drivers/base: cacheinfo: remove warning in resume

Submitted by Sumit Gupta on Aug. 29, 2016, 7:20 a.m.

Details

Message ID 1472455254-27692-1-git-send-email-sumitg@nvidia.com
State New
Headers show

Commit Message

Sumit Gupta Aug. 29, 2016, 7:20 a.m.
CPU notifier is present for creating device
entries for child node "cache" under parent
node "cpu" as per DT. During resume from
suspend, while booting all non-boot CPU's,
this notifier for adding cache device gets
called before cpu device is added by
device_resume. Because of this warning message
of "parent should not be sleeping" comes during
resume.

Removing the notifier to explicitly add/remove
cache device as CPU and cache device get
added/removed anyway as part of normal suspend
resume sequence.
dpm_resume_end - > dpm_resume -> device_resume

Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
---
 drivers/base/cacheinfo.c | 2 --
 1 file changed, 2 deletions(-)

Comments

Sudeep Holla Aug. 30, 2016, 10:20 a.m.
Hi Sumit,

I understand the warning we get but the patch is completely wrong.
One it removes the feature of adding/removing the cache devices
on cpu hotplug events. Have you tested your patch with simple cpu
hotplug and seen no change before and after this change ?

On 29/08/16 08:20, Sumit Gupta wrote:
> CPU notifier is present for creating device
> entries for child node "cache" under parent
> node "cpu" as per DT.

Again DT is only on few architectures not on all(e.g. x86)

> During resume from
> suspend, while booting all non-boot CPU's,
> this notifier for adding cache device gets
> called before cpu device is added by
> device_resume. Because of this warning message
> of "parent should not be sleeping" comes during
> resume.
>

Yes that's correct and needs to be fixed. I have seen this but haven't
spent much time to check in detail. It's harmless warning IMO.
See the comment in the code too:"This is a fib. But we'll allow new
children to be added below a resumed device, even if the device hasn't
been completed yet"

CPU devices are special and they have separate hotplug paths. So we need
to consider that for cpu devices and set is_prepared quite early.

> Removing the notifier to explicitly add/remove
> cache device as CPU and cache device get
> added/removed anyway as part of normal suspend
> resume sequence.
> dpm_resume_end - > dpm_resume -> device_resume
>

Yes, but:
1. the caches objects are visible even when the cpu is offline
2. how is this handled for normal cpu-hotplug events ? This patch
    breaks the existing feature.

> Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
> ---
>  drivers/base/cacheinfo.c | 2 --
>  1 file changed, 2 deletions(-)
>
> diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
> index e9fd32e91668..17d9c051a16f 100644
> --- a/drivers/base/cacheinfo.c
> +++ b/drivers/base/cacheinfo.c
> @@ -544,8 +544,6 @@ static int __init cacheinfo_sysfs_init(void)
>  			goto out;
>  		}
>  	}
> -	__hotcpu_notifier(cacheinfo_cpu_callback, 0);
> -
>  out:
>  	cpu_notifier_register_done();
>  	return rc;
>
Sumit Gupta Sept. 2, 2016, 12:58 p.m.
Hi Sudeep,

Thank you for your comments.
> 
> I understand the warning we get but the patch is completely wrong.
> One it removes the feature of adding/removing the cache devices on cpu
> hotplug events. Have you tested your patch with simple cpu hotplug and seen no
> change before and after this change ?
> 

I referred other node "cpufreq" under "cpu" node and that was also not getting deleted when core goes down.
So, thought the behavior should be similar as entries are only used to read data and it won't change after boot.

> On 29/08/16 08:20, Sumit Gupta wrote:
> > CPU notifier is present for creating device entries for child node
> > "cache" under parent node "cpu" as per DT.
> 
> Again DT is only on few architectures not on all(e.g. x86)
> 
> > During resume from
> > suspend, while booting all non-boot CPU's, this notifier for adding
> > cache device gets called before cpu device is added by device_resume.
> > Because of this warning message of "parent should not be sleeping"
> > comes during resume.
> >
> 
> Yes that's correct and needs to be fixed. I have seen this but haven't
> spent much time to check in detail. It's harmless warning IMO.
> See the comment in the code too:"This is a fib. But we'll allow new
> children to be added below a resumed device, even if the device hasn't
> been completed yet"
> 
> CPU devices are special and they have separate hotplug paths. So we need
> to consider that for cpu devices and set is_prepared quite early.
> 
> > Removing the notifier to explicitly add/remove
> > cache device as CPU and cache device get
> > added/removed anyway as part of normal suspend
> > resume sequence.
> > dpm_resume_end - > dpm_resume -> device_resume
> >
> 
> Yes, but:
> 1. the caches objects are visible even when the cpu is offline
> 2. how is this handled for normal cpu-hotplug events ? This patch
>     breaks the existing feature.

Compared with x86 now and there even cpufreq is getting deleted. 
As per the comments, right behavior seems to be where cache entries should be created during core on and removed during core off in hotplug.
I will try to find other way of doing it. Also will see why cpufreq is not getting deleted as in x86.

> 
> > Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
> > ---
> >  drivers/base/cacheinfo.c | 2 --
> >  1 file changed, 2 deletions(-)
> >
> > diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
> > index e9fd32e91668..17d9c051a16f 100644
> > --- a/drivers/base/cacheinfo.c
> > +++ b/drivers/base/cacheinfo.c
> > @@ -544,8 +544,6 @@ static int __init cacheinfo_sysfs_init(void)
> >  			goto out;
> >  		}
> >  	}
> > -	__hotcpu_notifier(cacheinfo_cpu_callback, 0);
> > -
> >  out:
> >  	cpu_notifier_register_done();
> >  	return rc;
> >
> 
> --
> Regards,
> Sudeep

--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Sudeep Holla Sept. 2, 2016, 3:02 p.m.
On 02/09/16 13:58, Sumit Gupta wrote:
> Hi Sudeep,
>
> Thank you for your comments.
>>
>> I understand the warning we get but the patch is completely wrong.
>> One it removes the feature of adding/removing the cache devices on
>> cpu hotplug events. Have you tested your patch with simple cpu
>> hotplug and seen no change before and after this change ?
>>
>
> I referred other node "cpufreq" under "cpu" node and that was also
> not getting deleted when core goes down. So, thought the behavior
> should be similar as entries are only used to read data and it won't
> change after boot.
>

I agree, but for caches it has been like this and in a way make sense.

>> On 29/08/16 08:20, Sumit Gupta wrote:
>>> CPU notifier is present for creating device entries for child
>>> node "cache" under parent node "cpu" as per DT.
>>
>> Again DT is only on few architectures not on all(e.g. x86)
>>
>>> During resume from suspend, while booting all non-boot CPU's,
>>> this notifier for adding cache device gets called before cpu
>>> device is added by device_resume. Because of this warning message
>>> of "parent should not be sleeping" comes during resume.
>>>
>>
>> Yes that's correct and needs to be fixed. I have seen this but
>> haven't spent much time to check in detail. It's harmless warning
>> IMO. See the comment in the code too:"This is a fib. But we'll
>> allow new children to be added below a resumed device, even if the
>> device hasn't been completed yet"
>>
>> CPU devices are special and they have separate hotplug paths. So we
>> need to consider that for cpu devices and set is_prepared quite
>> early.
>>
>>> Removing the notifier to explicitly add/remove cache device as
>>> CPU and cache device get added/removed anyway as part of normal
>>> suspend resume sequence. dpm_resume_end - > dpm_resume ->
>>> device_resume
>>>
>>
>> Yes, but: 1. the caches objects are visible even when the cpu is
>> offline 2. how is this handled for normal cpu-hotplug events ? This
>> patch breaks the existing feature.
>
> Compared with x86 now and there even cpufreq is getting deleted. As
> per the comments, right behavior seems to be where cache entries
> should be created during core on and removed during core off in
> hotplug. I will try to find other way of doing it. Also will see why
> cpufreq is not getting deleted as in x86.
>

IIUC you are saying cpufreq sysfs entries are deleted on x86 while not
on ARM{32,64} ? If so are you running same kernel version on both x86
and ARM ? I remember some recent change around cpufreq sysfs entries, it
could be result of that.

Patch hide | download patch | download mbox

diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
index e9fd32e91668..17d9c051a16f 100644
--- a/drivers/base/cacheinfo.c
+++ b/drivers/base/cacheinfo.c
@@ -544,8 +544,6 @@  static int __init cacheinfo_sysfs_init(void)
 			goto out;
 		}
 	}
-	__hotcpu_notifier(cacheinfo_cpu_callback, 0);
-
 out:
 	cpu_notifier_register_done();
 	return rc;