On Fri, Sep 14, 2012 at 09:05:10AM +0200, Paolo Bonzini wrote:
> Il 13/09/2012 22:12, Jason Baron ha scritto:
> > From: Isaku Yamahata <yamah...@valinux.co.jp>
> >
> > This was totally off: The CC registers are 16 bit (stored as little
> > endian), their offsets run in reverse order, and D26IR as well as D25IR
> > have 4 bytes offset to their successors.
> >
> > Reported-by: Jan Kiszka <jan.kis...@siemens.com>
> > Signed-off-by: Isaku Yamahata <yamah...@valinux.co.jp>
> > Signed-off-by: Jason Baron <jba...@redhat.com>
> > ---
> > hw/q35.c | 29 ++++++++++++++++++++---------
> > 1 files changed, 20 insertions(+), 9 deletions(-)
> >
> > diff --git a/hw/q35.c b/hw/q35.c
> > index 8b6a2e5..295344e 100644
> > --- a/hw/q35.c
> > +++ b/hw/q35.c
> > @@ -474,7 +474,7 @@ static void ich9_lpc_reset(DeviceState *qdev);
> > * Although it's not pci configuration space, it's little endian as Intel.
> > */
> >
> > -static void ich9_cc_update_ir(uint8_t irr[PCI_NUM_PINS], uint32_t ir)
> > +static void ich9_cc_update_ir(uint8_t irr[PCI_NUM_PINS], uint16_t ir)
> > {
> > int intx;
> > for (intx = 0; intx < PCI_NUM_PINS; intx++) {
> > @@ -485,15 +485,26 @@ static void ich9_cc_update_ir(uint8_t
> > irr[PCI_NUM_PINS], uint32_t ir)
> > static void ich9_cc_update(ICH9_LPCState *lpc)
> > {
> > int slot;
> > - int reg_offset;
> > - int intx;
> > + int pci_intx;
> > +
> > + const int reg_offsets[] = {
> > + ICH9_CC_D25IR,
> > + ICH9_CC_D26IR,
> > + ICH9_CC_D27IR,
> > + ICH9_CC_D28IR,
> > + ICH9_CC_D29IR,
> > + ICH9_CC_D30IR,
> > + ICH9_CC_D31IR,
> > + };
> > + const int *offset;
> >
> > /* D{25 - 31}IR, but D30IR is read only to 0. */
> > - for (slot = 25, reg_offset = 0; slot < 32; slot++, reg_offset++) {
> > - if (slot != 30) {
> > - ich9_cc_update_ir(lpc->irr[slot],
> > - lpc->chip_config[ICH9_CC_D31IR +
> > reg_offset]);
> > + for (slot = 25, offset = reg_offsets; slot < 32; slot++, offset++) {
> > + if (slot == 30) {
> > + continue;
> > }
> > + ich9_cc_update_ir(lpc->irr[slot],
> > + pci_get_word(lpc->chip_config + *offset));
> > }
> >
> > /*
> > @@ -502,8 +513,8 @@ static void ich9_cc_update(ICH9_LPCState *lpc)
> > * are connected to pirq lines. Our choice is PIRQ[E-H].
> > * INT[A-D] are connected to PIRQ[E-H]
> > */
> > - for (intx = 0; intx < PCI_NUM_PINS; intx++) {
> > - lpc->irr[30][intx] = intx + 4;
> > + for (pci_intx = 0; pci_intx < PCI_NUM_PINS; pci_intx++) {
> > + lpc->irr[30][pci_intx] = pci_intx + 4;
> > }
> > }
> >
> >
>
> I guess this patch and patch 12 could/should be squashed in patch 11
> (the one that introduces q35.c)?
>
> Paolo
Michael Tsirkin also suggested combining them. I kept them separate to
make it clear what Yamahata had written, and the re-base I had done. I
agree it would be cleaner to combine. That said, Michael also suggested
not adding the initial one to the build so its still bi-sectable, I
think that could be a reasonable option too.
Thanks,
-Jason
