| Message ID | 20200320151055.h32rh5hpnv7xyuey@macmini.local (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | Daniel Lezcano |
| Headers | show |
| Series | weird cooling_device/cur_state sysfs behaviour | expand |
On 20/03/2020 16:10, Willy Wolff wrote: > Hi all, > Hope you're doing well with the situation. > > I fill that my board get a cold too... > > When I write to a sysfs node, I have a weird behaviour about the function that is called behind. > This bug appears on an arm32 odroid-xu3, and only after v5.4, v5.3.18 behave correctly. > > > Here my modification to see what's going on: > > diff --git a/drivers/thermal/thermal_sysfs.c b/drivers/thermal/thermal_sysfs.c > index aa99edb4dff7..a437ae3f4b9f 100644 > --- a/drivers/thermal/thermal_sysfs.c > +++ b/drivers/thermal/thermal_sysfs.c > @@ -706,11 +706,22 @@ cur_state_store(struct device *dev, struct device_attribute *attr, [ ... ] > return result ? result : count; > } > > + [ ... ] > > As you can see, the function is recalled with one character less, until the "buffer is empty". > I don't understand why. Can anybody help me for this thing? > Many thanks in advance. [ ... ] Very likely the problem is coming from: result = cdev->ops->set_cur_state(cdev, state); which returns '1' as showed by the traces: drivers/thermal/thermal_sysfs.c:735:cur_state_store result = 1, count = 3 And because of the return condition above: return result ? result : count; the function is returning result, so '1', which is interpreted by the sysfs as "I wrote one character', so it recalls the function with the two remaining characters, etc ... The problem is from the governor AFAICT, which governor is it?
Adding Rafael, On 20/03/2020 18:43, Daniel Lezcano wrote: > On 20/03/2020 16:10, Willy Wolff wrote: >> Hi all, >> Hope you're doing well with the situation. >> >> I fill that my board get a cold too... >> >> When I write to a sysfs node, I have a weird behaviour about the function that is called behind. >> This bug appears on an arm32 odroid-xu3, and only after v5.4, v5.3.18 behave correctly. >> >> >> Here my modification to see what's going on: >> >> diff --git a/drivers/thermal/thermal_sysfs.c b/drivers/thermal/thermal_sysfs.c >> index aa99edb4dff7..a437ae3f4b9f 100644 >> --- a/drivers/thermal/thermal_sysfs.c >> +++ b/drivers/thermal/thermal_sysfs.c >> @@ -706,11 +706,22 @@ cur_state_store(struct device *dev, struct device_attribute *attr, > > [ ... ] > >> return result ? result : count; >> } >> >> + > > [ ... ] > >> >> As you can see, the function is recalled with one character less, until the "buffer is empty". >> I don't understand why. Can anybody help me for this thing? >> Many thanks in advance. > > [ ... ] > > Very likely the problem is coming from: > > result = cdev->ops->set_cur_state(cdev, state); > > which returns '1' as showed by the traces: > > drivers/thermal/thermal_sysfs.c:735:cur_state_store result = 1, count = 3 > > And because of the return condition above: > > return result ? result : count; > > the function is returning result, so '1', which is interpreted by the > sysfs as "I wrote one character', so it recalls the function with the > two remaining characters, etc ... > > The problem is from the governor AFAICT, which governor is it? I went through the code and I believe the problem is coming from: cpufreq_set_cur_state -> freq_qos_update_request -> freq_qos_apply -> pm_qos_update_target => " * This function returns 1 if the aggregated constraint value has changed," freq_qos_apply() does: ret = pm_qos_update_target() ... return ret; At the first glance, it is related to commit 77751a466ebd1 (Nov 2019).
Indeed, many thanks for you help. On Fri, Mar 20, 2020 at 09:17:47pm +0100, Daniel Lezcano wrote: > Adding Rafael, > > On 20/03/2020 18:43, Daniel Lezcano wrote: > > On 20/03/2020 16:10, Willy Wolff wrote: > >> Hi all, > >> Hope you're doing well with the situation. > >> > >> I fill that my board get a cold too... > >> > >> When I write to a sysfs node, I have a weird behaviour about the function that is called behind. > >> This bug appears on an arm32 odroid-xu3, and only after v5.4, v5.3.18 behave correctly. > >> > >> > >> Here my modification to see what's going on: > >> > >> diff --git a/drivers/thermal/thermal_sysfs.c b/drivers/thermal/thermal_sysfs.c > >> index aa99edb4dff7..a437ae3f4b9f 100644 > >> --- a/drivers/thermal/thermal_sysfs.c > >> +++ b/drivers/thermal/thermal_sysfs.c > >> @@ -706,11 +706,22 @@ cur_state_store(struct device *dev, struct device_attribute *attr, > > > > [ ... ] > > > >> return result ? result : count; > >> } > >> > >> + > > > > [ ... ] > > > >> > >> As you can see, the function is recalled with one character less, until the "buffer is empty". > >> I don't understand why. Can anybody help me for this thing? > >> Many thanks in advance. > > > > [ ... ] > > > > Very likely the problem is coming from: > > > > result = cdev->ops->set_cur_state(cdev, state); > > > > which returns '1' as showed by the traces: > > > > drivers/thermal/thermal_sysfs.c:735:cur_state_store result = 1, count = 3 > > > > And because of the return condition above: > > > > return result ? result : count; > > > > the function is returning result, so '1', which is interpreted by the > > sysfs as "I wrote one character', so it recalls the function with the > > two remaining characters, etc ... > > > > The problem is from the governor AFAICT, which governor is it? > > I went through the code and I believe the problem is coming from: > > cpufreq_set_cur_state > -> freq_qos_update_request > -> freq_qos_apply > -> pm_qos_update_target => > > " * This function returns 1 if the aggregated constraint value has changed," > > freq_qos_apply() does: > > ret = pm_qos_update_target() > ... > return ret; > > > At the first glance, it is related to commit 77751a466ebd1 (Nov 2019). > > > > > -- > <http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs > > Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | > <http://twitter.com/#!/linaroorg> Twitter | > <http://www.linaro.org/linaro-blog/> Blog >
diff --git a/drivers/thermal/thermal_sysfs.c b/drivers/thermal/thermal_sysfs.c index aa99edb4dff7..a437ae3f4b9f 100644 --- a/drivers/thermal/thermal_sysfs.c +++ b/drivers/thermal/thermal_sysfs.c @@ -706,11 +706,22 @@ cur_state_store(struct device *dev, struct device_attribute *attr, unsigned long state; int result; - if (sscanf(buf, "%ld\n", &state) != 1) + dump_stack(); + + printk(KERN_INFO "%s:%d:%s buf = \"%s\", count = %zu\n", + __FILE__, __LINE__, __func__, buf, count); + + if (sscanf(buf, "%ld\n", &state) != 1) { + printk(KERN_INFO "%s:%d:%s\n", + __FILE__, __LINE__, __func__); return -EINVAL; + } - if ((long)state < 0) + if ((long)state < 0) { + printk(KERN_INFO "%s:%d:%s\n", + __FILE__, __LINE__, __func__); return -EINVAL; + } mutex_lock(&cdev->lock); @@ -719,9 +730,14 @@ cur_state_store(struct device *dev, struct device_attribute *attr, thermal_cooling_device_stats_update(cdev, state); mutex_unlock(&cdev->lock); + + printk(KERN_INFO "%s:%d:%s result = %d, count = %zu\n", + __FILE__, __LINE__, __func__, result, count); + return result ? result : count; } + static struct device_attribute dev_attr_cdev_type = __ATTR(type, 0444, cdev_type_show, NULL); static DEVICE_ATTR_RO(max_state);