Zynq core changes v2
Subsystems (5)
Hi guys,
here is the v2 of patches I have sent yesterday. I have fixed hopefully all changes we have discussed. Please let me know if you find any problem.
Steffen: I have added your SOB line to that smp patch where you had the main contribution. We should probably also add it to that scu one or it is up to you. Just send me proper lines for that specific patches.
Thanks,
Michal
PATCH v2 1/10 arm: zynq: Use standard timer binding
Use cdns,ttc because this driver is Cadence Rev06 Triple Timer Counter and everybody can use it without xilinx specific function name or probing.
Also use standard dt description for timer
and also prepare for moving to clocksource
initialization.
---
v2: Update year in copyright
Fix typo fault
Remove all zynq specific names
Steffen: I have done this change based on our discussion. Moving to generic location will be done in the next patch.
Josh: We have talked about this change at ELC. Using standard dt binding as other timers.
I have also discussed this with Rob some time ago.
https://patchwork.kernel.org/patch/2112791/
--- arch/arm/boot/dts/zynq-7000.dtsi | 45 +------ arch/arm/boot/dts/zynq-zc702.dts | 10 -- arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/timer.c | 261 +++++++++++++++++++++++++++----------- 4 files changed, 195 insertions(+), 122 deletions(-)
arch/arm/boot/dts/zynq-7000.dtsi +6 -39
@@ -111,56 +111,23 @@ }; ttc0: ttc0 at f8001000 { - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; reg = <0xF8001000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc0_0: ttc0.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 10 4>; - }; - ttc0_1: ttc0.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 11 4>; - }; - ttc0_2: ttc0.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 12 4>; - }; }; ttc1: ttc1 at f8002000 { - #interrupt-parent = <&intc>; - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 37 4 0 38 4 0 39 4 >; + compatible = "cdns,ttc"; reg = <0xF8002000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc1_0: ttc1.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 37 4>; - }; - ttc1_1: ttc1.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 38 4>; - }; - ttc1_2: ttc1.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 39 4>; - }; }; }; };
arch/arm/boot/dts/zynq-zc702.dts +0 -10
@@ -32,13 +32,3 @@ &ps_clk { clock-frequency = <33333330>; }; - -&ttc0_0 { - status = "ok"; - compatible = "xlnx,ttc-counter-clocksource"; -}; - -&ttc0_1 { - status = "ok"; - compatible = "xlnx,ttc-counter-clockevent"; -};
arch/arm/mach-zynq/common.c +1 -0
@@ -20,6 +20,7 @@#include <linux/platform_device.h> #include <linux/clk.h> #include <linux/clk/zynq.h> +#include <linux/clocksource.h> #include <linux/of_address.h> #include <linux/of_irq.h> #include <linux/of_platform.h>
arch/arm/mach-zynq/timer.c +188 -73
374 lines not on this page · raw message
Signature · 1 line
1.7.9.7
PATCH v2 1/10 arm: zynq: Use standard timer binding
Quoted 86 lines · Michal Simek
Use cdns,ttc because this driver is Cadence Rev06 Triple Timer Counter and everybody can use it without xilinx specific function name or probing.
Also use standard dt description for timer
and also prepare for moving to clocksource
initialization.Signed-off-by Michal Simek <[email protected]>---
v2: Update year in copyright
Fix typo fault
Remove all zynq specific namesSteffen: I have done this change based on our discussion. Moving to generic location will be done in the next patch.
Josh: We have talked about this change at ELC. Using standard dt binding as other timers.
I have also discussed this with Rob some time ago.
https://patchwork.kernel.org/patch/2112791/--- arch/arm/boot/dts/zynq-7000.dtsi | 45 +------ arch/arm/boot/dts/zynq-zc702.dts | 10 -- arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/timer.c | 261 +++++++++++++++++++++++++++----------- 4 files changed, 195 insertions(+), 122 deletions(-)arch/arm/boot/dts/zynq-7000.dtsi +6
-32@@ -111,56 +111,23 @@ }; ttc0: ttc0 at f8001000 { - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; reg = <0xF8001000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc0_0: ttc0.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 10 4>; - }; - ttc0_1: ttc0.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 11 4>; - }; - ttc0_2: ttc0.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 12 4>; - }; }; ttc1: ttc1 at f8002000 { - #interrupt-parent = <&intc>; - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 37 4 0 38 4 0 39 4 >; + compatible = "cdns,ttc"; reg = <0xF8002000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc1_0: ttc1.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 37 4>; - }; - ttc1_1: ttc1.1 { - status = "disabled"; - reg = <1>;
arch/arm/boot/dts/zynq-7000.dtsi line 133
Quoted 10 lines
arch/arm/boot/dts/zynq-7000.dtsi +0
-7- interrupts = <0 38 4>; - }; - ttc1_2: ttc1.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 39 4>; - }; }; }; };
Hi Michal!
Thought about this a little more. I'm not seeing the improvment this gives us. The ttcs give us 6 counters that could be used however one likes. Linux wants a clocksource and clockevent provider, but I could use the Cortex timer as clocksource and would only have to sacrifice one of the 6 counters. With this patch I have to sacrifice 3 counters and would only use 2 of them. Then there is pinmuxing. That can be handled by one driver, so I think that is doable with both versions and I think I'm okay with that.
So what am I missing? Why would I want it like this and not the way it is right now?
Regards,
Steffen
Quoted 420 lines
arch/arm/boot/dts/zynq-zc702.dts +0
-10@@ -32,13 +32,3 @@ &ps_clk { clock-frequency = <33333330>; }; - -&ttc0_0 { - status = "ok"; - compatible = "xlnx,ttc-counter-clocksource"; -}; - -&ttc0_1 { - status = "ok"; - compatible = "xlnx,ttc-counter-clockevent"; -};arch/arm/mach-zynq/common.c +1
-0@@ -20,6 +20,7 @@#include <linux/platform_device.h> #include <linux/clk.h> #include <linux/clk/zynq.h> +#include <linux/clocksource.h> #include <linux/of_address.h> #include <linux/of_irq.h> #include <linux/of_platform.h> arch/arm/mach-zynq/timer.c +188
-73@@ -1,7 +1,7 @@/* * This file contains driver for the Xilinx PS Timer Counter IP. * -* Copyright (C) 2011 Xilinx +* Copyright (C) 2011-2013 Xilinx * * based on arch/mips/kernel/time.c timer driver * @@ -15,6 +15,7 @@* GNU General Public License for more details. */ +#include <linux/clk.h> #include <linux/interrupt.h> #include <linux/clockchips.h> #include <linux/of_address.h> @@ -24,6 +25,21 @@#include "common.h" /* +* This driver configures the 2 16-bit count-up timers as follows: +* +* T1: Timer 1, clocksource for generic timekeeping +* T2: Timer 2, clockevent source for hrtimers +* T3: Timer 3, <unused> +* +* The input frequency to the timer module for emulation is 2.5MHz which is +* common to all the timer channels (T1, T2, and T3). With a pre-scaler of 32, +* the timers are clocked at 78.125KHz (12.8 us resolution). + +* The input frequency to the timer module in silicon is configurable and +* obtained from device tree. The pre-scaler of 32 is used. +*/ + +/* * Timer Register Offset Definitions of Timer 1, Increment base address by 4 * and use same offsets for Timer 2 */ @@ -44,17 +60,24 @@#define PRESCALE 2048/* The exponent must match this */ #define CLK_CNTRL_PRESCALE ((PRESCALE_EXPONENT - 1) << 1)#define CLK_CNTRL_PRESCALE_EN 1 -#define CNT_CNTRL_RESET (1<<4) +#define CNT_CNTRL_RESET (1 << 4)/** * struct xttcps_timer - This definition defines local timer structure * * @base_addr: Base address of timer -**/ +* @clk: Associated clock source +* @clk_rate_change_nb Notifier block for clock rate changes +*/ struct xttcps_timer { -void __iomem *base_addr; +void __iomem *base_addr; +struct clk *clk; +struct notifier_block clk_rate_change_nb; }; +#define to_xttcps_timer (x) \ + container_of(x, struct xttcps_timer, clk_rate_change_nb) +struct xttcps_timer_clocksource {struct xttcps_timer xttc;struct clocksource cs;@@ -66,7 +89,6 @@ struct xttcps_timer_clocksource {struct xttcps_timer_clockevent {struct xttcps_timer xttc;struct clock_event_device ce; -struct clk *clk; };#define to_xttcps_timer_clkevent (x) \@@ -167,8 +189,8 @@ static void xttcps_set_mode(enum clock_event_mode mode,switch (mode) {case CLOCK_EVT_MODE_PERIODIC :xttcps_set_interval (timer, -DIV_ROUND_CLOSEST (clk_get_rate (xttce->clk), -PRESCALE *HZ )); +DIV_ROUND_CLOSEST (clk_get_rate (xttce->xttc.clk), +PRESCALE *HZ ));break ;case CLOCK_EVT_MODE_ONESHOT :case CLOCK_EVT_MODE_UNUSED :@@ -189,79 +211,148 @@ static void xttcps_set_mode(enum clock_event_mode mode, } } -static void __init zynq_ttc_setup_clocksource (struct device_node *np, -void __iomem *base) +static int xttcps_rate_change_clocksource_cb (struct notifier_block *nb, +unsigned long event,void *data) +{ +struct clk_notifier_data *ndata = data; +struct xttcps_timer *xttcps =to_xttcps_timer (nb); +struct xttcps_timer_clocksource *xttccs =container_of (xttcps, + struct xttcps_timer_clocksource, xttc); + +switch (event) { +case POST_RATE_CHANGE : +/* +* Do whatever is necessary to maintain a proper time base +* +* I cannot find a way to adjust the currently used clocksource +* to the new frequency. __clocksource_updatefreq_hz() sounds +* good, but does not work. Not sure what's that missing. +* +* This approach works, but triggers two clocksource switches. +* The first after unregister to clocksource jiffies. And +* another one after the register to the newly registered timer. +* +* Alternatively we could 'waste' another HW timer to ping pong +* between clock sources. That would also use one register and +* one unregister call, but only trigger one clocksource switch +* for the cost of another HW timer used by the OS. +*/ +clocksource_unregister (&xttccs->cs); +clocksource_register_hz (&xttccs->cs, + ndata->new_rate /PRESCALE ); +/* fall through */ +case PRE_RATE_CHANGE : +case ABORT_RATE_CHANGE : +default : +return NOTIFY_DONE ; + } +} + +static void __init xttc_setup_clocksource (struct clk *clk,void __iomem *base) {struct xttcps_timer_clocksource *ttccs; -struct clk *clk;int err; -u32 reg; ttccs =kzalloc (sizeof (*ttccs),GFP_KERNEL );if (WARN_ON (!ttccs))return ; - err =of_property_read_u32 (np,"reg" , ®); -if (WARN_ON (err)) -return ; + ttccs->xttc.clk = clk; - clk =of_clk_get_by_name (np,"cpu_1x" ); -if (WARN_ON (IS_ERR (clk))) -return ; - - err =clk_prepare_enable (clk); + err =clk_prepare_enable (ttccs->xttc.clk);if (WARN_ON (err))return ; - ttccs->xttc.base_addr = base + reg *4 ; + ttccs->xttc.clk_rate_change_nb.notifier_call = + xttcps_rate_change_clocksource_cb; + ttccs->xttc.clk_rate_change_nb.next =NULL ; +if (clk_notifier_register (ttccs->xttc.clk, + &ttccs->xttc.clk_rate_change_nb)) +pr_warn ("Unable to register clock notifier.\n" ); - ttccs->cs.name = np->name; + ttccs->xttc.base_addr = base; + ttccs->cs.name ="xttcps_clocksource" ; ttccs->cs.rating =200 ; ttccs->cs.read = __xttc_clocksource_read; ttccs->cs.mask =CLOCKSOURCE_MASK (16 ); ttccs->cs.flags =CLOCK_SOURCE_IS_CONTINUOUS ; +/* +* Setup the clock source counter to be an incrementing counter +* with no interrupt and it rolls over at 0xFFFF. Pre-scale +* it by 32 also. Let it start running now. +*/ __raw_writel (0x0 , ttccs->xttc.base_addr +XTTCPS_IER_OFFSET );__raw_writel (CLK_CNTRL_PRESCALE |CLK_CNTRL_PRESCALE_EN , ttccs->xttc.base_addr +XTTCPS_CLK_CNTRL_OFFSET );__raw_writel (CNT_CNTRL_RESET , ttccs->xttc.base_addr +XTTCPS_CNT_CNTRL_OFFSET ); - err =clocksource_register_hz (&ttccs->cs,clk_get_rate (clk) /PRESCALE ); + err =clocksource_register_hz (&ttccs->cs, +clk_get_rate (ttccs->xttc.clk) /PRESCALE );if (WARN_ON (err))return ; + } -static void __init zynq_ttc_setup_clockevent (struct device_node *np, -void __iomem *base) +static int xttcps_rate_change_clockevent_cb (struct notifier_block *nb, +unsigned long event,void *data) +{ +struct clk_notifier_data *ndata = data; +struct xttcps_timer *xttcps =to_xttcps_timer (nb); +struct xttcps_timer_clockevent *xttcce =container_of (xttcps, + struct xttcps_timer_clockevent, xttc); + +switch (event) { +case POST_RATE_CHANGE : + { +unsigned long flags; + +/* +* clockevents_update_freq should be called with IRQ disabled on +* the CPU the timer provides events for. The timer we use is +* common to both CPUs, not sure if we need to run on both +* cores. +*/ +local_irq_save (flags); +clockevents_update_freq (&xttcce->ce, + ndata->new_rate /PRESCALE ); +local_irq_restore (flags); + +/* fall through */ + } +case PRE_RATE_CHANGE : +case ABORT_RATE_CHANGE : +default : +return NOTIFY_DONE ; + } +} + +static void __init xttc_setup_clockevent (struct clk *clk, +void __iomem *base,u32 irq) {struct xttcps_timer_clockevent *ttcce; -int err, irq; -u32 reg; +int err; ttcce =kzalloc (sizeof (*ttcce),GFP_KERNEL );if (WARN_ON (!ttcce))return ; - err =of_property_read_u32 (np,"reg" , ®); -if (WARN_ON (err)) -return ; + ttcce->xttc.clk = clk; - ttcce->xttc.base_addr = base + reg *4 ; - - ttcce->clk =of_clk_get_by_name (np,"cpu_1x" ); -if (WARN_ON (IS_ERR (ttcce->clk))) -return ; - - err =clk_prepare_enable (ttcce->clk); + err =clk_prepare_enable (ttcce->xttc.clk);if (WARN_ON (err))return ; - irq =irq_of_parse_and_map (np,0 ); -if (WARN_ON (!irq)) -return ; + ttcce->xttc.clk_rate_change_nb.notifier_call = + xttcps_rate_change_clockevent_cb; + ttcce->xttc.clk_rate_change_nb.next =NULL ; +if (clk_notifier_register (ttcce->xttc.clk, + &ttcce->xttc.clk_rate_change_nb)) +pr_warn ("Unable to register clock notifier.\n" ); - ttcce->ce.name = np->name; + ttcce->xttc.base_addr = base; + ttcce->ce.name ="xttcps_clockevent" ; ttcce->ce.features =CLOCK_EVT_FEAT_PERIODIC |CLOCK_EVT_FEAT_ONESHOT ; ttcce->ce.set_next_event = xttcps_set_next_event; ttcce->ce.set_mode = xttcps_set_mode;@@ -269,56 +360,80 @@ static void __init zynq_ttc_setup_clockevent(struct device_node *np, ttcce->ce.irq = irq; ttcce->ce.cpumask = cpu_possible_mask; +/* +* Setup the clock event timer to be an interval timer which +* is prescaled by 32 using the interval interrupt. Leave it +* disabled for now. +*/ __raw_writel (0x23 , ttcce->xttc.base_addr +XTTCPS_CNT_CNTRL_OFFSET );__raw_writel (CLK_CNTRL_PRESCALE |CLK_CNTRL_PRESCALE_EN , ttcce->xttc.base_addr +XTTCPS_CLK_CNTRL_OFFSET );__raw_writel (0x1 , ttcce->xttc.base_addr +XTTCPS_IER_OFFSET ); - err =request_irq (irq, xttcps_clock_event_interrupt,IRQF_TIMER , - np->name, ttcce); + err =request_irq (irq, xttcps_clock_event_interrupt, +IRQF_DISABLED |IRQF_TIMER , + ttcce->ce.name, ttcce);if (WARN_ON (err))return ;clockevents_config_and_register (&ttcce->ce, -clk_get_rate (ttcce->clk) /PRESCALE , -1 ,0xfffe ); +clk_get_rate (ttcce->xttc.clk) /PRESCALE ,1 ,0xfffe ); } -static const __initconst struct of_device_id zynq_ttc_match[] = { - { .compatible ="xlnx,ttc-counter-clocksource" , - .data = zynq_ttc_setup_clocksource, }, - { .compatible ="xlnx,ttc-counter-clockevent" , - .data = zynq_ttc_setup_clockevent, }, - {} -}; -/** * xttcps_timer_init - Initialize the timer * * Initializes the timer hardware and register the clock source and clock event * timers with Linux kernal timer framework -**/ +*/ +static void __init xttcps_timer_init_of (struct device_node *timer) +{ +unsigned int irq; +void __iomem *timer_baseaddr; +struct clk *clk; + +/* +* Get the 1st Triple Timer Counter (TTC) block from the device tree +* and use it. Note that the event timer uses the interrupt and it's the +* 2nd TTC hence the irq_of_parse_and_map(,1) +*/ + timer_baseaddr =of_iomap (timer,0 ); +if (!timer_baseaddr) { +pr_err ("ERROR: invalid timer base address\n" ); +BUG (); + } + + irq =irq_of_parse_and_map (timer,1 ); +if (irq <=0 ) { +pr_err ("ERROR: invalid interrupt number\n" ); +BUG (); + } + + clk =of_clk_get_by_name (timer,"cpu_1x" ); +if (IS_ERR (clk)) { +pr_err ("ERROR: timer input clock not found\n" ); +BUG (); + } + +xttc_setup_clocksource (clk, timer_baseaddr); +xttc_setup_clockevent (clk, timer_baseaddr +4 , irq); + +pr_info ("%s #0 at %p, irq=%d\n" , timer->name, timer_baseaddr, irq); +} +void __init xttcps_timer_init (void ) { -struct device_node *np; - -for_each_compatible_node (np,NULL ,"xlnx,ttc" ) { -struct device_node *np_chld; -void __iomem *base; - - base =of_iomap (np,0 ); -if (WARN_ON (!base)) -return ; - -for_each_available_child_of_node (np, np_chld) { -int (*cb)(struct device_node *np,void __iomem *base); -const struct of_device_id *match; - - match =of_match_node (zynq_ttc_match, np_chld); -if (match) { - cb = match->data; -cb (np_chld, base); - } - } +const char *const timer_list[] = { +"cdns,ttc" , +NULL + }; +struct device_node *timer; + + timer =of_find_compatible_node (NULL ,NULL , timer_list[0 ]); +if (!timer) { +pr_err ("ERROR: no compatible timer found\n" ); +BUG (); } + +xttcps_timer_init_of (timer); }Signature · 1 line
1.7.9.7
Signature · 4 lines
Pengutronix e.K. | | Industrial Linux Solutions | http://www.pengutronix.de/ | Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
PATCH v2 1/10 arm: zynq: Use standard timer binding
2013/3/26 Steffen Trumtrar <[email protected]>:
Quoted 100 lines · Michal Simek
Quoted 86 lines
Use cdns,ttc because this driver is Cadence Rev06 Triple Timer Counter and everybody can use it without xilinx specific function name or probing.
Also use standard dt description for timer
and also prepare for moving to clocksource
initialization.Signed-off-by Michal Simek <[email protected]>---
v2: Update year in copyright
Fix typo fault
Remove all zynq specific namesSteffen: I have done this change based on our discussion. Moving to generic location will be done in the next patch.
Josh: We have talked about this change at ELC. Using standard dt binding as other timers.
I have also discussed this with Rob some time ago.
https://patchwork.kernel.org/patch/2112791/--- arch/arm/boot/dts/zynq-7000.dtsi | 45 +------ arch/arm/boot/dts/zynq-zc702.dts | 10 -- arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/timer.c | 261 +++++++++++++++++++++++++++----------- 4 files changed, 195 insertions(+), 122 deletions(-)arch/arm/boot/dts/zynq-7000.dtsi +6
-32@@ -111,56 +111,23 @@ }; ttc0: ttc0 at f8001000 { - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; reg = <0xF8001000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc0_0: ttc0.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 10 4>; - }; - ttc0_1: ttc0.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 11 4>; - }; - ttc0_2: ttc0.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 12 4>; - }; }; ttc1: ttc1 at f8002000 { - #interrupt-parent = <&intc>; - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 37 4 0 38 4 0 39 4 >; + compatible = "cdns,ttc"; reg = <0xF8002000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc1_0: ttc1.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 37 4>; - }; - ttc1_1: ttc1.1 { - status = "disabled"; - reg = <1>;arch/arm/boot/dts/zynq-7000.dtsi line 133
Quoted 10 lines
arch/arm/boot/dts/zynq-7000.dtsi +0
-7- interrupts = <0 38 4>; - }; - ttc1_2: ttc1.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 39 4>; - }; }; }; };Hi Michal!
Quoted 9 lines
Thought about this a little more. I'm not seeing the improvment this gives us. The ttcs give us 6 counters that could be used however one likes. Linux wants a clocksource and clockevent provider, but I could use the Cortex timer as clocksource and would only have to sacrifice one of the 6 counters. With this patch I have to sacrifice 3 counters and would only use 2 of them. Then there is pinmuxing. That can be handled by one driver, so I think that is doable with both versions and I think I'm okay with that.
So what am I missing? Why would I want it like this and not the way it is right now?
There is a big move because previous DT description was incorrect. Device-tree should describe hardware and there is no something like ttc-counter-clocksource or ttc-counter-clockevent compatible driver. Every ttc counter can be used as clocksource or clockevent. Changing/Setting compatible properties for linux purpose is not correct.
Just read this thread about.
https://patchwork.kernel.org/patch/2112791/
Utilization of the core or using different timers in the system for clock is different discussion. We can just try to utilize timers in the arm core or add bunch of them to PL to see how kernel behaves.
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 1/10 arm: zynq: Use standard timer binding
Quoted 111 lines · Michal Simek
2013/3/26 Steffen Trumtrar <[email protected]>:
Quoted 100 lines · Michal Simek
Quoted 86 lines
Use cdns,ttc because this driver is Cadence Rev06 Triple Timer Counter and everybody can use it without xilinx specific function name or probing.
Also use standard dt description for timer
and also prepare for moving to clocksource
initialization.Signed-off-by Michal Simek <[email protected]>---
v2: Update year in copyright
Fix typo fault
Remove all zynq specific namesSteffen: I have done this change based on our discussion. Moving to generic location will be done in the next patch.
Josh: We have talked about this change at ELC. Using standard dt binding as other timers.
I have also discussed this with Rob some time ago.
https://patchwork.kernel.org/patch/2112791/--- arch/arm/boot/dts/zynq-7000.dtsi | 45 +------ arch/arm/boot/dts/zynq-zc702.dts | 10 -- arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/timer.c | 261 +++++++++++++++++++++++++++----------- 4 files changed, 195 insertions(+), 122 deletions(-)arch/arm/boot/dts/zynq-7000.dtsi +6
-32@@ -111,56 +111,23 @@ }; ttc0: ttc0 at f8001000 { - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; reg = <0xF8001000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc0_0: ttc0.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 10 4>; - }; - ttc0_1: ttc0.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 11 4>; - }; - ttc0_2: ttc0.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 12 4>; - }; }; ttc1: ttc1 at f8002000 { - #interrupt-parent = <&intc>; - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 37 4 0 38 4 0 39 4 >; + compatible = "cdns,ttc"; reg = <0xF8002000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc1_0: ttc1.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 37 4>; - }; - ttc1_1: ttc1.1 { - status = "disabled"; - reg = <1>;arch/arm/boot/dts/zynq-7000.dtsi line 133
Quoted 10 lines
arch/arm/boot/dts/zynq-7000.dtsi +0
-7- interrupts = <0 38 4>; - }; - ttc1_2: ttc1.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 39 4>; - }; }; }; };Hi Michal!
Quoted 9 lines
Thought about this a little more. I'm not seeing the improvment this gives us. The ttcs give us 6 counters that could be used however one likes. Linux wants a clocksource and clockevent provider, but I could use the Cortex timer as clocksource and would only have to sacrifice one of the 6 counters. With this patch I have to sacrifice 3 counters and would only use 2 of them. Then there is pinmuxing. That can be handled by one driver, so I think that is doable with both versions and I think I'm okay with that.
So what am I missing? Why would I want it like this and not the way it is right now?
Quoted 6 lines
There is a big move because previous DT description was incorrect. Device-tree should describe hardware and there is no something like ttc-counter-clocksource or ttc-counter-clockevent compatible driver. Every ttc counter can be used as clocksource or clockevent. Changing/Setting compatible properties for linux purpose is not correct.
That is correct. DT describes the HW and the TTC includes three counters that could be used in any way a HW vendor likes. How would you decide that with just the driver?
The compatible property is uncommon and might even be wrong. How about adding for example a mode-property like usb does for otg,peripheral,host? Suppose I have some obscure board with a Zynq on it, that has ttc1_0 setup to count some external signal and ttc1_1 counts some signal from the PL. I mean it is possible to configure the SoC in that way, no?!
I would than have a board-specific dts where I would somehow have to describe this setup and mark the different ttcs accordingly.
What I'm going for is, what is wrong in having the subnodes in ttc and then iterate over these subnodes in the driver and registering the first two ttcs that I find and that support being used as clocksource/event?
Quoted 3 lines
Just read this thread about.
https://patchwork.kernel.org/patch/2112791/
Already did.
Quoted 4 lines
Utilization of the core or using different timers in the system for clock is different discussion. We can just try to utilize timers in the arm core or add bunch of them to PL to see how kernel behaves.
No problem there. I use the smp_twd on the ZedBoard and the kernel just does the right thing.
Regards,
Steffen
Signature · 4 lines
Pengutronix e.K. | | Industrial Linux Solutions | http://www.pengutronix.de/ | Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
PATCH v2 1/10 arm: zynq: Use standard timer binding
2013/3/27 Steffen Trumtrar <[email protected]>:
Quoted 119 lines · Michal Simek
Quoted 111 lines
2013/3/26 Steffen Trumtrar <[email protected]>:
Quoted 100 lines · Michal Simek
Quoted 86 lines
Use cdns,ttc because this driver is Cadence Rev06 Triple Timer Counter and everybody can use it without xilinx specific function name or probing.
Also use standard dt description for timer
and also prepare for moving to clocksource
initialization.Signed-off-by Michal Simek <[email protected]>---
v2: Update year in copyright
Fix typo fault
Remove all zynq specific namesSteffen: I have done this change based on our discussion. Moving to generic location will be done in the next patch.
Josh: We have talked about this change at ELC. Using standard dt binding as other timers.
I have also discussed this with Rob some time ago.
https://patchwork.kernel.org/patch/2112791/--- arch/arm/boot/dts/zynq-7000.dtsi | 45 +------ arch/arm/boot/dts/zynq-zc702.dts | 10 -- arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/timer.c | 261 +++++++++++++++++++++++++++----------- 4 files changed, 195 insertions(+), 122 deletions(-)arch/arm/boot/dts/zynq-7000.dtsi +6
-32@@ -111,56 +111,23 @@ }; ttc0: ttc0 at f8001000 { - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; reg = <0xF8001000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc0_0: ttc0.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 10 4>; - }; - ttc0_1: ttc0.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 11 4>; - }; - ttc0_2: ttc0.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 12 4>; - }; }; ttc1: ttc1 at f8002000 { - #interrupt-parent = <&intc>; - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 37 4 0 38 4 0 39 4 >; + compatible = "cdns,ttc"; reg = <0xF8002000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc1_0: ttc1.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 37 4>; - }; - ttc1_1: ttc1.1 { - status = "disabled"; - reg = <1>;arch/arm/boot/dts/zynq-7000.dtsi line 133
Quoted 10 lines
arch/arm/boot/dts/zynq-7000.dtsi +0
-7- interrupts = <0 38 4>; - }; - ttc1_2: ttc1.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 39 4>; - }; }; }; };Hi Michal!
Quoted 9 lines
Thought about this a little more. I'm not seeing the improvment this gives us. The ttcs give us 6 counters that could be used however one likes. Linux wants a clocksource and clockevent provider, but I could use the Cortex timer as clocksource and would only have to sacrifice one of the 6 counters. With this patch I have to sacrifice 3 counters and would only use 2 of them. Then there is pinmuxing. That can be handled by one driver, so I think that is doable with both versions and I think I'm okay with that.
So what am I missing? Why would I want it like this and not the way it is right now?Quoted 6 lines
There is a big move because previous DT description was incorrect. Device-tree should describe hardware and there is no something like ttc-counter-clocksource or ttc-counter-clockevent compatible driver. Every ttc counter can be used as clocksource or clockevent. Changing/Setting compatible properties for linux purpose is not correct.
Quoted 5 lines
That is correct. DT describes the HW and the TTC includes three counters that could be used in any way a HW vendor likes. How would you decide that with just the driver?
The compatible property is uncommon and might even be wrong. How about adding for example a mode-property like usb does for otg,peripheral,host?
I don't think that this is right thing to do.
Rob: What do you think?
Quoted 3 lines
Suppose I have some obscure board with a Zynq on it, that has
ttc1_0setup to count some external signal andttc1_1counts some signal from the PL. I mean it is possible to configure the SoC in that way, no?!
If you are on ttc1 then it should be fine because ttc0 should be probed first and ttc1 is not allocated but look below.
Quoted 2 lines
I would than have a board-specific dts where I would somehow have to describe this setup and mark the different ttcs accordingly.
I understand your setup and I am not aware if there is any standard way how to handles
other timers in the system to be able to work with them via standard api. What do you use? UIO for ttc setup and reading it back?
Quoted 3 lines
What I'm going for is, what is wrong in having the subnodes in ttc and then iterate over these subnodes in the driver and registering the first two ttcs that I find and that support being used as clocksource/event?
I understand your concern but the question is how to describe this in dts in a proper way.
Compatible property - no
Based on aliases order - possible but based on my discussion with Rob probably no
Additional mode options - it is also suggesting Linux specific usage.
What next?
Quoted 7 lines
Quoted 4 lines
Just read this thread about.
https://patchwork.kernel.org/patch/2112791/Already did.
Quoted 7 lines
Quoted 4 lines
Utilization of the core or using different timers in the system for clock is different discussion. We can just try to utilize timers in the arm core or add bunch of them to PL to see how kernel behaves.
No problem there. I use the
smp_twdon the ZedBoard and the kernel just does the right thing.
ok.
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 1/10 arm: zynq: Use standard timer binding
Quoted 136 lines · Michal Simek
2013/3/27 Steffen Trumtrar <[email protected]>:
Quoted 119 lines · Michal Simek
Quoted 111 lines
2013/3/26 Steffen Trumtrar <[email protected]>:
Quoted 100 lines · Michal Simek
Quoted 86 lines
Use cdns,ttc because this driver is Cadence Rev06 Triple Timer Counter and everybody can use it without xilinx specific function name or probing.
Also use standard dt description for timer
and also prepare for moving to clocksource
initialization.Signed-off-by Michal Simek <[email protected]>---
v2: Update year in copyright
Fix typo fault
Remove all zynq specific namesSteffen: I have done this change based on our discussion. Moving to generic location will be done in the next patch.
Josh: We have talked about this change at ELC. Using standard dt binding as other timers.
I have also discussed this with Rob some time ago.
https://patchwork.kernel.org/patch/2112791/--- arch/arm/boot/dts/zynq-7000.dtsi | 45 +------ arch/arm/boot/dts/zynq-zc702.dts | 10 -- arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/timer.c | 261 +++++++++++++++++++++++++++----------- 4 files changed, 195 insertions(+), 122 deletions(-)arch/arm/boot/dts/zynq-7000.dtsi +6
-32@@ -111,56 +111,23 @@ }; ttc0: ttc0 at f8001000 { - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; reg = <0xF8001000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc0_0: ttc0.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 10 4>; - }; - ttc0_1: ttc0.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 11 4>; - }; - ttc0_2: ttc0.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 12 4>; - }; }; ttc1: ttc1 at f8002000 { - #interrupt-parent = <&intc>; - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 37 4 0 38 4 0 39 4 >; + compatible = "cdns,ttc"; reg = <0xF8002000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc1_0: ttc1.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 37 4>; - }; - ttc1_1: ttc1.1 { - status = "disabled"; - reg = <1>;arch/arm/boot/dts/zynq-7000.dtsi line 133
Quoted 10 lines
arch/arm/boot/dts/zynq-7000.dtsi +0
-7- interrupts = <0 38 4>; - }; - ttc1_2: ttc1.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 39 4>; - }; }; }; };Hi Michal!
Quoted 9 lines
Thought about this a little more. I'm not seeing the improvment this gives us. The ttcs give us 6 counters that could be used however one likes. Linux wants a clocksource and clockevent provider, but I could use the Cortex timer as clocksource and would only have to sacrifice one of the 6 counters. With this patch I have to sacrifice 3 counters and would only use 2 of them. Then there is pinmuxing. That can be handled by one driver, so I think that is doable with both versions and I think I'm okay with that.
So what am I missing? Why would I want it like this and not the way it is right now?Quoted 6 lines
There is a big move because previous DT description was incorrect. Device-tree should describe hardware and there is no something like ttc-counter-clocksource or ttc-counter-clockevent compatible driver. Every ttc counter can be used as clocksource or clockevent. Changing/Setting compatible properties for linux purpose is not correct.
Quoted 5 lines
That is correct. DT describes the HW and the TTC includes three counters that could be used in any way a HW vendor likes. How would you decide that with just the driver?
The compatible property is uncommon and might even be wrong. How about adding for example a mode-property like usb does for otg,peripheral,host?I don't think that this is right thing to do.
Rob: What do you think?Quoted 3 lines
Suppose I have some obscure board with a Zynq on it, that has
ttc1_0setup to count some external signal andttc1_1counts some signal from the PL. I mean it is possible to configure the SoC in that way, no?!If you are on ttc1 then it should be fine because ttc0 should be probed first and ttc1 is not allocated but look below.
Quoted 8 lines
Quoted 2 lines
I would than have a board-specific dts where I would somehow have to describe this setup and mark the different ttcs accordingly.
I understand your setup and I am not aware if there is any standard way how to handles
other timers in the system to be able to work with them via standard api. What do you use? UIO for ttc setup and reading it back?
Oh, I don't have this setup. I'm just imagining things and how we could then handle this setup, if it ever occurs. When this binding is fixed it is fixed. At the moment it might be okay to rework this, as current mainline only works under certain circumstances and there is probably no one really using it as of yet.
Quoted 7 lines
Quoted 3 lines
What I'm going for is, what is wrong in having the subnodes in ttc and then iterate over these subnodes in the driver and registering the first two ttcs that I find and that support being used as clocksource/event?
I understand your concern but the question is how to describe this in dts in a proper way.
Compatible property - no
Agreed.
Quoted 2 lines
Based on aliases order - possible but based on my discussion with Rob probably no
Agreed.
Quoted 2 lines
Additional mode options - it is also suggesting Linux specific usage.
Hm, it would specify the capabilities of the ttc. Other users of the DT might be interested in this info, too. I'm not saying it is the best/only solution.
Quoted 2 lines
What next?
Regards,
Steffen
Signature · 4 lines
Pengutronix e.K. | | Industrial Linux Solutions | http://www.pengutronix.de/ | Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
PATCH v2 1/10 arm: zynq: Use standard timer binding
2013/3/27 Steffen Trumtrar <[email protected]>:
Quoted 164 lines · Michal Simek
Quoted 136 lines
2013/3/27 Steffen Trumtrar <[email protected]>:
Quoted 119 lines · Michal Simek
Quoted 111 lines
2013/3/26 Steffen Trumtrar <[email protected]>:
Quoted 100 lines · Michal Simek
Quoted 86 lines
Use cdns,ttc because this driver is Cadence Rev06 Triple Timer Counter and everybody can use it without xilinx specific function name or probing.
Also use standard dt description for timer
and also prepare for moving to clocksource
initialization.Signed-off-by Michal Simek <[email protected]>---
v2: Update year in copyright
Fix typo fault
Remove all zynq specific namesSteffen: I have done this change based on our discussion. Moving to generic location will be done in the next patch.
Josh: We have talked about this change at ELC. Using standard dt binding as other timers.
I have also discussed this with Rob some time ago.
https://patchwork.kernel.org/patch/2112791/--- arch/arm/boot/dts/zynq-7000.dtsi | 45 +------ arch/arm/boot/dts/zynq-zc702.dts | 10 -- arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/timer.c | 261 +++++++++++++++++++++++++++----------- 4 files changed, 195 insertions(+), 122 deletions(-)arch/arm/boot/dts/zynq-7000.dtsi +6
-32@@ -111,56 +111,23 @@ }; ttc0: ttc0 at f8001000 { - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; reg = <0xF8001000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc0_0: ttc0.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 10 4>; - }; - ttc0_1: ttc0.1 { - status = "disabled"; - reg = <1>; - interrupts = <0 11 4>; - }; - ttc0_2: ttc0.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 12 4>; - }; }; ttc1: ttc1 at f8002000 { - #interrupt-parent = <&intc>; - #address-cells = <1>; - #size-cells = <0>; - compatible = "xlnx,ttc"; + interrupt-parent = <&intc>; + interrupts = < 0 37 4 0 38 4 0 39 4 >; + compatible = "cdns,ttc"; reg = <0xF8002000 0x1000>; clocks = <&cpu_clk 3>; clock-names = "cpu_1x"; clock-ranges; - - ttc1_0: ttc1.0 { - status = "disabled"; - reg = <0>; - interrupts = <0 37 4>; - }; - ttc1_1: ttc1.1 { - status = "disabled"; - reg = <1>;arch/arm/boot/dts/zynq-7000.dtsi line 133
Quoted 10 lines
arch/arm/boot/dts/zynq-7000.dtsi +0
-7- interrupts = <0 38 4>; - }; - ttc1_2: ttc1.2 { - status = "disabled"; - reg = <2>; - interrupts = <0 39 4>; - }; }; }; };Hi Michal!
Quoted 9 lines
Thought about this a little more. I'm not seeing the improvment this gives us. The ttcs give us 6 counters that could be used however one likes. Linux wants a clocksource and clockevent provider, but I could use the Cortex timer as clocksource and would only have to sacrifice one of the 6 counters. With this patch I have to sacrifice 3 counters and would only use 2 of them. Then there is pinmuxing. That can be handled by one driver, so I think that is doable with both versions and I think I'm okay with that.
So what am I missing? Why would I want it like this and not the way it is right now?Quoted 6 lines
There is a big move because previous DT description was incorrect. Device-tree should describe hardware and there is no something like ttc-counter-clocksource or ttc-counter-clockevent compatible driver. Every ttc counter can be used as clocksource or clockevent. Changing/Setting compatible properties for linux purpose is not correct.
Quoted 5 lines
That is correct. DT describes the HW and the TTC includes three counters that could be used in any way a HW vendor likes. How would you decide that with just the driver?
The compatible property is uncommon and might even be wrong. How about adding for example a mode-property like usb does for otg,peripheral,host?I don't think that this is right thing to do.
Rob: What do you think?Quoted 3 lines
Suppose I have some obscure board with a Zynq on it, that has
ttc1_0setup to count some external signal andttc1_1counts some signal from the PL. I mean it is possible to configure the SoC in that way, no?!If you are on ttc1 then it should be fine because ttc0 should be probed first and ttc1 is not allocated but look below.
Quoted 8 lines
Quoted 2 lines
I would than have a board-specific dts where I would somehow have to describe this setup and mark the different ttcs accordingly.
I understand your setup and I am not aware if there is any standard way how to handles
other timers in the system to be able to work with them via standard api. What do you use? UIO for ttc setup and reading it back?Oh, I don't have this setup. I'm just imagining things and how we could then handle this setup, if it ever occurs. When this binding is fixed it is fixed. At the moment it might be okay to rework this, as current mainline only works under certain circumstances and there is probably no one really using it as of yet.
Quoted 7 lines
Quoted 3 lines
What I'm going for is, what is wrong in having the subnodes in ttc and then iterate over these subnodes in the driver and registering the first two ttcs that I find and that support being used as clocksource/event?
I understand your concern but the question is how to describe this in dts in a proper way.
Compatible property - noAgreed.
Quoted 2 lines
Based on aliases order - possible but based on my discussion with Rob probably no
Quoted 8 lines
Agreed.
Quoted 2 lines
Additional mode options - it is also suggesting Linux specific usage.
Hm, it would specify the capabilities of the ttc. Other users of the DT might be interested in this info, too. I'm not saying it is the best/only solution.
But the capabilities for all 2 ttc (6 timers) will be the same because they are the same
and all of them can do the same things.
I am definitely willing to find out any generic solution and the reason for this patch is that the current dt description is not correct and based on my discussion with Rob this was the solution which others are also using.
In the valid case you have described above there are some things which can be easy to handle on zynq - if you know that the first two timers are used by linux because of the code then changing input to 3rd timer your hw design should be possible (not sure if ttc can handle input but in generic case definitely yes).
But I can imagine case where the system has some limitations and for this purpose you have to use the first counter and currently you have no option how to describe this in the device-tree.
It means there should be any user input to the system to say - not to use this timer
for Linux purpose. And DTS is only one user input to the kernel that's why it should be written there.
Describe this in chosen node?
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 2/10 arm: zynq: Move timer to clocksource interface
In reply to Michal SimekUse clocksource timer initialization.
---
v2: Free allocated space for structures
Till this time this was only one timer driver which could be possible to use on zynq.
--- arch/arm/mach-zynq/common.c | 2 +- arch/arm/mach-zynq/common.h | 2 -- arch/arm/mach-zynq/timer.c | 43 ++++++++++++++++++++----------------------- 3 files changed, 21 insertions(+), 26 deletions(-)
arch/arm/mach-zynq/common.c +1 -1
@@ -78,7 +78,7 @@ static void __init xilinx_zynq_timer_init(void)xilinx_zynq_clocks_init (slcr); -xttcps_timer_init (); +clocksource_of_init (); }/**
arch/arm/mach-zynq/common.h +0 -2
@@ -17,6 +17,4 @@#ifndef __MACH_ZYNQ_COMMON_H__#define __MACH_ZYNQ_COMMON_H__ -void __init xttcps_timer_init (void ); -#endif
arch/arm/mach-zynq/timer.c +20 -23
90 lines not on this page · raw message
Signature · 1 line
1.7.9.7
PATCH v2 3/10 arm: zynq: Move timer to generic location
In reply to Michal SimekMove zynq timer out of mach folder to generic location and enable it.
---
v2: Add binding documentation
Rename zynq_timer.c to cadence_ttc_timer.c
Use CADENCE_TTC_TIMER to be able to enable this driver for others
Arnd: This is the last patch as we discuss about moving this timer to generic location.
--- .../bindings/timer/cadence,ttc-timer.txt | 17 + arch/arm/mach-zynq/Kconfig | 1 + arch/arm/mach-zynq/Makefile | 2 +- arch/arm/mach-zynq/timer.c | 436 -------------------- drivers/clocksource/Kconfig | 3 + drivers/clocksource/Makefile | 1 + drivers/clocksource/cadence_ttc_timer.c | 436 ++++++++++++++++++++ 7 files changed, 459 insertions(+), 437 deletions(-) create mode 100644 Documentation/devicetree/bindings/timer/cadence,ttc-timer.txt delete mode 100644 arch/arm/mach-zynq/timer.c create mode 100644 drivers/clocksource/cadence_ttc_timer.c
Documentation/devicetree/bindings/timer/cadence,ttc-timer.txt new file +17 -0
@@ -0,0 +1,17 @@ +Cadence TTC - Triple Timer Counter + +Required properties: +- compatible : Should be "cdns,ttc". +- reg : Specifies base physical address and size of the registers. +- interrupts : A list of 3 interrupts; one per timer channel. +- clocks: phandle to the source clock + +Example: + +ttc0: ttc0 at f8001000 { + interrupt-parent = <&intc>; + interrupts = < 0 10 4 0 11 4 0 12 4 >; + compatible = "cdns,ttc"; + reg = <0xF8001000 0x1000>; + clocks = <&cpu_clk 3>; +};
arch/arm/mach-zynq/Kconfig +1 -0
@@ -9,5 +9,6 @@ config ARCH_ZYNQ select MIGHT_HAVE_CACHE_L2X0 select USE_OF select SPARSE_IRQ + select CADENCE_TTC_TIMER help Support for Xilinx Zynq ARM Cortex A9 Platform
arch/arm/mach-zynq/Makefile +1 -1
@@ -3,4 +3,4 @@ # # Common support -obj-y := common.o timer.o +obj-y := common.o
arch/arm/mach-zynq/timer.c deleted +0 -436
436 lines not on this page · raw message
drivers/clocksource/Kconfig +3 -0
@@ -31,6 +31,9 @@ config SUNXI_TIMER config VT8500_TIMER bool +config CADENCE_TTC_TIMER + bool + config CLKSRC_NOMADIK_MTU bool depends on (ARCH_NOMADIK || ARCH_U8500)
drivers/clocksource/Makefile +1 -0
@@ -19,6 +19,7 @@ obj-$(CONFIG_ARCH_BCM2835) += bcm2835_timer.o obj-$(CONFIG_SUNXI_TIMER) += sunxi_timer.o obj-$(CONFIG_ARCH_TEGRA) += tegra20_timer.o obj-$(CONFIG_VT8500_TIMER) += vt8500_timer.o +obj-$(CONFIG_CADENCE_TTC_TIMER) += cadence_ttc_timer.o obj-$(CONFIG_ARM_ARCH_TIMER) += arm_arch_timer.o obj-$(CONFIG_CLKSRC_METAG_GENERIC) += metag_generic.o
drivers/clocksource/cadence_ttc_timer.c new file +436 -0
436 lines not on this page · raw message
Signature · 1 line
1.7.9.7
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
In reply to Michal SimekUse Cortex a9 cp15 to read scu baseaddress.
---
v2: Remove dynamic mapping because it is probably not needed
(code taken from vexpress platsmp code)
Remove scu node from DTS
Use zynq_scu_map_io() instead of scu_init() as Steffen suggested Add comment to scu_init to be sure that we know what we are doing here
--- arch/arm/mach-zynq/common.c | 34 ++++++++++++++++++++++------------ arch/arm/mach-zynq/common.h | 2 ++ 2 files changed, 24 insertions(+), 12 deletions(-)
arch/arm/mach-zynq/common.c +22 -12
@@ -33,10 +33,13 @@#include <asm/mach-types.h> #include <asm/page.h> #include <asm/pgtable.h> +#include <asm/smp_scu.h> #include <asm/hardware/cache-l2x0.h> #include "common.h" +void __iomem *scu_base; +static struct of_device_id zynq_of_bus_ids[]__initdata = { { .compatible ="simple-bus" , }, {}
@@ -56,17 +59,6 @@ static void __init xilinx_init_machine(void)of_platform_bus_probe (NULL , zynq_of_bus_ids,NULL ); } -#define SCU_PERIPH_PHYS 0xF8F00000 -#define SCU_PERIPH_SIZE SZ_8K -#define SCU_PERIPH_VIRT (VMALLOC_END - SCU_PERIPH_SIZE) - -static struct map_desc scu_desc__initdata = { - .virtual =SCU_PERIPH_VIRT , - .pfn =__phys_to_pfn (SCU_PERIPH_PHYS ), - .length =SCU_PERIPH_SIZE , - .type =MT_DEVICE , -}; -static void __init xilinx_zynq_timer_init (void ) {struct device_node *np;
@@ -81,13 +73,31 @@ static void __init xilinx_zynq_timer_init(void)clocksource_of_init (); } +static struct map_desc zynq_cortex_a9_scu_map__initdata = { + .length =SZ_256 , + .type =MT_DEVICE , +}; + +static void __init zynq_scu_map_io (void ) +{ +unsigned long base; + + base =scu_a9_get_base (); + zynq_cortex_a9_scu_map.pfn =__phys_to_pfn (base); +/* Expected address is in vmalloc area that's why simple assign here */ + zynq_cortex_a9_scu_map.virtual = base; +iotable_init (&zynq_cortex_a9_scu_map,1 ); + scu_base = (void __iomem *)base; +BUG_ON (!scu_base); +} +/** * xilinx_map_io() - Create memory mappings needed for early I/O. */ static void __init xilinx_map_io (void ) {debug_ll_io_init (); -iotable_init (&scu_desc,1 ); +zynq_scu_map_io (); }static const char *xilinx_dt_match[] = {
arch/arm/mach-zynq/common.h +2 -0
@@ -17,4 +17,6 @@#ifndef __MACH_ZYNQ_COMMON_H__#define __MACH_ZYNQ_COMMON_H__ +extern void __iomem *scu_base; +#endif
Signature · 1 line
1.7.9.7
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
arch/arm/mach-zynq/common.c line 42
Quoted 2 lines · Michal Simek
arch/arm/mach-zynq/common.c +2
-0+void __iomem *scu_base; +
This cannot be a global variable in a multiplatform kernel. Can you make it static?
Arnd
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
2013/3/26 Arnd Bergmann <[email protected]>:
Quoted 6 lines · Michal Simek
arch/arm/mach-zynq/common.c line 42
Quoted 2 lines
arch/arm/mach-zynq/common.c +2
-0+void __iomem *scu_base; +This cannot be a global variable in a multiplatform kernel. Can you make it static?
This global variable is shared because I am using it in smp code and also it will be used in pm code we have.
What is it another way how to handle this?
Because the same problem could be with zynq_slcr_base.
Or the rule is just to use zynq_ prefix for all global variables?
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
Quoted 8 lines · Michal Simek
2013/3/26 Arnd Bergmann <[email protected]>:
Quoted 6 lines · Michal Simek
arch/arm/mach-zynq/common.c line 42
Quoted 2 lines
arch/arm/mach-zynq/common.c +2
-0+void __iomem *scu_base; +This cannot be a global variable in a multiplatform kernel. Can you make it static?
Quoted 9 lines
This global variable is shared because I am using it in smp code and also it will be used in pm code we have.
What is it another way how to handle this?
Because the same problem could be with
zynq_slcr_base.Or the rule is just to use
zynq_prefix for all global variables?
Yes. The best solution is to mke variables local to some context and only pointed to by other data structures. If that doesn't work, you should try using a static variable in one file and move all code using it there. If that doesn't work, you need a global variable, but that must be uniquely named.
Arnd
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
2013/3/27 Arnd Bergmann <[email protected]>:
Quoted 16 lines · Michal Simek
Quoted 15 lines
2013/3/26 Arnd Bergmann <[email protected]>:
Quoted 6 lines · Michal Simek
arch/arm/mach-zynq/common.c line 42
Quoted 2 lines
arch/arm/mach-zynq/common.c +2
-0+void __iomem *scu_base; +This cannot be a global variable in a multiplatform kernel. Can you make it static?
This global variable is shared because I am using it in smp code and also it will be used in pm code we have.
What is it another way how to handle this?
Because the same problem could be with
zynq_slcr_base.
Quoted 8 lines
Quoted 2 lines
Or the rule is just to use
zynq_prefix for all global variables?Yes. The best solution is to mke variables local to some context and only pointed to by other data structures. If that doesn't work, you should try using a static variable in one file and move all code using it there. If that doesn't work, you need a global variable, but that must be uniquely named.
FYI: I have looked at some code and I saw that Rob is using scu_base_addr in highbank. And then pointing to it in cpuidle-calxeda.c.
Moving everything to one file is probably impossible.
And in connection to symbols/functions/variables. It means that all specific soc functions in mach
should also use specific prefix for everything.
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
Quoted 2 lines · Michal Simek
FYI: I have looked at some code and I saw that Rob is using
scu_base_addrin highbank. And then pointing to it incpuidle-calxeda.c.
Yes, the point is that it works as long as only one person uses that identifier, so we should either not use it at all, or have a single global definition shared by all ARM platforms.
Quoted 5 lines
Moving everything to one file is probably impossible.
And in connection to symbols/functions/variables. It means that all specific soc functions in mach
should also use specific prefix for everything.
That is a good rule, although for static symbols it is not a bug if they don't have a prefix.
Arnd
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
2013/3/27 Arnd Bergmann <[email protected]>:
Quoted 7 lines · Michal Simek
Quoted 2 lines
FYI: I have looked at some code and I saw that Rob is using
scu_base_addrin highbank. And then pointing to it incpuidle-calxeda.c.Yes, the point is that it works as long as only one person uses that identifier, so we should either not use it at all, or have a single global definition shared by all ARM platforms.
yep.
Quoted 8 lines
Quoted 5 lines
Moving everything to one file is probably impossible.
And in connection to symbols/functions/variables. It means that all specific soc functions in mach
should also use specific prefix for everything.That is a good rule, although for static symbols it is not a bug if they don't have a prefix.
ok. Let me check all my patches and Add there zynq_ prefix everywhere.
BTW: Have you added to zero day testing system arm multi platform or is there any arm multi platform config which developer should run to be sure that armv7 code can be compiled for armv6 and also that there is no problem with symbols?
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
Quoted 13 lines · Michal Simek
2013/3/27 Arnd Bergmann <[email protected]>:
Quoted 7 lines · Michal Simek
Quoted 2 lines
FYI: I have looked at some code and I saw that Rob is using
scu_base_addrin highbank. And then pointing to it incpuidle-calxeda.c.Yes, the point is that it works as long as only one person uses that identifier, so we should either not use it at all, or have a single global definition shared by all ARM platforms.
yep.
Quoted 2 lines
Quoted 2 lines
Moving everything to one file is probably impossible.
Quoted 9 lines
Quoted 6 lines
Quoted 3 lines
And in connection to symbols/functions/variables. It means that all specific soc functions in mach
should also use specific prefix for everything.That is a good rule, although for static symbols it is not a bug if they don't have a prefix.
ok. Let me check all my patches and Add there
zynq_prefix everywhere.
While you're at, how about getting rid of the xilinx in the names and use just zynq_* ?
I would also suggest s/xttc/ttc/g in your timer patch. It is up to you, but I think the x doesn't make any sense anymore.
Regards,
Steffen
Signature · 4 lines
Pengutronix e.K. | | Industrial Linux Solutions | http://www.pengutronix.de/ | Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
2013/3/27 Steffen Trumtrar <[email protected]>:
Quoted 18 lines · Michal Simek
Quoted 17 lines
2013/3/27 Arnd Bergmann <[email protected]>:
Quoted 7 lines · Michal Simek
Quoted 2 lines
FYI: I have looked at some code and I saw that Rob is using
scu_base_addrin highbank. And then pointing to it incpuidle-calxeda.c.Yes, the point is that it works as long as only one person uses that identifier, so we should either not use it at all, or have a single global definition shared by all ARM platforms.
yep.
Quoted 6 lines
Quoted 5 lines
Moving everything to one file is probably impossible.
And in connection to symbols/functions/variables. It means that all specific soc functions in mach
should also use specific prefix for everything.
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
That is a good rule, although for static symbols it is not a bug if they don't have a prefix.
ok. Let me check all my patches and Add there
zynq_prefix everywhere.While you're at, how about getting rid of the xilinx in the names and use just zynq_* ?
yep. That's what I am going to do.
Quoted 2 lines
I would also suggest s/xttc/ttc/g in your timer patch. It is up to you, but I think the x doesn't make any sense anymore.
Agree. I don't know the historical reason for this. Also in xilinx git repo there was PSS instead of PS everywhere.
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
In reply to Michal SimekQuoted 4 lines · Michal Simek
BTW: Have you added to zero day testing system arm multi platform or is there any arm multi platform config which developer should run to be sure that armv7 code can be compiled for armv6 and also that there is no problem with symbols?
I'm only occasionally doing build tested right now, not all the time.
Arnd
PATCH v2 4/10 arm: zynq: Load scu baseaddress at run time
2013/3/27 Arnd Bergmann <[email protected]>:
Quoted 7 lines · Michal Simek
Quoted 4 lines
BTW: Have you added to zero day testing system arm multi platform or is there any arm multi platform config which developer should run to be sure that armv7 code can be compiled for armv6 and also that there is no problem with symbols?
I'm only occasionally doing build tested right now, not all the time.
What about to provide this config to Fengguang to add this to zero day testing system which will check this all the time?
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
In reply to Michal SimekCreate separate slcr driver instead of pollute common code.
--- arch/arm/mach-zynq/Makefile | 2 +- arch/arm/mach-zynq/common.c | 10 +------ arch/arm/mach-zynq/common.h | 3 ++ arch/arm/mach-zynq/slcr.c | 69 +++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 74 insertions(+), 10 deletions(-) create mode 100644 arch/arm/mach-zynq/slcr.c
arch/arm/mach-zynq/Makefile +1 -1
@@ -3,4 +3,4 @@ # # Common support -obj-y := common.o +obj-y := common.o slcr.o
arch/arm/mach-zynq/common.c +1 -9
@@ -61,15 +61,7 @@ static void __init xilinx_init_machine(void)static void __init xilinx_zynq_timer_init (void ) { -struct device_node *np; -void __iomem *slcr; - - np =of_find_compatible_node (NULL ,NULL ,"xlnx,zynq-slcr" ); - slcr =of_iomap (np,0 ); -WARN_ON (!slcr); - -xilinx_zynq_clocks_init (slcr); - +slcr_init ();clocksource_of_init (); }
arch/arm/mach-zynq/common.h +3 -0
@@ -17,6 +17,9 @@#ifndef __MACH_ZYNQ_COMMON_H__#define __MACH_ZYNQ_COMMON_H__ +extern int slcr_init (void ); + +extern void __iomem *zynq_slcr_base;extern void __iomem *scu_base;#endif
arch/arm/mach-zynq/slcr.c new file +69 -0
69 lines not on this page · raw message
Signature · 1 line
1.7.9.7
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
Quoted 3 lines · Michal Simek
Create separate slcr driver instead of pollute common code.
Signed-off-by Michal Simek <[email protected]>
Can't you move that code into the zynq_cpu_clk_setup function instead, and only call of_clk_init(NULL) from platform code?
I would like to eventually add a default call to of_clk_init(NULL) for all platforms, and that would be a step in that direction.
Arnd
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
Hi Arnd,
Quoted 8 lines · Arnd Bergmann
Quoted 3 lines · Michal Simek
Create separate slcr driver instead of pollute common code.
Signed-off-by Michal Simek <[email protected]>Can't you move that code into the
zynq_cpu_clk_setupfunction instead, and only callof_clk_init(NULL)from platform code?
if you are talking about the slcr function, than moving it into a separate file is the right move. This should actually become a real driver. The slcr is master over all clock, reset, pinmux and ddr registers. And as all those registers can be locked/unlocked via a slcr register (for whatever reason one would do that), there should be one master that controls this space.
Quoted 3 lines
I would like to eventually add a default call to
of_clk_init(NULL)for all platforms, and that would be a step in that direction.
Regards,
Steffen
Signature · 4 lines
Pengutronix e.K. | | Industrial Linux Solutions | http://www.pengutronix.de/ | Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
Quoted 9 lines · Steffen Trumtrar
Quoted 8 lines · Arnd Bergmann
Quoted 3 lines · Michal Simek
Create separate slcr driver instead of pollute common code.
Signed-off-by Michal Simek <[email protected]>Can't you move that code into the
zynq_cpu_clk_setupfunction instead, and only callof_clk_init(NULL)from platform code?
Quoted 6 lines
if you are talking about the slcr function, than moving it into a separate file is the right move. This should actually become a real driver. The slcr is master over all clock, reset, pinmux and ddr registers. And as all those registers can be locked/unlocked via a slcr register (for whatever reason one would do that), there should be one master that controls this space.
Ok, I see. Thanks for the explanation. Should this be using the drivers/mfd/syscon.c infrastructure then?
Arnd
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
Quoted 17 lines · Arnd Bergmann
Quoted 9 lines · Steffen Trumtrar
Quoted 8 lines · Arnd Bergmann
Quoted 3 lines · Michal Simek
Create separate slcr driver instead of pollute common code.
Signed-off-by Michal Simek <[email protected]>Can't you move that code into the
zynq_cpu_clk_setupfunction instead, and only callof_clk_init(NULL)from platform code?Quoted 6 lines
if you are talking about the slcr function, than moving it into a separate file is the right move. This should actually become a real driver. The slcr is master over all clock, reset, pinmux and ddr registers. And as all those registers can be locked/unlocked via a slcr register (for whatever reason one would do that), there should be one master that controls this space.
Quoted 2 lines
Ok, I see. Thanks for the explanation. Should this be using the
drivers/mfd/syscon.cinfrastructure then?
A quick look suggests that this might be the way to go. I wasn't aware of that.
Thanks,
Steffen
Signature · 4 lines
Pengutronix e.K. | | Industrial Linux Solutions | http://www.pengutronix.de/ | Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
2013/3/27 Steffen Trumtrar <[email protected]>:
Quoted 18 lines · Arnd Bergmann
Quoted 17 lines · Steffen Trumtrar
Quoted 15 lines · Arnd Bergmann
Quoted 8 lines · Michal Simek
Quoted 3 lines
Create separate slcr driver instead of pollute common code.
Signed-off-by Michal Simek <[email protected]>Can't you move that code into the
zynq_cpu_clk_setupfunction instead, and only callof_clk_init(NULL)from platform code?if you are talking about the slcr function, than moving it into a separate file is the right move. This should actually become a real driver. The slcr is master over all clock, reset, pinmux and ddr registers. And as all those registers can be locked/unlocked via a slcr register (for whatever reason one would do that), there should be one master that controls this space.
Quoted 4 lines
Quoted 2 lines
Ok, I see. Thanks for the explanation. Should this be using the
drivers/mfd/syscon.cinfrastructure then?A quick look suggests that this might be the way to go. I wasn't aware of that.
I have looked at syscon driver but the problem is when this driver is ready to use.
It is too late because bus probing is done in init_machine but we need to unlock slcr before clk and timer init functions are called. For unlocking we have to also map it.
of_clk_init(NULL) is called from zynq clock. It is no problem to move it to common code but still we have to map slcr register in the driver and pass this address to clk driver.
Currently we are passing slcr reg base as xilinx_zynq_clocks_init() parameter to avoid to have extern in C file.
I see that Rob in highbank clk choose a way with extern. of_iomap is done in highbank_timer_init and in the same function calls of_clk_init()
and clocksource_of_init().
Syscon will be nice to use much later but not for this core stuff. But maybe you know nicer way how to handle it.
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
In reply to Michal SimekQuoted 1 line · Michal Simek
Create separate slcr driver instead of pollute common code.
Nitpicking: A native speaker would say "polluting common code".
Philip
Quoted 131 lines
Signed-off-by Michal Simek <[email protected]>--- arch/arm/mach-zynq/Makefile | 2 +- arch/arm/mach-zynq/common.c | 10 +------ arch/arm/mach-zynq/common.h | 3 ++ arch/arm/mach-zynq/slcr.c | 69 +++++++++++++++++++++++++++++++++++++++++++ 4 files changed, 74 insertions(+), 10 deletions(-)create mode 100644
arch/arm/mach-zynq/slcr.carch/arm/mach-zynq/Makefile +1
-1@@ -3,4 +3,4 @@ # # Common support -obj-y := common.o +obj-y := common.o slcr.oarch/arm/mach-zynq/common.c +1
-9@@ -61,15 +61,7 @@ static void __init xilinx_init_machine(void)static void __init xilinx_zynq_timer_init (void ) { -struct device_node *np; -void __iomem *slcr; - - np =of_find_compatible_node (NULL ,NULL ,"xlnx,zynq-slcr" ); - slcr =of_iomap (np,0 ); -WARN_ON (!slcr); - -xilinx_zynq_clocks_init (slcr); - +slcr_init ();clocksource_of_init (); }arch/arm/mach-zynq/common.h +3
-0@@ -17,6 +17,9 @@#ifndef __MACH_ZYNQ_COMMON_H__#define __MACH_ZYNQ_COMMON_H__ +extern int slcr_init (void ); + +extern void __iomem *zynq_slcr_base;extern void __iomem *scu_base;#endif arch/arm/mach-zynq/slcr.c new file +69
-0@@ -0,0 +1,69 @@ +/* +* Xilinx SLCR driver +* +* Copyright (c) 2011-2013 Xilinx Inc. +* +* This program is free software; you can redistribute it and/or +* modify it under the terms of the GNU General Public License +* as published by the Free Software Foundation; either version +* 2 of the License, or (at your option) any later version. +* +* You should have received a copy of the GNU General Public +* License along with this program; if not, write to the Free +* Software Foundation, Inc., 675 Mass Ave, Cambridge, MA +* 02139, USA. +*/ + +#include <linux/export.h> +#include <linux/io.h> +#include <linux/fs.h> +#include <linux/interrupt.h> +#include <linux/init.h> +#include <linux/kernel.h> +#include <linux/module.h> +#include <linux/of_address.h> +#include <linux/uaccess.h> +#include <linux/platform_device.h> +#include <linux/slab.h> +#include <linux/string.h> +#include <linux/clk/zynq.h> +#include "common.h" + +#define SLCR_UNLOCK_MAGIC 0xDF0D +#define SLCR_UNLOCK 0x8/* SCLR unlock register */ + +void __iomem *zynq_slcr_base; + +/** +* xslcr_init() +* Returns 0 on success, negative errno otherwise. +* +* Called early during boot from platform code to remap SLCR area. +*/ +int __init slcr_init (void ) +{ +struct device_node *np; + + np =of_find_compatible_node (NULL ,NULL ,"xlnx,zynq-slcr" ); +if (!np) { +pr_err ("%s: no slcr node found\n" , __func__); +BUG (); + } + + zynq_slcr_base =of_iomap (np,0 ); +if (!zynq_slcr_base) { +pr_err ("%s: Unable to map I/O memory\n" , __func__); +BUG (); + } + +/* unlock the SLCR so that registers can be changed */ +writel (SLCR_UNLOCK_MAGIC , zynq_slcr_base +SLCR_UNLOCK ); + +pr_info ("%s mapped to %p\n" , np->name, zynq_slcr_base); + +xilinx_zynq_clocks_init (zynq_slcr_base); + +of_node_put (np); + +return 0 ; +}
PATCH v2 5/10 arm: zynq: Move slcr initialization to separate file
Hi Philip,
2013/3/27 Philip Balister <[email protected]>:
Quoted 6 lines · Michal Simek
Quoted 2 lines
Create separate slcr driver instead of pollute common code.
Nitpicking: A native speaker would say "polluting common code".
Feel free to check my english. :-)
M
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 6/10 arm: zynq: Add support for system reset
In reply to Michal SimekDo system reset via slcr registers.
---
v2: Fix comment
use writel instead of __raw_writel
Do not use PSS - use PS instead
--- arch/arm/mach-zynq/common.c | 6 ++++++ arch/arm/mach-zynq/common.h | 1 + arch/arm/mach-zynq/slcr.c | 28 ++++++++++++++++++++++++++++ 3 files changed, 35 insertions(+)
arch/arm/mach-zynq/common.c +6 -0
@@ -92,6 +92,11 @@ static void __init xilinx_map_io(void)zynq_scu_map_io (); } +static void xilinx_system_reset (char mode,const char *cmd) +{ +slcr_system_reset (); +} +static const char *xilinx_dt_match[] = {"xlnx,zynq-zc702" ,"xlnx,zynq-7000" ,
@@ -104,4 +109,5 @@ MACHINE_START(XILINX_EP107, "Xilinx Zynq Platform") .init_machine = xilinx_init_machine, .init_time = xilinx_zynq_timer_init, .dt_compat = xilinx_dt_match, + .restart = xilinx_system_reset,MACHINE_END
arch/arm/mach-zynq/common.h +1 -0
@@ -18,6 +18,7 @@#define __MACH_ZYNQ_COMMON_H__extern int slcr_init (void ); +extern void slcr_system_reset (void );extern void __iomem *zynq_slcr_base;extern void __iomem *scu_base;
arch/arm/mach-zynq/slcr.c +28 -0
@@ -32,9 +32,37 @@#define SLCR_UNLOCK_MAGIC 0xDF0D#define SLCR_UNLOCK 0x8/* SCLR unlock register */ +#define SLCR_PS_RST_CTRL_OFFSET 0x200/* PS Software Reset Control */ +#define SLCR_REBOOT_STATUS 0x258/* PS Reboot Status */ +void __iomem *zynq_slcr_base;/** +* xslcr_system_reset - Reset the entire system. +* +*/ +void slcr_system_reset (void ) +{ +u32 reboot; + +/* +* Unlock the SLCR then reset the system. +* Note that this seems to require raw i/o +* functions or there's a lockup? +*/ +writel (SLCR_UNLOCK_MAGIC , zynq_slcr_base +SLCR_UNLOCK ); + +/* +* Clear 0x0F000000 bits of reboot status register to workaround +* the FSBL not loading the bitstream after soft-reboot +* This is a temporary solution until we know more. +*/ + reboot =readl (zynq_slcr_base +SLCR_REBOOT_STATUS ); +writel (reboot &0xF0FFFFFF , zynq_slcr_base +SLCR_REBOOT_STATUS ); +writel (1 , zynq_slcr_base +SLCR_PS_RST_CTRL_OFFSET ); +} + +/** * xslcr_init() * Returns 0 on success, negative errno otherwise. *
Signature · 1 line
1.7.9.7
PATCH v2 7/10 arm: zynq: Add support for pmu
In reply to Michal SimekZynq is standard PMU with 2 interrupt per core. There is also access via register which is not used right now.
--- arch/arm/boot/dts/zynq-7000.dtsi | 7 +++++++ 1 file changed, 7 insertions(+)
arch/arm/boot/dts/zynq-7000.dtsi +7 -0
@@ -15,6 +15,13 @@ / { compatible = "xlnx,zynq-7000"; + pmu { + compatible = "arm,cortex-a9-pmu"; + interrupts = <0 5 4>, <0 6 4>; + interrupt-parent = <&intc>; + reg = < 0xf8891000 0x1000 0xf8893000 0x1000 >; + }; + amba { compatible = "simple-bus"; #address-cells = <1>;
Signature · 1 line
1.7.9.7
PATCH v2 8/10 arm: zynq: Add smp support
In reply to Michal SimekZynq is dual core Cortex A9 which starts always
at zero. Using simple trampoline ensure long jump
to secondary_startup code.
---
v2: Remove boot_lock from secondary_init
Incorporate Steffen changes in smp code
Use flush range instead of simple flush
Add HAVE_SMP
Add support for the booting kernels with
non zero starting address
--- arch/arm/mach-zynq/Kconfig | 1 + arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/common.h | 11 ++++ arch/arm/mach-zynq/headsmp.S | 24 +++++++ arch/arm/mach-zynq/platsmp.c | 149 ++++++++++++++++++++++++++++++++++++++++++ arch/arm/mach-zynq/slcr.c | 29 ++++++++ 7 files changed, 216 insertions(+) create mode 100644 arch/arm/mach-zynq/headsmp.S create mode 100644 arch/arm/mach-zynq/platsmp.c
arch/arm/mach-zynq/Kconfig +1 -0
@@ -8,6 +8,7 @@ config ARCH_ZYNQ select ICST select MIGHT_HAVE_CACHE_L2X0 select USE_OF + select HAVE_SMP select SPARSE_IRQ select CADENCE_TTC_TIMER help
arch/arm/mach-zynq/Makefile +1 -0
@@ -4,3 +4,4 @@ # Common support obj-y := common.o slcr.o +obj-$(CONFIG_SMP) += headsmp.o platsmp.o
arch/arm/mach-zynq/common.c +1 -0
@@ -104,6 +104,7 @@ static const char *xilinx_dt_match[] = { };MACHINE_START (XILINX_EP107 ,"Xilinx Zynq Platform" ) + .smp = smp_ops(zynq_smp_ops), .map_io = xilinx_map_io, .init_irq = irqchip_init, .init_machine = xilinx_init_machine,
arch/arm/mach-zynq/common.h +11 -0
@@ -19,6 +19,17 @@extern int slcr_init (void );extern void slcr_system_reset (void ); +extern void slcr_cpu_stop (int cpu); +extern void slcr_cpu_start (int cpu); + +#ifdef CONFIG_SMP +extern void secondary_startup (void ); +extern char zynq_secondary_trampoline; +extern char zynq_secondary_trampoline_jump; +extern char zynq_secondary_trampoline_end; +extern int __cpuinit zynq_cpun_start (u32 address,int cpu); +extern struct smp_operations zynq_smp_ops__initdata ; +#endif extern void __iomem *zynq_slcr_base;extern void __iomem *scu_base;
arch/arm/mach-zynq/headsmp.S new file +24 -0
@@ -0,0 +1,24 @@ +/* + * Copyright (c) 2013 Steffen Trumtrar <[email protected]> + * Copyright (c) 2012-2013 Xilinx + * + * 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/linkage.h> +#include <linux/init.h> + + __CPUINIT + +ENTRY(zynq_secondary_trampoline) + ldr r0, [pc] + bx r0 +.globl zynq_secondary_trampoline_jump +zynq_secondary_trampoline_jump: + /* Space for jumping address */ + .word /* cpu 1 */ +.globl zynq_secondary_trampoline_end +zynq_secondary_trampoline_end: + +ENDPROC(zynq_secondary_trampoline)
arch/arm/mach-zynq/platsmp.c new file +149 -0
149 lines not on this page · raw message
arch/arm/mach-zynq/slcr.c +29 -0
@@ -33,6 +33,11 @@#define SLCR_UNLOCK 0x8/* SCLR unlock register */ #define SLCR_PS_RST_CTRL_OFFSET 0x200/* PS Software Reset Control */ + +#define SLCR_A9_CPU_CLKSTOP 0x10 +#define SLCR_A9_CPU_RST 0x1 + +#define SLCR_A9_CPU_RST_CTRL 0x244/* CPU Software Reset Control */ #define SLCR_REBOOT_STATUS 0x258/* PS Reboot Status */ void __iomem *zynq_slcr_base;
@@ -63,6 +68,30 @@ void slcr_system_reset(void) }/** +* slcr_cpu_start - Start cpu +* @cpu: cpu number +*/ +void slcr_cpu_start (int cpu) +{ +/* enable CPUn */ +writel (SLCR_A9_CPU_CLKSTOP << cpu, + zynq_slcr_base +SLCR_A9_CPU_RST_CTRL ); +/* enable CLK for CPUn */ +writel (0x0 << cpu, zynq_slcr_base +SLCR_A9_CPU_RST_CTRL ); +} + +/** +* slcr_cpu_stop - Stop cpu +* @cpu: cpu number +*/ +void slcr_cpu_stop (int cpu) +{ +/* stop CLK and reset CPUn */ +writel ((SLCR_A9_CPU_CLKSTOP |SLCR_A9_CPU_RST ) << cpu, + zynq_slcr_base +SLCR_A9_CPU_RST_CTRL ); +} + +/** * xslcr_init() * Returns 0 on success, negative errno otherwise. *
Signature · 1 line
1.7.9.7
PATCH v2 8/10 arm: zynq: Add smp support
2013/3/26 Michal Simek <[email protected]>:
Quoted 196 lines
Zynq is dual core Cortex A9 which starts always
at zero. Using simple trampoline ensure long jump
tosecondary_startupcode.Signed-off-by Michal Simek <[email protected]>Signed-off-by Steffen Trumtrar <[email protected]>---
v2: Removeboot_lockfromsecondary_init
Incorporate Steffen changes in smp code
Use flush range instead of simple flush
AddHAVE_SMP
Add support for the booting kernels with
non zero starting address--- arch/arm/mach-zynq/Kconfig | 1 + arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/common.c | 1 + arch/arm/mach-zynq/common.h | 11 ++++ arch/arm/mach-zynq/headsmp.S | 24 +++++++ arch/arm/mach-zynq/platsmp.c | 149 ++++++++++++++++++++++++++++++++++++++++++ arch/arm/mach-zynq/slcr.c | 29 ++++++++ 7 files changed, 216 insertions(+) create mode 100644 arch/arm/mach-zynq/headsmp.S create mode 100644 arch/arm/mach-zynq/platsmp.carch/arm/mach-zynq/Kconfig +1
-0@@ -8,6 +8,7 @@ config ARCH_ZYNQ select ICST select MIGHT_HAVE_CACHE_L2X0 select USE_OF + select HAVE_SMP select SPARSE_IRQ select CADENCE_TTC_TIMER helparch/arm/mach-zynq/Makefile +1
-0@@ -4,3 +4,4 @@ # Common support obj-y := common.o slcr.o +obj-$(CONFIG_SMP) += headsmp.o platsmp.oarch/arm/mach-zynq/common.c +1
-0@@ -104,6 +104,7 @@ static const char *xilinx_dt_match[] = { };MACHINE_START (XILINX_EP107 ,"Xilinx Zynq Platform" ) + .smp = smp_ops(zynq_smp_ops), .map_io = xilinx_map_io, .init_irq = irqchip_init, .init_machine = xilinx_init_machine,arch/arm/mach-zynq/common.h +11
-0@@ -19,6 +19,17 @@extern int slcr_init (void );extern void slcr_system_reset (void ); +extern void slcr_cpu_stop (int cpu); +extern void slcr_cpu_start (int cpu); + +#ifdef CONFIG_SMP +extern void secondary_startup (void ); +extern char zynq_secondary_trampoline; +extern char zynq_secondary_trampoline_jump; +extern char zynq_secondary_trampoline_end; +extern int __cpuinit zynq_cpun_start (u32 address,int cpu); +extern struct smp_operations zynq_smp_ops__initdata ; +#endif extern void __iomem *zynq_slcr_base;extern void __iomem *scu_base;arch/arm/mach-zynq/headsmp.S new file +24
-0@@ -0,0 +1,24 @@ +/* + * Copyright (c) 2013 Steffen Trumtrar <[email protected]> + * Copyright (c) 2012-2013 Xilinx + * + * 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/linkage.h> +#include <linux/init.h> + + __CPUINIT + +ENTRY(zynq_secondary_trampoline) + ldr r0, [pc] + bx r0 +.globl zynq_secondary_trampoline_jump +zynq_secondary_trampoline_jump: + /* Space for jumping address */ + .word /* cpu 1 */ +.globl zynq_secondary_trampoline_end +zynq_secondary_trampoline_end: + +ENDPROC(zynq_secondary_trampoline)arch/arm/mach-zynq/platsmp.c new file +80
-0@@ -0,0 +1,149 @@ +/* +* This file contains Xilinx specific SMP code, used to start up +* the second processor. +* +* Copyright (C) 2011-2013 Xilinx +* +* based on linux/arch/arm/mach-realview/platsmp.c +* +* Copyright (C) 2002 ARM Ltd. +* +* This software is licensed under the terms of the GNU General Public +* License version 2, as published by the Free Software Foundation, and +* may be copied, distributed, and modified under those terms. +* +* This program is distributed in the hope that it will be useful, +* but WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +* GNU General Public License for more details. +*/ + +#include <linux/export.h> +#include <linux/jiffies.h> +#include <linux/init.h> +#include <linux/io.h> +#include <asm/cacheflush.h> +#include <asm/smp_scu.h> +#include <linux/irqchip/arm-gic.h> +#include "common.h" + +/* +* Store number of cores in the system +* Because of scu_get_core_count() must be in __init section and can't +* be called from zynq_cpun_start() because it is in __cpuinit section. +*/ +static int ncores; + +/* Secondary CPU kernel startup is a 2 step process. The primary CPU +* starts the secondary CPU by giving it the address of the kernel and +* then sending it an event to wake it up. The secondary CPU then +* starts the kernel and tells the primary CPU it's up and running. +*/ +static void __cpuinit zynq_secondary_init (unsigned int cpu) +{ +/* +* if any interrupts are already enabled for the primary +* core (e.g. timer irq), then they will not have been enabled +* for us: do so +*/ +gic_secondary_init (0 ); +} + +int __cpuinit zynq_cpun_start (u32 address,int cpu) +{ +u32 trampoline_code_size = &zynq_secondary_trampoline_end - + &zynq_secondary_trampoline; + +if (cpu > ncores) { +pr_warn ("CPU No. is not available in the system\n" ); +return -1 ; + } + +/* MS: Expectation that SLCR are directly map and accessible */ +/* Not possible to jump to non aligned address */ +if (!(address &3 ) && (!address || (address >= trampoline_code_size))) { +/* Store pointer to ioremap area which points to address 0x0 */ +static u8 __iomem *zero; +u32 trampoline_size = &zynq_secondary_trampoline_jump - + &zynq_secondary_trampoline; + +slcr_cpu_stop (cpu); + +if (__pa (PAGE_OFFSET )) { + zero =ioremap (0 , trampoline_code_size); +if (!zero) { +pr_warn ("BOOTUP jump vectors not accessible\n" ); +return -1 ; + } + }else { + zero = (__force u8 __iomem *)PAGE_OFFSET ; + }
arch/arm/mach-zynq/platsmp.c line 90
Quoted 10 lines
arch/arm/mach-zynq/platsmp.c new file +10
-0+ +/* +* This is elegant way how to jump to any address +* 0x0: Load address at 0x8 to r0 +* 0x4: Jump by mov instruction +* 0x8: Jumping address +*/ +memcpy ((__force void *)zero, &zynq_secondary_trampoline, + trampoline_size); +writel (address, zero + trampoline_size +sizeof (int ));
Please remove this + sizeof(int). I forget to remove it from this version. Because I tested Rob's idea to have jump table based on cpu id.
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 9/10 arm: zynq: Add hotplug support
In reply to Michal Simek---
v2: Change platform_cpu_die name function to zynq_platform_cpu_dio
Fix header
Add cpu compilation flags to be able to build on V6 too
--- arch/arm/mach-zynq/Makefile | 3 ++ arch/arm/mach-zynq/common.h | 3 ++ arch/arm/mach-zynq/hotplug.c | 104 ++++++++++++++++++++++++++++++++++++++++++ arch/arm/mach-zynq/platsmp.c | 3 ++ 4 files changed, 113 insertions(+) create mode 100644 arch/arm/mach-zynq/hotplug.c
arch/arm/mach-zynq/Makefile +3 -0
@@ -4,4 +4,7 @@ # Common support obj-y := common.o slcr.o +CFLAGS_REMOVE_hotplug.o =-march=armv6k +CFLAGS_hotplug.o =-Wa,-march=armv7-a -mcpu=cortex-a9 +obj-$(CONFIG_HOTPLUG_CPU) += hotplug.o obj-$(CONFIG_SMP) += headsmp.o platsmp.o
arch/arm/mach-zynq/common.h +3 -0
@@ -34,4 +34,7 @@ extern struct smp_operations zynq_smp_ops __initdata;extern void __iomem *zynq_slcr_base;extern void __iomem *scu_base; +/* Hotplug */ +extern void zynq_platform_cpu_die (unsigned int cpu); +#endif
arch/arm/mach-zynq/hotplug.c new file +104 -0
104 lines not on this page · raw message
arch/arm/mach-zynq/platsmp.c +3 -0
@@ -146,4 +146,7 @@ struct smp_operations zynq_smp_ops __initdata = { .smp_prepare_cpus = zynq_smp_prepare_cpus, .smp_secondary_init = zynq_secondary_init, .smp_boot_secondary = zynq_boot_secondary, +#ifdef CONFIG_HOTPLUG_CPU + .cpu_die = zynq_platform_cpu_die, +#endif };
Signature · 1 line
1.7.9.7
PATCH v2 10/10 arm: zynq: Add cpuidle support
In reply to Michal SimekAdd support for cpuidle.
---
v2: Fix file header
--- arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.c
arch/arm/mach-zynq/Makefile +1 -0
@@ -7,4 +7,5 @@ obj-y := common.o slcr.o CFLAGS_REMOVE_hotplug.o =-march=armv6k CFLAGS_hotplug.o =-Wa,-march=armv7-a -mcpu=cortex-a9 obj-$(CONFIG_HOTPLUG_CPU) += hotplug.o +obj-$(CONFIG_CPU_IDLE) += cpuidle.o obj-$(CONFIG_SMP) += headsmp.o platsmp.o
arch/arm/mach-zynq/cpuidle.c new file +133 -0
133 lines not on this page · raw message
Signature · 1 line
1.7.9.7
PATCH v2 10/10 arm: zynq: Add cpuidle support
Quoted 10 lines · Michal Simek
Add support for cpuidle.
Signed-off-by Michal Simek <[email protected]>---
v2: Fix file header--- arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.c
Can you move that file to drivers/cpuidle instead?
arch/arm/mach-zynq/cpuidle.c line 117
Quoted 12 lines
arch/arm/mach-zynq/cpuidle.c new file +12
-0+/* Initialize CPU idle by registering the idle states */ +static int xilinx_init_cpuidle (void ) +{ +unsigned int cpu; +struct cpuidle_device *device; +int ret; + + ret =cpuidle_register_driver (&xilinx_idle_driver); +if (ret) { +pr_err ("Registering Xilinx CpuIdle Driver failed.\n" ); +return ret; + }
I think you have to check that you actually run on a Zynq system before registering the driver.
Arnd
PATCH v2 10/10 arm: zynq: Add cpuidle support
2013/3/26 Arnd Bergmann <[email protected]>:
Quoted 12 lines · Michal Simek
Quoted 10 lines
Add support for cpuidle.
Signed-off-by Michal Simek <[email protected]>---
v2: Fix file header--- arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.c
Quoted 1 line
Can you move that file to
drivers/cpuidleinstead?
I can't see any problem in it right now.
Quoted 7 lines
Quoted 7 lines
arch/arm/mach-zynq/cpuidle.c new file +7
-0+/* Initialize CPU idle by registering the idle states */ +static int xilinx_init_cpuidle (void ) +{ +unsigned int cpu; +struct cpuidle_device *device; +int ret; +
Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 117
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +5
-0+ ret =cpuidle_register_driver (&xilinx_idle_driver); +if (ret) { +pr_err ("Registering Xilinx CpuIdle Driver failed.\n" ); +return ret; + }I think you have to check that you actually run on a Zynq system before registering the driver.
Is there any elegant way how to do it?
I see that Rob is checking compatible machine with of_machine_is_compatible().
Thanks,
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 10/10 arm: zynq: Add cpuidle support
Quoted 10 lines · Michal Simek
2013/3/26 Arnd Bergmann <[email protected]>:
Quoted 7 lines · Michal Simek
Quoted 4 lines
arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.cCan you move that file to
drivers/cpuidleinstead?I can't see any problem in it right now.
Ok.
Adding Daniel Lezcano to CC, he's currently doing a lot of work on the cpuidle drivers and may have some input as well.
Quoted 16 lines
Quoted 7 lines
Quoted 7 lines
arch/arm/mach-zynq/cpuidle.c new file +7
-0+/* Initialize CPU idle by registering the idle states */ +static int xilinx_init_cpuidle (void ) +{ +unsigned int cpu; +struct cpuidle_device *device; +int ret; +Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 117
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +5
-0+ ret =cpuidle_register_driver (&xilinx_idle_driver); +if (ret) { +pr_err ("Registering Xilinx CpuIdle Driver failed.\n" ); +return ret; + }I think you have to check that you actually run on a Zynq system before registering the driver.
Quoted 3 lines
Is there any elegant way how to do it?
I see that Rob is checking compatible machine withof_machine_is_compatible().
Most drivers use some resource that they can check the presence of, which is better than checking the global "compatible" property of the system. I don't see any of that in your driver, but I may be missing something. What is it specifically that makes this cpuidle driver special to Zynq and different from other cpuidle drivers? Is that difference something we can describe using the device tree in more specific terms than the root compatible property?
Arnd
PATCH v2 10/10 arm: zynq: Add cpuidle support
Quoted 8 lines · Arnd Bergmann
Quoted 7 lines · Michal Simek
2013/3/26 Arnd Bergmann <[email protected]>:
Quoted 6 lines · Michal Simek
Quoted 4 lines
arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.c
Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Can you move that file to
drivers/cpuidleinstead?I can't see any problem in it right now.
Ok.
Adding Daniel Lezcano to CC, he's currently doing a lot of work on the cpuidle drivers and may have some input as well.
Thanks Arnd,
I commented the driver.
I am in favor for moving also this driver to drivers/cpuidle.
I suggest in the future, you nack any new drivers which is not located in drivers/cpuidle. Let's try to move the current ones out of arch/arm/mach-* and concentrate the location in a single place.
Maintaining all these drivers by jumping branch to branch like a monkey is really painful :)
Quoted 30 lines
Quoted 16 lines
Quoted 7 lines
Quoted 7 lines
arch/arm/mach-zynq/cpuidle.c new file +7
-0+/* Initialize CPU idle by registering the idle states */ +static int xilinx_init_cpuidle (void ) +{ +unsigned int cpu; +struct cpuidle_device *device; +int ret; +Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 117
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +5
-0+ ret =cpuidle_register_driver (&xilinx_idle_driver); +if (ret) { +pr_err ("Registering Xilinx CpuIdle Driver failed.\n" ); +return ret; + }I think you have to check that you actually run on a Zynq system before registering the driver.
Quoted 3 lines
Is there any elegant way how to do it?
I see that Rob is checking compatible machine withof_machine_is_compatible().Most drivers use some resource that they can check the presence of, which is better than checking the global "compatible" property of the system. I don't see any of that in your driver, but I may be missing something. What is it specifically that makes this cpuidle driver special to Zynq and different from other cpuidle drivers? Is that difference something we can describe using the device tree in more specific terms than the root compatible property?
Arnd
Signature · 5 lines
<http://www.linaro.org/> Linaro.org ? Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
PATCH v2 10/10 arm: zynq: Add cpuidle support
In reply to Michal SimekQuoted 3 lines · Michal Simek
Add support for cpuidle.
Signed-off-by Michal Simek <[email protected]>
Hi Michal,
at the first glance, the driver won't enter in the self refresh state.
If I am right, by checking:
/sys/devices/system/cpu/cpu0/cpuidle/state1/usage, it should be always zero.
Also, I see there is self refresh but you save the cpu context, so it is cpu shutdown no ?
In this case, 3 states would be needed, WFI, self-refresh, powerdown, no ?
The comments below applies for this driver based on linux-pm-next.
https://git.kernel.org/cgit/linux/kernel/git/rafael/linux-pm.git/
Quoted 49 lines
---
v2: Fix file header--- arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.carch/arm/mach-zynq/Makefile +1
-0@@ -7,4 +7,5 @@ obj-y := common.o slcr.o CFLAGS_REMOVE_hotplug.o =-march=armv6k CFLAGS_hotplug.o =-Wa,-march=armv7-a -mcpu=cortex-a9 obj-$(CONFIG_HOTPLUG_CPU) += hotplug.o +obj-$(CONFIG_CPU_IDLE) += cpuidle.o obj-$(CONFIG_SMP) += headsmp.o platsmp.oarch/arm/mach-zynq/cpuidle.c new file +25
-0@@ -0,0 +1,133 @@ +/* +* Copyright (C) 2012-2013 Xilinx +* +* CPU idle support for Xilinx +* +* based on arch/arm/mach-at91/cpuidle.c +* +* This file is licensed under the terms of the GNU General Public +* License version 2. This program is licensed "as is" without any +* warranty of any kind, whether express or implied. +* +* The cpu idle uses wait-for-interrupt and RAM self refresh in order +* to implement two idle states - +* #1 wait-for-interrupt +* #2 wait-for-interrupt and RAM self refresh +* +* Note that this code is only intended as a prototype and is not tested +* well yet, or tuned for the Xilinx Cortex A9. Also note that for a +* tickless kernel, high res timers must not be turned on. The cpuidle +* framework must also be turned on in the kernel. +*/ + +#include <linux/kernel.h> +#include <linux/init.h> +#include <linux/platform_device.h>
arch/arm/mach-zynq/cpuidle.c line 35
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+#include <linux/cpuidle.h> +#include <linux/io.h> +#include <linux/export.h> +#include <linux/clockchips.h> +#include <linux/cpu_pm.h> +#include <linux/clk.h> +#include <linux/err.h> +#include <asm/proc-fns.h> + +#define XILINX_MAX_STATES 1
You defined XILINX_MAX_STATES as 1, it should be 2.
arch/arm/mach-zynq/cpuidle.c line 46
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+static DEFINE_PER_CPU(struct cpuidle_device, xilinx_cpuidle_device); + +/* Actual code that puts the SoC in different idle states */ +static int xilinx_enter_idle (struct cpuidle_device *dev, +struct cpuidle_driver *drv,int index) +{ +struct timeval before, after; +int idle_time; + +local_irq_disable ();
Remove this line, it is done by the caller.
arch/arm/mach-zynq/cpuidle.c line 47
Quoted 1 line
arch/arm/mach-zynq/cpuidle.c new file +1
-0+do_gettimeofday (&before);
Remove this line, that will be taken into account by the driver's en_core_tk_irqen flag.
arch/arm/mach-zynq/cpuidle.c line 51
Quoted 4 lines
arch/arm/mach-zynq/cpuidle.c new file +4
-0+ +if (index ==0 ) +/* Wait for interrupt state */ +cpu_do_idle ();
Don't check the index, see in the driver initialization comment below the defaul wfi state. You will have one state entering WFI and the xilinx_enter_idle function entering with always index 1.
arch/arm/mach-zynq/cpuidle.c line 56
Quoted 4 lines
arch/arm/mach-zynq/cpuidle.c new file +4
-0+else if (index ==1 ) { +unsigned int cpu_id =smp_processor_id (); + +clockevents_notify (CLOCK_EVT_NOTIFY_BROADCAST_ENTER , &cpu_id);
Remove this line, CPUIDLE_FLAG_TIMER_STOP will tell the cpuidle framework to handle that.
arch/arm/mach-zynq/cpuidle.c line 70
Quoted 13 lines
arch/arm/mach-zynq/cpuidle.c new file +13
-0+/* Devices must be stopped here */ +cpu_pm_enter (); + +/* Add code for DDR self refresh start */ + +cpu_do_idle (); +/*cpu_suspend(foo, bar);*/ + +/* Add code for DDR self refresh stop */ + +cpu_pm_exit (); + +clockevents_notify (CLOCK_EVT_NOTIFY_BROADCAST_EXIT , &cpu_id);
Remove this line, CPUIDLE_FLAG_TIMER_STOP will tell the cpuidle framework to handle that.
arch/arm/mach-zynq/cpuidle.c line 78
Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c new file +8
-0+ } + +do_gettimeofday (&after); +local_irq_enable (); + idle_time = (after.tv_sec - before.tv_sec) *USEC_PER_SEC + + (after.tv_usec - before.tv_usec); + + dev->last_residency = idle_time;
Remove these lines about time computation, that is handled by the caller.
Quoted 16 lines
arch/arm/mach-zynq/cpuidle.c new file +16
-0+return index; +} + +static struct cpuidle_driver xilinx_idle_driver = { + .name ="xilinx_idle" , + .owner =THIS_MODULE , + .state_count =XILINX_MAX_STATES , +/* Wait for interrupt state */ + .states[0 ] = { + .enter = xilinx_enter_idle, + .exit_latency =1 , + .target_residency =10000 , + .flags =CPUIDLE_FLAG_TIME_VALID , + .name ="WFI" , + .desc ="Wait for interrupt" , + },
arch/arm/mach-zynq/cpuidle.c line 104
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+/* Wait for interrupt and RAM self refresh state */ + .states[1] = { + .enter = xilinx_enter_idle, + .exit_latency = 10, + .target_residency = 10000, + .flags = CPUIDLE_FLAG_TIME_VALID, + .name = "RAM_SR", + .desc = "WFI and RAM Self Refresh", + }, +};
static struct cpuidle_driver xilinx_idle_driver = {
.name = "xilinx_idle",
.owner = THIS_MODULE,
.states = {
ARM_CPUIDLE_WFI_STATE,
{
.enter = xilinx_enter_idle,
.exit_latency = 10,
.target_residency = 10000,
.flags = CPUIDLE_FLAG_TIME_VALID,
CPUIDLE_FLAG_TIMER_STOP,
.name = "RAM_SR",
.desc = "WFI and RAM Self Refresh",
},
},
.safe_state_index = 0,
.state_count = XILINX_MAX_STATES,
};Quoted 6 lines
arch/arm/mach-zynq/cpuidle.c new file +6
-0+/* Initialize CPU idle by registering the idle states */ +static int xilinx_init_cpuidle (void ) +{ +unsigned int cpu; +struct cpuidle_device *device; +int ret;
arch/arm/mach-zynq/cpuidle.c line 121
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+ + ret =cpuidle_register_driver (&xilinx_idle_driver); +if (ret) { +pr_err ("Registering Xilinx CpuIdle Driver failed.\n" ); +return ret; + } + +for_each_possible_cpu (cpu) { + device = &per_cpu (xilinx_cpuidle_device, cpu); + device->state_count =XILINX_MAX_STATES ;
This is initialization is done from the cpuidle_register_device. As drv->state_count == device->state_count.
Quoted 13 lines
arch/arm/mach-zynq/cpuidle.c new file +12
-0+ device->cpu = cpu; + ret =cpuidle_register_device (device); +if (ret) { +pr_err ("xilinx_init_cpuidle: Failed registering\n" ); +return ret; + } + } + +pr_info ("Xilinx CpuIdle Driver started\n" ); +return 0 ; +} +device_initcall (xilinx_init_cpuidle);
Signature · 5 lines
<http://www.linaro.org/> Linaro.org ? Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
PATCH v2 10/10 arm: zynq: Add cpuidle support
Hi Daniel,
2013/3/27 Daniel Lezcano <[email protected]>:
Quoted 5 lines · Michal Simek
Quoted 3 lines
Add support for cpuidle.
Signed-off-by Michal Simek <[email protected]>
Quoted 7 lines
Hi Michal,
at the first glance, the driver won't enter in the self refresh state.
If I am right, by checking:
/sys/devices/system/cpu/cpu0/cpuidle/state1/usage, it should be always zero.
I have found this some mins ago.
I means that currently it is not zero.
zynq> cat /sys/devices/system/cpu/cpu0/cpuidle/state1/usage 96931
zynq> cat /sys/devices/system/cpu/cpu0/cpuidle/state0/usage 8396
Quoted 4 lines
Also, I see there is self refresh but you save the cpu context, so it is cpu shutdown no ?
In this case, 3 states would be needed, WFI, self-refresh, powerdown, no ?
Let me talk to Soren how this should be done on zynq.
Quoted 3 lines
The comments below applies for this driver based on linux-pm-next.
https://git.kernel.org/cgit/linux/kernel/git/rafael/linux-pm.git/
Appreciate that - the origin author of this code is not longer with Xilinx that's why hard to tell why it was done in this way. This version was done 2011-02. But that's not excuse just explanation.
It mean that this driver should also go through this tree, right?
Quoted 54 lines
Quoted 54 lines
---
v2: Fix file header--- arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.carch/arm/mach-zynq/Makefile +1
-0@@ -7,4 +7,5 @@ obj-y := common.o slcr.o CFLAGS_REMOVE_hotplug.o =-march=armv6k CFLAGS_hotplug.o =-Wa,-march=armv7-a -mcpu=cortex-a9 obj-$(CONFIG_HOTPLUG_CPU) += hotplug.o +obj-$(CONFIG_CPU_IDLE) += cpuidle.o obj-$(CONFIG_SMP) += headsmp.o platsmp.oarch/arm/mach-zynq/cpuidle.c new file +29
-0@@ -0,0 +1,133 @@ +/* +* Copyright (C) 2012-2013 Xilinx +* +* CPU idle support for Xilinx +* +* based on arch/arm/mach-at91/cpuidle.c +* +* This file is licensed under the terms of the GNU General Public +* License version 2. This program is licensed "as is" without any +* warranty of any kind, whether express or implied. +* +* The cpu idle uses wait-for-interrupt and RAM self refresh in order +* to implement two idle states - +* #1 wait-for-interrupt +* #2 wait-for-interrupt and RAM self refresh +* +* Note that this code is only intended as a prototype and is not tested +* well yet, or tuned for the Xilinx Cortex A9. Also note that for a +* tickless kernel, high res timers must not be turned on. The cpuidle +* framework must also be turned on in the kernel. +*/ + +#include <linux/kernel.h> +#include <linux/init.h> +#include <linux/platform_device.h> +#include <linux/cpuidle.h> +#include <linux/io.h> +#include <linux/export.h> +#include <linux/clockchips.h>
Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 35
Quoted 6 lines
arch/arm/mach-zynq/cpuidle.c new file +6
-0+#include <linux/cpu_pm.h> +#include <linux/clk.h> +#include <linux/err.h> +#include <asm/proc-fns.h> + +#define XILINX_MAX_STATES 1You defined
XILINX_MAX_STATESas 1, it should be 2.
yep. Done.
Quoted 13 lines
arch/arm/mach-zynq/cpuidle.c line 46
Quoted 11 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+static DEFINE_PER_CPU(struct cpuidle_device, xilinx_cpuidle_device); + +/* Actual code that puts the SoC in different idle states */ +static int xilinx_enter_idle (struct cpuidle_device *dev, +struct cpuidle_driver *drv,int index) +{ +struct timeval before, after; +int idle_time; + +local_irq_disable ();Remove this line, it is done by the caller.
Where is that caller function? I can trace it but this maybe faster way. :-)
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c line 47
Quoted 2 lines
arch/arm/mach-zynq/cpuidle.c new file +1
-0+do_gettimeofday (&before);Remove this line, that will be taken into account by the driver's
en_core_tk_irqenflag.
ok
Quoted 9 lines
arch/arm/mach-zynq/cpuidle.c line 51
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +4
-0+ +if (index ==0 ) +/* Wait for interrupt state */ +cpu_do_idle ();Don't check the index, see in the driver initialization comment below the defaul wfi state. You will have one state entering WFI and the
xilinx_enter_idlefunction entering with always index 1.
ok
Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 56
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +4
-0+else if (index ==1 ) { +unsigned int cpu_id =smp_processor_id (); + +clockevents_notify (CLOCK_EVT_NOTIFY_BROADCAST_ENTER , &cpu_id);Remove this line,
CPUIDLE_FLAG_TIMER_STOPwill tell the cpuidle framework to handle that.
ok
Quoted 17 lines
Quoted 4 lines
arch/arm/mach-zynq/cpuidle.c new file +3
-0+/* Devices must be stopped here */ +cpu_pm_enter (); +arch/arm/mach-zynq/cpuidle.c line 70
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+/* Add code for DDR self refresh start */ + +cpu_do_idle (); +/*cpu_suspend(foo, bar);*/ + +/* Add code for DDR self refresh stop */ + +cpu_pm_exit (); + +clockevents_notify (CLOCK_EVT_NOTIFY_BROADCAST_EXIT , &cpu_id);Remove this line,
CPUIDLE_FLAG_TIMER_STOPwill tell the cpuidle framework to handle that.
ok
Quoted 11 lines
arch/arm/mach-zynq/cpuidle.c line 78
Quoted 9 lines
arch/arm/mach-zynq/cpuidle.c new file +8
-0+ } + +do_gettimeofday (&after); +local_irq_enable (); + idle_time = (after.tv_sec - before.tv_sec) *USEC_PER_SEC + + (after.tv_usec - before.tv_usec); + + dev->last_residency = idle_time;Remove these lines about time computation, that is handled by the caller.
ok
Quoted 38 lines
Quoted 17 lines
arch/arm/mach-zynq/cpuidle.c new file +16
-0+return index; +} + +static struct cpuidle_driver xilinx_idle_driver = { + .name ="xilinx_idle" , + .owner =THIS_MODULE , + .state_count =XILINX_MAX_STATES , +/* Wait for interrupt state */ + .states[0 ] = { + .enter = xilinx_enter_idle, + .exit_latency =1 , + .target_residency =10000 , + .flags =CPUIDLE_FLAG_TIME_VALID , + .name ="WFI" , + .desc ="Wait for interrupt" , + },arch/arm/mach-zynq/cpuidle.c line 104
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+/* Wait for interrupt and RAM self refresh state */ + .states[1] = { + .enter = xilinx_enter_idle, + .exit_latency = 10, + .target_residency = 10000, + .flags = CPUIDLE_FLAG_TIME_VALID, + .name = "RAM_SR", + .desc = "WFI and RAM Self Refresh", + }, +};static struct cpuidle_driver xilinx_idle_driver = { .name = "xilinx_idle", .owner = THIS_MODULE, .states = { ARM_CPUIDLE_WFI_STATE, { .enter = xilinx_enter_idle, .exit_latency = 10, .target_residency = 10000, .flags = CPUIDLE_FLAG_TIME_VALID,
Quoted 9 lines
CPUIDLE_FLAG_TIMER_STOP,
.name = "RAM_SR",
.desc = "WFI and RAM Self Refresh",
},
},
.safe_state_index = 0,
.state_count = XILINX_MAX_STATES,
};ok
Quoted 12 lines
Quoted 12 lines
arch/arm/mach-zynq/cpuidle.c new file +11
-0+/* Initialize CPU idle by registering the idle states */ +static int xilinx_init_cpuidle (void ) +{ +unsigned int cpu; +struct cpuidle_device *device; +int ret; + + ret =cpuidle_register_driver (&xilinx_idle_driver); +if (ret) { +pr_err ("Registering Xilinx CpuIdle Driver failed.\n" ); +return ret;
Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 121
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +5
-0+ } + +for_each_possible_cpu (cpu) { + device = &per_cpu (xilinx_cpuidle_device, cpu); + device->state_count =XILINX_MAX_STATES ;This is initialization is done from the
cpuidle_register_device. Asdrv->state_count==device->state_count.
ok. Thanks you very much for your comments.
Michal
Signature · 5 lines
Michal Simek, Ing. (M.Eng) w: www.monstr.eu p: +42-0-721842854 Maintainer of Linux kernel - Microblaze cpu - http://www.monstr.eu/fdt/ Maintainer of Linux kernel - Xilinx Zynq ARM architecture Microblaze U-BOOT custodian and responsible for u-boot arm zynq platform
PATCH v2 10/10 arm: zynq: Add cpuidle support
Quoted 123 lines · Michal Simek
Hi Daniel,
2013/3/27 Daniel Lezcano <[email protected]>:
Quoted 5 lines · Michal Simek
Quoted 3 lines
Add support for cpuidle.
Signed-off-by Michal Simek <[email protected]>Quoted 7 lines
Hi Michal,
at the first glance, the driver won't enter in the self refresh state.
If I am right, by checking:
/sys/devices/system/cpu/cpu0/cpuidle/state1/usage, it should be always zero.I have found this some mins ago.
I means that currently it is not zero.
zynq> cat
/sys/devices/system/cpu/cpu0/cpuidle/state1/usage96931
zynq> cat/sys/devices/system/cpu/cpu0/cpuidle/state0/usage8396Quoted 4 lines
Also, I see there is self refresh but you save the cpu context, so it is cpu shutdown no ?
In this case, 3 states would be needed, WFI, self-refresh, powerdown, no ?
Let me talk to Soren how this should be done on zynq.
Quoted 3 lines
The comments below applies for this driver based on linux-pm-next.
https://git.kernel.org/cgit/linux/kernel/git/rafael/linux-pm.git/
Appreciate that - the origin author of this code is not longer with Xilinx that's why hard to tell why it was done in this way. This version was done 2011-02. But that's not excuse just explanation.
It mean that this driver should also go through this tree, right?
Quoted 54 lines
Quoted 54 lines
---
v2: Fix file header--- arch/arm/mach-zynq/Makefile | 1 + arch/arm/mach-zynq/cpuidle.c | 133 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 134 insertions(+) create mode 100644 arch/arm/mach-zynq/cpuidle.carch/arm/mach-zynq/Makefile +1
-0@@ -7,4 +7,5 @@ obj-y := common.o slcr.o CFLAGS_REMOVE_hotplug.o =-march=armv6k CFLAGS_hotplug.o =-Wa,-march=armv7-a -mcpu=cortex-a9 obj-$(CONFIG_HOTPLUG_CPU) += hotplug.o +obj-$(CONFIG_CPU_IDLE) += cpuidle.o obj-$(CONFIG_SMP) += headsmp.o platsmp.oarch/arm/mach-zynq/cpuidle.c new file +29
-0@@ -0,0 +1,133 @@ +/* +* Copyright (C) 2012-2013 Xilinx +* +* CPU idle support for Xilinx +* +* based on arch/arm/mach-at91/cpuidle.c +* +* This file is licensed under the terms of the GNU General Public +* License version 2. This program is licensed "as is" without any +* warranty of any kind, whether express or implied. +* +* The cpu idle uses wait-for-interrupt and RAM self refresh in order +* to implement two idle states - +* #1 wait-for-interrupt +* #2 wait-for-interrupt and RAM self refresh +* +* Note that this code is only intended as a prototype and is not tested +* well yet, or tuned for the Xilinx Cortex A9. Also note that for a +* tickless kernel, high res timers must not be turned on. The cpuidle +* framework must also be turned on in the kernel. +*/ + +#include <linux/kernel.h> +#include <linux/init.h> +#include <linux/platform_device.h> +#include <linux/cpuidle.h> +#include <linux/io.h> +#include <linux/export.h> +#include <linux/clockchips.h> Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 35
Quoted 6 lines
arch/arm/mach-zynq/cpuidle.c new file +6
-0+#include <linux/cpu_pm.h> +#include <linux/clk.h> +#include <linux/err.h> +#include <asm/proc-fns.h> + +#define XILINX_MAX_STATES 1You defined
XILINX_MAX_STATESas 1, it should be 2.yep. Done.
Quoted 13 lines
arch/arm/mach-zynq/cpuidle.c line 46
Quoted 11 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+static DEFINE_PER_CPU(struct cpuidle_device, xilinx_cpuidle_device); + +/* Actual code that puts the SoC in different idle states */ +static int xilinx_enter_idle (struct cpuidle_device *dev, +struct cpuidle_driver *drv,int index) +{ +struct timeval before, after; +int idle_time; + +local_irq_disable ();Remove this line, it is done by the caller.
Quoted 1 line
Where is that caller function? I can trace it but this maybe faster way. :-)
In the idle loop arch/arm/kernel/process.c
Quoted 138 lines
Quoted 4 lines
arch/arm/mach-zynq/cpuidle.c line 47
Quoted 1 line
arch/arm/mach-zynq/cpuidle.c new file +1
-0+do_gettimeofday (&before);Remove this line, that will be taken into account by the driver's
en_core_tk_irqenflag.ok
Quoted 9 lines
arch/arm/mach-zynq/cpuidle.c line 51
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +4
-0+ +if (index ==0 ) +/* Wait for interrupt state */ +cpu_do_idle ();Don't check the index, see in the driver initialization comment below the defaul wfi state. You will have one state entering WFI and the
xilinx_enter_idlefunction entering with always index 1.ok
Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 56
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +4
-0+else if (index ==1 ) { +unsigned int cpu_id =smp_processor_id (); + +clockevents_notify (CLOCK_EVT_NOTIFY_BROADCAST_ENTER , &cpu_id);Remove this line,
CPUIDLE_FLAG_TIMER_STOPwill tell the cpuidle framework to handle that.ok
Quoted 17 lines
Quoted 4 lines
arch/arm/mach-zynq/cpuidle.c new file +3
-0+/* Devices must be stopped here */ +cpu_pm_enter (); +arch/arm/mach-zynq/cpuidle.c line 70
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+/* Add code for DDR self refresh start */ + +cpu_do_idle (); +/*cpu_suspend(foo, bar);*/ + +/* Add code for DDR self refresh stop */ + +cpu_pm_exit (); + +clockevents_notify (CLOCK_EVT_NOTIFY_BROADCAST_EXIT , &cpu_id);Remove this line,
CPUIDLE_FLAG_TIMER_STOPwill tell the cpuidle framework to handle that.ok
Quoted 11 lines
arch/arm/mach-zynq/cpuidle.c line 78
Quoted 9 lines
arch/arm/mach-zynq/cpuidle.c new file +8
-0+ } + +do_gettimeofday (&after); +local_irq_enable (); + idle_time = (after.tv_sec - before.tv_sec) *USEC_PER_SEC + + (after.tv_usec - before.tv_usec); + + dev->last_residency = idle_time;Remove these lines about time computation, that is handled by the caller.
ok
Quoted 38 lines
Quoted 17 lines
arch/arm/mach-zynq/cpuidle.c new file +16
-0+return index; +} + +static struct cpuidle_driver xilinx_idle_driver = { + .name ="xilinx_idle" , + .owner =THIS_MODULE , + .state_count =XILINX_MAX_STATES , +/* Wait for interrupt state */ + .states[0 ] = { + .enter = xilinx_enter_idle, + .exit_latency =1 , + .target_residency =10000 , + .flags =CPUIDLE_FLAG_TIME_VALID , + .name ="WFI" , + .desc ="Wait for interrupt" , + },arch/arm/mach-zynq/cpuidle.c line 104
Quoted 10 lines
arch/arm/mach-zynq/cpuidle.c new file +10
-0+/* Wait for interrupt and RAM self refresh state */ + .states[1] = { + .enter = xilinx_enter_idle, + .exit_latency = 10, + .target_residency = 10000, + .flags = CPUIDLE_FLAG_TIME_VALID, + .name = "RAM_SR", + .desc = "WFI and RAM Self Refresh", + }, +};static struct cpuidle_driver xilinx_idle_driver = { .name = "xilinx_idle", .owner = THIS_MODULE, .states = { ARM_CPUIDLE_WFI_STATE, { .enter = xilinx_enter_idle, .exit_latency = 10, .target_residency = 10000, .flags = CPUIDLE_FLAG_TIME_VALID,Quoted 9 lines
CPUIDLE_FLAG_TIMER_STOP, .name = "RAM_SR", .desc = "WFI and RAM Self Refresh", }, }, .safe_state_index = 0, .state_count = XILINX_MAX_STATES, };ok
Quoted 12 lines
Quoted 12 lines
arch/arm/mach-zynq/cpuidle.c new file +11
-0+/* Initialize CPU idle by registering the idle states */ +static int xilinx_init_cpuidle (void ) +{ +unsigned int cpu; +struct cpuidle_device *device; +int ret; + + ret =cpuidle_register_driver (&xilinx_idle_driver); +if (ret) { +pr_err ("Registering Xilinx CpuIdle Driver failed.\n" ); +return ret;Quoted 8 lines
arch/arm/mach-zynq/cpuidle.c line 121
Quoted 5 lines
arch/arm/mach-zynq/cpuidle.c new file +5
-0+ } + +for_each_possible_cpu (cpu) { + device = &per_cpu (xilinx_cpuidle_device, cpu); + device->state_count =XILINX_MAX_STATES ;This is initialization is done from the
cpuidle_register_device. Asdrv->state_count==device->state_count.ok. Thanks you very much for your comments.
Michal
Signature · 5 lines
<http://www.linaro.org/> Linaro.org ? Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog