[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-devel] [PATCH] MP interrupt delivery fix for xen-3.4
Please backport the following xen-4.0-testing.hg changesets to xen-3.4. They are necessary for kexec/kdump to work on MP guests. Backporting is mostly obvious but the changeset 19691:c4b2a3c9a61a, which is slightly different from 19912:721c14d7f60b, will cause a conflict. A combined diff which is ready to be applyed on xen-3.4-testing.hg is also attached. changeset: 19877:80839a223746 user: Keir Fraser <keir.fraser@xxxxxxxxxx> date: Wed Jul 01 20:22:29 2009 +0100 files: tools/firmware/hvmloader/hvmloader.c tools/firmware/hvmloader/mp_tables.c xen/arch/x86/hvm/vioapic.c xen/arch/x86/hvm/vlapic.c xen/arch/x86/hvm/vpic.c xen/include/asm-x86/hvm/domain.h xen/include/asm-x86/hvm/vlapic.h description: x86 hvm: Allow delivery of legacy 8259 interrupts to VCPUs != 0. changeset: 19906:2fadef1b008f user: Keir Fraser <keir.fraser@xxxxxxxxxx> date: Tue Jul 07 14:08:47 2009 +0100 files: xen/arch/x86/hvm/hpet.c xen/include/asm-x86/hvm/vpt.h description: x86,hvm: cleanup hpet.c vcpu handling same as i8254.c/rtc.c - introduce macros: domain_vhpet, vcpu_vhpet, vhpet_domain, vhpet_vcpu - remove *vcpu field from struct HPETState - modify guest_time_hpet() takes *vhpet instead of *vcpu as 1st - argument changeset: 19907:a29bb4efff00 user: Keir Fraser <keir.fraser@xxxxxxxxxx> date: Tue Jul 07 14:21:16 2009 +0100 files: xen/arch/x86/hvm/hpet.c xen/arch/x86/hvm/i8254.c xen/arch/x86/hvm/rtc.c xen/arch/x86/hvm/vlapic.c xen/arch/x86/hvm/vpt.c xen/include/asm-x86/hvm/vpt.h description: x86,hvm: Allow delivery of timer interrupts to VCPUs != 0 This patch is needed for kexec/kdump since VCPU#0 is halted. changeset: 19912:721c14d7f60b user: Keir Fraser <keir.fraser@xxxxxxxxxx> date: Wed Jul 08 14:22:00 2009 +0100 files: xen/arch/x86/hvm/hpet.c xen/arch/x86/hvm/i8254.c xen/arch/x86/hvm/rtc.c xen/arch/x86/hvm/vlapic.c xen/include/asm-x86/hvm/vlapic.h description: x86 hvm: Use 'x' as parameter name for macros converting between {vcpu,domain} and {vlapic,vpic,vrtc,hpet}. Completely avoids accidental aliasing. diff -r 14709d196e43 tools/firmware/hvmloader/hvmloader.c --- a/tools/firmware/hvmloader/hvmloader.c Wed Jun 30 18:26:13 2010 +0100 +++ b/tools/firmware/hvmloader/hvmloader.c Thu Jan 13 14:17:37 2011 +0900 @@ -141,13 +141,17 @@ static void apic_setup(void) { - /* Set the IOAPIC ID to tha static value used in the MP/ACPI tables. */ + /* Set the IOAPIC ID to the static value used in the MP/ACPI tables. */ ioapic_write(0x00, IOAPIC_ID); - /* Set up Virtual Wire mode. */ + /* NMIs are delivered direct to the BSP. */ lapic_write(APIC_SPIV, APIC_SPIV_APIC_ENABLED | 0xFF); - lapic_write(APIC_LVT0, APIC_MODE_EXTINT << 8); - lapic_write(APIC_LVT1, APIC_MODE_NMI << 8); + lapic_write(APIC_LVT0, (APIC_MODE_EXTINT << 8) | APIC_LVT_MASKED); + lapic_write(APIC_LVT1, APIC_MODE_NMI << 8); + + /* 8259A ExtInts are delivered through IOAPIC pin 0 (Virtual Wire Mode). */ + ioapic_write(0x10, APIC_DM_EXTINT); + ioapic_write(0x11, SET_APIC_ID(LAPIC_ID(0))); } static void pci_setup(void) diff -r 14709d196e43 tools/firmware/hvmloader/mp_tables.c --- a/tools/firmware/hvmloader/mp_tables.c Wed Jun 30 18:26:13 2010 +0100 +++ b/tools/firmware/hvmloader/mp_tables.c Thu Jan 13 14:17:37 2011 +0900 @@ -233,21 +233,6 @@ } -/* fills in an IO interrupt entry for IOAPIC 'ioapic_id' */ -static void fill_mp_io_intr_entry( - struct mp_io_intr_entry *mpiie, - int src_bus_id, int src_bus_irq, int ioapic_id, int dst_ioapic_intin) -{ - mpiie->type = ENTRY_TYPE_IO_INTR; - mpiie->intr_type = INTR_TYPE_INT; - mpiie->io_intr_flags = (PCI_ISA_IRQ_MASK & (1U<<src_bus_irq)) ? 0xf : 0x0; - mpiie->src_bus_id = src_bus_id; - mpiie->src_bus_irq = src_bus_irq; - mpiie->dst_ioapic_id = ioapic_id; - mpiie->dst_ioapic_intin = dst_ioapic_intin; -} - - /* fill in the mp floating processor structure */ static void fill_mpfps(struct mp_floating_pointer_struct *mpfps, uint32_t mpct) { @@ -316,6 +301,7 @@ void *mp_table_base; char *p; int vcpu_nr, i, length; + struct mp_io_intr_entry *mpiie; vcpu_nr = hvm_info->nr_vcpus; @@ -343,12 +329,28 @@ fill_mp_ioapic_entry((struct mp_ioapic_entry *)p); p += sizeof(struct mp_ioapic_entry); + /* I/O interrupt assignment: IOAPIC pin 0 is connected to 8259 ExtInt. */ + mpiie = (struct mp_io_intr_entry *)p; + memset(mpiie, 0, sizeof(*mpiie)); + mpiie->type = ENTRY_TYPE_IO_INTR; + mpiie->intr_type = INTR_TYPE_EXTINT; + mpiie->dst_ioapic_id = IOAPIC_ID; + p += sizeof(*mpiie); + + /* I/O interrupt assignment for every legacy 8259 interrupt source. */ for ( i = 0; i < 16; i++ ) { - if ( i == 2 ) continue; /* skip the slave PIC connection */ - fill_mp_io_intr_entry((struct mp_io_intr_entry *)p, - BUS_ID_ISA, i, IOAPIC_ID, (i == 0) ? 2 : i); - p += sizeof(struct mp_io_intr_entry); + if ( i == 2 ) + continue; /* skip the slave PIC connection */ + mpiie = (struct mp_io_intr_entry *)p; + mpiie->type = ENTRY_TYPE_IO_INTR; + mpiie->intr_type = INTR_TYPE_EXTINT; + mpiie->io_intr_flags = (PCI_ISA_IRQ_MASK & (1U << i)) ? 0xf : 0x0; + mpiie->src_bus_id = BUS_ID_ISA; + mpiie->src_bus_irq = i; + mpiie->dst_ioapic_id = IOAPIC_ID; + mpiie->dst_ioapic_intin = (i == 0) ? 2 : i; + p += sizeof(*mpiie); } length = p - (char *)mp_table_base; diff -r 14709d196e43 xen/arch/x86/hvm/hpet.c --- a/xen/arch/x86/hvm/hpet.c Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/arch/x86/hvm/hpet.c Thu Jan 13 14:17:37 2011 +0900 @@ -24,6 +24,12 @@ #include <xen/sched.h> #include <xen/event.h> +#define domain_vhpet(x) (&(x)->arch.hvm_domain.pl_time.vhpet) +#define vcpu_vhpet(x) (domain_vhpet((x)->domain)) +#define vhpet_domain(x) (container_of((x), struct domain, \ + arch.hvm_domain.pl_time.vhpet)) +#define vhpet_vcpu(x) (pt_global_vcpu_target(vhpet_domain(x))) + #define HPET_BASE_ADDRESS 0xfed00000ULL #define HPET_MMAP_SIZE 1024 #define S_TO_NS 1000000000ULL /* 1s = 10^9 ns */ @@ -31,7 +37,8 @@ /* Frequency_of_Xen_systeme_time / frequency_of_HPET = 16 */ #define STIME_PER_HPET_TICK 16 -#define guest_time_hpet(v) (hvm_get_guest_time(v) / STIME_PER_HPET_TICK) +#define guest_time_hpet(hpet) \ + (hvm_get_guest_time(vhpet_vcpu(hpet)) / STIME_PER_HPET_TICK) #define HPET_ID 0x000 #define HPET_PERIOD 0x004 @@ -94,7 +101,7 @@ ASSERT(spin_is_locked(&h->lock)); if ( hpet_enabled(h) ) - return guest_time_hpet(h->vcpu) + h->mc_offset; + return guest_time_hpet(h) + h->mc_offset; else return h->hpet.mc64; } @@ -176,7 +183,7 @@ struct vcpu *v, unsigned long addr, unsigned long length, unsigned long *pval) { - HPETState *h = &v->domain->arch.hvm_domain.pl_time.vhpet; + HPETState *h = vcpu_vhpet(v); unsigned long result; uint64_t val; @@ -230,7 +237,7 @@ { /* HPET specification requires PIT shouldn't generate * interrupts if LegacyReplacementRoute is set for timer0 */ - PITState *pit = &h->vcpu->domain->arch.hvm_domain.pl_time.vpit; + PITState *pit = &vhpet_domain(h)->arch.hvm_domain.pl_time.vpit; pit_stop_channel0_irq(pit); } @@ -272,7 +279,7 @@ * being enabled (now). */ oneshot = !timer_is_periodic(h, tn); - create_periodic_time(h->vcpu, &h->pt[tn], + create_periodic_time(vhpet_vcpu(h), &h->pt[tn], hpet_tick_to_ns(h, diff), oneshot ? 0 : hpet_tick_to_ns(h, h->hpet.period[tn]), irq, NULL, NULL); @@ -290,7 +297,7 @@ struct vcpu *v, unsigned long addr, unsigned long length, unsigned long val) { - HPETState *h = &v->domain->arch.hvm_domain.pl_time.vhpet; + HPETState *h = vcpu_vhpet(v); uint64_t old_val, new_val; int tn, i; @@ -323,7 +330,7 @@ if ( !(old_val & HPET_CFG_ENABLE) && (new_val & HPET_CFG_ENABLE) ) { /* Enable main counter and interrupt generation. */ - h->mc_offset = h->hpet.mc64 - guest_time_hpet(h->vcpu); + h->mc_offset = h->hpet.mc64 - guest_time_hpet(h); for ( i = 0; i < HPET_TIMER_NUM; i++ ) { h->hpet.comparator64[i] = @@ -337,7 +344,7 @@ else if ( (old_val & HPET_CFG_ENABLE) && !(new_val & HPET_CFG_ENABLE) ) { /* Halt main counter and disable interrupt generation. */ - h->hpet.mc64 = h->mc_offset + guest_time_hpet(h->vcpu); + h->hpet.mc64 = h->mc_offset + guest_time_hpet(h); for ( i = 0; i < HPET_TIMER_NUM; i++ ) if ( timer_enabled(h, i) ) set_stop_timer(i); @@ -487,13 +494,13 @@ static int hpet_save(struct domain *d, hvm_domain_context_t *h) { - HPETState *hp = &d->arch.hvm_domain.pl_time.vhpet; + HPETState *hp = domain_vhpet(d); int rc; spin_lock(&hp->lock); /* Write the proper value into the main counter */ - hp->hpet.mc64 = hp->mc_offset + guest_time_hpet(hp->vcpu); + hp->hpet.mc64 = hp->mc_offset + guest_time_hpet(hp); /* Save the HPET registers */ rc = _hvm_init_entry(h, HVM_SAVE_CODE(HPET), 0, HVM_SAVE_LENGTH(HPET)); @@ -531,7 +538,7 @@ static int hpet_load(struct domain *d, hvm_domain_context_t *h) { - HPETState *hp = &d->arch.hvm_domain.pl_time.vhpet; + HPETState *hp = domain_vhpet(d); struct hvm_hw_hpet *rec; uint64_t cmp; int i; @@ -572,7 +579,7 @@ #undef C /* Recalculate the offset between the main counter and guest time */ - hp->mc_offset = hp->hpet.mc64 - guest_time_hpet(hp->vcpu); + hp->mc_offset = hp->hpet.mc64 - guest_time_hpet(hp); /* restart all timers */ @@ -590,14 +597,13 @@ void hpet_init(struct vcpu *v) { - HPETState *h = &v->domain->arch.hvm_domain.pl_time.vhpet; + HPETState *h = vcpu_vhpet(v); int i; memset(h, 0, sizeof(HPETState)); spin_lock_init(&h->lock); - h->vcpu = v; h->stime_freq = S_TO_NS; h->hpet_to_ns_scale = ((S_TO_NS * STIME_PER_HPET_TICK) << 10) / h->stime_freq; @@ -622,7 +628,7 @@ void hpet_deinit(struct domain *d) { int i; - HPETState *h = &d->arch.hvm_domain.pl_time.vhpet; + HPETState *h = domain_vhpet(d); spin_lock(&h->lock); diff -r 14709d196e43 xen/arch/x86/hvm/i8254.c --- a/xen/arch/x86/hvm/i8254.c Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/arch/x86/hvm/i8254.c Thu Jan 13 14:17:37 2011 +0900 @@ -42,7 +42,7 @@ #define vcpu_vpit(x) (domain_vpit((x)->domain)) #define vpit_domain(x) (container_of((x), struct domain, \ arch.hvm_domain.pl_time.vpit)) -#define vpit_vcpu(x) (vpit_domain(x)->vcpu[0]) +#define vpit_vcpu(x) (pt_global_vcpu_target(vpit_domain(x))) #define RW_STATE_LSB 1 #define RW_STATE_MSB 2 diff -r 14709d196e43 xen/arch/x86/hvm/rtc.c --- a/xen/arch/x86/hvm/rtc.c Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/arch/x86/hvm/rtc.c Thu Jan 13 14:17:37 2011 +0900 @@ -32,7 +32,7 @@ #define vcpu_vrtc(x) (domain_vrtc((x)->domain)) #define vrtc_domain(x) (container_of((x), struct domain, \ arch.hvm_domain.pl_time.vrtc)) -#define vrtc_vcpu(x) (vrtc_domain(x)->vcpu[0]) +#define vrtc_vcpu(x) (pt_global_vcpu_target(vrtc_domain(x))) static void rtc_periodic_cb(struct vcpu *v, void *opaque) { diff -r 14709d196e43 xen/arch/x86/hvm/vioapic.c --- a/xen/arch/x86/hvm/vioapic.c Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/arch/x86/hvm/vioapic.c Thu Jan 13 14:17:37 2011 +0900 @@ -146,10 +146,14 @@ *pent = ent; - if ( (ent.fields.trig_mode == VIOAPIC_LEVEL_TRIG) && - !ent.fields.mask && - !ent.fields.remote_irr && - hvm_irq->gsi_assert_count[idx] ) + if ( idx == 0 ) + { + vlapic_adjust_i8259_target(d); + } + else if ( (ent.fields.trig_mode == VIOAPIC_LEVEL_TRIG) && + !ent.fields.mask && + !ent.fields.remote_irr && + hvm_irq->gsi_assert_count[idx] ) { pent->fields.remote_irr = 1; vioapic_deliver(vioapic, idx); @@ -159,8 +163,7 @@ } static void vioapic_write_indirect( - struct hvm_hw_vioapic *vioapic, unsigned long addr, - unsigned long length, unsigned long val) + struct hvm_hw_vioapic *vioapic, unsigned long length, unsigned long val) { switch ( vioapic->ioregsel ) { @@ -213,7 +216,7 @@ break; case VIOAPIC_REG_WINDOW: - vioapic_write_indirect(vioapic, addr, length, val); + vioapic_write_indirect(vioapic, length, val); break; #if VIOAPIC_IS_IOSAPIC diff -r 14709d196e43 xen/arch/x86/hvm/vlapic.c --- a/xen/arch/x86/hvm/vlapic.c Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/arch/x86/hvm/vlapic.c Thu Jan 13 14:17:37 2011 +0900 @@ -697,6 +697,8 @@ val |= APIC_LVT_MASKED; val &= vlapic_lvt_mask[(offset - APIC_LVTT) >> 4]; vlapic_set_reg(vlapic, offset, val); + if ( offset == APIC_LVT0 ) + vlapic_adjust_i8259_target(v->domain); break; case APIC_TMICT: @@ -776,18 +778,46 @@ "apic base msr is 0x%016"PRIx64, vlapic->hw.apic_base_msr); } +static int __vlapic_accept_pic_intr(struct vcpu *v) +{ + struct domain *d = v->domain; + struct vlapic *vlapic = vcpu_vlapic(v); + uint32_t lvt0 = vlapic_get_reg(vlapic, APIC_LVT0); + union vioapic_redir_entry redir0 = domain_vioapic(d)->redirtbl[0]; + + /* We deliver 8259 interrupts to the appropriate CPU as follows. */ + return ((/* IOAPIC pin0 is unmasked and routing to this LAPIC? */ + ((redir0.fields.delivery_mode == dest_ExtINT) && + !redir0.fields.mask && + redir0.fields.dest_id == VLAPIC_ID(vlapic) && + !vlapic_disabled(vlapic)) || + /* LAPIC has LVT0 unmasked for ExtInts? */ + ((lvt0 & (APIC_MODE_MASK|APIC_LVT_MASKED)) == APIC_DM_EXTINT) || + /* LAPIC is fully disabled? */ + vlapic_hw_disabled(vlapic))); +} + int vlapic_accept_pic_intr(struct vcpu *v) { - struct vlapic *vlapic = vcpu_vlapic(v); - uint32_t lvt0 = vlapic_get_reg(vlapic, APIC_LVT0); + return ((v == v->domain->arch.hvm_domain.i8259_target) && + __vlapic_accept_pic_intr(v)); +} - /* - * Only CPU0 is wired to the 8259A. INTA cycles occur if LINT0 is set up - * accept ExtInts, or if the LAPIC is disabled (so LINT0 behaves as INTR). - */ - return ((v->vcpu_id == 0) && - (((lvt0 & (APIC_MODE_MASK|APIC_LVT_MASKED)) == APIC_DM_EXTINT) || - vlapic_hw_disabled(vlapic))); +void vlapic_adjust_i8259_target(struct domain *d) +{ + struct vcpu *v; + + for_each_vcpu ( d, v ) + if ( __vlapic_accept_pic_intr(v) ) + goto found; + + v = d->vcpu ? d->vcpu[0] : NULL; + + found: + if ( d->arch.hvm_domain.i8259_target == v ) + return; + d->arch.hvm_domain.i8259_target = v; + pt_adjust_global_vcpu_target(v); } int vlapic_has_pending_irq(struct vcpu *v) @@ -946,6 +976,7 @@ if ( hvm_load_entry(LAPIC_REGS, h, s->regs) != 0 ) return -EINVAL; + vlapic_adjust_i8259_target(d); lapic_rearm(s); return 0; } diff -r 14709d196e43 xen/arch/x86/hvm/vpic.c --- a/xen/arch/x86/hvm/vpic.c Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/arch/x86/hvm/vpic.c Thu Jan 13 14:17:37 2011 +0900 @@ -109,9 +109,9 @@ { if ( vpic->is_master ) { - /* Master INT line is connected to VCPU0's VLAPIC LVT0. */ - struct vcpu *v = vpic_domain(vpic)->vcpu[0]; - if ( (v != NULL) && vlapic_accept_pic_intr(v) ) + /* Master INT line is connected in Virtual Wire Mode. */ + struct vcpu *v = vpic_domain(vpic)->arch.hvm_domain.i8259_target; + if ( v != NULL ) vcpu_kick(v); } else diff -r 14709d196e43 xen/arch/x86/hvm/vpt.c --- a/xen/arch/x86/hvm/vpt.c Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/arch/x86/hvm/vpt.c Thu Jan 13 14:17:37 2011 +0900 @@ -434,3 +434,53 @@ */ kill_timer(&pt->timer); } + +static void pt_adjust_vcpu(struct periodic_time *pt, struct vcpu *v) +{ + int on_list; + + ASSERT(pt->source == PTSRC_isa); + + if ( pt->vcpu == NULL ) + return; + + pt_lock(pt); + on_list = pt->on_list; + if ( pt->on_list ) + list_del(&pt->list); + pt->on_list = 0; + pt_unlock(pt); + + spin_lock(&v->arch.hvm_vcpu.tm_lock); + pt->vcpu = v; + if ( on_list ) + { + pt->on_list = 1; + list_add(&pt->list, &v->arch.hvm_vcpu.tm_list); + + migrate_timer(&pt->timer, v->processor); + } + spin_unlock(&v->arch.hvm_vcpu.tm_lock); +} + +void pt_adjust_global_vcpu_target(struct vcpu *v) +{ + struct pl_time *pl_time = &v->domain->arch.hvm_domain.pl_time; + int i; + + if ( v == NULL ) + return; + + spin_lock(&pl_time->vpit.lock); + pt_adjust_vcpu(&pl_time->vpit.pt0, v); + spin_unlock(&pl_time->vpit.lock); + + spin_lock(&pl_time->vrtc.lock); + pt_adjust_vcpu(&pl_time->vrtc.pt, v); + spin_unlock(&pl_time->vrtc.lock); + + spin_lock(&pl_time->vhpet.lock); + for ( i = 0; i < HPET_TIMER_NUM; i++ ) + pt_adjust_vcpu(&pl_time->vhpet.pt[i], v); + spin_unlock(&pl_time->vhpet.lock); +} diff -r 14709d196e43 xen/include/asm-x86/hvm/domain.h --- a/xen/include/asm-x86/hvm/domain.h Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/include/asm-x86/hvm/domain.h Thu Jan 13 14:17:37 2011 +0900 @@ -56,6 +56,9 @@ struct hvm_vioapic *vioapic; struct hvm_hw_stdvga stdvga; + /* VCPU which is current target for 8259 interrupts. */ + struct vcpu *i8259_target; + /* hvm_print_line() logging. */ char pbuf[80]; int pbuf_idx; diff -r 14709d196e43 xen/include/asm-x86/hvm/vlapic.h --- a/xen/include/asm-x86/hvm/vlapic.h Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/include/asm-x86/hvm/vlapic.h Thu Jan 13 14:17:37 2011 +0900 @@ -92,6 +92,8 @@ int vlapic_accept_pic_intr(struct vcpu *v); +void vlapic_adjust_i8259_target(struct domain *d); + struct vlapic *apic_lowest_prio(struct domain *d, uint32_t bitmap); int vlapic_match_logical_addr(struct vlapic *vlapic, uint8_t mda); diff -r 14709d196e43 xen/include/asm-x86/hvm/vpt.h --- a/xen/include/asm-x86/hvm/vpt.h Wed Jun 30 18:26:13 2010 +0100 +++ b/xen/include/asm-x86/hvm/vpt.h Thu Jan 13 14:17:37 2011 +0900 @@ -92,7 +92,6 @@ typedef struct HPETState { struct hpet_registers hpet; - struct vcpu *vcpu; uint64_t stime_freq; uint64_t hpet_to_ns_scale; /* hpet ticks to ns (multiplied by 2^10) */ uint64_t hpet_to_ns_limit; /* max hpet ticks convertable to ns */ @@ -145,6 +144,10 @@ void pt_reset(struct vcpu *v); void pt_migrate(struct vcpu *v); +void pt_adjust_global_vcpu_target(struct vcpu *v); +#define pt_global_vcpu_target(d) \ + ((d)->arch.hvm_domain.i8259_target ? : (d)->vcpu ? (d)->vcpu[0] : NULL) + /* Is given periodic timer active? */ #define pt_active(pt) ((pt)->on_list) _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |