Re: [PATCH v2 1/2] hw/usb/hcd-xhci-pci: Make PCI device more configurable

[Nicholas Piggin]

On Thu Dec 12, 2024 at 8:41 PM AEST, Phil Dennis-Jordan wrote:
> Hey Nicholas,
>
> I'm not an XHCI & PCI expert (yet?) so apologies if I've got some of this
> wrong, but I've asked some questions and made some comments inline:

Hey Phil,

Thanks for the review, looks like you are the expert now :)

>
> On Thu, 12 Dec 2024 at 09:52, Nicholas Piggin <npiggin@gmail.com> wrote:
>
> > To prepare to support another USB PCI Host Controller, make some PCI
> > configuration dynamic.
> >
> > Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
> > ---
> >  hw/usb/hcd-xhci-pci.h |  9 ++++++
> >  hw/usb/hcd-xhci-nec.c | 10 +++++++
> >  hw/usb/hcd-xhci-pci.c | 69 ++++++++++++++++++++++++++++++++++++-------
> >  3 files changed, 78 insertions(+), 10 deletions(-)
> >
> > diff --git a/hw/usb/hcd-xhci-pci.h b/hw/usb/hcd-xhci-pci.h
> > index 08f70ce97cc..213076aabf6 100644
> > --- a/hw/usb/hcd-xhci-pci.h
> > +++ b/hw/usb/hcd-xhci-pci.h
> > @@ -40,6 +40,15 @@ typedef struct XHCIPciState {
> >      XHCIState xhci;
> >      OnOffAuto msi;
> >      OnOffAuto msix;
> > +    uint8_t cache_line_size;
> > +    uint8_t pm_cap_off;
> > +    uint8_t pcie_cap_off;
> > +    uint8_t msi_cap_off;
> > +    uint8_t msix_cap_off;
> > +    int msix_bar_nr;
> > +    uint64_t msix_bar_size;
> > +    uint32_t msix_table_off;
> > +    uint32_t msix_pba_off;
> >  } XHCIPciState;
> >
> >  #endif
> > diff --git a/hw/usb/hcd-xhci-nec.c b/hw/usb/hcd-xhci-nec.c
> > index 0e61c6c4f06..6ac1dc7764c 100644
> > --- a/hw/usb/hcd-xhci-nec.c
> > +++ b/hw/usb/hcd-xhci-nec.c
> > @@ -52,6 +52,16 @@ static void nec_xhci_instance_init(Object *obj)
> >
> >      pci->xhci.numintrs = nec->intrs;
> >      pci->xhci.numslots = nec->slots;
> > +
> > +    pci->cache_line_size = 0x10;
> > +    pci->pm_cap_off = 0;
> > +    pci->pcie_cap_off = 0xa0;
> > +    pci->msi_cap_off = 0x70;
> > +    pci->msix_cap_off = 0x90;
> > +    pci->msix_bar_nr = 0;
> > +    pci->msix_bar_size = 0;
> > +    pci->msix_table_off = 0x3000;
> > +    pci->msix_pba_off = 0x3800;
> >  }
>
>
> What about the "qemu-xhci" device, does that need similar treatment? I
> suspect it does at least for a bunch of these settings. Perhaps
> xhci_instance_init() in the abstract "pci-xhci" base might be a better
> place for these "sensible defaults" and then override them only in the
> specific implementations that need to do so, such as the new TI model?
> And/or have suitably named helper init function for configuring single-BAR
> PCI XHCI controllers so we can get some meaning behind all these magic
> numbers?

No you're right, I missed this entirely and the qemu-xhci dev is
indeed broken after this patch. Just moving it into the parent
instance init gets it to work.

> >  static void nec_xhci_class_init(ObjectClass *klass, void *data)
> > diff --git a/hw/usb/hcd-xhci-pci.c b/hw/usb/hcd-xhci-pci.c
> > index a039f5778a6..948d75b7379 100644
> > --- a/hw/usb/hcd-xhci-pci.c
> > +++ b/hw/usb/hcd-xhci-pci.c
> > @@ -32,8 +32,9 @@
> >  #include "trace.h"
> >  #include "qapi/error.h"
> >
> > -#define OFF_MSIX_TABLE  0x3000
> > -#define OFF_MSIX_PBA    0x3800
> > +#define MSIX_BAR_SIZE   0x800000
> >
>
> MSIX_BAR_SIZE doesn't seem to be used anywhere, and patch 2/2 uses 0x800000
> explicitly. (8 MiB also seems… huge? But I'm guessing you're matching this
> with the physical TI controller hardware - either way I don't think it
> belongs in this file.)
>
>
> > +#define OFF_MSIX_TABLE  0x0000
> > +#define OFF_MSIX_PBA    0x1000
> >
>
> Maybe instead of redefining these constants to only apply to the split BAR
> device variants, there should be 2 variants of them, one for single-BAR
> controllers, and one for controllers with separate BARs. That would also
> help make sense of the "magic numbers" in nec_xhci_instance_init().

You're right on both counts, I tidied these up.

> >  static void xhci_pci_intr_update(XHCIState *xhci, int n, bool enable)
> >  {
> > @@ -104,6 +105,31 @@ static int xhci_pci_vmstate_post_load(void *opaque,
> > int version_id)
> >     return 0;
> >  }
> >
> > +static int xhci_pci_add_pm_capability(PCIDevice *pci_dev, uint8_t offset,
> > +                                      Error **errp)
> > +{
> > +    int err;
> > +
> > +    err = pci_add_capability(pci_dev, PCI_CAP_ID_PM, offset,
> > +                             PCI_PM_SIZEOF, errp);
> > +    if (err < 0) {
> > +        return err;
> > +    }
> > +
> > +    pci_set_word(pci_dev->config + offset + PCI_PM_PMC,
> > +                 PCI_PM_CAP_VER_1_2 |
> > +                 PCI_PM_CAP_D1 | PCI_PM_CAP_D2 |
> > +                 PCI_PM_CAP_PME_D0 | PCI_PM_CAP_PME_D1 |
> > +                 PCI_PM_CAP_PME_D2 | PCI_PM_CAP_PME_D3hot);
> > +    pci_set_word(pci_dev->wmask + offset + PCI_PM_PMC, 0);
> > +    pci_set_word(pci_dev->config + offset + PCI_PM_CTRL,
> > +                 PCI_PM_CTRL_NO_SOFT_RESET);
> > +    pci_set_word(pci_dev->wmask + offset + PCI_PM_CTRL,
> > +                 PCI_PM_CTRL_STATE_MASK);
> > +
> > +    return 0;
> > +}
> > +
> >  static void usb_xhci_pci_realize(struct PCIDevice *dev, Error **errp)
> >  {
> >      int ret;
> > @@ -112,7 +138,7 @@ static void usb_xhci_pci_realize(struct PCIDevice
> > *dev, Error **errp)
> >
> >      dev->config[PCI_CLASS_PROG] = 0x30;    /* xHCI */
> >      dev->config[PCI_INTERRUPT_PIN] = 0x01; /* interrupt pin 1 */
> > -    dev->config[PCI_CACHE_LINE_SIZE] = 0x10;
> > +    dev->config[PCI_CACHE_LINE_SIZE] = s->cache_line_size;
> >      dev->config[0x60] = 0x30; /* release number */
> >
> >      object_property_set_link(OBJECT(&s->xhci), "host", OBJECT(s), NULL);
> > @@ -125,8 +151,16 @@ static void usb_xhci_pci_realize(struct PCIDevice
> > *dev, Error **errp)
> >          s->xhci.nec_quirks = true;
> >      }
> >
> > +    if (s->pm_cap_off) {
> > +        if (xhci_pci_add_pm_capability(dev, s->pm_cap_off, &err)) {
> > +            error_propagate(errp, err);
> > +            return;
> >
>
> Can't we just pass errp straight to xhci_pci_add_pm_capability and skip the
> error_propagate() here?

Yes I think so.

> > +        }
> > +    }
> > +
> >      if (s->msi != ON_OFF_AUTO_OFF) {
> > -        ret = msi_init(dev, 0x70, s->xhci.numintrs, true, false, &err);
> > +        ret = msi_init(dev, s->msi_cap_off, s->xhci.numintrs,
> > +                       true, false, &err);
> >          /*
> >           * Any error other than -ENOTSUP(board's MSI support is broken)
> >           * is a programming error
> > @@ -143,22 +177,37 @@ static void usb_xhci_pci_realize(struct PCIDevice
> > *dev, Error **errp)
> >          /* With msi=auto, we fall back to MSI off silently */
> >          error_free(err);
> >      }
> > +
> >      pci_register_bar(dev, 0,
> >                       PCI_BASE_ADDRESS_SPACE_MEMORY |
> >                       PCI_BASE_ADDRESS_MEM_TYPE_64,
> >                       &s->xhci.mem);
> >
> >      if (pci_bus_is_express(pci_get_bus(dev))) {
> > -        ret = pcie_endpoint_cap_init(dev, 0xa0);
> > +        ret = pcie_endpoint_cap_init(dev, s->pcie_cap_off);
> >          assert(ret > 0);
> >      }
> >
> >      if (s->msix != ON_OFF_AUTO_OFF) {
> > -        /* TODO check for errors, and should fail when msix=on */
> > -        msix_init(dev, s->xhci.numintrs,
> > -                  &s->xhci.mem, 0, OFF_MSIX_TABLE,
> > -                  &s->xhci.mem, 0, OFF_MSIX_PBA,
> > -                  0x90, NULL);
> > +        MemoryRegion *msix_bar = &s->xhci.mem;
> > +        if (s->msix_bar_nr != 0) {
> > +            memory_region_init(&dev->msix_exclusive_bar, OBJECT(dev),
> > +                               "xhci-msix", s->msix_bar_size);
> > +            msix_bar = &dev->msix_exclusive_bar;
> > +        }
> > +
> > +        ret = msix_init(dev, s->xhci.numintrs,
> > +                        msix_bar, s->msix_bar_nr, s->msix_table_off,
> > +                        msix_bar, s->msix_bar_nr, s->msix_pba_off,
> > +                        s->msix_cap_off, errp);
> > +        if (ret) {
> > +            return;
> > +        }
> >
>
> Surely we should only propagate the error and fail realize() iff s->msix is
> ON_OFF_AUTO_ON?
>
> For ON_OFF_AUTO_AUTO, msix_init returning failure isn't a critical error.

Yep you're right... you had been testing with msix disabled. I wonder if
there is a good way to force fail this in qtests?

> > +
> > +        pci_register_bar(dev, s->msix_bar_nr,
> > +                         PCI_BASE_ADDRESS_SPACE_MEMORY |
> > +                         PCI_BASE_ADDRESS_MEM_TYPE_64,
> > +                         msix_bar);
> >
>
> Is it safe to call pci_register_bar() again for the msix_bar_nr = 0 case?
> Even if it is safe, is it sensible? If we're calling it twice for the same
> BAR, and the arguments of either of the calls changes in future, the other
> needs to change too. Doesn't seem ideal.

Good catch. It looks like it "works" so long as the bar wasn't mapped,
but I'm sure bad practice... Interesting there is no assertion in
there though. I'll fix it though.

Thanks,
Nick