source ? qemu_plugin_insn_vaddr(insn) : 0);
+ qemu_plugin_register_vcpu_mem_cb(insn, vcpu_haddr,
+ QEMU_PLUGIN_CB_NO_REGS,
+ rw, udata);
+ }
+}
+
+QEMU_PLUGIN_EXPORT
+int qemu_plugin_install(qemu_plugin_id_t id, const qemu_info_t *info,
+ int argc, char **argv)
+{
+ int i;
+
+ for (i = 0; i < argc; i++) {
+ char *opt = argv[i];
+ if (g_strcmp0(opt, "read") == 0) {
+ rw = QEMU_PLUGIN_MEM_R;
+ } else if (g_strcmp0(opt, "write") == 0) {
+ rw = QEMU_PLUGIN_MEM_W;
+ } else if (g_strcmp0(opt, "pattern") == 0) {
+ pattern = true;
+ } else if (g_strcmp0(opt, "source") == 0) {
+ source = true;
+ } else if (g_str_has_prefix(opt, "match")) {
+ gchar **parts = g_strsplit(opt, "=", 2);
+ check_match = true;
+ matches = g_strsplit(parts[1], ",", -1);
+ g_strfreev(parts);
+ } else {
+ fprintf(stderr, "option parsing failed: %s\n", opt);
+ return -1;
+ }
+ }
+
+ if (source && pattern) {
+ fprintf(stderr, "can only currently track either source or pattern.\n");
+ return -1;
+ }
+
+ if (!info->system_emulation) {
+ fprintf(stderr, "hwprofile: plugin only useful for system emulation\n");
+ return -1;
+ }
+
+ /* Just warn about overflow */
+ if (info->system.smp_vcpus > 64 ||
+ info->system.max_vcpus > 64) {
+ fprintf(stderr, "hwprofile: can only track up to 64 CPUs\n");
+ }
+
+ plugin_init();
+
+ qemu_plugin_register_vcpu_tb_trans_cb(id, vcpu_tb_trans);
+ qemu_plugin_register_atexit_cb(id, plugin_exit, NULL);
+ return 0;
+}
diff --git a/tests/plugin/Makefile b/tests/plugin/Makefile
index e9348fde4a..e9d1947751 100644
--- a/tests/plugin/Makefile
+++ b/tests/plugin/Makefile
@@ -21,6 +21,7 @@ NAMES += hotblocks
NAMES += howvec
NAMES += hotpages
NAMES += lockstep
+NAMES += hwprofile
SONAMES := $(addsuffix .so,$(addprefix lib,$(NAMES)))
diff --git a/tests/tcg/Makefile.target b/tests/tcg/Makefile.target
index 4b2b696fce..bc595b0588 100644
--- a/tests/tcg/Makefile.target
+++ b/tests/tcg/Makefile.target
@@ -129,8 +129,16 @@ ifeq ($(CONFIG_PLUGIN),y)
PLUGIN_SRC=$(SRC_PATH)/tests/plugin
PLUGIN_LIB=../../plugin
VPATH+=$(PLUGIN_LIB)
-PLUGINS=$(filter-out liblockstep.so,\
- $(patsubst %.c, lib%.so, $(notdir $(wildcard $(PLUGIN_SRC)/*.c))))
+
+# Some plugins aren't testable automatically
+SKIP_PLUGINS=liblockstep.so
+ifdef CONFIG_USER_ONLY
+SKIP_PLUGINS+=libhwprofile.so
+endif
+
+PLUGINS=$(filter-out $(SKIP_PLUGINS),\
+ $(patsubst %.c, lib%.so, \
+ $(notdir $(wildcard $(PLUGIN_SRC)/*.c))))
# We need to ensure expand the run-plugin-TEST-with-PLUGIN
# pre-requistes manually here as we can't use stems to handle it. We
--
2.20.1
prev parent reply other threads:[~2020-07-13 20:09 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-07-13 20:04 [PATCH v2 00/11] misc fixes for rc0 (docker, plugins, softfloat) Alex Bennée
2020-07-13 20:04 ` [PATCH v2 01/11] configure: remove all dependencies on a (re)configure Alex Bennée
2020-07-13 20:15 ` Philippe Mathieu-Daudé
2020-07-13 20:04 ` [PATCH v2 02/11] tests/docker: Remove the libssh workaround from the ubuntu 20.04 image Alex Bennée
2020-07-13 20:04 ` [PATCH v2 03/11] docker.py: fix fetching of FROM layers Alex Bennée
2020-07-13 20:04 ` [PATCH v2 04/11] fpu/softfloat: fix up float16 nan recognition Alex Bennée
2020-07-13 20:04 ` [PATCH v2 05/11] tests/plugins: don't unconditionally add -Wpsabi Alex Bennée
2020-07-14 5:31 ` Thomas Huth
2020-07-13 20:04 ` [PATCH v2 06/11] cputlb: ensure we save the IOTLB data in case of reset Alex Bennée
2020-07-13 21:58 ` Richard Henderson
2020-07-18 20:51 ` Emilio G. Cota
2020-07-13 20:04 ` [PATCH v2 07/11] plugins: expand the bb plugin to be thread safe and track per-cpu Alex Bennée
2020-07-13 20:04 ` [PATCH v2 08/11] docs/devel: fix grammar in multi-thread-tcg Alex Bennée
2020-07-13 22:01 ` Richard Henderson
2020-07-14 5:41 ` Thomas Huth
2020-07-14 10:20 ` Philippe Mathieu-Daudé
2020-07-13 20:04 ` [PATCH v2 09/11] hw/virtio/pci: include vdev name in registered PCI sections Alex Bennée
2020-07-14 9:07 ` Michael S. Tsirkin
2020-07-14 9:49 ` Philippe Mathieu-Daudé
2020-07-13 20:04 ` [PATCH v2 10/11] plugins: add API to return a name for a IO device Alex Bennée
2020-07-13 22:04 ` Richard Henderson
2020-07-13 20:04 ` Alex Bennée [this message]
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=20200713200415.26214-12-alex.bennee@linaro.org \
--to=alex.bennee@linaro.org \
--cc=aaron@os.amperecomputing.com \
--cc=aurelien@aurel32.net \
--cc=berrange@redhat.com \
--cc=cota@braap.org \
--cc=f4bug@amsat.org \
--cc=fam@euphon.net \
--cc=kuhn.chenqun@huawei.com \
--cc=peter.puhov@linaro.org \
--cc=philmd@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=robert.foley@linaro.org \
--cc=robhenry@microsoft.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;
as well as URLs for NNTP newsgroup(s).