linaro-dev January 2013

linaro-dev@lists.linaro.org

82 participants
82 discussions

Re: [PATCH] Genericise CPUIdle for ARM big.LITTLE

by Lorenzo Pieralisi

On Tue, Jan 15, 2013 at 06:17:37PM +0000, Mark Hambleton wrote: > Hi Lorenzo, > > > > +obj-$(CONFIG_BIG_LITTLE) += arm_big_little.o > > There is nothing big.LITTLE specific in all of this, so arm_idle.c would > >be better. > > I figured that because the current version calls into the big.little platform power framework (bL_entry.c) and makes calls into that framework that this wasn't totally generic and is dependant upon that code. The version of the cpuidle driver won't build unless that code is built in, so I still think this is more appropriate naming, I could call it bL_* but I suspect someone will object to that upstream because of the mixed case. Well, calling it big.LITTLE, bL or whatever else describing a big and a LITTLE is wrong IMHO because it makes people think this a driver for big.LITTLE ARM platforms. And it is not, if we ever manage to make it generic. > > > > +static int bl_enter_powerdown(struct cpuidle_device *dev, > > > + struct cpuidle_driver *drv, int idx); > > > + > > > +static struct cpuidle_state bl_cpuidle_set[] __initdata = { > > > + [0] = { > > > + .enter = bl_cpuidle_simple_enter, > > > + .exit_latency = 1, > > > + .target_residency = 1, > > > + .power_usage = UINT_MAX, > > > + .flags = CPUIDLE_FLAG_TIME_VALID, > > > + .name = "WFI", > > > + .desc = "ARM WFI", > > > + }, > > > + [1] = { > > > + .enter = bl_enter_powerdown, > > > + .exit_latency = 300, > > > + .target_residency = 1000, > > > + .flags = CPUIDLE_FLAG_TIME_VALID, > > > + .name = "C1", > > > + .desc = "ARM power down", > > > + }, > > > +}; > > > First problem, these states are not standard, in particular > > target_residency, exit_latency and the enter function. We could > > have a stab at initializing them from DT (or provide a way to probe > > them from HW like Intel does), that's the only way to make this stuff > > generic. > > Ok, I will extend with a DT node to retrieve the states, it means I no longer need the generic compatibility node. Extending the states require a standard and we need to adapt kernel code to comply and carry out required actions, it is not just a matter of adding some latencies in DT bindings. I fail to understand why we want to make this code generic NOW, ARM kernel is not ready for that and to be honest I do not see why it has to be done now. > > > > +struct cpuidle_driver bl_idle_driver = { > > > + .name = "bl_idle", > > > + .owner = THIS_MODULE, > > > + .safe_state_index = 0 > > > +}; > > > Different clusters have different drivers. This can be solved with > > DT and a couple of bindings. > > Not sure I understand this one, I am looking at the latest 3.8-rc3 tree and there is only 1 driver registered. Do I need to look at another version? A cpuidle driver basically is an array of C-states. I am merging code that defines different states so different drivers for A15 and A7 clusters. cpuidle framework now supports multiple drivers, that's what we want for big.LITTLE platforms (but the generic idle driver should support as many drivers as required by DT bindings, which is 1 for symmetric systems. > > > > + bL_set_entry_vector(cpu, cluster, cpu_resume); > > > This might not be what we want. We might want to resume from an > > address different from cpu_resume (although it might well be > > standardized since the power API allows to carry out platform > > specific operations before jumping back to the kernel so I reckon > > this can be and will be generic). > > Ok, this was a patch to the linaro tree, it wasn't intended as a re-write of the driver. I still do not understand why this file move has to be done now. > > > > + ret = cpu_suspend((unsigned long) dev, bl_powerdown_finisher); > > This is where things start getting a little more complicated. If we do > > not know what the power state implies (e.g. L2 retained or not, global > > timer or emulated local timers, GIC state lost in low-power or not) we > > end up saving/restoring state or executing actions that may be useless. > > > > In Intel world executing "mwait" does the trick for all C-states. Here > > we need to define what a state is and what it implies to carry out the > > required actions in a generic way, provided we can pull it off. > > > > PSCI and the power API are a MAJOR step in this direction. > > Same again, this was based on what was already there... if you are re-working for PSCI please just treat this as a request to be more generic where possible. To be generic we need multiple platforms and understand what are they doing with the respective C-states. I will update the code for PSCI. > > > This should be improved to cater for multiple drivers but that's not my > > major concern. > Ok, as I said this was just a move of tc2-cpuidle.c from mach-vexpress into drivers/cpuidle, if that driver is deficient then I suspect someone else is looking to fix it? The driver is not deficient it is just a perfectly working cpuidle driver, TC2 specific, as tens of other cpuidle ARM drivers. It should be improved and made generic in due course, when we have a standard for C-states and the kernel save/restore and cache flushing mechanism is coded to follow suit. Lorenzo

12 years, 1 month

ssh on Foundation v8 model filesystem

by Tim Northover

Hi there, We're hoping to use the foundation model and its filesystem for our LLVM AArch64 testing, but I'm having some difficulties getting reliable ssh access. I'm running the suggested kernel image and filesystem with options: $ Foundation_v8pkg/Foundation_v8 --image img-foundation.axf --block-device vexpress64-openembedded_sdk-armv8_20121213-134.img --network=nat --network- nat-ports=8022=22 However, when I login to the system: $ ssh -p 8022 root@localhost I can execute commands reasonably, but the session appears to hang when I exit. I can, of course, exit with "<Ret>~.", but that leaves the process running on the model. Similar problems occur with scp, and even switching to bridged networking doesn't resolve the issue. Does anyone have any experience with this or a solution? Cheers. Tim. -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.

12 years, 1 month

RE: [PATCH] Genericise CPUIdle for ARM big.LITTLE

by Nicolas Pitre

On Tue, 15 Jan 2013, Mark Hambleton wrote: > Hi Lorenzo, > > > > +obj-$(CONFIG_BIG_LITTLE) += arm_big_little.o > > There is nothing big.LITTLE specific in all of this, so arm_idle.c would > >be better. > > I figured that because the current version calls into the big.little > platform power framework (bL_entry.c) and makes calls into that > framework that this wasn't totally generic and is dependant upon that > code. The version of the cpuidle driver won't build unless that code > is built in, so I still think this is more appropriate naming, I could > call it bL_* but I suspect someone will object to that upstream > because of the mixed case. I'll wait that someone with a cluebat. Semantically, "bL_" is the most efficient prefix you could find to refer to b"ig.LITTLE". And no one eported any issue with that from the initial public review so far. Nicolas

12 years, 1 month

Re: [PATCH] Genericise CPUIdle for ARM big.LITTLE

by Steve Bannister

Thanks for this Mark, CCing Lorenzo as this doesn't seem to have come through on the linaro-devs list. Cheers, Steve On 15 Jan 2013, at 13:35, Mark Hambleton <mark.hambleton(a)broadcom.com> wrote: > Take the TC2 version of CPUIdle and remove TC2 specific line. > Move the file to drivers/cpuidle. > Introduce the concept of "arm,generic" compatible drivers to avoid > the need for a dummy dts node. > > Build the new CPUIdle based on CONFIG_BIG_LITTLE. > > Signed-off-by: mark <mahamble(a)broadcom.com> > --- > Hi Nicolas / Dave, > > After plugging in our ASIC support to the new big.LITTLE frameworks I realised that the TC2 CPUIdle driver was fundamentally generic, this patch moves it from mach-vexpress into drivers/cpuidle and then checks for the presence of a compatible node in the devicetree instead of checking for an SPC. > > I haven't added a new config define as I figured it was specific to the big.LITTLE framework, so just used that instead. > > I have tested the change on TC2 and am looking at extending the DCSCB code to make it work on fastmodels. > > Any comments? > > Regards, > > Mark > > arch/arm/boot/dts/vexpress-v2p-ca15-tc2.dts | 2 +- > arch/arm/mach-vexpress/Kconfig | 12 -- > arch/arm/mach-vexpress/Makefile | 1 - > arch/arm/mach-vexpress/cpuidle-tc2.c | 184 --------------------------- > drivers/cpuidle/Makefile | 2 +- > drivers/cpuidle/arm_big_little.c | 183 ++++++++++++++++++++++++++ > 6 files changed, 185 insertions(+), 199 deletions(-) > delete mode 100644 arch/arm/mach-vexpress/cpuidle-tc2.c > create mode 100755 drivers/cpuidle/arm_big_little.c > > diff --git a/arch/arm/boot/dts/vexpress-v2p-ca15-tc2.dts b/arch/arm/boot/dts/vexpress-v2p-ca15-tc2.dts > index 3b7cd86..b261bde 100644 > --- a/arch/arm/boot/dts/vexpress-v2p-ca15-tc2.dts > +++ b/arch/arm/boot/dts/vexpress-v2p-ca15-tc2.dts > @@ -15,7 +15,7 @@ > model = "V2P-CA15_CA7"; > arm,hbi = <0x249>; > arm,vexpress,site = <0xf>; > - compatible = "arm,vexpress,v2p-ca15,tc2", "arm,vexpress,v2p-ca15", "arm,vexpress"; > + compatible = "arm,vexpress,v2p-ca15,tc2", "arm,vexpress,v2p-ca15", "arm,vexpress", "arm,generic"; > interrupt-parent = <&gic>; > #address-cells = <2>; > #size-cells = <2>; > diff --git a/arch/arm/mach-vexpress/Kconfig b/arch/arm/mach-vexpress/Kconfig > index f302fa9..3fd44c4 100644 > --- a/arch/arm/mach-vexpress/Kconfig > +++ b/arch/arm/mach-vexpress/Kconfig > @@ -61,18 +61,6 @@ config ARCH_VEXPRESS_TC2 > help > Support for CPU and cluster power management on TC2. > > -config VEXPRESS_TC2_CPUIDLE > - bool "cpuidle support for TC2 test-chip (EXPERIMENTAL)" > - depends on EXPERIMENTAL > - depends on CPU_IDLE && PM && ARCH_VEXPRESS_TC2 > - select ARM_CPU_SUSPEND > - select ARM_SPC > - help > - Provides code that enables CPU idle power management on the > - TC2 testchip. It enables the CPU idle driver so that the kernel > - can enter cluster power down states provided by the power > - controller. > - > config ARCH_VEXPRESS_DCSCB > bool "Dual Cluster System Control Block (DCSCB) support" > depends on BIG_LITTLE > diff --git a/arch/arm/mach-vexpress/Makefile b/arch/arm/mach-vexpress/Makefile > index 05c6fed..b10390e 100644 > --- a/arch/arm/mach-vexpress/Makefile > +++ b/arch/arm/mach-vexpress/Makefile > @@ -10,4 +10,3 @@ obj-$(CONFIG_ARCH_VEXPRESS_DCSCB) += dcscb.o dcscb_setup.o > obj-$(CONFIG_ARCH_VEXPRESS_TC2) += tc2_pm.o tc2_pm_setup.o > obj-$(CONFIG_SMP) += platsmp.o > obj-$(CONFIG_HOTPLUG_CPU) += hotplug.o > -obj-$(CONFIG_VEXPRESS_TC2_CPUIDLE) += cpuidle-tc2.o > diff --git a/arch/arm/mach-vexpress/cpuidle-tc2.c b/arch/arm/mach-vexpress/cpuidle-tc2.c > deleted file mode 100644 > index 1f1efc4..0000000 > --- a/arch/arm/mach-vexpress/cpuidle-tc2.c > +++ /dev/null > @@ -1,184 +0,0 @@ > -/* > - * TC2 CPU idle driver. > - * > - * Copyright (C) 2012 ARM Ltd. > - * Author: Lorenzo Pieralisi <lorenzo.pieralisi(a)arm.com> > - * > - * 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. > - */ > - > -#include <linux/arm-cci.h> > -#include <linux/bitmap.h> > -#include <linux/cpuidle.h> > -#include <linux/cpu_pm.h> > -#include <linux/clockchips.h> > -#include <linux/debugfs.h> > -#include <linux/hrtimer.h> > -#include <linux/kernel.h> > -#include <linux/module.h> > -#include <linux/tick.h> > -#include <linux/vexpress.h> > -#include <asm/bL_entry.h> > -#include <asm/cpuidle.h> > -#include <asm/cputype.h> > -#include <asm/idmap.h> > -#include <asm/proc-fns.h> > -#include <asm/suspend.h> > - > -#include <mach/motherboard.h> > - > -static int tc2_cpuidle_simple_enter(struct cpuidle_device *dev, > - struct cpuidle_driver *drv, int index) > -{ > - ktime_t time_start, time_end; > - s64 diff; > - > - time_start = ktime_get(); > - > - cpu_do_idle(); > - > - time_end = ktime_get(); > - > - local_irq_enable(); > - > - diff = ktime_to_us(ktime_sub(time_end, time_start)); > - if (diff > INT_MAX) > - diff = INT_MAX; > - > - dev->last_residency = (int) diff; > - > - return index; > -} > - > -static int tc2_enter_powerdown(struct cpuidle_device *dev, > - struct cpuidle_driver *drv, int idx); > - > -static struct cpuidle_state tc2_cpuidle_set[] __initdata = { > - [0] = { > - .enter = tc2_cpuidle_simple_enter, > - .exit_latency = 1, > - .target_residency = 1, > - .power_usage = UINT_MAX, > - .flags = CPUIDLE_FLAG_TIME_VALID, > - .name = "WFI", > - .desc = "ARM WFI", > - }, > - [1] = { > - .enter = tc2_enter_powerdown, > - .exit_latency = 300, > - .target_residency = 1000, > - .flags = CPUIDLE_FLAG_TIME_VALID, > - .name = "C1", > - .desc = "ARM power down", > - }, > -}; > - > -struct cpuidle_driver tc2_idle_driver = { > - .name = "tc2_idle", > - .owner = THIS_MODULE, > - .safe_state_index = 0 > -}; > - > -static DEFINE_PER_CPU(struct cpuidle_device, tc2_idle_dev); > - > -static int notrace tc2_powerdown_finisher(unsigned long arg) > -{ > - unsigned int mpidr = read_cpuid_mpidr(); > - unsigned int cluster = (mpidr >> 8) & 0xf; > - unsigned int cpu = mpidr & 0xf; > - > - bL_set_entry_vector(cpu, cluster, cpu_resume); > - bL_cpu_suspend(0); /* 0 should be replaced with better value here */ > - return 1; > -} > - > -/* > - * tc2_enter_powerdown - Programs CPU to enter the specified state > - * @dev: cpuidle device > - * @drv: The target state to be programmed > - * @idx: state index > - * > - * Called from the CPUidle framework to program the device to the > - * specified target state selected by the governor. > - */ > -static int tc2_enter_powerdown(struct cpuidle_device *dev, > - struct cpuidle_driver *drv, int idx) > -{ > - struct timespec ts_preidle, ts_postidle, ts_idle; > - int ret; > - > - /* Used to keep track of the total time in idle */ > - getnstimeofday(&ts_preidle); > - > - BUG_ON(!irqs_disabled()); > - > - cpu_pm_enter(); > - > - clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu); > - > - ret = cpu_suspend((unsigned long) dev, tc2_powerdown_finisher); > - if (ret) > - BUG(); > - > - bL_cpu_powered_up(); > - > - clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu); > - > - cpu_pm_exit(); > - > - getnstimeofday(&ts_postidle); > - local_irq_enable(); > - ts_idle = timespec_sub(ts_postidle, ts_preidle); > - > - dev->last_residency = ts_idle.tv_nsec / NSEC_PER_USEC + > - ts_idle.tv_sec * USEC_PER_SEC; > - return idx; > -} > - > -/* > - * tc2_idle_init > - * > - * Registers the TC2 specific cpuidle driver with the cpuidle > - * framework with the valid set of states. > - */ > -int __init tc2_idle_init(void) > -{ > - struct cpuidle_device *dev; > - int i, cpu_id; > - struct cpuidle_driver *drv = &tc2_idle_driver; > - > - if (!vexpress_spc_check_loaded()) { > - pr_info("TC2 CPUidle not registered because no SPC found\n"); > - return -ENODEV; > - } > - > - drv->state_count = (sizeof(tc2_cpuidle_set) / > - sizeof(struct cpuidle_state)); > - > - for (i = 0; i < drv->state_count; i++) { > - memcpy(&drv->states[i], &tc2_cpuidle_set[i], > - sizeof(struct cpuidle_state)); > - } > - > - cpuidle_register_driver(drv); > - > - for_each_cpu(cpu_id, cpu_online_mask) { > - pr_err("CPUidle for CPU%d registered\n", cpu_id); > - dev = &per_cpu(tc2_idle_dev, cpu_id); > - dev->cpu = cpu_id; > - > - dev->state_count = drv->state_count; > - > - if (cpuidle_register_device(dev)) { > - printk(KERN_ERR "%s: Cpuidle register device failed\n", > - __func__); > - return -EIO; > - } > - } > - > - return 0; > -} > - > -late_initcall(tc2_idle_init); > diff --git a/drivers/cpuidle/Makefile b/drivers/cpuidle/Makefile > index 03ee874..398ece7 100644 > --- a/drivers/cpuidle/Makefile > +++ b/drivers/cpuidle/Makefile > @@ -4,5 +4,5 @@ > > obj-y += cpuidle.o driver.o governor.o sysfs.o governors/ > obj-$(CONFIG_ARCH_NEEDS_CPU_IDLE_COUPLED) += coupled.o > - > +obj-$(CONFIG_BIG_LITTLE) += arm_big_little.o > obj-$(CONFIG_CPU_IDLE_CALXEDA) += cpuidle-calxeda.o > diff --git a/drivers/cpuidle/arm_big_little.c b/drivers/cpuidle/arm_big_little.c > new file mode 100755 > index 0000000..b97ebe0 > --- /dev/null > +++ b/drivers/cpuidle/arm_big_little.c > @@ -0,0 +1,183 @@ > +/* > + * big.LITTLE CPU idle driver. > + * > + * Copyright (C) 2012 ARM Ltd. > + * Author: Lorenzo Pieralisi <lorenzo.pieralisi(a)arm.com> > + * > + * 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. > + */ > + > +#include <linux/arm-cci.h> > +#include <linux/bitmap.h> > +#include <linux/cpuidle.h> > +#include <linux/cpu_pm.h> > +#include <linux/clockchips.h> > +#include <linux/debugfs.h> > +#include <linux/hrtimer.h> > +#include <linux/kernel.h> > +#include <linux/module.h> > +#include <linux/tick.h> > +#include <linux/vexpress.h> > +#include <asm/bL_entry.h> > +#include <asm/cpuidle.h> > +#include <asm/cputype.h> > +#include <asm/idmap.h> > +#include <asm/proc-fns.h> > +#include <asm/suspend.h> > +#include <linux/of.h> > + > +static int bl_cpuidle_simple_enter(struct cpuidle_device *dev, > + struct cpuidle_driver *drv, int index) > +{ > + ktime_t time_start, time_end; > + s64 diff; > + > + time_start = ktime_get(); > + > + cpu_do_idle(); > + > + time_end = ktime_get(); > + > + local_irq_enable(); > + > + diff = ktime_to_us(ktime_sub(time_end, time_start)); > + if (diff > INT_MAX) > + diff = INT_MAX; > + > + dev->last_residency = (int) diff; > + > + return index; > +} > + > +static int bl_enter_powerdown(struct cpuidle_device *dev, > + struct cpuidle_driver *drv, int idx); > + > +static struct cpuidle_state bl_cpuidle_set[] __initdata = { > + [0] = { > + .enter = bl_cpuidle_simple_enter, > + .exit_latency = 1, > + .target_residency = 1, > + .power_usage = UINT_MAX, > + .flags = CPUIDLE_FLAG_TIME_VALID, > + .name = "WFI", > + .desc = "ARM WFI", > + }, > + [1] = { > + .enter = bl_enter_powerdown, > + .exit_latency = 300, > + .target_residency = 1000, > + .flags = CPUIDLE_FLAG_TIME_VALID, > + .name = "C1", > + .desc = "ARM power down", > + }, > +}; > + > +struct cpuidle_driver bl_idle_driver = { > + .name = "bl_idle", > + .owner = THIS_MODULE, > + .safe_state_index = 0 > +}; > + > +static DEFINE_PER_CPU(struct cpuidle_device, bl_idle_dev); > + > +static int notrace bl_powerdown_finisher(unsigned long arg) > +{ > + unsigned int mpidr = read_cpuid_mpidr(); > + unsigned int cluster = (mpidr >> 8) & 0xf; > + unsigned int cpu = mpidr & 0xf; > + > + bL_set_entry_vector(cpu, cluster, cpu_resume); > + bL_cpu_suspend(0); /* 0 should be replaced with better value here */ > + return 1; > +} > + > +/* > + * bl_enter_powerdown - Programs CPU to enter the specified state > + * @dev: cpuidle device > + * @drv: The target state to be programmed > + * @idx: state index > + * > + * Called from the CPUidle framework to program the device to the > + * specified target state selected by the governor. > + */ > +static int bl_enter_powerdown(struct cpuidle_device *dev, > + struct cpuidle_driver *drv, int idx) > +{ > + struct timespec ts_preidle, ts_postidle, ts_idle; > + int ret; > + > + /* Used to keep track of the total time in idle */ > + getnstimeofday(&ts_preidle); > + > + BUG_ON(!irqs_disabled()); > + > + cpu_pm_enter(); > + > + clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &dev->cpu); > + > + ret = cpu_suspend((unsigned long) dev, bl_powerdown_finisher); > + if (ret) > + BUG(); > + > + bL_cpu_powered_up(); > + > + clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &dev->cpu); > + > + cpu_pm_exit(); > + > + getnstimeofday(&ts_postidle); > + local_irq_enable(); > + ts_idle = timespec_sub(ts_postidle, ts_preidle); > + > + dev->last_residency = ts_idle.tv_nsec / NSEC_PER_USEC + > + ts_idle.tv_sec * USEC_PER_SEC; > + return idx; > +} > + > +/* > + * bl_idle_init > + * > + * Registers the bl specific cpuidle driver with the cpuidle > + * framework with the valid set of states. > + */ > +int __init bl_idle_init(void) > +{ > + struct cpuidle_device *dev; > + int i, cpu_id; > + struct cpuidle_driver *drv = &bl_idle_driver; > + > + if (!of_find_compatible_node(NULL, NULL, "arm,generic")) { > + pr_info("%s: No compatible node found\n", __func__); > + return -ENODEV; > + } > + > + drv->state_count = (sizeof(bl_cpuidle_set) / > + sizeof(struct cpuidle_state)); > + > + for (i = 0; i < drv->state_count; i++) { > + memcpy(&drv->states[i], &bl_cpuidle_set[i], > + sizeof(struct cpuidle_state)); > + } > + > + cpuidle_register_driver(drv); > + > + for_each_cpu(cpu_id, cpu_online_mask) { > + pr_err("CPUidle for CPU%d registered\n", cpu_id); > + dev = &per_cpu(bl_idle_dev, cpu_id); > + dev->cpu = cpu_id; > + > + dev->state_count = drv->state_count; > + > + if (cpuidle_register_device(dev)) { > + printk(KERN_ERR "%s: Cpuidle register device failed\n", > + __func__); > + return -EIO; > + } > + } > + > + return 0; > +} > + > +late_initcall(bl_idle_init); > -- > 1.7.9.5 > > > -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.

12 years, 1 month

Fwd: [PATCH] vexpress: Setup IRQ affinity to A7s for TC2

by Punit Agrawal

(Forward to the right address for linaro-dev) -------- Original Message -------- Subject: [PATCH] vexpress: Setup IRQ affinity to A7s for TC2 Date: Wed, 9 Jan 2013 14:20:19 +0000 From: Punit Agrawal <punit.agrawal(a)arm.com><mailto:punit.agrawal@arm.com> To: tixy(a)linaro.org<mailto:tixy@linaro.org> <tixy(a)linaro.org><mailto:tixy@linaro.org> CC: linaro-dev(a)linaro.org<mailto:linaro-dev@linaro.org> <linaro-dev(a)linaro.org><mailto:linaro-dev@linaro.org>, Punit Agrawal <Punit.Agrawal(a)arm.com><mailto:Punit.Agrawal@arm.com> On TC2, to maximise power savings, it helps to affine the IRQs to the A7s. Now that the cpu to parts mapping is available via /proc/cpuinfo, this patch introduces a script to parse this file and build a mask of particular CPU type and change the IRQ affinity to this mask. The patch also makes the necessary changes to platform specific init script to call above script and affine the IRQs to the A7s. Lastly, it changes the makefile to include the new script in the target. Change-Id: I2fd2840acd4602d9145c7c2f2e839e5dda09ff99 Signed-off-by: Punit Agrawal <punit.agrawal(a)arm.com><mailto:punit.agrawal@arm.com> --- Hi Tixy, I am not quite sure where to send this patch to get it included in Linaro. But you seem to have a few patches in this repo, so maybe you can apply this one. Let me know if you have any feedback on this as well. I would be happy to incorporate that as long as the intent of the patches carries through. Thanks, Punit device.mk | 3 ++- init.arm-versatileexpress.rc | 6 ++++++ set_irq_affinity.sh | 45 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 53 insertions(+), 1 deletion(-) create mode 100755 set_irq_affinity.sh diff --git a/device.mk b/device.mk index 6e6c6dd..9ca73e3 100644 --- a/device.mk +++ b/device.mk @@ -9,7 +9,8 @@ PRODUCT_COPY_FILES += \ device/linaro/vexpress/init.v2p-aarch64.rc:root/init.v2p-aarch64.rc \ device/linaro/vexpress/ueventd.v2p-aarch64.rc:root/ueventd.v2p-aarch64.rc \ device/linaro/vexpress/init.vexpress.sh:system/etc/init.vexpress.sh \ - device/linaro/vexpress/initlogo.rle:root/initlogo.rle + device/linaro/vexpress/initlogo.rle:root/initlogo.rle \ + device/linaro/vexpress/set_irq_affinity.sh:root/set_irq_affinity.sh PRODUCT_CHARACTERISTICS := tablet,nosdcard diff --git a/init.arm-versatileexpress.rc b/init.arm-versatileexpress.rc index a71e491..d710f36 100644 --- a/init.arm-versatileexpress.rc +++ b/init.arm-versatileexpress.rc @@ -33,6 +33,12 @@ on boot chown system system /sys/class/graphics/fb0/fit_to_screen chown system system /sys/class/graphics/fb1/overlays +# setup IRQ affinity to the A7s +service setirqaffinity /set_irq_affinity 0xc07 + class main + user root + oneshot + service faketsd /system/bin/faketsd class main user bluetooth diff --git a/set_irq_affinity.sh b/set_irq_affinity.sh new file mode 100755 index 0000000..4a09d61 --- /dev/null +++ b/set_irq_affinity.sh @@ -0,0 +1,45 @@ +#!/system/bin/sh + +# This script sets the default affinity to the processors with the given part id. +# - part id is in hex (as seen in /proc/cpuinfo) + +function build_mask_from_part_id { + local IFS + local mask + local ref_part_id + + ref_part_id=$1 + IFS=$'\n' + + for line in `cat /proc/cpuinfo` + do + IFS=':' + set -A tokens $line + + if [ "${line#'processor'}" != "$line" ] + then + cpu="${tokens[1]##' '}" + elif [ "${line#'CPU part'}" != "$line" ] + then + part_id="${tokens[1]##' '}" + + if [ "$part_id" == "$ref_part_id" ] + then + (( mask |= 1 << $cpu )) + fi + fi + done + echo $(printf "%x" $mask) +} + +ref_part_id=$(echo $1 | tr '[A-Z]' '[a-z]') +mask=$(build_mask_from_part_id $ref_part_id) +[ -z "$mask" ] && exit + +echo $mask > /proc/irq/default_smp_affinity + +for i in `ls /proc/irq` +do + affinity_file="/proc/irq/$i/smp_affinity" + [ -e $affinity_file ] && echo $mask > $affinity_file +done -- 1.7.9.5 -- IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.

12 years, 1 month

[PATCH V2 1/2] cpufreq: Simplify __cpufreq_remove_dev()

by Viresh Kumar

__cpufreq_remove_dev() is called on multiple occasions: cpufreq_driver unregister and cpu removals. Current implementation of this routine is overly complex without much need. If the cpu to be removed is the policy->cpu, we remove the policy first and add all other cpus again from policy->cpus and then finally call __cpufreq_remove_dev() again to remove the cpu to be deleted. Haahhhh.. There exist a simple solution to removal of a cpu: - Simply use the old policy structure - update its fields like: policy->cpu, etc. - notify any users of cpufreq, which depend on changing policy->cpu Hence this patch, which tries to implement the above theory. It is tested well by myself on ARM big.LITTLE TC2 SoC, which has 5 cores (2 A15 and 3 A7). Both A15's share same struct policy and all A7's share same policy structure. Signed-off-by: Viresh Kumar <viresh.kumar(a)linaro.org> Tested-by: Shawn Guo <shawn.guo(a)linaro.org> --- V1->V2: - #ifdef,endif CONFIG_CPU_FREQ_TABLE added across cpufreq_frequency_table_update_policy_cpu() drivers/cpufreq/cpufreq.c | 161 +++++++++++++++++----------------------- drivers/cpufreq/cpufreq_stats.c | 27 ++++++- drivers/cpufreq/freq_table.c | 9 +++ include/linux/cpufreq.h | 14 ++-- 4 files changed, 113 insertions(+), 98 deletions(-) diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c index 034d183..9af14a8 100644 --- a/drivers/cpufreq/cpufreq.c +++ b/drivers/cpufreq/cpufreq.c @@ -1036,6 +1036,25 @@ module_out: return ret; } +static void update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu) +{ + int j; + + policy->last_cpu = policy->cpu; + policy->cpu = cpu; + + for_each_cpu(j, policy->cpus) { + if (!cpu_online(j)) + continue; + per_cpu(cpufreq_policy_cpu, j) = cpu; + } + +#ifdef CONFIG_CPU_FREQ_TABLE + cpufreq_frequency_table_update_policy_cpu(policy); +#endif + blocking_notifier_call_chain(&cpufreq_policy_notifier_list, + CPUFREQ_UPDATE_POLICY_CPU, policy); +} /** * __cpufreq_remove_dev - remove a CPU device @@ -1046,132 +1065,92 @@ module_out: */ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif) { - unsigned int cpu = dev->id; + unsigned int cpu = dev->id, ret, cpus; unsigned long flags; struct cpufreq_policy *data; struct kobject *kobj; struct completion *cmp; -#ifdef CONFIG_SMP struct device *cpu_dev; - unsigned int j; -#endif - pr_debug("unregistering CPU %u\n", cpu); + pr_debug("%s: unregistering CPU %u\n", __func__, cpu); spin_lock_irqsave(&cpufreq_driver_lock, flags); data = per_cpu(cpufreq_cpu_data, cpu); if (!data) { + pr_debug("%s: No cpu_data found\n", __func__); spin_unlock_irqrestore(&cpufreq_driver_lock, flags); unlock_policy_rwsem_write(cpu); return -EINVAL; } - per_cpu(cpufreq_cpu_data, cpu) = NULL; -#ifdef CONFIG_SMP - /* if this isn't the CPU which is the parent of the kobj, we - * only need to unlink, put and exit - */ - if (unlikely(cpu != data->cpu)) { - pr_debug("removing link\n"); + if (cpufreq_driver->target) __cpufreq_governor(data, CPUFREQ_GOV_STOP); - cpumask_clear_cpu(cpu, data->cpus); - spin_unlock_irqrestore(&cpufreq_driver_lock, flags); - - __cpufreq_governor(data, CPUFREQ_GOV_START); - __cpufreq_governor(data, CPUFREQ_GOV_LIMITS); - - kobj = &dev->kobj; - cpufreq_cpu_put(data); - unlock_policy_rwsem_write(cpu); - sysfs_remove_link(kobj, "cpufreq"); - return 0; - } -#endif - -#ifdef CONFIG_SMP #ifdef CONFIG_HOTPLUG_CPU strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name, CPUFREQ_NAME_LEN); #endif - /* if we have other CPUs still registered, we need to unlink them, - * or else wait_for_completion below will lock up. Clean the - * per_cpu(cpufreq_cpu_data) while holding the lock, and remove - * the sysfs links afterwards. - */ - if (unlikely(cpumask_weight(data->cpus) > 1)) { - for_each_cpu(j, data->cpus) { - if (j == cpu) - continue; - per_cpu(cpufreq_cpu_data, j) = NULL; - } - } - - spin_unlock_irqrestore(&cpufreq_driver_lock, flags); + per_cpu(cpufreq_cpu_data, cpu) = NULL; + cpus = cpumask_weight(data->cpus); + cpumask_clear_cpu(cpu, data->cpus); - if (unlikely(cpumask_weight(data->cpus) > 1)) { - for_each_cpu(j, data->cpus) { - if (j == cpu) - continue; - pr_debug("removing link for cpu %u\n", j); -#ifdef CONFIG_HOTPLUG_CPU - strncpy(per_cpu(cpufreq_cpu_governor, j), - data->governor->name, CPUFREQ_NAME_LEN); -#endif - cpu_dev = get_cpu_device(j); - kobj = &cpu_dev->kobj; + if (unlikely((cpu == data->cpu) && (cpus > 1))) { + /* first sibling now owns the new sysfs dir */ + cpu_dev = get_cpu_device(cpumask_first(data->cpus)); + sysfs_remove_link(&cpu_dev->kobj, "cpufreq"); + ret = kobject_move(&data->kobj, &cpu_dev->kobj); + if (ret) { + pr_err("%s: Failed to move kobj: %d", __func__, ret); + cpumask_set_cpu(cpu, data->cpus); + ret = sysfs_create_link(&cpu_dev->kobj, &data->kobj, + "cpufreq"); + spin_unlock_irqrestore(&cpufreq_driver_lock, flags); unlock_policy_rwsem_write(cpu); - sysfs_remove_link(kobj, "cpufreq"); - lock_policy_rwsem_write(cpu); - cpufreq_cpu_put(data); + return -EINVAL; } + + update_policy_cpu(data, cpu_dev->id); + pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n", + __func__, cpu_dev->id, cpu); } -#else - spin_unlock_irqrestore(&cpufreq_driver_lock, flags); -#endif - if (cpufreq_driver->target) - __cpufreq_governor(data, CPUFREQ_GOV_STOP); + spin_unlock_irqrestore(&cpufreq_driver_lock, flags); - kobj = &data->kobj; - cmp = &data->kobj_unregister; + pr_debug("%s: removing link, cpu: %d\n", __func__, cpu); + cpufreq_cpu_put(data); unlock_policy_rwsem_write(cpu); - kobject_put(kobj); + sysfs_remove_link(&dev->kobj, "cpufreq"); - /* we need to make sure that the underlying kobj is actually - * not referenced anymore by anybody before we proceed with - * unloading. - */ - pr_debug("waiting for dropping of refcount\n"); - wait_for_completion(cmp); - pr_debug("wait complete\n"); - - lock_policy_rwsem_write(cpu); - if (cpufreq_driver->exit) - cpufreq_driver->exit(data); - unlock_policy_rwsem_write(cpu); + /* If cpu is last user of policy, free policy */ + if (cpus == 1) { + lock_policy_rwsem_write(cpu); + kobj = &data->kobj; + cmp = &data->kobj_unregister; + unlock_policy_rwsem_write(cpu); + kobject_put(kobj); -#ifdef CONFIG_HOTPLUG_CPU - /* when the CPU which is the parent of the kobj is hotplugged - * offline, check for siblings, and create cpufreq sysfs interface - * and symlinks - */ - if (unlikely(cpumask_weight(data->cpus) > 1)) { - /* first sibling now owns the new sysfs dir */ - cpumask_clear_cpu(cpu, data->cpus); - cpufreq_add_dev(get_cpu_device(cpumask_first(data->cpus)), NULL); + /* we need to make sure that the underlying kobj is actually + * not referenced anymore by anybody before we proceed with + * unloading. + */ + pr_debug("waiting for dropping of refcount\n"); + wait_for_completion(cmp); + pr_debug("wait complete\n"); - /* finally remove our own symlink */ lock_policy_rwsem_write(cpu); - __cpufreq_remove_dev(dev, sif); - } -#endif + if (cpufreq_driver->exit) + cpufreq_driver->exit(data); + unlock_policy_rwsem_write(cpu); - free_cpumask_var(data->related_cpus); - free_cpumask_var(data->cpus); - kfree(data); + free_cpumask_var(data->related_cpus); + free_cpumask_var(data->cpus); + kfree(data); + } else if (cpufreq_driver->target) { + __cpufreq_governor(data, CPUFREQ_GOV_START); + __cpufreq_governor(data, CPUFREQ_GOV_LIMITS); + } return 0; } diff --git a/drivers/cpufreq/cpufreq_stats.c b/drivers/cpufreq/cpufreq_stats.c index e40e508..251719a 100644 --- a/drivers/cpufreq/cpufreq_stats.c +++ b/drivers/cpufreq/cpufreq_stats.c @@ -170,11 +170,13 @@ static int freq_table_get_index(struct cpufreq_stats *stat, unsigned int freq) static void cpufreq_stats_free_table(unsigned int cpu) { struct cpufreq_stats *stat = per_cpu(cpufreq_stats_table, cpu); + if (stat) { + pr_debug("%s: Free stat table\n", __func__); kfree(stat->time_in_state); kfree(stat); + per_cpu(cpufreq_stats_table, cpu) = NULL; } - per_cpu(cpufreq_stats_table, cpu) = NULL; } /* must be called early in the CPU removal sequence (before @@ -183,8 +185,10 @@ static void cpufreq_stats_free_table(unsigned int cpu) static void cpufreq_stats_free_sysfs(unsigned int cpu) { struct cpufreq_policy *policy = cpufreq_cpu_get(cpu); - if (policy && policy->cpu == cpu) + if (policy && (cpumask_weight(policy->cpus) == 1)) { + pr_debug("%s: Free sysfs stat\n", __func__); sysfs_remove_group(&policy->kobj, &stats_attr_group); + } if (policy) cpufreq_cpu_put(policy); } @@ -262,6 +266,19 @@ error_get_fail: return ret; } +static void cpufreq_stats_update_policy_cpu(struct cpufreq_policy *policy) +{ + struct cpufreq_stats *stat = per_cpu(cpufreq_stats_table, + policy->last_cpu); + + pr_debug("Updating stats_table for new_cpu %u from last_cpu %u\n", + policy->cpu, policy->last_cpu); + per_cpu(cpufreq_stats_table, policy->cpu) = per_cpu(cpufreq_stats_table, + policy->last_cpu); + per_cpu(cpufreq_stats_table, policy->last_cpu) = NULL; + stat->cpu = policy->cpu; +} + static int cpufreq_stat_notifier_policy(struct notifier_block *nb, unsigned long val, void *data) { @@ -269,6 +286,12 @@ static int cpufreq_stat_notifier_policy(struct notifier_block *nb, struct cpufreq_policy *policy = data; struct cpufreq_frequency_table *table; unsigned int cpu = policy->cpu; + + if (val == CPUFREQ_UPDATE_POLICY_CPU) { + cpufreq_stats_update_policy_cpu(policy); + return 0; + } + if (val != CPUFREQ_NOTIFY) return 0; table = cpufreq_frequency_get_table(cpu); diff --git a/drivers/cpufreq/freq_table.c b/drivers/cpufreq/freq_table.c index 49cda25..aa5bd39 100644 --- a/drivers/cpufreq/freq_table.c +++ b/drivers/cpufreq/freq_table.c @@ -227,6 +227,15 @@ void cpufreq_frequency_table_put_attr(unsigned int cpu) } EXPORT_SYMBOL_GPL(cpufreq_frequency_table_put_attr); +void cpufreq_frequency_table_update_policy_cpu(struct cpufreq_policy *policy) +{ + pr_debug("Updating show_table for new_cpu %u from last_cpu %u\n", + policy->cpu, policy->last_cpu); + per_cpu(cpufreq_show_table, policy->cpu) = per_cpu(cpufreq_show_table, + policy->last_cpu); + per_cpu(cpufreq_show_table, policy->last_cpu) = NULL; +} + struct cpufreq_frequency_table *cpufreq_frequency_get_table(unsigned int cpu) { return per_cpu(cpufreq_show_table, cpu); diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h index a55b88e..52be2d0 100644 --- a/include/linux/cpufreq.h +++ b/include/linux/cpufreq.h @@ -93,7 +93,9 @@ struct cpufreq_policy { cpumask_var_t related_cpus; /* CPUs with any coordination */ unsigned int shared_type; /* ANY or ALL affected CPUs should set cpufreq */ - unsigned int cpu; /* cpu nr of registered CPU */ + unsigned int cpu; /* cpu nr of CPU managing this policy */ + unsigned int last_cpu; /* cpu nr of previous CPU that managed + * this policy */ struct cpufreq_cpuinfo cpuinfo;/* see above */ unsigned int min; /* in kHz */ @@ -112,10 +114,11 @@ struct cpufreq_policy { struct completion kobj_unregister; }; -#define CPUFREQ_ADJUST (0) -#define CPUFREQ_INCOMPATIBLE (1) -#define CPUFREQ_NOTIFY (2) -#define CPUFREQ_START (3) +#define CPUFREQ_ADJUST (0) +#define CPUFREQ_INCOMPATIBLE (1) +#define CPUFREQ_NOTIFY (2) +#define CPUFREQ_START (3) +#define CPUFREQ_UPDATE_POLICY_CPU (4) #define CPUFREQ_SHARED_TYPE_NONE (0) /* None */ #define CPUFREQ_SHARED_TYPE_HW (1) /* HW does needed coordination */ @@ -405,6 +408,7 @@ extern struct freq_attr cpufreq_freq_attr_scaling_available_freqs; void cpufreq_frequency_table_get_attr(struct cpufreq_frequency_table *table, unsigned int cpu); +void cpufreq_frequency_table_update_policy_cpu(struct cpufreq_policy *policy); void cpufreq_frequency_table_put_attr(unsigned int cpu); #endif /* _LINUX_CPUFREQ_H */ -- 1.7.12.rc2.18.g61b472e

12 years, 2 months

[PATCH 0/5] Cpufreq Fixes for 3.9

by Viresh Kumar

Hi Rafael, As requested, i am resending all my cpufreq fixes together. I have added correct Tested-by flags. Viresh Kumar (5): cpufreq: Manage only online cpus cpufreq: Notify governors when cpus are hot-[un]plugged cpufreq: Don't use cpu removed during cpufreq_driver_unregister cpufreq: Simplify __cpufreq_remove_dev() cpufreq: Simplify cpufreq_add_dev() drivers/cpufreq/cpufreq.c | 316 +++++++++++++++++----------------------- drivers/cpufreq/cpufreq_stats.c | 27 +++- drivers/cpufreq/freq_table.c | 9 ++ include/linux/cpufreq.h | 14 +- 4 files changed, 178 insertions(+), 188 deletions(-) -- 1.7.12.rc2.18.g61b472e

12 years, 2 months

[PATCH 1/3] cpufreq: Manage only online cpus

by Viresh Kumar

cpufreq core doesn't manage offline cpus and if driver->init() has returned mask including offline cpus, it may result in unwanted behavior by cpufreq core or governors. We need to get only online cpus in this mask. There are two places to fix this mask, cpufreq core and cpufreq driver. It makes sense to do this at common place and hence is done in core. Signed-off-by: Viresh Kumar <viresh.kumar(a)linaro.org> --- drivers/cpufreq/cpufreq.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c index 1f93dbd..de99517 100644 --- a/drivers/cpufreq/cpufreq.c +++ b/drivers/cpufreq/cpufreq.c @@ -970,6 +970,13 @@ static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif) pr_debug("initialization failed\n"); goto err_unlock_policy; } + + /* + * affected cpus must always be the one, which are online. We aren't + * managing offline cpus here. + */ + cpumask_and(policy->cpus, policy->cpus, cpu_online_mask); + policy->user_policy.min = policy->min; policy->user_policy.max = policy->max; -- 1.7.12.rc2.18.g61b472e

12 years, 2 months

[PATCH 0/2][V2] cpuidle - remove power from the select equation

by Daniel Lezcano

The states could be sorted in the power backward order, we get rid of some lines of code and by this way that fixes also a bug in the dynamic c-states. Changelog: V2: * added the optimization in the select menu governor Daniel Lezcano (2): cpuidle - remove the power_specified field in the driver cpuidle - optimize the select function for the 'menu' governor drivers/cpuidle/cpuidle.c | 17 ++++------------- drivers/cpuidle/driver.c | 25 ------------------------- drivers/cpuidle/governors/menu.c | 20 ++++++++------------ include/linux/cpuidle.h | 2 +- 4 files changed, 13 insertions(+), 51 deletions(-) -- 1.7.5.4

12 years, 2 months

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

12 years, 2 months

← Newer
1
2
3
4
5
6
7
8
9
Older →

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

linaro-dev January 2013