Hi Phil,
-----Original Message----- From: Phil Edworthy [mailto:phil.edworthy@renesas.com] Sent: Tuesday, February 07, 2017 6:34 AM To: Tom Rini trini@konsulko.com Cc: Kamil Lulko kamil.lulko@gmail.com; Vikas MANOCHA vikas.manocha@st.com; Michael Kurz michi.kurz@gmail.com; Albert Aribaud albert.u.boot@aribaud.net; U-Boot Mailing List u-boot@lists.denx.de Subject: RE: [U-Boot] [PATCH] armv7m: Add SysTick timer driver
Hi Tom,
On 07 February 2017 14:23, Tom Rini wrote:
On Tue, Feb 07, 2017 at 02:19:39PM +0000, Phil Edworthy wrote:
Hi Kamil,
On 07 February 2017 14:12, Kamil Lulko wrote:
On Mon, Feb 6, 2017 at 12:16 AM, Tom Rini trini@konsulko.com wrote:
On Fri, Feb 03, 2017 at 02:48:40PM +0000, Phil Edworthy wrote:
The SysTick is a 24-bit down counter that is found on all ARM Cortex M3, M4, M7 devices and is always located at a fixed address.
Signed-off-by: Phil Edworthy phil.edworthy@renesas.com
arch/arm/cpu/armv7m/Makefile | 2 + arch/arm/cpu/armv7m/systick-timer.c | 91
+++++++++++++++++++++++++++++++++++++
2 files changed, 93 insertions(+) create mode 100644 arch/arm/cpu/armv7m/systick-timer.c
diff --git a/arch/arm/cpu/armv7m/Makefile
b/arch/arm/cpu/armv7m/Makefile
index aff60e8..e1a6c40 100644 --- a/arch/arm/cpu/armv7m/Makefile +++ b/arch/arm/cpu/armv7m/Makefile @@ -7,3 +7,5 @@
extra-y := start.o obj-y += cpu.o
+obj-$(CONFIG_SYS_ARCH_TIMER) += systick-timer.o diff --git a/arch/arm/cpu/armv7m/systick-timer.c
b/arch/arm/cpu/armv7m/systick-timer.c
new file mode 100644 index 0000000..6ccc2fb --- /dev/null +++ b/arch/arm/cpu/armv7m/systick-timer.c @@ -0,0 +1,91 @@ +/*
- ARM Cortex M3/M4/M7 SysTick timer driver
- (C) Copyright 2017 Renesas Electronics Europe Ltd
- Based on arch/arm/mach-stm32/stm32f1/timer.c
- (C) Copyright 2015
- Kamil Lulko, kamil.lulko@gmail.com
- Copyright 2015 ATS Advanced Telematics Systems GmbH
- Copyright 2015 Konsulko Group, Matt Porter
- SPDX-License-Identifier: GPL-2.0+
- */
+#include <common.h> +#include <asm/io.h>
+DECLARE_GLOBAL_DATA_PTR;
+/* SysTick Base Address - fixed for all Cortex M3, M4 and M7 devices */ +#define SYSTICK_BASE 0xE000E010
+struct cm3_systick {
uint32_t ctrl;uint32_t reload;uint32_t val;uint32_t calibration;+};
+#define TIMER_LOAD_VAL 0x00FFFFFF +#define SYSTICK_CTRL_EN BIT(0) +/* Clock source: 0 = Ref clock, 1 = CPU clock */ #define +SYSTICK_CTRL_CPU_CLK BIT(2)
+/* read the 24-bit timer */ +static ulong read_timer(void) {
struct cm3_systick *systick = (struct cm3_systick+*)SYSTICK_BASE;
/* The timer counts down, therefore convert to an- incrementing timer
*/
return TIMER_LOAD_VAL - readl(&systick->val); }+int timer_init(void) +{
struct cm3_systick *systick = (struct cm3_systick+*)SYSTICK_BASE;
writel(TIMER_LOAD_VAL, &systick->reload);writel(TIMER_LOAD_VAL, &systick->val);+#ifdef CONFIG_ARMCORTEXM3_SYSTICK_CPU
/* Use CPU clock, no interrupts */writel(SYSTICK_CTRL_EN | SYSTICK_CTRL_CPU_CLK,+&systick->ctrl); #else
/* Use external clock, no interrupts */writel(SYSTICK_CTRL_EN, &systick->ctrl); #endifgd->arch.tbl = 0;gd->arch.tbu = 0;gd->arch.lastinc = read_timer();return 0;+}
+/* return milli-seconds timer value */ ulong get_timer(ulong +base) {
unsigned long long t = get_ticks() * 1000;return (ulong)((t / CONFIG_SYS_HZ_CLOCK)) - base; }+unsigned long long get_ticks(void) {
u32 now = read_timer();if (now >= gd->arch.lastinc)gd->arch.tbl += (now - gd->arch.lastinc);elsegd->arch.tbl += (TIMER_LOAD_VAL -- gd->arch.lastinc) + now;
gd->arch.lastinc = now;return gd->arch.tbl;+}
+ulong get_tbclk(void) +{
return CONFIG_SYS_HZ_CLOCK; }And (cc'ing maintainers) we could use this in place of arch/arm/mach-stm32/*/timer.c I assume, yes? Thanks!
-- Tom
Yes, however as far as I remember the systick is only 24-bit and usually runs on CPU clock so it will roll over too fast to be useful without using interrupts. Maybe I have missed something?
Yes, it is only a 24-bit counter. On the device I am using, the clock is not the cpu clock, but a low speed clock at 6.25MHz. Even at that low clock it will still wrap every ~2.6 seconds.
So are you able to use the timer then, given the quick overflow?
Yes, I didn't see any problems. All of the udelay/mdelay calls should work because they resolve to constantly calling get_ticks.
I guess it will not work if you attempt to time something that is longer than 2.6 seconds though.
Hmm, I wish it had some way to increase overflow time...it doesn't have the pre-scalar to divide the clock..
As this timer is same for all cortex-Mx, It makes sense to use this timer for stm32 and other cortexMx. I didn't check calibration register : could it be useful here ?
Cheers, Vikas
Thanks Phil