arm_pmu: fix lockdep issues with ACPI systems
Subsystems (4)
Mark Rutland
maintainerv2 cover arm_pmu: fix lockdep issues with ACPI systems
Currently the arm_pmu ACPI code is somewhat dubious. It attempts to allocate memory and manipulate IRQs in a hotplug callback, which is an atomic context.
These patches (based on the arm64 for-next/core branch [1]) attempt to address this by moving work out of hotplug callback, requiring a reorganisation of the common arm_pmu code.
I've given these a boot-test on a Juno R1 system, both with DT and ACPI. In either case the PMU works as expected, and lockdep seems happy.
I've pushed the series out to my arm64/acpi-pmu-lockdep branch [2].
Since v1 [3]:
* Kill arm_pmu_platdata (and ux500 IRQ bouncer)
* Make PMU/CPU binding implicit in hotplug notifier
* Explicitly enable/disable SPIs at hogplug
* Add armpmu_alloc_atomic
* Cleanup per Will's comments
Thanks,
Mark.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git/log/?h=for-next/core
[2] https://git.kernel.org/pub/scm/linux/kernel/git/mark/linux.git/log/?h=arm64/acpi-pmu-lockdep
[3] https://lkml.kernel.org/r/20171101141239.45340-1-mark.rutland at arm.com
Mark Rutland (8):
ARM: ux500: remove PMU IRQ bouncer
arm_pmu: kill arm_pmu_platdata
arm_pmu: fold platform helpers into platform code
arm_pmu: add armpmu_alloc_atomic()
arm_pmu: acpi: check for mismatched PPIs
arm_pmu: explicitly enable/disable SPIs at hotplug
arm_pmu: note IRQs and PMUs per-cpu
arm_pmu: acpi: request IRQs up-front
arch/arm/mach-ux500/cpu-db8500.c | 35 --------- drivers/perf/arm_pmu.c | 149 +++++++++++++++++---------------------- drivers/perf/arm_pmu_acpi.c | 61 +++++++++++----- drivers/perf/arm_pmu_platform.c | 37 ++++++++-- include/linux/perf/arm_pmu.h | 26 ++----- 5 files changed, 142 insertions(+), 166 deletions(-)
Signature · 1 line
2.11.0
Mark Rutland
maintainerPATCH v2 1/8 ARM: ux500: remove PMU IRQ bouncer
The ux500 PMU IRQ bouncer is getting in the way of some fundametnal changes to the ARM PMU driver, and it's the only special case that exists today. Let's remove it.
--- arch/arm/mach-ux500/cpu-db8500.c | 35 ----------------------------------- 1 file changed, 35 deletions(-)
arch/arm/mach-ux500/cpu-db8500.c +0 -35
@@ -23,7 +23,6 @@#include <linux/of.h> #include <linux/of_address.h> #include <linux/of_platform.h> -#include <linux/perf/arm_pmu.h> #include <linux/regulator/machine.h> #include <asm/outercache.h>
@@ -112,37 +111,6 @@ static void ux500_restart(enum reboot_mode mode, const char *cmd)prcmu_system_reset (0 ); } -/* -* The PMU IRQ lines of two cores are wired together into a single interrupt. -* Bounce the interrupt to the other core if it's not ours. -*/ -static irqreturn_t db8500_pmu_handler (int irq,void *dev,irq_handler_t handler) -{ -irqreturn_t ret =handler (irq, dev); -int other = !smp_processor_id (); - -if (ret ==IRQ_NONE &&cpu_online (other)) -irq_set_affinity (irq,cpumask_of (other)); - -/* -* We should be able to get away with the amount of IRQ_NONEs we give, -* while still having the spurious IRQ detection code kick in if the -* interrupt really starts hitting spuriously. -*/ -return ret; -} - -static struct arm_pmu_platdata db8500_pmu_platdata = { - .handle_irq = db8500_pmu_handler, - .irq_flags =IRQF_NOBALANCING |IRQF_NO_THREAD , -}; - -static struct of_dev_auxdata u8500_auxdata_lookup[]__initdata = { -/* Requires call-back bindings. */ -OF_DEV_AUXDATA ("arm,cortex-a9-pmu" ,0 ,"arm-pmu" , &db8500_pmu_platdata), - {}, -}; -static struct of_dev_auxdata u8540_auxdata_lookup[]__initdata = {OF_DEV_AUXDATA ("stericsson,db8500-prcmu" ,0x80157000 ,"db8500-prcmu" ,NULL ), {},
@@ -165,9 +133,6 @@ static void __init u8500_init_machine(void)if (of_machine_is_compatible ("st-ericsson,u8540" ))of_platform_populate (NULL , u8500_local_bus_nodes, u8540_auxdata_lookup,NULL ); -else -of_platform_populate (NULL , u8500_local_bus_nodes, - u8500_auxdata_lookup,NULL ); }static const char * stericsson_dt_platform_compat[] = {
Signature · 1 line
2.11.0
PATCH v2 1/8 ARM: ux500: remove PMU IRQ bouncer
Quoted 7 lines · Mark Rutland
The ux500 PMU IRQ bouncer is getting in the way of some fundametnal changes to the ARM PMU driver, and it's the only special case that exists today. Let's remove it.
Signed-off-by Mark Rutland <[email protected]>Cc Linus Walleij <[email protected]>Cc Will Deacon <[email protected]>
It's OK this silicon design was especially strange it seems. We are done with the days of performance optimization for the platform anyways.
Yours,
Linus Walleij
Mark Rutland
maintainerPATCH v2 1/8 ARM: ux500: remove PMU IRQ bouncer
Quoted 6 lines · Linus Walleij
Quoted 4 lines · Mark Rutland
The ux500 PMU IRQ bouncer is getting in the way of some fundametnal changes to the ARM PMU driver, and it's the only special case that exists today. Let's remove it.
Quoted 8 lines
Quoted 3 lines
Signed-off-by Mark Rutland <[email protected]>Cc Linus Walleij <[email protected]>Cc Will Deacon <[email protected]>It's OK this silicon design was especially strange it seems. We are done with the days of performance optimization for the platform anyways.
Reviewed-by Linus Walleij <[email protected]>
Cheers!
Mark.
Mark Rutland
maintainerPATCH v2 2/8 arm_pmu: kill arm_pmu_platdata
In reply to Mark RutlandNow that we have no platforms passing platform data to the arm_pmu code, we can get rid of the platdata and associated hooks, paving the way for rework of our IRQ handling.
--- drivers/perf/arm_pmu.c | 27 ++++----------------------- include/linux/perf/arm_pmu.h | 17 ----------------- 2 files changed, 4 insertions(+), 40 deletions(-)
drivers/perf/arm_pmu.c +4 -23
@@ -17,7 +17,6 @@#include <linux/export.h> #include <linux/kernel.h> #include <linux/perf/arm_pmu.h> -#include <linux/platform_device.h> #include <linux/slab.h> #include <linux/sched/clock.h> #include <linux/spinlock.h>
@@ -320,17 +319,9 @@ validate_group(struct perf_event *event)return 0 ; } -static struct arm_pmu_platdata *armpmu_get_platdata (struct arm_pmu *armpmu) -{ -struct platform_device *pdev = armpmu->plat_device; - -return pdev ?dev_get_platdata (&pdev->dev) :NULL ; -} -static irqreturn_t armpmu_dispatch_irq (int irq,void *dev) {struct arm_pmu *armpmu; -struct arm_pmu_platdata *plat;int ret;u64 start_clock, finish_clock;
@@ -342,13 +333,8 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)*/ armpmu = *(void **)dev; - plat =armpmu_get_platdata (armpmu); - start_clock =sched_clock (); -if (plat && plat->handle_irq) - ret = plat->handle_irq (irq, armpmu, armpmu->handle_irq); -else - ret = armpmu->handle_irq (irq, armpmu); + ret = armpmu->handle_irq (irq, armpmu); finish_clock =sched_clock ();perf_sample_event_took (finish_clock - start_clock);
@@ -578,7 +564,6 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) goto err_out; } }else { -struct arm_pmu_platdata *platdata =armpmu_get_platdata (armpmu);unsigned long irq_flags; err =irq_force_affinity (irq,cpumask_of (cpu));
@@ -589,13 +574,9 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) goto err_out; } -if (platdata && platdata->irq_flags) { - irq_flags = platdata->irq_flags; - }else { - irq_flags =IRQF_PERCPU | -IRQF_NOBALANCING | -IRQF_NO_THREAD ; - } + irq_flags =IRQF_PERCPU | +IRQF_NOBALANCING | +IRQF_NO_THREAD ; err =request_irq (irq, handler, irq_flags,"arm-pmu" ,per_cpu_ptr (&hw_events->percpu_pmu, cpu));
include/linux/perf/arm_pmu.h +0 -17
@@ -17,23 +17,6 @@#include <linux/sysfs.h> #include <asm/cputype.h> -/* -* struct arm_pmu_platdata - ARM PMU platform data -* -* @handle_irq: an optional handler which will be called from the -* interrupt and passed the address of the low level handler, -* and can be used to implement any platform specific handling -* before or after calling it. -* -* @irq_flags: if non-zero, these flags will be passed to request_irq -* when requesting interrupts for this PMU device. -*/ -struct arm_pmu_platdata { -irqreturn_t (*handle_irq)(int irq,void *dev, -irq_handler_t pmu_handler); -unsigned long irq_flags; -}; -#ifdef CONFIG_ARM_PMU /*
Signature · 1 line
2.11.0
Mark Rutland
maintainerPATCH v2 3/8 arm_pmu: fold platform helpers into platform code
In reply to Mark RutlandThe armpmu_{request,free}_irqs() helpers are only used by arm_pmu_platform.c, so let's fold them in and make them static.
--- drivers/perf/arm_pmu.c | 21 --------------------- drivers/perf/arm_pmu_platform.c | 21 +++++++++++++++++++++ include/linux/perf/arm_pmu.h | 2 -- 3 files changed, 21 insertions(+), 23 deletions(-)
drivers/perf/arm_pmu.c +0 -21
@@ -534,14 +534,6 @@ void armpmu_free_irq(struct arm_pmu *armpmu, int cpu)free_irq (irq,per_cpu_ptr (&hw_events->percpu_pmu, cpu)); } -void armpmu_free_irqs (struct arm_pmu *armpmu) -{ -int cpu; - -for_each_cpu (cpu, &armpmu->supported_cpus) -armpmu_free_irq (armpmu, cpu); -} -int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) {int err =0 ;
@@ -593,19 +585,6 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)return err; } -int armpmu_request_irqs (struct arm_pmu *armpmu) -{ -int cpu, err; - -for_each_cpu (cpu, &armpmu->supported_cpus) { - err =armpmu_request_irq (armpmu, cpu); -if (err) -break ; - } - -return err; -} -static int armpmu_get_cpu_irq (struct arm_pmu *pmu,int cpu) {struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
drivers/perf/arm_pmu_platform.c +21 -0
@@ -164,6 +164,27 @@ static int pmu_parse_irqs(struct arm_pmu *pmu)return 0 ; } +static int armpmu_request_irqs (struct arm_pmu *armpmu) +{ +int cpu, err; + +for_each_cpu (cpu, &armpmu->supported_cpus) { + err =armpmu_request_irq (armpmu, cpu); +if (err) +break ; + } + +return err; +} + +static void armpmu_free_irqs (struct arm_pmu *armpmu) +{ +int cpu; + +for_each_cpu (cpu, &armpmu->supported_cpus) +armpmu_free_irq (armpmu, cpu); +} +int arm_pmu_device_probe (struct platform_device *pdev,const struct of_device_id *of_table,const struct pmu_probe_info *probe_table)
include/linux/perf/arm_pmu.h +0 -2
@@ -159,8 +159,6 @@ static inline int arm_pmu_acpi_probe(armpmu_init_fn init_fn) { return 0; }struct arm_pmu *armpmu_alloc (void );void armpmu_free (struct arm_pmu *pmu);int armpmu_register (struct arm_pmu *pmu); -int armpmu_request_irqs (struct arm_pmu *armpmu); -void armpmu_free_irqs (struct arm_pmu *armpmu);int armpmu_request_irq (struct arm_pmu *armpmu,int cpu);void armpmu_free_irq (struct arm_pmu *armpmu,int cpu);
Signature · 1 line
2.11.0
Mark Rutland
maintainerPATCH v2 4/8 arm_pmu: add armpmu_alloc_atomic()
In reply to Mark RutlandIn ACPI systems, we don't know the makeup of CPUs until we hotplug them on, and thus have to allocate the PMU datastructures at hotplug time. Thus, we must use GFP_ATOMIC allocations.
Let's add an armpmu_alloc_atomic() that we can use in this case.
--- drivers/perf/arm_pmu.c | 17 ++++++++++++++--- drivers/perf/arm_pmu_acpi.c | 2 +- include/linux/perf/arm_pmu.h | 1 + 3 files changed, 16 insertions(+), 4 deletions(-)
drivers/perf/arm_pmu.c +14 -3
@@ -760,18 +760,18 @@ static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu) &cpu_pmu->node); } -struct arm_pmu *armpmu_alloc (void ) +static struct arm_pmu *__armpmu_alloc (gfp_t flags) {struct arm_pmu *pmu;int cpu; - pmu =kzalloc (sizeof (*pmu),GFP_KERNEL ); + pmu =kzalloc (sizeof (*pmu), flags);if (!pmu) {pr_info ("failed to allocate PMU device!\n" ); goto out; } - pmu->hw_events = alloc_percpu(struct pmu_hw_events); + pmu->hw_events = alloc_percpu_gfp(struct pmu_hw_events, flags);if (!pmu->hw_events) {pr_info ("failed to allocate per-cpu PMU data.\n" ); goto out_free_pmu;
@@ -817,6 +817,17 @@ struct arm_pmu *armpmu_alloc(void)return NULL ; } +struct arm_pmu *armpmu_alloc (void ) +{ +return __armpmu_alloc (GFP_KERNEL ); +} + +struct arm_pmu *armpmu_alloc_atomic (void ) +{ +return __armpmu_alloc (GFP_ATOMIC ); +} + +void armpmu_free (struct arm_pmu *pmu) {free_percpu (pmu->hw_events);
drivers/perf/arm_pmu_acpi.c +1 -1
@@ -127,7 +127,7 @@ static struct arm_pmu *arm_pmu_acpi_find_alloc_pmu(void)return pmu; } - pmu =armpmu_alloc (); + pmu =armpmu_alloc_atomic ();if (!pmu) {pr_warn ("Unable to allocate PMU for CPU%d\n" ,smp_processor_id ());
include/linux/perf/arm_pmu.h +1 -0
@@ -157,6 +157,7 @@ static inline int arm_pmu_acpi_probe(armpmu_init_fn init_fn) { return 0; }/* Internal functions only for core arm_pmu code */ struct arm_pmu *armpmu_alloc (void ); +struct arm_pmu *armpmu_alloc_atomic (void );void armpmu_free (struct arm_pmu *pmu);int armpmu_register (struct arm_pmu *pmu);int armpmu_request_irq (struct arm_pmu *armpmu,int cpu);
Signature · 1 line
2.11.0
Mark Rutland
maintainerPATCH v2 5/8 arm_pmu: acpi: check for mismatched PPIs
In reply to Mark RutlandThe arm_pmu platform code explicitly checks for mismatched PPIs at probe time, while the ACPI code leaves this to the core code. Future refactoring will make this difficult for the core code to check, so let's have the ACPI code check this explicitly.
As before, upon a failure we'll continue on without an interrupt. Ho hum.
--- drivers/perf/arm_pmu.c | 17 ++++------------- drivers/perf/arm_pmu_acpi.c | 42 +++++++++++++++++++++++++++++++++++++---- drivers/perf/arm_pmu_platform.c | 7 ------- 3 files changed, 42 insertions(+), 24 deletions(-)
drivers/perf/arm_pmu.c +4 -13
@@ -543,19 +543,7 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)if (!irq)return 0 ; -if (irq_is_percpu_devid (irq) &&cpumask_empty (&armpmu->active_irqs)) { - err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); - }else if (irq_is_percpu_devid (irq)) { -int other_cpu =cpumask_first (&armpmu->active_irqs); -int other_irq =per_cpu (hw_events->irq, other_cpu); - -if (irq != other_irq) { -pr_warn ("mismatched PPIs detected.\n" ); - err = -EINVAL ; - goto err_out; - } - }else { +if (!irq_is_percpu_devid (irq)) {unsigned long irq_flags; err =irq_force_affinity (irq,cpumask_of (cpu));
@@ -572,6 +560,9 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) err =request_irq (irq, handler, irq_flags,"arm-pmu" ,per_cpu_ptr (&hw_events->percpu_pmu, cpu)); + }else if (cpumask_empty (&armpmu->active_irqs)) { + err =request_percpu_irq (irq, handler,"arm-pmu" , + &hw_events->percpu_pmu); }if (err)
drivers/perf/arm_pmu_acpi.c +38 -4
68 lines not on this page · raw message
drivers/perf/arm_pmu_platform.c +0 -7
@@ -127,13 +127,6 @@ static int pmu_parse_irqs(struct arm_pmu *pmu) pdev->dev.of_node); } -/* -* Some platforms have all PMU IRQs OR'd into a single IRQ, with a -* special platdata function that attempts to demux them. -*/ -if (dev_get_platdata (&pdev->dev)) -cpumask_setall (&pmu->supported_cpus); -for (i =0 ; i < num_irqs; i++) {int cpu, irq;
Signature · 1 line
2.11.0
Mark Rutland
maintainerPATCH v2 6/8 arm_pmu: explicitly enable/disable SPIs at hotplug
In reply to Mark RutlandTo support ACPI systems, we need to request IRQs before CPUs are hotplugged, and thus we need to request IRQs before we know their associated PMU.
This is problematic if a PMU IRQ is pending out of reset, as it may be taken before we know the PMU, and thus the IRQ handler won't be able to handle it, leaving it screaming.
To avoid such problems, lets request all IRQs in a disabled state, and explicitly enable/disable them at hotplug time, when we're sure the PMU has been probed.
--- drivers/perf/arm_pmu.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)
drivers/perf/arm_pmu.c +10 -5
@@ -558,6 +558,7 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)IRQF_NOBALANCING |IRQF_NO_THREAD ; +irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" ,per_cpu_ptr (&hw_events->percpu_pmu, cpu)); }else if (cpumask_empty (&armpmu->active_irqs)) {
@@ -600,10 +601,10 @@ static int arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node) irq =armpmu_get_cpu_irq (pmu, cpu);if (irq) { -if (irq_is_percpu_devid (irq)) { +if (irq_is_percpu_devid (irq))enable_percpu_irq (irq,IRQ_TYPE_NONE ); -return 0 ; - } +else +enable_irq (irq); }return 0 ;
@@ -618,8 +619,12 @@ static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)return 0 ; irq =armpmu_get_cpu_irq (pmu, cpu); -if (irq &&irq_is_percpu_devid (irq)) -disable_percpu_irq (irq); +if (irq) { +if (irq_is_percpu_devid (irq)) +disable_percpu_irq (irq); +else +disable_irq (irq); + }return 0 ; }
Signature · 1 line
2.11.0
PATCH v2 6/8 arm_pmu: explicitly enable/disable SPIs at hotplug
Hi Mark,
Quoted 8 lines · Mark Rutland
To support ACPI systems, we need to request IRQs before CPUs are hotplugged, and thus we need to request IRQs before we know their associated PMU.
This is problematic if a PMU IRQ is pending out of reset, as it may be taken before we know the PMU, and thus the IRQ handler won't be able to handle it, leaving it screaming.
Quoted 5 lines
To avoid such problems, lets request all IRQs in a disabled state, and explicitly enable/disable them at hotplug time, when we're sure the PMU has been probed.
Signed-off-by Mark Rutland <[email protected]>
This is now commit 6de3f79112cc26bf in v4.16-rc3, and causes a BUG during CPU offlining (e.g. during system suspend, or during boot with CONFIG_ARM_PSCI_CHECKER=y).
With CONFIG_ARM_PSCI_CHECKER=y:
psci_checker: PSCI checker started using 6 CPUspsci_checker: Starting hotplug testspsci_checker: Trying to turn off and on again all CPUs
BUG: sleeping function called from invalid context at kernel/irq/manage.c:112 in_atomic(): 1, irqs_disabled(): 128, pid: 15, name: migration/1 no locks held by migration/1/15. irq event stamp: 192 hardirqs last enabled at (191): [<00000000803c2507>] _raw_spin_unlock_irq+0x2c/0x4c hardirqs last disabled at (192): [<000000007f57ad28>] multi_cpu_stop+0x9c/0x140 softirqs last enabled at (0): [<0000000004ee1b58>] copy_process.isra.77.part.78+0x43c/0x1504 softirqs last disabled at (0): [< (null)>] (null) CPU: 1 PID: 15 Comm: migration/1 Not tainted 4.16.0-rc3-salvator-x #1651 Hardware name: Renesas Salvator-X board based on r8a7796 (DT) Call trace: dump_backtrace+0x0/0x140 show_stack+0x14/0x1c dump_stack+0xb4/0xf0 ___might_sleep+0x1fc/0x218 __might_sleep+0x70/0x80 synchronize_irq+0x40/0xa8 disable_irq+0x20/0x2c arm_perf_teardown_cpu+0x80/0xac cpuhp_invoke_callback+0x5a0/0xd18 take_cpu_down+0x84/0xc4 multi_cpu_stop+0xb0/0x140 cpu_stopper_thread+0xbc/0x128 smpboot_thread_fn+0x218/0x234 kthread+0x11c/0x124 ret_from_fork+0x10/0x18
CPU1: shutdown
psci: CPU1 killed.
Reverting that commit on v4.16-rc3 fixes the issue.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
PATCH v2 6/8 arm_pmu: explicitly enable/disable SPIs at hotplug
Quoted 36 lines · Geert Uytterhoeven
Quoted 8 lines · Mark Rutland
To support ACPI systems, we need to request IRQs before CPUs are hotplugged, and thus we need to request IRQs before we know their associated PMU.
This is problematic if a PMU IRQ is pending out of reset, as it may be taken before we know the PMU, and thus the IRQ handler won't be able to handle it, leaving it screaming.
Quoted 5 lines
To avoid such problems, lets request all IRQs in a disabled state, and explicitly enable/disable them at hotplug time, when we're sure the PMU has been probed.
Signed-off-by Mark Rutland <[email protected]>This is now commit
6de3f79112cc26bfin v4.16-rc3, and causes a BUG during CPU offlining (e.g. during system suspend, or during boot withCONFIG_ARM_PSCI_CHECKER=y).With
CONFIG_ARM_PSCI_CHECKER=y:
psci_checker: PSCI checker started using 6 CPUspsci_checker: Starting hotplug testspsci_checker: Trying to turn off and on again all CPUsBUG: sleeping function called from invalid context at kernel/irq/manage.c:112 in_atomic(): 1, irqs_disabled(): 128, pid: 15, name: migration/1 no locks held by migration/1/15. irq event stamp: 192 hardirqs last enabled at (191): [<00000000803c2507>] _raw_spin_unlock_irq+0x2c/0x4c hardirqs last disabled at (192): [<000000007f57ad28>] multi_cpu_stop+0x9c/0x140 softirqs last enabled at (0): [<0000000004ee1b58>] copy_process.isra.77.part.78+0x43c/0x1504 softirqs last disabled at (0): [< (null)>] (null) CPU: 1 PID: 15 Comm: migration/1 Not tainted 4.16.0-rc3-salvator-x #1651 Hardware name: Renesas Salvator-X board based on r8a7796 (DT)
Quoted 8 lines
Call trace: dump_backtrace+0x0/0x140 show_stack+0x14/0x1c dump_stack+0xb4/0xf0 ___might_sleep+0x1fc/0x218 __might_sleep+0x70/0x80 synchronize_irq+0x40/0xa8 disable_irq+0x20/0x2c
Given that these things are CPU-affine, I reckon this should be disable_irq_nosync. Mark?
Will
--->8
drivers/perf/arm_pmu.c +1 -1
@@ -638,7 +638,7 @@ static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)if (irq_is_percpu_devid (irq))disable_percpu_irq (irq);else -disable_irq (irq); +disable_irq_nosync (irq); }per_cpu (cpu_armpmu, cpu) =NULL ;
PATCH v2 6/8 arm_pmu: explicitly enable/disable SPIs at hotplug
Hi Will,
Quoted 58 lines · Will Deacon
Quoted 36 lines · Geert Uytterhoeven
Quoted 8 lines · Mark Rutland
To support ACPI systems, we need to request IRQs before CPUs are hotplugged, and thus we need to request IRQs before we know their associated PMU.
This is problematic if a PMU IRQ is pending out of reset, as it may be taken before we know the PMU, and thus the IRQ handler won't be able to handle it, leaving it screaming.
Quoted 5 lines
To avoid such problems, lets request all IRQs in a disabled state, and explicitly enable/disable them at hotplug time, when we're sure the PMU has been probed.
Signed-off-by Mark Rutland <[email protected]>This is now commit
6de3f79112cc26bfin v4.16-rc3, and causes a BUG during CPU offlining (e.g. during system suspend, or during boot withCONFIG_ARM_PSCI_CHECKER=y).With
CONFIG_ARM_PSCI_CHECKER=y:
psci_checker: PSCI checker started using 6 CPUspsci_checker: Starting hotplug testspsci_checker: Trying to turn off and on again all CPUsBUG: sleeping function called from invalid context at kernel/irq/manage.c:112 in_atomic(): 1, irqs_disabled(): 128, pid: 15, name: migration/1 no locks held by migration/1/15. irq event stamp: 192 hardirqs last enabled at (191): [<00000000803c2507>] _raw_spin_unlock_irq+0x2c/0x4c hardirqs last disabled at (192): [<000000007f57ad28>] multi_cpu_stop+0x9c/0x140 softirqs last enabled at (0): [<0000000004ee1b58>] copy_process.isra.77.part.78+0x43c/0x1504 softirqs last disabled at (0): [< (null)>] (null) CPU: 1 PID: 15 Comm: migration/1 Not tainted 4.16.0-rc3-salvator-x #1651 Hardware name: Renesas Salvator-X board based on r8a7796 (DT)Quoted 8 lines
Call trace: dump_backtrace+0x0/0x140 show_stack+0x14/0x1c dump_stack+0xb4/0xf0 ___might_sleep+0x1fc/0x218 __might_sleep+0x70/0x80 synchronize_irq+0x40/0xa8 disable_irq+0x20/0x2cGiven that these things are CPU-affine, I reckon this should be
disable_irq_nosync. Mark?Will
--->8
drivers/perf/arm_pmu.c +0
-0
drivers/perf/arm_pmu.c static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node) · line 644
Quoted 8 lines
drivers/perf/arm_pmu.c +1
-1@@ -638,7 +638,7 @@ static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)if (irq_is_percpu_devid (irq))disable_percpu_irq (irq);else -disable_irq (irq); +disable_irq_nosync (irq); }per_cpu (cpu_armpmu, cpu) =NULL ;
Thanks!
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Mark Rutland
maintainerPATCH v2 6/8 arm_pmu: explicitly enable/disable SPIs at hotplug
In reply to Will DeaconQuoted 46 lines · Will Deacon
Quoted 36 lines · Geert Uytterhoeven
Quoted 8 lines · Mark Rutland
To support ACPI systems, we need to request IRQs before CPUs are hotplugged, and thus we need to request IRQs before we know their associated PMU.
This is problematic if a PMU IRQ is pending out of reset, as it may be taken before we know the PMU, and thus the IRQ handler won't be able to handle it, leaving it screaming.
Quoted 5 lines
To avoid such problems, lets request all IRQs in a disabled state, and explicitly enable/disable them at hotplug time, when we're sure the PMU has been probed.
Signed-off-by Mark Rutland <[email protected]>This is now commit
6de3f79112cc26bfin v4.16-rc3, and causes a BUG during CPU offlining (e.g. during system suspend, or during boot withCONFIG_ARM_PSCI_CHECKER=y).With
CONFIG_ARM_PSCI_CHECKER=y:
psci_checker: PSCI checker started using 6 CPUspsci_checker: Starting hotplug testspsci_checker: Trying to turn off and on again all CPUsBUG: sleeping function called from invalid context at kernel/irq/manage.c:112 in_atomic(): 1, irqs_disabled(): 128, pid: 15, name: migration/1 no locks held by migration/1/15. irq event stamp: 192 hardirqs last enabled at (191): [<00000000803c2507>] _raw_spin_unlock_irq+0x2c/0x4c hardirqs last disabled at (192): [<000000007f57ad28>] multi_cpu_stop+0x9c/0x140 softirqs last enabled at (0): [<0000000004ee1b58>] copy_process.isra.77.part.78+0x43c/0x1504 softirqs last disabled at (0): [< (null)>] (null) CPU: 1 PID: 15 Comm: migration/1 Not tainted 4.16.0-rc3-salvator-x #1651 Hardware name: Renesas Salvator-X board based on r8a7796 (DT)Quoted 8 lines
Call trace: dump_backtrace+0x0/0x140 show_stack+0x14/0x1c dump_stack+0xb4/0xf0 ___might_sleep+0x1fc/0x218 __might_sleep+0x70/0x80 synchronize_irq+0x40/0xa8 disable_irq+0x20/0x2c
Quoted 2 lines
Given that these things are CPU-affine, I reckon this should be
disable_irq_nosync. Mark?
Given IRQs are disabled, this should be fine, yes.
FWIW, if you spin this as a patch:
Mark.
Quoted 18 lines
Will
--->8
drivers/perf/arm_pmu.c +1
-1@@ -638,7 +638,7 @@ static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)if (irq_is_percpu_devid (irq))disable_percpu_irq (irq);else -disable_irq (irq); +disable_irq_nosync (irq); }per_cpu (cpu_armpmu, cpu) =NULL ;
Mark Rutland
maintainerPATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
In reply to Mark RutlandTo support ACPI systems, we need to request IRQs before we know the associated PMU, and thus we need some percpu variable that the IRQ handler can find the PMU from.
As we're going to request IRQs without the PMU, we can't rely on the arm_pmu::active_irqs mask, and similarly need to track requested IRQs with a percpu variable.
--- drivers/perf/arm_pmu.c | 76 +++++++++++++++++++++++++++++++------------- include/linux/perf/arm_pmu.h | 1 - 2 files changed, 54 insertions(+), 23 deletions(-)
drivers/perf/arm_pmu.c +54 -22
@@ -25,6 +25,9 @@#include <asm/irq_regs.h> +static DEFINE_PER_CPU(struct arm_pmu *, cpu_armpmu); +static DEFINE_PER_CPU(int, cpu_irq); +static int armpmu_map_cache_event (const unsigned (*cache_map) [PERF_COUNT_HW_CACHE_MAX]
@@ -325,13 +328,9 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)int ret;u64 start_clock, finish_clock; -/* -* we request the IRQ with a (possibly percpu) struct arm_pmu**, but -* the handlers expect a struct arm_pmu*. The percpu_irq framework will -* do any necessary shifting, we just need to perform the first -* dereference. -*/ - armpmu = *(void **)dev; + armpmu =this_cpu_read (cpu_armpmu); +if (WARN_ON_ONCE (!armpmu)) +return IRQ_NONE ; start_clock =sched_clock (); ret = armpmu->handle_irq (irq, armpmu);
@@ -517,29 +516,47 @@ int perf_num_counters(void) }EXPORT_SYMBOL_GPL (perf_num_counters); -void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) +int armpmu_count_irq_users (const int irq) { -struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -int irq =per_cpu (hw_events->irq, cpu); +int cpu, count =0 ; + +for_each_possible_cpu (cpu) { +if (per_cpu (cpu_irq, cpu) == irq) + count++; + } -if (!cpumask_test_and_clear_cpu (cpu, &armpmu->active_irqs)) +return count; +} + +void armpmu_free_cpu_irq (int irq,int cpu) +{ +if (per_cpu (cpu_irq, cpu) ==0 ) +return ; +if (WARN_ON (irq !=per_cpu (cpu_irq, cpu)))return ;if (irq_is_percpu_devid (irq)) { -free_percpu_irq (irq, &hw_events->percpu_pmu); -cpumask_clear (&armpmu->active_irqs); -return ; +if (armpmu_count_irq_users (irq) ==1 ) +free_percpu_irq (irq, &cpu_armpmu); + }else { +free_irq (irq,NULL ); } -free_irq (irq,per_cpu_ptr (&hw_events->percpu_pmu, cpu)); +per_cpu (cpu_irq, cpu) =0 ; } -int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) +void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) { -int err =0 ;struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -const irq_handler_t handler = armpmu_dispatch_irq;int irq =per_cpu (hw_events->irq, cpu); + +armpmu_free_cpu_irq (irq, cpu); +} + +int armpmu_request_cpu_irq (int irq,int cpu) +{ +int err =0 ; +const irq_handler_t handler = armpmu_dispatch_irq;if (!irq)return 0 ;
@@ -560,16 +577,16 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" , -per_cpu_ptr (&hw_events->percpu_pmu, cpu)); - }else if (cpumask_empty (&armpmu->active_irqs)) { +NULL ); + }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); + cpu_armpmu); }if (err) goto err_out; -cpumask_set_cpu (cpu, &armpmu->active_irqs); +per_cpu (cpu_irq, cpu) = irq;return 0 ; err_out:
@@ -577,6 +594,17 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)return err; } +int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) +{ +struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; +int irq =per_cpu (hw_events->irq, cpu); +if (!irq) +return 0 ; + + +return armpmu_request_cpu_irq (irq, cpu); +} +static int armpmu_get_cpu_irq (struct arm_pmu *pmu,int cpu) {struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
@@ -599,6 +627,8 @@ static int arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node)if (pmu->reset) pmu->reset (pmu); +per_cpu (cpu_armpmu, cpu) = pmu; + irq =armpmu_get_cpu_irq (pmu, cpu);if (irq) {if (irq_is_percpu_devid (irq))
@@ -626,6 +656,8 @@ static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)disable_irq (irq); } +per_cpu (cpu_armpmu, cpu) =NULL ; +return 0 ; }
include/linux/perf/arm_pmu.h +0 -1
@@ -75,7 +75,6 @@ enum armpmu_attr_groups {struct arm_pmu {struct pmu pmu; -cpumask_t active_irqs;cpumask_t supported_cpus;char *name;irqreturn_t (*handle_irq)(int irq_num,void *dev);
Signature · 1 line
2.11.0
PATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
Hi Mark,
Just a couple of nits spotted in passing...
Quoted 72 lines · Mark Rutland
To support ACPI systems, we need to request IRQs before we know the associated PMU, and thus we need some percpu variable that the IRQ handler can find the PMU from.
As we're going to request IRQs without the PMU, we can't rely on the
arm_pmu::active_irqsmask, and similarly need to track requested IRQs with a percpu variable.Signed-off-by Mark Rutland <[email protected]>Cc Will Deacon <[email protected]>--- drivers/perf/arm_pmu.c | 76 +++++++++++++++++++++++++++++++------------- include/linux/perf/arm_pmu.h | 1 - 2 files changed, 54 insertions(+), 23 deletions(-)drivers/perf/arm_pmu.c +21
-11@@ -25,6 +25,9 @@#include <asm/irq_regs.h> +static DEFINE_PER_CPU(struct arm_pmu *, cpu_armpmu); +static DEFINE_PER_CPU(int, cpu_irq); +static int armpmu_map_cache_event (const unsigned (*cache_map) [PERF_COUNT_HW_CACHE_MAX]@@ -325,13 +328,9 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)int ret;u64 start_clock, finish_clock; -/* -* we request the IRQ with a (possibly percpu) struct arm_pmu**, but -* the handlers expect a struct arm_pmu*. The percpu_irq framework will -* do any necessary shifting, we just need to perform the first -* dereference. -*/ - armpmu = *(void **)dev; + armpmu =this_cpu_read (cpu_armpmu); +if (WARN_ON_ONCE (!armpmu)) +return IRQ_NONE ; start_clock =sched_clock (); ret = armpmu->handle_irq (irq, armpmu);@@ -517,29 +516,47 @@ int perf_num_counters(void) }EXPORT_SYMBOL_GPL (perf_num_counters); -void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) +int armpmu_count_irq_users (const int irq) { -struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -int irq =per_cpu (hw_events->irq, cpu); +int cpu, count =0 ; + +for_each_possible_cpu (cpu) { +if (per_cpu (cpu_irq, cpu) == irq) + count++; + } -if (!cpumask_test_and_clear_cpu (cpu, &armpmu->active_irqs)) +return count; +} + +void armpmu_free_cpu_irq (int irq,int cpu) +{ +if (per_cpu (cpu_irq, cpu) ==0 ) +return ; +if (WARN_ON (irq !=per_cpu (cpu_irq, cpu)))return ;
drivers/perf/arm_pmu.c int perf_num_counters(void) · line 543
Quoted 10 lines
drivers/perf/arm_pmu.c +4
-3if (irq_is_percpu_devid (irq)) { -free_percpu_irq (irq, &hw_events->percpu_pmu); -cpumask_clear (&armpmu->active_irqs); -return ; +if (armpmu_count_irq_users (irq) ==1 ) +free_percpu_irq (irq, &cpu_armpmu); + }else { +free_irq (irq,NULL ); }
Following the same pattern as the request would make this marginally simpler (and nicely symmetric), i.e.
if (!irq_is_percpu_devid(irq)) free_irq(irq, NULL); else if (armpmu_count_irq_users(irq) == 1) free_percpu_irq(irq, &cpu_armpmu);
Quoted 45 lines
drivers/perf/arm_pmu.c +14
-8-free_irq (irq,per_cpu_ptr (&hw_events->percpu_pmu, cpu)); +per_cpu (cpu_irq, cpu) =0 ; } -int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) +void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) { -int err =0 ;struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -const irq_handler_t handler = armpmu_dispatch_irq;int irq =per_cpu (hw_events->irq, cpu); + +armpmu_free_cpu_irq (irq, cpu); +} + +int armpmu_request_cpu_irq (int irq,int cpu) +{ +int err =0 ; +const irq_handler_t handler = armpmu_dispatch_irq;if (!irq)return 0 ;@@ -560,16 +577,16 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" , -per_cpu_ptr (&hw_events->percpu_pmu, cpu)); - }else if (cpumask_empty (&armpmu->active_irqs)) { +NULL ); + }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); + cpu_armpmu); }if (err) goto err_out; -cpumask_set_cpu (cpu, &armpmu->active_irqs); +per_cpu (cpu_irq, cpu) = irq;return 0 ; err_out:@@ -577,6 +594,17 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)return err;
drivers/perf/arm_pmu.c int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) · line 604
Quoted 10 lines
drivers/perf/arm_pmu.c +8
-0} +int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) +{ +struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; +int irq =per_cpu (hw_events->irq, cpu); +if (!irq) +return 0 ; + +
Double blank line here.
Robin.
Quoted 37 lines
drivers/perf/arm_pmu.c +7
-0+return armpmu_request_cpu_irq (irq, cpu); +} +static int armpmu_get_cpu_irq (struct arm_pmu *pmu,int cpu) {struct pmu_hw_events __percpu *hw_events = pmu->hw_events;@@ -599,6 +627,8 @@ static int arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node)if (pmu->reset) pmu->reset (pmu); +per_cpu (cpu_armpmu, cpu) = pmu; + irq =armpmu_get_cpu_irq (pmu, cpu);if (irq) {if (irq_is_percpu_devid (irq))@@ -626,6 +656,8 @@ static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)disable_irq (irq); } +per_cpu (cpu_armpmu, cpu) =NULL ; +return 0 ; }include/linux/perf/arm_pmu.h +0
-1@@ -75,7 +75,6 @@ enum armpmu_attr_groups {struct arm_pmu {struct pmu pmu; -cpumask_t active_irqs;cpumask_t supported_cpus;char *name;irqreturn_t (*handle_irq)(int irq_num,void *dev);
Mark Rutland
maintainerPATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
Quoted 25 lines · Robin Murphy
Hi Mark,
Just a couple of nits spotted in passing...
Quoted 5 lines · Mark Rutland
drivers/perf/arm_pmu.c +5
-0+void armpmu_free_cpu_irq (int irq,int cpu) +{ +if (per_cpu (cpu_irq, cpu) ==0 ) +return ; +if (WARN_ON (irq !=per_cpu (cpu_irq, cpu)))drivers/perf/arm_pmu.c int perf_num_counters(void) · line 543
Quoted 10 lines
drivers/perf/arm_pmu.c +4
-3return ;if (irq_is_percpu_devid (irq)) { -free_percpu_irq (irq, &hw_events->percpu_pmu); -cpumask_clear (&armpmu->active_irqs); -return ; +if (armpmu_count_irq_users (irq) ==1 ) +free_percpu_irq (irq, &cpu_armpmu); + }else { +free_irq (irq,NULL ); }Following the same pattern as the request would make this marginally simpler (and nicely symmetric), i.e.
Quoted 4 lines
if (!irq_is_percpu_devid(irq)) free_irq(irq, NULL); else if (armpmu_count_irq_users(irq) == 1) free_percpu_irq(irq, &cpu_armpmu);
Sure; done.
Quoted 10 lines
drivers/perf/arm_pmu.c int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) · line 604
Quoted 8 lines
drivers/perf/arm_pmu.c +8
-0+int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) +{ +struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; +int irq =per_cpu (hw_events->irq, cpu); +if (!irq) +return 0 ; + +Double blank line here.
... and gone.
Thanks,
Mark.
PATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
In reply to Mark RutlandQuoted 32 lines · Mark Rutland
drivers/perf/arm_pmu.c +6
-7@@ -25,6 +25,9 @@#include <asm/irq_regs.h> +static DEFINE_PER_CPU(struct arm_pmu *, cpu_armpmu); +static DEFINE_PER_CPU(int, cpu_irq); +static int armpmu_map_cache_event (const unsigned (*cache_map) [PERF_COUNT_HW_CACHE_MAX]@@ -325,13 +328,9 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)int ret;u64 start_clock, finish_clock; -/* -* we request the IRQ with a (possibly percpu) struct arm_pmu**, but -* the handlers expect a struct arm_pmu*. The percpu_irq framework will -* do any necessary shifting, we just need to perform the first -* dereference. -*/ - armpmu = *(void **)dev; + armpmu =this_cpu_read (cpu_armpmu); +if (WARN_ON_ONCE (!armpmu)) +return IRQ_NONE ; start_clock =sched_clock (); ret = armpmu->handle_irq (irq, armpmu);
drivers/perf/arm_pmu.c int perf_num_counters(void) · line 519
Quoted 5 lines
drivers/perf/arm_pmu.c +1
-1@@ -517,29 +516,47 @@ int perf_num_counters(void) }EXPORT_SYMBOL_GPL (perf_num_counters); -void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) +int armpmu_count_irq_users (const int irq)
This can be static.
Quoted 54 lines
drivers/perf/arm_pmu.c +28
-10{ -struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -int irq =per_cpu (hw_events->irq, cpu); +int cpu, count =0 ; + +for_each_possible_cpu (cpu) { +if (per_cpu (cpu_irq, cpu) == irq) + count++; + } -if (!cpumask_test_and_clear_cpu (cpu, &armpmu->active_irqs)) +return count; +} + +void armpmu_free_cpu_irq (int irq,int cpu) +{ +if (per_cpu (cpu_irq, cpu) ==0 ) +return ; +if (WARN_ON (irq !=per_cpu (cpu_irq, cpu)))return ;if (irq_is_percpu_devid (irq)) { -free_percpu_irq (irq, &hw_events->percpu_pmu); -cpumask_clear (&armpmu->active_irqs); -return ; +if (armpmu_count_irq_users (irq) ==1 ) +free_percpu_irq (irq, &cpu_armpmu); + }else { +free_irq (irq,NULL ); } -free_irq (irq,per_cpu_ptr (&hw_events->percpu_pmu, cpu)); +per_cpu (cpu_irq, cpu) =0 ; } -int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) +void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) { -int err =0 ;struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -const irq_handler_t handler = armpmu_dispatch_irq;int irq =per_cpu (hw_events->irq, cpu); + +armpmu_free_cpu_irq (irq, cpu); +} + +int armpmu_request_cpu_irq (int irq,int cpu) +{ +int err =0 ; +const irq_handler_t handler = armpmu_dispatch_irq;if (!irq)return 0 ;
drivers/perf/arm_pmu.c int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) · line 583
Quoted 10 lines
drivers/perf/arm_pmu.c +3
-3@@ -560,16 +577,16 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" , -per_cpu_ptr (&hw_events->percpu_pmu, cpu)); - }else if (cpumask_empty (&armpmu->active_irqs)) { +NULL ); + }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); + cpu_armpmu);
This should be &cpu_armpmu.
Would it be possible to pass &cpu_armpmu as the devid even in the normal request_irq case and have the dispatcher just pass it through to the underlying handler, rather than access cpu_armpmu directly?
Will
Mark Rutland
maintainerPATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
Quoted 36 lines · Will Deacon
Quoted 35 lines · Mark Rutland
drivers/perf/arm_pmu.c +6
-7@@ -25,6 +25,9 @@#include <asm/irq_regs.h> +static DEFINE_PER_CPU(struct arm_pmu *, cpu_armpmu); +static DEFINE_PER_CPU(int, cpu_irq); +static int armpmu_map_cache_event (const unsigned (*cache_map) [PERF_COUNT_HW_CACHE_MAX]@@ -325,13 +328,9 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)int ret;u64 start_clock, finish_clock; -/* -* we request the IRQ with a (possibly percpu) struct arm_pmu**, but -* the handlers expect a struct arm_pmu*. The percpu_irq framework will -* do any necessary shifting, we just need to perform the first -* dereference. -*/ - armpmu = *(void **)dev; + armpmu =this_cpu_read (cpu_armpmu); +if (WARN_ON_ONCE (!armpmu)) +return IRQ_NONE ; start_clock =sched_clock (); ret = armpmu->handle_irq (irq, armpmu);@@ -517,29 +516,47 @@ int perf_num_counters(void) }EXPORT_SYMBOL_GPL (perf_num_counters);
Quoted 4 lines
drivers/perf/arm_pmu.c int perf_num_counters(void) · line 519
Quoted 2 lines
drivers/perf/arm_pmu.c +1
-1-void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) +int armpmu_count_irq_users (const int irq)This can be static.
sure.
Quoted 13 lines
drivers/perf/arm_pmu.c int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) · line 583
Quoted 11 lines
drivers/perf/arm_pmu.c +3
-3@@ -560,16 +577,16 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" , -per_cpu_ptr (&hw_events->percpu_pmu, cpu)); - }else if (cpumask_empty (&armpmu->active_irqs)) { +NULL ); + }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); + cpu_armpmu);This should be
&cpu_armpmu.
I'm not sure that's the case, given the way we statically define cpu_armpmu, but I'll try to figure that out.
Quoted 3 lines
Would it be possible to pass
&cpu_armpmuas the devid even in the normalrequest_irqcase and have the dispatcher just pass it through to the underlying handler, rather than accesscpu_armpmudirectly?
Do you mean so that the dispatcher takes a struct arm_pmu ** in all cases?
Otherwise, we don't have the struct pmu * early enough in the ACPI case.
Thanks,
Mark.
PATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
Quoted 16 lines · Mark Rutland
Quoted 14 lines · Will Deacon
drivers/perf/arm_pmu.c int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) · line 583
Quoted 11 lines · Mark Rutland
drivers/perf/arm_pmu.c +3
-3@@ -560,16 +577,16 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" , -per_cpu_ptr (&hw_events->percpu_pmu, cpu)); - }else if (cpumask_empty (&armpmu->active_irqs)) { +NULL ); + }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); + cpu_armpmu);This should be
&cpu_armpmu.
Quoted 2 lines
I'm not sure that's the case, given the way we statically define
cpu_armpmu, but I'll try to figure that out.
Well that's what you pass to free_percpu_irq :p
Quoted 6 lines
Quoted 3 lines
Would it be possible to pass
&cpu_armpmuas the devid even in the normalrequest_irqcase and have the dispatcher just pass it through to the underlying handler, rather than accesscpu_armpmudirectly?Do you mean so that the dispatcher takes a struct
arm_pmu** in all cases?
Yes.
Will
Mark Rutland
maintainerPATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
Quoted 17 lines · Will Deacon
Quoted 16 lines · Mark Rutland
Quoted 14 lines · Will Deacon
drivers/perf/arm_pmu.c int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) · line 583
Quoted 11 lines · Mark Rutland
drivers/perf/arm_pmu.c +3
-3@@ -560,16 +577,16 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" , -per_cpu_ptr (&hw_events->percpu_pmu, cpu)); - }else if (cpumask_empty (&armpmu->active_irqs)) { +NULL ); + }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); + cpu_armpmu);This should be
&cpu_armpmu.
Quoted 4 lines
Quoted 2 lines
I'm not sure that's the case, given the way we statically define
cpu_armpmu, but I'll try to figure that out.Well that's what you pass to
free_percpu_irq:p
Ugh. One of the two is clearly wrnog...
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Would it be possible to pass
&cpu_armpmuas the devid even in the normalrequest_irqcase and have the dispatcher just pass it through to the underlying handler, rather than accesscpu_armpmudirectly?Do you mean so that the dispatcher takes a struct
arm_pmu** in all cases?Yes.
Ok. I'll try to piece that together (which should allow us to revert some changes to the dispatcher).
Thanks,
Mark.
Mark Rutland
maintainerPATCH v2 7/8 arm_pmu: note IRQs and PMUs per-cpu
In reply to Will DeaconQuoted 17 lines · Will Deacon
Quoted 16 lines · Mark Rutland
Quoted 14 lines · Will Deacon
drivers/perf/arm_pmu.c int armpmu_request_irq(struct arm_pmu *armpmu, int cpu) · line 583
Quoted 11 lines · Mark Rutland
drivers/perf/arm_pmu.c +3
-3@@ -560,16 +577,16 @@ int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)irq_set_status_flags (irq,IRQ_NOAUTOEN ); err =request_irq (irq, handler, irq_flags,"arm-pmu" , -per_cpu_ptr (&hw_events->percpu_pmu, cpu)); - }else if (cpumask_empty (&armpmu->active_irqs)) { +NULL ); + }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , - &hw_events->percpu_pmu); + cpu_armpmu);This should be
&cpu_armpmu.
Quoted 4 lines
Quoted 2 lines
I'm not sure that's the case, given the way we statically define
cpu_armpmu, but I'll try to figure that out.Well that's what you pass to
free_percpu_irq:p
Ugh; the mismatch is clearly a bug...
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Would it be possible to pass
&cpu_armpmuas the devid even in the normalrequest_irqcase and have the dispatcher just pass it through to the underlying handler, rather than accesscpu_armpmudirectly?Do you mean so that the dispatcher takes a struct
arm_pmu** in all cases?Yes.
Done, and pushed out to my arm64/acpi-pmu-lockdep branch.
Let me know if you want me to repost the series.
Mark.
Mark Rutland
maintainerPATCH v2 8/8 arm_pmu: acpi: request IRQs up-front
In reply to Mark RutlandWe can't request IRQs in atomic context, so for ACPI systems we'll have to request them up-front, and later associate them with CPUs.
This patch reorganises the arm_pmu code to do so. As we no longer have the arm_pmu structure at probe time, a number of prototypes need to be adjusted, requiring changes to the common arm_pmu code and arm_pmu platform code.
--- drivers/perf/arm_pmu.c | 26 +++----------------------- drivers/perf/arm_pmu_acpi.c | 19 ++++++------------- drivers/perf/arm_pmu_platform.c | 15 ++++++++++++--- include/linux/perf/arm_pmu.h | 5 +++-- 4 files changed, 24 insertions(+), 41 deletions(-)
drivers/perf/arm_pmu.c +3 -23
@@ -528,7 +528,7 @@ int armpmu_count_irq_users(const int irq)return count; } -void armpmu_free_cpu_irq (int irq,int cpu) +void armpmu_free_irq (int irq,int cpu) {if (per_cpu (cpu_irq, cpu) ==0 )return ;
@@ -545,15 +545,7 @@ void armpmu_free_cpu_irq(int irq, int cpu)per_cpu (cpu_irq, cpu) =0 ; } -void armpmu_free_irq (struct arm_pmu *armpmu,int cpu) -{ -struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -int irq =per_cpu (hw_events->irq, cpu); - -armpmu_free_cpu_irq (irq, cpu); -} - -int armpmu_request_cpu_irq (int irq,int cpu) +int armpmu_request_irq (int irq,int cpu) {int err =0 ;const irq_handler_t handler = armpmu_dispatch_irq;
@@ -576,8 +568,7 @@ int armpmu_request_cpu_irq(int irq, int cpu)IRQF_NO_THREAD ;irq_set_status_flags (irq,IRQ_NOAUTOEN ); - err =request_irq (irq, handler, irq_flags,"arm-pmu" , -NULL ); + err =request_irq (irq, handler, irq_flags,"arm-pmu" ,NULL ); }else if (armpmu_count_irq_users (irq) ==0 ) { err =request_percpu_irq (irq, handler,"arm-pmu" , cpu_armpmu);
@@ -594,17 +585,6 @@ int armpmu_request_cpu_irq(int irq, int cpu)return err; } -int armpmu_request_irq (struct arm_pmu *armpmu,int cpu) -{ -struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -int irq =per_cpu (hw_events->irq, cpu); -if (!irq) -return 0 ; - - -return armpmu_request_cpu_irq (irq, cpu); -} -static int armpmu_get_cpu_irq (struct arm_pmu *pmu,int cpu) {struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
drivers/perf/arm_pmu_acpi.c +6 -13
@@ -89,7 +89,13 @@ static int arm_pmu_acpi_parse_irqs(void)pr_warn ("No ACPI PMU IRQ for CPU%d\n" , cpu); } +/* +* Log and request the IRQ so the core arm_pmu code can manage +* it. We'll have to sanity-check IRQs later when we associate +* them with their PMUs. +*/ per_cpu (pmu_irqs, cpu) = irq; +armpmu_request_irq (irq, cpu); }return 0 ;
@@ -205,14 +211,6 @@ static int arm_pmu_acpi_cpu_starting(unsigned int cpu)cpumask_set_cpu (cpu, &pmu->supported_cpus);/* -* Log and request the IRQ so the core arm_pmu code can manage it. In -* some situations (e.g. mismatched PPIs), we may fail to request the -* IRQ. However, it may be too late for us to do anything about it. -* The common ARM PMU code will log a warning in this case. -*/ -armpmu_request_irq (pmu, cpu); - -/* * Ideally, we'd probe the PMU here when we find the first matching * CPU. We can't do that for several reasons; see the comment in * arm_pmu_acpi_init().
@@ -281,11 +279,6 @@ static int arm_pmu_acpi_init(void)if (acpi_disabled)return 0 ; -/* -* We can't request IRQs yet, since we don't know the cookie value -* until we know which CPUs share the same logical PMU. We'll handle -* that in arm_pmu_acpi_cpu_starting(). -*/ ret =arm_pmu_acpi_parse_irqs ();if (ret)return ret;
drivers/perf/arm_pmu_platform.c +12 -3
@@ -159,10 +159,15 @@ static int pmu_parse_irqs(struct arm_pmu *pmu)static int armpmu_request_irqs (struct arm_pmu *armpmu) { +struct pmu_hw_events __percpu *hw_events = armpmu->hw_events;int cpu, err;for_each_cpu (cpu, &armpmu->supported_cpus) { - err =armpmu_request_irq (armpmu, cpu); +int irq =per_cpu (hw_events->irq, cpu); +if (!irq) +continue ; + + err =armpmu_request_irq (irq, cpu);if (err)break ; }
@@ -173,9 +178,13 @@ static int armpmu_request_irqs(struct arm_pmu *armpmu)static void armpmu_free_irqs (struct arm_pmu *armpmu) {int cpu; +struct pmu_hw_events __percpu *hw_events = armpmu->hw_events; -for_each_cpu (cpu, &armpmu->supported_cpus) -armpmu_free_irq (armpmu, cpu); +for_each_cpu (cpu, &armpmu->supported_cpus) { +int irq =per_cpu (hw_events->irq, cpu); + +armpmu_free_irq (irq, cpu); + } } int arm_pmu_device_probe(struct platform_device *pdev,
include/linux/perf/arm_pmu.h +3 -2
@@ -14,6 +14,7 @@#include <linux/interrupt.h> #include <linux/perf_event.h> +#include <linux/platform_device.h> #include <linux/sysfs.h> #include <asm/cputype.h>
@@ -159,8 +160,8 @@ struct arm_pmu *armpmu_alloc(void);struct arm_pmu *armpmu_alloc_atomic (void );void armpmu_free (struct arm_pmu *pmu);int armpmu_register (struct arm_pmu *pmu); -int armpmu_request_irq (struct arm_pmu *armpmu,int cpu); -void armpmu_free_irq (struct arm_pmu *armpmu,int cpu); +int armpmu_request_irq (int irq,int cpu); +void armpmu_free_irq (int irq,int cpu);#define ARMV8_PMU_PDEV_NAME "armv8-pmu"
Signature · 1 line
2.11.0