Re: [PATCH v2 10/11] PCI: Stop caching ATS Invalidate Queue Depth

[Bjorn Helgaas <bhelgaas@google.com>]

Hi Don,

On Mon, Jul 27, 2015 at 10:00:53AM -0400, Don Dutile wrote:
> On 07/20/2015 08:15 PM, Bjorn Helgaas wrote:
> >Stop caching the Invalidate Queue Depth in struct pci_dev.
> >pci_ats_queue_depth() is typically called only once per device, and it
> >returns a fixed value per-device, so callers who need the value frequently
> >can cache it themselves.
> >
> >Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
> >---
> >  drivers/pci/ats.c   |    9 ++++-----
> >  include/linux/pci.h |    1 -
> >  2 files changed, 4 insertions(+), 6 deletions(-)
> >
> >diff --git a/drivers/pci/ats.c b/drivers/pci/ats.c
> >index 67524a7..bdb1383 100644
> >--- a/drivers/pci/ats.c
> >+++ b/drivers/pci/ats.c
> >@@ -20,16 +20,12 @@
> >  void pci_ats_init(struct pci_dev *dev)
> >  {
> >  	int pos;
> >-	u16 cap;
> >
> >  	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ATS);
> >  	if (!pos)
> >  		return;
> >
> >  	dev->ats_cap = pos;
> >-	pci_read_config_word(dev, dev->ats_cap + PCI_ATS_CAP, &cap);
> >-	dev->ats_qdep = PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) :
> >-					    PCI_ATS_MAX_QDEP;
> >  }
> >
> >  /**
> >@@ -131,13 +127,16 @@ EXPORT_SYMBOL_GPL(pci_restore_ats_state);
> >   */
> >  int pci_ats_queue_depth(struct pci_dev *dev)
> >  {
> >+	u16 cap;
> >+
> >  	if (!dev->ats_cap)
> >  		return -EINVAL;
> >
> >  	if (dev->is_virtfn)
> >  		return 0;
> >
> hmmm, isn't one of the fixes in this patch set to
> change the caching of ats queue depth ?  won't above
> make it 0 for all VF's ?  is that correct, or desired (virtual) value?

Per spec (SR-IOV r1.1., sec 3.7.4), the ATS queue depth register on a VF
always contains zero.

Here's the v4.1 code:

    int pci_ats_queue_depth(struct pci_dev *dev)
    {
        if (dev->is_virtfn)
                return 0;

        if (dev->ats)
                return dev->ats->qdep;

	...


In v4.1, pci_ats_queue_depth() always returned 0 for a VF.  For VFs, we
didn't look at the dev->ats->qdep cache.  So I don't think this changes
anything for the caller.

A previous patch changed this path so we return -EINVAL instead of 0 for
VFs that don't support ATS.  I think the previous behavior there was wrong,
but I doubt anybody noticed.

Bjorn

> >-	return dev->ats_qdep;
> >+	pci_read_config_word(dev, dev->ats_cap + PCI_ATS_CAP, &cap);
> >+	return PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) : PCI_ATS_MAX_QDEP;
> >  }
> >  EXPORT_SYMBOL_GPL(pci_ats_queue_depth);
> >
> >diff --git a/include/linux/pci.h b/include/linux/pci.h
> >index 307f96a..4b484fd 100644
> >--- a/include/linux/pci.h
> >+++ b/include/linux/pci.h
> >@@ -378,7 +378,6 @@ struct pci_dev {
> >  	};
> >  	u16		ats_cap;	/* ATS Capability offset */
> >  	u8		ats_stu;	/* ATS Smallest Translation Unit */
> >-	u8		ats_qdep;	/* ATS Invalidate Queue Depth */
> >  	atomic_t	ats_ref_cnt;	/* number of VFs with ATS enabled */
> >  #endif
> >  	phys_addr_t rom; /* Physical address of ROM if it's not from the BAR */
> >
> >--
> >To unsubscribe from this list: send the line "unsubscribe linux-pci" in
> >the body of a message to majordomo@vger.kernel.org
> >More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >
>