Re: [PATCH 2/3 v2] iommu/fsl: Enable default DMA window for PCIedevices

[Alex Williamson]

On Tue, 2013-12-03 at 18:15 +0000, Varun Sethi wrote:
> 
> > -----Original Message-----
> > From: Alex Williamson [mailto:alex.williamson@xxxxxxxxxx]
> > Sent: Tuesday, December 03, 2013 3:04 AM
> > To: Sethi Varun-B16395
> > Cc: joro@xxxxxxxxxx; iommu@xxxxxxxxxxxxxxxxxxxxxxxxxx; linuxppc-
> > dev@xxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Yoder Stuart-B08248;
> > Wood Scott-B07421; Bhushan Bharat-R65777
> > Subject: Re: [PATCH 2/3 v2] iommu/fsl: Enable default DMA window for PCIe
> > devices
> > 
> > On Wed, 2013-10-16 at 16:53 +0530, Varun Sethi wrote:
> > > Once the PCIe device assigned to a guest VM (via VFIO) gets detached
> > > from the iommu domain (when guest terminates), its PAMU table entry is
> > > disabled. So, this would prevent the device from being used once it's
> > assigned back to the host.
> > >
> > > This patch allows for creation of a default DMA window corresponding
> > > to the device and subsequently enabling the PAMU table entry. Before
> > > we enable the entry, we ensure that the device's bus master capability
> > is disabled (device quiesced).
> > >
> > > Signed-off-by: Varun Sethi <Varun.Sethi@xxxxxxxxxxxxx>
> > > ---
> > >  drivers/iommu/fsl_pamu.c        |   43 ++++++++++++++++++++++++++++---
> > -----
> > >  drivers/iommu/fsl_pamu.h        |    1 +
> > >  drivers/iommu/fsl_pamu_domain.c |   46
> > ++++++++++++++++++++++++++++++++++++---
> > >  3 files changed, 78 insertions(+), 12 deletions(-)
> > >
> > > diff --git a/drivers/iommu/fsl_pamu.c b/drivers/iommu/fsl_pamu.c index
> > > cba0498..fb4a031 100644
> > > --- a/drivers/iommu/fsl_pamu.c
> > > +++ b/drivers/iommu/fsl_pamu.c
> > > @@ -225,6 +225,21 @@ static struct paace *pamu_get_spaace(struct paace
> > *paace, u32 wnum)
> > >  	return spaace;
> > >  }
> > >
> > > +/*
> > > + * Defaul PPAACE settings for an LIODN.
> > > + */
> > > +static void setup_default_ppaace(struct paace *ppaace) {
> > > +	pamu_init_ppaace(ppaace);
> > > +	/* window size is 2^(WSE+1) bytes */
> > > +	set_bf(ppaace->addr_bitfields, PPAACE_AF_WSE, 35);
> > > +	ppaace->wbah = 0;
> > > +	set_bf(ppaace->addr_bitfields, PPAACE_AF_WBAL, 0);
> > > +	set_bf(ppaace->impl_attr, PAACE_IA_ATM,
> > > +		PAACE_ATM_NO_XLATE);
> > > +	set_bf(ppaace->addr_bitfields, PAACE_AF_AP,
> > > +		PAACE_AP_PERMS_ALL);
> > > +}
> > >  /**
> > >   * pamu_get_fspi_and_allocate() - Allocates fspi index and reserves
> > subwindows
> > >   *                                required for primary PAACE in the
> > secondary
> > > @@ -253,6 +268,24 @@ static unsigned long
> > pamu_get_fspi_and_allocate(u32 subwin_cnt)
> > >  	return (spaace_addr - (unsigned long)spaact) / (sizeof(struct
> > > paace));  }
> > >
> > > +/* Reset the PAACE entry to the default state */ void
> > > +enable_default_dma_window(int liodn) {
> > > +	struct paace *ppaace;
> > > +
> > > +	ppaace = pamu_get_ppaace(liodn);
> > > +	if (!ppaace) {
> > > +		pr_debug("Invalid liodn entry\n");
> > > +		return;
> > > +	}
> > > +
> > > +	memset(ppaace, 0, sizeof(struct paace));
> > > +
> > > +	setup_default_ppaace(ppaace);
> > > +	mb();
> > > +	pamu_enable_liodn(liodn);
> > > +}
> > > +
> > >  /* Release the subwindows reserved for a particular LIODN */  void
> > > pamu_free_subwins(int liodn)  { @@ -752,15 +785,7 @@ static void
> > > __init setup_liodns(void)
> > >  				continue;
> > >  			}
> > >  			ppaace = pamu_get_ppaace(liodn);
> > > -			pamu_init_ppaace(ppaace);
> > > -			/* window size is 2^(WSE+1) bytes */
> > > -			set_bf(ppaace->addr_bitfields, PPAACE_AF_WSE, 35);
> > > -			ppaace->wbah = 0;
> > > -			set_bf(ppaace->addr_bitfields, PPAACE_AF_WBAL, 0);
> > > -			set_bf(ppaace->impl_attr, PAACE_IA_ATM,
> > > -				PAACE_ATM_NO_XLATE);
> > > -			set_bf(ppaace->addr_bitfields, PAACE_AF_AP,
> > > -				PAACE_AP_PERMS_ALL);
> > > +			setup_default_ppaace(ppaace);
> > >  			if (of_device_is_compatible(node, "fsl,qman-portal"))
> > >  				setup_qbman_paace(ppaace, QMAN_PORTAL_PAACE);
> > >  			if (of_device_is_compatible(node, "fsl,qman")) diff --
> > git
> > > a/drivers/iommu/fsl_pamu.h b/drivers/iommu/fsl_pamu.h index
> > > 8fc1a12..0edcbbbb 100644
> > > --- a/drivers/iommu/fsl_pamu.h
> > > +++ b/drivers/iommu/fsl_pamu.h
> > > @@ -406,5 +406,6 @@ void get_ome_index(u32 *omi_index, struct device
> > > *dev);  int  pamu_update_paace_stash(int liodn, u32 subwin, u32
> > > value);  int pamu_disable_spaace(int liodn, u32 subwin);
> > >  u32 pamu_get_max_subwin_cnt(void);
> > > +void enable_default_dma_window(int liodn);
> > >
> > >  #endif  /* __FSL_PAMU_H */
> > > diff --git a/drivers/iommu/fsl_pamu_domain.c
> > > b/drivers/iommu/fsl_pamu_domain.c index 966ae70..dd6cafc 100644
> > > --- a/drivers/iommu/fsl_pamu_domain.c
> > > +++ b/drivers/iommu/fsl_pamu_domain.c
> > > @@ -340,17 +340,57 @@ static inline struct device_domain_info
> > *find_domain(struct device *dev)
> > >  	return dev->archdata.iommu_domain;
> > >  }
> > >
> > > +/* Disable device DMA capability and enable default DMA window */
> > > +static void disable_device_dma(struct device_domain_info *info,
> > > +				int enable_dma_window)
> > 
> > nit, bool?
> > 
> > > +{
> > > +#ifdef CONFIG_PCI
> > > +	if (info->dev->bus == &pci_bus_type) {
> > > +		struct pci_dev *pdev = NULL;
> > 
> > nit, unnecessary initialization
> > 
> > > +		pdev = to_pci_dev(info->dev);
> > > +		if (pci_is_enabled(pdev))
> > > +			pci_disable_device(pdev);
> > 
> > This looks suspicious.  What's the case where you're finding devices
> > where this is needed?  Logically the driver that called
> > pci_enable_device() should also call pci_disabled_device() when it's
> > done.  Maybe there's a case to be made that we can't rely on the driver,
> > but then maybe there should be a pr_debug/info here to notify that
> > somebody left the device running.  There might also be the question of
> > whether you should test the busmaster bit in the command register rather
> > than trusting these indirect paths if it's really for cleanup.  Thanks,
> > 
> Once the device is detached from the domain, we enable a default DMA
> window for the device. Before enabling the default window, I want to
> ensure that device DMA is disabled. VFIO subsystem handles this for
> vfio-pci. But, can we assume that this would always be the case?

That's what I'm asking.  It seems like pci_enable/disable_device is part
of the driver API and every enable should have a matching disable.  If
this is to catch cases where the disable never happened, then there
should be some indication that the previous driver did something wrong
and we should have some sort of printed warning.  Then there's also the
question of whether this fully does what you want it to.  This is an
indirect way of checking the busmaster bit in the command register.  If
you really want to make sure it's disabled, is it sufficient or should
we be reading the actual command register?  Thanks,

Alex


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/