Fixes for for dra7 timer wrap errata i940

Hi all,

Here is v2 set of fixes for dra7 ARM architected timer wrap errata i940 where the timer fails to wrap after 388 days. The workaround is to use two dmtimers as the local timers instead.

Note that these patches depend on timer posted mode fixes series "[PATCH 0/3] Fixes for timer-ti-dm systimer posted mode" for the write status register check fix. Also the spurious timer interrupt fix is good to have from that series.

Regards,

Tony

Changes since v1:
- Drop pointless irqflags and IRQF_NOBALANCING as noted by Daniel

Tony Lindgren (2):
clocksource/drivers/timer-ti-dm: Prepare to handle dra7 timer wrap
issue
clocksource/drivers/timer-ti-dm: Handle dra7 timer wrap errata i940

 arch/arm/boot/dts/dra7-l4.dtsi             |   4 +-
 arch/arm/boot/dts/dra7.dtsi                |  20 +++
 drivers/clocksource/timer-ti-dm-systimer.c | 142 +++++++++++++++++----
 include/linux/cpuhotplug.h                 |   1 +
 4 files changed, 142 insertions(+), 25 deletions(-)

2.31.0

There is a timer wrap issue on dra7 for the ARM architected timer. In a typical clock configuration the timer fails to wrap after 388 days.

To work around the issue, we need to use timer-ti-dm timers instead.

Let's prepare for adding support for percpu timers by adding a common dmtimer_clkevt_init_common() and call it from dmtimer_clockevent_init(). This patch makes no intentional functional changes.

---
 drivers/clocksource/timer-ti-dm-systimer.c | 66 ++++++++++++++--------
 1 file changed, 43 insertions(+), 23 deletions(-)

@@ -530,17 +530,17 @@ static void omap_clockevent_unidle(struct clock_event_device *evt)
 	writel_relaxed(OMAP_TIMER_INT_OVERFLOW, t->base + t->wakeup);
 }
 
-static int __init dmtimer_clockevent_init(struct device_node *np)
+static int __init dmtimer_clkevt_init_common(struct dmtimer_clockevent *clkevt,
+					     struct device_node *np,
+					     unsigned int features,
+					     const struct cpumask *cpumask,
+					     const char *name,
+					     int rating)
 {
-	struct dmtimer_clockevent *clkevt;
 	struct clock_event_device *dev;
 	struct dmtimer_systimer *t;
 	int error;
 
-	clkevt = kzalloc(sizeof(*clkevt), GFP_KERNEL);
-	if (!clkevt)
-		return -ENOMEM;
-
 	t = &clkevt->t;
 	dev = &clkevt->dev;
 

@@ -548,25 +548,23 @@ static int __init dmtimer_clockevent_init(struct device_node *np)
 	 * We mostly use cpuidle_coupled with ARM local timers for runtime,
 	 * so there's probably no use for CLOCK_EVT_FEAT_DYNIRQ here.
 	 */
-	dev->features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT;
-	dev->rating = 300;
+	dev->features = features;
+	dev->rating = rating;
 	dev->set_next_event = dmtimer_set_next_event;
 	dev->set_state_shutdown = dmtimer_clockevent_shutdown;
 	dev->set_state_periodic = dmtimer_set_periodic;
 	dev->set_state_oneshot = dmtimer_clockevent_shutdown;
 	dev->set_state_oneshot_stopped = dmtimer_clockevent_shutdown;
 	dev->tick_resume = dmtimer_clockevent_shutdown;
-	dev->cpumask = cpu_possible_mask;
+	dev->cpumask = cpumask;
 
 	dev->irq = irq_of_parse_and_map(np, 0);
-	if (!dev->irq) {
-		error = -ENXIO;
-		goto err_out_free;
-	}
+	if (!dev->irq)
+		return -ENXIO;
 
 	error = dmtimer_systimer_setup(np, &clkevt->t);
 	if (error)
-		goto err_out_free;
+		return error;
 
 	clkevt->period = 0xffffffff - DIV_ROUND_CLOSEST(t->rate, HZ);
 

@@ -578,32 +576,54 @@ static int __init dmtimer_clockevent_init(struct device_node *np)
 	writel_relaxed(OMAP_TIMER_CTRL_POSTED, t->base + t->ifctrl);
 
 	error = request_irq(dev->irq, dmtimer_clockevent_interrupt,
-			    IRQF_TIMER, "clockevent", clkevt);
+			    IRQF_TIMER, name, clkevt);
 	if (error)
 		goto err_out_unmap;
 
 	writel_relaxed(OMAP_TIMER_INT_OVERFLOW, t->base + t->irq_ena);
 	writel_relaxed(OMAP_TIMER_INT_OVERFLOW, t->base + t->wakeup);
 
-	pr_info("TI gptimer clockevent: %s%lu Hz at %pOF\n",
-		of_find_property(np, "ti,timer-alwon", NULL) ?
+	pr_info("TI gptimer %s: %s%lu Hz at %pOF\n",
+		name, of_find_property(np, "ti,timer-alwon", NULL) ?
 		"always-on " : "", t->rate, np->parent);
 
-	clockevents_config_and_register(dev, t->rate,
+	return 0;
+
+err_out_unmap:
+	iounmap(t->base);
+
+	return error;
+}
+
+static int __init dmtimer_clockevent_init(struct device_node *np)
+{
+	struct dmtimer_clockevent *clkevt;
+	int error;
+
+	clkevt = kzalloc(sizeof(*clkevt), GFP_KERNEL);
+	if (!clkevt)
+		return -ENOMEM;
+
+	error = dmtimer_clkevt_init_common(clkevt, np,
+					   CLOCK_EVT_FEAT_PERIODIC |
+					   CLOCK_EVT_FEAT_ONESHOT,
+					   cpu_possible_mask, "clockevent",
+					   300);
+	if (error)
+		goto err_out_free;
+
+	clockevents_config_and_register(&clkevt->dev, clkevt->t.rate,
 					3, /* Timer internal resynch latency */
 					0xffffffff);
 
 	if (of_machine_is_compatible("ti,am33xx") ||
 	    of_machine_is_compatible("ti,am43")) {
-		dev->suspend = omap_clockevent_idle;
-		dev->resume = omap_clockevent_unidle;
+		clkevt->dev.suspend = omap_clockevent_idle;
+		clkevt->dev.resume = omap_clockevent_unidle;
 	}
 
 	return 0;
 
-err_out_unmap:
-	iounmap(t->base);
-
 err_out_free:
 	kfree(clkevt);
 

2.31.0

There is a timer wrap issue on dra7 for the ARM architected timer. In a typical clock configuration the timer fails to wrap after 388 days.

To work around the issue, we need to use timer-ti-dm percpu timers instead.

Let's configure dmtimer3 and 4 as percpu timers by default, and warn about the issue if the dtb is not configured properly.

Let's do this as a single patch so it can be backported to v5.8 and later kernels easily. Note that this patch depends on earlier timer-ti-dm systimer posted mode fixes, and a preparatory clockevent patch "clocksource/drivers/timer-ti-dm: Prepare to handle dra7 timer wrap issue".

For more information, please see the errata for "AM572x Sitara Processors Silicon Revisions 1.1, 2.0":

https://www.ti.com/lit/er/sprz429m/sprz429m.pdf

The concept is based on earlier reference patches done by Tero Kristo and Keerthy.

---
 arch/arm/boot/dts/dra7-l4.dtsi             |  4 +-
 arch/arm/boot/dts/dra7.dtsi                | 20 ++++++
 drivers/clocksource/timer-ti-dm-systimer.c | 76 ++++++++++++++++++++++
 include/linux/cpuhotplug.h                 |  1 +
 4 files changed, 99 insertions(+), 2 deletions(-)

@@ -1168,7 +1168,7 @@ timer2: timer@0 {
 			};
 		};
 
-		target-module@34000 {			/* 0x48034000, ap 7 46.0 */
+		timer3_target: target-module@34000 {	/* 0x48034000, ap 7 46.0 */
 			compatible = "ti,sysc-omap4-timer", "ti,sysc";
 			reg = <0x34000 0x4>,
 			      <0x34010 0x4>;

@@ -1195,7 +1195,7 @@ timer3: timer@0 {
 			};
 		};
 
-		target-module@36000 {			/* 0x48036000, ap 9 4e.0 */
+		timer4_target: target-module@36000 {	/* 0x48036000, ap 9 4e.0 */
 			compatible = "ti,sysc-omap4-timer", "ti,sysc";
 			reg = <0x36000 0x4>,
 			      <0x36010 0x4>;

@@ -46,6 +46,7 @@ aliases {
 
 	timer {
 		compatible = "arm,armv7-timer";
+		status = "disabled";	/* See ARM architected timer wrap erratum i940 */
 		interrupts = <GIC_PPI 13 (GIC_CPU_MASK_SIMPLE(2) | IRQ_TYPE_LEVEL_LOW)>,
 			     <GIC_PPI 14 (GIC_CPU_MASK_SIMPLE(2) | IRQ_TYPE_LEVEL_LOW)>,
 			     <GIC_PPI 11 (GIC_CPU_MASK_SIMPLE(2) | IRQ_TYPE_LEVEL_LOW)>,

@@ -1241,3 +1242,22 @@ timer@0 {
 		assigned-clock-parents = <&sys_32k_ck>;
 	};
 };
+
+/* Local timers, see ARM architected timer wrap erratum i940 */
+&timer3_target {
+	ti,no-reset-on-init;
+	ti,no-idle;
+	timer@0 {
+		assigned-clocks = <&l4per_clkctrl DRA7_L4PER_TIMER3_CLKCTRL 24>;
+		assigned-clock-parents = <&timer_sys_clk_div>;
+	};
+};
+
+&timer4_target {
+	ti,no-reset-on-init;
+	ti,no-idle;
+	timer@0 {
+		assigned-clocks = <&l4per_clkctrl DRA7_L4PER_TIMER4_CLKCTRL 24>;
+		assigned-clock-parents = <&timer_sys_clk_div>;
+	};
+};

@@ -135,6 +135,7 @@ enum cpuhp_state {
 	CPUHP_AP_RISCV_TIMER_STARTING,
 	CPUHP_AP_CLINT_TIMER_STARTING,
 	CPUHP_AP_CSKY_TIMER_STARTING,
+	CPUHP_AP_TI_GP_TIMER_STARTING,
 	CPUHP_AP_HYPERV_TIMER_STARTING,
 	CPUHP_AP_KVM_STARTING,
 	CPUHP_AP_KVM_ARM_VGIC_INIT_STARTING,

2.31.0

Hi Tony,

This causes a conflict for IPU2 which is using timer 3 and 4.

From arch/arm/boot/dts/dra7-ipu-dsp-common.dtsi:

  &ipu2 {
          mboxes = <&mailbox6 &mbox_ipu2_ipc3x>;
          ti,timers = <&timer3>;
          ti,watchdog-timers = <&timer4>, <&timer9>;
  };

I noticed an error ("could not get timer platform device") when booting mainline on a BeagleBoard X15 (AM578):

omap-rproc 55020000.ipu: assigned reserved memory node ipu2-memory@95800000 remoteproc remoteproc1: 55020000.ipu is available
remoteproc remoteproc1: powering up 55020000.ipu
remoteproc remoteproc1: Booting fw image dra7-ipu2-fw.xem4, size 3747220 omap-rproc 55020000.ipu: could not get timer platform device omap-rproc 55020000.ipu: omap_rproc_enable_timers failed: -19 remoteproc remoteproc1: can't start rproc 55020000.ipu: -19

I switched this errata fix to use timer 15 and 16 instead which resolves the error. Do you think that is an acceptable solution?

If so, I will post the patch to the list.

Thanks,
Drew

Hi,

OK, sorry I missed that part.

I think the only difference is that timers 15 and 16 are in l4_per3 instead of l4_per1. I doubt that matters as they are pretty much always clocked in this case. If you want to check you can run cyclictest :)

OK thanks,

Tony

I ran this with existing errata fix with dmtimer 3 and 4:

root@am57xx-evm:~# cyclictest --mlockall --smp --priority=80 --interval=200 --distance=0
# /dev/cpu_dma_latency set to 0us
policy: fifo: loadavg: 0.02 0.03 0.05 

T: 0 ( 1449) P:80 I:200 C: 800368 Min: 0 Act: 32 Avg: 22 Max: 128
T: 1 ( 1450) P:80 I:200 C: 800301 Min: 0 Act: 12 Avg: 23 Max: 70

And the results after my switch to dmtimer 15 and 16:

root@am57xx-evm:~# cyclictest --mlockall --smp --priority=80 --interval=200 --distance=0
# /dev/cpu_dma_latency set to 0us
policy: fifo: loadavg: 0.36 0.19 0.07

T: 0 ( 1711) P:80 I:200 C: 759599 Min: 0 Act: 6 Avg: 22 Max: 108
T: 1 ( 1712) P:80 I:200 C: 759539 Min: 0 Act: 19 Avg: 23 Max: 79

This doesn't appear to show any latency regression.

Any other options for cyclictest that that you'd recommend trying?

Thank you,
Drew