Skip to content

a13: replay immediate IPIs dropped while the destination was busy (port from MakrSas/Inferno) #58

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
dylanneve1 merged 1 commit into from
Oct 22, 2026
Merged

a13: replay immediate IPIs dropped while the destination was busy (port from MakrSas/Inferno) #58

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
43 changes: 36 additions & 7 deletions hw/arm/a13.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -173,19 +173,36 @@ static void apple_a13_cluster_cpreg_write(CPUARMState* env, const ARMCPRegInfo*
*(uint64_t*)((char*)(c) + (ri)->fieldoffset) = value;
}

/* Deliver IPI */
static void apple_a13_deliver_ipi(AppleA13State* cpu, uint64_t src_cpu, uint64_t flag)
/* Deliver IPI. Returns false if the destination already had one in flight --
* the caller (an immediate send) is on its own to remember the source and
* retry, since nothing here revisits a drop by itself. */
static bool apple_a13_deliver_ipi(AppleA13State* cpu, uint64_t src_cpu, uint64_t flag)
{
trace_apple_a13_ipi_deliver(cpu->cpu_id, src_cpu, flag, cpu->ipi_sr != 0);
if (cpu->ipi_sr) { return; }
if (cpu->ipi_sr) { return false; }

cpu->ipi_sr = 1LL | (src_cpu << IPI_SR_SRC_CPU_SHIFT) | flag;
qemu_irq_raise(cpu->fast_ipi);
return true;
}

static void apple_a13_cluster_deliver_ipi(AppleA13Cluster* c, uint64_t cpu_id, uint64_t src_cpu, uint64_t flag)
{ apple_a13_deliver_ipi(c->cpus[cpu_id], src_cpu, flag); }

/* An immediate IPI whose destination was already busy has nowhere else to
* wait -- deferred/no-wake sends are retried every tick by
* apple_a13_cluster_tick, but an immediate send that apple_a13_deliver_ipi
* drops is otherwise gone for good. Recorded here, indexed by destination,
* and replayed from apple_a13_ipi_write_sr the moment that destination
* acknowledges its current IPI and has room for another. */
static void apple_a13_send_immediate_ipi(AppleA13Cluster* c, AppleA13State* dst_acpu, uint64_t dst_cpu_id,
uint64_t src_cpu)
{
if (!apple_a13_deliver_ipi(dst_acpu, src_cpu, IPI_RR_TYPE_IMMEDIATE)) {
c->pendingImmediateIPI[dst_cpu_id] |= BIT32(src_cpu);
}
}

static int add_cpu_to_cluster(Object* obj, void* opaque)
{
AppleA13Cluster* cluster = opaque;
Expand Down Expand Up @@ -289,15 +306,15 @@ static void apple_a13_ipi_rr_local(CPUARMState* env, const ARMCPRegInfo* ri, uin
case IPI_RR_TYPE_NOWAKE:
if (apple_a13_is_asleep(dst_acpu)) { c->noWakeIPI[acpu->cpu_id] |= BIT32(dst_cpu_id); }
else {
apple_a13_deliver_ipi(dst_acpu, acpu->cpu_id, IPI_RR_TYPE_IMMEDIATE);
apple_a13_send_immediate_ipi(c, dst_acpu, dst_cpu_id, acpu->cpu_id);
}
break;
case IPI_RR_TYPE_DEFERRED: c->deferredIPI[acpu->cpu_id] |= BIT32(dst_cpu_id); break;
case IPI_RR_TYPE_RETRACT:
c->deferredIPI[acpu->cpu_id] &= ~BIT32(dst_cpu_id);
c->noWakeIPI[acpu->cpu_id] &= ~BIT32(dst_cpu_id);
break;
case IPI_RR_TYPE_IMMEDIATE: apple_a13_deliver_ipi(dst_acpu, acpu->cpu_id, IPI_RR_TYPE_IMMEDIATE); break;
case IPI_RR_TYPE_IMMEDIATE: apple_a13_send_immediate_ipi(c, dst_acpu, dst_cpu_id, acpu->cpu_id); break;
default : assert_not_reached();
}
}
Expand Down Expand Up @@ -335,15 +352,15 @@ static void apple_a13_ipi_rr_global(CPUARMState* env, const ARMCPRegInfo* ri, ui
case IPI_RR_TYPE_NOWAKE:
if (apple_a13_is_asleep(dst_acpu)) { cluster->noWakeIPI[acpu->cpu_id] |= BIT32(dst_cpu_id); }
else {
apple_a13_deliver_ipi(dst_acpu, acpu->cpu_id, IPI_RR_TYPE_IMMEDIATE);
apple_a13_send_immediate_ipi(cluster, dst_acpu, dst_cpu_id, acpu->cpu_id);
}
break;
case IPI_RR_TYPE_DEFERRED: cluster->deferredIPI[acpu->cpu_id] |= BIT32(dst_cpu_id); break;
case IPI_RR_TYPE_RETRACT:
cluster->deferredIPI[acpu->cpu_id] &= ~BIT32(dst_cpu_id);
cluster->noWakeIPI[acpu->cpu_id] &= ~BIT32(dst_cpu_id);
break;
case IPI_RR_TYPE_IMMEDIATE: apple_a13_deliver_ipi(dst_acpu, acpu->cpu_id, IPI_RR_TYPE_IMMEDIATE); break;
case IPI_RR_TYPE_IMMEDIATE: apple_a13_send_immediate_ipi(cluster, dst_acpu, dst_cpu_id, acpu->cpu_id); break;
default : assert_not_reached();
}
}
Expand Down Expand Up @@ -372,6 +389,17 @@ static void apple_a13_ipi_write_sr(CPUARMState* env, const ARMCPRegInfo* ri, uin
case IPI_RR_TYPE_DEFERRED: c->deferredIPI[src_cpu] &= ~BIT32(acpu->cpu_id); break;
default : break;
}

/* This CPU just freed up its one IPI slot. Anything an immediate send
* found busy earlier and left waiting here gets its turn now -- without
* this, that sender's wakeup is gone for good, and MTTCG's wider window
* between "found busy" and "would have retried" makes hitting it far
* likelier than on real hardware. */
if (c->pendingImmediateIPI[acpu->cpu_id]) {
uint32_t pending_src = ctz32(c->pendingImmediateIPI[acpu->cpu_id]);
c->pendingImmediateIPI[acpu->cpu_id] &= ~BIT32(pending_src);
apple_a13_deliver_ipi(acpu, pending_src, IPI_RR_TYPE_IMMEDIATE);
}
}

/* Read deferred interrupt timeout (global) */
Expand Down Expand Up @@ -664,6 +692,7 @@ static void apple_a13_cluster_reset_enter(Object* obj, ResetType type)
AppleA13Cluster* cluster = APPLE_A13_CLUSTER(obj);
memset(cluster->deferredIPI, 0, sizeof(cluster->deferredIPI));
memset(cluster->noWakeIPI, 0, sizeof(cluster->noWakeIPI));
memset(cluster->pendingImmediateIPI, 0, sizeof(cluster->pendingImmediateIPI));
}

static void apple_a13_cluster_class_init(ObjectClass* klass, const void* data)
Expand Down
4 changes: 4 additions & 0 deletions include/hw/arm/a13.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,6 +130,10 @@ typedef struct AppleA13Cluster
AppleA13State* cpus[A13_MAX_CPU];
uint32_t deferredIPI[A13_MAX_CPU];
uint32_t noWakeIPI[A13_MAX_CPU];
/* Immediate IPIs dropped because the destination already had one
* in-flight, indexed by destination, bit per waiting source -- retried
* from apple_a13_ipi_write_sr once the destination acknowledges. */
uint32_t pendingImmediateIPI[A13_MAX_CPU];
uint64_t ipi_cr;
QTAILQ_ENTRY(AppleA13Cluster) next;
A13_CPREG_VAR_DEF(CTRR_A_LWR_EL1);
Expand Down
Loading