Re: [PATCH V2 01/11] cxl/pci: Add generic MSI-X/MSI irq support

[Ira Weiny <ira.weiny@intel.com>]

On Thu, Dec 01, 2022 at 04:23:21PM -0800, Dan Williams wrote:
> ira.weiny@ wrote:
> > From: Davidlohr Bueso <dave@stgolabs.net>
> > 
> > Currently the only CXL features targeted for irq support require their
> > message numbers to be within the first 16 entries.  The device may
> > however support less than 16 entries depending on the support it
> > provides.
> > 
> > Attempt to allocate these 16 irq vectors.  If the device supports less
> > then the PCI infrastructure will allocate that number.
> 
> What happens if the device supports 16, but irq-core allocates less? I
> believe the answer is with the first user, but this patch does not
> include a user.
> 
> > Store the number of vectors actually allocated in the device state for
> > later use by individual functions.
> 
> The patch does not do that.

Sorry missed updating this message.

> 
> I know this patch has gone through a lot of discussion, but this
> mismatch shows it should really be squashed with the first user because
> it does not stand on its own anymore.

It is separate because it was Davidlohr's to begin with.

I'll squash it back.

> 
> > Upon successful allocation, users can plug in their respective isr at
> > any point thereafter, for example, if the irq setup is not done in the
> > PCI driver, such as the case of the CXL-PMU.
> > 
> > Cc: Bjorn Helgaas <helgaas@kernel.org>
> > Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
> > Co-developed-by: Ira Weiny <ira.weiny@intel.com>
> > Signed-off-by: Ira Weiny <ira.weiny@intel.com>
> > Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
> > 
> > ---
> > Changes from V1:
> > 	Jonathan
> > 		pci_alloc_irq_vectors() cleans up the vectors automatically
> > 		use msi_enabled rather than nr_irq_vecs
> > 
> > Changes from Ira
> > 	Remove reviews
> > 	Allocate up to a static 16 vectors.
> > 	Change cover letter
> > ---
> >  drivers/cxl/cxlmem.h |  3 +++
> >  drivers/cxl/cxlpci.h |  6 ++++++
> >  drivers/cxl/pci.c    | 23 +++++++++++++++++++++++
> >  3 files changed, 32 insertions(+)
> > 
> > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > index 88e3a8e54b6a..cd35f43fedd4 100644
> > --- a/drivers/cxl/cxlmem.h
> > +++ b/drivers/cxl/cxlmem.h
> > @@ -211,6 +211,7 @@ struct cxl_endpoint_dvsec_info {
> >   * @info: Cached DVSEC information about the device.
> >   * @serial: PCIe Device Serial Number
> >   * @doe_mbs: PCI DOE mailbox array
> > + * @msi_enabled: MSI-X/MSI has been enabled
> >   * @mbox_send: @dev specific transport for transmitting mailbox commands
> >   *
> >   * See section 8.2.9.5.2 Capacity Configuration and Label Storage for
> > @@ -247,6 +248,8 @@ struct cxl_dev_state {
> >  
> >  	struct xarray doe_mbs;
> >  
> > +	bool msi_enabled;
> > +
> 
> This goes unused in this patch and it also duplicates what the core
> offers with pdev->{msi,msix}_enabled.

I tried to argue that with Jonathan and lost.  What I had in V1 was to store
the number actually allocated.  Then if a function reports something higher
later it can't be used.

I admit that at this point I really don't understand PCI interrupts at all.
Every time this patch is discussed I get (what is to me) confusing information.
And I've been unable to discern from the spec how exactly this is all supposed
to work.

> 
> >  	int (*mbox_send)(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd);
> >  };
> >  
> > diff --git a/drivers/cxl/cxlpci.h b/drivers/cxl/cxlpci.h
> > index eec597dbe763..b7f4e2f417d3 100644
> > --- a/drivers/cxl/cxlpci.h
> > +++ b/drivers/cxl/cxlpci.h
> > @@ -53,6 +53,12 @@
> >  #define	    CXL_DVSEC_REG_LOCATOR_BLOCK_ID_MASK			GENMASK(15, 8)
> >  #define     CXL_DVSEC_REG_LOCATOR_BLOCK_OFF_LOW_MASK		GENMASK(31, 16)
> >  
> > +/*
> > + * NOTE: Currently all the functions which are enabled for CXL require their
> > + * vectors to be in the first 16.  Use this as the max.
> > + */
> > +#define CXL_PCI_REQUIRED_VECTORS 16
> > +
> >  /* Register Block Identifier (RBI) */
> >  enum cxl_regloc_type {
> >  	CXL_REGLOC_RBI_EMPTY = 0,
> > diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
> > index faeb5d9d7a7a..8f86f85d89c7 100644
> > --- a/drivers/cxl/pci.c
> > +++ b/drivers/cxl/pci.c
> > @@ -428,6 +428,27 @@ static void devm_cxl_pci_create_doe(struct cxl_dev_state *cxlds)
> >  	}
> >  }
> >  
> > +static void cxl_pci_alloc_irq_vectors(struct cxl_dev_state *cxlds)
> > +{
> > +	struct device *dev = cxlds->dev;
> > +	struct pci_dev *pdev = to_pci_dev(dev);
> > +	int nvecs;
> > +
> > +	/*
> > +	 * NOTE: pci_alloc_irq_vectors() handles calling pci_free_irq_vectors()
> > +	 * automatically despite not being called pcim_*.  See
> > +	 * pci_setup_msi_context().
> > +	 */
> > +	nvecs = pci_alloc_irq_vectors(pdev, 1, CXL_PCI_REQUIRED_VECTORS,
> > +				   PCI_IRQ_MSIX | PCI_IRQ_MSI);
> 
> clang-format would scooch that second line in for you.
> 
> Might also be worth a comment for the next person that goes looking for
> why this isn't PCI_IRQ_ALL_TYPES.
> 
> From CXL 3.0 3.1.1 CXL.io Endpoint:
> A Function on a CXL device must not generate INTx messages if that
> Function participates in CXL.cache protocol or CXL.mem protocols.

Seems reasonable.

Ira

> 
> 
> > +	if (nvecs < 0) {
> > +		dev_dbg(dev, "Failed to alloc irq vectors; use polling instead.\n");
> > +		return;
> > +	}
> > +
> > +	cxlds->msi_enabled = true;
> > +}
> > +
> >  static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> >  {
> >  	struct cxl_register_map map;
> > @@ -494,6 +515,8 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> >  	if (rc)
> >  		return rc;
> >  
> > +	cxl_pci_alloc_irq_vectors(cxlds);
> > +
> >  	cxlmd = devm_cxl_add_memdev(cxlds);
> >  	if (IS_ERR(cxlmd))
> >  		return PTR_ERR(cxlmd);
> > -- 
> > 2.37.2
> > 
> 
>