Re: [Intel-gfx] [PATCH v5 1/2] module: update dependencies at try_module_get()

From: Mauro Carvalho Chehab <mchehab@kernel.org>
To: Andi Shyti <andi.shyti@linux.intel.com>
Cc: Luis Chamberlain <mcgrof@kernel.org>,
	alsa-devel@alsa-project.org, mauro.chehab@linux.intel.com,
	David Airlie <airlied@linux.ie>,
	Greg KH <gregkh@linuxfoundation.org>,
	intel-gfx@lists.freedesktop.org,
	Lucas De Marchi <lucas.demarchi@intel.com>,
	Takashi Iwai <tiwai@suse.com>,
	dri-devel@lists.freedesktop.org, Jaroslav Kysela <perex@perex.cz>,
	Kai Vehmanen <kai.vehmanen@intel.com>,
	linux-modules@vger.kernel.org,
	Dan Williams <dan.j.williams@intel.com>,
	linux-kernel@vger.kernel.org,
	Pierre-Louis Bossart <pierre-louis.bossart@intel.com>
Subject: Re: [Intel-gfx] [PATCH v5 1/2] module: update dependencies at try_module_get()
Date: Mon, 9 May 2022 18:56:20 +0200	[thread overview]
Message-ID: <20220509185620.05567716@coco.lan> (raw)
In-Reply-To: <YnRDIfthGJXdY23h@intel.intel>

Em Thu, 5 May 2022 23:35:29 +0200
Andi Shyti <andi.shyti@linux.intel.com> escreveu:

> Hi Mauro,
> 
> [...]
> 
> > +static int ref_module_dependency(struct module *mod, struct module *this)
> > +{
> > +	int ret;
> > +
> > +	if (!this || !this->name)
> > +		return -EINVAL;
> > +
> > +	if (mod == this)
> > +		return 0;
> > +
> > +	mutex_lock(&module_mutex);
> > +
> > +	ret = ref_module(this, mod);
> > +
> > +#ifdef CONFIG_MODULE_UNLOAD
> > +	if (ret)
> > +		goto ret;
> > +
> > +	ret = sysfs_create_link(mod->holders_dir,
> > +				&this->mkobj.kobj, this->name);
> > +#endif
> > +
> > +ret:
> > +	mutex_unlock(&module_mutex);
> > +	return ret;
> > +}
> > +
> >  /* Clear the unload stuff of the module. */
> >  static void module_unload_free(struct module *mod)
> >  {
> > @@ -841,24 +886,16 @@ void __module_get(struct module *module)
> >  }
> >  EXPORT_SYMBOL(__module_get);
> >  
> > -bool try_module_get(struct module *module)
> > +bool try_module_get_owner(struct module *module, struct module *this)
> >  {
> > -	bool ret = true;
> > +	int ret = __try_module_get(module);
> >  
> > -	if (module) {
> > -		preempt_disable();
> > -		/* Note: here, we can fail to get a reference */
> > -		if (likely(module_is_live(module) &&
> > -			   atomic_inc_not_zero(&module->refcnt) != 0))
> > -			trace_module_get(module, _RET_IP_);
> > -		else
> > -			ret = false;
> > +	if (ret)
> > +		ref_module_dependency(module, this);  
> 
> do we care about the return value here?

I don't think it should care about the return value, as a failure to
create a sysfs node for the holder or to add it to the holders list
is not fatal: modules can still continue working without that.

Also, I opted to be conservative here: currently, not creating these
doesn't cause try_module_get() to fail. I'm not sure what would be the
impact if this starts to fail.

So, right now, I'm opting to just ignore the return value. Perhaps
in the future this could a warning (similarly to what sysfs create
link does).

Regards,
Mauro

> 
> Andi
> 
> >  
> > -		preempt_enable();
> > -	}
> >  	return ret;
> >  }
> > -EXPORT_SYMBOL(try_module_get);
> > +EXPORT_SYMBOL(try_module_get_owner);
> >  
> >  void module_put(struct module *module)
> >  {
> > -- 
> > 2.35.1  

Thanks,
Mauro