| Message ID | 1481930319-4796-9-git-send-email-boris.ostrovsky@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
>>> On 17.12.16 at 00:18, <boris.ostrovsky@oracle.com> wrote: > @@ -128,6 +130,13 @@ static int acpi_access_common(struct domain *d, bool is_guest_access, > *en = (((v & 0xff) << 8) | (*en & 0xff)) & *mask_en; > break; > } > + > + /* > + * If a new bit has been set in status register and corresponding > + * event is enabled then an SCI is sent to the guest. > + */ > + if ( !is_guest_access && ((*sts & ~sts_orig) & *en) ) > + send_guest_global_virq(d, VIRQ_SCI); From an abstract pov: How a bit gets set in either register shouldn't matter. Hence the !is_guest_access part here is at least questionable. However, iirc the guest can't itself set any status bits (since they're write-1-clear). Otoh, when an enable bit transitions from 0 to 1 (which the guest can effect) and the respective status bit is set, an SCI should occur according to my reading of the spec. Also - indentation. Jan
On 12/20/2016 08:37 AM, Jan Beulich wrote: >>>> On 17.12.16 at 00:18, <boris.ostrovsky@oracle.com> wrote: >> @@ -128,6 +130,13 @@ static int acpi_access_common(struct domain *d, bool is_guest_access, >> *en = (((v & 0xff) << 8) | (*en & 0xff)) & *mask_en; >> break; >> } >> + >> + /* >> + * If a new bit has been set in status register and corresponding >> + * event is enabled then an SCI is sent to the guest. >> + */ >> + if ( !is_guest_access && ((*sts & ~sts_orig) & *en) ) >> + send_guest_global_virq(d, VIRQ_SCI); > From an abstract pov: How a bit gets set in either register shouldn't > matter. Hence the !is_guest_access part here is at least questionable. > However, iirc the guest can't itself set any status bits (since they're > write-1-clear). Otoh, when an enable bit transitions from 0 to 1 (which > the guest can effect) and the respective status bit is set, an SCI > should occur according to my reading of the spec. I haven't considered the case of enabling with a pending status. Yes, that should trigger an SCI too, irrespective of is_guest_access. And I guess since guests can't set status bits I can drop the is_guest_access test in the check above too. -boris
diff --git a/xen/arch/x86/hvm/acpi.c b/xen/arch/x86/hvm/acpi.c index 044699d..c3c5ed7 100644 --- a/xen/arch/x86/hvm/acpi.c +++ b/xen/arch/x86/hvm/acpi.c @@ -4,6 +4,7 @@ */ #include <xen/acpi.h> #include <xen/errno.h> +#include <xen/event.h> #include <xen/lib.h> #include <xen/sched.h> @@ -94,6 +95,7 @@ static int acpi_access_common(struct domain *d, bool is_guest_access, else { uint32_t v = *val; + uint16_t sts_orig = *sts; /* Status register is write-1-to-clear by guests */ switch ( port & 3 ) @@ -128,6 +130,13 @@ static int acpi_access_common(struct domain *d, bool is_guest_access, *en = (((v & 0xff) << 8) | (*en & 0xff)) & *mask_en; break; } + + /* + * If a new bit has been set in status register and corresponding + * event is enabled then an SCI is sent to the guest. + */ + if ( !is_guest_access && ((*sts & ~sts_orig) & *en) ) + send_guest_global_virq(d, VIRQ_SCI); } return X86EMUL_OKAY; diff --git a/xen/common/event_channel.c b/xen/common/event_channel.c index 638dc5e..1d77373 100644 --- a/xen/common/event_channel.c +++ b/xen/common/event_channel.c @@ -727,7 +727,7 @@ void send_guest_vcpu_virq(struct vcpu *v, uint32_t virq) spin_unlock_irqrestore(&v->virq_lock, flags); } -static void send_guest_global_virq(struct domain *d, uint32_t virq) +void send_guest_global_virq(struct domain *d, uint32_t virq) { unsigned long flags; int port; @@ -739,7 +739,10 @@ static void send_guest_global_virq(struct domain *d, uint32_t virq) if ( unlikely(d == NULL) || unlikely(d->vcpu == NULL) ) return; - v = d->vcpu[0]; + /* Send to first available VCPU */ + for_each_vcpu(d, v) + if ( is_vcpu_online(v) ) + break; if ( unlikely(v == NULL) ) return; diff --git a/xen/include/xen/domain.h b/xen/include/xen/domain.h index bce0ea1..b386038 100644 --- a/xen/include/xen/domain.h +++ b/xen/include/xen/domain.h @@ -52,6 +52,7 @@ void vcpu_destroy(struct vcpu *v); int map_vcpu_info(struct vcpu *v, unsigned long gfn, unsigned offset); void unmap_vcpu_info(struct vcpu *v); +int arch_update_avail_vcpus(struct domain *d); int arch_domain_create(struct domain *d, unsigned int domcr_flags, struct xen_arch_domainconfig *config); diff --git a/xen/include/xen/event.h b/xen/include/xen/event.h index 5008c80..74bd605 100644 --- a/xen/include/xen/event.h +++ b/xen/include/xen/event.h @@ -23,6 +23,14 @@ void send_guest_vcpu_virq(struct vcpu *v, uint32_t virq); /* + * send_guest_global_virq: Notify guest via a global VIRQ. + * @d: domain to which virtual IRQ should be sent. First + * online VCPU will be selected. + * @virq: Virtual IRQ number (VIRQ_*) + */ +void send_guest_global_virq(struct domain *d, uint32_t virq); + +/* * send_global_virq: Notify the domain handling a global VIRQ. * @virq: Virtual IRQ number (VIRQ_*) */
When GPE0 status register gets a bit set (currently XEN_GPE0_CPUHP_BIT only) send an SCI to the guest. Also update send_guest_global_virq() to handle cases when VCPU0 is offlined. Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> --- Changes in v5: * Fixed SCI-triggering logic xen/arch/x86/hvm/acpi.c | 9 +++++++++ xen/common/event_channel.c | 7 +++++-- xen/include/xen/domain.h | 1 + xen/include/xen/event.h | 8 ++++++++ 4 files changed, 23 insertions(+), 2 deletions(-)