From patchwork Thu Mar 29 18:46:22 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Tvrtko Ursulin X-Patchwork-Id: 10316057 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork.web.codeaurora.org (Postfix) with ESMTP id 8590460353 for ; Thu, 29 Mar 2018 18:46:40 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 792952A491 for ; Thu, 29 Mar 2018 18:46:40 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 6D5922A49C; Thu, 29 Mar 2018 18:46:40 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-4.1 required=2.0 tests=BAYES_00,DKIM_SIGNED, RCVD_IN_DNSWL_MED,T_DKIM_INVALID autolearn=ham version=3.3.1 Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher DHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.wl.linuxfoundation.org (Postfix) with ESMTPS id 8FE4A2A491 for ; Thu, 29 Mar 2018 18:46:38 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 5CA186E2D3; Thu, 29 Mar 2018 18:46:36 +0000 (UTC) X-Original-To: Intel-gfx@lists.freedesktop.org Delivered-To: Intel-gfx@lists.freedesktop.org Received: from mail-wm0-x235.google.com (mail-wm0-x235.google.com [IPv6:2a00:1450:400c:c09::235]) by gabe.freedesktop.org (Postfix) with ESMTPS id D448C88DF5 for ; Thu, 29 Mar 2018 18:46:33 +0000 (UTC) Received: by mail-wm0-x235.google.com with SMTP id f125so13271922wme.4 for ; Thu, 29 Mar 2018 11:46:33 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=ursulin-net.20150623.gappssmtp.com; s=20150623; h=from:to:cc:subject:date:message-id:in-reply-to:references :mime-version:content-transfer-encoding; bh=TlGyfuu2SYiSEK9y+6T727DRxYjJquSZXWrOmYfEiug=; b=MIUPQDlAEAV2hpodwgpk05gt/Buf2RdScw4dc09IW1wRbEtg+JTWjTTNKYS5C7naUA DENe0auCSqmrmxsWFsGCluMRHTUkiylDamGNnYIzFYO5gbMDYA/P1fLKV1XL351DhIr2 66rgIN8xAr8/DIVSaN54aTEpDXNR/5rXB974p6jzofZg40wZvkMpwYHH44NZYinULnRD aUdLlQd8cFwAs5TjKPBDhy+xrGAFBpH1uJdzBPHKcn+73NK8mSzHVi4XRIyiwGMfoKrK emUFuJWxWQB8APQIswxlN42lP9KTn1IwaajnwliJGd/Hg3YNIv9bA1mUAhm6vyHNUgXP QA6A== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=TlGyfuu2SYiSEK9y+6T727DRxYjJquSZXWrOmYfEiug=; b=Vr+O2kFgnlmMF4dfAdAMeApMbV2gHXQMCorBJ41oeeXi1IIXFLMBTUf6IJzPK/l7Sh tJ4+Bpf0kjCM6/2h2HAUOymn/rfJvyVV2tv716rRYKLigbDcUZdXSQ/sRKDgFTkzjsnv +rj7x5SdiWmaBvXc6X6aH2RZtmgf4ajoyx1hqStyyVMNqfHVrPtZ+xqdRQeXTENCiV41 pBgFxhijqSfmWoNb3SDF1v3ZcvaPgZwwgX7OZ6D/gJuC4TEFuAXT799MJVwBOFaUT9OO ZwrA53SoXomyu+HbkQ/Y6MvF2NRZlDu0GSwn+yXLB2ViKUa5ZFELgTSLqkrvdd86IOry P7Kw== X-Gm-Message-State: AElRT7EaTYLo5ktb8EipT2tDu46u5uxtuQcTl5/xPH6t1q88f+5Tldkt fm9kI8KE5iMwS0Pa0cQzhIZ/jw== X-Google-Smtp-Source: AIpwx4/k5AWdxeMd0cuzs0QwJp68/yRslN8NTjG1JX5NbKjNDujlF/XaC7p5AlcNCpDEayUo/1mQ+Q== X-Received: by 10.28.109.131 with SMTP id b3mr65295wmi.145.1522349192181; Thu, 29 Mar 2018 11:46:32 -0700 (PDT) Received: from localhost.localdomain ([95.146.144.186]) by smtp.gmail.com with ESMTPSA id y144sm4086289wmd.47.2018.03.29.11.46.31 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Thu, 29 Mar 2018 11:46:31 -0700 (PDT) From: Tvrtko Ursulin X-Google-Original-From: Tvrtko Ursulin To: igt-dev@lists.freedesktop.org Date: Thu, 29 Mar 2018 19:46:22 +0100 Message-Id: <20180329184622.10480-1-tvrtko.ursulin@linux.intel.com> X-Mailer: git-send-email 2.14.1 In-Reply-To: <63199548-08b6-446f-4185-082a87287ed3@intel.com> References: <63199548-08b6-446f-4185-082a87287ed3@intel.com> MIME-Version: 1.0 Subject: [Intel-gfx] [PATCH i-g-t v3] intel-gpu-top: Rewrite the tool to be safe to use X-BeenThere: intel-gfx@lists.freedesktop.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: Intel graphics driver community testing & development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Rinat Ibragimov , Eero Tamminen , Intel-gfx@lists.freedesktop.org Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" X-Virus-Scanned: ClamAV using ClamSMTP From: Tvrtko Ursulin intel-gpu-top is a dangerous tool which can hang machines due unsafe mmio register access. This patch rewrites it to use only PMU. Only overall command streamer busyness and GPU global data such as power and frequencies are included in this new version. For access to more GPU functional unit level data, an OA metric based tool like gpu-top should be used instead. v2: * Sort engines by class and instance. * Do not wait for one sampling period to display something on screen. * Move code out of the asserts. (Rinat Ibragimov) * Continuously adapt to terminal size. (Rinat Ibgragimov) v3: * Change layout and precision of some field. (Chris Wilson) Eero Tamminen: * Use more user friendly engine names. * Don't error out if a counter is missing. * Add IMC read/write bandwidth. * Report minimum required kernel version. Signed-off-by: Tvrtko Ursulin Cc: Chris Wilson Cc: Lionel Landwerlin Cc: Petri Latvala Cc: Eero Tamminen Cc: Rinat Ibragimov Reviewed-by: Lionel Landwerlin # v1 Reviewed-by: Chris Wilson # v0.5 --- lib/igt_perf.c | 6 + lib/igt_perf.h | 1 + tools/Makefile.am | 2 + tools/intel_gpu_top.c | 1216 ++++++++++++++++++++++++++----------------------- tools/meson.build | 6 +- 5 files changed, 657 insertions(+), 574 deletions(-) diff --git a/lib/igt_perf.c b/lib/igt_perf.c index 99d82ea51c9b..e3dec2cc29c7 100644 --- a/lib/igt_perf.c +++ b/lib/igt_perf.c @@ -69,3 +69,9 @@ int igt_perf_open(uint64_t type, uint64_t config) return _perf_open(type, config, -1, PERF_FORMAT_TOTAL_TIME_ENABLED); } + +int igt_perf_open_group(uint64_t type, uint64_t config, int group) +{ + return _perf_open(type, config, group, + PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_FORMAT_GROUP); +} diff --git a/lib/igt_perf.h b/lib/igt_perf.h index 614ea5d23fa6..e00718f4769a 100644 --- a/lib/igt_perf.h +++ b/lib/igt_perf.h @@ -55,5 +55,6 @@ uint64_t i915_type_id(void); int perf_i915_open(uint64_t config); int perf_i915_open_group(uint64_t config, int group); int igt_perf_open(uint64_t type, uint64_t config); +int igt_perf_open_group(uint64_t type, uint64_t config, int group); #endif /* I915_PERF_H */ diff --git a/tools/Makefile.am b/tools/Makefile.am index 09b6dbcc3ece..a0b016ddd7ff 100644 --- a/tools/Makefile.am +++ b/tools/Makefile.am @@ -28,6 +28,8 @@ intel_aubdump_la_LDFLAGS = -module -avoid-version -no-undefined intel_aubdump_la_SOURCES = aubdump.c intel_aubdump_la_LIBADD = $(top_builddir)/lib/libintel_tools.la -ldl +intel_gpu_top_LDADD = $(top_builddir)/lib/libigt_perf.la + bin_SCRIPTS = intel_aubdump CLEANFILES = $(bin_SCRIPTS) diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c index 098e6ce3ff86..46a3db42d3d6 100644 --- a/tools/intel_gpu_top.c +++ b/tools/intel_gpu_top.c @@ -1,6 +1,5 @@ /* - * Copyright © 2007 Intel Corporation - * Copyright © 2011 Intel Corporation + * Copyright © 2018 Intel Corporation * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -18,701 +17,772 @@ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER - * DEALINGS IN THE SOFTWARE. - * - * Authors: - * Eric Anholt - * Eugeni Dodonov - * + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. */ -#include "config.h" - -#include -#include -#include #include -#include -#include -#include -#include +#include +#include +#include +#include #include -#ifdef HAVE_TERMIOS_H -#include -#endif -#include "intel_io.h" -#include "instdone.h" -#include "intel_reg.h" -#include "intel_chipset.h" -#include "drmtest.h" - -#define FORCEWAKE 0xA18C -#define FORCEWAKE_ACK 0x130090 - -#define SAMPLES_PER_SEC 10000 -#define SAMPLES_TO_PERCENT_RATIO (SAMPLES_PER_SEC / 100) - -#define MAX_NUM_TOP_BITS 100 - -#define HAS_STATS_REGS(devid) IS_965(devid) - -struct top_bit { - struct instdone_bit *bit; - int count; -} top_bits[MAX_NUM_TOP_BITS]; -struct top_bit *top_bits_sorted[MAX_NUM_TOP_BITS]; - -static uint32_t instdone, instdone1; - -static const char *bars[] = { - " ", - "▏", - "▎", - "▍", - "▌", - "▋", - "▊", - "▉", - "█" -}; +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "igt_perf.h" -enum stats_counts { - IA_VERTICES, - IA_PRIMITIVES, - VS_INVOCATION, - GS_INVOCATION, - GS_PRIMITIVES, - CL_INVOCATION, - CL_PRIMITIVES, - PS_INVOCATION, - PS_DEPTH, - STATS_COUNT +struct pmu_pair { + uint64_t cur; + uint64_t prev; }; -const uint32_t stats_regs[STATS_COUNT] = { - IA_VERTICES_COUNT_QW, - IA_PRIMITIVES_COUNT_QW, - VS_INVOCATION_COUNT_QW, - GS_INVOCATION_COUNT_QW, - GS_PRIMITIVES_COUNT_QW, - CL_INVOCATION_COUNT_QW, - CL_PRIMITIVES_COUNT_QW, - PS_INVOCATION_COUNT_QW, - PS_DEPTH_COUNT_QW, +struct pmu_counter { + bool present; + uint64_t config; + unsigned int idx; + struct pmu_pair val; }; -const char *stats_reg_names[STATS_COUNT] = { - "vert fetch", - "prim fetch", - "VS invocations", - "GS invocations", - "GS prims", - "CL invocations", - "CL prims", - "PS invocations", - "PS depth pass", +struct engine { + const char *name; + const char *display_name; + + unsigned int class; + unsigned int instance; + + struct pmu_counter busy; + struct pmu_counter wait; + struct pmu_counter sema; }; -uint64_t stats[STATS_COUNT]; -uint64_t last_stats[STATS_COUNT]; +struct engines { + unsigned int num_engines; + unsigned int num_counters; + DIR *root; + int fd; + struct pmu_pair ts; + + int rapl_fd; + double rapl_scale; + + int imc_fd; + double imc_reads_scale; + const char *imc_reads_unit; + double imc_writes_scale; + const char *imc_writes_unit; + + struct pmu_counter freq_req; + struct pmu_counter freq_act; + struct pmu_counter irq; + struct pmu_counter rc6; + struct pmu_counter rapl; + struct pmu_counter imc_reads; + struct pmu_counter imc_writes; + + struct engine engine; +}; -static unsigned long -gettime(void) +static uint64_t +get_pmu_config(int dirfd, const char *name, const char *counter) { - struct timeval t; - gettimeofday(&t, NULL); - return (t.tv_usec + (t.tv_sec * 1000000)); -} + char buf[128], *p; + int fd, ret; -static int -top_bits_sort(const void *a, const void *b) -{ - struct top_bit * const *bit_a = a; - struct top_bit * const *bit_b = b; - int a_count = (*bit_a)->count; - int b_count = (*bit_b)->count; + ret = snprintf(buf, sizeof(buf), "%s-%s", name, counter); + if (ret < 0 || ret == sizeof(buf)) + return -1; - if (a_count < b_count) - return 1; - else if (a_count == b_count) - return 0; - else + fd = openat(dirfd, buf, O_RDONLY); + if (fd < 0) return -1; -} -static void -update_idle_bit(struct top_bit *top_bit) -{ - uint32_t reg_val; + ret = read(fd, buf, sizeof(buf)); + close(fd); + if (ret <= 0) + return -1; - if (top_bit->bit->reg == INSTDONE_1) - reg_val = instdone1; - else - reg_val = instdone; + p = index(buf, '0'); + if (!p) + return -1; - if ((reg_val & top_bit->bit->bit) == 0) - top_bit->count++; + return strtoul(p, NULL, 0); } -static void -print_clock(const char *name, int clock) { - if (clock == -1) - printf("%s clock: unknown", name); +#define engine_ptr(engines, n) \ + ((struct engine *)((unsigned char *)(&engines->engine) + (n) * sizeof(struct engine))) + +static const char *class_display_name(unsigned int class) +{ + switch (class) { + case I915_ENGINE_CLASS_RENDER: + return "Render/3D"; + case I915_ENGINE_CLASS_COPY: + return "Blitter"; + case I915_ENGINE_CLASS_VIDEO: + return "Video"; + case I915_ENGINE_CLASS_VIDEO_ENHANCE: + return "VideoEnhance"; + default: + return "[unknown]"; + } +} + +static int engine_cmp(const void *__a, const void *__b) +{ + const struct engine *a = (struct engine *)__a; + const struct engine *b = (struct engine *)__b; + + if (a->class != b->class) + return a->class - b->class; else - printf("%s clock: %d Mhz", name, clock); + return a->instance - b->instance; } -static int -print_clock_info(struct pci_device *pci_dev) +static struct engines *discover_engines(void) { - uint32_t devid = pci_dev->device_id; - uint16_t gcfgc; + const char *sysfs_root = "/sys/devices/i915/events"; + struct engines *engines; + struct dirent *dent; + int ret = 0; + DIR *d; - if (IS_GM45(devid)) { - int core_clock = -1; + engines = malloc(sizeof(struct engines)); + if (!engines) + return NULL; - pci_device_cfg_read_u16(pci_dev, &gcfgc, I915_GCFGC); + memset(engines, 0, sizeof(*engines)); - switch (gcfgc & 0xf) { - case 8: - core_clock = 266; - break; - case 9: - core_clock = 320; - break; - case 11: - core_clock = 400; - break; - case 13: - core_clock = 533; - break; - } - print_clock("core", core_clock); - } else if (IS_965(devid) && IS_MOBILE(devid)) { - int render_clock = -1, sampler_clock = -1; + engines->num_engines = 0; - pci_device_cfg_read_u16(pci_dev, &gcfgc, I915_GCFGC); + d = opendir(sysfs_root); + if (!d) + return NULL; - switch (gcfgc & 0xf) { - case 2: - render_clock = 250; sampler_clock = 267; - break; - case 3: - render_clock = 320; sampler_clock = 333; - break; - case 4: - render_clock = 400; sampler_clock = 444; - break; - case 5: - render_clock = 500; sampler_clock = 533; + while ((dent = readdir(d)) != NULL) { + const char *endswith = "-busy"; + const unsigned int endlen = strlen(endswith); + struct engine *engine = + engine_ptr(engines, engines->num_engines); + char buf[256]; + + if (dent->d_type != DT_REG) + continue; + + if (strlen(dent->d_name) >= sizeof(buf)) { + ret = -1; break; } - print_clock("render", render_clock); - printf(" "); - print_clock("sampler", sampler_clock); - } else if (IS_945(devid) && IS_MOBILE(devid)) { - int render_clock = -1, display_clock = -1; + strcpy(buf, dent->d_name); - pci_device_cfg_read_u16(pci_dev, &gcfgc, I915_GCFGC); + /* xxxN-busy */ + if (strlen(buf) < (endlen + 4)) + continue; + if (strcmp(&buf[strlen(buf) - endlen], endswith)) + continue; - switch (gcfgc & 0x7) { - case 0: - render_clock = 166; - break; - case 1: - render_clock = 200; - break; - case 3: - render_clock = 250; - break; - case 5: - render_clock = 400; + memset(engine, 0, sizeof(*engine)); + + buf[strlen(buf) - endlen] = 0; + engine->name = strdup(buf); + if (!engine->name) { + ret = -1; break; } - switch (gcfgc & 0x70) { - case 0: - display_clock = 200; - break; - case 4: - display_clock = 320; + engine->busy.config = get_pmu_config(dirfd(d), engine->name, + "busy"); + if (engine->busy.config == -1) { + ret = -1; break; } - if (gcfgc & (1 << 7)) - display_clock = 133; - print_clock("render", render_clock); - printf(" "); - print_clock("display", display_clock); - } else if (IS_915(devid) && IS_MOBILE(devid)) { - int render_clock = -1, display_clock = -1; + engine->class = (engine->busy.config & + (__I915_PMU_OTHER(0) - 1)) >> + I915_PMU_CLASS_SHIFT; - pci_device_cfg_read_u16(pci_dev, &gcfgc, I915_GCFGC); + engine->instance = (engine->busy.config >> + I915_PMU_SAMPLE_BITS) & + ((1 << I915_PMU_SAMPLE_INSTANCE_BITS) - 1); - switch (gcfgc & 0x7) { - case 0: - render_clock = 160; - break; - case 1: - render_clock = 190; - break; - case 4: - render_clock = 333; + ret = snprintf(buf, sizeof(buf), "%s/%u", + class_display_name(engine->class), + engine->instance); + if (ret < 0 || ret == sizeof(buf)) { + ret = -1; break; } - if (gcfgc & (1 << 13)) - render_clock = 133; + ret = 0; - switch (gcfgc & 0x70) { - case 0: - display_clock = 190; + engine->display_name = strdup(buf); + if (!engine->display_name) { + ret = -1; break; - case 4: - display_clock = 333; + } + + engines->num_engines++; + engines = realloc(engines, sizeof(struct engines) + + engines->num_engines * sizeof(struct engine)); + if (!engines) { + ret = -ENOMEM; break; } - if (gcfgc & (1 << 7)) - display_clock = 133; + } - print_clock("render", render_clock); - printf(" "); - print_clock("display", display_clock); + if (ret) + free(engines); + else { + qsort(engine_ptr(engines, 0), engines->num_engines, + sizeof(struct engine), engine_cmp); + + engines->root = d; } + return ret == 0 ? engines : NULL; +} + +static int +filename_to_buf(const char *filename, char *buf, unsigned int bufsize) +{ + int fd; + ssize_t ret; + + fd = open(filename, O_RDONLY); + if (fd < 0) + return -1; + + ret = read(fd, buf, bufsize - 1); + close(fd); + if (ret < 1) + return -1; + + buf[ret] = '\0'; - printf("\n"); - return -1; + return 0; } -#define STATS_LEN (20) -#define PERCENTAGE_BAR_END (79 - STATS_LEN) +static uint64_t filename_to_u64(const char *filename, int base) +{ + char buf[64], *b; -static void -print_percentage_bar(float percent, int cur_line_len) + if (filename_to_buf(filename, buf, sizeof(buf))) + return 0; + + /* + * Handle both single integer and key=value formats by skipping + * leading non-digits. + */ + b = buf; + while (*b && !isdigit(*b)) + b++; + + return strtoull(b, NULL, base); +} + +static double filename_to_double(const char *filename) { - int bar_avail_len = (PERCENTAGE_BAR_END - cur_line_len - 1) * 8; - int bar_len = bar_avail_len * (percent + .5) / 100.0; - int i; + char *oldlocale; + char buf[80]; + double v; - for (i = bar_len; i >= 8; i -= 8) { - printf("%s", bars[8]); - cur_line_len++; - } - if (i) { - printf("%s", bars[i]); - cur_line_len++; - } + if (filename_to_buf(filename, buf, sizeof(buf))) + return 0; + + oldlocale = setlocale(LC_ALL, "C"); + v = strtod(buf, NULL); + setlocale(LC_ALL, oldlocale); - /* NB: We can't use a field width with utf8 so we manually - * guarantee a field with of 45 chars for any bar. */ - printf("%*s", PERCENTAGE_BAR_END - cur_line_len, ""); + return v; } -struct ring { - const char *name; - uint32_t mmio; - int head, tail, size; - uint64_t full; - int idle; -}; +static uint64_t rapl_type_id(void) +{ + return filename_to_u64("/sys/devices/power/type", 10); +} -static uint32_t ring_read(struct ring *ring, uint32_t reg) +static uint64_t rapl_gpu_power(void) { - return INREG(ring->mmio + reg); + return filename_to_u64("/sys/devices/power/events/energy-gpu", 0); } -static void ring_init(struct ring *ring) +static double rapl_gpu_power_scale(void) { - ring->size = (((ring_read(ring, RING_LEN) & RING_NR_PAGES) >> 12) + 1) * 4096; + return filename_to_double("/sys/devices/power/events/energy-gpu.scale"); } -static void ring_reset(struct ring *ring) +static uint64_t imc_type_id(void) { - ring->idle = ring->full = 0; + return filename_to_u64("/sys/devices/uncore_imc/type", 10); } -static void ring_sample(struct ring *ring) +static uint64_t imc_data_reads(void) { - int full; + return filename_to_u64("/sys/devices/uncore_imc/events/data_reads", 0); +} - if (!ring->size) - return; +static double imc_data_reads_scale(void) +{ + return filename_to_double("/sys/devices/uncore_imc/events/data_reads.scale"); +} - ring->head = ring_read(ring, RING_HEAD) & HEAD_ADDR; - ring->tail = ring_read(ring, RING_TAIL) & TAIL_ADDR; +static const char *imc_data_reads_unit(void) +{ + char buf[32]; - if (ring->tail == ring->head) - ring->idle++; + if (filename_to_buf("/sys/devices/uncore_imc/events/data_reads.unit", + buf, sizeof(buf)) == 0) + return strdup(buf); + else + return NULL; +} - full = ring->tail - ring->head; - if (full < 0) - full += ring->size; - ring->full += full; +static uint64_t imc_data_writes(void) +{ + return filename_to_u64("/sys/devices/uncore_imc/events/data_writes", 0); } -static void ring_print_header(FILE *out, struct ring *ring) +static double imc_data_writes_scale(void) { - fprintf(out, "%.6s%%\tops\t", - ring->name - ); + return filename_to_double("/sys/devices/uncore_imc/events/data_writes.scale"); } -static void ring_print(struct ring *ring, unsigned long samples_per_sec) +static const char *imc_data_writes_unit(void) { - int percent_busy, len; + char buf[32]; + + if (filename_to_buf("/sys/devices/uncore_imc/events/data_writes.unit", + buf, sizeof(buf)) == 0) + return strdup(buf); + else + return NULL; +} + +#define _open_pmu(cnt, pmu, fd) \ +({ \ + int fd__; \ +\ + fd__ = perf_i915_open_group((pmu)->config, (fd)); \ + if (fd__ >= 0) { \ + if ((fd) == -1) \ + (fd) = fd__; \ + (pmu)->present = true; \ + (pmu)->idx = (cnt)++; \ + } \ +\ + fd__; \ +}) + +#define _open_imc(cnt, pmu, fd) \ +({ \ + int fd__; \ +\ + fd__ = igt_perf_open_group(imc_type_id(), (pmu)->config, (fd)); \ + if (fd__ >= 0) { \ + if ((fd) == -1) \ + (fd) = fd__; \ + (pmu)->present = true; \ + (pmu)->idx = (cnt)++; \ + } \ +\ + fd__; \ +}) + +static int pmu_init(struct engines *engines) +{ + unsigned int i; + int fd; + + engines->fd = -1; + engines->num_counters = 0; + + engines->irq.config = I915_PMU_INTERRUPTS; + fd = _open_pmu(engines->num_counters, &engines->irq, engines->fd); + if (fd < 0) + return -1; - if (!ring->size) + engines->freq_req.config = I915_PMU_REQUESTED_FREQUENCY; + _open_pmu(engines->num_counters, &engines->freq_req, engines->fd); + + engines->freq_act.config = I915_PMU_ACTUAL_FREQUENCY; + _open_pmu(engines->num_counters, &engines->freq_act, engines->fd); + + engines->rc6.config = I915_PMU_RC6_RESIDENCY; + _open_pmu(engines->num_counters, &engines->rc6, engines->fd); + + for (i = 0; i < engines->num_engines; i++) { + struct engine *engine = engine_ptr(engines, i); + struct { + struct pmu_counter *pmu; + const char *counter; + bool optional; + } *cnt, counters[] = { + { .pmu = &engine->busy, .counter = "busy" }, + { .pmu = &engine->wait, .counter = "wait" }, + { .pmu = &engine->sema, .counter = "sema", .optional = true }, + { .pmu = NULL, .counter = NULL }, + }; + + for (cnt = counters; cnt->pmu; cnt++) { + if (!cnt->pmu->config) + cnt->pmu->config = + get_pmu_config(dirfd(engines->root), + engine->name, + cnt->counter); + fd = _open_pmu(engines->num_counters, cnt->pmu, + engines->fd); + if (fd < 0 && !cnt->optional) + return -1; + } + } + + engines->rapl_fd = -1; + if (rapl_type_id()) { + engines->rapl_scale = rapl_gpu_power_scale(); + if (engines->rapl_scale != NAN) + engines->rapl_scale *= 1e3; /* from nano to micro */ + + engines->rapl.config = rapl_gpu_power(); + if (!engines->rapl.config) + return -1; + + engines->rapl_fd = igt_perf_open(rapl_type_id(), + engines->rapl.config); + if (engines->rapl_fd < 0) + return -1; + + engines->rapl.present = true; + } + + engines->imc_fd = -1; + if (imc_type_id()) { + unsigned int num = 0; + + engines->imc_reads_scale = imc_data_reads_scale(); + engines->imc_writes_scale = imc_data_writes_scale(); + + engines->imc_reads_unit = imc_data_reads_unit(); + if (!engines->imc_reads_unit) + return -1; + + engines->imc_writes_unit = imc_data_writes_unit(); + if (!engines->imc_writes_unit) + return -1; + + engines->imc_reads.config = imc_data_reads(); + if (!engines->imc_reads.config) + return -1; + + engines->imc_writes.config = imc_data_writes(); + if (!engines->imc_writes.config) + return -1; + + fd = _open_imc(num, &engines->imc_reads, engines->imc_fd); + if (fd < 0) + return -1; + fd = _open_imc(num, &engines->imc_writes, engines->imc_fd); + if (fd < 0) + return -1; + + engines->imc_reads.present = true; + engines->imc_writes.present = true; + } + + return 0; +} + +static uint64_t pmu_read_multi(int fd, unsigned int num, uint64_t *val) +{ + uint64_t buf[2 + num]; + unsigned int i; + ssize_t len; + + memset(buf, 0, sizeof(buf)); + + len = read(fd, buf, sizeof(buf)); + assert(len == sizeof(buf)); + + for (i = 0; i < num; i++) + val[i] = buf[2 + i]; + + return buf[1]; +} + +static double __pmu_calc(struct pmu_pair *p, double d, double t, double s) +{ + double v; + + v = p->cur - p->prev; + v /= d; + v /= t; + v *= s; + + if (s == 100.0 && v > 100.0) + v = 100.0; + + return v; +} + +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wformat-nonliteral" +static void pmu_calc(struct pmu_counter *cnt, + char *buf, unsigned int bufsz, const char *fmt, + double d, double t, double s) +{ + double val; + int len; + + if (!cnt->present) { + strncpy(buf, "---", bufsz); return; + } - percent_busy = 100 - 100 * ring->idle / samples_per_sec; + val = __pmu_calc(&cnt->val, d, t, s); - len = printf("%25s busy: %3d%%: ", ring->name, percent_busy); - print_percentage_bar (percent_busy, len); - printf("%24s space: %d/%d\n", - ring->name, - (int)(ring->full / samples_per_sec), - ring->size); + len = snprintf(buf, bufsz, fmt, val); + if (len < 0 || len == bufsz) { + strncpy(buf, "XXX", bufsz); + return; + } } +#pragma GCC diagnostic pop -static void ring_log(struct ring *ring, unsigned long samples_per_sec, - FILE *output) +static uint64_t __pmu_read_single(int fd, uint64_t *ts) { - if (ring->size) - fprintf(output, "%3d\t%d\t", - (int)(100 - 100 * ring->idle / samples_per_sec), - (int)(ring->full / samples_per_sec)); - else - fprintf(output, "-1\t-1\t"); + uint64_t data[2] = { }; + ssize_t len; + + len = read(fd, data, sizeof(data)); + assert(len == sizeof(data)); + + if (ts) + *ts = data[1]; + + return data[0]; } +static uint64_t pmu_read_single(int fd) +{ + return __pmu_read_single(fd, NULL); +} + +static void __update_sample(struct pmu_counter *counter, uint64_t val) +{ + counter->val.prev = counter->val.cur; + counter->val.cur = val; +} + +static void update_sample(struct pmu_counter *counter, uint64_t *val) +{ + if (counter->present) + __update_sample(counter, val[counter->idx]); +} + +static void pmu_sample(struct engines *engines) +{ + const int num_val = engines->num_counters; + uint64_t val[2 + num_val]; + unsigned int i; + + engines->ts.prev = engines->ts.cur; + + if (engines->rapl_fd >= 0) + __update_sample(&engines->rapl, + pmu_read_single(engines->rapl_fd)); + + if (engines->imc_fd >= 0) { + pmu_read_multi(engines->imc_fd, 2, val); + update_sample(&engines->imc_reads, val); + update_sample(&engines->imc_writes, val); + } + + engines->ts.cur = pmu_read_multi(engines->fd, num_val, val); + + update_sample(&engines->freq_req, val); + update_sample(&engines->freq_act, val); + update_sample(&engines->irq, val); + update_sample(&engines->rc6, val); + + for (i = 0; i < engines->num_engines; i++) { + struct engine *engine = engine_ptr(engines, i); + + update_sample(&engine->busy, val); + update_sample(&engine->sema, val); + update_sample(&engine->wait, val); + } +} + +static const char *bars[] = { " ", "▏", "▎", "▍", "▌", "▋", "▊", "▉", "█" }; + +static void +print_percentage_bar(double percent, int max_len) +{ + int bar_len = percent * (8 * (max_len - 2)) / 100.0; + int i; + + putchar('|'); + + for (i = bar_len; i >= 8; i -= 8) + printf("%s", bars[8]); + if (i) + printf("%s", bars[i]); + + for (i = 0; i < (max_len - 2 - (bar_len + 7) / 8); i++) + putchar(' '); + + putchar('|'); +} + +#define DEFAULT_PERIOD_MS (1000) + static void usage(const char *appname) { printf("intel_gpu_top - Display a top-like summary of Intel GPU usage\n" - "\n" - "usage: %s [parameters]\n" - "\n" - "The following parameters apply:\n" - "[-s ] samples per seconds (default %d)\n" - "[-e ] command to profile\n" - "[-o ] output statistics to file. If file is '-'," - " run in batch mode and output statistics to stdio only \n" - "[-h] show this help screen\n" - "\n", - appname, - SAMPLES_PER_SEC - ); - return; + "\n" + "Usage: %s [parameters]\n" + "\n" + "\tThe following parameters are optional:\n" + "\t[-s ] refresh period in ms (default %ums)\n" + "\t[-h] show this help text\n" + "\n", + appname, DEFAULT_PERIOD_MS); } int main(int argc, char **argv) { - uint32_t devid; - struct pci_device *pci_dev; - struct ring render_ring = { - .name = "render", - .mmio = 0x2030, - }, bsd_ring = { - .name = "bitstream", - .mmio = 0x4030, - }, bsd6_ring = { - .name = "bitstream", - .mmio = 0x12030, - }, blt_ring = { - .name = "blitter", - .mmio = 0x22030, - }; - int i, ch; - int samples_per_sec = SAMPLES_PER_SEC; - FILE *output = NULL; - double elapsed_time=0; - int print_headers=1; - pid_t child_pid=-1; - int child_stat; - char *cmd=NULL; - int interactive=1; - - /* Parse options? */ - while ((ch = getopt(argc, argv, "s:o:e:h")) != -1) { + unsigned int period_us = DEFAULT_PERIOD_MS * 1000; + int con_w = -1, con_h = -1; + struct engines *engines; + unsigned int i; + int ret, ch; + + /* Parse options */ + while ((ch = getopt(argc, argv, "s:h")) != -1) { switch (ch) { - case 'e': cmd = strdup(optarg); - break; - case 's': samples_per_sec = atoi(optarg); - if (samples_per_sec < 100) { - fprintf(stderr, "Error: samples per second must be >= 100\n"); - exit(1); - } - break; - case 'o': - if (!strcmp(optarg, "-")) { - /* Running in non-interactive mode */ - interactive = 0; - output = stdout; - } - else - output = fopen(optarg, "w"); - if (!output) - { - perror("fopen"); - exit(1); - } + case 's': + period_us = atoi(optarg) * 1000; break; case 'h': usage(argv[0]); exit(0); - break; default: - fprintf(stderr, "Invalid flag %c!\n", (char)optopt); + fprintf(stderr, "Invalid option %c!\n", (char)optopt); usage(argv[0]); exit(1); - break; } } - pci_dev = intel_get_pci_device(); - devid = pci_dev->device_id; - intel_mmio_use_pci_bar(pci_dev); - init_instdone_definitions(devid); - - /* Do we have a command to run? */ - if (cmd != NULL) { - if (output) { - fprintf(output, "# Profiling: %s\n", cmd); - fflush(output); - } - child_pid = fork(); - if (child_pid < 0) { - perror("fork"); - exit(1); - } - else if (child_pid == 0) { - int res; - res = system(cmd); - if (res < 0) - perror("running command"); - if (output) { - fflush(output); - fprintf(output, "# %s exited with status %d\n", cmd, res); - fflush(output); - } - free(cmd); - exit(0); - } else { - free(cmd); - } - } - - for (i = 0; i < num_instdone_bits; i++) { - top_bits[i].bit = &instdone_bits[i]; - top_bits[i].count = 0; - top_bits_sorted[i] = &top_bits[i]; + engines = discover_engines(); + if (!engines) { + fprintf(stderr, + "Failed to detect engines! Kernel 4.16 or newer?\n"); + return 1; } - /* Grab access to the registers */ - intel_register_access_init(pci_dev, 0, -1); - - ring_init(&render_ring); - if (IS_GEN4(devid) || IS_GEN5(devid)) - ring_init(&bsd_ring); - if (IS_GEN6(devid) || IS_GEN7(devid)) { - ring_init(&bsd6_ring); - ring_init(&blt_ring); + ret = pmu_init(engines); + if (ret) { + fprintf(stderr, + "Failed to initialize PMU! Kernel 4.16 or newer?\n"); + return 1; } - /* Initialize GPU stats */ - if (HAS_STATS_REGS(devid)) { - for (i = 0; i < STATS_COUNT; i++) { - uint32_t stats_high, stats_low, stats_high_2; - - do { - stats_high = INREG(stats_regs[i] + 4); - stats_low = INREG(stats_regs[i]); - stats_high_2 = INREG(stats_regs[i] + 4); - } while (stats_high != stats_high_2); - - last_stats[i] = (uint64_t)stats_high << 32 | - stats_low; - } - } + pmu_sample(engines); for (;;) { - int j; - unsigned long long t1, ti, tf, t2; - unsigned long long def_sleep = 1000000 / samples_per_sec; - unsigned long long last_samples_per_sec = samples_per_sec; - unsigned short int max_lines; + double t; +#define BUFSZ 16 + char freq[BUFSZ]; + char fact[BUFSZ]; + char irq[BUFSZ]; + char rc6[BUFSZ]; + char power[BUFSZ]; + char reads[BUFSZ]; + char writes[BUFSZ]; struct winsize ws; - char clear_screen[] = {0x1b, '[', 'H', - 0x1b, '[', 'J', - 0x0}; - int percent; - int len; - - t1 = gettime(); - - ring_reset(&render_ring); - ring_reset(&bsd_ring); - ring_reset(&bsd6_ring); - ring_reset(&blt_ring); - - for (i = 0; i < samples_per_sec; i++) { - long long interval; - ti = gettime(); - if (IS_965(devid)) { - instdone = INREG(INSTDONE_I965); - instdone1 = INREG(INSTDONE_1); - } else - instdone = INREG(INSTDONE); - - for (j = 0; j < num_instdone_bits; j++) - update_idle_bit(&top_bits[j]); - - ring_sample(&render_ring); - ring_sample(&bsd_ring); - ring_sample(&bsd6_ring); - ring_sample(&blt_ring); - - tf = gettime(); - if (tf - t1 >= 1000000) { - /* We are out of sync, bail out */ - last_samples_per_sec = i+1; - break; - } - interval = def_sleep - (tf - ti); - if (interval > 0) - usleep(interval); - } - - if (HAS_STATS_REGS(devid)) { - for (i = 0; i < STATS_COUNT; i++) { - uint32_t stats_high, stats_low, stats_high_2; + int lines = 0; - do { - stats_high = INREG(stats_regs[i] + 4); - stats_low = INREG(stats_regs[i]); - stats_high_2 = INREG(stats_regs[i] + 4); - } while (stats_high != stats_high_2); - - stats[i] = (uint64_t)stats_high << 32 | - stats_low; - } + /* Update terminal size. */ + if (ioctl(0, TIOCGWINSZ, &ws) != -1) { + con_w = ws.ws_col; + con_h = ws.ws_row; } - qsort(top_bits_sorted, num_instdone_bits, - sizeof(struct top_bit *), top_bits_sort); - - /* Limit the number of lines printed to the terminal height so the - * most important info (at the top) will stay on screen. */ - max_lines = -1; - if (ioctl(0, TIOCGWINSZ, &ws) != -1) - max_lines = ws.ws_row - 6; /* exclude header lines */ - if (max_lines >= num_instdone_bits) - max_lines = num_instdone_bits; - - t2 = gettime(); - elapsed_time += (t2 - t1) / 1000000.0; - - if (interactive) { - printf("%s", clear_screen); - print_clock_info(pci_dev); - - ring_print(&render_ring, last_samples_per_sec); - ring_print(&bsd_ring, last_samples_per_sec); - ring_print(&bsd6_ring, last_samples_per_sec); - ring_print(&blt_ring, last_samples_per_sec); - - printf("\n%30s %s\n", "task", "percent busy"); - for (i = 0; i < max_lines; i++) { - if (top_bits_sorted[i]->count > 0) { - percent = (top_bits_sorted[i]->count * 100) / - last_samples_per_sec; - len = printf("%30s: %3d%%: ", - top_bits_sorted[i]->bit->name, - percent); - print_percentage_bar (percent, len); - } else { - printf("%*s", PERCENTAGE_BAR_END, ""); - } - - if (i < STATS_COUNT && HAS_STATS_REGS(devid)) { - printf("%13s: %llu (%lld/sec)", - stats_reg_names[i], - (long long)stats[i], - (long long)(stats[i] - last_stats[i])); - last_stats[i] = stats[i]; - } else { - if (!top_bits_sorted[i]->count) - break; - } - printf("\n"); - } - } - if (output) { - /* Print headers for columns at first run */ - if (print_headers) { - fprintf(output, "# time\t"); - ring_print_header(output, &render_ring); - ring_print_header(output, &bsd_ring); - ring_print_header(output, &bsd6_ring); - ring_print_header(output, &blt_ring); - for (i = 0; i < MAX_NUM_TOP_BITS; i++) { - if (i < STATS_COUNT && HAS_STATS_REGS(devid)) { - fprintf(output, "%.6s\t", - stats_reg_names[i] - ); - } - if (!top_bits[i].count) - continue; - } - fprintf(output, "\n"); - print_headers = 0; - } - - /* Print statistics */ - fprintf(output, "%.2f\t", elapsed_time); - ring_log(&render_ring, last_samples_per_sec, output); - ring_log(&bsd_ring, last_samples_per_sec, output); - ring_log(&bsd6_ring, last_samples_per_sec, output); - ring_log(&blt_ring, last_samples_per_sec, output); - - for (i = 0; i < MAX_NUM_TOP_BITS; i++) { - if (i < STATS_COUNT && HAS_STATS_REGS(devid)) { - fprintf(output, "%"PRIu64"\t", - stats[i] - last_stats[i]); - last_stats[i] = stats[i]; - } - if (!top_bits[i].count) - continue; - } - fprintf(output, "\n"); - fflush(output); + pmu_sample(engines); + t = (double)(engines->ts.cur - engines->ts.prev) / 1e9; + + printf("\033[H\033[J"); + + pmu_calc(&engines->freq_req, freq, BUFSZ, "%4.0f", 1.0, t, 1); + pmu_calc(&engines->freq_act, fact, BUFSZ, "%4.0f", 1.0, t, 1); + pmu_calc(&engines->irq, irq, BUFSZ, "%8.0f", 1.0, t, 1); + pmu_calc(&engines->rc6, rc6, BUFSZ, "%3.0f", 1e9, t, 100); + pmu_calc(&engines->rapl, power, BUFSZ, "%6.0f", 1.0, t, + engines->rapl_scale); + pmu_calc(&engines->imc_reads, reads, BUFSZ, "%6.0f", 1.0, t, + engines->imc_reads_scale); + pmu_calc(&engines->imc_writes, writes, BUFSZ, "%6.0f", 1.0, t, + engines->imc_writes_scale); + + printf("intel-gpu-top - %s/%s MHz; %s%% RC6; %smW; %s irqs/s\n", + fact, freq, rc6, power, irq); + lines++; + + printf("\n"); + lines++; + + printf("%16s %s %s/s\n", + "IMC reads:", reads, engines->imc_reads_unit); + lines++; + + printf("%16s %s %s/s\n", + "IMC writes:", writes, engines->imc_writes_unit); + lines++; + + printf("\n"); + lines++; + + for (i = 0; i < engines->num_engines && lines < con_h; i++) { + struct engine *engine = engine_ptr(engines, i); + unsigned int max_w = con_w - 1; + unsigned int len; + char sema[BUFSZ]; + char wait[BUFSZ]; + char busy[BUFSZ]; + char buf[128]; + double val; + + pmu_calc(&engine->sema, sema, BUFSZ, "%3.0f", 1e9, t, + 100); + pmu_calc(&engine->wait, wait, BUFSZ, "%3.0f", 1e9, t, + 100); + len = snprintf(buf, sizeof(buf), "%s%% sema, %s%% wait", + sema, wait); + + pmu_calc(&engine->busy, busy, BUFSZ, "%6.2f", 1e9, t, + 100); + len += printf("%16s %s%% ", engine->display_name, busy); + + val = __pmu_calc(&engine->busy.val, 1e9, t, 100); + print_percentage_bar(val, max_w - len); + + printf("%s\n", buf); + + lines++; } - for (i = 0; i < num_instdone_bits; i++) { - top_bits_sorted[i]->count = 0; - - if (i < STATS_COUNT) - last_stats[i] = stats[i]; - } + printf("\n"); - /* Check if child has gone */ - if (child_pid > 0) { - int res; - if ((res = waitpid(child_pid, &child_stat, WNOHANG)) == -1) { - perror("waitpid"); - exit(1); - } - if (res == 0) - continue; - if (WIFEXITED(child_stat)) - break; - } + usleep(period_us); } - fclose(output); - - intel_register_access_fini(); return 0; } diff --git a/tools/meson.build b/tools/meson.build index bd2d313d5156..a918eeb0bef1 100644 --- a/tools/meson.build +++ b/tools/meson.build @@ -23,7 +23,6 @@ tools_progs = [ 'intel_gpu_frequency', 'intel_firmware_decode', 'intel_gpu_time', - 'intel_gpu_top', 'intel_gtt', 'intel_guc_logger', 'intel_infoframes', @@ -117,6 +116,11 @@ shared_library('intel_aubdump', 'aubdump.c', name_prefix : '', install : true) +executable('intel_gpu_top', 'intel_gpu_top.c', + install : true, + install_rpath : rpathdir, + dependencies : tool_deps + [ lib_igt_perf ]) + conf_data = configuration_data() conf_data.set('prefix', prefix) conf_data.set('exec_prefix', '${prefix}')