From owner-svn-src-all@FreeBSD.ORG Thu Jul 16 08:27:12 2009 Return-Path: Delivered-To: svn-src-all@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id BE96E106566C; Thu, 16 Jul 2009 08:27:12 +0000 (UTC) (envelope-from mav@FreeBSD.org) Received: from svn.freebsd.org (svn.freebsd.org [IPv6:2001:4f8:fff6::2c]) by mx1.freebsd.org (Postfix) with ESMTP id ABA7A8FC14; Thu, 16 Jul 2009 08:27:12 +0000 (UTC) (envelope-from mav@FreeBSD.org) Received: from svn.freebsd.org (localhost [127.0.0.1]) by svn.freebsd.org (8.14.3/8.14.3) with ESMTP id n6G8RCYW017164; Thu, 16 Jul 2009 08:27:12 GMT (envelope-from mav@svn.freebsd.org) Received: (from mav@localhost) by svn.freebsd.org (8.14.3/8.14.3/Submit) id n6G8RC05017163; Thu, 16 Jul 2009 08:27:12 GMT (envelope-from mav@svn.freebsd.org) Message-Id: <200907160827.n6G8RC05017163@svn.freebsd.org> From: Alexander Motin Date: Thu, 16 Jul 2009 08:27:12 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-7@freebsd.org X-SVN-Group: stable-7 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cc: Subject: svn commit: r195714 - in stable/7/sys: . contrib/pf dev/acpica X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 16 Jul 2009 08:27:13 -0000 Author: mav Date: Thu Jul 16 08:27:12 2009 New Revision: 195714 URL: http://svn.freebsd.org/changeset/base/195714 Log: MFC rev. 191760, 191763, 191764. Avoid comparing negative signed to positive unsignad values. It was leading to a bug, when C-state does not decrease on sleep shorter then declared transition latency. Fixing this deprecates workaround for broken C-states on some hardware. By the way, change state selecting logic a bit. Instead of last sleep time use short-time average of it. Global interrupts rate in system is a quite random value, to corellate subsequent sleeps so directly. Make dev.cpu.X.cx_usage sysctl also report current average of sleep time. Modified: stable/7/sys/ (props changed) stable/7/sys/contrib/pf/ (props changed) stable/7/sys/dev/acpica/acpi_cpu.c Modified: stable/7/sys/dev/acpica/acpi_cpu.c ============================================================================== --- stable/7/sys/dev/acpica/acpi_cpu.c Wed Jul 15 22:27:36 2009 (r195713) +++ stable/7/sys/dev/acpica/acpi_cpu.c Thu Jul 16 08:27:12 2009 (r195714) @@ -79,7 +79,6 @@ struct acpi_cpu_softc { int cpu_features; /* Child driver supported features. */ /* Runtime state. */ int cpu_non_c3; /* Index of lowest non-C3 state. */ - int cpu_short_slp; /* Count of < 1us sleeps. */ u_int cpu_cx_stats[MAX_CX_STATES];/* Cx usage history. */ /* Values for sysctl. */ struct sysctl_ctx_list cpu_sysctl_ctx; @@ -875,43 +874,13 @@ acpi_cpu_idle() return; } - /* - * If we slept 100 us or more, use the lowest Cx state. Otherwise, - * find the lowest state that has a latency less than or equal to - * the length of our last sleep. - */ - cx_next_idx = sc->cpu_cx_lowest; - if (sc->cpu_prev_sleep < 100) { - /* - * If we sleep too short all the time, this system may not implement - * C2/3 correctly (i.e. reads return immediately). In this case, - * back off and use the next higher level. - * It seems that when you have a dual core cpu (like the Intel Core Duo) - * that both cores will get out of C3 state as soon as one of them - * requires it. This breaks the sleep detection logic as the sleep - * counter is local to each cpu. Disable the sleep logic for now as a - * workaround if there's more than one CPU. The right fix would probably - * be to add quirks for system that don't really support C3 state. - */ - if (mp_ncpus < 2 && sc->cpu_prev_sleep <= 1) { - sc->cpu_short_slp++; - if (sc->cpu_short_slp == 1000 && sc->cpu_cx_lowest != 0) { - if (sc->cpu_non_c3 == sc->cpu_cx_lowest && sc->cpu_non_c3 != 0) - sc->cpu_non_c3--; - sc->cpu_cx_lowest--; - sc->cpu_short_slp = 0; - device_printf(sc->cpu_dev, - "too many short sleeps, backing off to C%d\n", - sc->cpu_cx_lowest + 1); - } - } else - sc->cpu_short_slp = 0; - - for (i = sc->cpu_cx_lowest; i >= 0; i--) - if (sc->cpu_cx_states[i].trans_lat <= sc->cpu_prev_sleep) { - cx_next_idx = i; - break; - } + /* Find the lowest state that has small enough latency. */ + cx_next_idx = 0; + for (i = sc->cpu_cx_lowest; i >= 0; i--) { + if (sc->cpu_cx_states[i].trans_lat * 3 <= sc->cpu_prev_sleep) { + cx_next_idx = i; + break; + } } /* @@ -936,10 +905,10 @@ acpi_cpu_idle() /* * Execute HLT (or equivalent) and wait for an interrupt. We can't * calculate the time spent in C1 since the place we wake up is an - * ISR. Assume we slept one quantum and return. + * ISR. Assume we slept half of quantum and return. */ if (cx_next->type == ACPI_STATE_C1) { - sc->cpu_prev_sleep = 1000000 / hz; + sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + 500000 / hz) / 4; acpi_cpu_c1(); return; } @@ -982,9 +951,9 @@ acpi_cpu_idle() } ACPI_ENABLE_IRQS(); - /* Find the actual time asleep in microseconds, minus overhead. */ + /* Find the actual time asleep in microseconds. */ end_time = acpi_TimerDelta(end_time, start_time); - sc->cpu_prev_sleep = PM_USEC(end_time) - cx_next->trans_lat; + sc->cpu_prev_sleep = (sc->cpu_prev_sleep * 3 + PM_USEC(end_time)) / 4; } /* @@ -1106,8 +1075,9 @@ acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARG sbuf_printf(&sb, "%u.%02u%% ", (u_int)(whole / sum), (u_int)(fract / sum)); } else - sbuf_printf(&sb, "0%% "); + sbuf_printf(&sb, "0.00%% "); } + sbuf_printf(&sb, "last %dus", sc->cpu_prev_sleep); sbuf_trim(&sb); sbuf_finish(&sb); sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);