Re: [RFT][PATCH 1/3] PM: QoS: Introduce frequency QoS

["Rafael J. Wysocki" <rjw@rjwysocki.net>]

On Tuesday, November 19, 2019 8:17:05 PM CET Rafael J. Wysocki wrote:
> On Tue, Nov 19, 2019 at 3:35 PM Doug Smythies <dsmythies@telus.net> wrote:
> >
> > On 2019.11.17 08:13 Doug Smythies wrote:
> > > On 2019.11.16 23:35 Doug Smythies wrote:
> >
> > >> Hi Rafael,
> > >>
> > >> Not sure, but I think it is this one that
> > >> causes complaining when I try to set the
> > >> intel_pstate driver to passive mode.
> > >> I started from active mode, powersave governor,
> > >> no HWP.
> > >>
> > >> Kernel: 5.4-rc7
> > >>
> > >> I did not go back and try previous 5.4 RCs.
> >
> > After looking at the git tags for this patch,
> > I tried kernel 5.4-rc2, which was the closest
> > Kernel I had to before the patch set was added.
> > It worked fine, as expected.
> >
> > >> I did try kernel 5.3-rc8, because I already had
> > >> it installed, and it worked fine.
> > >>
> > >> I use a script (for years), run as sudo:
> > >>
> > >> doug@s15:~/temp$ cat set_cpu_passive
> > >> #! /bin/bash
> > >> cat /sys/devices/system/cpu/intel_pstate/status
> > >> echo passive > /sys/devices/system/cpu/intel_pstate/status
> > >> cat /sys/devices/system/cpu/intel_pstate/status
> > >>
> > >> And I get this (very small excerpt):
> > >>
> > >> freq_qos_add_request() called for active request
> > >> WARNING: CPU: 1 PID: 2758 at kernel/power/qos.c:763 freq_qos_add_request+0x4c/0xa0
> > >> CPU: 1 PID: 2758 Comm: set_cpu_passive Not tainted 5.4.0-rc7-stock #727
> > >> Failed to add freq constraint for CPU0 (-22)
> > >>
> > >> freq_qos_add_request() called for active request
> > >> WARNING: CPU: 1 PID: 2758 at kernel/power/qos.c:763 freq_qos_add_request+0x4c/0xa0
> > >> CPU: 1 PID: 2758 Comm: set_cpu_passive Tainted: G        W         5.4.0-rc7-stock #727
> > >> Failed to add freq constraint for CPU1 (-22)
> >
> > Updated summary of previous emails:
> > This patch or patch set breaks the after boot
> > ability to change CPU frequency scaling drivers.
> >
> > Using a workaround of booting with
> > "intel_pstate=passive" seems to prevent the errors.
> >
> > Changing between the intel_pstate and intel_cpufreq drivers
> > (i.e. between active and passive modes)
> > after boot, either way, causes the errors. i.e.
> >
> > Failed to add freq constraint for CPU7 (-22)
> > (2 per CPU per attempt)
> 
> These messages come from acpi_processor_ppc_init() and
> acpi_thermal_cpufreq_init(), AFAICS, which are invoked by
> acpi_processor_notifier() and that is invoked by the
> blocking_notifier_call_chain() in cpufreq_online() which tirggers for
> new policies after adding the max freq QoS request to
> policy->constraints.
> 
> The requests added by them should be removed by
> acpi_processor_ppc_exit() and acpi_thermal_cpufreq_exit(),
> respectively, invoked by the blocking_notifier_call_chain() in
> cpufreq_policy_free(), but it looks like that doesn't happen.
> 
> However, I now also see that freq_qos_remove_request() doesn't clear
> the qos field in req which is should do, so freq_qos_add_request()
> will complain and fail if the object pointed to by req is passed to it
> again.
> 
> I'll send a patch to test for this later today.
> 

The patch is appended.  Please test it (on top of 5.4-rc8) and report back.


---
 kernel/power/qos.c |    8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

Index: linux-pm/kernel/power/qos.c
===================================================================
--- linux-pm.orig/kernel/power/qos.c
+++ linux-pm/kernel/power/qos.c
@@ -814,6 +814,8 @@ EXPORT_SYMBOL_GPL(freq_qos_update_reques
  */
 int freq_qos_remove_request(struct freq_qos_request *req)
 {
+	int ret;
+
 	if (!req)
 		return -EINVAL;
 
@@ -821,7 +823,11 @@ int freq_qos_remove_request(struct freq_
 		 "%s() called for unknown object\n", __func__))
 		return -EINVAL;
 
-	return freq_qos_apply(req, PM_QOS_REMOVE_REQ, PM_QOS_DEFAULT_VALUE);
+	ret = freq_qos_apply(req, PM_QOS_REMOVE_REQ, PM_QOS_DEFAULT_VALUE);
+	req->qos = NULL;
+	req->type = 0;
+
+	return ret;
 }
 EXPORT_SYMBOL_GPL(freq_qos_remove_request);