
Hi,
On 29 March 2018 at 18:52, Andy Shevchenko andriy.shevchenko@linux.intel.com wrote:
On Wed, 2018-03-28 at 17:58 -0700, Ivan Gorinov wrote:
Adding HPET as an alternative timer for x86 (default is TSC). HPET main counter has constant clock frequency, calibration is not required. This change also makes TSC timer driver optional on x86 platforms. If X86_TSC is disabled, early timer functions are provided by HPET.
HPET can be selected as the tick timer in the Device Tree "chosen" node:
/include/ "hpet.dtsi"
chosen { tick-timer = "/hpet0"; };
First question is how this will work in case of Broadwell and Ivybridge that have something to do with HPET in their CPU code, i.e.
arch/x86/cpu/broadwell/pch.c arch/x86/cpu/ivybridge/lpc.c
?
Look for
clrsetbits_le32(RCB_REG(HPTC), 3, 1 << 7);
+static int hpet_timer_get_count(struct udevice *dev, u64 *count)
+static int hpet_timer_probe(struct udevice *dev)
+#ifndef CONFIG_X86_TSC_TIMER
+#define EARLY_HPET_BASE 0xfed00000
HPET address is fixed, AFAIU, on x86, and provided by config option in U-Boot.
+unsigned long notrace timer_early_get_rate(void)
+u64 notrace timer_early_get_count(void)
+int timer_init(void)
These functions have too much code duplication with above.
Perhaps that code can be removed in favour of this driver? I am happy to test on these platforms if it helps.
Regards, Simon