Re: [PATCH V2 3/9] PCI/portdrv: Add pcie_walk_rcec() to walk RCiEPs associated with RCEC

[Sean V Kelley]

On 5 Aug 2020, at 10:43, Bjorn Helgaas wrote:

> On Tue, Aug 04, 2020 at 12:40:46PM -0700, Sean V Kelley wrote:
> > From: Qiuxu Zhuo <qiuxu.zhuo@xxxxxxxxx>
> >
> > When an RCEC device signals error(s) to a CPU core, the CPU core
> > needs to walk all the RCiEPs associated with that RCEC to check
> > errors. So add the function pcie_walk_rcec() to walk all RCiEPs
> > associated with the RCEC device.
> >
> > Co-developed-by: Sean V Kelley <sean.v.kelley@xxxxxxxxx>
> > Signed-off-by: Qiuxu Zhuo <qiuxu.zhuo@xxxxxxxxx>
> > Signed-off-by: Sean V Kelley <sean.v.kelley@xxxxxxxxx>
> > Reviewed-by: Jonathan Cameron <Jonathan.Cameron@xxxxxxxxxx>
> > ---
> >  drivers/pci/pcie/portdrv.h      |  2 +
> >  drivers/pci/pcie/portdrv_core.c | 82 
> > +++++++++++++++++++++++++++++++++
> >  2 files changed, 84 insertions(+)
> >
> > diff --git a/drivers/pci/pcie/portdrv.h b/drivers/pci/pcie/portdrv.h
> > index af7cf237432a..c11d5ecbad76 100644
> > --- a/drivers/pci/pcie/portdrv.h
> > +++ b/drivers/pci/pcie/portdrv.h
> > @@ -116,6 +116,8 @@ void pcie_port_service_unregister(struct 
> > pcie_port_service_driver *new);
> >
> >  extern struct bus_type pcie_port_bus_type;
> >  int pcie_port_device_register(struct pci_dev *dev);
> > +void pcie_walk_rcec(struct pci_dev *rcec, int (*cb)(struct pci_dev 
> > *, void *),
> > +		    void *userdata);
> >  #ifdef CONFIG_PM
> >  int pcie_port_device_suspend(struct device *dev);
> >  int pcie_port_device_resume_noirq(struct device *dev);
> > diff --git a/drivers/pci/pcie/portdrv_core.c 
> > b/drivers/pci/pcie/portdrv_core.c
> > index 5d4a400094fc..daa2dfa83a0b 100644
> > --- a/drivers/pci/pcie/portdrv_core.c
> > +++ b/drivers/pci/pcie/portdrv_core.c
> > @@ -14,6 +14,7 @@
> >  #include <linux/pm_runtime.h>
> >  #include <linux/string.h>
> >  #include <linux/slab.h>
> > +#include <linux/bitops.h>
> >  #include <linux/aer.h>
> >
> >  #include "../pci.h"
> > @@ -365,6 +366,87 @@ int pcie_port_device_register(struct pci_dev 
> > *dev)
> >  	return status;
> >  }
> >
> > +static int pcie_walk_rciep_devfn(struct pci_bus *pbus, int 
> > (*cb)(struct pci_dev *, void *),
>
> In drivers/pci, the typical name for a "struct pci_bus *" is "bus",
> and for a "struct pci_dev *" is "dev" (below).
>
> > +				 void *userdata, unsigned long bitmap)

Will correct.

> Use "u32 bitmap" so the width is explicit.  Looks like this would also
> get rid of the cast in the caller.  So maybe you used "unsigned long"
> here for some reason?

Will correct. Somehow one cast led to another.  Should have been u32.

> > +{
> > +	unsigned int dev, fn;
> > +	struct pci_dev *pdev;
> > +	int retval;
> > +
> > +	for_each_set_bit(dev, &bitmap, 32) {
> > +		for (fn = 0; fn < 8; fn++) {
> > +			pdev = pci_get_slot(pbus, PCI_DEVFN(dev, fn));
>
> This needs a matching pci_dev_put(), according to the pci_get_slot()
> function comment.

Will add.

> > +
> > +			if (!pdev || pci_pcie_type(pdev) != PCI_EXP_TYPE_RC_END)
> > +				continue;
> > +
> > +			retval = cb(pdev, userdata);
> > +			if (retval)
> > +				return retval;
> > +		}
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +/**
> > + * pcie_walk_rcec - Walk RCiEP devices associating with RCEC and 
> > call callback.
> > + * @rcec     RCEC whose RCiEP devices should be walked.
> > + * @cb       Callback to be called for each RCiEP device found.
> > + * @userdata Arbitrary pointer to be passed to callback.
> > + *
> > + * Walk the given RCEC. Call the provided callback on each RCiEP 
> > device found.
> > + *
> > + * We check the return of @cb each time. If it returns anything
> > + * other than 0, we break out.
> > + */
> > +void pcie_walk_rcec(struct pci_dev *rcec, int (*cb)(struct pci_dev 
> > *, void *),
> > +		    void *userdata)
> > +{
> > +	u32 pos, bitmap, hdr, busn;
> > +	u8 ver, nextbusn, lastbusn;
> > +	struct pci_bus *pbus;
> > +	unsigned int bnr;
> > +
> > +	pos = pci_find_ext_capability(rcec, PCI_EXT_CAP_ID_RCEC);
> > +	if (!pos)
> > +		return;
>
> I think all the registers you care about here are read-only.  If so, 
> we
> should find the capability, read the registers (bitmap & bus numbers),
> and save them at enumeration-time, e.g., in pci_init_capabilities().
> I hope we don't have to dig all this out every time we process an
> error.

I originally went that way by caching the rcec cap (pos) in pci_dev on 
probe. Will make changes to cache at enumeration time.

> > +	pbus = pci_find_bus(pci_domain_nr(rcec->bus), rcec->bus->number);
> > +	if (!pbus)
> > +		return;
>
> This seems like a really complicated way to write:
>
>   pbus = rcec->bus;
>
> "rcec->bus" cannot be NULL, so there's no need to check.

For some reason, at the time it appeared I needed to call pci_find_bus.  
But that makes sense if it is always valid. Will correct.

> > +	pci_read_config_dword(rcec, pos + PCI_RCEC_RCIEP_BITMAP, &bitmap);
> > +
> > +	/* Find RCiEP devices on the same bus as the RCEC */
> > +	if (pcie_walk_rciep_devfn(pbus, cb, userdata, (unsigned 
> > long)bitmap))
> > +		return;
> > +
> > +	/* Check whether RCEC BUSN register is present */
> > +	pci_read_config_dword(rcec, pos, &hdr);
> > +	ver = PCI_EXT_CAP_VER(hdr);
> > +	if (ver < PCI_RCEC_BUSN_REG_VER)
> > +		return;
> > +
> > +	pci_read_config_dword(rcec, pos + PCI_RCEC_BUSN, &busn);
> > +	nextbusn = PCI_RCEC_BUSN_NEXT(busn);
> > +	lastbusn = PCI_RCEC_BUSN_LAST(busn);
> > +
> > +	/* All RCiEP devices are on the same bus as the RCEC */
> > +	if (nextbusn == 0xff && lastbusn == 0x00)
> > +		return;
> > +
> > +	for (bnr = nextbusn; bnr <= lastbusn; bnr++) {
> > +		pbus = pci_find_bus(pci_domain_nr(rcec->bus), bnr);
> > +		if (!pbus)
> > +			continue;
> > +
> > +		/* Find RCiEP devices on the given bus */
> > +		if (pcie_walk_rciep_devfn(pbus, cb, userdata, 0xffffffff))
> > +			return;
> > +	}
> > +}
> > +
> >  #ifdef CONFIG_PM
> >  typedef int (*pcie_pm_callback_t)(struct pcie_device *);
> >
> > --
> > 2.27.0