| Message ID | 2717750.dCEzHT3DVQ@kreacher (mailing list archive) |
|---|---|
| State | Mainlined, archived |
| Headers | show |
| Series | cpuidle: Consolidate disabled state checks | expand |
On Mon, Nov 04, 2019 at 12:16:17PM +0100, Rafael J. Wysocki wrote: > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > There are two reasons why CPU idle states may be disabled: either > because the driver has disabled them or because they have been > disabled by user space via sysfs. > > In the former case, the state's "disabled" flag is set once during > the initialization of the driver and it is never cleared later (it > is read-only effectively). In the latter case, the "disable" field > of the given state's cpuidle_state_usage struct is set and it may be > changed via sysfs. Thus checking whether or not an idle state has > been disabled involves reading these two flags every time. > > In order to avoid the additional check of the state's "disabled" flag > (which is effectively read-only anyway), use the value of it at the > init time to set a (new) flag in the "disable" field of that state's > cpuidle_state_usage structure and use the sysfs interface to > manipulate another (new) flag in it. This way the state is disabled > whenever the "disable" field of its cpuidle_state_usage structure is > nonzero, whatever the reason, and it is the only place to look into > to check whether or not the state has been disabled. > > Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org> Much thanks, that always bugged me. Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
On Mon, Nov 4, 2019 at 6:16 AM Rafael J. Wysocki <rjw@rjwysocki.net> wrote: > > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > There are two reasons why CPU idle states may be disabled: either > because the driver has disabled them or because they have been > disabled by user space via sysfs. > > In the former case, the state's "disabled" flag is set once during > the initialization of the driver and it is never cleared later (it > is read-only effectively). for x86 (intel_idle and acpi_idle), no states with disabled=1 are registered with cpuidle. Instead, intel_idle (currently) skips them in the loop that registers states. (and acpi_idle never touches the disabled field) And so for x86, governors checking for drv->states[i].disabled is a NOP, and the condition described by CPUIDLE_STATE_DISABLED_BY_DRIVER does not (yet) exist. Looking at the ARM code, it seems that cpuidle-imx6q.c and cpuidle-tegra20.c reach into the cpuidle states at run time and toggle the drv->states[i].disabled. It seems that this patch takes the initial value of drv->states->disabled, and sets the (per cpu) usage.disable=..BY_DRIVER, but that subsequent run-time toggles in drv->states[i]disabled by these drivers would be missed, because you're removed the run-time checking of drv->states->disabled? Finally, I'd like to change intel_idle so that it *can* register a state that is disabled, by default. If I change the driver to NOT skip registering disabled states, and the cpuidle copy has cpuidle_state.disabled=1, then the state is indeed, unused at run-time. But as you said, it is effectively read-only, and is not indicated in sysfs, and can not be changed via sysfs. One way to do this is to do what you do here and initialize usage.disabled to drv->state.disabled. (not distinguishing between DRIVER and USER) That way the user could later over-ride what a driver set, by clearing the disabled attribute. However, the ARM drivers, at least, seem to want to reserve the right to set and clear the drv->state.disabled, and to have them continue to have that right, we have to continue checking that field at run-time. And giving drivers the opportunity to do that disabling driver-wide, instead of per-cpu (usage) wide, seems to be something we may want to keep. -Len
On Mon, Nov 18, 2019 at 5:46 AM Len Brown <lenb@kernel.org> wrote: > > On Mon, Nov 4, 2019 at 6:16 AM Rafael J. Wysocki <rjw@rjwysocki.net> wrote: > > > > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > > > There are two reasons why CPU idle states may be disabled: either > > because the driver has disabled them or because they have been > > disabled by user space via sysfs. > > > > In the former case, the state's "disabled" flag is set once during > > the initialization of the driver and it is never cleared later (it > > is read-only effectively). > > for x86 (intel_idle and acpi_idle), no states with disabled=1 are registered > with cpuidle. Instead, intel_idle (currently) skips them in the loop > that registers states. > (and acpi_idle never touches the disabled field) > > And so for x86, governors checking for drv->states[i].disabled is a NOP, > and the condition described by CPUIDLE_STATE_DISABLED_BY_DRIVER > does not (yet) exist. OK > Looking at the ARM code, it seems that cpuidle-imx6q.c and cpuidle-tegra20.c > reach into the cpuidle states at run time and toggle the > drv->states[i].disabled. I might have overlooked that, let me check. > It seems that this patch takes the initial value of > drv->states->disabled, and sets the (per cpu) > usage.disable=..BY_DRIVER, > but that subsequent run-time toggles in drv->states[i]disabled by > these drivers would be missed, > because you're removed the run-time checking of drv->states->disabled? If it is updated at run time, then yes, the updates will be missed, so thanks for pointing that out. > Finally, I'd like to change intel_idle so that it *can* register a > state that is disabled, by default. > If I change the driver to NOT skip registering disabled states, and > the cpuidle copy has cpuidle_state.disabled=1, > then the state is indeed, unused at run-time. But as you said, > it is effectively read-only, and is not indicated in sysfs, and can > not be changed via sysfs. > > One way to do this is to do what you do here and initialize > usage.disabled to drv->state.disabled. (not distinguishing between > DRIVER and USER) > That way the user could later over-ride what a driver set, by clearing > the disabled attribute. > > However, the ARM drivers, at least, seem to want to reserve the right > to set and clear the drv->state.disabled, > and to have them continue to have that right, we have to continue > checking that field at run-time. Alternatively, the drivers in question can be changed to update the disable field in state_usage instead (maybe under a lock to prevent them from racing with user space). > And giving drivers the opportunity to do that disabling driver-wide, > instead of per-cpu (usage) wide, > seems to be something we may want to keep. So it looks like you want me to revert this patch which is something that I really don't want to do, because of the extra checks all over the place which are simply pointless in the majority of cases. Cheers, Rafael
On Mon, Nov 18, 2019 at 10:22 AM Rafael J. Wysocki <rafael@kernel.org> wrote: > > On Mon, Nov 18, 2019 at 5:46 AM Len Brown <lenb@kernel.org> wrote: > > > > On Mon, Nov 4, 2019 at 6:16 AM Rafael J. Wysocki <rjw@rjwysocki.net> wrote: > > > > > > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > > > > > There are two reasons why CPU idle states may be disabled: either > > > because the driver has disabled them or because they have been > > > disabled by user space via sysfs. > > > > > > In the former case, the state's "disabled" flag is set once during > > > the initialization of the driver and it is never cleared later (it > > > is read-only effectively). > > > > for x86 (intel_idle and acpi_idle), no states with disabled=1 are registered > > with cpuidle. Instead, intel_idle (currently) skips them in the loop > > that registers states. > > (and acpi_idle never touches the disabled field) > > > > And so for x86, governors checking for drv->states[i].disabled is a NOP, > > and the condition described by CPUIDLE_STATE_DISABLED_BY_DRIVER > > does not (yet) exist. > > OK > > > Looking at the ARM code, it seems that cpuidle-imx6q.c and cpuidle-tegra20.c > > reach into the cpuidle states at run time and toggle the > > drv->states[i].disabled. > > I might have overlooked that, let me check. > > > It seems that this patch takes the initial value of > > drv->states->disabled, and sets the (per cpu) > > usage.disable=..BY_DRIVER, > > but that subsequent run-time toggles in drv->states[i]disabled by > > these drivers would be missed, > > because you're removed the run-time checking of drv->states->disabled? > > If it is updated at run time, then yes, the updates will be missed, so > thanks for pointing that out. > > > Finally, I'd like to change intel_idle so that it *can* register a > > state that is disabled, by default. > > If I change the driver to NOT skip registering disabled states, and > > the cpuidle copy has cpuidle_state.disabled=1, > > then the state is indeed, unused at run-time. But as you said, > > it is effectively read-only, and is not indicated in sysfs, and can > > not be changed via sysfs. > > > > One way to do this is to do what you do here and initialize > > usage.disabled to drv->state.disabled. (not distinguishing between > > DRIVER and USER) > > That way the user could later over-ride what a driver set, by clearing > > the disabled attribute. I'd rather get rid of the "disabled" field from struct cpuidle_state entirely and introduce a new state flag to indicate the "disabled by default" status. I also would expose that new flag in a new sysfs attribute of idle states, say "disable_default". Then, the DISABLED_BY_DRIVER bit would be reserved for driver quirks (as per https://patchwork.kernel.org/patch/11249519/) and the DISABLED_BY_USER one could be used for all of the other purposes. Cheers, Rafael
Index: linux-pm/drivers/cpuidle/cpuidle.c =================================================================== --- linux-pm.orig/drivers/cpuidle/cpuidle.c +++ linux-pm/drivers/cpuidle/cpuidle.c @@ -84,12 +84,12 @@ static int find_deepest_state(struct cpu for (i = 1; i < drv->state_count; i++) { struct cpuidle_state *s = &drv->states[i]; - struct cpuidle_state_usage *su = &dev->states_usage[i]; - if (s->disabled || su->disable || s->exit_latency <= latency_req - || s->exit_latency > max_latency - || (s->flags & forbidden_flags) - || (s2idle && !s->enter_s2idle)) + if (dev->states_usage[i].disable || + s->exit_latency <= latency_req || + s->exit_latency > max_latency || + (s->flags & forbidden_flags) || + (s2idle && !s->enter_s2idle)) continue; latency_req = s->exit_latency; @@ -265,8 +265,7 @@ int cpuidle_enter_state(struct cpuidle_d if (diff < drv->states[entered_state].target_residency) { for (i = entered_state - 1; i >= 0; i--) { - if (drv->states[i].disabled || - dev->states_usage[i].disable) + if (dev->states_usage[i].disable) continue; /* Shallower states are enabled, so update. */ @@ -275,8 +274,7 @@ int cpuidle_enter_state(struct cpuidle_d } } else if (diff > delay) { for (i = entered_state + 1; i < drv->state_count; i++) { - if (drv->states[i].disabled || - dev->states_usage[i].disable) + if (dev->states_usage[i].disable) continue; /* @@ -380,7 +378,7 @@ u64 cpuidle_poll_time(struct cpuidle_dri limit_ns = TICK_NSEC; for (i = 1; i < drv->state_count; i++) { - if (drv->states[i].disabled || dev->states_usage[i].disable) + if (dev->states_usage[i].disable) continue; limit_ns = (u64)drv->states[i].target_residency * NSEC_PER_USEC; @@ -567,12 +565,16 @@ static void __cpuidle_device_init(struct */ static int __cpuidle_register_device(struct cpuidle_device *dev) { - int ret; struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev); + int i, ret; if (!try_module_get(drv->owner)) return -EINVAL; + for (i = 0; i < drv->state_count; i++) + if (drv->states[i].disabled) + dev->states_usage[i].disable |= CPUIDLE_STATE_DISABLED_BY_DRIVER; + per_cpu(cpuidle_devices, dev->cpu) = dev; list_add(&dev->device_list, &cpuidle_detected_devices); Index: linux-pm/drivers/cpuidle/sysfs.c =================================================================== --- linux-pm.orig/drivers/cpuidle/sysfs.c +++ linux-pm/drivers/cpuidle/sysfs.c @@ -255,25 +255,6 @@ static ssize_t show_state_##_name(struct return sprintf(buf, "%u\n", state->_name);\ } -#define define_store_state_ull_function(_name) \ -static ssize_t store_state_##_name(struct cpuidle_state *state, \ - struct cpuidle_state_usage *state_usage, \ - const char *buf, size_t size) \ -{ \ - unsigned long long value; \ - int err; \ - if (!capable(CAP_SYS_ADMIN)) \ - return -EPERM; \ - err = kstrtoull(buf, 0, &value); \ - if (err) \ - return err; \ - if (value) \ - state_usage->_name = 1; \ - else \ - state_usage->_name = 0; \ - return size; \ -} - #define define_show_state_ull_function(_name) \ static ssize_t show_state_##_name(struct cpuidle_state *state, \ struct cpuidle_state_usage *state_usage, \ @@ -299,11 +280,39 @@ define_show_state_ull_function(usage) define_show_state_ull_function(time) define_show_state_str_function(name) define_show_state_str_function(desc) -define_show_state_ull_function(disable) -define_store_state_ull_function(disable) define_show_state_ull_function(above) define_show_state_ull_function(below) +static ssize_t show_state_disable(struct cpuidle_state *state, + struct cpuidle_state_usage *state_usage, + char *buf) +{ + return sprintf(buf, "%llu\n", + state_usage->disable & CPUIDLE_STATE_DISABLED_BY_USER); +} + +static ssize_t store_state_disable(struct cpuidle_state *state, + struct cpuidle_state_usage *state_usage, + const char *buf, size_t size) +{ + unsigned int value; + int err; + + if (!capable(CAP_SYS_ADMIN)) + return -EPERM; + + err = kstrtouint(buf, 0, &value); + if (err) + return err; + + if (value) + state_usage->disable |= CPUIDLE_STATE_DISABLED_BY_USER; + else + state_usage->disable &= ~CPUIDLE_STATE_DISABLED_BY_USER; + + return size; +} + define_one_state_ro(name, show_state_name); define_one_state_ro(desc, show_state_desc); define_one_state_ro(latency, show_state_exit_latency); Index: linux-pm/include/linux/cpuidle.h =================================================================== --- linux-pm.orig/include/linux/cpuidle.h +++ linux-pm/include/linux/cpuidle.h @@ -29,6 +29,9 @@ struct cpuidle_driver; * CPUIDLE DEVICE INTERFACE * ****************************/ +#define CPUIDLE_STATE_DISABLED_BY_USER BIT(0) +#define CPUIDLE_STATE_DISABLED_BY_DRIVER BIT(1) + struct cpuidle_state_usage { unsigned long long disable; unsigned long long usage; Index: linux-pm/drivers/cpuidle/governors/menu.c =================================================================== --- linux-pm.orig/drivers/cpuidle/governors/menu.c +++ linux-pm/drivers/cpuidle/governors/menu.c @@ -298,7 +298,7 @@ static int menu_select(struct cpuidle_dr if (unlikely(drv->state_count <= 1 || latency_req == 0) || ((data->next_timer_us < drv->states[1].target_residency || latency_req < drv->states[1].exit_latency) && - !drv->states[0].disabled && !dev->states_usage[0].disable)) { + !dev->states_usage[0].disable)) { /* * In this case state[0] will be used no matter what, so return * it right away and keep the tick running if state[0] is a @@ -349,9 +349,8 @@ static int menu_select(struct cpuidle_dr idx = -1; for (i = 0; i < drv->state_count; i++) { struct cpuidle_state *s = &drv->states[i]; - struct cpuidle_state_usage *su = &dev->states_usage[i]; - if (s->disabled || su->disable) + if (dev->states_usage[i].disable) continue; if (idx == -1) @@ -422,8 +421,7 @@ static int menu_select(struct cpuidle_dr * tick, so try to correct that. */ for (i = idx - 1; i >= 0; i--) { - if (drv->states[i].disabled || - dev->states_usage[i].disable) + if (dev->states_usage[i].disable) continue; idx = i; Index: linux-pm/drivers/cpuidle/governors/teo.c =================================================================== --- linux-pm.orig/drivers/cpuidle/governors/teo.c +++ linux-pm/drivers/cpuidle/governors/teo.c @@ -212,7 +212,7 @@ static int teo_find_shallower_state(stru int i; for (i = state_idx - 1; i >= 0; i--) { - if (drv->states[i].disabled || dev->states_usage[i].disable) + if (dev->states_usage[i].disable) continue; state_idx = i; @@ -254,9 +254,8 @@ static int teo_select(struct cpuidle_dri for (i = 0; i < drv->state_count; i++) { struct cpuidle_state *s = &drv->states[i]; - struct cpuidle_state_usage *su = &dev->states_usage[i]; - if (s->disabled || su->disable) { + if (dev->states_usage[i].disable) { /* * If the "early hits" metric of a disabled state is * greater than the current maximum, it should be taken Index: linux-pm/drivers/cpuidle/governors/ladder.c =================================================================== --- linux-pm.orig/drivers/cpuidle/governors/ladder.c +++ linux-pm/drivers/cpuidle/governors/ladder.c @@ -84,7 +84,6 @@ static int ladder_select_state(struct cp /* consider promotion */ if (last_idx < drv->state_count - 1 && - !drv->states[last_idx + 1].disabled && !dev->states_usage[last_idx + 1].disable && last_residency > last_state->threshold.promotion_time && drv->states[last_idx + 1].exit_latency <= latency_req) { @@ -98,8 +97,7 @@ static int ladder_select_state(struct cp /* consider demotion */ if (last_idx > first_idx && - (drv->states[last_idx].disabled || - dev->states_usage[last_idx].disable || + (dev->states_usage[last_idx].disable || drv->states[last_idx].exit_latency > latency_req)) { int i; Index: linux-pm/drivers/cpuidle/cpuidle-powernv.c =================================================================== --- linux-pm.orig/drivers/cpuidle/cpuidle-powernv.c +++ linux-pm/drivers/cpuidle/cpuidle-powernv.c @@ -56,13 +56,10 @@ static u64 get_snooze_timeout(struct cpu return default_snooze_timeout; for (i = index + 1; i < drv->state_count; i++) { - struct cpuidle_state *s = &drv->states[i]; - struct cpuidle_state_usage *su = &dev->states_usage[i]; - - if (s->disabled || su->disable) + if (dev->states_usage[i].disable) continue; - return s->target_residency * tb_ticks_per_usec; + return drv->states[i].target_residency * tb_ticks_per_usec; } return default_snooze_timeout;