linaro-dev May 2012

linaro-dev@lists.linaro.org

118 participants
90 discussions

Booting linux-linaro-2.6.37 on Beagle Board

by Avik Sil

Hi, I tried booting linux-linaro-2.6.37 kernel on my beagle board C4. I executed following: 1. Installed linaro on a 4 GB SD card using linaro-image-tools 0.4.1 with hwpack daily snapshot hwpack_linaro-omap3_20110125-0_armel_supported.tar.gz and linaro-natty-headless-tar-20101202-1.tar.gz. It was booting properly on my BB. 2. Cloned linux-linaro-2.6.37. Changed to source directory 3. make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- omap2plus_defconfig 4. make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- menuconfig (enabled EARLY_PRINTK) 5. make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- uImage 6. make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- modules 7. make ARCH=arm CROSS_COMPILE=arm-linux-gnueabi- modules_install INSTALL_MOD_PATH=/media/rootfs 8. cp arch/arm/boot/uImage /media/boot; sync Everything went on smoothly. Then I put the SD card on BB and powered it on. I got a kernel panic: http://paste.ubuntu.com/560562 Please help me figuring out the problem. Is it because I didn't create uInitrd? If so, then how to create it for ARM? Regards, Avik

11 years, 3 months

ARM A9 oprofile

by Dmitry Antipov

Hello Ming, could you please give some pointers to observe an overall status of oprofile support on ARM A9 cores? IIUC, now it doesn't work without oprofile.timer=1 kernel option, at least for Linus' tree; searching gives a lot of discussion/patches fragments and similar stuff, but I was unable to find a complete patch/git tree/whatever else to try. Thanks in advance, Dmitry

11 years, 6 months

[PATCH v4 0/5] thermal: exynos: Add kernel thermal support for exynos platform

by Amit Daniel Kachhap

Hi Andrew/Rui, As discussed with Rui Zhang, I dropped the patch for new trip type THERMAL_TRIP_STATE_INSTANCE and made the necessary state magnagement changes in cpufreq cooling functions. Also I fixed all the review comments suggested by Andrew. If any other changes please let me know. This patchset introduces a new generic cooling device based on cpufreq that can be used on non-ACPI platforms. As a proof of concept, we have drivers for the following platforms using this mechanism now: * Samsung Exynos (Exynos4 and Exynos5) in the current patchset. * TI OMAP (git://git.linaro.org/people/amitdanielk/linux.git omap4460_thermal) * Freescale i.MX (git://git.linaro.org/people/amitdanielk/linux.git imx6q_thermal) The is a small change in cpufreq cooling registration APIs, so a minor change is needed for OMAP and Freescale platforms. Thanks, Amit Daniel Changes since V3: * Dropped the concept of using new trip type THERMAL_TRIP_STATE_INSTANCE as discussed with Rui Zhang. This requires adding some state management logic in cpufreq cooling implementation. * Many review comments suggested by Andrew Morton * More documentation added in cpufreq cooling codes. Changes since V2: *Added Exynos5 TMU sensor support by enhancing the exynos4 tmu driver. Exynos5 TMU driver was internally developed by SangWook Ju <sw.ju(a)samsung.com>. *Removed cpuhotplug cooling code in this patchset. *Rebased the patches against 3.4-rc6 kernel. Changes since V1: *Moved the sensor driver to driver/thermal folder from driver/hwmon folder as suggested by Mark Brown and Guenter Roeck *Added notifier support to notify the registered drivers of any cpu cooling action. The driver can modify the default cooling behaviour(eg set different max clip frequency). *The percentage based frequency replaced with absolute clipped frequency. *Some more conditional checks when setting max frequency. *Renamed the new trip type THERMAL_TRIP_STATE_ACTIVE to THERMAL_TRIP_STATE_INSTANCE *Many review comments from R, Durgadoss <durgadoss.r(a)intel.com> and eduardo.valentin(a)ti.com implemented. *Removed cooling stats through debugfs patch *The V1 based can be found here, https://lkml.org/lkml/2012/2/22/123 http://lkml.org/lkml/2012/3/3/32 Changes since RFC: *Changed the cpu cooling registration/unregistration API's to instance based *Changed the STATE_ACTIVE trip type to pass correct instance id *Adding support to restore back the policy->max_freq after doing frequency clipping. *Moved the trip cooling stats from sysfs node to debugfs node as suggested by Greg KH greg(a)kroah.com *Incorporated several review comments from eduardo.valentin(a)ti.com *Moved the Temperature sensor driver from driver/hwmon/ to driver/mfd as discussed with Guenter Roeck <guenter.roeck(a)ericsson.com> and Donggeun Kim <dg77.kim(a)samsung.com> (https://lkml.org/lkml/2012/1/5/7) *Some changes according to the changes in common cpu cooling APIs *The RFC based patches can be found here, https://lkml.org/lkml/2011/12/13/186 https://lkml.org/lkml/2011/12/21/169 Brief Description: 1) The generic cooling devices code is placed inside driver/thermal/* as placing inside acpi folder will need un-necessary enabling of acpi code. This codes is architecture independent. 2) This patchset adds generic cpu cooling low level implementation through frequency clipping. In future, other cpu related cooling devices may be added here. An ACPI version of this already exists (drivers/acpi/processor_thermal.c) . But this will be useful for platforms like ARM using the generic thermal interface along with the generic cpu cooling devices. The cooling device registration API's return cooling device pointers which can be easily binded with the thermal zone trip points. The important APIs exposed are, a)struct thermal_cooling_device *cpufreq_cooling_register( struct freq_clip_table *tab_ptr, unsigned int tab_size) b)void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev) 3) Samsung exynos platform thermal implementation is done using the generic cpu cooling APIs and the new trip type. The temperature sensor driver present in the hwmon folder(registered as hwmon driver) is moved to thermal folder and registered as a thermal driver. All this patchset is based on Kernel version 3.4-rc6 A simple data/control flow diagrams is shown below, Core Linux thermal <-----> Exynos thermal interface <----- Temperature Sensor | | \|/ | Cpufreq cooling device <--------------- TODO: *Will send the DT enablement patches later after the driver is merged. Amit Daniel Kachhap (5): thermal: Add generic cpufreq cooling implementation hwmon: exynos4: Move thermal sensor driver to driver/thermal directory thermal: exynos5: Add exynos5 thermal sensor driver support thermal: exynos: Register the tmu sensor with the kernel thermal layer ARM: exynos: Add thermal sensor driver platform data support Documentation/hwmon/exynos4_tmu | 81 --- Documentation/thermal/cpu-cooling-api.txt | 60 ++ Documentation/thermal/exynos_thermal | 52 ++ drivers/hwmon/Kconfig | 10 - drivers/hwmon/Makefile | 1 - drivers/hwmon/exynos4_tmu.c | 514 -------------- drivers/thermal/Kconfig | 20 + drivers/thermal/Makefile | 4 +- drivers/thermal/cpu_cooling.c | 483 +++++++++++++ drivers/thermal/exynos_thermal.c | 956 ++++++++++++++++++++++++++ include/linux/cpu_cooling.h | 99 +++ include/linux/platform_data/exynos4_tmu.h | 83 --- include/linux/platform_data/exynos_thermal.h | 100 +++ 13 files changed, 1773 insertions(+), 690 deletions(-) delete mode 100644 Documentation/hwmon/exynos4_tmu create mode 100644 Documentation/thermal/cpu-cooling-api.txt create mode 100644 Documentation/thermal/exynos_thermal delete mode 100644 drivers/hwmon/exynos4_tmu.c create mode 100644 drivers/thermal/cpu_cooling.c create mode 100644 drivers/thermal/exynos_thermal.c create mode 100644 include/linux/cpu_cooling.h delete mode 100644 include/linux/platform_data/exynos4_tmu.h create mode 100644 include/linux/platform_data/exynos_thermal.h

11 years, 8 months

[PATCH 1/2] mfd: Add Freescale's PMIC MC34708 support

by Ying-Chun Liu (PaulLiu)

From: "Ying-Chun Liu (PaulLiu)" <paul.liu(a)linaro.org> Freescale MC34708 is a PMIC which supports the following features: * 6 multi-mode buck regulators * Boost regulator for USB OTG. * 8 regulators for thermal budget optimization * 10-bit ADC * Real time clock Signed-off-by: Robin Gong <B38343(a)freescale.com> Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu(a)linaro.org> Cc: Samuel Ortiz <sameo(a)linux.intel.com> Cc: Mark Brown <broonie(a)opensource.wolfsonmicro.com> Cc: Shawn Guo <shawn.guo(a)linaro.org> --- Documentation/devicetree/bindings/mfd/mc34708.txt | 61 ++ drivers/mfd/Kconfig | 11 + drivers/mfd/Makefile | 1 + drivers/mfd/mc34708-core.c | 634 +++++++++++++++++++++ include/linux/mfd/mc34708.h | 134 +++++ 5 files changed, 841 insertions(+) create mode 100644 Documentation/devicetree/bindings/mfd/mc34708.txt create mode 100644 drivers/mfd/mc34708-core.c create mode 100644 include/linux/mfd/mc34708.h diff --git a/Documentation/devicetree/bindings/mfd/mc34708.txt b/Documentation/devicetree/bindings/mfd/mc34708.txt new file mode 100644 index 0000000..2bb5c9e --- /dev/null +++ b/Documentation/devicetree/bindings/mfd/mc34708.txt @@ -0,0 +1,61 @@ +* Freescale MC34708 Power Management Integrated Circuit (PMIC) + +Required properties: +- compatible : Must be "fsl,mc34708" + +Optional properties: +- fsl,mc34708-uses-adc : Indicate the ADC is being used +- fsl,mc34708-uses-rtc : Indicate the RTC is being used +- fsl,mc34708-uses-ts : Indicate the touchscreen controller is being used + +Sub-nodes: +- regulators : Contain the regulator nodes. The MC34708 regulators are + bound using their names as listed below for enabling. + + mc34708__sw1a : regulator SW1A + mc34708__sw1b : regulator SW1B + mc34708__sw2 : regulator SW2 + mc34708__sw3 : regulator SW3 + mc34708__sw4A : regulator SW4A + mc34708__sw4b : regulator SW4B + mc34708__swbst : regulator SWBST + mc34708__vpll : regulator VPLL + mc34708__vrefddr : regulator VREFDDR + mc34708__vusb : regulator VUSB + mc34708__vusb2 : regulator VUSB2 + mc34708__vdac : regulator VDAC + mc34708__vgen1 : regulator VGEN1 + mc34708__vgen2 : regulator VGEN2 + + The bindings details of individual regulator device can be found in: + Documentation/devicetree/bindings/regulator/regulator.txt + +Examples: + +i2c@63fc8000 { + #address-cells = <1>; + #size-cells = <0>; + compatible = "fsl,imx53-i2c", "fsl,imx1-i2c"; + reg = <0x63fc8000 0x4000>; + interrupts = <62>; + status = "okay"; + + pmic: mc34708@8 { + compatible = "fsl,mc34708"; + reg = <0x08>; + + regulators { + mc34708__sw1a { + regulator-min-microvolt = <650000>; + regulator-max-microvolt = <1437500>; + regulator-boot-on; + regulator-always-on; + }; + + mc34708__vusb { + regulator-boot-on; + regulator-always-on; + }; + }; + }; +}; diff --git a/drivers/mfd/Kconfig b/drivers/mfd/Kconfig index 29f463c..17b6cc5 100644 --- a/drivers/mfd/Kconfig +++ b/drivers/mfd/Kconfig @@ -600,6 +600,17 @@ config MFD_MC13XXX additional drivers must be enabled in order to use the functionality of the device. +config MFD_MC34708 + tristate "Support for Freescale's PMIC MC34708" + depends on I2C + depends on OF + select MFD_CORE + help + Support for the Freescale's PMIC MC34708 + This driver provides common support for accessing the device, + additional drivers must be enabled in order to use the + functionality of the device. + config ABX500_CORE bool "ST-Ericsson ABX500 Mixed Signal Circuit register functions" default y if ARCH_U300 || ARCH_U8500 diff --git a/drivers/mfd/Makefile b/drivers/mfd/Makefile index 05fa538..b98d943 100644 --- a/drivers/mfd/Makefile +++ b/drivers/mfd/Makefile @@ -54,6 +54,7 @@ obj-$(CONFIG_TWL6030_PWM) += twl6030-pwm.o obj-$(CONFIG_TWL6040_CORE) += twl6040-core.o twl6040-irq.o obj-$(CONFIG_MFD_MC13XXX) += mc13xxx-core.o +obj-$(CONFIG_MFD_MC34708) += mc34708-core.o obj-$(CONFIG_MFD_CORE) += mfd-core.o diff --git a/drivers/mfd/mc34708-core.c b/drivers/mfd/mc34708-core.c new file mode 100644 index 0000000..54f469b --- /dev/null +++ b/drivers/mfd/mc34708-core.c @@ -0,0 +1,634 @@ +/* + * Copyright (C) 2011 Freescale Semiconductor, Inc. All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + */ + +#include <linux/slab.h> +#include <linux/module.h> +#include <linux/platform_device.h> +#include <linux/mutex.h> +#include <linux/interrupt.h> +#include <linux/i2c.h> +#include <linux/mfd/core.h> +#include <linux/mfd/mc34708.h> +#include <linux/delay.h> +#include <linux/of.h> +#include <linux/of_device.h> + +struct mc34708 { + struct i2c_client *i2c_client; + struct mutex lock; + int irq; + + irq_handler_t irqhandler[MC34708_NUM_IRQ]; + void *irqdata[MC34708_NUM_IRQ]; +}; + +/*! + * This is the enumeration of versions of PMIC + */ +enum mc34708_id { + MC_PMIC_ID_MC34708, + MC_PMIC_ID_INVALID, +}; + +struct mc34708_version_t { + /*! + * PMIC version identifier. + */ + enum mc34708_id id; + /*! + * Revision of the PMIC. + */ + int revision; +}; + +#define PMIC_I2C_RETRY_TIMES 10 + +static const struct i2c_device_id mc34708_device_id[] = { + {"mc34708", MC_PMIC_ID_MC34708}, + {}, +}; + +static const char * const mc34708_chipname[] = { + [MC_PMIC_ID_MC34708] = "mc34708", +}; + +void mc34708_lock(struct mc34708 *mc_pmic) +{ + if (!mutex_trylock(&mc_pmic->lock)) { + dev_dbg(&mc_pmic->i2c_client->dev, "wait for %s from %pf\n", + __func__, __builtin_return_address(0)); + + mutex_lock(&mc_pmic->lock); + } + dev_dbg(&mc_pmic->i2c_client->dev, "%s from %pf\n", + __func__, __builtin_return_address(0)); +} +EXPORT_SYMBOL(mc34708_lock); + +void mc34708_unlock(struct mc34708 *mc_pmic) +{ + dev_dbg(&mc_pmic->i2c_client->dev, "%s from %pf\n", + __func__, __builtin_return_address(0)); + mutex_unlock(&mc_pmic->lock); +} +EXPORT_SYMBOL(mc34708_unlock); + +static int mc34708_i2c_24bit_read(struct i2c_client *client, + unsigned int offset, + unsigned int *value) +{ + unsigned char buf[3]; + int ret; + int i; + + memset(buf, 0, 3); + for (i = 0; i < PMIC_I2C_RETRY_TIMES; i++) { + ret = i2c_smbus_read_i2c_block_data(client, offset, 3, buf); + if (ret == 3) + break; + usleep_range(1000, 1500); + } + + if (ret == 3) { + *value = buf[0] << 16 | buf[1] << 8 | buf[2]; + return ret; + } else { + dev_err(&client->dev, "24bit read error, ret = %d\n", ret); + return -1; /* return -1 on failure */ + } +} + +int mc34708_reg_read(struct mc34708 *mc_pmic, unsigned int offset, + u32 *val) +{ + BUG_ON(!mutex_is_locked(&mc_pmic->lock)); + + if (offset > MC34708_NUMREGS) + return -EINVAL; + + if (mc34708_i2c_24bit_read(mc_pmic->i2c_client, offset, val) == -1) + return -1; + *val &= 0xffffff; + + dev_vdbg(&mc_pmic->i2c_client->dev, "mc_pmic read [%02d] -> 0x%06x\n", + offset, *val); + + return 0; +} +EXPORT_SYMBOL(mc34708_reg_read); + +static int mc34708_i2c_24bit_write(struct i2c_client *client, + unsigned int offset, unsigned int reg_val) +{ + char buf[3]; + int ret; + int i; + + buf[0] = (reg_val >> 16) & 0xff; + buf[1] = (reg_val >> 8) & 0xff; + buf[2] = (reg_val) & 0xff; + + for (i = 0; i < PMIC_I2C_RETRY_TIMES; i++) { + ret = i2c_smbus_write_i2c_block_data(client, offset, 3, buf); + if (ret == 0) + break; + usleep_range(1000, 1500); + } + if (i == PMIC_I2C_RETRY_TIMES) + dev_err(&client->dev, "24bit write error, ret = %d\n", ret); + + return ret; +} + +int mc34708_reg_write(struct mc34708 *mc_pmic, unsigned int offset, u32 val) +{ + BUG_ON(!mutex_is_locked(&mc_pmic->lock)); + + if (offset > MC34708_NUMREGS) + return -EINVAL; + + if (mc34708_i2c_24bit_write(mc_pmic->i2c_client, offset, val)) + return -1; + val &= 0xffffff; + + dev_vdbg(&mc_pmic->i2c_client->dev, "mc_pmic write[%02d] -> 0x%06x\n", + offset, val); + + return 0; +} +EXPORT_SYMBOL(mc34708_reg_write); + +int mc34708_reg_rmw(struct mc34708 *mc_pmic, unsigned int offset, u32 mask, + u32 val) +{ + int ret; + u32 valread; + + BUG_ON(val & ~mask); + + ret = mc34708_reg_read(mc_pmic, offset, &valread); + if (ret) + return ret; + + valread = (valread & ~mask) | val; + + return mc34708_reg_write(mc_pmic, offset, valread); +} +EXPORT_SYMBOL(mc34708_reg_rmw); + +int mc34708_irq_mask(struct mc34708 *mc_pmic, int irq) +{ + int ret; + unsigned int offmask = + irq < 24 ? MC34708_REG_INT_MASK0 : MC34708_REG_INT_MASK1; + u32 irqbit = 1 << (irq < 24 ? irq : irq - 24); + u32 mask; + + if (irq < 0 || irq >= MC34708_NUM_IRQ) + return -EINVAL; + + ret = mc34708_reg_read(mc_pmic, offmask, &mask); + if (ret) + return ret; + + if (mask & irqbit) + /* already masked */ + return 0; + + return mc34708_reg_write(mc_pmic, offmask, mask | irqbit); +} +EXPORT_SYMBOL(mc34708_irq_mask); + +int mc34708_irq_unmask(struct mc34708 *mc_pmic, int irq) +{ + int ret; + unsigned int offmask = + irq < 24 ? MC34708_REG_INT_MASK0 : MC34708_REG_INT_MASK1; + u32 irqbit = 1 << (irq < 24 ? irq : irq - 24); + u32 mask; + + if (irq < 0 || irq >= MC34708_NUM_IRQ) + return -EINVAL; + + ret = mc34708_reg_read(mc_pmic, offmask, &mask); + if (ret) + return ret; + + if (!(mask & irqbit)) + /* already unmasked */ + return 0; + + return mc34708_reg_write(mc_pmic, offmask, mask & ~irqbit); +} +EXPORT_SYMBOL(mc34708_irq_unmask); + +int mc34708_irq_status(struct mc34708 *mc_pmic, int irq, int *enabled, + int *pending) +{ + int ret; + unsigned int offmask = + irq < 24 ? MC34708_REG_INT_MASK0 : MC34708_REG_INT_MASK1; + unsigned int offstat = + irq < 24 ? MC34708_REG_INT_STATUS0 : MC34708_REG_INT_STATUS1; + u32 irqbit = 1 << (irq < 24 ? irq : irq - 24); + + if (irq < 0 || irq >= MC34708_NUM_IRQ) + return -EINVAL; + + if (enabled) { + u32 mask; + + ret = mc34708_reg_read(mc_pmic, offmask, &mask); + if (ret) + return ret; + + *enabled = mask & irqbit; + } + + if (pending) { + u32 stat; + + ret = mc34708_reg_read(mc_pmic, offstat, &stat); + if (ret) + return ret; + + *pending = stat & irqbit; + } + + return 0; +} +EXPORT_SYMBOL(mc34708_irq_status); + +int mc34708_irq_ack(struct mc34708 *mc_pmic, int irq) +{ + unsigned int offstat = + irq < 24 ? MC34708_REG_INT_STATUS0 : MC34708_REG_INT_STATUS1; + unsigned int val = 1 << (irq < 24 ? irq : irq - 24); + + BUG_ON(irq < 0 || irq >= MC34708_NUM_IRQ); + + return mc34708_reg_write(mc_pmic, offstat, val); +} +EXPORT_SYMBOL(mc34708_irq_ack); + +int mc34708_irq_request_nounmask(struct mc34708 *mc_pmic, int irq, + irq_handler_t handler, const char *name, + void *dev) +{ + BUG_ON(!mutex_is_locked(&mc_pmic->lock)); + BUG_ON(!handler); + + if (irq < 0 || irq >= MC34708_NUM_IRQ) + return -EINVAL; + + if (mc_pmic->irqhandler[irq]) + return -EBUSY; + + mc_pmic->irqhandler[irq] = handler; + mc_pmic->irqdata[irq] = dev; + + return 0; +} +EXPORT_SYMBOL(mc34708_irq_request_nounmask); + +int mc34708_irq_request(struct mc34708 *mc_pmic, int irq, + irq_handler_t handler, const char *name, void *dev) +{ + int ret; + + ret = mc34708_irq_request_nounmask(mc_pmic, irq, handler, name, dev); + if (ret) + return ret; + + ret = mc34708_irq_unmask(mc_pmic, irq); + if (ret) { + mc_pmic->irqhandler[irq] = NULL; + mc_pmic->irqdata[irq] = NULL; + return ret; + } + + return 0; +} +EXPORT_SYMBOL(mc34708_irq_request); + +int mc34708_irq_free(struct mc34708 *mc_pmic, int irq, void *dev) +{ + int ret; + BUG_ON(!mutex_is_locked(&mc_pmic->lock)); + + if (irq < 0 || irq >= MC34708_NUM_IRQ || !mc_pmic->irqhandler[irq] || + mc_pmic->irqdata[irq] != dev) + return -EINVAL; + + ret = mc34708_irq_mask(mc_pmic, irq); + if (ret) + return ret; + + mc_pmic->irqhandler[irq] = NULL; + mc_pmic->irqdata[irq] = NULL; + + return 0; +} +EXPORT_SYMBOL(mc34708_irq_free); + +static inline irqreturn_t mc34708_irqhandler(struct mc34708 *mc_pmic, int irq) +{ + return mc_pmic->irqhandler[irq] (irq, mc_pmic->irqdata[irq]); +} + +/* + * returns: number of handled irqs or negative error + * locking: holds mc_pmic->lock + */ +static int mc34708_irq_handle(struct mc34708 *mc_pmic, + unsigned int offstat, unsigned int offmask, + int baseirq) +{ + u32 stat, mask; + int ret = mc34708_reg_read(mc_pmic, offstat, &stat); + int num_handled = 0; + + if (ret) + return ret; + + ret = mc34708_reg_read(mc_pmic, offmask, &mask); + if (ret) + return ret; + + while (stat & ~mask) { + int irq = __ffs(stat & ~mask); + + stat &= ~(1 << irq); + + if (likely(mc_pmic->irqhandler[baseirq + irq])) { + irqreturn_t handled; + + handled = mc34708_irqhandler(mc_pmic, baseirq + irq); + if (handled == IRQ_HANDLED) + num_handled++; + } else { + dev_err(&mc_pmic->i2c_client->dev, + "BUG: irq %u but no handler\n", baseirq + irq); + + mask |= 1 << irq; + + ret = mc34708_reg_write(mc_pmic, offmask, mask); + } + } + + return num_handled; +} + +static irqreturn_t mc34708_irq_thread(int irq, void *data) +{ + struct mc34708 *mc_pmic = data; + irqreturn_t ret; + int handled = 0; + + mc34708_lock(mc_pmic); + + ret = mc34708_irq_handle(mc_pmic, MC34708_REG_INT_STATUS0, + MC34708_REG_INT_MASK0, 0); + if (ret > 0) + handled = 1; + + ret = mc34708_irq_handle(mc_pmic, MC34708_REG_INT_STATUS1, + MC34708_REG_INT_MASK1, 24); + if (ret > 0) + handled = 1; + + mc34708_unlock(mc_pmic); + + return IRQ_RETVAL(handled); +} + +#define maskval(reg, mask) (((reg) & (mask)) >> __ffs(mask)) +static int mc34708_identify(struct mc34708 *mc_pmic, + struct mc34708_version_t *ver) +{ + int rev_id = 0; + int rev1 = 0; + int rev2 = 0; + int finid = 0; + int icid = 0; + int ret; + ret = mc34708_reg_read(mc_pmic, MC34708_REG_IDENTIFICATION, &rev_id); + if (ret) { + dev_dbg(&mc_pmic->i2c_client->dev, "read ID error!%x\n", ret); + return ret; + } + rev1 = (rev_id & 0x018) >> 3; + rev2 = (rev_id & 0x007); + icid = (rev_id & 0x01C0) >> 6; + finid = (rev_id & 0x01E00) >> 9; + ver->id = MC_PMIC_ID_MC34708; + + ver->revision = ((rev1 * 10) + rev2); + dev_dbg(&mc_pmic->i2c_client->dev, + "mc_pmic Rev %d.%d FinVer %x detected\n", rev1, rev2, finid); + + return 0; +} + +static const struct i2c_device_id *i2c_match_id(const struct i2c_device_id *id, + const struct i2c_client *client) +{ + while (id->name[0]) { + if (strcmp(client->name, id->name) == 0) + return id; + id++; + } + + return NULL; +} + +static const struct i2c_device_id *i2c_get_device_id(const struct i2c_client + *idev) +{ + const struct i2c_driver *idrv = to_i2c_driver(idev->dev.driver); + + return i2c_match_id(idrv->id_table, idev); +} + +static const char *mc34708_get_chipname(struct mc34708 *mc_pmic) +{ + const struct i2c_device_id *devid = + i2c_get_device_id(mc_pmic->i2c_client); + + if (!devid) + return NULL; + + return mc34708_chipname[devid->driver_data]; +} + +int mc34708_get_flags(struct mc34708 *mc_pmic) +{ + struct mc34708_platform_data *pdata = + dev_get_platdata(&mc_pmic->i2c_client->dev); + + return pdata->flags; +} +EXPORT_SYMBOL(mc34708_get_flags); + +static int mc34708_add_subdevice_pdata(struct mc34708 *mc_pmic, + const char *format, void *pdata, + size_t pdata_size) +{ + char buf[30]; + const char *name = mc34708_get_chipname(mc_pmic); + + struct mfd_cell cell = { + .platform_data = pdata, + .pdata_size = pdata_size, + }; + + /* there is no asnprintf in the kernel :-( */ + if (snprintf(buf, sizeof(buf), format, name) > sizeof(buf)) + return -E2BIG; + + cell.name = kmemdup(buf, strlen(buf) + 1, GFP_KERNEL); + if (!cell.name) + return -ENOMEM; + + return mfd_add_devices(&mc_pmic->i2c_client->dev, -1, &cell, 1, NULL, + 0); +} + +static int mc34708_add_subdevice(struct mc34708 *mc_pmic, const char *format) +{ + return mc34708_add_subdevice_pdata(mc_pmic, format, NULL, 0); +} + +static int mc34708_probe(struct i2c_client *client, + const struct i2c_device_id *id) +{ + struct mc34708 *mc_pmic; + struct mc34708_version_t version; + struct mc34708_platform_data *pdata = client->dev.platform_data; + struct device_node *np = client->dev.of_node; + int ret; + + mc_pmic = kzalloc(sizeof(*mc_pmic), GFP_KERNEL); + if (!mc_pmic) + return -ENOMEM; + i2c_set_clientdata(client, mc_pmic); + mc_pmic->i2c_client = client; + + mutex_init(&mc_pmic->lock); + mc34708_lock(mc_pmic); + mc34708_identify(mc_pmic, &version); + /* mask all irqs */ + ret = mc34708_reg_write(mc_pmic, MC34708_REG_INT_MASK0, 0x00ffffff); + if (ret) + goto err_mask; + + ret = mc34708_reg_write(mc_pmic, MC34708_REG_INT_MASK1, 0x00ffffff); + if (ret) + goto err_mask; + + ret = request_threaded_irq(client->irq, NULL, mc34708_irq_thread, + IRQF_ONESHOT | IRQF_TRIGGER_HIGH, "mc_pmic", + mc_pmic); + + if (ret) { + err_mask: + mc34708_unlock(mc_pmic); + dev_set_drvdata(&client->dev, NULL); + kfree(mc_pmic); + return ret; + } + + mc34708_unlock(mc_pmic); + + if (pdata && pdata->flags & MC34708_USE_ADC) + mc34708_add_subdevice(mc_pmic, "%s-adc"); + else if (of_get_property(np, "fsl,mc34708-uses-adc", NULL)) + mc34708_add_subdevice(mc_pmic, "%s-adc"); + + + if (pdata && pdata->flags & MC34708_USE_REGULATOR) { + struct mc34708_regulator_platform_data regulator_pdata = { + .num_regulators = pdata->num_regulators, + .regulators = pdata->regulators, + }; + + mc34708_add_subdevice_pdata(mc_pmic, "%s-regulator", + &regulator_pdata, + sizeof(regulator_pdata)); + } else if (of_find_node_by_name(np, "regulators")) { + mc34708_add_subdevice(mc_pmic, "%s-regulator"); + } + + if (pdata && pdata->flags & MC34708_USE_RTC) + mc34708_add_subdevice(mc_pmic, "%s-rtc"); + else if (of_get_property(np, "fsl,mc34708-uses-rtc", NULL)) + mc34708_add_subdevice(mc_pmic, "%s-rtc"); + + if (pdata && pdata->flags & MC34708_USE_TOUCHSCREEN) + mc34708_add_subdevice(mc_pmic, "%s-ts"); + else if (of_get_property(np, "fsl,mc34708-uses-ts", NULL)) + mc34708_add_subdevice(mc_pmic, "%s-ts"); + + return 0; +} + +static int __devexit mc34708_remove(struct i2c_client *client) +{ + struct mc34708 *mc_pmic = dev_get_drvdata(&client->dev); + + free_irq(mc_pmic->i2c_client->irq, mc_pmic); + + mfd_remove_devices(&client->dev); + + kfree(mc_pmic); + + return 0; +} + +static const struct of_device_id mc34708_dt_ids[] = { + { .compatible = "fsl,mc34708" }, + { /* sentinel */ } +}; + +static struct i2c_driver mc34708_driver = { + .id_table = mc34708_device_id, + .driver = { + .name = "mc34708", + .owner = THIS_MODULE, + .of_match_table = mc34708_dt_ids, + }, + .probe = mc34708_probe, + .remove = __devexit_p(mc34708_remove), +}; + +static int __init mc34708_init(void) +{ + return i2c_add_driver(&mc34708_driver); +} +subsys_initcall(mc34708_init); + +static void __exit mc34708_exit(void) +{ + i2c_del_driver(&mc34708_driver); +} +module_exit(mc34708_exit); + +MODULE_DESCRIPTION("Core driver for Freescale MC34708 PMIC"); +MODULE_AUTHOR("Robin Gong <B38343(a)freescale.com>, " + "Ying-Chun Liu (PaulLiu) <paul.liu(a)linaro.org>"); +MODULE_LICENSE("GPL v2"); diff --git a/include/linux/mfd/mc34708.h b/include/linux/mfd/mc34708.h new file mode 100644 index 0000000..505813d --- /dev/null +++ b/include/linux/mfd/mc34708.h @@ -0,0 +1,134 @@ +/* For mc34708's pmic driver + * Copyright (C) 2004-2011 Freescale Semiconductor, Inc. + * + * based on: + * Copyright 2009-2010 Pengutronix, Uwe Kleine-Koenig + * <u.kleine-koenig(a)pengutronix.de> + * + * This program is free software; you can redistribute it and/or modify it under + * the terms of the GNU General Public License version 2 as published by the + * Free Software Foundation. + */ +#ifndef __LINUX_MFD_MC_34708_H +#define __LINUX_MFD_MC_34708_H + +#include <linux/interrupt.h> +#include <linux/regulator/driver.h> + +struct mc34708; + +void mc34708_lock(struct mc34708 *mc_pmic); +void mc34708_unlock(struct mc34708 *mc_pmic); + +int mc34708_reg_read(struct mc34708 *mc_pmic, unsigned int offset, u32 *val); +int mc34708_reg_write(struct mc34708 *mc_pmic, unsigned int offset, u32 val); +int mc34708_reg_rmw(struct mc34708 *mc_pmic, unsigned int offset, + u32 mask, u32 val); + +int mc34708_get_flags(struct mc34708 *mc_pmic); + +int mc34708_irq_request(struct mc34708 *mc_pmic, int irq, + irq_handler_t handler, const char *name, void *dev); +int mc34708_irq_request_nounmask(struct mc34708 *mc_pmic, int irq, + irq_handler_t handler, const char *name, + void *dev); +int mc34708_irq_free(struct mc34708 *mc_pmic, int irq, void *dev); +int mc34708_irq_mask(struct mc34708 *mc_pmic, int irq); +int mc34708_irq_unmask(struct mc34708 *mc_pmic, int irq); +int mc34708_irq_status(struct mc34708 *mc_pmic, int irq, + int *enabled, int *pending); +int mc34708_irq_ack(struct mc34708 *mc_pmic, int irq); + +int mc34708_get_flags(struct mc34708 *mc_pmic); + +#define MC34708_SW1A 0 +#define MC34708_SW1B 1 +#define MC34708_SW2 2 +#define MC34708_SW3 3 +#define MC34708_SW4A 4 +#define MC34708_SW4B 5 +#define MC34708_SW5 6 +#define MC34708_SWBST 7 +#define MC34708_VPLL 8 +#define MC34708_VREFDDR 9 +#define MC34708_VUSB 10 +#define MC34708_VUSB2 11 +#define MC34708_VDAC 12 +#define MC34708_VGEN1 13 +#define MC34708_VGEN2 14 +#define MC34708_REGU_NUM 15 + +#define MC34708_REG_INT_STATUS0 0 +#define MC34708_REG_INT_MASK0 1 +#define MC34708_REG_INT_STATUS1 3 +#define MC34708_REG_INT_MASK1 4 +#define MC34708_REG_IDENTIFICATION 7 + +#define MC34708_IRQ_ADCDONE 0 +#define MC34708_IRQ_TSDONE 1 +#define MC34708_IRQ_TSPENDET 2 +#define MC34708_IRQ_USBDET 3 +#define MC34708_IRQ_AUXDET 4 +#define MC34708_IRQ_USBOVP 5 +#define MC34708_IRQ_AUXOVP 6 +#define MC34708_IRQ_CHRTIMEEXP 7 +#define MC34708_IRQ_BATTOTP 8 +#define MC34708_IRQ_BATTOVP 9 +#define MC34708_IRQ_CHRCMPL 10 +#define MC34708_IRQ_WKVBUSDET 11 +#define MC34708_IRQ_WKAUXDET 12 +#define MC34708_IRQ_LOWBATT 13 +#define MC34708_IRQ_VBUSREGMI 14 +#define MC34708_IRQ_ATTACH 15 +#define MC34708_IRQ_DETACH 16 +#define MC34708_IRQ_KP 17 +#define MC34708_IRQ_LKP 18 +#define MC34708_IRQ_LKR 19 +#define MC34708_IRQ_UKNOW_ATTA 20 +#define MC34708_IRQ_ADC_CHANGE 21 +#define MC34708_IRQ_STUCK_KEY 22 +#define MC34708_IRQ_STUCK_KEY_RCV 23 +#define MC34708_IRQ_1HZ 24 +#define MC34708_IRQ_TODA 25 +#define MC34708_IRQ_UNUSED1 26 +#define MC34708_IRQ_PWRON1 27 +#define MC34708_IRQ_PWRON2 28 +#define MC34708_IRQ_WDIRESET 29 +#define MC34708_IRQ_SYSRST 30 +#define MC34708_IRQ_RTCRST 31 +#define MC34708_IRQ_PCI 32 +#define MC34708_IRQ_WARM 33 +#define MC34708_IRQ_MEMHLD 34 +#define MC34708_IRQ_UNUSED2 35 +#define MC34708_IRQ_THWARNL 36 +#define MC34708_IRQ_THWARNH 37 +#define MC34708_IRQ_CLK 38 +#define MC34708_IRQ_UNUSED3 39 +#define MC34708_IRQ_SCP 40 +#define MC34708_NUMREGS 0x3f +#define MC34708_NUM_IRQ 46 + +struct mc34708_regulator_init_data { + int id; + struct regulator_init_data *init_data; +}; + +struct mc34708_regulator_platform_data { + int num_regulators; + struct mc34708_regulator_init_data *regulators; +}; + +struct mc34708_platform_data { +#define MC34708_USE_TOUCHSCREEN (1 << 0) +#define MC34708_USE_CODEC (1 << 1) +#define MC34708_USE_ADC (1 << 2) +#define MC34708_USE_RTC (1 << 3) +#define MC34708_USE_REGULATOR (1 << 4) +#define MC34708_USE_LED (1 << 5) + unsigned int flags; + + int num_regulators; + struct mc34708_regulator_init_data *regulators; +}; + +#endif /* ifndef __LINUX_MFD_MC_PMIC_H */ -- 1.7.9.5

11 years, 9 months

Linaro recommended (tm) brand of SD card?

by Michael Hudson-Doyle

Hi, The SD card I routinely use for testing which I got at some Linaro meeting or other has fallen apart (physically), so I'm on the hunt for a new one. Does anyone have a recommendation of a brand of card I should be looking for? For LAVA stuff, it needs to be at least 8 gigs. Cheers, mwh

11 years, 10 months

[PATCH v5 0/7] cleanup imx5 idle, add imx5/6 cpuidle

by Robert Lee

Cleanup up imx5 idle code and enable imx5 low powe idle for imx53. Add common imx cpuidle initialization functionality and add a i.MX5 and i.MX6Q platform cpuidle implementation. Based on v3.4 plus recently submitted/accepted MACHINE_INIT late_initcall patch series: http://www.spinics.net/lists/arm-kernel/msg171620.html and common clock patch series: http://comments.gmane.org/gmane.linux.ports.arm.kernel/161043 v5 changes: * rebase to v3.4 plus common clock patch series * various trivial changes/fixes from v4 submssion * Add separate pm_init patch for imx51 and imx53 v4 changes: * Added several imx5 idle cleanups to series. * Modified imx_io_p2v function to allow common imx5 AIPS2 bus virutal address * Added comment to tzic_wakeup_enable(). * Movied imx5 idle code from mm-imx5.c to pm-imx5.c. * Removed unnecessary time consuming code execution that ran on each idle instance. * modified imx5_pm_init to be a late_initcall * added late_initcall to all imx53 MACHINE_START entries. * enabled imx5 low power idle for imx53 * rebased cpuidle driver on top of imx5 cleanup changes. * modified cpuidle driver exit time to reflect removal of unnecessary code http://comments.gmane.org/gmane.linux.linaro.devel/11858 v3 changes: * removed file introduced in v1 no longer needed after v2 [per Shawn Guo] * re-ordered added #includes in alphabetical order [per Shawn Guo] * Remove if(!drv) check to allow handling by stack trace [per Sascha Hauer] * Added missing return value in error meesage [per Shawn Guo] * fixed (void *) casting problem pointed out Sasha Hauer. Used a different type of casting to properly give warning on improper casting: arm_pm_idle = (void (*)(void))imx5_idle; Used this casting instead of adding a dummy caller function because adding the dummy function appears to unnecessarily introduce the following additional operations: static void imx5_pm_idle(void) { a0: e1a0c00d mov ip, sp a4: e92dd800 push {fp, ip, lr, pc} a8: e24cb004 sub fp, ip, #4 imx5_idle(); ac: ebffffd3 bl 0 <imx5_idle> } b0: e89da800 ldm sp, {fp, sp, pc} http://permalink.gmane.org/gmane.linux.linaro.devel/11653 v2 changes: * Removed some unnecessary spaces [per Jess Juhl] * Added return value for an error message [per Sascha Hauer] * Reworked init scheme to use device tree late_initcall [per Sascha and Shawn] * Moved imx6q and imx5 cpuidle functionality to existing files. https://lkml.org/lkml/2012/5/1/363 v1 initial submission: https://lkml.org/lkml/2012/4/16/644 Robert Lee (7): ARM: imx: Modify IMX_IO_P2V macro ARM: imx: Add comments to tzic_enable_waker() ARM: imx: clean and consolidate imx5 suspend and idle code ARM: imx: Enable imx53 low power idle ARM: imx: Add common imx cpuidle init functionality. ARM: imx: Add imx5 cpuidle ARM: imx: Add imx6q cpuidle driver arch/arm/mach-imx/clk-imx51-imx53.c | 2 +- arch/arm/mach-imx/clock-mx51-mx53.c | 1 + arch/arm/mach-imx/imx53-dt.c | 1 + arch/arm/mach-imx/mach-imx6q.c | 19 +++++ arch/arm/mach-imx/mach-mx53_ard.c | 1 + arch/arm/mach-imx/mach-mx53_evk.c | 1 + arch/arm/mach-imx/mach-mx53_loco.c | 1 + arch/arm/mach-imx/mach-mx53_smd.c | 1 + arch/arm/mach-imx/mm-imx5.c | 26 ++----- arch/arm/mach-imx/pm-imx5.c | 111 ++++++++++++++++++++++------- arch/arm/plat-mxc/Makefile | 1 + arch/arm/plat-mxc/cpuidle.c | 80 +++++++++++++++++++++ arch/arm/plat-mxc/include/mach/common.h | 6 +- arch/arm/plat-mxc/include/mach/cpuidle.h | 22 ++++++ arch/arm/plat-mxc/include/mach/hardware.h | 25 ++++--- arch/arm/plat-mxc/tzic.c | 4 ++ 16 files changed, 243 insertions(+), 59 deletions(-) create mode 100644 arch/arm/plat-mxc/cpuidle.c create mode 100644 arch/arm/plat-mxc/include/mach/cpuidle.h -- 1.7.10

11 years, 10 months

Linaro rootfs / kernel version combination

by Suresh Kumar SHUKLA

Hi all, I need some help in choosing right version of rootfs for my environment. I have a development board with ST SoC with Cortex-A9 dual core and VFPv3-D16. Kernel version available to me is 2.6.37. I am using Debian 6 arm-hf port, but would like to try Linaro rootfs. 1. What is the version of linaro filesytem I should pick (optimum for -mfpu=vfp and -mfloat-abi=hard) ? 2. I don't have a hwpack, is it possible to use vexpress or similar rootfs. Regards, Suresh

11 years, 10 months

android-3.4 or android-3.4-compat

by John Stultz

Hey Andrey, Zach, So I'm back from my vacation, and have found that the Android team has released a -compat tree for their 3.4 kernel. Basically this tree re-adds some items like earlysuspend and classic wakelocks in order to provide better compatibility with old (and by old, I really mean current as far as we see - so ICS and earlier) Android userland. Since we're still shipping ICS, and have no access to whatever the Android 5.0 userland will be, it seems merging in the -compat tree would make sense. However, I know Tixy and others have already tried to address the lack of earlysuspend in the android-3.3+ kernels, so I wanted to double check that this wouldn't cause additional pain (since those adjustments might need to be reverted). So I just wanted to check first with folks to make sure there are no objections to merging in the -compat changes, and that the timing of merging in these changes isn't problematic (I can happily hold off till this months release is done, so we don't risk any last minute gotchas). thanks -john

11 years, 10 months

12.05 linux-linaro kernel tree

by Andrey Konovalov

Greetings, So far I wasn't updating the linux-linaro tree since the 12.04 release. (The generic topic updates were being done to the linux-linaro-core-tracking tree) Now it is time to move the focus to the linux-linaro tree. For one week it will use the mainline tip as the base. Then, on next Thursday the most recent -rc will be selected as the base, and won't be changed until 12.05 is released. Most probably it will be v3.4-rc7. The 12.05 linux-linaro tree will get the ARM and Samsung LTs topics plus the 7 generic topics currently included into the linux-linaro-core-tracking tree: ufs (ufs-for-linux-linaro) emmc (emmc-for-linux-linaro) thermal_exynos4_imx6 (thermal_exynos4_imx6_work) linaro-android-3.4 armlt-gator (tracking-armlt-gator) umm-wip (umm-3.4rc4-wip) linaro-configs-3.4 If you don't see your generic topic in this list, but you think it should be there, please let me know ASAP. If you have a new topic to add, please send me the request before the next Thursday, May 17; the sooner, the better. The requirements for a topic can be found here: https://wiki.linaro.org/Platform/DevPlatform/LinuxLinaroKernelTreeProcess#A… The landing teams - please update your topic branches if needed: ARM: tracking-armlt-hdlcd tracking-armlt-mmc tracking-armlt-arm-arch-fixes tracking-armlt-misc-fixes tracking-armlt-ubuntu-config tracking-armlt-android-config Samsung: topic/base topic/core topic/bl topic/dt topic/fb topic/pd topic/s2ram topic/asv_cpufreq topic/led topic/dummy_reg topic/gadget topic/touch topic/wlan topic/audio topic/hdmi topic/mali topic/cma_v24 topic/android_config Thanks, Andrey

11 years, 10 months

[PATCH] [perf] sched spr-replay load record/playback.

by Pantelis Antoniou

This is an attempt to create a portable scheduler load recorder and playback. While there exists a perf recorder and playback functionality in mainline perf it is not very useful in trying to debug the scheduler. There are a number of problems that this patch solves. 1. The format used by default is the binary perf trace format which is both non-portable between arches and non human editable. The format spr-replay uses is text based and easily understandable, and completely portable between arches & kernel version. It can been used to collect load data from an android ICS setup and then have the load analyzed in a normal linux x86 host. 2. For accurate load capture, it is important that the implementation details of the original machine that the load was captured to be removed as much as possible. For example if you capture a trace of two cpu hogs in a single core machine, the amount of time the processes were blocked waiting for the cpu should be removed, and when the load replayed on a dual core machine both tasks should be run in parallel. There is a switch to preserve the timings if you desire to compare. 3. When capturing a load you get to see various possibly spurious wakeups and interruptions by kernel tasks. It is important to be possible to select only the tasks that are important, and then have all the other tasks effects removed. 4. This patch allows capture of forking activity which is important and affects the scheduler operation quite a lot. Example use: Record load to a perf.data file (on target) $ perf record \ -a -R -f -m 8192 -c 1 \ -e sched:sched_switch \ -e sched:sched_process_exit \ -e sched:sched_process_fork \ -e sched:sched_wakeup \ -e sched:sched_migrate_task List the tasks in the perf.data capture file (on target) $ perf sched spr-replay -l -n ksoftirqd/0/3] S R:7839561 [events/0/6] S R:16961982 [sync_supers/12] S R:31803 [bdi-default/13] S R:48753 .... Process the perf.data capture file to generate an spr load program (on target) $ perf sched spr-replay -g -n >capture.spr Execute the recorded load from the perf.data file (on target) $ perf sched spr-replay [-d] # -d debugging mode Copy the capture.spr file and execute it (on host) $ perf sched spr-replay -f capture.spr Process the perf.data capture file to generate an spr load program for of a single (or more) tasks (on target). $ perf sched spr-replay -g -n -s <task-pid|task-name|~regexp> >capture.spr For example capturing cyclictest (cyclictest -t1 -p 80 -n -i 100000 -l 1000) Generates a capture.spr program like this: [cyclictest/4646] burn 11703 sleep 99927688 burn 30594 sleep 99943711 burn 55277 sleep 99744841 burn 76059 sleep 99782827 burn 93262 sleep 99868136 burn 112340 sleep 99825780 burn 132841 sleep 99799446 burn 150585 sleep 99850653 burn 163754 sleep 80514018 burn 215810 exit 0 end Process the perf.data capture file to generate an spr load program which excludes load from a couple of kernel tasks: $ perf sched spr-replay -g -n -x '~flush-*' -x '~events/*' \ -x '~ksoftirqd/*' -x swapper >capture.spr Signed-off-by: Pantelis Antoniou <panto(a)antoniou-consulting.com> --- tools/perf/builtin-sched.c | 2352 +++++++++++++++++++++++++++++++++++++++++++- 1 files changed, 2332 insertions(+), 20 deletions(-) diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c index b125e07..3b23062 100644 --- a/tools/perf/builtin-sched.c +++ b/tools/perf/builtin-sched.c @@ -22,6 +22,13 @@ #include <semaphore.h> #include <pthread.h> #include <math.h> +#include <regex.h> + +#include <sys/mman.h> +#include <sys/prctl.h> +#include <linux/futex.h> +#include <sys/syscall.h> +#include <sys/time.h> static const char *input_name; @@ -43,6 +50,9 @@ static u64 sleep_measurement_overhead; static unsigned long nr_tasks; +#define mb() asm volatile ("" : : : "memory") +static u64 bogoloops = 0; + struct sched_atom; struct task_desc { @@ -61,6 +71,11 @@ struct task_desc { sem_t work_done_sem; u64 cpu_usage; + + u64 task_state; + + unsigned int exited : 1; /* exited */ + unsigned int selected : 1; /* part of playback */ }; enum sched_event_type { @@ -68,6 +83,9 @@ enum sched_event_type { SCHED_EVENT_SLEEP, SCHED_EVENT_WAKEUP, SCHED_EVENT_MIGRATION, + SCHED_EVENT_EXIT, + SCHED_EVENT_FORK_PARENT, + SCHED_EVENT_FORK_CHILD, }; struct sched_atom { @@ -78,11 +96,39 @@ struct sched_atom { unsigned long nr; sem_t *wait_sem; struct task_desc *wakee; + struct sched_atom *wakee_event; + u64 task_state; + int pid; + struct task_desc *parent; + struct task_desc *child; + + /* extra */ + unsigned int exited; /* event was generated after task has called exit() */ + char *msg; + + unsigned int waker_count; + struct task_desc **wakers; + struct sched_atom **waker_events; + unsigned long wake_id; }; +static const char *sched_atom_str(const struct sched_atom *atom); + static struct task_desc *pid_to_task[MAX_PID]; +static u64 replay_start_time; +static u64 replay_end_time; static struct task_desc **tasks; +static unsigned long next_wake_id; +static bool preserve_time = false; +static bool dry_run = false; +static bool generate = false; +static int debug = 0; +static bool spr_list = false; +static bool spr_select_list = false; +static const char *spr_filename = NULL; +static bool bogoburn = false; +static bool spr_replay = false; static pthread_mutex_t start_work_mutex = PTHREAD_MUTEX_INITIALIZER; static u64 start_time; @@ -92,6 +138,9 @@ static pthread_mutex_t work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER; static unsigned long nr_run_events; static unsigned long nr_sleep_events; static unsigned long nr_wakeup_events; +static unsigned long nr_exit_events; +static unsigned long nr_fork_parent_events; +static unsigned long nr_fork_child_events; static unsigned long nr_sleep_corrections; static unsigned long nr_run_events_optimized; @@ -218,6 +267,120 @@ static void calibrate_sleep_measurement_overhead(void) printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); } +static u64 bogoloops_measure_single(u64 loops) +{ + u64 cnt; + u64 ns1, ns2; + + /* loop to make sure you start close to the monotonic clock */ + ns2 = get_nsecs(); + while ((ns1 = get_nsecs()) - ns2 == 0) + ; + cnt = loops; + while (cnt-- > 0) + mb(); + ns2 = get_nsecs(); + return ns2 - ns1; +} + +static u64 bogoloops_measure_stable(u64 loops) +{ + int i; + unsigned int j, k; + u64 delta; + u64 delta_samples[16]; /* 16 is an adequate number */ + u64 estimate, maxdiff, new_estimate; + int imin, imax; /* indices of min & max */ + + i = 0; + do { + imin = 0; + imax = 0; + estimate = 0; + for (j = 0; j < ARRAY_SIZE(delta_samples); j++) { + /* measure a single time */ + delta = bogoloops_measure_single(loops); + delta_samples[j] = delta; + if (delta < delta_samples[imin]) + imin = j; + if (delta > delta_samples[imax]) + imax = j; + estimate += delta; + } + estimate /= ARRAY_SIZE(delta_samples); + + maxdiff = estimate >> 3; /* maxdiff is 12.5% of estimate */ + + /* only keep the ones that are with the proper range */ + k = 0; + new_estimate = 0; + for (j = 0; j < ARRAY_SIZE(delta_samples); j++) { + if (delta_samples[j] > estimate) + delta = delta_samples[j] - estimate; + else + delta = estimate - delta_samples[j]; + + /* too bad, continue */ + if (delta > maxdiff) + continue; + k++; + new_estimate += delta_samples[j]; + } + + /* we need at least half the samples to be right */ + if (k >= ARRAY_SIZE(delta_samples) / 2) + return new_estimate / k; + + } while (i++ < 8); /* try for 8 times */ + + /* we have failed to find a stable measurement */ + /* (this is normal for low loops values) */ + return 0; +} + +static void calculate_bogoloops_value(void) +{ + u64 loops, last_loops, new_loops; + u64 sample_period; + u64 ns1, ns2, delta, delta_new, delta_diff; + + /* 1 ms */ + sample_period = 1000000; + + ns1 = get_nsecs(); + loops = 0; + while (((ns2 = get_nsecs()) - ns1) < sample_period) + loops++; + + /* it is guaranteed that the loops value would be lower */ + /* we grow until we get right after the sample period */ + /* then we interpolate for the period */ + do { + last_loops = loops; + delta = bogoloops_measure_stable(loops); + loops *= 2; + } while (delta < sample_period); + + new_loops = ((last_loops * sample_period) + (delta / 2)) / delta; + delta_new = bogoloops_measure_stable(new_loops); + + if (delta_new > sample_period) + delta_diff = delta_new - sample_period; + else + delta_diff = sample_period - delta_new; + + bogoloops = new_loops; +} + +static void bogoburn_nsecs(u64 nsecs) +{ + u64 cnt; + + cnt = (bogoloops * nsecs) / 1000000LLU; + while (cnt-- > 0) + mb(); +} + static struct sched_atom * get_new_event(struct task_desc *task, u64 timestamp) { @@ -228,6 +391,8 @@ get_new_event(struct task_desc *task, u64 timestamp) event->timestamp = timestamp; event->nr = idx; + event->exited = task->exited; + task->nr_events++; size = sizeof(struct sched_atom *) * task->nr_events; task->atoms = realloc(task->atoms, size); @@ -249,7 +414,8 @@ static struct sched_atom *last_event(struct task_desc *task) static void add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration) { - struct sched_atom *event, *curr_event = last_event(task); + struct sched_atom *curr_event = last_event(task); + struct sched_atom *event; /* * optimize an existing RUN event by merging this one @@ -270,6 +436,18 @@ add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration) } static void +add_sched_event_exit(struct task_desc *task, u64 timestamp) +{ + struct sched_atom *event; + + event = get_new_event(task, timestamp); + + event->type = SCHED_EVENT_EXIT; + + nr_exit_events++; +} + +static int add_sched_event_wakeup(struct task_desc *task, u64 timestamp, struct task_desc *wakee) { @@ -282,11 +460,26 @@ add_sched_event_wakeup(struct task_desc *task, u64 timestamp, wakee_event = last_event(wakee); if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { targetless_wakeups++; - return; + return 1; /* targetless wakeup */ } + event->wakee_event = wakee_event; + + /* add to the waker array */ + wakee_event->waker_count++; + wakee_event->wakers = realloc(wakee_event->wakers, + wakee_event->waker_count * sizeof(*wakee_event->wakers)); + BUG_ON(wakee_event->wakers == NULL); + wakee_event->wakers[wakee_event->waker_count-1] = task; + + /* add to the waker event array */ + wakee_event->waker_events = realloc(wakee_event->waker_events, + wakee_event->waker_count * sizeof(*wakee_event->waker_events)); + BUG_ON(wakee_event->waker_events == NULL); + wakee_event->waker_events[wakee_event->waker_count-1] = event; + if (wakee_event->wait_sem) { multitarget_wakeups++; - return; + return 2; /* multi target wakup */ } wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); @@ -295,19 +488,51 @@ add_sched_event_wakeup(struct task_desc *task, u64 timestamp, event->wait_sem = wakee_event->wait_sem; nr_wakeup_events++; + + return 0; /* normal wakeup with both waker & wakee */ } static void add_sched_event_sleep(struct task_desc *task, u64 timestamp, - u64 task_state __used) + u64 task_state) { struct sched_atom *event = get_new_event(task, timestamp); event->type = SCHED_EVENT_SLEEP; + event->task_state = task_state; + + event->wake_id = ++next_wake_id; + BUG_ON(next_wake_id == 0); /* do not allow overflow */ nr_sleep_events++; } +static void +add_sched_event_fork_parent(struct task_desc *task, u64 timestamp, int child_pid) +{ + struct sched_atom *event; + + event = get_new_event(task, timestamp); + + event->type = SCHED_EVENT_FORK_PARENT; + event->pid = child_pid; + + nr_fork_parent_events++; +} + +static void +add_sched_event_fork_child(struct task_desc *task, u64 timestamp, int parent_pid) +{ + struct sched_atom *event; + + event = get_new_event(task, timestamp); + + event->type = SCHED_EVENT_FORK_CHILD; + event->pid = parent_pid; + + nr_fork_child_events++; +} + static struct task_desc *register_pid(unsigned long pid, const char *comm) { struct task_desc *task; @@ -316,18 +541,29 @@ static struct task_desc *register_pid(unsigned long pid, const char *comm) task = pid_to_task[pid]; - if (task) + if (task) { + /* update task name if it was unknown before */ + if (comm != NULL && strcmp(task->comm, "<unknown>") == 0) { + if (verbose) + printf("task #%lu with pid %ld updates name from '%s' to '%s'\n", + task->nr, task->pid, task->comm, comm); + strcpy(task->comm, comm); + } return task; + } task = zalloc(sizeof(*task)); task->pid = pid; task->nr = nr_tasks; - strcpy(task->comm, comm); + strcpy(task->comm, comm ? comm : "<unknown>"); + /* * every task starts in sleeping state - this gets ignored * if there's no wakeup pointing to this sleep state: + * NOTE: spr-replay get's awfully messy with this; removed. */ - add_sched_event_sleep(task, 0, 0); + if (!spr_replay) + add_sched_event_sleep(task, 0, 0); pid_to_task[pid] = task; nr_tasks++; @@ -336,7 +572,7 @@ static struct task_desc *register_pid(unsigned long pid, const char *comm) tasks[task->nr] = task; if (verbose) - printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm); + printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, task->comm); return task; } @@ -390,6 +626,11 @@ process_sched_event(struct task_desc *this_task __used, struct sched_atom *atom) break; case SCHED_EVENT_MIGRATION: break; + /* unused */ + case SCHED_EVENT_EXIT: + case SCHED_EVENT_FORK_PARENT: + case SCHED_EVENT_FORK_CHILD: + break; default: BUG_ON(1); } @@ -709,6 +950,21 @@ struct trace_fork_event { u32 child_pid; }; +struct trace_exit_event { + u32 size; + + u16 common_type; + u8 common_flags; + u8 common_preempt_count; + u32 common_pid; + u32 common_tgid; + + u32 pid, ppid; + u32 tid, ptid; + u64 time; +}; + + struct trace_migrate_task_event { u32 size; @@ -726,6 +982,9 @@ struct trace_migrate_task_event { }; struct trace_sched_handler { + int (*comm_event)(struct machine *machine, + u32 tid, const char *comm); + void (*switch_event)(struct trace_switch_event *, struct machine *, struct event_format *, @@ -759,18 +1018,80 @@ struct trace_sched_handler { int cpu, u64 timestamp, struct thread *thread); + + void (*exit_event)(struct trace_exit_event *, + struct machine *machine, + struct event_format *, + int cpu, + u64 timestamp, + struct thread *thread); }; + __used static void dump_event(struct event_format *event) +{ + printf("dump of event=%p\n", event); + printf(" %-20s %s\n", "name", event->name); + printf(" %-20s %d\n", "id", event->id); + printf(" %-20s %d (%s|%s|%s|%s|%s|%s)\n", "flags", event->flags, + (event->flags & EVENT_FL_ISFTRACE) ? "ISFTRACE" : "", + (event->flags & EVENT_FL_ISPRINT) ? "ISPRINT" : "", + (event->flags & EVENT_FL_ISBPRINT) ? "ISBPRINT" : "", + (event->flags & EVENT_FL_ISFUNCENT) ? "ISFUNCENT" : "", + (event->flags & EVENT_FL_ISFUNCRET) ? "ISFUNCRET" : "", + (event->flags & EVENT_FL_FAILED) ? "FAILED" : ""); + printf(" %-20s %s\n", "system", event->system); +} + +__used static void dump_perf_sample(struct perf_sample *sample) +{ + printf("dump of perf_sample=%p\n", sample); + printf(" %-20s %" PRIu64 "\n", "ip", sample->ip); + printf(" %-20s %" PRIu32 "\n", "pid", sample->pid); + printf(" %-20s %" PRIu32 "\n", "tid", sample->tid); + printf(" %-20s %" PRIu64 "\n", "time", sample->time); + printf(" %-20s %" PRIu64 "\n", "addr", sample->addr); + printf(" %-20s %" PRIu64 "\n", "id", sample->id); + printf(" %-20s %" PRIu64 "\n", "stream_id", sample->stream_id); + printf(" %-20s %" PRIu64 "\n", "period", sample->period); + printf(" %-20s %" PRIu32 "\n", "cpu", sample->cpu); + printf(" %-20s %" PRIu32 "\n", "raw_size", sample->raw_size); + printf(" %-20s %p\n", "raw_data", sample->raw_data); +} + +static u64 cpu_last_switched[MAX_CPUS]; +struct task_desc *cpu_last_task[MAX_CPUS]; + +static int +replay_comm_event(struct machine *machine __used, u32 tid, const char *comm) +{ + struct task_desc *task; + + if (verbose) { + printf("replay_comm_event tid=%u comm=%s\n", + tid, comm); + } + + task = register_pid(tid, comm); + if (task == NULL) { + dump_printf("problem processing replay_comm_event, skipping event.\n"); + return -1; + } + return 0; +} static void replay_wakeup_event(struct trace_wakeup_event *wakeup_event, struct machine *machine __used, struct event_format *event, int cpu __used, - u64 timestamp __used, + u64 timestamp, struct thread *thread __used) { struct task_desc *waker, *wakee; + struct sched_atom *waker_event, *wakee_event; + char *s, *e; + char buf[BUFSIZ]; + int ret; if (verbose) { printf("sched_wakeup event %p\n", event); @@ -781,13 +1102,47 @@ replay_wakeup_event(struct trace_wakeup_event *wakeup_event, wakeup_event->pid); } - waker = register_pid(wakeup_event->common_pid, "<unknown>"); + waker = register_pid(wakeup_event->common_pid, NULL); wakee = register_pid(wakeup_event->pid, wakeup_event->comm); - add_sched_event_wakeup(waker, timestamp, wakee); -} + ret = add_sched_event_wakeup(waker, timestamp, wakee); -static u64 cpu_last_switched[MAX_CPUS]; + waker_event = last_event(waker); + BUG_ON(waker_event == NULL); + + wakee_event = last_event(wakee); + + if (debug > 2) { + s = buf; + e = &buf[ARRAY_SIZE(buf)]; + e[-1] = '\0'; + + snprintf(s, e - s, "wakee %d preempt=%u prio=%" PRIu32 " succ=%" PRIu32 "", + wakeup_event->pid, + (unsigned int)wakeup_event->common_preempt_count, + wakeup_event->prio, wakeup_event->cpu); + e[-1] = '\0'; + s += strlen(s); + + if (ret == 1) { + snprintf(s, e - s, " (targetless )"); + e[-1] = '\0'; + s += strlen(s); + } else if (ret == 2) { + snprintf(s, e - s, " (multitarget)"); + e[-1] = '\0'; + s += strlen(s); + } + + if (wakee_event != NULL) { + snprintf(s, e - s, " (%s)", sched_atom_str(wakee_event)); + e[-1] = '\0'; + s += strlen(s); + } + + waker_event->msg = strdup(buf); + } +} static void replay_switch_event(struct trace_switch_event *switch_event, @@ -797,9 +1152,11 @@ replay_switch_event(struct trace_switch_event *switch_event, u64 timestamp, struct thread *thread __used) { - struct task_desc *prev, __used *next; + struct task_desc *prev, *next; + struct sched_atom *prev_atom; u64 timestamp0; s64 delta; + char buf[BUFSIZ]; if (verbose) printf("sched_switch event %p\n", event); @@ -827,9 +1184,20 @@ replay_switch_event(struct trace_switch_event *switch_event, next = register_pid(switch_event->next_pid, switch_event->next_comm); cpu_last_switched[cpu] = timestamp; + cpu_last_task[cpu] = next; add_sched_event_run(prev, timestamp, delta); - add_sched_event_sleep(prev, timestamp, switch_event->prev_state); + add_sched_event_sleep(prev, timestamp, switch_event->prev_state); /* was adding +delta */ + prev->task_state = switch_event->prev_state; + next->task_state = 0; /* task running */ + + prev_atom = last_event(prev); + BUG_ON(prev_atom == NULL); + + if (debug > 2) { + snprintf(buf, sizeof(buf), "switch to %d", switch_event->next_pid); + prev_atom->msg = strdup(buf); + } } @@ -837,22 +1205,66 @@ static void replay_fork_event(struct trace_fork_event *fork_event, struct event_format *event, int cpu __used, - u64 timestamp __used, + u64 timestamp, struct thread *thread __used) { + struct task_desc *parent, *child; + struct sched_atom *atom; + char buf[BUFSIZ]; + if (verbose) { printf("sched_fork event %p\n", event); printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid); printf("... child: %s/%d\n", fork_event->child_comm, fork_event->child_pid); } - register_pid(fork_event->parent_pid, fork_event->parent_comm); - register_pid(fork_event->child_pid, fork_event->child_comm); + + parent = register_pid(fork_event->parent_pid, fork_event->parent_comm); + child = register_pid(fork_event->child_pid, fork_event->child_comm); + + add_sched_event_fork_parent(parent, timestamp, fork_event->child_pid); + + atom = last_event(parent); + BUG_ON(atom == NULL); + atom->child = child; + + if (debug > 2) { + snprintf(buf, sizeof(buf), "child %d", fork_event->child_pid); + atom->msg = strdup(buf); + } + + add_sched_event_fork_child(child, timestamp, fork_event->parent_pid); + + atom = last_event(child); + BUG_ON(atom == NULL); + atom->parent = parent; + + if (debug > 2) { + snprintf(buf, sizeof(buf), "parent %d", fork_event->parent_pid); + atom->msg = strdup(buf); + } +} + +static void +replay_exit_event(struct trace_exit_event *exit_event __used, + struct machine *machine __used, + struct event_format *event __used, + int cpu __used, + u64 timestamp, + struct thread *thread __used) +{ + struct task_desc *task; + + task = register_pid(exit_event->pid, NULL); /* should never pick up unkown */ + add_sched_event_exit(task, timestamp); + task->exited = 1; /* mark that this task has exited */ } static struct trace_sched_handler replay_ops = { + .comm_event = replay_comm_event, .wakeup_event = replay_wakeup_event, .switch_event = replay_switch_event, .fork_event = replay_fork_event, + .exit_event = replay_exit_event, }; struct sort_dimension { @@ -1559,8 +1971,23 @@ process_sched_exit_event(struct perf_tool *tool __used, struct machine *machine __used, struct thread *thread __used) { + void *data = sample->raw_data; + struct trace_exit_event exit_event; + if (verbose) printf("sched_exit event %p\n", event); + + FILL_COMMON_FIELDS(exit_event, event, data); + + FILL_FIELD(exit_event, pid, event, data); + FILL_FIELD(exit_event, ppid, event, data); + FILL_FIELD(exit_event, tid, event, data); + FILL_FIELD(exit_event, ptid, event, data); + FILL_FIELD(exit_event, time, event, data); + + if (trace_handler->exit_event) + trace_handler->exit_event(&exit_event, machine, event, + sample->cpu, sample->time, thread); } static void @@ -1620,11 +2047,26 @@ static int perf_sched__process_tracepoint_sample(struct perf_tool *tool, return 0; } +static int perf_sched__process_comm(struct perf_tool *tool, + union perf_event *event, + struct perf_sample *sample, + struct machine *machine) +{ + if (verbose) + printf("\t*** %s:%d\n", event->comm.comm, event->comm.tid); + + if (trace_handler->comm_event) + trace_handler->comm_event(machine, event->comm.tid, event->comm.comm); + + return perf_event__process_comm(tool, event, sample, machine); +} + static struct perf_tool perf_sched = { .sample = perf_sched__process_tracepoint_sample, - .comm = perf_event__process_comm, + .comm = perf_sched__process_comm, .lost = perf_event__process_lost, .fork = perf_event__process_task, + .exit = perf_event__process_task, .ordered_samples = true, }; @@ -1780,7 +2222,7 @@ static void __cmd_replay(void) static const char * const sched_usage[] = { - "perf sched [<options>] {record|latency|map|replay|script}", + "perf sched [<options>] {record|latency|map|replay|script|spr-replay}", NULL }; @@ -1883,6 +2325,1868 @@ static int __cmd_record(int argc, const char **argv) return cmd_record(i, rec_argv, NULL); } +/**************************************************************************/ + +enum task_action_type { + TA_END, + TA_BURN, + TA_SLEEP, + TA_SPAWN, + TA_EXIT, + TA_CLONE_PARENT, + TA_CLONE_CHILD, + TA_WAIT_ID, + TA_SIGNAL_ID +}; + +struct task_action { + enum task_action_type action; + union { + struct { u64 nsecs; } burn; + struct { u64 nsecs; } sleep; + struct { u64 nsecs; } spawn; + struct { int ret; } exit; + struct { int child_pid; } clone_parent; + struct { int parent_pid; } clone_child; + struct { unsigned long id; } wait_id; + struct { unsigned long id; } signal_id; + } u; +}; + +#define T(x) ((u64)((double)(x) * 1E9)) + +struct task { + const char *name; + int pid; + const struct task_action *actions; +}; + +void *work_area; +struct playback *playback; + +enum match_type { + MT_PID, + MT_NAME, + MT_REGEX, +}; + +struct match_list_entry { + struct list_head node; + enum match_type type; + int pid; + char *name; + regex_t regex; +}; + +LIST_HEAD(select_list); +LIST_HEAD(exclude_list); + +static int match_task(struct list_head *lh, const char *name, int pid, int empty_list_val) +{ + int matched; + struct match_list_entry *mle; + + if (list_empty(lh)) + return empty_list_val; + + list_for_each_entry(mle, lh, node) { + switch (mle->type) { + case MT_PID: + matched = mle->pid == pid; + break; + case MT_NAME: + matched = strcmp(mle->name, name) == 0; + break; + case MT_REGEX: + matched = regexec(&mle->regex, name, 0, NULL, 0) == 0; + break; + default: + matched = 0; + break; + } + if (matched) + return 1; + } + return 0; +} + +struct task_run { + struct playback *playback; /* point back to the playback */ + const struct task *task; /* task */ + const struct task_action *current_action; + char name[BUFSIZ]; /* lots of space for a name */ + int id; + pid_t real_pid; + pid_t parent_pid; /* when created by CLONE_PARENT */ + u64 start_time; /* time when the current task starts */ + u64 local_time; /* local time (beginning from 0) */ + int children_count; +}; + +struct playback_info { + int task_cnt; + int sem_cnt; + int task_off; + int sem_off; + int work_area_size; +}; + +struct playback { + struct playback_info info; + + u64 origin_time; /* time when the simulation starts */ + int task_count; + struct task_run *task_runs; + unsigned long sem_count; + sem_t *semas; + + unsigned int flags; /* flags */ +#define PF_ABSOLUTE_TIME 1 +#define PF_RELATIVE_TIME 2 +#define PF_TIME_MASK 3 + + int debug_level; + + int spawn_count; /* left to spawn */ + int exited_count; /* left to exit */ + int forked_count; /* initially forked */ +}; + +static inline int playback_absolute_time(const struct playback *p) +{ + return ((p->flags & PF_TIME_MASK) == PF_ABSOLUTE_TIME); +} + +static inline int playback_relative_time(const struct playback *p) +{ + return ((p->flags & PF_TIME_MASK) == PF_RELATIVE_TIME); +} + +static inline void playback_set_absolute_time(struct playback *p) +{ + p->flags = (p->flags & ~PF_TIME_MASK) | PF_ABSOLUTE_TIME; +} + +static inline void playback_set_relative_time(struct playback *p) +{ + p->flags = (p->flags & ~PF_TIME_MASK) | PF_ABSOLUTE_TIME; +} + +static inline void playback_set_debug_level(struct playback *p, int debug_level) +{ + p->debug_level = debug_level; +} + +static const char *task_action_str(u64 local_ts, const struct task_action *ta); +const struct task_action *task_action_current(struct task_run *tr); +void task_action_advance(struct task_run *tr); + +void vtprintf(struct task_run *tr, const char *fmt, va_list ap); +void tprintf(struct task_run *tr, const char *fmt, ...) __attribute__((__format__(__printf__,2,3))); + +void vtprintf(struct task_run *tr, const char *fmt, va_list ap) +{ + struct playback *p = tr->playback; + int n, size; + char *str = NULL; + + if (p->debug_level <= 0) + return; + + size = 256; /* start with 80 chars (40 * 2) */ + do { + size *= 2; + str = realloc(str, size); + BUG_ON(str == NULL); + /* Try to print in the allocated space. */ + n = vsnprintf(str, size, fmt, ap); + } while (n < 0 || n >= size); + + printf("#%2d [%16" PRIu64 " %16" PRIu64 "] %s", tr->id, + tr->local_time, + get_nsecs() - p->origin_time, + str); + free(str); +} + +void tprintf(struct task_run *tr, const char *fmt, ...) +{ + struct playback *p; + va_list ap; + + p = tr->playback; + + if (p->debug_level <= 0) + return; + + va_start(ap, fmt); + vtprintf(tr, fmt, ap); + va_end(ap); +} + + +const struct task_action *task_action_current(struct task_run *tr) +{ + if (tr->current_action->action == TA_END) + return NULL; + return tr->current_action; +} + +void task_action_advance(struct task_run *tr) +{ + if (tr->current_action->action != TA_END) + tr->current_action++; +} + +struct task_action_desc { + unsigned int id; + const char *name; + void (*execute)(struct task_run *tr, const struct task_action *ta); +}; + +static void execute_task(struct task_run *tr) __attribute__((__noreturn__)); + +/* END is not processed */ + +static void execute_spawn(struct task_run *tr __used, const struct task_action *ta __used) +{ + /* we do nothing */ +} + +static void execute_burn(struct task_run *tr, const struct task_action *ta) +{ + struct playback *p = tr->playback; + u64 t1, dt, target; + + t1 = get_nsecs(); + dt = ta->u.burn.nsecs; + + if (playback_absolute_time(p)) + target = p->origin_time + tr->local_time + dt; + else + target = t1 + dt; + + if (target <= t1) { + tprintf(tr, "%s %" PRIu64 " - SKIPPED; slip of %" PRIu64 "ns\n", + "BURN", ta->u.burn.nsecs, t1 - target); + return; + } + + tprintf(tr, "%s %" PRIu64 ":%" PRIi64 "\n", "BURN", + ta->u.burn.nsecs, (int64_t)(target - t1)); + + if (!bogoburn) + burn_nsecs(target - t1); + else + bogoburn_nsecs(target - t1); + + tr->local_time += dt; +} + +static void execute_sleep(struct task_run *tr, const struct task_action *ta) +{ + struct playback *p = tr->playback; + u64 t1, dt, target; + + t1 = get_nsecs(); + dt = ta->u.sleep.nsecs; + + if (playback_absolute_time(p)) + target = p->origin_time + tr->local_time + dt; + else + target = t1 + dt; + + if (target <= t1) { + tprintf(tr, "%s %" PRIu64 " - SKIPPED; slip of %" PRIu64 "ns\n", + "SLEEP", ta->u.sleep.nsecs, t1 - target); + return; + } + + tprintf(tr, "%s %" PRIu64 ":%" PRIi64 "\n", "SLEEP", + ta->u.burn.nsecs, (int64_t)(target - t1)); + + sleep_nsecs(target - t1); /* bogomips */ + + tr->local_time += dt; +} + +static void execute_exit(struct task_run *tr, const struct task_action *ta) +{ + struct playback *p = tr->playback; + struct task_run *tr1; + pid_t real_pid; + int i, status; + + tprintf(tr, "%s\n", "EXIT"); + + while (tr->children_count > 0) { + real_pid = wait(&status); + if (real_pid < 0) { + fprintf(stderr, "panto-perf: Wait failed: %s:%d\n", __func__, __LINE__); + exit(EXIT_FAILURE); + } + if (!WIFEXITED(status)) + continue; + + /* locate child */ + for (i = 0, tr1 = p->task_runs; i < p->task_count; i++, tr1++) + if (real_pid == tr1->real_pid) + break; + assert(i < p->task_count); + + tr->children_count--; + } + exit(ta->u.exit.ret); +} + +static void execute_clone_parent(struct task_run *tr, const struct task_action *ta) +{ + struct playback *p = tr->playback; + struct task_run *tr1; + int i, pid; + pid_t real_pid; + + pid = ta->u.clone_parent.child_pid; + + tprintf(tr, "%s %d\n", "CLONE_PARENT", ta->u.clone_parent.child_pid); + + /* locate child */ + for (i = 0, tr1 = p->task_runs; i < p->task_count; i++, tr1++) + if (pid == tr1->task->pid) + break; + assert(i < p->task_count); + + tr1->parent_pid = getpid(); + + /* all clones are for now forks */ + real_pid = fork(); + assert(real_pid >= 0); + + if (real_pid == 0) { /* child */ + tr1->real_pid = getpid(); + tr1->start_time = tr->start_time + tr->local_time; + tr1->local_time = tr->local_time; + execute_task(tr1); + } + tr1->real_pid = real_pid; + tr->children_count++; +} + +static void execute_clone_child(struct task_run *tr, const struct task_action *ta) +{ + struct playback *p = tr->playback; + + (void)p; + + /* nothing */ + tprintf(tr, "%s %d\n", "CLONE_CHILD", ta->u.clone_child.parent_pid); +} + +static void execute_wait_id(struct task_run *tr, const struct task_action *ta) +{ + struct playback *p = tr->playback; + sem_t *sem_ptr; + u64 t1, dt; + int ret; + + t1 = get_nsecs(); + + tprintf(tr, "%s %lu\n", "WAIT_ID", ta->u.wait_id.id); + + /* wait for someone to wake us up */ + assert(ta->u.wait_id.id < p->sem_count); + sem_ptr = p->semas + ta->u.wait_id.id; + do { + ret = sem_wait(sem_ptr); + } while (ret == -1 && (errno == EINTR || errno == EAGAIN)); + if (ret != 0) { + printf("sem_wait failed (%s wait_id=%lu) = %d errno=%d(%s)\n", tr->name, + ta->u.wait_id.id, ret, errno, strerror(errno)); + assert(0); + } + + dt = get_nsecs() - t1; + + tr->local_time += dt; +} + +static void execute_signal_id(struct task_run *tr, const struct task_action *ta) +{ + struct playback *p = tr->playback; + sem_t *sem_ptr; + int ret; + + tprintf(tr, "%s %lu\n", "SIGNAL_ID", ta->u.signal_id.id); + + /* signal wake up for the other thread */ + assert(ta->u.signal_id.id < p->sem_count); + sem_ptr = p->semas + ta->u.signal_id.id; + do { + ret = sem_post(sem_ptr); + } while (ret == -1 && (errno == EINTR || errno == EAGAIN)); + if (ret == -1) { + printf("sem_post failed (%s signal_id=%lu) = %d errno=%d(%s)\n", tr->name, + ta->u.signal_id.id, ret, errno, strerror(errno)); + assert(0); + } + + /* local time does not advance */ +} + +static const struct task_action_desc action_table[] = { + [TA_END] = { + .id = TA_END, + .name = "END", + }, + [TA_SPAWN] = { + .id = TA_SPAWN, + .name = "SPAWN", + .execute = execute_spawn, + }, + [TA_BURN] = { + .id = TA_BURN, + .name = "BURN", + .execute = execute_burn, + }, + [TA_SLEEP] = { + .id = TA_SLEEP, + .name = "SLEEP", + .execute = execute_sleep, + }, + [TA_EXIT] = { + .id = TA_EXIT, + .name = "EXIT", + .execute = execute_exit, + }, + [TA_CLONE_PARENT] = { + .id = TA_CLONE_PARENT, + .name = "CLONE_PARENT", + .execute = execute_clone_parent, + }, + [TA_CLONE_CHILD] = { + .id = TA_CLONE_CHILD, + .name = "CLONE_CHILD", + .execute = execute_clone_child, + }, + [TA_WAIT_ID] = { + .id = TA_WAIT_ID, + .name = "WAIT_ID", + .execute = execute_wait_id, + }, + [TA_SIGNAL_ID] = { + .id = TA_SIGNAL_ID, + .name = "SIGNAL_ID", + .execute = execute_signal_id, + }, +}; + +static void execute_task(struct task_run *tr) +{ + const struct task *t; + const struct task_action *ta; + struct playback *p; + + t = tr->task; + assert(t != NULL); + + p = tr->playback; + assert(p != NULL); + + /* copy name from the task and inform */ + snprintf(tr->name, sizeof(tr->name) - 1, ":%s-%d", t->name, t->pid); + tr->name[sizeof(tr->name) - 1] = '\0'; + prctl(PR_SET_NAME, tr->name); + + /* skip over spawn */ + while ((ta = task_action_current(tr)) != NULL && ta->action == TA_SPAWN) { + tr->local_time += ta->u.spawn.nsecs; + task_action_advance(tr); + } + + tprintf(tr, "start: name %s, pid %d\n", t->name, t->pid); + + while ((ta = task_action_current(tr)) != NULL) { + /* verify size */ + assert((unsigned int)ta->action < ARRAY_SIZE(action_table)); + (*action_table[ta->action].execute)(tr, ta); + + task_action_advance(tr); + } + + exit(0); +} + +static void dump_task(const struct task *t) +{ + const struct task_action *ta; + u64 ts; + + printf("TASK: name %s, pid %d\n", t->name, t->pid); + + ts = 0; + for (ta = t->actions; ; ta++) { + + if (ta->action == TA_SPAWN) + ts = ta->u.spawn.nsecs; + + printf("%s\n", task_action_str(ts, ta)); + if (ta->action == TA_END) + break; + + if (ta->action == TA_BURN) + ts += ta->u.burn.nsecs; + else if (ta->action == TA_SLEEP) + ts += ta->u.sleep.nsecs; + } +} + +static void generate_task(const struct task *t) +{ + const struct task_action *ta; + + printf("[%s/%d]\n", t->name, t->pid); + + for (ta = t->actions; ; ta++) { + + switch (ta->action) { + case TA_BURN: + printf("\tburn %" PRIu64 "\n", ta->u.burn.nsecs); + break; + case TA_SLEEP: + printf("\tsleep %" PRIu64 "\n", ta->u.sleep.nsecs); + break; + case TA_SPAWN: + printf("\tspawn %" PRIu64 "\n", ta->u.spawn.nsecs); + break; + case TA_CLONE_PARENT: + printf("\tclone-parent %d\n", ta->u.clone_parent.child_pid); + break; + case TA_CLONE_CHILD: + printf("\tclone-child %d\n", ta->u.clone_child.parent_pid); + break; + case TA_WAIT_ID: + printf("\twait-id %lu\n", ta->u.wait_id.id); + break; + case TA_SIGNAL_ID: + printf("\tsignal-id %lu\n", ta->u.signal_id.id); + break; + case TA_EXIT: + printf("\texit %d\n", ta->u.exit.ret); + break; + case TA_END: + printf("\tend\n"); + default: + break; + } + + if (ta->action == TA_END) + break; + } +} + +static void fill_playback_info(struct playback_info *pi, const struct task * const * tasks_array) +{ + const struct task *t; + const struct task * const *tt; + const struct task_action *ta; + unsigned long id, maxid; + int pagesize; + + /* get page size */ + pagesize = getpagesize(); + + /* make sure it's a power of two */ + assert((pagesize & (pagesize - 1)) == 0); + + /* align work area to 16 bytes */ + pi->work_area_size = ALIGN(sizeof(struct playback), 16); + pi->task_off = pi->work_area_size; + + /* count number of tasks (and find maximum futex id) */ + maxid = -1; + pi->task_cnt = 0; + tt = tasks_array; + while ((t = *tt++) != NULL) { + pi->task_cnt++; + for (ta = t->actions; ta != NULL && ta->action != TA_END; ta++) { + if (ta->action == TA_WAIT_ID) + id = ta->u.wait_id.id; + else if (ta->action == TA_SIGNAL_ID) + id = ta->u.signal_id.id; + else + continue; + + if (id > maxid || maxid == (unsigned long)-1) + maxid = id; + } + } + pi->sem_cnt = maxid + 1; + + /* add the size of the task structs */ + pi->work_area_size += ALIGN(sizeof(struct task_run) * pi->task_cnt, 16); + pi->sem_off = pi->work_area_size; + + /* add the size of the semas */ + pi->work_area_size += ALIGN(sizeof(sem_t) * pi->sem_cnt, 16); + + /* now align to the pagesize (we have verified that pagesize is a power of two) */ + pi->work_area_size = ALIGN(pi->work_area_size, pagesize); + + printf("#%d tasks, #%d sems, total work area size %d\n", + pi->task_cnt, pi->sem_cnt, pi->work_area_size); + +} + +static struct playback *playback_create(const struct task * const *tasks_array) +{ + struct playback_info pi; + const struct task * const *tt; + struct task_run *tr; + int i; + unsigned long j; + struct playback *p; + + assert(tasks_array != NULL); + + /* layout playback structure according to inputs */ + fill_playback_info(&pi, tasks_array); + + /* map the shared memory across all processes */ + work_area = mmap(NULL, pi.work_area_size, PROT_READ | PROT_WRITE, MAP_ANON | MAP_SHARED, -1, 0); + assert(work_area != MAP_FAILED); + + memset(work_area, 0, pi.work_area_size); + + playback = work_area; + + p = playback; + memcpy(&p->info, &pi, sizeof(pi)); + + p->origin_time = 0; + p->task_count = pi.task_cnt; + p->task_runs = work_area + pi.task_off; + p->sem_count = pi.sem_cnt; + p->semas = work_area + pi.sem_off; + for (j = 0; j < p->sem_count; j++) + sem_init(&p->semas[j], 1, 0); + + p->spawn_count = 0; + p->exited_count = 0; + p->forked_count = 0; + + for (i = 0, tr = p->task_runs, tt = tasks_array; i < p->task_count; + i++, tr++, tt++) { + + tr->playback = p; + tr->task = *tt; + tr->current_action = tr->task->actions; + tr->id = i; + tr->real_pid = -1; + tr->start_time = 0; + tr->local_time = 0; + } + + return p; +} + +static void playback_destroy(struct playback *p) +{ + assert(p != NULL); + + munmap(p, p->info.work_area_size); +} + +static void playback_run(struct playback *p) +{ + struct task_run *tr, *tr1; + const struct task *t; + const struct task_action *ta; + int i, status; + pid_t real_pid; + uint64_t curr_nsecs; + + p->origin_time = get_nsecs(); + + p->spawn_count = 0; + for (i = 0; i < p->task_count; i++) { + tr = &p->task_runs[i]; + t = tr->task; + if (t->actions->action == TA_SPAWN) { + p->spawn_count++; + continue; + } + + /* don't bother with children */ + if (t->actions->action == TA_CLONE_CHILD) + continue; + + real_pid = fork(); + assert(real_pid >= 0); + + if (real_pid == 0) { + /* fill up for the child */ + tr->start_time = get_nsecs(); + tr->real_pid = getpid(); + execute_task(tr); + } + tr->real_pid = real_pid; + + p->forked_count++; + } + + curr_nsecs = 0; + while (p->spawn_count > 0) { + tr1 = NULL; + for (i = 0; i < p->task_count; i++) { + tr = &p->task_runs[i]; + t = tr->task; + ta = t->actions; + + if (tr->real_pid >= 0 || ta->action != TA_SPAWN) + continue; + + if (tr1 == NULL || tr1->task->actions->u.spawn.nsecs > ta->u.spawn.nsecs) + tr1 = tr; + } + assert(tr1 != NULL); + + tr = tr1; + ta = t->actions; + sleep_nsecs(ta->u.spawn.nsecs - curr_nsecs); + + real_pid = fork(); + assert(real_pid >= 0); + + if (real_pid == 0) { + /* fill up for the child */ + tr->start_time = get_nsecs(); + tr->real_pid = getpid(); + execute_task(tr); + } + tr->real_pid = real_pid; + + curr_nsecs += (ta->u.spawn.nsecs - curr_nsecs); + p->spawn_count--; + p->forked_count++; + } + + while (p->forked_count > 0) { + real_pid = wait(&status); + if (real_pid < 0) { + fprintf(stderr, "panto-perf: Wait failed: %s:%d\n", __func__, __LINE__); + exit(EXIT_FAILURE); + } + if (!WIFEXITED(status)) + continue; + + for (i = 0; i < p->task_count; i++) { + tr = &p->task_runs[i]; + t = tr->task; + if (tr->real_pid == real_pid) + break; + } + + assert(i < p->task_count); + tr->real_pid = -1; + p->forked_count--; + } +} + +static const char * const spr_replay_usage[] = { + "perf sched spr-replay [<options>]", + NULL +}; + +static int parse_match_option(const struct option *opt, const char *str, + int unset __used) +{ + struct list_head *lh = (struct list_head *)opt->value; + struct match_list_entry *mle = NULL; + const char *s; + int ret; + + mle = zalloc(sizeof(*mle)); + if (mle == NULL) + goto error; + + INIT_LIST_HEAD(&mle->node); + mle->pid = -1; + + /* first check if everything is a digit (pid) */ + for (s = str; *s != '\0'; s++) + if (!isdigit(*s)) + break; + if (*s == '\0') { + mle->type = MT_PID; + mle->pid = atoi(str); + goto add; + } + + /* name is either a name, or a pattern */ + if (*str != '~') { + mle->type = MT_NAME; + mle->name = strdup(str); + if (mle->name == NULL) + goto error; + goto add; + } + + /* it is a pattern after all of the form '~regex' */ + ret = regcomp(&mle->regex, str + 1, REG_EXTENDED); + if (ret) + goto error; + mle->type = MT_REGEX; +add: + list_add_tail(&mle->node, lh); + return 0; + +error: + free(mle); /* NULL is properly handled */ + return -1; +} + +static const struct option spr_replay_options[] = { + OPT_CALLBACK('s', "select", &select_list, "select", + "Number pid, 'name | ~regex' matches comm name", + parse_match_option), + OPT_CALLBACK('x', "exclude", &exclude_list, "exclude", + "Number pid, 'name | ~regex' matches comm name", + parse_match_option), + OPT_UINTEGER('r', "repeat", &replay_repeat, + "repeat the workload replay N times (-1: infinite)"), + OPT_INCR('v', "verbose", &verbose, + "be more verbose (show symbol address, etc)"), + OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, + "dump raw trace in ASCII"), + OPT_BOOLEAN('p', "preserve-time", &preserve_time, + "Preserve time (do not erase sleeps while in running state)"), + OPT_BOOLEAN('n', "dry-run", &dry_run, + "do not execute"), + OPT_INCR('d', "debug", &debug, + "be more verbose"), + OPT_BOOLEAN('g', "generate", &generate, + "generate an spr program at stdout"), + OPT_STRING('f', "spr-filename", &spr_filename, "file", + "spr file name (instead of trace)"), + OPT_BOOLEAN('l', "list", &spr_list, + "list tasks & pids"), + OPT_BOOLEAN('L', "list-selected", &spr_select_list, + "list selected tasks & pids"), + OPT_BOOLEAN('b', "bogoburn", &bogoburn, + "burn time using a bogo-mips like busy loop"), + OPT_U64('B', "bogoloops", &bogoloops, + "set bogoloops value directly without re-calculating"), + OPT_END() +}; + +static const char *task_state_str(unsigned int task_state) +{ + static const char * const task_state_array[] = { + "R (running)", /* 0 */ + "S (sleeping)", /* 1 */ + "D (disk sleep)", /* 2 */ + "T (stopped)", /* 4 */ + "t (tracing stop)", /* 8 */ + "Z (zombie)", /* 16 */ + "X (dead)", /* 32 */ + "x (dead)", /* 64 */ + "K (wakekill)", /* 128 */ + "W (waking)", /* 256 */ + }; + const char * const *p = &task_state_array[0]; + unsigned int state = task_state & 511; + + while (state) { + p++; + state >>= 1; + } + return *p; +} + +static void calculate_replay_start_time(void) +{ + struct sched_atom *atom; + struct task_desc *task; + unsigned long i; + + replay_start_time = (u64)-1; /* maximum possible */ + for (i = 0; i < nr_tasks; i++) { + task = tasks[i]; + + if (task->nr_events == 0) + continue; + + atom = task->atoms[0]; + + if (atom->timestamp < replay_start_time) + replay_start_time = atom->timestamp; + + } + if (replay_start_time == (u64)-1) + replay_start_time = 0; +} + +static void calculate_replay_end_time(void) +{ + struct sched_atom *atom; + struct task_desc *task; + unsigned long i; + u64 ts_end; + + replay_end_time = 0; /* minimum */ + for (i = 0; i < nr_tasks; i++) { + task = tasks[i]; + + if (task->nr_events == 0) + continue; + + atom = task->atoms[task->nr_events - 1]; + + ts_end = atom->timestamp; + if (atom->type == SCHED_EVENT_RUN) + ts_end += atom->duration; + + if (ts_end > replay_end_time) + replay_end_time = ts_end; + } +} + +static const char *sched_atom_str(const struct sched_atom *atom) +{ + static char buf[BUFSIZ]; + char *s, *e; + unsigned int i; + + if (replay_start_time == 0) + calculate_replay_start_time(); + + if (replay_end_time == 0) + calculate_replay_end_time(); + + s = buf; + e = &buf[ARRAY_SIZE(buf)]; + + snprintf(s, e - s, "\t%014" PRIu64 " ", atom->timestamp - replay_start_time); + e[-1] = '\0'; + s += strlen(s); + + switch (atom->type) { + case SCHED_EVENT_RUN: + snprintf(s, e - s, "%-10s%014" PRIu64, "RUN", + atom->duration); + e[-1] = '\0'; + s += strlen(s); + break; + case SCHED_EVENT_SLEEP: + snprintf(s, e - s, "%-10s %s %d", "SLEEP", + task_state_str(atom->task_state), + atom->waker_count); + e[-1] = '\0'; + s += strlen(s); + for (i = 0; i < atom->waker_count; i++) { + snprintf(s, e - s, " [%ld@%014" PRIu64 "]", + atom->wakers[i]->pid, + atom->waker_events[i]->timestamp - replay_start_time); + e[-1] = '\0'; + s += strlen(s); + } + break; + case SCHED_EVENT_WAKEUP: + snprintf(s, e - s, "%-10s", "WAKEUP"); + e[-1] = '\0'; + s += strlen(s); + if (atom->wakee != NULL && atom->wakee_event != NULL) { + snprintf(s, e - s, " [%ld@%014" PRIu64 "]", + atom->wakee->pid, + atom->wakee_event->timestamp - replay_start_time); + e[-1] = '\0'; + s += strlen(s); + } + break; + case SCHED_EVENT_MIGRATION: + snprintf(s, e - s, "%-10s", "MIGRATION"); + e[-1] = '\0'; + s += strlen(s); + break; + case SCHED_EVENT_EXIT: + snprintf(s, e - s, "%-10s", "EXIT"); + e[-1] = '\0'; + s += strlen(s); + break; + case SCHED_EVENT_FORK_PARENT: + snprintf(s, e - s, "%-10s %d", "FORK_PARENT", atom->pid); + e[-1] = '\0'; + s += strlen(s); + break; + case SCHED_EVENT_FORK_CHILD: + snprintf(s, e - s, "%-10s %d", "FORK_CHILD", atom->pid); + e[-1] = '\0'; + s += strlen(s); + break; + default: + /* GCC whines about missing default case without this one */ + break; + } + + if (debug > 2 && atom->msg) { + snprintf(s, e - s, " | %s", atom->msg); + e[-1] = '\0'; + s += strlen(s); + } + + return buf; +} + +static void dump_sched_atom_task(struct task_desc *task) +{ + struct sched_atom *atom; + unsigned long j; + unsigned int exited; + + exited = 0; + for (j = 0; j < task->nr_events && !exited; j++) { + + atom = task->atoms[j]; + + printf("%s\n", sched_atom_str(atom)); + + /* exited = atom->type == SCHED_EVENT_EXIT; */ + } +} + +static const char *task_action_str(u64 local_ts, const struct task_action *ta) +{ + static char buf[BUFSIZ]; + char *s, *e; + + s = buf; + e = &buf[ARRAY_SIZE(buf)]; + + snprintf(s, e - s, "[%014" PRIu64 "] %-12s", local_ts, action_table[ta->action].name); + e[-1] = '\0'; + s += strlen(s); + + switch (ta->action) { + case TA_BURN: + snprintf(s, e - s, "%014" PRIu64 "ns", ta->u.burn.nsecs); + break; + case TA_SLEEP: + snprintf(s, e - s, "%014" PRIu64 "ns", ta->u.sleep.nsecs); + break; + case TA_SPAWN: + snprintf(s, e - s, "%014" PRIu64 "ns", ta->u.spawn.nsecs); + break; + case TA_CLONE_PARENT: + snprintf(s, e - s, "%d", ta->u.clone_parent.child_pid); + break; + case TA_CLONE_CHILD: + snprintf(s, e - s, "%d", ta->u.clone_child.parent_pid); + break; + case TA_WAIT_ID: + snprintf(s, e - s, "%lu", ta->u.wait_id.id); + break; + case TA_SIGNAL_ID: + snprintf(s, e - s, "%lu", ta->u.signal_id.id); + break; + case TA_EXIT: + case TA_END: + *s = '\0'; + break; + default: + /* GCC is whining about missing default case */ + break; + } + e[-1] = '\0'; + s += strlen(s); + + return buf; +} + +static struct task *generate_spr_program(struct task_desc *task, unsigned long *map_id_fwd, unsigned long *next_opt_id) +{ + struct sched_atom *atom, *atom_last; + unsigned long eventnr; + unsigned int exited; + unsigned int blocked; + unsigned int last_blocked; + unsigned int last_was_runnable; + unsigned int selected_wakers; + unsigned int exited_wakers; + struct task *t; + struct task_action *ta_alloc; + int ta_count, ta_max; + int task_state; + u64 replay_ts, replay_ts_new; + u64 local_ts; + u64 delta_ts; + u64 start_run_ts; + int pending_count; + struct task_action ta_work; + unsigned int i; + u64 burn_run_accum; + + BUG_ON(task == NULL); + + BUG_ON(task->nr_events == 0); + + if (replay_start_time == 0) + calculate_replay_start_time(); + + if (replay_end_time == 0) + calculate_replay_end_time(); + + t = zalloc(sizeof(*t)); + BUG_ON(t == NULL); + t->pid = task->pid; + t->name = strdup(task->comm); + BUG_ON(t->name == NULL); + + ta_count = 0; + ta_max = 256; /* starting point */ + ta_alloc = zalloc(sizeof(*ta_alloc) * ta_max); + BUG_ON(ta_alloc == NULL); + t->actions = ta_alloc; + +#undef APPEND_TA_WORK +#define APPEND_TA_WORK() \ + do { \ + if (ta_count >= ta_max) { \ + ta_alloc = realloc(ta_alloc, ta_max * 2 * (sizeof(*t->actions))); \ + BUG_ON(ta_alloc == NULL); \ + ta_max *= 2; \ + t->actions = ta_alloc; \ + } \ + ta_alloc[ta_count++] = ta_work; \ + if (debug > 2) \ + printf("\t\t%4d: GEN: %4d: [%012" PRIu64 "] %s\n", __LINE__, \ + pending_count, start_run_ts, \ + task_action_str(start_run_ts, &ta_work)); \ + } while (0) + +#undef GET_OPTIMIZED_ID +#define GET_OPTIMIZED_ID(x) \ + ({ \ + unsigned long _x = (x); \ + BUG_ON(_x >= (next_wake_id + 1)); \ + if (map_id_fwd[_x] == 0) { \ + map_id_fwd[_x] = ++(*next_opt_id); \ + BUG_ON(*next_opt_id == 0); \ + } \ + map_id_fwd[_x]; \ + }) + + exited = 0; + replay_ts = 0; + local_ts = 0; + task_state = -1; + atom = NULL; + start_run_ts = 0; + pending_count = 0; + last_was_runnable = 0; + last_blocked = 0; + burn_run_accum = 0; + + memset(&ta_work, 0, sizeof(ta_work)); + ta_work.action = TA_END; + + for (eventnr = 0; eventnr < task->nr_events && !exited; eventnr++) { + + atom_last = atom; + + atom = task->atoms[eventnr]; + + exited = atom->type == SCHED_EVENT_EXIT; + + blocked = 0; + selected_wakers = 0; + exited_wakers = 0; + + if (atom->type == SCHED_EVENT_SLEEP) { + + /* by default we're blocked */ + blocked = 1; + + /* but if we don't preserve time and we're running... */ + if (!preserve_time && (atom->task_state & 511) == 0) { + /* check for exited wakers */ + selected_wakers = 0; + exited_wakers = 0; + for (i = 0; i < atom->waker_count; i++) { + if (atom->wakers[i]->selected) { + selected_wakers++; + if (atom->waker_events[i]->exited) + exited_wakers++; + } + } + blocked = selected_wakers > 0 && selected_wakers > exited_wakers; + } + } + + replay_ts_new = atom->timestamp - replay_start_time; + + /* first one */ + if (eventnr == 0) + replay_ts = replay_ts_new; + + if (last_was_runnable && !preserve_time) /* we don't count the time we were runnable */ + delta_ts = 0; + else + delta_ts = replay_ts_new - replay_ts; + last_was_runnable = 0; + replay_ts = replay_ts_new; + + if (debug > 2) + printf("[%012" PRIu64 " %012" PRIu64 "] %s\n", + replay_ts, delta_ts, sched_atom_str(atom)); + + /* first one is special */ + if (eventnr == 0) { + if (atom->type == SCHED_EVENT_EXIT) { + ta_work.action = TA_SLEEP; + ta_work.u.sleep.nsecs = 0; + start_run_ts = 0; + pending_count = 1; + } else if (atom->type != SCHED_EVENT_FORK_CHILD) { + ta_work.action = TA_BURN; + ta_work.u.burn.nsecs = replay_ts; + start_run_ts = 0; + pending_count = 1; + burn_run_accum = 0; + } + } + + if (eventnr > 0 && pending_count > 0 && blocked != last_blocked) { + + BUG_ON(ta_work.action != TA_BURN && ta_work.action != TA_SLEEP); + + if (ta_work.action == TA_BURN) { + ta_work.u.burn.nsecs += burn_run_accum + delta_ts; + if (ta_work.u.burn.nsecs > 0) + APPEND_TA_WORK(); + } else if (ta_work.action == TA_SLEEP) { + ta_work.u.sleep.nsecs += delta_ts; + if (ta_work.u.sleep.nsecs > 0) + APPEND_TA_WORK(); + } + start_run_ts = local_ts; + pending_count = 0; + } + + switch (atom->type) { + case SCHED_EVENT_RUN: + task_state = 0; + if (pending_count == 0) { + ta_work.action = TA_BURN; + if (atom->duration < delta_ts) + start_run_ts = local_ts - atom->duration; + else + start_run_ts = local_ts - delta_ts; + } else { + BUG_ON(ta_work.action != TA_BURN); + } + ta_work.u.burn.nsecs = 0; /* always reset */ + burn_run_accum += atom->duration; + pending_count++; + break; + + case SCHED_EVENT_SLEEP: + task_state = atom->task_state & 511; + + /* set when we were running, and switched out */ + last_was_runnable = task_state == 0; + + if (pending_count > 0) { + BUG_ON(ta_work.action != TA_BURN); + ta_work.u.burn.nsecs += delta_ts; + if (blocked) { + ta_work.u.burn.nsecs += burn_run_accum; + if (ta_work.u.burn.nsecs > 0) + APPEND_TA_WORK(); + pending_count = 0; + burn_run_accum = 0; + } + } + + /* if we're not blocked, we skip if we're still running */ + if (!blocked) + break; + + if (selected_wakers > 0 && selected_wakers > exited_wakers) { + ta_work.action = TA_WAIT_ID; + ta_work.u.wait_id.id = GET_OPTIMIZED_ID(atom->wake_id); + start_run_ts = local_ts; + APPEND_TA_WORK(); + pending_count = 0; + } else { + ta_work.action = TA_SLEEP; + ta_work.u.sleep.nsecs = 0; + start_run_ts = local_ts; + pending_count = 1; + } + + break; + + case SCHED_EVENT_WAKEUP: + /* we were blocked? */ + if (pending_count > 0 && last_blocked) { + BUG_ON(ta_work.action != TA_SLEEP); + ta_work.u.sleep.nsecs += delta_ts; + if (ta_work.u.sleep.nsecs > 0) + APPEND_TA_WORK(); + start_run_ts = local_ts; + pending_count = 0; + } + + /* no target; ignore */ + if (atom->wakee == NULL || atom->wakee_event == NULL || + !atom->wakee->selected) { + if (pending_count > 0) { + BUG_ON(ta_work.action != TA_BURN); + ta_work.u.burn.nsecs += delta_ts; + } else { + ta_work.action = TA_BURN; + ta_work.u.burn.nsecs = 0; + start_run_ts = local_ts; + burn_run_accum = 0; + } + pending_count++; + break; + } + + /* target; need to finish the run */ + if (pending_count > 0) { + BUG_ON(ta_work.action != TA_BURN); + ta_work.u.burn.nsecs += burn_run_accum + delta_ts; + if (ta_work.u.burn.nsecs > 0) + APPEND_TA_WORK(); + pending_count = 0; + } + + ta_work.action = TA_SIGNAL_ID; + ta_work.u.signal_id.id = GET_OPTIMIZED_ID(atom->wakee_event->wake_id); + start_run_ts = local_ts; + APPEND_TA_WORK(); + + /* and we're running */ + ta_work.action = TA_BURN; + ta_work.u.burn.nsecs = 0; + start_run_ts = local_ts; + pending_count = 1; + burn_run_accum = 0; + + break; + + case SCHED_EVENT_MIGRATION: + break; + case SCHED_EVENT_EXIT: + if (pending_count > 0) { + if (ta_work.action == TA_BURN) { + ta_work.u.burn.nsecs += burn_run_accum + delta_ts; + if (ta_work.u.burn.nsecs > 0) + APPEND_TA_WORK(); + } else if (ta_work.action == TA_SLEEP) { + ta_work.u.sleep.nsecs += delta_ts; + if (ta_work.u.sleep.nsecs > 0) + APPEND_TA_WORK(); + } else { + BUG_ON(ta_work.action != TA_BURN && + ta_work.action != TA_SLEEP); + } + pending_count = 0; + } + + ta_work.action = TA_EXIT; + ta_work.u.exit.ret = 0; + start_run_ts = local_ts; + APPEND_TA_WORK(); + break; + case SCHED_EVENT_FORK_PARENT: + BUG_ON(atom->child == NULL); + + if (pending_count > 0) { + BUG_ON(ta_work.action != TA_BURN); + ta_work.u.burn.nsecs += delta_ts; + + if (atom->child->selected) { + ta_work.u.burn.nsecs += burn_run_accum; + if (ta_work.u.burn.nsecs > 0) + APPEND_TA_WORK(); + pending_count = 0; + } else + pending_count++; + } + + if (!atom->child->selected) { + if (verbose) + printf("Forking parent (%s/%ld) but child (%s/%ld) not selected; ignoring\n", + task->comm, task->pid, atom->child->comm, atom->child->pid); + break; + } + + ta_work.action = TA_CLONE_PARENT; + ta_work.u.clone_parent.child_pid = atom->pid; + start_run_ts = local_ts; + APPEND_TA_WORK(); + + /* and we're running */ + ta_work.action = TA_BURN; + ta_work.u.burn.nsecs = 0; + start_run_ts = local_ts; + pending_count = 1; + burn_run_accum = 0; + + task_state = 0; /* running */ + + break; + case SCHED_EVENT_FORK_CHILD: + BUG_ON(eventnr != 0); /* this _must_ be the first one */ + BUG_ON(atom->parent == NULL); + + if (!atom->parent->selected) { + if (verbose) + printf("Forking child (%s/%ld) but parent (%s/%ld) not selected; converting to run\n", + task->comm, task->pid, atom->parent->comm, atom->parent->pid); + + task_state = 0; + ta_work.action = TA_BURN; + ta_work.u.burn.nsecs = 0; + start_run_ts = local_ts; + pending_count = 1; + burn_run_accum = 0; + break; + } + + ta_work.action = TA_CLONE_CHILD; + ta_work.u.clone_child.parent_pid = atom->pid; + start_run_ts = local_ts; + APPEND_TA_WORK(); + + /* and we're running */ + ta_work.action = TA_BURN; + ta_work.u.burn.nsecs = 0; + start_run_ts = local_ts; + pending_count = 1; + burn_run_accum = 0; + + task_state = 0; /* running */ + + break; + default: + break; + } + + if (preserve_time || !last_was_runnable) + local_ts += delta_ts; + + last_blocked = blocked; + } + + /* stop any runs */ + if (pending_count > 0) { + + BUG_ON(ta_work.action != TA_BURN && ta_work.action != TA_SLEEP); + + /* if the process was running extend till end */ + delta_ts = !exited ? (replay_end_time - (replay_ts + replay_start_time)) : 0; + + if (ta_work.action == TA_BURN) { + ta_work.u.burn.nsecs += burn_run_accum + delta_ts; + if (ta_work.u.burn.nsecs > 0) + APPEND_TA_WORK(); + } else if (ta_work.action == TA_SLEEP) { + ta_work.u.sleep.nsecs += delta_ts; + if (ta_work.u.sleep.nsecs > 0) + APPEND_TA_WORK(); + } + pending_count = 0; + } + + ta_work.action = TA_END; + ta_work.u.exit.ret = 0; + start_run_ts = local_ts; + APPEND_TA_WORK(); + + return t; +} + +static int read_spr_program(const char *file, struct task ***ttt) +{ + FILE *fp; + char buf[BUFSIZ]; + char orig_buf[BUFSIZ]; + char *name; + struct task *t, **tt; + struct task_action ta_work, *ta_alloc; + int task_count, task_max; + int ta_count, ta_max; + char *s, *e, *start, *end; + int line, pid; + +#undef APPEND_TASK +#define APPEND_TASK() \ + do { \ + if (task_count >= task_max) { \ + tt = realloc(tt, (task_max + 32) * (sizeof(*tt))); \ + BUG_ON(tt == NULL); \ + task_max += 32; \ + } \ + tt[task_count++] = t; \ + } while (0) + +#undef APPEND_TA_WORK +#define APPEND_TA_WORK() \ + do { \ + if (ta_count >= ta_max) { \ + ta_alloc = realloc(ta_alloc, (ta_max + 4096) * (sizeof(*ta_alloc))); \ + BUG_ON(ta_alloc == NULL); \ + ta_max += 4096; \ + } \ + ta_alloc[ta_count++] = ta_work; \ + } while (0) + + + fp = fopen(file, "ra"); + if (fp == NULL) { + fprintf(stderr, "Could not open file '%s'\n", file); + return -1; + } + + task_count = 0; + task_max = 0; + tt = NULL; + ta_count = 0; + ta_max = 0; + ta_alloc = NULL; + pid = -1; + name = NULL; + + t = NULL; + line = 0; + while (fgets(buf, sizeof(buf) - 1, fp) != NULL) { + line++; + + buf[sizeof(buf) - 1] = '\0'; + + /* remove trailing newline */ + s = strrchr(buf, '\n'); + if (s != NULL) + *s = '\0'; + + strcpy(orig_buf, buf); + + /* remove comments */ + s = strchr(buf, '#'); + if (s != NULL) + *s = '\0'; + + /* remove trailing spaces */ + s = buf + strlen(buf) - 1; + while (s > buf && isspace(*s)) + *s-- = '\0'; + + /* skip over spaces in the front */ + s = buf; + while (isspace(*s)) + s++; + start = s; + end = s + strlen(s); + + /* empty line */ + if (start >= end - 1) + continue; + + /* waiting for task */ + if (t == NULL) { + if (*start != '[' || end[-1] != ']') + goto syntax_error; + s = start + 1; + e = strrchr(s, '/'); + if (e == NULL) + goto syntax_error; + if (e - s < 1) + goto syntax_error; + + name = malloc(e - s + 1); + BUG_ON(name == NULL); + + memcpy(name, s, e - s); + name[e - s] = '\0'; + + pid = atoi(e + 1); + if (pid < 0) + goto syntax_error; + + ta_count = 0; + ta_max = 0; + ta_alloc = NULL; + + /* allocate but don't do anything with it */ + t = malloc(sizeof(*t)); + BUG_ON(t == NULL); + + } else { + s = start; + e = start; + while (!isspace(*e)) + e++; + *e++ = '\0'; + if (strcmp(s, "burn") == 0) { + ta_work.action = TA_BURN; + ta_work.u.burn.nsecs = strtoull(e, NULL, 10); + } else if (strcmp(s, "sleep") == 0) { + ta_work.action = TA_SLEEP; + ta_work.u.sleep.nsecs = strtoull(e, NULL, 10); + } else if (strcmp(s, "spawn") == 0) { + ta_work.action = TA_SPAWN; + ta_work.u.spawn.nsecs = strtoull(e, NULL, 10); + } else if (strcmp(s, "clone-parent") == 0) { + ta_work.action = TA_CLONE_PARENT; + ta_work.u.clone_parent.child_pid = (int)strtoul(e, NULL, 10); + } else if (strcmp(s, "clone-child") == 0) { + ta_work.action = TA_CLONE_CHILD; + ta_work.u.clone_child.parent_pid = (int)strtoul(e, NULL, 10); + } else if (strcmp(s, "wait-id") == 0) { + ta_work.action = TA_WAIT_ID; + ta_work.u.wait_id.id = strtoul(e, NULL, 10); + } else if (strcmp(s, "signal-id") == 0) { + ta_work.action = TA_SIGNAL_ID; + ta_work.u.signal_id.id = strtoul(e, NULL, 10); + } else if (strcmp(s, "exit") == 0) { + ta_work.action = TA_EXIT; + ta_work.u.exit.ret = (int)strtoul(e, NULL, 10); + } else if (strcmp(s, "end") == 0) { + ta_work.action = TA_END; + } else + goto syntax_error; + + APPEND_TA_WORK(); + + if (ta_work.action == TA_END) { + BUG_ON(t == NULL); + BUG_ON(name == NULL); + BUG_ON(pid < 0); + + if (ta_count == 0) + goto syntax_error; + + t->name = name; + t->pid = pid; + t->actions = ta_alloc; + + APPEND_TASK(); + t = NULL; + } + } + } + t = NULL; + APPEND_TASK(); + + fclose(fp); + + /* no program */ + if (tt == NULL) + return -1; + + *ttt = tt; + return 0; + +syntax_error: + fprintf(stderr, "syntax error at line %d: %s\n", line, orig_buf); + *ttt = NULL; + fclose(fp); + return -1; +} + +static void __cmd_spr_replay(void) +{ + struct task_desc *task; + struct task **tt; + const struct task * const *task_array; + int matched, excluded; + unsigned long i, j; + int selected_task_count; + struct playback *p; + unsigned long *map_id_fwd; + unsigned long next_opt_id; + struct sched_atom *atom; + u64 run_nsecs; + int ret; + + /* mark that we're not the standard replay */ + spr_replay = true; + + if (!dry_run && !(spr_list || spr_select_list)) { + calibrate_run_measurement_overhead(); + calibrate_sleep_measurement_overhead(); + + test_calibrations(); + } + + if (spr_filename == NULL) { + + read_events(true, NULL); + + if (debug > 0) { + printf("nr_run_events: %ld\n", nr_run_events); + printf("nr_sleep_events: %ld\n", nr_sleep_events); + printf("nr_wakeup_events: %ld\n", nr_wakeup_events); + + if (targetless_wakeups) + printf("target-less wakeups: %ld\n", targetless_wakeups); + if (multitarget_wakeups) + printf("multi-target wakeups: %ld\n", multitarget_wakeups); + if (nr_run_events_optimized) + printf("run atoms optimized: %ld\n", + nr_run_events_optimized); + } + + calculate_replay_start_time(); + calculate_replay_end_time(); + + selected_task_count = 0; + for (i = 0; i < nr_tasks; i++) { + task = tasks[i]; + + matched = match_task(&select_list, task->comm, task->pid, 1); + excluded = match_task(&exclude_list, task->comm, task->pid, 0); + + task->selected = matched && !excluded; + selected_task_count += task->selected; + } + + tt = zalloc((selected_task_count + 1) * sizeof(*tt)); + BUG_ON(tt == NULL); + + if (spr_list || spr_select_list) { + for (j = 0; j < nr_tasks; j++) { + task = tasks[j]; + + /* task with no events, doesn't show up */ + if (task->nr_events == 0) + continue; + + /* we have a match list */ + if (spr_select_list && !task->selected) + continue; + + run_nsecs = 0; + for (i = 0; i < task->nr_events; i++) { + atom = task->atoms[i]; + if (atom->type == SCHED_EVENT_EXIT) + break; + if (atom->type == SCHED_EVENT_RUN) + run_nsecs += atom->duration; + } + + printf("[%s/%ld] %c R:%" PRIu64 "\n", + task->comm, task->pid, + task->selected ? 'S' : '-', + run_nsecs); + } + } + + if (debug > 1) { + printf("Dump of scheduling atoms\n"); + for (i = 0; i < nr_tasks; i++) { + task = tasks[i]; + + if (task->nr_events == 0 || !task->selected) + continue; + + printf("task %6ld (%20s:%10ld), nr_events: %ld\n", + task->nr, task->comm, task->pid, task->nr_events); + + dump_sched_atom_task(task); + printf("\n"); + } + printf("\n"); + } + + /* map of a wake id to an optimized wake id */ + map_id_fwd = zalloc((next_wake_id + 1) * sizeof(*map_id_fwd)); + BUG_ON(map_id_fwd == NULL); + + next_opt_id = 0; + + if (verbose) + printf("Generating replay program\n"); + + for (i = 0, j = 0; i < nr_tasks; i++) { + task = tasks[i]; + + if (task->nr_events == 0 || !task->selected) + continue; + + if (verbose) + printf("task %6ld (%20s:%10ld), nr_events: %ld\n", + task->nr, task->comm, task->pid, task->nr_events); + + tt[j] = generate_spr_program(task, map_id_fwd, &next_opt_id); + BUG_ON(tt[j] == NULL); + + if (debug > 1) { + dump_task(tt[j]); + printf("\n"); + } + + j++; + } + + if (generate) { + for (i = 0; tt[i] != NULL; i++) + generate_task(tt[i]); + } + + free(map_id_fwd); + + } else { + ret = read_spr_program(spr_filename, &tt); + BUG_ON(ret != 0); + + if (generate) { + for (i = 0; tt[i] != NULL; i++) + generate_task(tt[i]); + } + + } + + if (!dry_run) { + + if (bogoburn) { + + if (bogoloops == 0) + calculate_bogoloops_value(); + + if (debug > 0) + printf("bogoloops at %" PRIu64 "\n", bogoloops); + } + + task_array = (void *)tt; + + p = playback_create(task_array); + BUG_ON(p == NULL); + + playback_set_debug_level(p, debug); + + if (verbose) + printf("Running...\n"); + playback_run(p); + + if (verbose) + printf("Done...\n"); + + playback_destroy(p); + } + + /* free */ + for (i = 0; tt[i] != NULL; i++) { + free((void *)tt[i]->actions); + free((void *)tt[i]->name); + free(tt[i]); + } + free(tt); +} + int cmd_sched(int argc, const char **argv, const char *prefix __used) { argc = parse_options(argc, argv, sched_options, sched_usage, @@ -1920,6 +4224,14 @@ int cmd_sched(int argc, const char **argv, const char *prefix __used) usage_with_options(replay_usage, replay_options); } __cmd_replay(); + } else if (!strncmp(argv[0], "spr-rep", 7)) { + trace_handler = &replay_ops; + if (argc) { + argc = parse_options(argc, argv, spr_replay_options, spr_replay_usage, 0); + if (argc) + usage_with_options(spr_replay_usage, spr_replay_options); + } + __cmd_spr_replay(); } else { usage_with_options(sched_usage, sched_options); } -- 1.7.1

11 years, 10 months

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

linaro-dev May 2012