Re: [PATCH 4/7] dt-bindings: chosen: Add clocksource and clockevent selection
On Tue, Sep 10, 2019 at 4:11 PM Alexandre Belloni wrote: > On 10/09/2019 16:08:26+0100, Sudeep Holla wrote: > > On Tue, Sep 10, 2019 at 02:51:50PM +, claudiu.bez...@microchip.com > > wrote: > > In that case, why can't we identify capability that with the compatibles > > for this timer IP ? > > > > IOW, I don't like the proposal as it's hardware limitation. > > To be clear, bot timers are exactly the same but can't be clocksource > and clockevent at the same time. Why would we have different compatibles > for the exact same IP? In that case why not just pick the first one you find as clocksource and the second one as clock event? As they all come to the same timer of init function two simple local state variables can solve that: static bool registered_clocksource; static bool registered_clockevent; probe(timer) { if (!registered_clocksource) { register_clocksource(timer); registrered_clocksource = true; return; } if (!registered_clockevent) { register_clockevent(timer); registered_clockevent = true; return; } pr_info("surplus timer %p\n", timer); } Clocksource and clockevent are natural singletons so there is no need to handle more than one of each in a driver for identical hardware. With the Integrator AP timer there is a real reason to select one over the other but as I replied to that patch it is pretty easy to just identify which block has this limitation by simply commenting out the IRQ line for it from the device tree. Maybe there is something about this I don't understand. Yours, Linus Walleij ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 7/7] clocksource/drivers/integrator-ap: parse the chosen node
On Tue, Sep 10, 2019 at 2:50 PM Claudiu Beznea wrote: > From: Alexandre Belloni > > The driver currently uses aliases to know whether the timer is the > clocksource or the clockevent. OK maybe that wasn't the most elegant solution. > Add the /chosen/linux,clocksource and > /chosen/linux,clockevent parsing while keeping backward compatibility. This is not how I would solve this today. I would simply remove/comment out the IRQ from the timer that cannot be used for clockevent from the device tree (apparently it doesn't work anyway), and make the code only pick a timer with a valid interrupt assigned as clock event, while a timer without interrupt can be used for clock source. This has the upside of not needing any special aliases or chosen things. Yours, Linus Walleij ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 0/7] add support for clocksource/clockevent DT selection
On Tue, Sep 10, 2019 at 6:47 AM Claudiu Beznea wrote: > > This series adds support to permit the selection of clocksource/clockevent > via DT. Sorry about this, but could you try to include more of a rational for *why* this would be useful in your cover-letter/commit messages? I'm not sure I understand the limitation that requires such an option to be added to the dts. thanks -john ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 4/7] dt-bindings: chosen: Add clocksource and clockevent selection
On 10/09/2019 16:08:26+0100, Sudeep Holla wrote: > On Tue, Sep 10, 2019 at 02:51:50PM +, claudiu.bez...@microchip.com wrote: > > > > > > On 10.09.2019 17:32, Sudeep Holla wrote: > > > External E-Mail > > > > > > > > > On Tue, Sep 10, 2019 at 04:47:13PM +0300, Claudiu Beznea wrote: > > >> From: Alexandre Belloni > > >> > > >> Some timer drivers may behave either as clocksource or clockevent > > >> or both. Until now, in case of platforms with multiple hardware > > >> resources of the same type, the drivers were chosing the first > > >> registered hardware resource as clocksource/clockevent and the > > >> next one as clockevent/clocksource. Other were using different > > >> compatibles (one for each functionality, although its about the > > >> same hardware). Add DT bindings to be able to choose the > > >> functionality of a timer. > > >> > > > > > > Is the piece of hardware not capable of serving as both clocksource > > > and clockevent or is it just the platform choice ? > > > > In my case, the hardware is not capable of serving at the same time > > a clocksource device and a clockevent device. > > > > First, I published v1 for a hardware device having this behavior at > > [1] requesting 1st probed hardware device to work as clocksource and > > the 2nd one to work as clockevent. The discussion at [1] ended up with > > the idea of having a mechanism to specify which hardware device behaves > > as clocksource and which one behaves as clockevent. > > > > In that case, why can't we identify capability that with the compatibles > for this timer IP ? > > IOW, I don't like the proposal as it's hardware limitation. > To be clear, bot timers are exactly the same but can't be clocksource and clockevent at the same time. Why would we have different compatibles for the exact same IP? -- Alexandre Belloni, Bootlin Embedded Linux and Kernel engineering https://bootlin.com ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 4/7] dt-bindings: chosen: Add clocksource and clockevent selection
On Tue, Sep 10, 2019 at 02:51:50PM +, claudiu.bez...@microchip.com wrote: > > > On 10.09.2019 17:32, Sudeep Holla wrote: > > External E-Mail > > > > > > On Tue, Sep 10, 2019 at 04:47:13PM +0300, Claudiu Beznea wrote: > >> From: Alexandre Belloni > >> > >> Some timer drivers may behave either as clocksource or clockevent > >> or both. Until now, in case of platforms with multiple hardware > >> resources of the same type, the drivers were chosing the first > >> registered hardware resource as clocksource/clockevent and the > >> next one as clockevent/clocksource. Other were using different > >> compatibles (one for each functionality, although its about the > >> same hardware). Add DT bindings to be able to choose the > >> functionality of a timer. > >> > > > > Is the piece of hardware not capable of serving as both clocksource > > and clockevent or is it just the platform choice ? > > In my case, the hardware is not capable of serving at the same time > a clocksource device and a clockevent device. > > First, I published v1 for a hardware device having this behavior at > [1] requesting 1st probed hardware device to work as clocksource and > the 2nd one to work as clockevent. The discussion at [1] ended up with > the idea of having a mechanism to specify which hardware device behaves > as clocksource and which one behaves as clockevent. > In that case, why can't we identify capability that with the compatibles for this timer IP ? IOW, I don't like the proposal as it's hardware limitation. -- Regards, Sudeep ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 2/7] clocksource: change timer registration macros
On 10.09.2019 17:49, Marc Zyngier wrote: > External E-Mail > > > [crazy Cc list, not sure it'll go anywhere] It is what get_maintainer.pl script returned to. > > On Tue, 10 Sep 2019 14:47:11 +0100, > Claudiu Beznea wrote: >> >> Change timer registration macros (TIMER_OF_DECLARE() and >> CLOCKSOURCE_OF_DECLARE()) by adding a new argument. This new argument >> is a pointer to an object of type struct timer_of and is used in >> timer_probe(). Based on the flags filled in the struct timer_of object >> the probing process will parse different DT bindings. Later on the >> drivers will use the result of this parsing. Even at the moment only >> few drivers are using this functionality there are other that could >> be converted to use it. >> >> Signed-off-by: Claudiu Beznea > > Why don't you introduce a new registration macro that does what you > want instead of creating this unnecessary churn all over the place? > I though it would be good to keep only one interface for all drivers. I'll keep in mind your proposal for next version. Thank you, Claudiu Beznea > M. > ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 4/7] dt-bindings: chosen: Add clocksource and clockevent selection
On 10.09.2019 17:32, Sudeep Holla wrote: > External E-Mail > > > On Tue, Sep 10, 2019 at 04:47:13PM +0300, Claudiu Beznea wrote: >> From: Alexandre Belloni >> >> Some timer drivers may behave either as clocksource or clockevent >> or both. Until now, in case of platforms with multiple hardware >> resources of the same type, the drivers were chosing the first >> registered hardware resource as clocksource/clockevent and the >> next one as clockevent/clocksource. Other were using different >> compatibles (one for each functionality, although its about the >> same hardware). Add DT bindings to be able to choose the >> functionality of a timer. >> > > Is the piece of hardware not capable of serving as both clocksource > and clockevent or is it just the platform choice ? In my case, the hardware is not capable of serving at the same time a clocksource device and a clockevent device. First, I published v1 for a hardware device having this behavior at [1] requesting 1st probed hardware device to work as clocksource and the 2nd one to work as clockevent. The discussion at [1] ended up with the idea of having a mechanism to specify which hardware device behaves as clocksource and which one behaves as clockevent. Thank you, Claudiu Beznea [1] https://lore.kernel.org/lkml/1552580772-8499-1-git-send-email-claudiu.bez...@microchip.com/ > > -- > Regards, > Sudeep > > ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 2/7] clocksource: change timer registration macros
[crazy Cc list, not sure it'll go anywhere] On Tue, 10 Sep 2019 14:47:11 +0100, Claudiu Beznea wrote: > > Change timer registration macros (TIMER_OF_DECLARE() and > CLOCKSOURCE_OF_DECLARE()) by adding a new argument. This new argument > is a pointer to an object of type struct timer_of and is used in > timer_probe(). Based on the flags filled in the struct timer_of object > the probing process will parse different DT bindings. Later on the > drivers will use the result of this parsing. Even at the moment only > few drivers are using this functionality there are other that could > be converted to use it. > > Signed-off-by: Claudiu Beznea Why don't you introduce a new registration macro that does what you want instead of creating this unnecessary churn all over the place? M. -- Jazz is not dead, it just smells funny. ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Re: [PATCH 4/7] dt-bindings: chosen: Add clocksource and clockevent selection
On Tue, Sep 10, 2019 at 04:47:13PM +0300, Claudiu Beznea wrote: > From: Alexandre Belloni > > Some timer drivers may behave either as clocksource or clockevent > or both. Until now, in case of platforms with multiple hardware > resources of the same type, the drivers were chosing the first > registered hardware resource as clocksource/clockevent and the > next one as clockevent/clocksource. Other were using different > compatibles (one for each functionality, although its about the > same hardware). Add DT bindings to be able to choose the > functionality of a timer. > Is the piece of hardware not capable of serving as both clocksource and clockevent or is it just the platform choice ? -- Regards, Sudeep ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
[PATCH 7/7] clocksource/drivers/integrator-ap: parse the chosen node
From: Alexandre Belloni The driver currently uses aliases to know whether the timer is the clocksource or the clockevent. Add the /chosen/linux,clocksource and /chosen/linux,clockevent parsing while keeping backward compatibility. Signed-off-by: Alexandre Belloni Signed-off-by: Claudiu Beznea --- drivers/clocksource/Kconfig | 1 + drivers/clocksource/timer-integrator-ap.c | 21 - 2 files changed, 21 insertions(+), 1 deletion(-) diff --git a/drivers/clocksource/Kconfig b/drivers/clocksource/Kconfig index a642c23b2fba..e1742c0abb03 100644 --- a/drivers/clocksource/Kconfig +++ b/drivers/clocksource/Kconfig @@ -240,6 +240,7 @@ config KEYSTONE_TIMER config INTEGRATOR_AP_TIMER bool "Integrator-ap timer driver" if COMPILE_TEST select CLKSRC_MMIO + select TIMER_OF help Enables support for the Integrator-ap timer. diff --git a/drivers/clocksource/timer-integrator-ap.c b/drivers/clocksource/timer-integrator-ap.c index 8d6f814ace36..78af89e73125 100644 --- a/drivers/clocksource/timer-integrator-ap.c +++ b/drivers/clocksource/timer-integrator-ap.c @@ -14,6 +14,7 @@ #include #include +#include "timer-of.h" #include "timer-sp.h" static void __iomem * sched_clk_base; @@ -160,6 +161,12 @@ static int integrator_clockevent_init(unsigned long inrate, return 0; } +static struct timer_of to[] = { + { .flags = TIMER_OF_TYPE_CS, }, + { .flags = TIMER_OF_TYPE_CE, }, + { /* sentinel */ } +}; + static int __init integrator_ap_timer_init_of(struct device_node *node) { const char *path; @@ -169,6 +176,7 @@ static int __init integrator_ap_timer_init_of(struct device_node *node) struct clk *clk; unsigned long rate; struct device_node *alias_node; + struct timer_of *to = node->data; base = of_io_request_and_map(node, 0, "integrator-timer"); if (IS_ERR(base)) @@ -183,6 +191,17 @@ static int __init integrator_ap_timer_init_of(struct device_node *node) rate = clk_get_rate(clk); writel(0, base + TIMER_CTRL); + if (timer_of_is_clocksource(to)) + /* The primary timer lacks IRQ, use as clocksource */ + return integrator_clocksource_init(rate, base); + + if (timer_of_is_clockevent(to)) { + /* The secondary timer will drive the clock event */ + irq = irq_of_parse_and_map(node, 0); + return integrator_clockevent_init(rate, base, irq); + } + + /* DT ABI compatibility below */ err = of_property_read_string(of_aliases, "arm,timer-primary", ); if (err) { @@ -227,4 +246,4 @@ static int __init integrator_ap_timer_init_of(struct device_node *node) } TIMER_OF_DECLARE(integrator_ap_timer, "arm,integrator-timer", - integrator_ap_timer_init_of, NULL); + integrator_ap_timer_init_of, to); -- 2.7.4 ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
[PATCH 6/7] drivers/clocksource/timer-of: keep declaration on one line
timer_of_init() declaration could be kept on one line (80 chars per line rule is not broken). Signed-off-by: Claudiu Beznea --- drivers/clocksource/timer-of.h | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/drivers/clocksource/timer-of.h b/drivers/clocksource/timer-of.h index 9fb8c5523150..af72e29e2ea4 100644 --- a/drivers/clocksource/timer-of.h +++ b/drivers/clocksource/timer-of.h @@ -84,8 +84,7 @@ static inline unsigned int timer_of_is_clockevent(struct timer_of *to) (TIMER_OF_TYPE_CE | TIMER_OF_TYPE_CE_AND_CS)); } -extern int __init timer_of_init(struct device_node *np, - struct timer_of *to); +extern int __init timer_of_init(struct device_node *np, struct timer_of *to); extern void __init timer_of_cleanup(struct timer_of *to); -- 2.7.4 ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
[PATCH 4/7] dt-bindings: chosen: Add clocksource and clockevent selection
From: Alexandre Belloni Some timer drivers may behave either as clocksource or clockevent or both. Until now, in case of platforms with multiple hardware resources of the same type, the drivers were chosing the first registered hardware resource as clocksource/clockevent and the next one as clockevent/clocksource. Other were using different compatibles (one for each functionality, although its about the same hardware). Add DT bindings to be able to choose the functionality of a timer. Signed-off-by: Alexandre Belloni Signed-off-by: Claudiu Beznea --- Documentation/devicetree/bindings/chosen.txt | 20 1 file changed, 20 insertions(+) diff --git a/Documentation/devicetree/bindings/chosen.txt b/Documentation/devicetree/bindings/chosen.txt index 45e79172a646..aad3034cdbdf 100644 --- a/Documentation/devicetree/bindings/chosen.txt +++ b/Documentation/devicetree/bindings/chosen.txt @@ -135,3 +135,23 @@ e.g. linux,initrd-end = <0x8280>; }; }; + +linux,clocksource and linux,clockevent +-- + +Those nodes have a timer property. This property is a phandle to the timer to be +chosen as the clocksource or clockevent. This is only useful when the platform +has multiple identical timers and it is not possible to let linux make the +correct choice. + +/ { + chosen { + linux,clocksource { + timer = <>; + }; + + linux,clockevent { + timer = <>; + }; + }; +}; -- 2.7.4 ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
[PATCH 2/7] clocksource: change timer registration macros
Change timer registration macros (TIMER_OF_DECLARE() and CLOCKSOURCE_OF_DECLARE()) by adding a new argument. This new argument is a pointer to an object of type struct timer_of and is used in timer_probe(). Based on the flags filled in the struct timer_of object the probing process will parse different DT bindings. Later on the drivers will use the result of this parsing. Even at the moment only few drivers are using this functionality there are other that could be converted to use it. Signed-off-by: Claudiu Beznea --- arch/arm/kernel/smp_twd.c | 10 ++-- arch/arm/mach-davinci/time.c | 2 +- arch/microblaze/kernel/timer.c| 2 +- arch/mips/ralink/cevt-rt3352.c| 2 +- arch/nios2/kernel/time.c | 2 +- drivers/clocksource/arc_timer.c | 6 +-- drivers/clocksource/arm_arch_timer.c | 6 +-- drivers/clocksource/arm_global_timer.c| 2 +- drivers/clocksource/armv7m_systick.c | 2 +- drivers/clocksource/asm9260_timer.c | 2 +- drivers/clocksource/bcm2835_timer.c | 2 +- drivers/clocksource/bcm_kona_timer.c | 4 +- drivers/clocksource/clksrc-dbx500-prcmu.c | 2 +- drivers/clocksource/clksrc_st_lpc.c | 2 +- drivers/clocksource/clps711x-timer.c | 2 +- drivers/clocksource/dw_apb_timer_of.c | 9 ++-- drivers/clocksource/exynos_mct.c | 4 +- drivers/clocksource/h8300_timer16.c | 2 +- drivers/clocksource/h8300_timer8.c| 2 +- drivers/clocksource/h8300_tpu.c | 2 +- drivers/clocksource/jcore-pit.c | 2 +- drivers/clocksource/mips-gic-timer.c | 2 +- drivers/clocksource/mps2-timer.c | 2 +- drivers/clocksource/mxs_timer.c | 2 +- drivers/clocksource/nomadik-mtu.c | 2 +- drivers/clocksource/renesas-ostm.c| 2 +- drivers/clocksource/samsung_pwm_timer.c | 12 +++-- drivers/clocksource/timer-armada-370-xp.c | 6 +-- drivers/clocksource/timer-atcpit100.c | 6 +-- drivers/clocksource/timer-atlas7.c| 3 +- drivers/clocksource/timer-atmel-pit.c | 2 +- drivers/clocksource/timer-atmel-st.c | 2 +- drivers/clocksource/timer-atmel-tcb.c | 2 +- drivers/clocksource/timer-cadence-ttc.c | 2 +- drivers/clocksource/timer-davinci.c | 3 +- drivers/clocksource/timer-digicolor.c | 2 +- drivers/clocksource/timer-efm32.c | 4 +- drivers/clocksource/timer-fsl-ftm.c | 2 +- drivers/clocksource/timer-fttmr010.c | 10 ++-- drivers/clocksource/timer-gx6605s.c | 7 +-- drivers/clocksource/timer-imx-gpt.c | 24 - drivers/clocksource/timer-imx-sysctr.c| 9 +--- drivers/clocksource/timer-imx-tpm.c | 6 +-- drivers/clocksource/timer-integrator-ap.c | 2 +- drivers/clocksource/timer-ixp4xx.c| 2 +- drivers/clocksource/timer-keystone.c | 2 +- drivers/clocksource/timer-lpc32xx.c | 2 +- drivers/clocksource/timer-mediatek.c | 84 --- drivers/clocksource/timer-meson6.c| 2 +- drivers/clocksource/timer-milbeaut.c | 7 +-- drivers/clocksource/timer-mp-csky.c | 15 ++ drivers/clocksource/timer-npcm7xx.c | 9 +--- drivers/clocksource/timer-nps.c | 6 +-- drivers/clocksource/timer-of.c| 8 ++- drivers/clocksource/timer-of.h| 11 drivers/clocksource/timer-orion.c | 2 +- drivers/clocksource/timer-owl.c | 6 +-- drivers/clocksource/timer-oxnas-rps.c | 4 +- drivers/clocksource/timer-pistachio.c | 2 +- drivers/clocksource/timer-prima2.c| 2 +- drivers/clocksource/timer-probe.c | 17 ++- drivers/clocksource/timer-pxa.c | 2 +- drivers/clocksource/timer-qcom.c | 4 +- drivers/clocksource/timer-rda.c | 8 +-- drivers/clocksource/timer-riscv.c | 2 +- drivers/clocksource/timer-rockchip.c | 4 +- drivers/clocksource/timer-sp804.c | 4 +- drivers/clocksource/timer-sprd.c | 16 +- drivers/clocksource/timer-stm32.c | 44 +++- drivers/clocksource/timer-sun4i.c | 12 ++--- drivers/clocksource/timer-sun5i.c | 4 +- drivers/clocksource/timer-tango-xtal.c| 2 +- drivers/clocksource/timer-tegra.c | 20 +++- drivers/clocksource/timer-ti-32k.c| 2 +- drivers/clocksource/timer-u300.c | 2 +- drivers/clocksource/timer-versatile.c | 4 +- drivers/clocksource/timer-vf-pit.c| 2 +- drivers/clocksource/timer-vt8500.c| 2 +- drivers/clocksource/timer-zevio.c | 2 +- include/linux/clocksource.h | 30 ++- 80 files changed, 266 insertions(+), 264 deletions(-) diff --git a/arch/arm/kernel/smp_twd.c b/arch/arm/kernel/smp_twd.c index 9a14f721a2b0..02ea5ac511ff 100644 --- a/arch/arm/kernel/smp_twd.c +++ b/arch/arm/kernel/smp_twd.c @@ -335,6 +335,10 @@ static int __init
[PATCH 3/7] clocksource/timer_of: use BIT() macro
Use BIT() macro for timer_of flags. Signed-off-by: Claudiu Beznea --- drivers/clocksource/timer-of.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/drivers/clocksource/timer-of.h b/drivers/clocksource/timer-of.h index ee467bb16ca3..df861ea2ec42 100644 --- a/drivers/clocksource/timer-of.h +++ b/drivers/clocksource/timer-of.h @@ -4,9 +4,9 @@ #include -#define TIMER_OF_BASE 0x1 -#define TIMER_OF_CLOCK 0x2 -#define TIMER_OF_IRQ 0x4 +#define TIMER_OF_BASE BIT(0) +#define TIMER_OF_CLOCK BIT(1) +#define TIMER_OF_IRQ BIT(2) struct of_timer_irq { int irq; -- 2.7.4 ___ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
[PATCH 1/7] clocksource/drivers/c-sky: request timer_of_init only for probing CPU
timer_of_init() was initially called for all possible CPUs although it was requested clock with index 0 for the same device_node on behalf of all possible CPUs. This patch keeps the timer_of_init() only for probing CPU and use the information obtained by timer_of_init() to also initialize the timer_of structure for the rest of CPUs. Since the probing CPU was requested also a per CPU interrupt, and the timer_of_init() has such a mechanism implemented, the patch took also the chance to pass TIMER_OF_IRQ flag to timer_of_init(). Apart from this csky_mptimer_irq variable was removed and information in per CPU timer_of objects was used instead (to->clkevt.irq). Signed-off-by: Claudiu Beznea --- drivers/clocksource/timer-mp-csky.c | 45 +++-- 1 file changed, 23 insertions(+), 22 deletions(-) diff --git a/drivers/clocksource/timer-mp-csky.c b/drivers/clocksource/timer-mp-csky.c index 183a9955160a..dd263c8de580 100644 --- a/drivers/clocksource/timer-mp-csky.c +++ b/drivers/clocksource/timer-mp-csky.c @@ -15,7 +15,7 @@ #define PTIM_LVR "cr<6, 14>" #define PTIM_TSR "cr<1, 14>" -static int csky_mptimer_irq; +static irqreturn_t csky_timer_interrupt(int irq, void *dev); static int csky_mptimer_set_next_event(unsigned long delta, struct clock_event_device *ce) @@ -47,7 +47,7 @@ static int csky_mptimer_oneshot_stopped(struct clock_event_device *ce) } static DEFINE_PER_CPU(struct timer_of, csky_to) = { - .flags = TIMER_OF_CLOCK, + .flags = TIMER_OF_CLOCK | TIMER_OF_IRQ, .clkevt = { .rating = 300, .features = CLOCK_EVT_FEAT_PERCPU | @@ -57,6 +57,10 @@ static DEFINE_PER_CPU(struct timer_of, csky_to) = { .set_state_oneshot_stopped = csky_mptimer_oneshot_stopped, .set_next_event = csky_mptimer_set_next_event, }, + .of_irq = { + .percpu = true, + .handler= csky_timer_interrupt, + }, }; static irqreturn_t csky_timer_interrupt(int irq, void *dev) @@ -79,7 +83,7 @@ static int csky_mptimer_starting_cpu(unsigned int cpu) to->clkevt.cpumask = cpumask_of(cpu); - enable_percpu_irq(csky_mptimer_irq, 0); + enable_percpu_irq(to->clkevt.irq, 0); clockevents_config_and_register(>clkevt, timer_of_rate(to), 2, ULONG_MAX); @@ -89,7 +93,9 @@ static int csky_mptimer_starting_cpu(unsigned int cpu) static int csky_mptimer_dying_cpu(unsigned int cpu) { - disable_percpu_irq(csky_mptimer_irq); + struct timer_of *to = per_cpu_ptr(_to, cpu); + + disable_percpu_irq(to->clkevt.irq); return 0; } @@ -117,8 +123,8 @@ struct clocksource csky_clocksource = { static int __init csky_mptimer_init(struct device_node *np) { - int ret, cpu, cpu_rollback; - struct timer_of *to = NULL; + struct timer_of *to = this_cpu_ptr(_to); + int ret, cpu; /* * Csky_mptimer is designed for C-SKY SMP multi-processors and @@ -132,20 +138,20 @@ static int __init csky_mptimer_init(struct device_node *np) * We use private irq for the mptimer and irq number is the same * for every core. So we use request_percpu_irq() in timer_of_init. */ - csky_mptimer_irq = irq_of_parse_and_map(np, 0); - if (csky_mptimer_irq <= 0) - return -EINVAL; - ret = request_percpu_irq(csky_mptimer_irq, csky_timer_interrupt, -"csky_mp_timer", _to); + ret = timer_of_init(np, to); if (ret) return -EINVAL; for_each_possible_cpu(cpu) { - to = per_cpu_ptr(_to, cpu); - ret = timer_of_init(np, to); - if (ret) - goto rollback; + struct timer_of *cpu_to = per_cpu_ptr(_to, cpu); + + if (to == cpu_to) + continue; + + cpu_to->clkevt.irq = to->of_irq.irq; + cpu_to->of_clk.rate = to->of_clk.rate; + cpu_to->of_clk.period = to->of_clk.period; } clocksource_register_hz(_clocksource, timer_of_rate(to)); @@ -156,18 +162,13 @@ static int __init csky_mptimer_init(struct device_node *np) csky_mptimer_starting_cpu, csky_mptimer_dying_cpu); if (ret) - return -EINVAL; + goto rollback; return 0; rollback: - for_each_possible_cpu(cpu_rollback) { - if (cpu_rollback == cpu) - break; + timer_of_cleanup(to); - to = per_cpu_ptr(_to, cpu_rollback); - timer_of_cleanup(to); - }
[PATCH 0/7] add support for clocksource/clockevent DT selection
Hi, This series adds support to permit the selection of clocksource/clockevent via DT. In [1] I proposed a solution other than the one in this series, with parsing DT bindings and at probe time and passing it to timer specific probe function. Looking forward though the clocksource/clockevent drivers implementation and taking into account the response at [2] I sticked the implementation to timer-of specific library. The implementation in this series is using timer-of specific library to parse the DT bindings related to timers functions: clocksource or clockevent. With this implementation a timer's driver registers for probing an array of objects of type struct timer_of. In flags member of struct timer_of object it has to be passed the following new flags (related to how the driver will behave) as follows: - TIMER_OF_TYPE_CS: timer could work only as clocksource at a time - TIMER_OF_TYPE_CE: timer could work only as clockevent at a time - TIMER_OF_TYPE_CE_AND_CE: timer could work at a time as both, clocksource and clockevent. The timer registration macro (for probing) now gets a new argument which should be an array of struct timer_of objects: TIMER_OF_DECLARE(name, compat, handler, to) CLOCKSOURCE_OF_DECLARE(name, compat, handler, to) In case driver could work to feed only the clocksource subsystem or only the clockevent subsystem the struct timer_of array passed to TIMER_OF_DECLARE()/ CLOCKSOURCE_OF_DECLARE() should contain 2 entries: one to be filled if probed timer works as clocksource device, one to be filled if probed timer works as clockevent device. For such a case, the minimal format of struct timer_of array is as follows: struct timer_of to[] = { { .flags = TIMER_OF_TYPE_CS } { .flags = TIMER_OF_TYPE_CE } { /* sentinel. */ }; If timer could work as both, clocksource and clockevent at the same time, the struct timer_of array should look as follows: struct timer_of to[] = { { .flags = TIMER_OF_TYPE_CE_AND_CS } { /* sentinel. */ }; And in device tree there should be added chosen bindings as follows: chosen { linux,clocksource { timer = <> }; linux,clockevent { timer = <>; } }; timer1: t1@123 { compatible = "timerx-compatible"; /* the rest of DT bindings here */ }; timer2: t1@234 { compatible = "timerx-compatible"; /* the rest of DT bindings here */ }; At probing time (timer_probe()), timer_of_init() will check the DT bindings and try to match with one of the entries in struct timer_of array passed at probe. The used entry will be considered used if the timers' device_node pointer is set. If no matching b/w DT and what has been passed for probe via struct timer_of array then probe should fail. The patches in this series are organized as follows: 1/7 - avoid calling timer_of_init() for every CPU since it should not be needed 2/7 - changes timer registration macro by adding a new argument (pointer to struct timer_of) 3/7 - use BIT() macro 4/7 - add clocksource/clockevent selection documentation [3] 5/7 - implement support described above 6/7 - remove an unnecessary line 7/7 - implement this support for integrator-ap timer I implemented this support for timer published at [4]. Thank you, Claudiu Beznea [1] https://lore.kernel.org/lkml/34574b0f-7d09-eb92-ea62-4199c293b...@microchip.com/ [2] https://lore.kernel.org/lkml/1ebaa306-8a7f-fd58-56e0-a61b76735...@linaro.org/ [3] https://lore.kernel.org/lkml/20171213185313.20017-1-alexandre.bell...@free-electrons.com/ [4] https://lore.kernel.org/lkml/1552580772-8499-1-git-send-email-claudiu.bez...@microchip.com/ Alexandre Belloni (2): dt-bindings: chosen: Add clocksource and clockevent selection clocksource/drivers/integrator-ap: parse the chosen node Claudiu Beznea (5): clocksource/drivers/c-sky: request timer_of_init only for probing CPU clocksource: change timer registration macros clocksource/timer_of: use BIT() macro clocksource/drivers/timer-of: add support support for timer's functionalities drivers/clocksource/timer-of: keep declaration on one line Documentation/devicetree/bindings/chosen.txt | 20 + arch/arm/kernel/smp_twd.c| 10 ++- arch/arm/mach-davinci/time.c | 2 +- arch/microblaze/kernel/timer.c | 2 +- arch/mips/ralink/cevt-rt3352.c | 2 +- arch/nios2/kernel/time.c | 2 +- drivers/clocksource/Kconfig | 1 + drivers/clocksource/arc_timer.c | 6 +- drivers/clocksource/arm_arch_timer.c | 6 +- drivers/clocksource/arm_global_timer.c | 2 +- drivers/clocksource/armv7m_systick.c | 2 +- drivers/clocksource/asm9260_timer.c | 2 +- drivers/clocksource/bcm2835_timer.c | 2 +- drivers/clocksource/bcm_kona_timer.c | 4 +- drivers/clocksource/clksrc-dbx500-prcmu.c| 2 +-