First and foremost this series is trying to deal with CPU offlining issues, which have become more prominent with the recently added SMT enable/disable operation in xen-hptool. Later patches in the series then carry out more or less unrelated changes (hopefully improvements) noticed while looking at various pieces of involved code. 01: deal with move-in-progress state in fixup_irqs() 02: deal with move cleanup count state in fixup_irqs() 03: improve dump_irqs() 04: desc->affinity should strictly represent the requested value 05: consolidate use of ->arch.cpu_mask 06: fix locking around vector management 07: target online CPUs when binding guest IRQ 08: correct/tighten vector check in _clear_irq_vector() 09: make fixup_irqs() skip unconnected internally used interrupts 10: drop redundant cpumask_empty() from move_masked_irq() 11: simplify and rename pirq_acktype() 12: add explict tracing-enabled check to trace_irq_mask() 13: tighten vector checks 14: eliminate some on-stack cpumask_t instances 15: move {,_}clear_irq_vector() In principle patches 1, 2, 4-7, and maybe 9 are backporting candidates. Their intrusive nature makes wanting to do so questionable, though. This (in particular patch 14) depends on "[PATCH 0/4] x86: EOI timer corrections / improvements". For v3 specific information please see the individual patches. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
The flag being set may prevent affinity changes, as these often imply assignment of a new vector. When there's no possible destination left for the IRQ, the clearing of the flag needs to happen right from fixup_irqs(). Additionally _assign_irq_vector() needs to avoid setting the flag when there's no online CPU left in what gets put into ->arch.old_cpu_mask. The old vector can be released right away in this case. Also extend the log message about broken affinity to include the new affinity as well, allowing to notice issues with affinity changes not actually having taken place. Swap the if/else-if order there at the same time to reduce the amount of conditions checked. At the same time replace two open coded instances of the new helper function. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- v3: Move release_old_vec() further up (so a later patch won't need to). Re-base. v2: Add/use valid_irq_vector(). v1b: Also update vector_irq[] in the code added to fixup_irqs(). --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -99,6 +99,27 @@ void unlock_vector_lock(void) spin_unlock(&vector_lock); } +static inline bool valid_irq_vector(unsigned int vector) +{ + return vector >= FIRST_DYNAMIC_VECTOR && vector <= LAST_HIPRIORITY_VECTOR; +} + +static void release_old_vec(struct irq_desc *desc) +{ + unsigned int vector = desc->arch.old_vector; + + desc->arch.old_vector = IRQ_VECTOR_UNASSIGNED; + cpumask_clear(desc->arch.old_cpu_mask); + + if ( !valid_irq_vector(vector) ) + ASSERT_UNREACHABLE(); + else if ( desc->arch.used_vectors ) + { + ASSERT(test_bit(vector, desc->arch.used_vectors)); + clear_bit(vector, desc->arch.used_vectors); + } +} + static void trace_irq_mask(uint32_t event, int irq, int vector, const cpumask_t *mask) { @@ -288,14 +309,7 @@ static void __clear_irq_vector(int irq) per_cpu(vector_irq, cpu)[old_vector] = ~irq; } - desc->arch.old_vector = IRQ_VECTOR_UNASSIGNED; - cpumask_clear(desc->arch.old_cpu_mask); - - if ( desc->arch.used_vectors ) - { - ASSERT(test_bit(old_vector, desc->arch.used_vectors)); - clear_bit(old_vector, desc->arch.used_vectors); - } + release_old_vec(desc); desc->arch.move_in_progress = 0; } @@ -520,12 +534,21 @@ next: /* Found one! */ current_vector = vector; current_offset = offset; - if (old_vector > 0) { - desc->arch.move_in_progress = 1; - cpumask_copy(desc->arch.old_cpu_mask, desc->arch.cpu_mask); + + if ( old_vector > 0 ) + { + cpumask_and(desc->arch.old_cpu_mask, desc->arch.cpu_mask, + &cpu_online_map); desc->arch.old_vector = desc->arch.vector; + if ( !cpumask_empty(desc->arch.old_cpu_mask) ) + desc->arch.move_in_progress = 1; + else + /* This can happen while offlining a CPU. */ + release_old_vec(desc); } + trace_irq_mask(TRC_HW_IRQ_ASSIGN_VECTOR, irq, vector, &tmp_mask); + for_each_cpu(new_cpu, &tmp_mask) per_cpu(vector_irq, new_cpu)[vector] = irq; desc->arch.vector = vector; @@ -694,14 +717,8 @@ void irq_move_cleanup_interrupt(struct c if ( desc->arch.move_cleanup_count == 0 ) { - desc->arch.old_vector = IRQ_VECTOR_UNASSIGNED; - cpumask_clear(desc->arch.old_cpu_mask); - - if ( desc->arch.used_vectors ) - { - ASSERT(test_bit(vector, desc->arch.used_vectors)); - clear_bit(vector, desc->arch.used_vectors); - } + ASSERT(vector == desc->arch.old_vector); + release_old_vec(desc); } unlock: spin_unlock(&desc->lock); @@ -2400,6 +2417,33 @@ void fixup_irqs(const cpumask_t *mask, b continue; } + /* + * In order for the affinity adjustment below to be successful, we + * need __assign_irq_vector() to succeed. This in particular means + * clearing desc->arch.move_in_progress if this would otherwise + * prevent the function from succeeding. Since there's no way for the + * flag to get cleared anymore when there's no possible destination + * left (the only possibility then would be the IRQs enabled window + * after this loop), there's then also no race with us doing it here. + * + * Therefore the logic here and there need to remain in sync. + */ + if ( desc->arch.move_in_progress && + !cpumask_intersects(mask, desc->arch.cpu_mask) ) + { + unsigned int cpu; + + cpumask_and(&affinity, desc->arch.old_cpu_mask, &cpu_online_map); + + spin_lock(&vector_lock); + for_each_cpu(cpu, &affinity) + per_cpu(vector_irq, cpu)[desc->arch.old_vector] = ~irq; + spin_unlock(&vector_lock); + + release_old_vec(desc); + desc->arch.move_in_progress = 0; + } + cpumask_and(&affinity, &affinity, mask); if ( cpumask_empty(&affinity) ) { @@ -2418,15 +2462,18 @@ void fixup_irqs(const cpumask_t *mask, b if ( desc->handler->enable ) desc->handler->enable(desc); + cpumask_copy(&affinity, desc->affinity); + spin_unlock(&desc->lock); if ( !verbose ) continue; - if ( break_affinity && set_affinity ) - printk("Broke affinity for irq %i\n", irq); - else if ( !set_affinity ) - printk("Cannot set affinity for irq %i\n", irq); + if ( !set_affinity ) + printk("Cannot set affinity for IRQ%u\n", irq); + else if ( break_affinity ) + printk("Broke affinity for IRQ%u, new: %*pb\n", + irq, nr_cpu_ids, &affinity); } /* That doesn't seem sufficient. Give it 1ms. */ _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:44, Jan Beulich wrote: > The flag being set may prevent affinity changes, as these often imply > assignment of a new vector. When there's no possible destination left > for the IRQ, the clearing of the flag needs to happen right from > fixup_irqs(). > > Additionally _assign_irq_vector() needs to avoid setting the flag when > there's no online CPU left in what gets put into ->arch.old_cpu_mask. > The old vector can be released right away in this case. This suggests that it is a bugfix, but it isn't clear what happens when things go wrong. > --- a/xen/arch/x86/irq.c > +++ b/xen/arch/x86/irq.c > @@ -2418,15 +2462,18 @@ void fixup_irqs(const cpumask_t *mask, b > if ( desc->handler->enable ) > desc->handler->enable(desc); > > + cpumask_copy(&affinity, desc->affinity); > + > spin_unlock(&desc->lock); > > if ( !verbose ) > continue; > > - if ( break_affinity && set_affinity ) > - printk("Broke affinity for irq %i\n", irq); > - else if ( !set_affinity ) > - printk("Cannot set affinity for irq %i\n", irq); > + if ( !set_affinity ) > + printk("Cannot set affinity for IRQ%u\n", irq); > + else if ( break_affinity ) > + printk("Broke affinity for IRQ%u, new: %*pb\n", > + irq, nr_cpu_ids, &affinity); While I certainly prefer this version, I should point out that you refused to accept my patches like this, and for consistency with the rest of the codebase, you should be using cpumask_bits(). ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 03.07.2019 17:39, Andrew Cooper wrote: > On 17/05/2019 11:44, Jan Beulich wrote: >> The flag being set may prevent affinity changes, as these often imply >> assignment of a new vector. When there's no possible destination left >> for the IRQ, the clearing of the flag needs to happen right from >> fixup_irqs(). >> >> Additionally _assign_irq_vector() needs to avoid setting the flag when >> there's no online CPU left in what gets put into ->arch.old_cpu_mask. >> The old vector can be released right away in this case. > > This suggests that it is a bugfix, but it isn't clear what happens when > things go wrong. The vector cleanup wouldn't ever trigger, as the IRQ wouldn't get raised anymore to any of its prior target CPUs. Hence the immediate cleanup that gets done in that case. I thought the 2nd sentence would make this clear. If it doesn't, do you have a suggestion on how to improve the text? >> --- a/xen/arch/x86/irq.c >> +++ b/xen/arch/x86/irq.c >> @@ -2418,15 +2462,18 @@ void fixup_irqs(const cpumask_t *mask, b >> if ( desc->handler->enable ) >> desc->handler->enable(desc); >> >> + cpumask_copy(&affinity, desc->affinity); >> + >> spin_unlock(&desc->lock); >> >> if ( !verbose ) >> continue; >> >> - if ( break_affinity && set_affinity ) >> - printk("Broke affinity for irq %i\n", irq); >> - else if ( !set_affinity ) >> - printk("Cannot set affinity for irq %i\n", irq); >> + if ( !set_affinity ) >> + printk("Cannot set affinity for IRQ%u\n", irq); >> + else if ( break_affinity ) >> + printk("Broke affinity for IRQ%u, new: %*pb\n", >> + irq, nr_cpu_ids, &affinity); > > While I certainly prefer this version, I should point out that you > refused to accept my patches like this, and for consistency with the > rest of the codebase, you should be using cpumask_bits(). Oh, indeed. I guess I had converted a debugging only printk() into this one without noticing the necessary tidying, the more that elsewhere in the series I'm actually doing so already. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
The cleanup IPI may get sent immediately before a CPU gets removed from the online map. In such a case the IPI would get handled on the CPU being offlined no earlier than in the interrupts disabled window after fixup_irqs()' main loop. This is too late, however, because a possible affinity change may incur the need for vector assignment, which will fail when the IRQ's move cleanup count is still non-zero. To fix this - record the set of CPUs the cleanup IPIs gets actually sent to alongside setting their count, - adjust the count in fixup_irqs(), accounting for all CPUs that the cleanup IPI was sent to, but that are no longer online, - bail early from the cleanup IPI handler when the CPU is no longer online, to prevent double accounting. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -668,6 +668,9 @@ void irq_move_cleanup_interrupt(struct c ack_APIC_irq(); me = smp_processor_id(); + if ( !cpu_online(me) ) + return; + for ( vector = FIRST_DYNAMIC_VECTOR; vector <= LAST_HIPRIORITY_VECTOR; vector++) { @@ -727,11 +730,14 @@ unlock: static void send_cleanup_vector(struct irq_desc *desc) { - cpumask_t cleanup_mask; + cpumask_and(desc->arch.old_cpu_mask, desc->arch.old_cpu_mask, + &cpu_online_map); + desc->arch.move_cleanup_count = cpumask_weight(desc->arch.old_cpu_mask); - cpumask_and(&cleanup_mask, desc->arch.old_cpu_mask, &cpu_online_map); - desc->arch.move_cleanup_count = cpumask_weight(&cleanup_mask); - send_IPI_mask(&cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR); + if ( desc->arch.move_cleanup_count ) + send_IPI_mask(desc->arch.old_cpu_mask, IRQ_MOVE_CLEANUP_VECTOR); + else + release_old_vec(desc); desc->arch.move_in_progress = 0; } @@ -2410,6 +2416,16 @@ void fixup_irqs(const cpumask_t *mask, b vector <= LAST_HIPRIORITY_VECTOR ) cpumask_and(desc->arch.cpu_mask, desc->arch.cpu_mask, mask); + if ( desc->arch.move_cleanup_count ) + { + /* The cleanup IPI may have got sent while we were still online. */ + cpumask_andnot(&affinity, desc->arch.old_cpu_mask, + &cpu_online_map); + desc->arch.move_cleanup_count -= cpumask_weight(&affinity); + if ( !desc->arch.move_cleanup_count ) + release_old_vec(desc); + } + cpumask_copy(&affinity, desc->affinity); if ( !desc->action || cpumask_subset(&affinity, mask) ) { _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:45, Jan Beulich wrote: > The cleanup IPI may get sent immediately before a CPU gets removed from > the online map. In such a case the IPI would get handled on the CPU > being offlined no earlier than in the interrupts disabled window after > fixup_irqs()' main loop. This is too late, however, because a possible > affinity change may incur the need for vector assignment, which will > fail when the IRQ's move cleanup count is still non-zero. > > To fix this > - record the set of CPUs the cleanup IPIs gets actually sent to alongside > setting their count, > - adjust the count in fixup_irqs(), accounting for all CPUs that the > cleanup IPI was sent to, but that are no longer online, > - bail early from the cleanup IPI handler when the CPU is no longer > online, to prevent double accounting. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Don't log a stray trailing comma. Shorten a few fields. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -2334,7 +2334,7 @@ static void dump_irqs(unsigned char key) spin_lock_irqsave(&desc->lock, flags); - printk(" IRQ:%4d affinity:%*pb vec:%02x type=%-15s status=%08x ", + printk(" IRQ:%4d aff:%*pb vec:%02x %-15s status=%03x ", irq, nr_cpu_ids, cpumask_bits(desc->affinity), desc->arch.vector, desc->handler->typename, desc->status); @@ -2345,23 +2345,21 @@ static void dump_irqs(unsigned char key) { action = (irq_guest_action_t *)desc->action; - printk("in-flight=%d domain-list=", action->in_flight); + printk("in-flight=%d%c", + action->in_flight, action->nr_guests ? ' ' : '\n'); - for ( i = 0; i < action->nr_guests; i++ ) + for ( i = 0; i < action->nr_guests; ) { - d = action->guest[i]; + d = action->guest[i++]; pirq = domain_irq_to_pirq(d, irq); info = pirq_info(d, pirq); - printk("%u:%3d(%c%c%c)", + printk("d%d:%3d(%c%c%c)%c", d->domain_id, pirq, evtchn_port_is_pending(d, info->evtchn) ? 'P' : '-', evtchn_port_is_masked(d, info->evtchn) ? 'M' : '-', - (info->masked ? 'M' : '-')); - if ( i != action->nr_guests ) - printk(","); + info->masked ? 'M' : '-', + i < action->nr_guests ? ',' : '\n'); } - - printk("\n"); } else if ( desc->action ) printk("%ps()\n", desc->action->handler); _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:46, Jan Beulich wrote: > Don't log a stray trailing comma. Shorten a few fields. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
desc->arch.cpu_mask reflects the actual set of target CPUs. Don't ever fiddle with desc->affinity itself, except to store caller requested values. Note that assign_irq_vector() now takes a NULL incoming CPU mask to mean "all CPUs" now, rather than just "all currently online CPUs". This way no further affinity adjustment is needed after onlining further CPUs. This renders both set_native_irq_info() uses (which weren't using proper locking anyway) redundant - drop the function altogether. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- a/xen/arch/x86/io_apic.c +++ b/xen/arch/x86/io_apic.c @@ -1039,7 +1039,6 @@ static void __init setup_IO_APIC_irqs(vo SET_DEST(entry, logical, cpu_mask_to_apicid(TARGET_CPUS)); spin_lock_irqsave(&ioapic_lock, flags); __ioapic_write_entry(apic, pin, 0, entry); - set_native_irq_info(irq, TARGET_CPUS); spin_unlock_irqrestore(&ioapic_lock, flags); } } @@ -2248,7 +2247,6 @@ int io_apic_set_pci_routing (int ioapic, spin_lock_irqsave(&ioapic_lock, flags); __ioapic_write_entry(ioapic, pin, 0, entry); - set_native_irq_info(irq, TARGET_CPUS); spin_unlock(&ioapic_lock); spin_lock(&desc->lock); --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -582,11 +582,16 @@ int assign_irq_vector(int irq, const cpu spin_lock_irqsave(&vector_lock, flags); ret = __assign_irq_vector(irq, desc, mask ?: TARGET_CPUS); - if (!ret) { + if ( !ret ) + { ret = desc->arch.vector; - cpumask_copy(desc->affinity, desc->arch.cpu_mask); + if ( mask ) + cpumask_copy(desc->affinity, mask); + else + cpumask_setall(desc->affinity); } spin_unlock_irqrestore(&vector_lock, flags); + return ret; } @@ -2334,9 +2339,10 @@ static void dump_irqs(unsigned char key) spin_lock_irqsave(&desc->lock, flags); - printk(" IRQ:%4d aff:%*pb vec:%02x %-15s status=%03x ", - irq, nr_cpu_ids, cpumask_bits(desc->affinity), desc->arch.vector, - desc->handler->typename, desc->status); + printk(" IRQ:%4d aff:%*pb/%*pb vec:%02x %-15s status=%03x ", + irq, nr_cpu_ids, cpumask_bits(desc->affinity), + nr_cpu_ids, cpumask_bits(desc->arch.cpu_mask), + desc->arch.vector, desc->handler->typename, desc->status); if ( ssid ) printk("Z=%-25s ", ssid); @@ -2424,8 +2430,7 @@ void fixup_irqs(const cpumask_t *mask, b release_old_vec(desc); } - cpumask_copy(&affinity, desc->affinity); - if ( !desc->action || cpumask_subset(&affinity, mask) ) + if ( !desc->action || cpumask_subset(desc->affinity, mask) ) { spin_unlock(&desc->lock); continue; @@ -2458,12 +2463,13 @@ void fixup_irqs(const cpumask_t *mask, b desc->arch.move_in_progress = 0; } - cpumask_and(&affinity, &affinity, mask); - if ( cpumask_empty(&affinity) ) + if ( !cpumask_intersects(mask, desc->affinity) ) { break_affinity = true; - cpumask_copy(&affinity, mask); + cpumask_setall(&affinity); } + else + cpumask_copy(&affinity, desc->affinity); if ( desc->handler->disable ) desc->handler->disable(desc); --- a/xen/include/xen/irq.h +++ b/xen/include/xen/irq.h @@ -162,11 +162,6 @@ extern irq_desc_t *domain_spin_lock_irq_ extern irq_desc_t *pirq_spin_lock_irq_desc( const struct pirq *, unsigned long *pflags); -static inline void set_native_irq_info(unsigned int irq, const cpumask_t *mask) -{ - cpumask_copy(irq_to_desc(irq)->affinity, mask); -} - unsigned int set_desc_affinity(struct irq_desc *, const cpumask_t *); #ifndef arch_hwdom_irqs _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:46, Jan Beulich wrote: > @@ -2334,9 +2339,10 @@ static void dump_irqs(unsigned char key) > > spin_lock_irqsave(&desc->lock, flags); > > - printk(" IRQ:%4d aff:%*pb vec:%02x %-15s status=%03x ", > - irq, nr_cpu_ids, cpumask_bits(desc->affinity), desc->arch.vector, > - desc->handler->typename, desc->status); > + printk(" IRQ:%4d aff:%*pb/%*pb vec:%02x %-15s status=%03x ", > + irq, nr_cpu_ids, cpumask_bits(desc->affinity), > + nr_cpu_ids, cpumask_bits(desc->arch.cpu_mask), > + desc->arch.vector, desc->handler->typename, desc->status); Taking a sample large system (Rome, with your x2apic series to be specific), which is only half as large as typical high-end Skylake systems. (XEN) IRQ information: (XEN) IRQ: 0 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:f0 type=IO-APIC-edge status=00000000 time.c#timer_interrupt() (XEN) IRQ: 1 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:68 type=IO-APIC-edge status=00000002 mapped, unbound (XEN) IRQ: 3 affinity:ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff vec:70 type=IO-APIC-edge status=00000002 mapped, unbound (XEN) IRQ: 4 affinity:ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff vec:f1 type=IO-APIC-edge status=00000000 ns16550.c#ns16550_interrupt() (XEN) IRQ: 5 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:78 type=IO-APIC-edge status=00000002 mapped, unbound (XEN) IRQ: 6 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:88 type=IO-APIC-edge status=00000002 mapped, unbound (XEN) IRQ: 7 affinity:ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff vec:90 type=IO-APIC-level status=00000002 mapped, unbound (XEN) IRQ: 8 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:98 type=IO-APIC-edge status=00000030 in-flight=0 domain-list=0: 8(---), (XEN) IRQ: 9 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:a0 type=IO-APIC-level status=00000030 in-flight=1 domain-list=0: 9(PMM), This change is going to double up the affinity block, which will make the lines even longer. Given that all examples I've ever spotted are either a single bit, or a fully set block, {%*pbl} will render in a much shorter, and keep the line length reasonable. (This in practice applies to the previous patch as well). ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 03.07.2019 19:58, Andrew Cooper wrote: > On 17/05/2019 11:46, Jan Beulich wrote: >> @@ -2334,9 +2339,10 @@ static void dump_irqs(unsigned char key) >> >> spin_lock_irqsave(&desc->lock, flags); >> >> - printk(" IRQ:%4d aff:%*pb vec:%02x %-15s status=%03x ", >> - irq, nr_cpu_ids, cpumask_bits(desc->affinity), desc->arch.vector, >> - desc->handler->typename, desc->status); >> + printk(" IRQ:%4d aff:%*pb/%*pb vec:%02x %-15s status=%03x ", >> + irq, nr_cpu_ids, cpumask_bits(desc->affinity), >> + nr_cpu_ids, cpumask_bits(desc->arch.cpu_mask), >> + desc->arch.vector, desc->handler->typename, desc->status); > > Taking a sample large system (Rome, with your x2apic series to be > specific), which is only half as large as typical high-end Skylake systems. > > (XEN) IRQ information: > (XEN) IRQ: 0 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:f0 type=IO-APIC-edge status=00000000 time.c#timer_interrupt() > (XEN) IRQ: 1 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:68 type=IO-APIC-edge status=00000002 mapped, unbound > (XEN) IRQ: 3 affinity:ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff vec:70 type=IO-APIC-edge status=00000002 mapped, unbound > (XEN) IRQ: 4 affinity:ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff vec:f1 type=IO-APIC-edge status=00000000 ns16550.c#ns16550_interrupt() > (XEN) IRQ: 5 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:78 type=IO-APIC-edge status=00000002 mapped, unbound > (XEN) IRQ: 6 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:88 type=IO-APIC-edge status=00000002 mapped, unbound > (XEN) IRQ: 7 affinity:ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff,ffffffff vec:90 type=IO-APIC-level status=00000002 mapped, unbound > (XEN) IRQ: 8 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:98 type=IO-APIC-edge status=00000030 in-flight=0 domain-list=0: 8(---), > (XEN) IRQ: 9 affinity:00000000,00000000,00000000,00000000,00000000,00000000,00000000,00000001 vec:a0 type=IO-APIC-level status=00000030 in-flight=1 domain-list=0: 9(PMM), > > This change is going to double up the affinity block, which will make > the lines even longer. > > Given that all examples I've ever spotted are either a single bit, or a > fully set block, {%*pbl} will render in a much shorter, and keep the > line length reasonable. (This in practice applies to the previous patch > as well). With SMT off (on Intel systems) I've certainly observed every other bit being set, which is why I had specifically decided against %*pbl. Plus using %*pbl would break the tabular formatting. The only middle ground I could see (still having the undesirable latter effect) would be to pick between both forms based on the ratio between set bits and total number of them (and perhaps using %*pb as long as the total number of them is below a certain threshold). Thoughts? Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Mixed meaning was implied so far by different pieces of code - disagreement was in particular about whether to expect offline CPUs' bits to possibly be set. Switch to a mostly consistent meaning (exception being high priority interrupts, which would perhaps better be switched to the same model as well in due course). Use the field to record the vector allocation mask, i.e. potentially including bits of offline (parked) CPUs. This implies that before passing the mask to certain functions (most notably cpu_mask_to_apicid()) it needs to be further reduced to the online subset. The exception of high priority interrupts is also why for the moment _bind_irq_vector() is left as is, despite looking wrong: It's used exclusively for IRQ0, which isn't supposed to move off CPU0 at any time. The prior lack of restricting to online CPUs in set_desc_affinity() before calling cpu_mask_to_apicid() in particular allowed (in x2APIC clustered mode) offlined CPUs to end up enabled in an IRQ's destination field. (I wonder whether vector_allocation_cpumask_flat() shouldn't follow a similar model, using cpu_present_map in favor of cpu_online_map.) For IO-APIC code it was definitely wrong to potentially store, as a fallback, TARGET_CPUS (i.e. all online ones) into the field, as that would have caused problems when determining on which CPUs to release vectors when they've gone out of use. Disable interrupts instead when no valid target CPU can be established (which code elsewhere should guarantee to never happen), and log a message in such an unlikely event. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- v2: New. --- a/xen/arch/x86/io_apic.c +++ b/xen/arch/x86/io_apic.c @@ -680,7 +680,7 @@ void /*__init*/ setup_ioapic_dest(void) continue; irq = pin_2_irq(irq_entry, ioapic, pin); desc = irq_to_desc(irq); - BUG_ON(cpumask_empty(desc->arch.cpu_mask)); + BUG_ON(!cpumask_intersects(desc->arch.cpu_mask, &cpu_online_map)); set_ioapic_affinity_irq(desc, desc->arch.cpu_mask); } @@ -2194,7 +2194,6 @@ int io_apic_set_pci_routing (int ioapic, { struct irq_desc *desc = irq_to_desc(irq); struct IO_APIC_route_entry entry; - cpumask_t mask; unsigned long flags; int vector; @@ -2229,11 +2228,17 @@ int io_apic_set_pci_routing (int ioapic, return vector; entry.vector = vector; - cpumask_copy(&mask, TARGET_CPUS); - /* Don't chance ending up with an empty mask. */ - if (cpumask_intersects(&mask, desc->arch.cpu_mask)) - cpumask_and(&mask, &mask, desc->arch.cpu_mask); - SET_DEST(entry, logical, cpu_mask_to_apicid(&mask)); + if (cpumask_intersects(desc->arch.cpu_mask, TARGET_CPUS)) { + cpumask_t *mask = this_cpu(scratch_cpumask); + + cpumask_and(mask, desc->arch.cpu_mask, TARGET_CPUS); + SET_DEST(entry, logical, cpu_mask_to_apicid(mask)); + } else { + printk(XENLOG_ERR "IRQ%d: no target CPU (%*pb vs %*pb)\n", + irq, nr_cpu_ids, cpumask_bits(desc->arch.cpu_mask), + nr_cpu_ids, cpumask_bits(TARGET_CPUS)); + desc->status |= IRQ_DISABLED; + } apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry " "(%d-%d -> %#x -> IRQ %d Mode:%i Active:%i)\n", ioapic, @@ -2419,7 +2424,21 @@ int ioapic_guest_write(unsigned long phy /* Set the vector field to the real vector! */ rte.vector = desc->arch.vector; - SET_DEST(rte, logical, cpu_mask_to_apicid(desc->arch.cpu_mask)); + if ( cpumask_intersects(desc->arch.cpu_mask, TARGET_CPUS) ) + { + cpumask_t *mask = this_cpu(scratch_cpumask); + + cpumask_and(mask, desc->arch.cpu_mask, TARGET_CPUS); + SET_DEST(rte, logical, cpu_mask_to_apicid(mask)); + } + else + { + gprintk(XENLOG_ERR, "IRQ%d: no target CPU (%*pb vs %*pb)\n", + irq, nr_cpu_ids, cpumask_bits(desc->arch.cpu_mask), + nr_cpu_ids, cpumask_bits(TARGET_CPUS)); + desc->status |= IRQ_DISABLED; + rte.mask = 1; + } __ioapic_write_entry(apic, pin, 0, rte); --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -471,11 +471,13 @@ static int __assign_irq_vector( */ static int current_vector = FIRST_DYNAMIC_VECTOR, current_offset = 0; int cpu, err, old_vector; - cpumask_t tmp_mask; vmask_t *irq_used_vectors = NULL; old_vector = irq_to_vector(irq); - if (old_vector > 0) { + if ( old_vector > 0 ) + { + cpumask_t tmp_mask; + cpumask_and(&tmp_mask, mask, &cpu_online_map); if (cpumask_intersects(&tmp_mask, desc->arch.cpu_mask)) { desc->arch.vector = old_vector; @@ -498,7 +500,9 @@ static int __assign_irq_vector( else irq_used_vectors = irq_get_used_vector_mask(irq); - for_each_cpu(cpu, mask) { + for_each_cpu(cpu, mask) + { + const cpumask_t *vec_mask; int new_cpu; int vector, offset; @@ -506,8 +510,7 @@ static int __assign_irq_vector( if (!cpu_online(cpu)) continue; - cpumask_and(&tmp_mask, vector_allocation_cpumask(cpu), - &cpu_online_map); + vec_mask = vector_allocation_cpumask(cpu); vector = current_vector; offset = current_offset; @@ -528,7 +531,7 @@ next: && test_bit(vector, irq_used_vectors) ) goto next; - for_each_cpu(new_cpu, &tmp_mask) + for_each_cpu(new_cpu, vec_mask) if (per_cpu(vector_irq, new_cpu)[vector] >= 0) goto next; /* Found one! */ @@ -547,12 +550,12 @@ next: release_old_vec(desc); } - trace_irq_mask(TRC_HW_IRQ_ASSIGN_VECTOR, irq, vector, &tmp_mask); + trace_irq_mask(TRC_HW_IRQ_ASSIGN_VECTOR, irq, vector, vec_mask); - for_each_cpu(new_cpu, &tmp_mask) + for_each_cpu(new_cpu, vec_mask) per_cpu(vector_irq, new_cpu)[vector] = irq; desc->arch.vector = vector; - cpumask_copy(desc->arch.cpu_mask, &tmp_mask); + cpumask_copy(desc->arch.cpu_mask, vec_mask); desc->arch.used = IRQ_USED; ASSERT((desc->arch.used_vectors == NULL) @@ -783,6 +786,7 @@ unsigned int set_desc_affinity(struct ir cpumask_copy(desc->affinity, mask); cpumask_and(&dest_mask, mask, desc->arch.cpu_mask); + cpumask_and(&dest_mask, &dest_mask, &cpu_online_map); return cpu_mask_to_apicid(&dest_mask); } --- a/xen/include/asm-x86/irq.h +++ b/xen/include/asm-x86/irq.h @@ -32,6 +32,12 @@ struct irq_desc; struct arch_irq_desc { s16 vector; /* vector itself is only 8 bits, */ s16 old_vector; /* but we use -1 for unassigned */ + /* + * Except for high priority interrupts @cpu_mask may have bits set for + * offline CPUs. Consumers need to be careful to mask this down to + * online ones as necessary. There is supposed to always be a non- + * empty intersection with cpu_online_map. + */ cpumask_var_t cpu_mask; cpumask_var_t old_cpu_mask; cpumask_var_t pending_mask; _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:47, Jan Beulich wrote: > Mixed meaning was implied so far by different pieces of code - > disagreement was in particular about whether to expect offline CPUs' > bits to possibly be set. Switch to a mostly consistent meaning > (exception being high priority interrupts, which would perhaps better > be switched to the same model as well in due course). Use the field to > record the vector allocation mask, i.e. potentially including bits of > offline (parked) CPUs. This implies that before passing the mask to > certain functions (most notably cpu_mask_to_apicid()) it needs to be > further reduced to the online subset. > > The exception of high priority interrupts is also why for the moment > _bind_irq_vector() is left as is, despite looking wrong: It's used > exclusively for IRQ0, which isn't supposed to move off CPU0 at any time. > > The prior lack of restricting to online CPUs in set_desc_affinity() > before calling cpu_mask_to_apicid() in particular allowed (in x2APIC > clustered mode) offlined CPUs to end up enabled in an IRQ's destination > field. (I wonder whether vector_allocation_cpumask_flat() shouldn't > follow a similar model, using cpu_present_map in favor of > cpu_online_map.) > > For IO-APIC code it was definitely wrong to potentially store, as a > fallback, TARGET_CPUS (i.e. all online ones) into the field, as that > would have caused problems when determining on which CPUs to release > vectors when they've gone out of use. Disable interrupts instead when > no valid target CPU can be established (which code elsewhere should > guarantee to never happen), and log a message in such an unlikely event. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
All of __{assign,bind,clear}_irq_vector() manipulate struct irq_desc fields, and hence ought to be called with the descriptor lock held in addition to vector_lock. This is currently the case for only set_desc_affinity() (in the common case) and destroy_irq(), which also clarifies what the nesting behavior between the locks has to be. Reflect the new expectation by having these functions all take a descriptor as parameter instead of an interrupt number. Also take care of the two special cases of calls to set_desc_affinity(): set_ioapic_affinity_irq() and VT-d's dma_msi_set_affinity() get called directly as well, and in these cases the descriptor locks hadn't got acquired till now. For set_ioapic_affinity_irq() this means acquiring / releasing of the IO-APIC lock can be plain spin_{,un}lock() then. Drop one of the two leading underscores from all three functions at the same time. There's one case left where descriptors get manipulated with just vector_lock held: setup_vector_irq() assumes its caller to acquire vector_lock, and hence can't itself acquire the descriptor locks (wrong lock order). I don't currently see how to address this. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Kevin Tian <kevin.tian@intel.com> [VT-d] Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- v3: Also drop one leading underscore from a comment. Re-base. v2: Also adjust set_ioapic_affinity_irq() and VT-d's dma_msi_set_affinity(). --- a/xen/arch/x86/io_apic.c +++ b/xen/arch/x86/io_apic.c @@ -550,14 +550,14 @@ static void clear_IO_APIC (void) static void set_ioapic_affinity_irq(struct irq_desc *desc, const cpumask_t *mask) { - unsigned long flags; unsigned int dest; int pin, irq; struct irq_pin_list *entry; irq = desc->irq; - spin_lock_irqsave(&ioapic_lock, flags); + spin_lock(&ioapic_lock); + dest = set_desc_affinity(desc, mask); if (dest != BAD_APICID) { if ( !x2apic_enabled ) @@ -580,8 +580,8 @@ set_ioapic_affinity_irq(struct irq_desc entry = irq_2_pin + entry->next; } } - spin_unlock_irqrestore(&ioapic_lock, flags); + spin_unlock(&ioapic_lock); } /* @@ -674,16 +674,19 @@ void /*__init*/ setup_ioapic_dest(void) for (ioapic = 0; ioapic < nr_ioapics; ioapic++) { for (pin = 0; pin < nr_ioapic_entries[ioapic]; pin++) { struct irq_desc *desc; + unsigned long flags; irq_entry = find_irq_entry(ioapic, pin, mp_INT); if (irq_entry == -1) continue; irq = pin_2_irq(irq_entry, ioapic, pin); desc = irq_to_desc(irq); + + spin_lock_irqsave(&desc->lock, flags); BUG_ON(!cpumask_intersects(desc->arch.cpu_mask, &cpu_online_map)); set_ioapic_affinity_irq(desc, desc->arch.cpu_mask); + spin_unlock_irqrestore(&desc->lock, flags); } - } } --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -27,6 +27,7 @@ #include <public/physdev.h> static int parse_irq_vector_map_param(const char *s); +static void _clear_irq_vector(struct irq_desc *desc); /* opt_noirqbalance: If true, software IRQ balancing/affinity is disabled. */ bool __read_mostly opt_noirqbalance; @@ -136,13 +137,12 @@ static void trace_irq_mask(uint32_t even trace_var(event, 1, sizeof(d), &d); } -static int __init __bind_irq_vector(int irq, int vector, const cpumask_t *cpu_mask) +static int __init _bind_irq_vector(struct irq_desc *desc, int vector, + const cpumask_t *cpu_mask) { cpumask_t online_mask; int cpu; - struct irq_desc *desc = irq_to_desc(irq); - BUG_ON((unsigned)irq >= nr_irqs); BUG_ON((unsigned)vector >= NR_VECTORS); cpumask_and(&online_mask, cpu_mask, &cpu_online_map); @@ -153,9 +153,9 @@ static int __init __bind_irq_vector(int return 0; if ( desc->arch.vector != IRQ_VECTOR_UNASSIGNED ) return -EBUSY; - trace_irq_mask(TRC_HW_IRQ_BIND_VECTOR, irq, vector, &online_mask); + trace_irq_mask(TRC_HW_IRQ_BIND_VECTOR, desc->irq, vector, &online_mask); for_each_cpu(cpu, &online_mask) - per_cpu(vector_irq, cpu)[vector] = irq; + per_cpu(vector_irq, cpu)[vector] = desc->irq; desc->arch.vector = vector; cpumask_copy(desc->arch.cpu_mask, &online_mask); if ( desc->arch.used_vectors ) @@ -169,12 +169,18 @@ static int __init __bind_irq_vector(int int __init bind_irq_vector(int irq, int vector, const cpumask_t *cpu_mask) { + struct irq_desc *desc = irq_to_desc(irq); unsigned long flags; int ret; - spin_lock_irqsave(&vector_lock, flags); - ret = __bind_irq_vector(irq, vector, cpu_mask); - spin_unlock_irqrestore(&vector_lock, flags); + BUG_ON((unsigned)irq >= nr_irqs); + + spin_lock_irqsave(&desc->lock, flags); + spin_lock(&vector_lock); + ret = _bind_irq_vector(desc, vector, cpu_mask); + spin_unlock(&vector_lock); + spin_unlock_irqrestore(&desc->lock, flags); + return ret; } @@ -259,18 +265,20 @@ void destroy_irq(unsigned int irq) spin_lock_irqsave(&desc->lock, flags); desc->handler = &no_irq_type; - clear_irq_vector(irq); + spin_lock(&vector_lock); + _clear_irq_vector(desc); + spin_unlock(&vector_lock); desc->arch.used_vectors = NULL; spin_unlock_irqrestore(&desc->lock, flags); xfree(action); } -static void __clear_irq_vector(int irq) +static void _clear_irq_vector(struct irq_desc *desc) { - int cpu, vector, old_vector; + unsigned int cpu; + int vector, old_vector, irq = desc->irq; cpumask_t tmp_mask; - struct irq_desc *desc = irq_to_desc(irq); BUG_ON(!desc->arch.vector); @@ -316,11 +324,14 @@ static void __clear_irq_vector(int irq) void clear_irq_vector(int irq) { + struct irq_desc *desc = irq_to_desc(irq); unsigned long flags; - spin_lock_irqsave(&vector_lock, flags); - __clear_irq_vector(irq); - spin_unlock_irqrestore(&vector_lock, flags); + spin_lock_irqsave(&desc->lock, flags); + spin_lock(&vector_lock); + _clear_irq_vector(desc); + spin_unlock(&vector_lock); + spin_unlock_irqrestore(&desc->lock, flags); } int irq_to_vector(int irq) @@ -455,8 +466,7 @@ static vmask_t *irq_get_used_vector_mask return ret; } -static int __assign_irq_vector( - int irq, struct irq_desc *desc, const cpumask_t *mask) +static int _assign_irq_vector(struct irq_desc *desc, const cpumask_t *mask) { /* * NOTE! The local APIC isn't very good at handling @@ -470,7 +480,8 @@ static int __assign_irq_vector( * 0x80, because int 0x80 is hm, kind of importantish. ;) */ static int current_vector = FIRST_DYNAMIC_VECTOR, current_offset = 0; - int cpu, err, old_vector; + unsigned int cpu; + int err, old_vector, irq = desc->irq; vmask_t *irq_used_vectors = NULL; old_vector = irq_to_vector(irq); @@ -583,8 +594,12 @@ int assign_irq_vector(int irq, const cpu BUG_ON(irq >= nr_irqs || irq <0); - spin_lock_irqsave(&vector_lock, flags); - ret = __assign_irq_vector(irq, desc, mask ?: TARGET_CPUS); + spin_lock_irqsave(&desc->lock, flags); + + spin_lock(&vector_lock); + ret = _assign_irq_vector(desc, mask ?: TARGET_CPUS); + spin_unlock(&vector_lock); + if ( !ret ) { ret = desc->arch.vector; @@ -593,7 +608,8 @@ int assign_irq_vector(int irq, const cpu else cpumask_setall(desc->affinity); } - spin_unlock_irqrestore(&vector_lock, flags); + + spin_unlock_irqrestore(&desc->lock, flags); return ret; } @@ -767,7 +783,6 @@ void irq_complete_move(struct irq_desc * unsigned int set_desc_affinity(struct irq_desc *desc, const cpumask_t *mask) { - unsigned int irq; int ret; unsigned long flags; cpumask_t dest_mask; @@ -775,10 +790,8 @@ unsigned int set_desc_affinity(struct ir if (!cpumask_intersects(mask, &cpu_online_map)) return BAD_APICID; - irq = desc->irq; - spin_lock_irqsave(&vector_lock, flags); - ret = __assign_irq_vector(irq, desc, mask); + ret = _assign_irq_vector(desc, mask); spin_unlock_irqrestore(&vector_lock, flags); if (ret < 0) @@ -2442,7 +2455,7 @@ void fixup_irqs(const cpumask_t *mask, b /* * In order for the affinity adjustment below to be successful, we - * need __assign_irq_vector() to succeed. This in particular means + * need _assign_irq_vector() to succeed. This in particular means * clearing desc->arch.move_in_progress if this would otherwise * prevent the function from succeeding. Since there's no way for the * flag to get cleared anymore when there's no possible destination --- a/xen/drivers/passthrough/vtd/iommu.c +++ b/xen/drivers/passthrough/vtd/iommu.c @@ -2134,11 +2134,16 @@ static void adjust_irq_affinity(struct a unsigned int node = rhsa ? pxm_to_node(rhsa->proximity_domain) : NUMA_NO_NODE; const cpumask_t *cpumask = &cpu_online_map; + struct irq_desc *desc; if ( node < MAX_NUMNODES && node_online(node) && cpumask_intersects(&node_to_cpumask(node), cpumask) ) cpumask = &node_to_cpumask(node); - dma_msi_set_affinity(irq_to_desc(drhd->iommu->msi.irq), cpumask); + + desc = irq_to_desc(drhd->iommu->msi.irq); + spin_lock_irq(&desc->lock); + dma_msi_set_affinity(desc, cpumask); + spin_unlock_irq(&desc->lock); } static int adjust_vtd_irq_affinities(void) _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:47, Jan Beulich wrote: > All of __{assign,bind,clear}_irq_vector() manipulate struct irq_desc > fields, and hence ought to be called with the descriptor lock held in > addition to vector_lock. This is currently the case for only > set_desc_affinity() (in the common case) and destroy_irq(), which also > clarifies what the nesting behavior between the locks has to be. > Reflect the new expectation by having these functions all take a > descriptor as parameter instead of an interrupt number. > > Also take care of the two special cases of calls to set_desc_affinity(): > set_ioapic_affinity_irq() and VT-d's dma_msi_set_affinity() get called > directly as well, and in these cases the descriptor locks hadn't got > acquired till now. For set_ioapic_affinity_irq() this means acquiring / > releasing of the IO-APIC lock can be plain spin_{,un}lock() then. > > Drop one of the two leading underscores from all three functions at > the same time. > > There's one case left where descriptors get manipulated with just > vector_lock held: setup_vector_irq() assumes its caller to acquire > vector_lock, and hence can't itself acquire the descriptor locks (wrong > lock order). I don't currently see how to address this. In practice, the only mutation is setting a bit in cpu_mask for the shared high priority vectors, so looks to be safe in practice. The callers use of the vector_lock looks like a bodge though. However, this analysis needs to be added to the comment for setup_vector_irq(), because the behaviour is extremely fragile and mustn't change. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Kevin Tian <kevin.tian@intel.com> [VT-d] > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> With some form of adjustment to the comment for setup_vector_irq(), and ideally to the commit message about safety in practice, Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 03.07.2019 20:23, Andrew Cooper wrote: > On 17/05/2019 11:47, Jan Beulich wrote: >> All of __{assign,bind,clear}_irq_vector() manipulate struct irq_desc >> fields, and hence ought to be called with the descriptor lock held in >> addition to vector_lock. This is currently the case for only >> set_desc_affinity() (in the common case) and destroy_irq(), which also >> clarifies what the nesting behavior between the locks has to be. >> Reflect the new expectation by having these functions all take a >> descriptor as parameter instead of an interrupt number. >> >> Also take care of the two special cases of calls to set_desc_affinity(): >> set_ioapic_affinity_irq() and VT-d's dma_msi_set_affinity() get called >> directly as well, and in these cases the descriptor locks hadn't got >> acquired till now. For set_ioapic_affinity_irq() this means acquiring / >> releasing of the IO-APIC lock can be plain spin_{,un}lock() then. >> >> Drop one of the two leading underscores from all three functions at >> the same time. >> >> There's one case left where descriptors get manipulated with just >> vector_lock held: setup_vector_irq() assumes its caller to acquire >> vector_lock, and hence can't itself acquire the descriptor locks (wrong >> lock order). I don't currently see how to address this. > > In practice, the only mutation is setting a bit in cpu_mask for the > shared high priority vectors, so looks to be safe in practice. I had tried to convince myself that it's safe in practice, but I'm afraid I couldn't (and hence wouldn't want to say so in the patch description here). There's one important thing to pay attention to: Not all manipulations of ->arch.cpu_mask are atomic (and there's really no way for them to be, with our current cpumask infrastructure) - all of them assume to be done under lock. And other than when offlining a CPU we're not in a fully synchronized state while onlining one. > The callers use of the vector_lock looks like a bodge though. Well, it's definitely not nice, but unavoidable (afaict). > However, this analysis needs to be added to the comment for > setup_vector_irq(), because the behaviour is extremely fragile and > mustn't change. Will do. >> Signed-off-by: Jan Beulich <jbeulich@suse.com> >> Reviewed-by: Kevin Tian <kevin.tian@intel.com> [VT-d] >> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> > > With some form of adjustment to the comment for setup_vector_irq(), and > ideally to the commit message about safety in practice, Acked-by: Andrew > Cooper <andrew.cooper3@citrix.com> Thanks. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
fixup_irqs() skips interrupts without action. Hence such interrupts can retain affinity to just offline CPUs. With "noirqbalance" in effect, pirq_guest_bind() so far would have left them alone, resulting in a non- working interrupt. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- v3: New. --- I've not observed this problem in practice - the change is just the result of code inspection after having noticed action-less IRQs in 'i' debug key output pointing at all parked/offline CPUs. --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -1683,9 +1683,27 @@ int pirq_guest_bind(struct vcpu *v, stru desc->status |= IRQ_GUEST; - /* Attempt to bind the interrupt target to the correct CPU. */ - if ( !opt_noirqbalance && (desc->handler->set_affinity != NULL) ) - desc->handler->set_affinity(desc, cpumask_of(v->processor)); + /* + * Attempt to bind the interrupt target to the correct (or at least + * some online) CPU. + */ + if ( desc->handler->set_affinity ) + { + const cpumask_t *affinity = NULL; + + if ( !opt_noirqbalance ) + affinity = cpumask_of(v->processor); + else if ( !cpumask_intersects(desc->affinity, &cpu_online_map) ) + { + cpumask_setall(desc->affinity); + affinity = &cpumask_all; + } + else if ( !cpumask_intersects(desc->arch.cpu_mask, + &cpu_online_map) ) + affinity = desc->affinity; + if ( affinity ) + desc->handler->set_affinity(desc, affinity); + } desc->status &= ~IRQ_DISABLED; desc->handler->startup(desc); _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On Fri, May 17, 2019 at 04:48:21AM -0600, Jan Beulich wrote: > fixup_irqs() skips interrupts without action. Hence such interrupts can > retain affinity to just offline CPUs. With "noirqbalance" in effect, > pirq_guest_bind() so far would have left them alone, resulting in a non- > working interrupt. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > --- > v3: New. > --- > I've not observed this problem in practice - the change is just the > result of code inspection after having noticed action-less IRQs in 'i' > debug key output pointing at all parked/offline CPUs. > > --- a/xen/arch/x86/irq.c > +++ b/xen/arch/x86/irq.c > @@ -1683,9 +1683,27 @@ int pirq_guest_bind(struct vcpu *v, stru > > desc->status |= IRQ_GUEST; > > - /* Attempt to bind the interrupt target to the correct CPU. */ > - if ( !opt_noirqbalance && (desc->handler->set_affinity != NULL) ) > - desc->handler->set_affinity(desc, cpumask_of(v->processor)); > + /* > + * Attempt to bind the interrupt target to the correct (or at least > + * some online) CPU. > + */ > + if ( desc->handler->set_affinity ) > + { > + const cpumask_t *affinity = NULL; > + > + if ( !opt_noirqbalance ) > + affinity = cpumask_of(v->processor); > + else if ( !cpumask_intersects(desc->affinity, &cpu_online_map) ) > + { > + cpumask_setall(desc->affinity); > + affinity = &cpumask_all; > + } > + else if ( !cpumask_intersects(desc->arch.cpu_mask, > + &cpu_online_map) ) I'm not sure I see the purpose of the desc->arch.cpu_mask check, wouldn't it be better to just use else and set the affinity to desc->affinity? Or it's just an optimization to avoid doing the set_affinity call if the interrupt it already bound to an online CPU? Thanks, Roger. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
>>> On 20.05.19 at 13:40, <roger.pau@citrix.com> wrote: > On Fri, May 17, 2019 at 04:48:21AM -0600, Jan Beulich wrote: >> fixup_irqs() skips interrupts without action. Hence such interrupts can >> retain affinity to just offline CPUs. With "noirqbalance" in effect, >> pirq_guest_bind() so far would have left them alone, resulting in a non- >> working interrupt. >> >> Signed-off-by: Jan Beulich <jbeulich@suse.com> >> --- >> v3: New. >> --- >> I've not observed this problem in practice - the change is just the >> result of code inspection after having noticed action-less IRQs in 'i' >> debug key output pointing at all parked/offline CPUs. >> >> --- a/xen/arch/x86/irq.c >> +++ b/xen/arch/x86/irq.c >> @@ -1683,9 +1683,27 @@ int pirq_guest_bind(struct vcpu *v, stru >> >> desc->status |= IRQ_GUEST; >> >> - /* Attempt to bind the interrupt target to the correct CPU. */ >> - if ( !opt_noirqbalance && (desc->handler->set_affinity != NULL) ) >> - desc->handler->set_affinity(desc, cpumask_of(v->processor)); >> + /* >> + * Attempt to bind the interrupt target to the correct (or at least >> + * some online) CPU. >> + */ >> + if ( desc->handler->set_affinity ) >> + { >> + const cpumask_t *affinity = NULL; >> + >> + if ( !opt_noirqbalance ) >> + affinity = cpumask_of(v->processor); >> + else if ( !cpumask_intersects(desc->affinity, &cpu_online_map) ) >> + { >> + cpumask_setall(desc->affinity); >> + affinity = &cpumask_all; >> + } >> + else if ( !cpumask_intersects(desc->arch.cpu_mask, >> + &cpu_online_map) ) > > I'm not sure I see the purpose of the desc->arch.cpu_mask check, > wouldn't it be better to just use else and set the affinity to > desc->affinity? We should avoid clobbering desc->affinity whenever possible: It reflects (see the respective patch in this series) what was requested by whatever "outside" party. > Or it's just an optimization to avoid doing the set_affinity call if > the interrupt it already bound to an online CPU? This is a second aspect here indeed - why play with the IRQ if it has a valid destination? Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On Mon, May 20, 2019 at 09:17:19AM -0600, Jan Beulich wrote: > >>> On 20.05.19 at 13:40, <roger.pau@citrix.com> wrote: > > On Fri, May 17, 2019 at 04:48:21AM -0600, Jan Beulich wrote: > >> fixup_irqs() skips interrupts without action. Hence such interrupts can > >> retain affinity to just offline CPUs. With "noirqbalance" in effect, > >> pirq_guest_bind() so far would have left them alone, resulting in a non- > >> working interrupt. > >> > >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > >> --- > >> v3: New. > >> --- > >> I've not observed this problem in practice - the change is just the > >> result of code inspection after having noticed action-less IRQs in 'i' > >> debug key output pointing at all parked/offline CPUs. > >> > >> --- a/xen/arch/x86/irq.c > >> +++ b/xen/arch/x86/irq.c > >> @@ -1683,9 +1683,27 @@ int pirq_guest_bind(struct vcpu *v, stru > >> > >> desc->status |= IRQ_GUEST; > >> > >> - /* Attempt to bind the interrupt target to the correct CPU. */ > >> - if ( !opt_noirqbalance && (desc->handler->set_affinity != NULL) ) > >> - desc->handler->set_affinity(desc, cpumask_of(v->processor)); > >> + /* > >> + * Attempt to bind the interrupt target to the correct (or at least > >> + * some online) CPU. > >> + */ > >> + if ( desc->handler->set_affinity ) > >> + { > >> + const cpumask_t *affinity = NULL; > >> + > >> + if ( !opt_noirqbalance ) > >> + affinity = cpumask_of(v->processor); > >> + else if ( !cpumask_intersects(desc->affinity, &cpu_online_map) ) > >> + { > >> + cpumask_setall(desc->affinity); > >> + affinity = &cpumask_all; > >> + } > >> + else if ( !cpumask_intersects(desc->arch.cpu_mask, > >> + &cpu_online_map) ) > > > > I'm not sure I see the purpose of the desc->arch.cpu_mask check, > > wouldn't it be better to just use else and set the affinity to > > desc->affinity? > > We should avoid clobbering desc->affinity whenever possible: It > reflects (see the respective patch in this series) what was > requested by whatever "outside" party. > > > Or it's just an optimization to avoid doing the set_affinity call if > > the interrupt it already bound to an online CPU? > > This is a second aspect here indeed - why play with the IRQ if > it has a valid destination? Thanks for the clarification, that LGTM: Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Roger. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:48, Jan Beulich wrote: > fixup_irqs() skips interrupts without action. Hence such interrupts can > retain affinity to just offline CPUs. With "noirqbalance" in effect, > pirq_guest_bind() so far would have left them alone, resulting in a non- > working interrupt. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
If any particular value was to be checked against, it would need to be IRQ_VECTOR_UNASSIGNED. Reported-by: Roger Pau Monné <roger.pau@citrix.com> Be more strict though and use valid_irq_vector() instead. Take the opportunity and also convert local variables to unsigned int. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- v2: New. --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -276,14 +276,13 @@ void destroy_irq(unsigned int irq) static void _clear_irq_vector(struct irq_desc *desc) { - unsigned int cpu; - int vector, old_vector, irq = desc->irq; + unsigned int cpu, old_vector, irq = desc->irq; + unsigned int vector = desc->arch.vector; cpumask_t tmp_mask; - BUG_ON(!desc->arch.vector); + BUG_ON(!valid_irq_vector(vector)); /* Always clear desc->arch.vector */ - vector = desc->arch.vector; cpumask_and(&tmp_mask, desc->arch.cpu_mask, &cpu_online_map); for_each_cpu(cpu, &tmp_mask) { _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:49, Jan Beulich wrote: > If any particular value was to be checked against, it would need to be > IRQ_VECTOR_UNASSIGNED. > > Reported-by: Roger Pau Monné <roger.pau@citrix.com> > > Be more strict though and use valid_irq_vector() instead. > > Take the opportunity and also convert local variables to unsigned int. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Since the "Cannot set affinity ..." warning is a one time one, avoid triggering it already at boot time when parking secondary threads and the serial console uses a (still unconnected at that time) PCI IRQ. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -2452,8 +2452,20 @@ void fixup_irqs(const cpumask_t *mask, b vector = irq_to_vector(irq); if ( vector >= FIRST_HIPRIORITY_VECTOR && vector <= LAST_HIPRIORITY_VECTOR ) + { cpumask_and(desc->arch.cpu_mask, desc->arch.cpu_mask, mask); + /* + * This can in particular happen when parking secondary threads + * during boot and when the serial console wants to use a PCI IRQ. + */ + if ( desc->handler == &no_irq_type ) + { + spin_unlock(&desc->lock); + continue; + } + } + if ( desc->arch.move_cleanup_count ) { /* The cleanup IPI may have got sent while we were still online. */ _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:49, Jan Beulich wrote: > Since the "Cannot set affinity ..." warning is a one time one, avoid > triggering it already at boot time when parking secondary threads and > the serial console uses a (still unconnected at that time) PCI IRQ. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
The subsequent cpumask_intersects() covers the "empty" case quite fine. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -650,9 +650,6 @@ void move_masked_irq(struct irq_desc *de desc->status &= ~IRQ_MOVE_PENDING; - if (unlikely(cpumask_empty(pending_mask))) - return; - if (!desc->handler->set_affinity) return; _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:50, Jan Beulich wrote: > The subsequent cpumask_intersects() covers the "empty" case quite fine. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Its only caller already has the IRQ descriptor in its hands, so there's no need for the function to re-obtain it. As a result the leading p of its name is no longer appropriate and hence gets dropped. Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> --- v2: New. --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -1556,17 +1556,8 @@ int pirq_guest_unmask(struct domain *d) return 0; } -static int pirq_acktype(struct domain *d, int pirq) +static int irq_acktype(const struct irq_desc *desc) { - struct irq_desc *desc; - int irq; - - irq = domain_pirq_to_irq(d, pirq); - if ( irq <= 0 ) - return ACKTYPE_NONE; - - desc = irq_to_desc(irq); - if ( desc->handler == &no_irq_type ) return ACKTYPE_NONE; @@ -1597,7 +1588,8 @@ static int pirq_acktype(struct domain *d if ( !strcmp(desc->handler->typename, "XT-PIC") ) return ACKTYPE_UNMASK; - printk("Unknown PIC type '%s' for IRQ %d\n", desc->handler->typename, irq); + printk("Unknown PIC type '%s' for IRQ%d\n", + desc->handler->typename, desc->irq); BUG(); return 0; @@ -1674,7 +1666,7 @@ int pirq_guest_bind(struct vcpu *v, stru action->nr_guests = 0; action->in_flight = 0; action->shareable = will_share; - action->ack_type = pirq_acktype(v->domain, pirq->pirq); + action->ack_type = irq_acktype(desc); init_timer(&action->eoi_timer, irq_guest_eoi_timer_fn, desc, 0); desc->status |= IRQ_GUEST; _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:51, Jan Beulich wrote: > Its only caller already has the IRQ descriptor in its hands, so there's > no need for the function to re-obtain it. As a result the leading p of > its name is no longer appropriate and hence gets dropped. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
The setup for calling trace_var() (which itself checks tb_init_done) is non-negligible, and hence a separate outer-most check is warranted. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- v3: New. --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -121,8 +121,8 @@ static void release_old_vec(struct irq_d } } -static void trace_irq_mask(uint32_t event, int irq, int vector, - const cpumask_t *mask) +static void _trace_irq_mask(uint32_t event, int irq, int vector, + const cpumask_t *mask) { struct { unsigned int irq:16, vec:16; @@ -137,6 +137,13 @@ static void trace_irq_mask(uint32_t even trace_var(event, 1, sizeof(d), &d); } +static inline void trace_irq_mask(uint32_t event, int irq, int vector, + const cpumask_t *mask) +{ + if ( unlikely(tb_init_done) ) + _trace_irq_mask(event, irq, vector, mask); +} + static int __init _bind_irq_vector(struct irq_desc *desc, int vector, const cpumask_t *cpu_mask) { _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:51, Jan Beulich wrote: > --- a/xen/arch/x86/irq.c > +++ b/xen/arch/x86/irq.c > @@ -137,6 +137,13 @@ static void trace_irq_mask(uint32_t even > trace_var(event, 1, sizeof(d), &d); > } > > +static inline void trace_irq_mask(uint32_t event, int irq, int vector, No inline. Otherwise, Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> > + const cpumask_t *mask) > +{ > + if ( unlikely(tb_init_done) ) > + _trace_irq_mask(event, irq, vector, mask); > +} > + > static int __init _bind_irq_vector(struct irq_desc *desc, int vector, > const cpumask_t *cpu_mask) > { > > > > _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 03.07.2019 20:41, Andrew Cooper wrote: > On 17/05/2019 11:51, Jan Beulich wrote: >> --- a/xen/arch/x86/irq.c >> +++ b/xen/arch/x86/irq.c >> @@ -137,6 +137,13 @@ static void trace_irq_mask(uint32_t even >> trace_var(event, 1, sizeof(d), &d); >> } >> >> +static inline void trace_irq_mask(uint32_t event, int irq, int vector, > > No inline. Well, I think in cases like this one we really want it, but anyway, I'll drop it just to make progress here. > Otherwise, Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> Thanks, Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On Fri, May 17, 2019 at 04:51:50AM -0600, Jan Beulich wrote: > The setup for calling trace_var() (which itself checks tb_init_done) is > non-negligible, and hence a separate outer-most check is warranted. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> I think a macro or helper would be helpful: ie: trace_enabled or some such. Checking tb_init_done is as obvious as it could be IMO. Roger. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Use valid_irq_vector() rather than "> 0". Also replace an open-coded use of IRQ_VECTOR_UNASSIGNED. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- v3: New. --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -342,7 +342,7 @@ void clear_irq_vector(int irq) int irq_to_vector(int irq) { - int vector = -1; + int vector = IRQ_VECTOR_UNASSIGNED; BUG_ON(irq >= nr_irqs || irq < 0); @@ -452,15 +452,18 @@ static vmask_t *irq_get_used_vector_mask int vector; vector = irq_to_vector(irq); - if ( vector > 0 ) + if ( valid_irq_vector(vector) ) { - printk(XENLOG_INFO "IRQ %d already assigned vector %d\n", + printk(XENLOG_INFO "IRQ%d already assigned vector %02x\n", irq, vector); ASSERT(!test_bit(vector, ret)); set_bit(vector, ret); } + else if ( vector != IRQ_VECTOR_UNASSIGNED ) + printk(XENLOG_WARNING "IRQ%d mapped to bogus vector %02x\n", + irq, vector); } } else if ( IO_APIC_IRQ(irq) && @@ -491,7 +494,7 @@ static int _assign_irq_vector(struct irq vmask_t *irq_used_vectors = NULL; old_vector = irq_to_vector(irq); - if ( old_vector > 0 ) + if ( valid_irq_vector(old_vector) ) { cpumask_t tmp_mask; @@ -555,7 +558,7 @@ next: current_vector = vector; current_offset = offset; - if ( old_vector > 0 ) + if ( valid_irq_vector(old_vector) ) { cpumask_and(desc->arch.old_cpu_mask, desc->arch.cpu_mask, &cpu_online_map); _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:52, Jan Beulich wrote: > Use valid_irq_vector() rather than "> 0". > > Also replace an open-coded use of IRQ_VECTOR_UNASSIGNED. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On Fri, May 17, 2019 at 04:52:32AM -0600, Jan Beulich wrote: > Use valid_irq_vector() rather than "> 0". > > Also replace an open-coded use of IRQ_VECTOR_UNASSIGNED. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> The question I have below is not directly related to the usage of valid_irq_vector, but rather with the existing code. > --- > v3: New. > > --- a/xen/arch/x86/irq.c > +++ b/xen/arch/x86/irq.c > @@ -342,7 +342,7 @@ void clear_irq_vector(int irq) > > int irq_to_vector(int irq) > { > - int vector = -1; > + int vector = IRQ_VECTOR_UNASSIGNED; > > BUG_ON(irq >= nr_irqs || irq < 0); > > @@ -452,15 +452,18 @@ static vmask_t *irq_get_used_vector_mask > int vector; > > vector = irq_to_vector(irq); > - if ( vector > 0 ) > + if ( valid_irq_vector(vector) ) > { > - printk(XENLOG_INFO "IRQ %d already assigned vector %d\n", > + printk(XENLOG_INFO "IRQ%d already assigned vector %02x\n", > irq, vector); > > ASSERT(!test_bit(vector, ret)); > > set_bit(vector, ret); > } > + else if ( vector != IRQ_VECTOR_UNASSIGNED ) > + printk(XENLOG_WARNING "IRQ%d mapped to bogus vector %02x\n", > + irq, vector); Maybe add an assert_unreachable here? It seems really bogus to call irq_get_used_vector_mask with an unassigned vector. But I'm not sure I fully understand this piece of code, neither why a vector without a IRQ assigned can have a vector assigned. Is this covering up for the lack of cleanup elsewhere? Thanks, Roger. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
>>> On 20.05.19 at 16:04, <roger.pau@citrix.com> wrote: > On Fri, May 17, 2019 at 04:52:32AM -0600, Jan Beulich wrote: >> Use valid_irq_vector() rather than "> 0". >> >> Also replace an open-coded use of IRQ_VECTOR_UNASSIGNED. >> >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Thanks. > The question I have below is not directly related to the usage of > valid_irq_vector, but rather with the existing code. > >> @@ -452,15 +452,18 @@ static vmask_t *irq_get_used_vector_mask >> int vector; >> >> vector = irq_to_vector(irq); >> - if ( vector > 0 ) >> + if ( valid_irq_vector(vector) ) >> { >> - printk(XENLOG_INFO "IRQ %d already assigned vector %d\n", >> + printk(XENLOG_INFO "IRQ%d already assigned vector %02x\n", >> irq, vector); >> >> ASSERT(!test_bit(vector, ret)); >> >> set_bit(vector, ret); >> } >> + else if ( vector != IRQ_VECTOR_UNASSIGNED ) >> + printk(XENLOG_WARNING "IRQ%d mapped to bogus vector %02x\n", >> + irq, vector); > > Maybe add an assert_unreachable here? It seems really bogus to call > irq_get_used_vector_mask with an unassigned vector. How that? This would e.g. get called the very first time a vector is to be assigned. But I'm afraid I'm a little confused anyway by the wording you use - after all this is the code path dealing with an IRQ _not_ being marked as having no vector assigned, but also not having a valid vector. > But I'm not sure I fully understand this piece of code, neither why a > vector without a IRQ assigned can have a vector assigned. Is this > covering up for the lack of cleanup elsewhere? I don't think so, no. However, users of irq_to_vector() need to be careful: The function can legitimately return 0 (besides IRQ_VECTOR_UNASSIGNED) as an error indication. I've tried to do away with this, but quickly realized I'd better not do so. I've not seen the printk() trigger, but I'd rather see the printk() log a message telling us that we also need to exclude vector 0 than a wrong assertion to fire. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On Mon, May 20, 2019 at 09:26:37AM -0600, Jan Beulich wrote: > >>> On 20.05.19 at 16:04, <roger.pau@citrix.com> wrote: > > On Fri, May 17, 2019 at 04:52:32AM -0600, Jan Beulich wrote: > >> Use valid_irq_vector() rather than "> 0". > >> > >> Also replace an open-coded use of IRQ_VECTOR_UNASSIGNED. > >> > >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > > > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> > > Thanks. > > > The question I have below is not directly related to the usage of > > valid_irq_vector, but rather with the existing code. > > > >> @@ -452,15 +452,18 @@ static vmask_t *irq_get_used_vector_mask > >> int vector; > >> > >> vector = irq_to_vector(irq); > >> - if ( vector > 0 ) > >> + if ( valid_irq_vector(vector) ) > >> { > >> - printk(XENLOG_INFO "IRQ %d already assigned vector %d\n", > >> + printk(XENLOG_INFO "IRQ%d already assigned vector %02x\n", > >> irq, vector); > >> > >> ASSERT(!test_bit(vector, ret)); > >> > >> set_bit(vector, ret); > >> } > >> + else if ( vector != IRQ_VECTOR_UNASSIGNED ) > >> + printk(XENLOG_WARNING "IRQ%d mapped to bogus vector %02x\n", > >> + irq, vector); > > > > Maybe add an assert_unreachable here? It seems really bogus to call > > irq_get_used_vector_mask with an unassigned vector. > > How that? This would e.g. get called the very first time a vector > is to be assigned. But I'm afraid I'm a little confused anyway by > the wording you use - after all this is the code path dealing with > an IRQ _not_ being marked as having no vector assigned, but > also not having a valid vector. Thanks for the clarification, by the name of the function I assumed it must be called with an irq that has a vector assigned, if that's not the case then I think it's fine. Roger. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
>>> On 22.05.19 at 18:42, <roger.pau@citrix.com> wrote: > On Mon, May 20, 2019 at 09:26:37AM -0600, Jan Beulich wrote: >> >>> On 20.05.19 at 16:04, <roger.pau@citrix.com> wrote: >> > On Fri, May 17, 2019 at 04:52:32AM -0600, Jan Beulich wrote: >> >> @@ -452,15 +452,18 @@ static vmask_t *irq_get_used_vector_mask >> >> int vector; >> >> >> >> vector = irq_to_vector(irq); >> >> - if ( vector > 0 ) >> >> + if ( valid_irq_vector(vector) ) >> >> { >> >> - printk(XENLOG_INFO "IRQ %d already assigned vector %d\n", >> >> + printk(XENLOG_INFO "IRQ%d already assigned vector %02x\n", >> >> irq, vector); >> >> >> >> ASSERT(!test_bit(vector, ret)); >> >> >> >> set_bit(vector, ret); >> >> } >> >> + else if ( vector != IRQ_VECTOR_UNASSIGNED ) >> >> + printk(XENLOG_WARNING "IRQ%d mapped to bogus vector %02x\n", >> >> + irq, vector); >> > >> > Maybe add an assert_unreachable here? It seems really bogus to call >> > irq_get_used_vector_mask with an unassigned vector. >> >> How that? This would e.g. get called the very first time a vector >> is to be assigned. But I'm afraid I'm a little confused anyway by >> the wording you use - after all this is the code path dealing with >> an IRQ _not_ being marked as having no vector assigned, but >> also not having a valid vector. > > Thanks for the clarification, by the name of the function I assumed it > must be called with an irq that has a vector assigned, if that's not > the case then I think it's fine. > > Roger. Well, the names means "get the object where used vectors are to be tracked for this IRQ", which has no implication on whether a vector was already assigned. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Use scratch_cpumask where possible, to avoid creating these possibly large stack objects. We can't use it in _assign_irq_vector() and set_desc_affinity(), as these get called in IRQ context. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- v3: New. --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -285,14 +285,15 @@ static void _clear_irq_vector(struct irq { unsigned int cpu, old_vector, irq = desc->irq; unsigned int vector = desc->arch.vector; - cpumask_t tmp_mask; + cpumask_t *tmp_mask = this_cpu(scratch_cpumask); BUG_ON(!valid_irq_vector(vector)); /* Always clear desc->arch.vector */ - cpumask_and(&tmp_mask, desc->arch.cpu_mask, &cpu_online_map); + cpumask_and(tmp_mask, desc->arch.cpu_mask, &cpu_online_map); - for_each_cpu(cpu, &tmp_mask) { + for_each_cpu(cpu, tmp_mask) + { ASSERT( per_cpu(vector_irq, cpu)[vector] == irq ); per_cpu(vector_irq, cpu)[vector] = ~irq; } @@ -308,16 +309,17 @@ static void _clear_irq_vector(struct irq desc->arch.used = IRQ_UNUSED; - trace_irq_mask(TRC_HW_IRQ_CLEAR_VECTOR, irq, vector, &tmp_mask); + trace_irq_mask(TRC_HW_IRQ_CLEAR_VECTOR, irq, vector, tmp_mask); if ( likely(!desc->arch.move_in_progress) ) return; /* If we were in motion, also clear desc->arch.old_vector */ old_vector = desc->arch.old_vector; - cpumask_and(&tmp_mask, desc->arch.old_cpu_mask, &cpu_online_map); + cpumask_and(tmp_mask, desc->arch.old_cpu_mask, &cpu_online_map); - for_each_cpu(cpu, &tmp_mask) { + for_each_cpu(cpu, tmp_mask) + { ASSERT( per_cpu(vector_irq, cpu)[old_vector] == irq ); TRACE_3D(TRC_HW_IRQ_MOVE_FINISH, irq, old_vector, cpu); per_cpu(vector_irq, cpu)[old_vector] = ~irq; @@ -1159,7 +1161,6 @@ static void irq_guest_eoi_timer_fn(void struct irq_desc *desc = data; unsigned int i, irq = desc - irq_desc; irq_guest_action_t *action; - cpumask_t cpu_eoi_map; spin_lock_irq(&desc->lock); @@ -1189,14 +1190,18 @@ static void irq_guest_eoi_timer_fn(void switch ( action->ack_type ) { + cpumask_t *cpu_eoi_map; + case ACKTYPE_UNMASK: if ( desc->handler->end ) desc->handler->end(desc, 0); break; + case ACKTYPE_EOI: - cpumask_copy(&cpu_eoi_map, action->cpu_eoi_map); + cpu_eoi_map = this_cpu(scratch_cpumask); + cpumask_copy(cpu_eoi_map, action->cpu_eoi_map); spin_unlock_irq(&desc->lock); - on_selected_cpus(&cpu_eoi_map, set_eoi_ready, desc, 0); + on_selected_cpus(cpu_eoi_map, set_eoi_ready, desc, 0); return; } @@ -2437,7 +2442,7 @@ void fixup_irqs(const cpumask_t *mask, b { bool break_affinity = false, set_affinity = true; unsigned int vector; - cpumask_t affinity; + cpumask_t *affinity = this_cpu(scratch_cpumask); if ( irq == 2 ) continue; @@ -2468,9 +2473,9 @@ void fixup_irqs(const cpumask_t *mask, b if ( desc->arch.move_cleanup_count ) { /* The cleanup IPI may have got sent while we were still online. */ - cpumask_andnot(&affinity, desc->arch.old_cpu_mask, + cpumask_andnot(affinity, desc->arch.old_cpu_mask, &cpu_online_map); - desc->arch.move_cleanup_count -= cpumask_weight(&affinity); + desc->arch.move_cleanup_count -= cpumask_weight(affinity); if ( !desc->arch.move_cleanup_count ) release_old_vec(desc); } @@ -2497,10 +2502,10 @@ void fixup_irqs(const cpumask_t *mask, b { unsigned int cpu; - cpumask_and(&affinity, desc->arch.old_cpu_mask, &cpu_online_map); + cpumask_and(affinity, desc->arch.old_cpu_mask, &cpu_online_map); spin_lock(&vector_lock); - for_each_cpu(cpu, &affinity) + for_each_cpu(cpu, affinity) per_cpu(vector_irq, cpu)[desc->arch.old_vector] = ~irq; spin_unlock(&vector_lock); @@ -2511,23 +2516,23 @@ void fixup_irqs(const cpumask_t *mask, b if ( !cpumask_intersects(mask, desc->affinity) ) { break_affinity = true; - cpumask_setall(&affinity); + cpumask_setall(affinity); } else - cpumask_copy(&affinity, desc->affinity); + cpumask_copy(affinity, desc->affinity); if ( desc->handler->disable ) desc->handler->disable(desc); if ( desc->handler->set_affinity ) - desc->handler->set_affinity(desc, &affinity); + desc->handler->set_affinity(desc, affinity); else if ( !(warned++) ) set_affinity = false; if ( desc->handler->enable ) desc->handler->enable(desc); - cpumask_copy(&affinity, desc->affinity); + cpumask_copy(affinity, desc->affinity); spin_unlock(&desc->lock); @@ -2538,7 +2543,7 @@ void fixup_irqs(const cpumask_t *mask, b printk("Cannot set affinity for IRQ%u\n", irq); else if ( break_affinity ) printk("Broke affinity for IRQ%u, new: %*pb\n", - irq, nr_cpu_ids, &affinity); + irq, nr_cpu_ids, affinity); } /* That doesn't seem sufficient. Give it 1ms. */ _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On Fri, May 17, 2019 at 04:52:54AM -0600, Jan Beulich wrote: > Use scratch_cpumask where possible, to avoid creating these possibly > large stack objects. We can't use it in _assign_irq_vector() and > set_desc_affinity(), as these get called in IRQ context. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com Thanks, Roger. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 20/05/2019 15:22, Roger Pau Monné wrote: > On Fri, May 17, 2019 at 04:52:54AM -0600, Jan Beulich wrote: >> Use scratch_cpumask where possible, to avoid creating these possibly >> large stack objects. We can't use it in _assign_irq_vector() and >> set_desc_affinity(), as these get called in IRQ context. >> >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com Missing a trailing > Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 03.07.2019 20:44, Andrew Cooper wrote: > On 20/05/2019 15:22, Roger Pau Monné wrote: >> On Fri, May 17, 2019 at 04:52:54AM -0600, Jan Beulich wrote: >>> Use scratch_cpumask where possible, to avoid creating these possibly >>> large stack objects. We can't use it in _assign_irq_vector() and >>> set_desc_affinity(), as these get called in IRQ context. >>> >>> Signed-off-by: Jan Beulich <jbeulich@suse.com> >> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com > > Missing a trailing > I had added that one already. > Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> Thanks again. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
This is largely to drop a forward declaration. There's one functional change - clear_irq_vector() gets marked __init, as its only caller is check_timer(). Beyond this only a few stray blanks get removed. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- v3: New. --- a/xen/arch/x86/irq.c +++ b/xen/arch/x86/irq.c @@ -27,7 +27,6 @@ #include <public/physdev.h> static int parse_irq_vector_map_param(const char *s); -static void _clear_irq_vector(struct irq_desc *desc); /* opt_noirqbalance: If true, software IRQ balancing/affinity is disabled. */ bool __read_mostly opt_noirqbalance; @@ -191,6 +190,67 @@ int __init bind_irq_vector(int irq, int return ret; } +static void _clear_irq_vector(struct irq_desc *desc) +{ + unsigned int cpu, old_vector, irq = desc->irq; + unsigned int vector = desc->arch.vector; + cpumask_t *tmp_mask = this_cpu(scratch_cpumask); + + BUG_ON(!valid_irq_vector(vector)); + + /* Always clear desc->arch.vector */ + cpumask_and(tmp_mask, desc->arch.cpu_mask, &cpu_online_map); + + for_each_cpu(cpu, tmp_mask) + { + ASSERT(per_cpu(vector_irq, cpu)[vector] == irq); + per_cpu(vector_irq, cpu)[vector] = ~irq; + } + + desc->arch.vector = IRQ_VECTOR_UNASSIGNED; + cpumask_clear(desc->arch.cpu_mask); + + if ( desc->arch.used_vectors ) + { + ASSERT(test_bit(vector, desc->arch.used_vectors)); + clear_bit(vector, desc->arch.used_vectors); + } + + desc->arch.used = IRQ_UNUSED; + + trace_irq_mask(TRC_HW_IRQ_CLEAR_VECTOR, irq, vector, tmp_mask); + + if ( likely(!desc->arch.move_in_progress) ) + return; + + /* If we were in motion, also clear desc->arch.old_vector */ + old_vector = desc->arch.old_vector; + cpumask_and(tmp_mask, desc->arch.old_cpu_mask, &cpu_online_map); + + for_each_cpu(cpu, tmp_mask) + { + ASSERT(per_cpu(vector_irq, cpu)[old_vector] == irq); + TRACE_3D(TRC_HW_IRQ_MOVE_FINISH, irq, old_vector, cpu); + per_cpu(vector_irq, cpu)[old_vector] = ~irq; + } + + release_old_vec(desc); + + desc->arch.move_in_progress = 0; +} + +void __init clear_irq_vector(int irq) +{ + struct irq_desc *desc = irq_to_desc(irq); + unsigned long flags; + + spin_lock_irqsave(&desc->lock, flags); + spin_lock(&vector_lock); + _clear_irq_vector(desc); + spin_unlock(&vector_lock); + spin_unlock_irqrestore(&desc->lock, flags); +} + /* * Dynamic irq allocate and deallocation for MSI */ @@ -281,67 +341,6 @@ void destroy_irq(unsigned int irq) xfree(action); } -static void _clear_irq_vector(struct irq_desc *desc) -{ - unsigned int cpu, old_vector, irq = desc->irq; - unsigned int vector = desc->arch.vector; - cpumask_t *tmp_mask = this_cpu(scratch_cpumask); - - BUG_ON(!valid_irq_vector(vector)); - - /* Always clear desc->arch.vector */ - cpumask_and(tmp_mask, desc->arch.cpu_mask, &cpu_online_map); - - for_each_cpu(cpu, tmp_mask) - { - ASSERT( per_cpu(vector_irq, cpu)[vector] == irq ); - per_cpu(vector_irq, cpu)[vector] = ~irq; - } - - desc->arch.vector = IRQ_VECTOR_UNASSIGNED; - cpumask_clear(desc->arch.cpu_mask); - - if ( desc->arch.used_vectors ) - { - ASSERT(test_bit(vector, desc->arch.used_vectors)); - clear_bit(vector, desc->arch.used_vectors); - } - - desc->arch.used = IRQ_UNUSED; - - trace_irq_mask(TRC_HW_IRQ_CLEAR_VECTOR, irq, vector, tmp_mask); - - if ( likely(!desc->arch.move_in_progress) ) - return; - - /* If we were in motion, also clear desc->arch.old_vector */ - old_vector = desc->arch.old_vector; - cpumask_and(tmp_mask, desc->arch.old_cpu_mask, &cpu_online_map); - - for_each_cpu(cpu, tmp_mask) - { - ASSERT( per_cpu(vector_irq, cpu)[old_vector] == irq ); - TRACE_3D(TRC_HW_IRQ_MOVE_FINISH, irq, old_vector, cpu); - per_cpu(vector_irq, cpu)[old_vector] = ~irq; - } - - release_old_vec(desc); - - desc->arch.move_in_progress = 0; -} - -void clear_irq_vector(int irq) -{ - struct irq_desc *desc = irq_to_desc(irq); - unsigned long flags; - - spin_lock_irqsave(&desc->lock, flags); - spin_lock(&vector_lock); - _clear_irq_vector(desc); - spin_unlock(&vector_lock); - spin_unlock_irqrestore(&desc->lock, flags); -} - int irq_to_vector(int irq) { int vector = IRQ_VECTOR_UNASSIGNED; _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 17/05/2019 11:53, Jan Beulich wrote: > This is largely to drop a forward declaration. There's one functional > change - clear_irq_vector() gets marked __init, as its only caller is > check_timer(). Beyond this only a few stray blanks get removed. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
© 2016 - 2024 Red Hat, Inc.