Message-ID: <20201007081227.d6f6otfsnmlgvegi@pali>
Date:   Wed, 7 Oct 2020 10:12:27 +0200
From:   Pali Rohár <pali@...nel.org>
To:     Bjorn Helgaas <helgaas@...nel.org>
Cc:     Bjorn Helgaas <bhelgaas@...gle.com>, linux-pci@...r.kernel.org,
        linux-kernel@...r.kernel.org,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        Gregory Clement <gregory.clement@...tlin.com>,
        Andrew Lunn <andrew@...n.ch>,
        Krzysztof Wilczyński <kw@...ux.com>,
        Yinghai Lu <yinghai@...nel.org>
Subject: Re: PCI: Race condition in pci_create_sysfs_dev_files

On Tuesday 06 October 2020 17:22:22 Bjorn Helgaas wrote:
> [+cc Krzysztof, Yinghai]
> 
> On Wed, Sep 09, 2020 at 01:28:50PM +0200, Pali Rohár wrote:
> > Hello! I'm adding more people to loop.
> > 
> > Can somebody look at these race conditions and my patch?
> > 
> > On Friday 14 August 2020 10:08:24 Pali Rohár wrote:
> > > Hello! I would like to remind this issue which I reported month ago.
> > > 
> > > On Thursday 16 July 2020 13:04:23 Pali Rohár wrote:
> > > > Hello Bjorn!
> > > > 
> > > > I see following error message in dmesg which looks like a race condition:
> > > > 
> > > > sysfs: cannot create duplicate filename '/devices/platform/soc/d0070000.pcie/pci0000:00/0000:00:00.0/config'
> > > > 
> > > > I looked at it deeper and found out that in PCI subsystem code is race
> > > > condition between pci_bus_add_device() and pci_sysfs_init() calls. Both
> > > > of these functions calls pci_create_sysfs_dev_files() and calling this
> > > > function more times for same pci device throws above error message.
> > > > 
> > > > There can be two different race conditions:
> > > > 
> > > > 1. pci_bus_add_device() called pcibios_bus_add_device() or
> > > > pci_fixup_device() but have not called pci_create_sysfs_dev_files() yet.
> > > > Meanwhile pci_sysfs_init() is running and pci_create_sysfs_dev_files()
> > > > was called for newly registered device. In this case function
> > > > pci_create_sysfs_dev_files() is called two times, ones from
> > > > pci_bus_add_device() and once from pci_sysfs_init().
> > > > 
> > > > 2. pci_sysfs_init() is called. It first sets sysfs_initialized to 1
> > > > which unblock calling pci_create_sysfs_dev_files(). Then another bus
> > > > registers new PCI device and calls pci_bus_add_device() which calls
> > > > pci_create_sysfs_dev_files() and registers sysfs files. Function
> > > > pci_sysfs_init() continues execution and calls function
> > > > pci_create_sysfs_dev_files() also for this newly registered device. So
> > > > pci_create_sysfs_dev_files() is again called two times.
> > > > 
> > > > 
> > > > I workaround both race conditions I created following hack patch. After
> > > > applying it I'm not getting that 'sysfs: cannot create duplicate filename'
> > > > error message anymore.
> > > > 
> > > > Can you look at it how to fix both race conditions in proper way?
> > > 
> > > Is this workaround diff enough? Or are you going to prepare
> > > something better?
> > > 
> > > Please let me know if I should send this diff as regular patch.
> 
> I'm not really a fan of this because pci_sysfs_init() is a bit of a
> hack to begin with, and this makes it even more complicated.

I understand, but the two race conditions are in this pci_sysfs_init()
function. So either it needs to be fixed or code changed, so this
function can be removed.

I'm hitting these race conditions randomly on pci aardvark controller
driver- I prepared patch which speed up initialization of this driver,
but also increase probability that it hits above race conditions :-(

> It's not obvious from the code why we need pci_sysfs_init(), but
> Yinghai hinted [1] that we need to create sysfs after assigning
> resources.  I experimented by removing pci_sysfs_init() and skipping
> the ROM BAR sizing.  In that case, we create sysfs files in
> pci_bus_add_device() and later assign space for the ROM BAR, so we
> fail to create the "rom" sysfs file.

Thank you for explanation. I think it should be documented somewhere in
the code as it is really not obvious.

> The current solution to that is to delay the sysfs files until
> pci_sysfs_init(), a late_initcall(), which runs after resource
> assignments.  But I think it would be better if we could create the
> sysfs file when we assign the BAR.  Then we could get rid of the
> late_initcall() and that implicit ordering requirement.
> 
> But I haven't tried to code it up, so it's probably more complicated
> than this.  I guess ideally we would assign all the resources before
> pci_bus_add_device().  If we could do that, we could just remove
> pci_sysfs_init() and everything would just work, but I think that's a
> HUGE can of worms.
> 
> [1] https://lore.kernel.org/linux-pci/CAE9FiQWBXHgz-gWCmpWLaBOfQQJwtRZemV6Ut9GVw_KJ-dTGTA@mail.gmail.com/
> 
> > > > diff --git a/drivers/pci/bus.c b/drivers/pci/bus.c
> > > > index 8e40b3e6da77..691be2258c4e 100644
> > > > --- a/drivers/pci/bus.c
> > > > +++ b/drivers/pci/bus.c
> > > > @@ -316,7 +316,7 @@ void pci_bus_add_device(struct pci_dev *dev)
> > > >  	 */
> > > >  	pcibios_bus_add_device(dev);
> > > >  	pci_fixup_device(pci_fixup_final, dev);
> > > > -	pci_create_sysfs_dev_files(dev);
> > > > +	pci_create_sysfs_dev_files(dev, false);
> > > >  	pci_proc_attach_device(dev);
> > > >  	pci_bridge_d3_update(dev);
> > > >  
> > > > diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> > > > index 6d78df981d41..b0c4852a51dd 100644
> > > > --- a/drivers/pci/pci-sysfs.c
> > > > +++ b/drivers/pci/pci-sysfs.c
> > > > @@ -1328,13 +1328,13 @@ static int pci_create_capabilities_sysfs(struct pci_dev *dev)
> > > >  	return retval;
> > > >  }
> > > >  
> > > > -int __must_check pci_create_sysfs_dev_files(struct pci_dev *pdev)
> > > > +int __must_check pci_create_sysfs_dev_files(struct pci_dev *pdev, bool sysfs_initializing)
> > > >  {
> > > >  	int retval;
> > > >  	int rom_size;
> > > >  	struct bin_attribute *attr;
> > > >  
> > > > -	if (!sysfs_initialized)
> > > > +	if (!sysfs_initializing && !sysfs_initialized)
> > > >  		return -EACCES;
> > > >  
> > > >  	if (pdev->cfg_size > PCI_CFG_SPACE_SIZE)
> > > > @@ -1437,18 +1437,21 @@ void pci_remove_sysfs_dev_files(struct pci_dev *pdev)
> > > >  static int __init pci_sysfs_init(void)
> > > >  {
> > > >  	struct pci_dev *pdev = NULL;
> > > > -	int retval;
> > > > +	int retval = 0;
> > > >  
> > > > -	sysfs_initialized = 1;
> > > >  	for_each_pci_dev(pdev) {
> > > > -		retval = pci_create_sysfs_dev_files(pdev);
> > > > +		if (!pci_dev_is_added(pdev))
> > > > +			continue;
> > > > +		retval = pci_create_sysfs_dev_files(pdev, true);
> > > >  		if (retval) {
> > > >  			pci_dev_put(pdev);
> > > > -			return retval;
> > > > +			goto out;
> > > >  		}
> > > >  	}
> > > >  
> > > > -	return 0;
> > > > +out:
> > > > +	sysfs_initialized = 1;
> > > > +	return retval;
> > > >  }
> > > >  late_initcall(pci_sysfs_init);
> > > >  
> > > > diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
> > > > index 6d3f75867106..304294c7171e 100644
> > > > --- a/drivers/pci/pci.h
> > > > +++ b/drivers/pci/pci.h
> > > > @@ -19,7 +19,7 @@ bool pcie_cap_has_rtctl(const struct pci_dev *dev);
> > > >  
> > > >  /* Functions internal to the PCI core code */
> > > >  
> > > > -int pci_create_sysfs_dev_files(struct pci_dev *pdev);
> > > > +int pci_create_sysfs_dev_files(struct pci_dev *pdev, bool sysfs_initializing);
> > > >  void pci_remove_sysfs_dev_files(struct pci_dev *pdev);
> > > >  #if !defined(CONFIG_DMI) && !defined(CONFIG_ACPI)
> > > >  static inline void pci_create_firmware_label_files(struct pci_dev *pdev)
> > > > 

Powered by blists - more mailing lists