Re: [Linaro-acpi] [PATCH v19 05/15] clocksource/drivers/arm_arch_timer: rework PPI determination

Hi Mark,

On 17 January 2017 at 01:29, Mark Rutland mark.rutland@arm.com wrote:

On Wed, Dec 21, 2016 at 02:45:53PM +0800, fu.wei@linaro.org wrote: [...]

...

• if (is_hyp_mode_available() || !arch_timer_ppi[ARCH_TIMER_VIRT_PPI]) {
  

•         bool has_ppi;
  

• if (is_hyp_mode_available() && is_kernel_in_hyp_mode())
  

•         return ARCH_TIMER_HYP_PPI;
  
  

•         if (is_kernel_in_hyp_mode()) {
  

•                 arch_timer_uses_ppi = ARCH_TIMER_HYP_PPI;
  

•                 has_ppi = !!arch_timer_ppi[ARCH_TIMER_HYP_PPI];
  

•         } else {
  

•                 arch_timer_uses_ppi = ARCH_TIMER_PHYS_SECURE_PPI;
  

•                 has_ppi = (!!arch_timer_ppi[ARCH_TIMER_PHYS_SECURE_PPI] ||
  

•                            !!arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI]);
  

•         }
  

• if (arch_timer_ppi[ARCH_TIMER_VIRT_PPI])
  

•         return ARCH_TIMER_VIRT_PPI;
  
  

•         if (!has_ppi) {
  

•                 pr_warn("No interrupt available, giving up\n");
  

•                 return -EINVAL;
  

•         }
  

• }
  

• if (IS_ENABLED(CONFIG_ARM64))
  

•         return ARCH_TIMER_PHYS_NONSECURE_PPI;
  

• return ARCH_TIMER_PHYS_SECURE_PPI;
  

For a 32-bit platform booted at hyp (with a virt PPI available), the new logic will select ARCH_TIMER_VIRT_PPI. I beleive that will break KVM.

I think the logic should be:

    if (is_kernel_in_hyp_mode())
            return ARCH_TIMER_HYP_PPI;

    if (!is_hyp_mode_available() &&
        arch_timer_ppi[ARCH_TIMER_VIRT_PPI])
            return ARCH_TIMER_VIRT_PPI;

    if (IS_ENABLED(CONFIG_ARM64))
            return ARCH_TIMER_PHYS_NONSECURE_PPI;

    return ARCH_TIMER_PHYS_SECURE_PPI;

Please use that instead (keeping the comment you retained).

Great thanks for pointing it out, that is bug. also got this bug report from Huawei engineer.

I have fixed it using your example code, thanks!

...

+static int __init arch_timer_init(void) +{

• int ret;
  
  ret = arch_timer_register();
  if (ret)
  

@@ -904,6 +906,13 @@ static int __init arch_timer_of_init(struct device_node *np) if (IS_ENABLED(CONFIG_ARM) && of_property_read_bool(np, "arm,cpu-registers-not-fw-configured")) arch_timer_uses_ppi = ARCH_TIMER_PHYS_SECURE_PPI;

• else
  

•         arch_timer_uses_ppi = arch_timer_select_ppi();
  

• if (!arch_timer_ppi[arch_timer_uses_ppi]) {
  

•         pr_err("No interrupt available, giving up\n");
  

•         return -EINVAL;
  

• }
  
  /* On some systems, the counter stops ticking when in suspend. */
  arch_counter_suspend_stop = of_property_read_bool(np,
  

@@ -1049,6 +1058,12 @@ static int __init arch_timer_acpi_init(struct acpi_table_header *table) /* Get the frequency from CNTFRQ */ arch_timer_detect_rate(NULL, NULL);

• arch_timer_uses_ppi = arch_timer_select_ppi();
  

• if (!arch_timer_ppi[arch_timer_uses_ppi]) {
  

•         pr_err("No interrupt available, giving up\n");
  

•         return -EINVAL;
  

• }
  

I see that we have to duplicate this so we can special-case the DT-specific behaviour, so that's fine by me.

Yes, that is the reason of the duplication :-)

If you can fix the arch_timer_select_ppi() logic as above, this should be fine.

Done, thanks :-)

Thanks, Mark.