| Message ID | 1459324077-7116-1-git-send-email-jszhang@marvell.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 03/30/2016 09:47 AM, Jisheng Zhang wrote: > Currently, we check cpuidle_ops.suspend every time when entering a > low-power idle state. But this check could be avoided in this hot path > by moving it into arm_cpuidle_read_ops() to reduce arm_cpuidle_suspend > overhead a bit. This patch makes sense but the ARM64 part is missing.
Dear Daniel, On Wed, 30 Mar 2016 10:02:02 +0200 Daniel Lezcano wrote: > On 03/30/2016 09:47 AM, Jisheng Zhang wrote: > > Currently, we check cpuidle_ops.suspend every time when entering a > > low-power idle state. But this check could be avoided in this hot path > > by moving it into arm_cpuidle_read_ops() to reduce arm_cpuidle_suspend > > overhead a bit. > > This patch makes sense but the ARM64 part is missing. > Here is the ARM64 part, you are also in the "To" list ;) http://lists.infradead.org/pipermail/linux-arm-kernel/2016-March/417713.html Thanks, Jisheng
On 03/30/2016 10:05 AM, Jisheng Zhang wrote: > Dear Daniel, > > On Wed, 30 Mar 2016 10:02:02 +0200 Daniel Lezcano wrote: > >> On 03/30/2016 09:47 AM, Jisheng Zhang wrote: >>> Currently, we check cpuidle_ops.suspend every time when entering a >>> low-power idle state. But this check could be avoided in this hot path >>> by moving it into arm_cpuidle_read_ops() to reduce arm_cpuidle_suspend >>> overhead a bit. >> >> This patch makes sense but the ARM64 part is missing. >> > > Here is the ARM64 part, you are also in the "To" list ;) > > http://lists.infradead.org/pipermail/linux-arm-kernel/2016-March/417713.html Ah, yes. That happens when we read the email sequentially :)
diff --git a/arch/arm/kernel/cpuidle.c b/arch/arm/kernel/cpuidle.c index 703926e..e63388e 100644 --- a/arch/arm/kernel/cpuidle.c +++ b/arch/arm/kernel/cpuidle.c @@ -47,18 +47,13 @@ int arm_cpuidle_simple_enter(struct cpuidle_device *dev, * This function calls the underlying arch specific low level PM code as * registered at the init time. * - * Returns -EOPNOTSUPP if no suspend callback is defined, the result of the - * callback otherwise. + * Returns the result of the suspend callback. */ int arm_cpuidle_suspend(int index) { - int ret = -EOPNOTSUPP; int cpu = smp_processor_id(); - if (cpuidle_ops[cpu].suspend) - ret = cpuidle_ops[cpu].suspend(index); - - return ret; + return cpuidle_ops[cpu].suspend(index); } /** @@ -92,7 +87,8 @@ static struct cpuidle_ops *__init arm_cpuidle_get_ops(const char *method) * process. * * Return 0 on sucess, -ENOENT if no 'enable-method' is defined, -EOPNOTSUPP if - * no cpuidle_ops is registered for the 'enable-method'. + * no cpuidle_ops is registered for the 'enable-method', or if no suspend + * callback is defined. */ static int __init arm_cpuidle_read_ops(struct device_node *dn, int cpu) { @@ -110,6 +106,12 @@ static int __init arm_cpuidle_read_ops(struct device_node *dn, int cpu) return -EOPNOTSUPP; } + if (!ops->suspend) { + pr_warn("cpuidle_ops '%s': no suspend callback\n", + enable_method); + return -EOPNOTSUPP; + } + cpuidle_ops[cpu] = *ops; /* structure copy */ pr_notice("cpuidle: enable-method property '%s'" @@ -129,7 +131,8 @@ static int __init arm_cpuidle_read_ops(struct device_node *dn, int cpu) * Returns: * 0 on success, * -ENODEV if it fails to find the cpu node in the device tree, - * -EOPNOTSUPP if it does not find a registered cpuidle_ops for this cpu, + * -EOPNOTSUPP if it does not find a registered and valid cpuidle_ops for + * this cpu, * -ENOENT if it fails to find an 'enable-method' property, * -ENXIO if the HW reports a failure or a misconfiguration, * -ENOMEM if the HW report an memory allocation failure
Currently, we check cpuidle_ops.suspend every time when entering a low-power idle state. But this check could be avoided in this hot path by moving it into arm_cpuidle_read_ops() to reduce arm_cpuidle_suspend overhead a bit. Signed-off-by: Jisheng Zhang <jszhang@marvell.com> --- Since v1: - drop patch1, I.E "ARM: cpuidle: fix !cpuidle_ops[cpu].init case during init", because there may be platforms which doesn't need the init member at all. - adopt Daniel's suggestion to move the cpuidle_ops.suspend check to arm_cpuidle_read_ops(). arch/arm/kernel/cpuidle.c | 21 ++++++++++++--------- 1 file changed, 12 insertions(+), 9 deletions(-)