On Fri, Jun 16, 2017 at 01:44:43PM +0200, Ladi Prosek wrote: > The VT-d spec (section 6.5.2) prescribes software to zero the > Invalidation Queue Tail Register before enabling the VTD_GCMD_QIE > Global Command Register bit. Windows Server 2012 R2 and possibly > other older Windows versions violate the protocol and set a > non-zero queue tail first, which in effect makes them crash early > on boot with -device intel-iommu,intremap=on. > > This commit relaxes the check and instead of failing to enable > VTD_GCMD_QIE with "error: can't enable Queued Invalidation", it > behaves as if the tail register was set just after enabling > VTD_GCMD_QIE (see vtd_handle_iqt_write). > > Signed-off-by: Ladi Prosek <lpro...@redhat.com> > --- > hw/i386/intel_iommu.c | 32 +++++++++++++++----------------- > 1 file changed, 15 insertions(+), 17 deletions(-) > > diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c > index 15610b9..f2cb822 100644 > --- a/hw/i386/intel_iommu.c > +++ b/hw/i386/intel_iommu.c > @@ -1466,10 +1466,7 @@ static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, > uint64_t val) > return iaig; > } > > -static inline bool vtd_queued_inv_enable_check(IntelIOMMUState *s) > -{ > - return s->iq_tail == 0; > -} > +static void vtd_fetch_inv_desc(IntelIOMMUState *s); > > static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s) > { > @@ -1483,19 +1480,20 @@ static void vtd_handle_gcmd_qie(IntelIOMMUState *s, > bool en) > > VTD_DPRINTF(INV, "Queued Invalidation Enable %s", (en ? "on" : "off")); > if (en) { > - if (vtd_queued_inv_enable_check(s)) { > - s->iq = iqa_val & VTD_IQA_IQA_MASK; > - /* 2^(x+8) entries */ > - s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8); > - s->qi_enabled = true; > - VTD_DPRINTF(INV, "DMAR_IQA_REG 0x%"PRIx64, iqa_val); > - VTD_DPRINTF(INV, "Invalidation Queue addr 0x%"PRIx64 " size %d", > - s->iq, s->iq_size); > - /* Ok - report back to driver */ > - vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES); > - } else { > - VTD_DPRINTF(GENERAL, "error: can't enable Queued Invalidation: " > - "tail %"PRIu16, s->iq_tail); > + s->iq = iqa_val & VTD_IQA_IQA_MASK; > + /* 2^(x+8) entries */ > + s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8); > + s->qi_enabled = true; > + VTD_DPRINTF(INV, "DMAR_IQA_REG 0x%"PRIx64, iqa_val); > + VTD_DPRINTF(INV, "Invalidation Queue addr 0x%"PRIx64 " size %d", > + s->iq, s->iq_size); > + /* Ok - report back to driver */ > + vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES); > + > + if (s->iq_tail != 0 && > + !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) { > + VTD_DPRINTF(INV, "warning: QI enabled with non-zero tail"); > + vtd_fetch_inv_desc(s);
Since looks like we need another post, a tiny suggestion is that we can also add some comment to tell the reason why we didn't really check it, and a trace_vtd_warn_invalid_qi_tail() tracer to show that spec is violated (if you see, we have two other warnings already named vtd_warn_ir_* which also helps on tracing spec violations). I'll ack the next version after it's cooked. Thanks Ladi for the patch! Peter > } > } else { > if (vtd_queued_inv_disable_check(s)) { > -- > 2.9.3 > -- Peter Xu