Message-ID: <20170617174543.GA1723@westernst>
Date:   Sat, 17 Jun 2017 18:45:44 +0100
From:   Piotr Gregor <piotrgregor@...ncme.org>
To:     Bjorn Helgaas <helgaas@...nel.org>
Cc:     Alex Williamson <alex.williamson@...hat.com>,
        Bjorn Helgaas <bhelgaas@...gle.com>,
        "Michael S. Tsirkin" <mst@...hat.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Neo Jia <cjia@...dia.com>,
        Kirti Wankhede <kwankhede@...dia.com>,
        Vlad Tsyrklevich <vlad@...rklevich.net>,
        Arvind Yadav <arvind.yadav.cs@...il.com>,
        Yongji Xie <xyjxie@...ux.vnet.ibm.com>,
        linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org,
        kvm@...r.kernel.org
Subject: Re: [PATCH] PCI: Move test of INTx masking to pci_setup_device

Hi Bjorn,

The pci_cfg_access_lock is most likely not needed there.
The assignment by return type is indeed preferred in this case.

However, you have changed the meaning of returned boolean information
by pci_intx_mask_broken leaving pci_intx_mask_supported unchanged.
The test should be: 

    if (new != toggle) /* the test failed */
	        return 1;
	return 0;

I have attached patch (created on pci/enumeration branch,
should I commit this to your repo too?).

Regarding v2.3 - do you think it is worth to apply the check
so we would have something like

    if ((new == toggle) || PCI_VERSION_PRIOR_TO_23) /* test OK or PCI prior to r2.3 */
	        return 0;
	return 1;

cheers,
Piotr


On Fri, Jun 16, 2017 at 05:54:46PM -0500, Bjorn Helgaas wrote:
> On Fri, May 26, 2017 at 10:02:25PM +0100, Piotr Gregor wrote:
> > The test for INTx masking via config space command performed
> > in pci_intx_mask_supported() should be performed before PCI device
> > can be used. This is to avoid reading/writing of PCI_COMMAND_INTX_DISABLE
> > register which may collide with MSI/MSI-X interrupts.
> > 
> > This patch moves test performed in pci_intx_mask_supported() to
> > 
> > static void pci_test_intx_masking(struct pci_dev *dev)
> > 
> > defined in drivers/pci/probe.c.
> > 
> > This function is called from pci_setup_device(). It skips the test
> > if the device has been already marked to have broken INTx masking
> > feature. Otherwise the test is executed and broken_intx_masking
> > field of struct pci_dev is set accordingly. broken_intx_masking
> > meaning is: if it is true then the test has been either skipped
> > because the device has been already known to have broken INTx
> > masking support, or the test's been done and it has detected INTx
> > masking support to be broken.
> > The test result can be queried at any time later from the pci_dev
> > using same interface as before (though whith changed implementation)
> > 
> > static inline bool pci_intx_mask_supported(struct pci_dev *pdev)
> > {
> >         /*
> >          * INTx masking is supported if device has not been marked
> >          * to have this feature broken and it has passed
> >          * pci_test_intx_masking() test.
> >          */
> >         return !pdev->broken_intx_masking;
> > }
> > 
> > so current users of pci_intx_mask_supported: uio and vfio, keep
> > their code unchanged.
> > 
> > Signed-off-by: Piotr Gregor <piotrgregor@...ncme.org>
> > ---
> >  drivers/pci/pci.c   | 42 +-----------------------------------------
> >  drivers/pci/probe.c | 44 ++++++++++++++++++++++++++++++++++++++++++++
> >  include/linux/pci.h | 13 +++++++++++--
> >  3 files changed, 56 insertions(+), 43 deletions(-)
> > 
> > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> > index b01bd5b..7c4e1aa 100644
> > --- a/drivers/pci/pci.c
> > +++ b/drivers/pci/pci.c
> > @@ -3708,46 +3708,6 @@ void pci_intx(struct pci_dev *pdev, int enable)
> >  }
> >  EXPORT_SYMBOL_GPL(pci_intx);
> >  
> > -/**
> > - * pci_intx_mask_supported - probe for INTx masking support
> > - * @dev: the PCI device to operate on
> > - *
> > - * Check if the device dev support INTx masking via the config space
> > - * command word.
> > - */
> > -bool pci_intx_mask_supported(struct pci_dev *dev)
> > -{
> > -	bool mask_supported = false;
> > -	u16 orig, new;
> > -
> > -	if (dev->broken_intx_masking)
> > -		return false;
> > -
> > -	pci_cfg_access_lock(dev);
> > -
> > -	pci_read_config_word(dev, PCI_COMMAND, &orig);
> > -	pci_write_config_word(dev, PCI_COMMAND,
> > -			      orig ^ PCI_COMMAND_INTX_DISABLE);
> > -	pci_read_config_word(dev, PCI_COMMAND, &new);
> > -
> > -	/*
> > -	 * There's no way to protect against hardware bugs or detect them
> > -	 * reliably, but as long as we know what the value should be, let's
> > -	 * go ahead and check it.
> > -	 */
> > -	if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) {
> > -		dev_err(&dev->dev, "Command register changed from 0x%x to 0x%x: driver or hardware bug?\n",
> > -			orig, new);
> > -	} else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) {
> > -		mask_supported = true;
> > -		pci_write_config_word(dev, PCI_COMMAND, orig);
> > -	}
> > -
> > -	pci_cfg_access_unlock(dev);
> > -	return mask_supported;
> > -}
> > -EXPORT_SYMBOL_GPL(pci_intx_mask_supported);
> > -
> >  static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
> >  {
> >  	struct pci_bus *bus = dev->bus;
> > @@ -3798,7 +3758,7 @@ static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
> >   * @dev: the PCI device to operate on
> >   *
> >   * Check if the device dev has its INTx line asserted, mask it and
> > - * return true in that case. False is returned if not interrupt was
> > + * return true in that case. False is returned if no interrupt was
> >   * pending.
> >   */
> >  bool pci_check_and_mask_intx(struct pci_dev *dev)
> > diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> > index 19c8950..ee6b55c 100644
> > --- a/drivers/pci/probe.c
> > +++ b/drivers/pci/probe.c
> > @@ -1330,6 +1330,48 @@ static void pci_msi_setup_pci_dev(struct pci_dev *dev)
> >  }
> >  
> >  /**
> > + * pci_test_intx_masking - probe for INTx masking support
> > + * @dev: the PCI device to operate on
> > + *
> > + * Check if the @dev supports INTx masking via the config space
> > + * command word. Executed when PCI device is setup. Result is saved
> > + * in broken_intx_masking field of struct pci_dev and can be checked
> > + * with pci_intx_mask_supported at any time later, after the PCI device
> > + * has been setup (this avoids testing of PCI_COMMAND_INTX_DISABLE
> > + * register at runtime).
> > + */
> > +static void pci_test_intx_masking(struct pci_dev *dev)
> > +{
> > +	u16 orig, toggle, new;
> > +
> > +	/*
> > +	 * If device doesn't support this feature though it could pass the test.
> > +	 */
> > +	if (dev->broken_intx_masking)
> > +		return;
> 
> I don't think we need this check anymore, because broken_intx_masking
> is only set by "FINAL" quirks, which run long after this code.
> 
> > +
> > +	pci_cfg_access_lock(dev);
> 
> I don't think we need this locking either.  This lock protects against
> user-space config access, and the device isn't accessible to user
> space yet.
> 
> > +	/*
> > +	 * Perform the test.
> > +	 */
> > +	pci_read_config_word(dev, PCI_COMMAND, &orig);
> > +	toggle = orig ^ PCI_COMMAND_INTX_DISABLE;
> > +	pci_write_config_word(dev, PCI_COMMAND, toggle);
> > +	pci_read_config_word(dev, PCI_COMMAND, &new);
> > +
> > +	/*
> > +	 * Restore initial state.
> > +	 */
> > +	pci_write_config_word(dev, PCI_COMMAND, orig);
> > +
> > +	pci_cfg_access_unlock(dev);
> > +
> > +	if (new != toggle)
> > +		dev->broken_intx_masking = 1;
> > +}
> > +
> > +/**
> >   * pci_setup_device - fill in class and map information of a device
> >   * @dev: the device structure to fill
> >   *
> > @@ -1399,6 +1441,8 @@ int pci_setup_device(struct pci_dev *dev)
> >  		}
> >  	}
> >  
> > +	pci_test_intx_masking(dev);
> 
> I think it's nicer to read the code if the assignment isn't hidden
> away, e.g., something like this:
> 
>   dev->broken_intx_masking = pci_test_intx_masking(dev);
> 
> That's more like what we do for dev->cfg_size and similar things.
> 
> I provisionally made these changes and attached the updated patch below.
> Let me know if it looks OK.
> 
> This is on my pci/enumeration branch, intended for v4.13.
> 
> >  	switch (dev->hdr_type) {		    /* header type */
> >  	case PCI_HEADER_TYPE_NORMAL:		    /* standard header */
> >  		if (class == PCI_CLASS_BRIDGE_PCI)
> > diff --git a/include/linux/pci.h b/include/linux/pci.h
> > index 33c2b0b..8a307fc 100644
> > --- a/include/linux/pci.h
> > +++ b/include/linux/pci.h
> > @@ -366,7 +366,7 @@ struct pci_dev {
> >  	unsigned int	is_thunderbolt:1; /* Thunderbolt controller */
> >  	unsigned int    __aer_firmware_first_valid:1;
> >  	unsigned int	__aer_firmware_first:1;
> > -	unsigned int	broken_intx_masking:1;
> > +	unsigned int	broken_intx_masking:1; /* INTx masking can't be used */
> >  	unsigned int	io_window_1k:1;	/* Intel P2P bridge 1K I/O windows */
> >  	unsigned int	irq_managed:1;
> >  	unsigned int	has_secondary_link:1;
> > @@ -1003,6 +1003,16 @@ int __must_check pci_reenable_device(struct pci_dev *);
> >  int __must_check pcim_enable_device(struct pci_dev *pdev);
> >  void pcim_pin_device(struct pci_dev *pdev);
> >  
> > +static inline bool pci_intx_mask_supported(struct pci_dev *pdev)
> > +{
> > +	/*
> > +	 * INTx masking is supported if device has not been marked
> > +	 * to have this feature broken and it has passed
> > +	 * pci_test_intx_masking() test.
> > +	 */
> > +	return !pdev->broken_intx_masking;
> > +}
> > +
> >  static inline int pci_is_enabled(struct pci_dev *pdev)
> >  {
> >  	return (atomic_read(&pdev->enable_cnt) > 0);
> > @@ -1026,7 +1036,6 @@ int __must_check pci_set_mwi(struct pci_dev *dev);
> >  int pci_try_set_mwi(struct pci_dev *dev);
> >  void pci_clear_mwi(struct pci_dev *dev);
> >  void pci_intx(struct pci_dev *dev, int enable);
> > -bool pci_intx_mask_supported(struct pci_dev *dev);
> >  bool pci_check_and_mask_intx(struct pci_dev *dev);
> >  bool pci_check_and_unmask_intx(struct pci_dev *dev);
> >  int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask);
> 
> 
> commit 73ab6bc006ba966e233785577a1742c6cace9c8a
> Author: Piotr Gregor <piotrgregor@...ncme.org>
> Date:   Fri May 26 22:02:25 2017 +0100
> 
>     PCI: Test INTx masking during enumeration, not at run-time
>     
>     The test for INTx masking via PCI_COMMAND_INTX_DISABLE performed in
>     pci_intx_mask_supported() should be done before the device can be used.
>     This is to avoid writing PCI_COMMAND while the driver owns the device, in
>     case that has any effect on MSI/MSI-X interrupts.
>     
>     Move the content of pci_intx_mask_supported() to pci_intx_mask_broken() and
>     call it from pci_setup_device().
>     
>     The test result can be queried at any time later using the same
>     pci_intx_mask_supported() interface as before (though with changed
>     implementation), so callers (uio, vfio) should be unaffected.
>     
>     Signed-off-by: Piotr Gregor <piotrgregor@...ncme.org>
>     [bhelgaas: changelog, remove quirk check, remove locking, move
>     dev->broken_intx_masking assignment to caller]
>     Signed-off-by: Bjorn Helgaas <bhelgaas@...gle.com>
>     Reviewed-by: Alex Williamson <alex.williamson@...hat.com>
>     Acked-by: Michael S. Tsirkin <mst@...hat.com>
> 
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index b01bd5bba8e6..7c4e1aa67c67 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -3708,46 +3708,6 @@ void pci_intx(struct pci_dev *pdev, int enable)
>  }
>  EXPORT_SYMBOL_GPL(pci_intx);
>  
> -/**
> - * pci_intx_mask_supported - probe for INTx masking support
> - * @dev: the PCI device to operate on
> - *
> - * Check if the device dev support INTx masking via the config space
> - * command word.
> - */
> -bool pci_intx_mask_supported(struct pci_dev *dev)
> -{
> -	bool mask_supported = false;
> -	u16 orig, new;
> -
> -	if (dev->broken_intx_masking)
> -		return false;
> -
> -	pci_cfg_access_lock(dev);
> -
> -	pci_read_config_word(dev, PCI_COMMAND, &orig);
> -	pci_write_config_word(dev, PCI_COMMAND,
> -			      orig ^ PCI_COMMAND_INTX_DISABLE);
> -	pci_read_config_word(dev, PCI_COMMAND, &new);
> -
> -	/*
> -	 * There's no way to protect against hardware bugs or detect them
> -	 * reliably, but as long as we know what the value should be, let's
> -	 * go ahead and check it.
> -	 */
> -	if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) {
> -		dev_err(&dev->dev, "Command register changed from 0x%x to 0x%x: driver or hardware bug?\n",
> -			orig, new);
> -	} else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) {
> -		mask_supported = true;
> -		pci_write_config_word(dev, PCI_COMMAND, orig);
> -	}
> -
> -	pci_cfg_access_unlock(dev);
> -	return mask_supported;
> -}
> -EXPORT_SYMBOL_GPL(pci_intx_mask_supported);
> -
>  static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
>  {
>  	struct pci_bus *bus = dev->bus;
> @@ -3798,7 +3758,7 @@ static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
>   * @dev: the PCI device to operate on
>   *
>   * Check if the device dev has its INTx line asserted, mask it and
> - * return true in that case. False is returned if not interrupt was
> + * return true in that case. False is returned if no interrupt was
>   * pending.
>   */
>  bool pci_check_and_mask_intx(struct pci_dev *dev)
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index 19c8950c6c38..28c57b31261c 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -1329,6 +1329,34 @@ static void pci_msi_setup_pci_dev(struct pci_dev *dev)
>  		pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
>  }
>  
> +/**
> + * pci_intx_mask_broken - test PCI_COMMAND_INTX_DISABLE writability
> + * @dev: PCI device
> + *
> + * Test whether PCI_COMMAND_INTX_DISABLE is writable for @dev.  Check this
> + * at enumeration-time to avoid modifying PCI_COMMAND at run-time.
> + */
> +static int pci_intx_mask_broken(struct pci_dev *dev)
> +{
> +	u16 orig, toggle, new;
> +
> +	pci_read_config_word(dev, PCI_COMMAND, &orig);
> +	toggle = orig ^ PCI_COMMAND_INTX_DISABLE;
> +	pci_write_config_word(dev, PCI_COMMAND, toggle);
> +	pci_read_config_word(dev, PCI_COMMAND, &new);
> +
> +	pci_write_config_word(dev, PCI_COMMAND, orig);
> +
> +	/*
> +	 * PCI_COMMAND_INTX_DISABLE was reserved and read-only prior to PCI
> +	 * r2.3, so strictly speaking, a device is not *broken* if it's not
> +	 * writable.
> +	 */
> +	if (new == toggle)
> +		return 1;
> +	return 0;
> +}
> +
>  /**
>   * pci_setup_device - fill in class and map information of a device
>   * @dev: the device structure to fill
> @@ -1399,6 +1427,8 @@ int pci_setup_device(struct pci_dev *dev)
>  		}
>  	}
>  
> +	dev->broken_intx_masking = pci_intx_mask_broken(dev);
> +
>  	switch (dev->hdr_type) {		    /* header type */
>  	case PCI_HEADER_TYPE_NORMAL:		    /* standard header */
>  		if (class == PCI_CLASS_BRIDGE_PCI)
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index 33c2b0b77429..4f0613d5d2d9 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -366,7 +366,7 @@ struct pci_dev {
>  	unsigned int	is_thunderbolt:1; /* Thunderbolt controller */
>  	unsigned int    __aer_firmware_first_valid:1;
>  	unsigned int	__aer_firmware_first:1;
> -	unsigned int	broken_intx_masking:1;
> +	unsigned int	broken_intx_masking:1; /* INTx masking can't be used */
>  	unsigned int	io_window_1k:1;	/* Intel P2P bridge 1K I/O windows */
>  	unsigned int	irq_managed:1;
>  	unsigned int	has_secondary_link:1;
> @@ -1003,6 +1003,15 @@ int __must_check pci_reenable_device(struct pci_dev *);
>  int __must_check pcim_enable_device(struct pci_dev *pdev);
>  void pcim_pin_device(struct pci_dev *pdev);
>  
> +static inline bool pci_intx_mask_supported(struct pci_dev *pdev)
> +{
> +	/*
> +	 * INTx masking is supported if PCI_COMMAND_INTX_DISABLE is
> +	 * writable and no quirk has marked the feature broken.
> +	 */
> +	return !pdev->broken_intx_masking;
> +}
> +
>  static inline int pci_is_enabled(struct pci_dev *pdev)
>  {
>  	return (atomic_read(&pdev->enable_cnt) > 0);
> @@ -1026,7 +1035,6 @@ int __must_check pci_set_mwi(struct pci_dev *dev);
>  int pci_try_set_mwi(struct pci_dev *dev);
>  void pci_clear_mwi(struct pci_dev *dev);
>  void pci_intx(struct pci_dev *dev, int enable);
> -bool pci_intx_mask_supported(struct pci_dev *dev);
>  bool pci_check_and_mask_intx(struct pci_dev *dev);
>  bool pci_check_and_unmask_intx(struct pci_dev *dev);
>  int pci_wait_for_pending(struct pci_dev *dev, int pos, u16 mask);

View attachment "0001-PCI-Test-INTx-masking-during-enumeration-not-at-run-.patch" of type "text/x-diff" (6207 bytes)

Powered by blists - more mailing lists