Message-ID: <20210423114241.29cf1ab3@coco.lan>
Date:   Fri, 23 Apr 2021 11:42:55 +0200
From:   Mauro Carvalho Chehab <mchehab+huawei@...nel.org>
To:     Julia Lawall <julia.lawall@...ia.fr>
Cc:     Krzysztof Kozlowski <krzysztof.kozlowski@...onical.com>,
        Hans Verkuil <hverkuil-cisco@...all.nl>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        linux-kernel@...r.kernel.org,
        "Rafael J. Wysocki" <rjw@...ysocki.net>
Subject: Re: [PATCH 009/190] Revert "media: s5p-mfc: Fix a reference count
 leak"

(adding c/c to Rafael)

Em Fri, 23 Apr 2021 10:41:32 +0200 (CEST)
Julia Lawall <julia.lawall@...ia.fr> escreveu:

> On Fri, 23 Apr 2021, Krzysztof Kozlowski wrote:
> 
> > On 23/04/2021 10:10, Hans Verkuil wrote:  
> > > On 23/04/2021 10:07, Mauro Carvalho Chehab wrote:  
> > >> Em Fri, 23 Apr 2021 09:10:32 +0200
> > >> Greg Kroah-Hartman <gregkh@...uxfoundation.org> escreveu:
> > >>  
> > >>> On Fri, Apr 23, 2021 at 09:04:27AM +0200, Krzysztof Kozlowski wrote:  
> > >>>> On 21/04/2021 14:58, Greg Kroah-Hartman wrote:  
> > >>>>> This reverts commit 78741ce98c2e36188e2343434406b0e0bc50b0e7.
> > >>>>>
> > >>>>> Commits from @umn.edu addresses have been found to be submitted in "bad
> > >>>>> faith" to try to test the kernel community's ability to review "known
> > >>>>> malicious" changes.  The result of these submissions can be found in a
> > >>>>> paper published at the 42nd IEEE Symposium on Security and Privacy
> > >>>>> entitled, "Open Source Insecurity: Stealthily Introducing
> > >>>>> Vulnerabilities via Hypocrite Commits" written by Qiushi Wu (University
> > >>>>> of Minnesota) and Kangjie Lu (University of Minnesota).
> > >>>>>
> > >>>>> Because of this, all submissions from this group must be reverted from
> > >>>>> the kernel tree and will need to be re-reviewed again to determine if
> > >>>>> they actually are a valid fix.  Until that work is complete, remove this
> > >>>>> change to ensure that no problems are being introduced into the
> > >>>>> codebase.
> > >>>>>
> > >>>>> Cc: Qiushi Wu <wu000273@....edu>
> > >>>>> Cc: Hans Verkuil <hverkuil-cisco@...all.nl>
> > >>>>> Cc: Mauro Carvalho Chehab <mchehab+huawei@...nel.org>
> > >>>>> Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
> > >>>>> ---
> > >>>>>  drivers/media/platform/s5p-mfc/s5p_mfc_pm.c | 4 +---
> > >>>>>  1 file changed, 1 insertion(+), 3 deletions(-)
> > >>>>>  
> > >>>>
> > >>>> This looks like a good commit but should be done now in a different way
> > >>>> - using pm_runtime_resume_and_get().  Therefore I am fine with revert
> > >>>> and I can submit later better fix.  
> > >>>
> > >>> Great, thanks for letting me know, I can have someone work on the
> > >>> "better fix" at the same time.  
> > >>
> > >> IMO, it is better to keep the fix. I mean, there's no reason to
> > >> revert a fix that it is known to be good.
> > >>
> > >> The "better fix" patch can be produced anytime. A simple coccinelle
> > >> ruleset can replace patterns like:
> > >>
> > >> 	ret = pm_runtime_get_sync(pm->device);
> > >> 	if (ret < 0) {
> > >> 		pm_runtime_put_noidle(pm->device);
> > >> 		return ret;
> > >> 	}
> > >>
> > >> and the broken pattern:
> > >>
> > >> 	ret = pm_runtime_get_sync(pm->device);
> > >> 	if (ret < 0)
> > >> 		return ret;
> > >>
> > >> to:
> > >>
> > >> 	ret = pm_runtime_resume_and_get(pm->device);
> > >> 	if (ret < 0)
> > >> 		return ret;  
> > >
> > > That's my preference as well.  
> >
> > It won't be that easy because sometimes the error handling is via goto
> > (like in other patches here) but anyway I don't mind keeping the
> > original commits.  
> 
> I tried the following semantic patch:
> 
> @@
> expression ret,e;
> @@
> 
> -     ret = pm_runtime_get_sync(e);
> +     ret = pm_resume_and_get(e);
>       if (ret < 0) {
>               ...
> ?-            pm_runtime_put_noidle(e);
>               ...
>               return ret;
>       }
> 
> It has the following features:
> 
> * The ? means that if pm_runtime_put_noidle is absent, the transformation
> will happen anyway.
> 
> * The ... before the return means that the matching will jump over a goto.
> 
> It makes a lot of changes (in a kernel I had handy from March). 

I would expect lots of changes, as the pm_runtime_resume_and_get() was only
recently introduced on this changeset:

commit dd8088d5a8969dc2b42f71d7bc01c25c61a78066
Author: Zhang Qilong <zhangqilong3@...wei.com>
Date:   Tue Nov 10 17:29:32 2020 +0800

    PM: runtime: Add pm_runtime_resume_and_get to deal with usage counter
    
    In many case, we need to check return value of pm_runtime_get_sync, but
    it brings a trouble to the usage counter processing. Many callers forget
    to decrease the usage counter when it failed, which could resulted in
    reference leak. It has been discussed a lot[0][1]. So we add a function
    to deal with the usage counter for better coding.
    
    [0]https://lkml.org/lkml/2020/6/14/88
    [1]https://patchwork.ozlabs.org/project/linux-tegra/list/?series=178139
    Signed-off-by: Zhang Qilong <zhangqilong3@...wei.com>
    Acked-by: Rafael J. Wysocki  <rafael.j.wysocki@...el.com>
    Signed-off-by: Jakub Kicinski <kuba@...nel.org>

> This is a
> complicated API, however, and I don't know if there are any other issues
> to take into account, especially in the case where the call to
> pm_runtime_put_noidle is not present.

I double-checked the code, despite its name, pm_runtime_put_noidle() just
changes the refcount. See, the relevant code is here:

	static inline void pm_runtime_put_noidle(struct device *dev)
	{
		atomic_add_unless(&dev->power.usage_count, -1, 0);
	}

	static inline int pm_runtime_get_sync(struct device *dev)
	{
		return __pm_runtime_resume(dev, RPM_GET_PUT);
	}

	int __pm_runtime_resume(struct device *dev, int rpmflags)
	{
	        unsigned long flags;
	        int retval;

	        might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
	                        dev->power.runtime_status != RPM_ACTIVE);

	        if (rpmflags & RPM_GET_PUT)
	                atomic_inc(&dev->power.usage_count);

	        spin_lock_irqsave(&dev->power.lock, flags);
	        retval = rpm_resume(dev, rpmflags);
	        spin_unlock_irqrestore(&dev->power.lock, flags);

	        return retval;
	}

Not being an expert at the PM runtime API, at least on my eyes,
replacing pm_runtime_get_sync() by pm_runtime_resume_and_get()
seems to be the right thing to do, but Rafael should know more.

Thanks,
Mauro

Powered by blists - more mailing lists