[PATCH v2 00/20] KVM: x86/xen: Fix Xen/GP/PREEMPT_RT issues with rwlock_t

[PATCH v2 00/20] KVM: x86/xen: Fix Xen/GP/PREEMPT_RT issues with rwlock_t

[Sean Christopherson]

This series fixes sleeping-in-hardirq bugs in KVM's Xen emulation on
PREEMPT_RT, cleans up the now-unnecessary IRQ disabling in GPC lock usage
throughout KVM, and then adds CLASS()-based APIs for utilizing GPCs mappings
to dedup code and (hopefully) make it easier to use GPCs in other places.
  
The core issue is that kvm_xen_set_evtchn_fast() and the Xen timer
callback are called from hardirq/atomic context, but on PREEMPT_RT the
GPC rwlock_t is a sleeping lock.

Assuming I can get an Ack on patch 1, I'm planning on grabbing at least these

  KVM: Move {g,p}fn <=> {g,h}pa conversion helpers to kvm_types.h
  KVM: x86/xen: Don't dirty track "vCPU info" page
  KVM: x86/xen: Explicitly tag "shared info" page as never being dirty tracked
  KVM: x86/xen: Extract delivery of event to vCPU into a separate helper
  KVM: x86/xen: Use guard() to grab kvm->srcu around gpc critical sections
  KVM: Remove unnecessary IRQ disabling from GPC lock in pfncache.c
  KVM: x86: Remove unnecessary irqsave from kvm_setup_guest_pvclock()
  KVM: x86/xen: Remove unnecessary irqsave from GPC lock usage in xen.c
  KVM: x86/xen: Use read_trylock() for GPC locks in hardirq/atomic paths
  locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

for 7.2.  If people like the CLASS() stuff, I'll also probably grab these:

  KVM: Add "extended" gpc CLASS() APIs for sometimes-atomic cases
  KVM: x86/xen: Convert event injection to gpc's CLASS() APIs
  KVM: x86/xen: Drop local "kick_vcpu" from __kvm_xen_set_evtchn_fast()
  KVM: x86/xen: Convert xen_get_guest_pvclock() to gpc's CLASS() APIs
  KVM: x86/xen: Convert kvm_xen_set_evtchn_fast() to gpc's CLASS() APIs
  KVM: x86/xen: Convert wait_pending_event() to gpc's CLASS() APIs
  KVM: x86/xen: Don't bother waiting on gpc->lock in SCHEDOP_poll
  KVM: x86/xen: Convert kvm_xen_shared_info_init() to gpc's CLASS() APIs
  KVM: Add CLASS() constructs to automagically handle lock+check of gpc

I do NOT plan on grabbing the record_steal_time change for 7.2 no matter
what, even though I do like the end result, as I still have concerns over the
lack of range-based invalidation for GPCs.  I 100% agree that such problems
are really only due to flawed VMMs and/or setups, but unfortunately history
has shown that there are a suprising number of deployments running what I
would consider flawed setups, e.g. run with NUMA autobalancing and KSM.

I realize I'm being somewhat paranoid, as KVM already uses a GPC for PV
clocks.  But for modern setups, KVM_REQ_CLOCK_UPDATE is a rare event, whereas
KVM will update steal time (when enabled) on every vCPU load.  So I want a
high level of confidence that KVM won't regress "imperfect" setups before
switching to a GPC for steal time (though again, I definitely like the end
result and want to do so).

[*] https://lore.kernel.org/all/20240821202814.711673-2-dwmw2@infradead.org

v2:
 - Add the CLASS() APIs.
 - Move the steal time change to the very end.
 - "Fix" a dirty logging inconsistency with the Xen vCPU info page.

v1: https://lore.kernel.org/all/20260508181717.3230988-1-dwmw2@infradead.org

Carsten Stollmaier (1):
  KVM: x86: Use gfn_to_pfn_cache for record_steal_time

David Woodhouse (5):
  locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()
  KVM: x86/xen: Use read_trylock() for GPC locks in hardirq/atomic paths
  KVM: x86/xen: Remove unnecessary irqsave from GPC lock usage in xen.c
  KVM: x86: Remove unnecessary irqsave from kvm_setup_guest_pvclock()
  KVM: Remove unnecessary IRQ disabling from GPC lock in pfncache.c

Sean Christopherson (14):
  KVM: x86/xen: Use guard() to grab kvm->srcu around gpc critical
    sections
  KVM: x86/xen: Extract delivery of event to vCPU into a separate helper
  KVM: x86/xen: Explicitly tag "shared info" page as never being dirty
    tracked
  KVM: x86/xen: Don't dirty track "vCPU info" page
  KVM: Move {g,p}fn <=> {g,h}pa conversion helpers to kvm_types.h
  KVM: Add CLASS() constructs to automagically handle lock+check of gpc
  KVM: x86/xen: Convert kvm_xen_shared_info_init() to gpc's CLASS() APIs
  KVM: x86/xen: Don't bother waiting on gpc->lock in SCHEDOP_poll
  KVM: x86/xen: Convert wait_pending_event() to gpc's CLASS() APIs
  KVM: x86/xen: Convert kvm_xen_set_evtchn_fast() to gpc's CLASS() APIs
  KVM: x86/xen: Convert xen_get_guest_pvclock() to gpc's CLASS() APIs
  KVM: x86/xen: Drop local "kick_vcpu" from __kvm_xen_set_evtchn_fast()
  KVM: x86/xen: Convert event injection to gpc's CLASS() APIs
  KVM: Add "extended" gpc CLASS() APIs for sometimes-atomic cases

 arch/x86/include/asm/kvm_host.h |   2 +-
 arch/x86/kvm/x86.c              | 140 +++++++---------
 arch/x86/kvm/xen.c              | 288 +++++++++++++-------------------
 include/linux/kvm_host.h        |  84 +++++++---
 include/linux/kvm_types.h       |  17 ++
 kernel/locking/rwbase_rt.c      |   5 +-
 virt/kvm/pfncache.c             |  68 ++++++--
 7 files changed, 304 insertions(+), 300 deletions(-)


base-commit: d1568b1332b6b3b36b222c2868fc102727c12a34
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Sean Christopherson]

From: David Woodhouse <dwmw@amazon.co.uk>

__rwbase_read_unlock() uses raw_spin_lock_irq()/raw_spin_unlock_irq()
which unconditionally disables and re-enables interrupts. When
read_unlock() is called from hardirq context (e.g. after a successful
read_trylock() in a timer callback), the raw_spin_unlock_irq()
incorrectly re-enables interrupts within the hardirq handler.

This causes lockdep warnings ('hardirqs_on_prepare' from hardirq
context) and can lead to IRQ state corruption.

Using read_trylock() in hardirq context on PREEMPT_RT is safe because
it does not record the lock owner. The read_unlock() acquires the
wait_lock which is hardirq safe. This change additionally allows
rwlock_t during early boot.

Switch to raw_spin_lock_irqsave()/raw_spin_unlock_irqrestore() to
preserve the caller's IRQ state.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 kernel/locking/rwbase_rt.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/kernel/locking/rwbase_rt.c b/kernel/locking/rwbase_rt.c
index 82e078c0665a..25744862d627 100644
--- a/kernel/locking/rwbase_rt.c
+++ b/kernel/locking/rwbase_rt.c
@@ -153,8 +153,9 @@ static void __sched __rwbase_read_unlock(struct rwbase_rt *rwb,
 	struct rt_mutex_base *rtm = &rwb->rtmutex;
 	struct task_struct *owner;
 	DEFINE_RT_WAKE_Q(wqh);
+	unsigned long flags;
 
-	raw_spin_lock_irq(&rtm->wait_lock);
+	raw_spin_lock_irqsave(&rtm->wait_lock, flags);
 	/*
 	 * Wake the writer, i.e. the rtmutex owner. It might release the
 	 * rtmutex concurrently in the fast path (due to a signal), but to
@@ -167,7 +168,7 @@ static void __sched __rwbase_read_unlock(struct rwbase_rt *rwb,
 
 	/* Pairs with the preempt_enable in rt_mutex_wake_up_q() */
 	preempt_disable();
-	raw_spin_unlock_irq(&rtm->wait_lock);
+	raw_spin_unlock_irqrestore(&rtm->wait_lock, flags);
 	rt_mutex_wake_up_q(&wqh);
 }
 
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 02/20] KVM: x86/xen: Use read_trylock() for GPC locks in hardirq/atomic paths

[Sean Christopherson]

From: David Woodhouse <dwmw@amazon.co.uk>

kvm_xen_set_evtchn_fast() is called from hardirq context (timer
callback, kvm_arch_set_irq_inatomic()). On PREEMPT_RT, rwlock_t is a
sleeping lock, so read_lock_irqsave() cannot be used in this context.

Switch to read_trylock() and return -EWOULDBLOCK on contention, which is
the designed fallback — there is always a slow path for the case where
the GPC is invalid and needs to be refreshed.

Reported-by: syzbot+208f7f3e5f59c11aeb90@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=208f7f3e5f59c11aeb90
Fixes: 14243b387137 ("KVM: x86/xen: Add KVM_IRQ_ROUTING_XEN_EVTCHN and event channel delivery")
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 32 +++++++++++++++++++++++---------
 1 file changed, 23 insertions(+), 9 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 91fd3673c09a..9bdb8e3cad58 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -697,6 +697,7 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 {
 	struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
+	bool atomic = in_atomic() || !task_is_running(current);
 	unsigned long flags;
 	u8 rc = 0;
 
@@ -713,7 +714,15 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 	BUILD_BUG_ON(sizeof(rc) !=
 		     sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
 
-	read_lock_irqsave(&gpc->lock, flags);
+	if (atomic) {
+		local_irq_save(flags);
+		if (!read_trylock(&gpc->lock)) {
+			local_irq_restore(flags);
+			return 1;
+		}
+	} else {
+		read_lock_irqsave(&gpc->lock, flags);
+	}
 	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
 		read_unlock_irqrestore(&gpc->lock, flags);
 
@@ -725,7 +734,7 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 		 * and we'll end up getting called again from a context where we *can*
 		 * fault in the page and wait for it.
 		 */
-		if (in_atomic() || !task_is_running(current))
+		if (atomic)
 			return 1;
 
 		if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info))) {
@@ -1794,7 +1803,6 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 	struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
 	struct kvm_vcpu *vcpu;
 	unsigned long *pending_bits, *mask_bits;
-	unsigned long flags;
 	int port_word_bit;
 	bool kick_vcpu = false;
 	int vcpu_idx, idx, rc;
@@ -1816,9 +1824,10 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 
 	idx = srcu_read_lock(&kvm->srcu);
 
-	read_lock_irqsave(&gpc->lock, flags);
-	if (!kvm_gpc_check(gpc, PAGE_SIZE))
+	if (!read_trylock(&gpc->lock))
 		goto out_rcu;
+	if (!kvm_gpc_check(gpc, PAGE_SIZE))
+		goto out_unlock;
 
 	if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
 		struct shared_info *shinfo = gpc->khva;
@@ -1847,11 +1856,10 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 	} else {
 		rc = 1; /* Delivered to the bitmap in shared_info. */
 		/* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
-		read_unlock_irqrestore(&gpc->lock, flags);
+		read_unlock(&gpc->lock);
 		gpc = &vcpu->arch.xen.vcpu_info_cache;
 
-		read_lock_irqsave(&gpc->lock, flags);
-		if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
+		if (!read_trylock(&gpc->lock)) {
 			/*
 			 * Could not access the vcpu_info. Set the bit in-kernel
 			 * and prod the vCPU to deliver it for itself.
@@ -1860,6 +1868,11 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 				kick_vcpu = true;
 			goto out_rcu;
 		}
+		if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
+			if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
+				kick_vcpu = true;
+			goto out_unlock;
+		}
 
 		if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
 			struct vcpu_info *vcpu_info = gpc->khva;
@@ -1883,8 +1896,9 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 		}
 	}
 
+ out_unlock:
+	read_unlock(&gpc->lock);
  out_rcu:
-	read_unlock_irqrestore(&gpc->lock, flags);
 	srcu_read_unlock(&kvm->srcu, idx);
 
 	if (kick_vcpu) {
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 03/20] KVM: x86/xen: Remove unnecessary irqsave from GPC lock usage in xen.c

[Sean Christopherson]

From: David Woodhouse <dwmw@amazon.co.uk>

Now that the hardirq path (xen_timer_callback and set_evtchn_fast) uses
read_trylock() instead of read_lock_irqsave(), the remaining GPC lock
users in xen.c are only called from process context (vcpu_run, ioctls).
There is no need to disable interrupts to prevent concurrent access from
a hardirq user, since the hardirq path no longer takes the lock.

Convert read_lock_irqsave()/read_unlock_irqrestore() to plain
read_lock()/read_unlock() in:
 - kvm_xen_update_runstate_guest()
 - kvm_xen_shared_info_init()
 - xen_get_guest_pvclock()
 - kvm_xen_inject_pending_events()
 - __kvm_xen_has_interrupt()
 - wait_pending_event()

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 60 +++++++++++++++++++---------------------------
 1 file changed, 25 insertions(+), 35 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 9bdb8e3cad58..b1fae42bf295 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -45,15 +45,15 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	int ret = 0;
 	int idx = srcu_read_lock(&kvm->srcu);
 
-	read_lock_irq(&gpc->lock);
+	read_lock(&gpc->lock);
 	while (!kvm_gpc_check(gpc, PAGE_SIZE)) {
-		read_unlock_irq(&gpc->lock);
+		read_unlock(&gpc->lock);
 
 		ret = kvm_gpc_refresh(gpc, PAGE_SIZE);
 		if (ret)
 			goto out;
 
-		read_lock_irq(&gpc->lock);
+		read_lock(&gpc->lock);
 	}
 
 	/*
@@ -96,7 +96,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	smp_wmb();
 
 	wc->version = wc_version + 1;
-	read_unlock_irq(&gpc->lock);
+	read_unlock(&gpc->lock);
 
 	kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
 
@@ -155,22 +155,21 @@ static int xen_get_guest_pvclock(struct kvm_vcpu *vcpu,
 				 struct gfn_to_pfn_cache *gpc,
 				 unsigned int offset)
 {
-	unsigned long flags;
 	int r;
 
-	read_lock_irqsave(&gpc->lock, flags);
+	read_lock(&gpc->lock);
 	while (!kvm_gpc_check(gpc, offset + sizeof(*hv_clock))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
+		read_unlock(&gpc->lock);
 
 		r = kvm_gpc_refresh(gpc, offset + sizeof(*hv_clock));
 		if (r)
 			return r;
 
-		read_lock_irqsave(&gpc->lock, flags);
+		read_lock(&gpc->lock);
 	}
 
 	memcpy(hv_clock, gpc->khva + offset, sizeof(*hv_clock));
-	read_unlock_irqrestore(&gpc->lock, flags);
+	read_unlock(&gpc->lock);
 
 	/*
 	 * Sanity check TSC shift+multiplier to verify the guest's view of time
@@ -325,7 +324,6 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 	struct gfn_to_pfn_cache *gpc2 = &vx->runstate2_cache;
 	size_t user_len, user_len1, user_len2;
 	struct vcpu_runstate_info rs;
-	unsigned long flags;
 	size_t times_ofs;
 	uint8_t *update_bit = NULL;
 	uint64_t entry_time;
@@ -421,16 +419,14 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 	 * gfn_to_pfn caches that cover the region.
 	 */
 	if (atomic) {
-		local_irq_save(flags);
 		if (!read_trylock(&gpc1->lock)) {
-			local_irq_restore(flags);
 			return;
 		}
 	} else {
-		read_lock_irqsave(&gpc1->lock, flags);
+		read_lock(&gpc1->lock);
 	}
 	while (!kvm_gpc_check(gpc1, user_len1)) {
-		read_unlock_irqrestore(&gpc1->lock, flags);
+		read_unlock(&gpc1->lock);
 
 		/* When invoked from kvm_sched_out() we cannot sleep */
 		if (atomic)
@@ -439,7 +435,7 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 		if (kvm_gpc_refresh(gpc1, user_len1))
 			return;
 
-		read_lock_irqsave(&gpc1->lock, flags);
+		read_lock(&gpc1->lock);
 	}
 
 	if (likely(!user_len2)) {
@@ -467,7 +463,7 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 		lock_set_subclass(&gpc1->lock.dep_map, 1, _THIS_IP_);
 		if (atomic) {
 			if (!read_trylock(&gpc2->lock)) {
-				read_unlock_irqrestore(&gpc1->lock, flags);
+				read_unlock(&gpc1->lock);
 				return;
 			}
 		} else {
@@ -476,7 +472,7 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 
 		if (!kvm_gpc_check(gpc2, user_len2)) {
 			read_unlock(&gpc2->lock);
-			read_unlock_irqrestore(&gpc1->lock, flags);
+			read_unlock(&gpc1->lock);
 
 			/* When invoked from kvm_sched_out() we cannot sleep */
 			if (atomic)
@@ -581,7 +577,7 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 	}
 
 	kvm_gpc_mark_dirty_in_slot(gpc1);
-	read_unlock_irqrestore(&gpc1->lock, flags);
+	read_unlock(&gpc1->lock);
 }
 
 void kvm_xen_update_runstate(struct kvm_vcpu *v, int state)
@@ -640,7 +636,6 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 {
 	unsigned long evtchn_pending_sel = READ_ONCE(v->arch.xen.evtchn_pending_sel);
 	struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
-	unsigned long flags;
 
 	if (!evtchn_pending_sel)
 		return;
@@ -650,14 +645,14 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 	 * does anyway. Page it in and retry the instruction. We're just a
 	 * little more honest about it.
 	 */
-	read_lock_irqsave(&gpc->lock, flags);
+	read_lock(&gpc->lock);
 	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
+		read_unlock(&gpc->lock);
 
 		if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info)))
 			return;
 
-		read_lock_irqsave(&gpc->lock, flags);
+		read_lock(&gpc->lock);
 	}
 
 	/* Now gpc->khva is a valid kernel address for the vcpu_info */
@@ -687,7 +682,7 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 	}
 
 	kvm_gpc_mark_dirty_in_slot(gpc);
-	read_unlock_irqrestore(&gpc->lock, flags);
+	read_unlock(&gpc->lock);
 
 	/* For the per-vCPU lapic vector, deliver it as MSI. */
 	if (v->arch.xen.upcall_vector)
@@ -698,7 +693,6 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 {
 	struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
 	bool atomic = in_atomic() || !task_is_running(current);
-	unsigned long flags;
 	u8 rc = 0;
 
 	/*
@@ -715,16 +709,13 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 		     sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
 
 	if (atomic) {
-		local_irq_save(flags);
-		if (!read_trylock(&gpc->lock)) {
-			local_irq_restore(flags);
+		if (!read_trylock(&gpc->lock))
 			return 1;
-		}
 	} else {
-		read_lock_irqsave(&gpc->lock, flags);
+		read_lock(&gpc->lock);
 	}
 	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
+		read_unlock(&gpc->lock);
 
 		/*
 		 * This function gets called from kvm_vcpu_block() after setting the
@@ -744,11 +735,11 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 			 */
 			return 0;
 		}
-		read_lock_irqsave(&gpc->lock, flags);
+		read_lock(&gpc->lock);
 	}
 
 	rc = ((struct vcpu_info *)gpc->khva)->evtchn_upcall_pending;
-	read_unlock_irqrestore(&gpc->lock, flags);
+	read_unlock(&gpc->lock);
 	return rc;
 }
 
@@ -1445,12 +1436,11 @@ static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
 	struct kvm *kvm = vcpu->kvm;
 	struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
 	unsigned long *pending_bits;
-	unsigned long flags;
 	bool ret = true;
 	int idx, i;
 
 	idx = srcu_read_lock(&kvm->srcu);
-	read_lock_irqsave(&gpc->lock, flags);
+	read_lock(&gpc->lock);
 	if (!kvm_gpc_check(gpc, PAGE_SIZE))
 		goto out_rcu;
 
@@ -1471,7 +1461,7 @@ static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
 	}
 
  out_rcu:
-	read_unlock_irqrestore(&gpc->lock, flags);
+	read_unlock(&gpc->lock);
 	srcu_read_unlock(&kvm->srcu, idx);
 
 	return ret;
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 04/20] KVM: x86: Remove unnecessary irqsave from kvm_setup_guest_pvclock()

[Sean Christopherson]

From: David Woodhouse <dwmw@amazon.co.uk>

kvm_setup_guest_pvclock() is only called from kvm_guest_time_update()
which runs in process context (vcpu_enter_guest or ioctl). There is no
hardirq path that takes the GPC read lock for pvclock, so irqsave is
unnecessary.

Convert to plain read_lock()/read_unlock().

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/x86.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index e6f1dd84f22d..87e99756de0a 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3265,18 +3265,17 @@ static void kvm_setup_guest_pvclock(struct pvclock_vcpu_time_info *ref_hv_clock,
 {
 	struct pvclock_vcpu_time_info *guest_hv_clock;
 	struct pvclock_vcpu_time_info hv_clock;
-	unsigned long flags;
 
 	memcpy(&hv_clock, ref_hv_clock, sizeof(hv_clock));
 
-	read_lock_irqsave(&gpc->lock, flags);
+	read_lock(&gpc->lock);
 	while (!kvm_gpc_check(gpc, offset + sizeof(*guest_hv_clock))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
+		read_unlock(&gpc->lock);
 
 		if (kvm_gpc_refresh(gpc, offset + sizeof(*guest_hv_clock)))
 			return;
 
-		read_lock_irqsave(&gpc->lock, flags);
+		read_lock(&gpc->lock);
 	}
 
 	guest_hv_clock = (void *)(gpc->khva + offset);
@@ -3301,7 +3300,7 @@ static void kvm_setup_guest_pvclock(struct pvclock_vcpu_time_info *ref_hv_clock,
 	guest_hv_clock->version = ++hv_clock.version;
 
 	kvm_gpc_mark_dirty_in_slot(gpc);
-	read_unlock_irqrestore(&gpc->lock, flags);
+	read_unlock(&gpc->lock);
 
 	trace_kvm_pvclock_update(vcpu->vcpu_id, &hv_clock);
 }
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 05/20] KVM: Remove unnecessary IRQ disabling from GPC lock in pfncache.c

[Sean Christopherson]

From: David Woodhouse <dwmw@amazon.co.uk>

Now that all hardirq/atomic GPC users (xen_timer_callback,
kvm_xen_set_evtchn_fast) use read_trylock() instead of read_lock(), no
hardirq path ever holds the GPC rwlock. There is therefore no risk of
deadlock between the write side and a hardirq reader, and no need to
disable interrupts when taking the lock.

Convert all read_lock_irq()/write_lock_irq() and their unlock
counterparts to plain read_lock()/write_lock() in pfncache.c.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 virt/kvm/pfncache.c | 28 ++++++++++++++--------------
 1 file changed, 14 insertions(+), 14 deletions(-)

diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c
index 728d2c1b488a..70b102095173 100644
--- a/virt/kvm/pfncache.c
+++ b/virt/kvm/pfncache.c
@@ -29,12 +29,12 @@ void gfn_to_pfn_cache_invalidate_start(struct kvm *kvm, unsigned long start,
 
 	spin_lock(&kvm->gpc_lock);
 	list_for_each_entry(gpc, &kvm->gpc_list, list) {
-		read_lock_irq(&gpc->lock);
+		read_lock(&gpc->lock);
 
 		/* Only a single page so no need to care about length */
 		if (gpc->valid && !is_error_noslot_pfn(gpc->pfn) &&
 		    gpc->uhva >= start && gpc->uhva < end) {
-			read_unlock_irq(&gpc->lock);
+			read_unlock(&gpc->lock);
 
 			/*
 			 * There is a small window here where the cache could
@@ -44,15 +44,15 @@ void gfn_to_pfn_cache_invalidate_start(struct kvm *kvm, unsigned long start,
 			 * acquired.
 			 */
 
-			write_lock_irq(&gpc->lock);
+			write_lock(&gpc->lock);
 			if (gpc->valid && !is_error_noslot_pfn(gpc->pfn) &&
 			    gpc->uhva >= start && gpc->uhva < end)
 				gpc->valid = false;
-			write_unlock_irq(&gpc->lock);
+			write_unlock(&gpc->lock);
 			continue;
 		}
 
-		read_unlock_irq(&gpc->lock);
+		read_unlock(&gpc->lock);
 	}
 	spin_unlock(&kvm->gpc_lock);
 }
@@ -184,7 +184,7 @@ static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc)
 		mmu_seq = gpc->kvm->mmu_invalidate_seq;
 		smp_rmb();
 
-		write_unlock_irq(&gpc->lock);
+		write_unlock(&gpc->lock);
 
 		/*
 		 * If the previous iteration "failed" due to an mmu_notifier
@@ -225,7 +225,7 @@ static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc)
 			goto out_error;
 		}
 
-		write_lock_irq(&gpc->lock);
+		write_lock(&gpc->lock);
 
 		/*
 		 * Other tasks must wait for _this_ refresh to complete before
@@ -248,7 +248,7 @@ static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc)
 	return 0;
 
 out_error:
-	write_lock_irq(&gpc->lock);
+	write_lock(&gpc->lock);
 
 	return -EFAULT;
 }
@@ -269,7 +269,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l
 
 	lockdep_assert_held(&gpc->refresh_lock);
 
-	write_lock_irq(&gpc->lock);
+	write_lock(&gpc->lock);
 
 	if (!gpc->active) {
 		ret = -EINVAL;
@@ -355,7 +355,7 @@ static int __kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned l
 	unmap_old = (old_pfn != gpc->pfn);
 
 out_unlock:
-	write_unlock_irq(&gpc->lock);
+	write_unlock(&gpc->lock);
 
 	if (unmap_old)
 		gpc_unmap(old_pfn, old_khva);
@@ -417,9 +417,9 @@ static int __kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned
 		 * refresh must not establish a mapping until the cache is
 		 * reachable by mmu_notifier events.
 		 */
-		write_lock_irq(&gpc->lock);
+		write_lock(&gpc->lock);
 		gpc->active = true;
-		write_unlock_irq(&gpc->lock);
+		write_unlock(&gpc->lock);
 	}
 	return __kvm_gpc_refresh(gpc, gpa, uhva);
 }
@@ -458,7 +458,7 @@ void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc)
 		 * must stall mmu_notifier events until all users go away, i.e.
 		 * until gpc->lock is dropped and refresh is guaranteed to fail.
 		 */
-		write_lock_irq(&gpc->lock);
+		write_lock(&gpc->lock);
 		gpc->active = false;
 		gpc->valid = false;
 
@@ -473,7 +473,7 @@ void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc)
 
 		old_pfn = gpc->pfn;
 		gpc->pfn = KVM_PFN_ERR_FAULT;
-		write_unlock_irq(&gpc->lock);
+		write_unlock(&gpc->lock);
 
 		spin_lock(&kvm->gpc_lock);
 		list_del(&gpc->list);
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 06/20] KVM: x86/xen: Use guard() to grab kvm->srcu around gpc critical sections

[Sean Christopherson]

Use guard() to acquire and release kvm->srcu protection around gpc
critical sections, so that said critical sections can also use the fancy
__cleanup() functionality.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 32 ++++++++++++++------------------
 1 file changed, 14 insertions(+), 18 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index b1fae42bf295..0c6b74b97408 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -42,8 +42,9 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	u32 *wc_sec_hi;
 	u32 wc_version;
 	u64 wall_nsec;
-	int ret = 0;
-	int idx = srcu_read_lock(&kvm->srcu);
+	int ret;
+
+	guard(srcu)(&kvm->srcu);
 
 	read_lock(&gpc->lock);
 	while (!kvm_gpc_check(gpc, PAGE_SIZE)) {
@@ -51,7 +52,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 
 		ret = kvm_gpc_refresh(gpc, PAGE_SIZE);
 		if (ret)
-			goto out;
+			return ret;
 
 		read_lock(&gpc->lock);
 	}
@@ -99,10 +100,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	read_unlock(&gpc->lock);
 
 	kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
-
-out:
-	srcu_read_unlock(&kvm->srcu, idx);
-	return ret;
+	return 0;
 }
 
 void kvm_xen_inject_timer_irqs(struct kvm_vcpu *vcpu)
@@ -1437,9 +1435,10 @@ static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
 	struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
 	unsigned long *pending_bits;
 	bool ret = true;
-	int idx, i;
+	int i;
+
+	guard(srcu)(&kvm->srcu);
 
-	idx = srcu_read_lock(&kvm->srcu);
 	read_lock(&gpc->lock);
 	if (!kvm_gpc_check(gpc, PAGE_SIZE))
 		goto out_rcu;
@@ -1462,8 +1461,6 @@ static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
 
  out_rcu:
 	read_unlock(&gpc->lock);
-	srcu_read_unlock(&kvm->srcu, idx);
-
 	return ret;
 }
 
@@ -1795,7 +1792,7 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 	unsigned long *pending_bits, *mask_bits;
 	int port_word_bit;
 	bool kick_vcpu = false;
-	int vcpu_idx, idx, rc;
+	int vcpu_idx, rc;
 
 	vcpu_idx = READ_ONCE(xe->vcpu_idx);
 	if (vcpu_idx >= 0)
@@ -1812,10 +1809,11 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 
 	rc = -EWOULDBLOCK;
 
-	idx = srcu_read_lock(&kvm->srcu);
+	guard(srcu)(&kvm->srcu);
 
 	if (!read_trylock(&gpc->lock))
-		goto out_rcu;
+		return rc;
+
 	if (!kvm_gpc_check(gpc, PAGE_SIZE))
 		goto out_unlock;
 
@@ -1856,7 +1854,7 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 			 */
 			if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
 				kick_vcpu = true;
-			goto out_rcu;
+			goto out_kick;
 		}
 		if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
 			if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
@@ -1888,9 +1886,7 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 
  out_unlock:
 	read_unlock(&gpc->lock);
- out_rcu:
-	srcu_read_unlock(&kvm->srcu, idx);
-
+ out_kick:
 	if (kick_vcpu) {
 		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
 		kvm_vcpu_kick(vcpu);
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 07/20] KVM: x86/xen: Extract delivery of event to vCPU into a separate helper

[Sean Christopherson]

Hoist the actual fastpath delivery of an event to a vCPU into a separate
helper so that CLASS()-based gpc locking+checking can be used without
needing to implement scoped versions.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 100 ++++++++++++++++++++++++---------------------
 1 file changed, 54 insertions(+), 46 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 0c6b74b97408..020ef0ddab01 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -1775,6 +1775,57 @@ static void kvm_xen_check_poller(struct kvm_vcpu *vcpu, int port)
 	}
 }
 
+static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
+{
+	struct gfn_to_pfn_cache *gpc = &vcpu->arch.xen.vcpu_info_cache;
+	bool kick_vcpu = false;
+
+	/* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
+	if (!read_trylock(&gpc->lock)) {
+		/*
+		 * Could not access the vcpu_info. Set the bit in-kernel and
+		 * prod the vCPU to deliver it for itself.
+		 */
+		if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
+			kick_vcpu = true;
+		goto out_kick;
+	}
+	if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
+		if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
+			kick_vcpu = true;
+		goto out_unlock;
+	}
+
+	if (IS_ENABLED(CONFIG_64BIT) && vcpu->kvm->arch.xen.long_mode) {
+		struct vcpu_info *vcpu_info = gpc->khva;
+		if (!test_and_set_bit(port_word_bit, &vcpu_info->evtchn_pending_sel)) {
+			WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
+			kick_vcpu = true;
+		}
+	} else {
+		struct compat_vcpu_info *vcpu_info = gpc->khva;
+		if (!test_and_set_bit(port_word_bit,
+					(unsigned long *)&vcpu_info->evtchn_pending_sel)) {
+			WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
+			kick_vcpu = true;
+		}
+	}
+
+	/* For the per-vCPU lapic vector, deliver it as MSI. */
+	if (kick_vcpu && vcpu->arch.xen.upcall_vector) {
+		kvm_xen_inject_vcpu_vector(vcpu);
+		kick_vcpu = false;
+	}
+
+out_unlock:
+	read_unlock(&gpc->lock);
+out_kick:
+	if (kick_vcpu) {
+		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
+		kvm_vcpu_kick(vcpu);
+	}
+}
+
 /*
  * The return value from this function is propagated to kvm_set_irq() API,
  * so it returns:
@@ -1791,7 +1842,6 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 	struct kvm_vcpu *vcpu;
 	unsigned long *pending_bits, *mask_bits;
 	int port_word_bit;
-	bool kick_vcpu = false;
 	int vcpu_idx, rc;
 
 	vcpu_idx = READ_ONCE(xe->vcpu_idx);
@@ -1843,55 +1893,13 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 		kvm_xen_check_poller(vcpu, xe->port);
 	} else {
 		rc = 1; /* Delivered to the bitmap in shared_info. */
-		/* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
-		read_unlock(&gpc->lock);
-		gpc = &vcpu->arch.xen.vcpu_info_cache;
-
-		if (!read_trylock(&gpc->lock)) {
-			/*
-			 * Could not access the vcpu_info. Set the bit in-kernel
-			 * and prod the vCPU to deliver it for itself.
-			 */
-			if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
-				kick_vcpu = true;
-			goto out_kick;
-		}
-		if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-			if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
-				kick_vcpu = true;
-			goto out_unlock;
-		}
-
-		if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
-			struct vcpu_info *vcpu_info = gpc->khva;
-			if (!test_and_set_bit(port_word_bit, &vcpu_info->evtchn_pending_sel)) {
-				WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
-				kick_vcpu = true;
-			}
-		} else {
-			struct compat_vcpu_info *vcpu_info = gpc->khva;
-			if (!test_and_set_bit(port_word_bit,
-					      (unsigned long *)&vcpu_info->evtchn_pending_sel)) {
-				WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
-				kick_vcpu = true;
-			}
-		}
-
-		/* For the per-vCPU lapic vector, deliver it as MSI. */
-		if (kick_vcpu && vcpu->arch.xen.upcall_vector) {
-			kvm_xen_inject_vcpu_vector(vcpu);
-			kick_vcpu = false;
-		}
 	}
 
- out_unlock:
+out_unlock:
 	read_unlock(&gpc->lock);
- out_kick:
-	if (kick_vcpu) {
-		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
-		kvm_vcpu_kick(vcpu);
-	}
 
+	if (rc == 1)
+		__kvm_xen_set_evtchn_fast(vcpu, port_word_bit);
 	return rc;
 }
 
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 08/20] KVM: x86/xen: Explicitly tag "shared info" page as never being dirty tracked

[Sean Christopherson]

Explicitly mark the Xen shared info page as never being dirty tracked so
that higher-level gpc APIs can be added to automatically take care of
things like dirty tracking, without reintroducing the bug fixed by commit
55749769fe60 ("KVM: x86: Fix wall clock writes in Xen shared_info not to
mark page dirty").  And because the code _looks_ buggy.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c        |  2 +-
 include/linux/kvm_host.h  | 13 ++++++++++---
 include/linux/kvm_types.h |  1 +
 virt/kvm/pfncache.c       |  4 +++-
 4 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 020ef0ddab01..ab8e95647406 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -2334,7 +2334,7 @@ void kvm_xen_init_vm(struct kvm *kvm)
 {
 	mutex_init(&kvm->arch.xen.xen_lock);
 	idr_init(&kvm->arch.xen.evtchn_ports);
-	kvm_gpc_init(&kvm->arch.xen.shinfo_cache, kvm);
+	__kvm_gpc_init(&kvm->arch.xen.shinfo_cache, kvm, true);
 }
 
 void kvm_xen_destroy_vm(struct kvm *kvm)
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 27498e990dff..0dc4eb78b6d9 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -1427,16 +1427,23 @@ int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
 			 unsigned long len);
 
 /**
- * kvm_gpc_init - initialize gfn_to_pfn_cache.
+ * __kvm_gpc_init - initialize gfn_to_pfn_cache.
  *
  * @gpc:	   struct gfn_to_pfn_cache object.
  * @kvm:	   pointer to kvm instance.
+ * @never_dirty:   %true if the associated gfn should never be marked dirty
  *
  * This sets up a gfn_to_pfn_cache by initializing locks and assigning the
  * immutable attributes.  Note, the cache must be zero-allocated (or zeroed by
  * the caller before init).
  */
-void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm);
+void __kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm,
+		    bool never_dirty);
+
+static inline void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm)
+{
+	__kvm_gpc_init(gpc, kvm, false);
+}
 
 /**
  * kvm_gpc_activate - prepare a cached kernel mapping and HPA for a given guest
@@ -1942,7 +1949,7 @@ static inline void kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache *gpc)
 {
 	lockdep_assert_held(&gpc->lock);
 
-	if (!gpc->memslot)
+	if (!gpc->memslot || gpc->never_dirty)
 		return;
 
 	mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpa_to_gfn(gpc->gpa));
diff --git a/include/linux/kvm_types.h b/include/linux/kvm_types.h
index a568d8e6f4e8..e850adc3f47e 100644
--- a/include/linux/kvm_types.h
+++ b/include/linux/kvm_types.h
@@ -94,6 +94,7 @@ struct gfn_to_pfn_cache {
 	kvm_pfn_t pfn;
 	bool active;
 	bool valid;
+	bool never_dirty;
 };
 
 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c
index 70b102095173..9209f06c46b4 100644
--- a/virt/kvm/pfncache.c
+++ b/virt/kvm/pfncache.c
@@ -382,7 +382,8 @@ int kvm_gpc_refresh(struct gfn_to_pfn_cache *gpc, unsigned long len)
 	return __kvm_gpc_refresh(gpc, gpc->gpa, uhva);
 }
 
-void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm)
+void __kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm,
+		    bool never_dirty)
 {
 	rwlock_init(&gpc->lock);
 	mutex_init(&gpc->refresh_lock);
@@ -392,6 +393,7 @@ void kvm_gpc_init(struct gfn_to_pfn_cache *gpc, struct kvm *kvm)
 	gpc->gpa = INVALID_GPA;
 	gpc->uhva = KVM_HVA_ERR_BAD;
 	gpc->active = gpc->valid = false;
+	gpc->never_dirty = never_dirty;
 }
 
 static int __kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long uhva,
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 09/20] KVM: x86/xen: Don't dirty track "vCPU info" page

[Sean Christopherson]

Give the Xen per-vCPU info page the same treatment as the per-VM shared
info page, and never mark it dirty, as KVM clearly relies on userspace to
assume the page is always dirty.  While the page is marked dirty on writes
via kvm_xen_inject_pending_events(), it's not marked dirty when written by
__kvm_xen_set_evtchn_fast().

Furthermore, as was the case with the shared info page, writes in the event
channel fastpath may be done without an active vCPU, e.g. when called via
timer callback or irqfd injection.  I.e. attempting to fix the faspath
would run afoul of same issue that was fixed by commit ("KVM: x86: Fix wall
clock writes in Xen shared_info not to mark page dirty").

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index ab8e95647406..7b527a983cfc 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -679,7 +679,6 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 		WRITE_ONCE(vi->evtchn_upcall_pending, 1);
 	}
 
-	kvm_gpc_mark_dirty_in_slot(gpc);
 	read_unlock(&gpc->lock);
 
 	/* For the per-vCPU lapic vector, deliver it as MSI. */
@@ -2313,7 +2312,7 @@ void kvm_xen_init_vcpu(struct kvm_vcpu *vcpu)
 
 	kvm_gpc_init(&vcpu->arch.xen.runstate_cache, vcpu->kvm);
 	kvm_gpc_init(&vcpu->arch.xen.runstate2_cache, vcpu->kvm);
-	kvm_gpc_init(&vcpu->arch.xen.vcpu_info_cache, vcpu->kvm);
+	__kvm_gpc_init(&vcpu->arch.xen.vcpu_info_cache, vcpu->kvm, true);
 	kvm_gpc_init(&vcpu->arch.xen.vcpu_time_info_cache, vcpu->kvm);
 }
 
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 10/20] KVM: Move {g,p}fn <=> {g,h}pa conversion helpers to kvm_types.h

[Sean Christopherson]

Move the helpers to convert between pfns and physical addresses (for host
and guest) into kvm_types.h so that they can be used throughout kvm_host.h
(and any other KVM-related header) without having to worry about ordering.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 include/linux/kvm_host.h  | 15 ---------------
 include/linux/kvm_types.h | 16 ++++++++++++++++
 2 files changed, 16 insertions(+), 15 deletions(-)

diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 0dc4eb78b6d9..ffbae1e6e84e 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -1923,21 +1923,6 @@ hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
 	return slot->base_gfn + gfn_offset;
 }
 
-static inline gpa_t gfn_to_gpa(gfn_t gfn)
-{
-	return (gpa_t)gfn << PAGE_SHIFT;
-}
-
-static inline gfn_t gpa_to_gfn(gpa_t gpa)
-{
-	return (gfn_t)(gpa >> PAGE_SHIFT);
-}
-
-static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
-{
-	return (hpa_t)pfn << PAGE_SHIFT;
-}
-
 static inline bool kvm_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
 {
 	unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
diff --git a/include/linux/kvm_types.h b/include/linux/kvm_types.h
index e850adc3f47e..961572e102f0 100644
--- a/include/linux/kvm_types.h
+++ b/include/linux/kvm_types.h
@@ -6,6 +6,7 @@
 #include <linux/bits.h>
 #include <linux/export.h>
 #include <linux/types.h>
+#include <vdso/page.h>
 #include <asm/kvm_types.h>
 
 #ifdef KVM_SUB_MODULES
@@ -73,6 +74,21 @@ typedef u64            hfn_t;
 
 typedef hfn_t kvm_pfn_t;
 
+static inline gpa_t gfn_to_gpa(gfn_t gfn)
+{
+	return (gpa_t)gfn << PAGE_SHIFT;
+}
+
+static inline gfn_t gpa_to_gfn(gpa_t gpa)
+{
+	return (gfn_t)(gpa >> PAGE_SHIFT);
+}
+
+static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
+{
+	return (hpa_t)pfn << PAGE_SHIFT;
+}
+
 struct gfn_to_hva_cache {
 	u64 generation;
 	gpa_t gpa;
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 11/20] KVM: Add CLASS() constructs to automagically handle lock+check of gpc

[Sean Christopherson]

Add CLASS() definitions for locally mapping a PFN given a gpc (gfn_to_pfn
cache), to eventually deduplicate a number of users that do:

  lock();
  while (!check()) {
    unlock();

    if (refresh())
	return err;

    lock()
  }

  ...

  mark_dirty();
  unlock();

Implement read-only (for cases where KVM is only reading) and "try" (for
use in atomic code where rwlock might sleep due to PREEMPT_RT) variations.

Use "map local" as the primary terminology as the basic concept is more or
less the same as kmap_local(): ensure the current CPU has a kernel mapping
to the underlying memory.

Convert the pvclock code as the first user, as it is straightforward and
thus easier to audit for correctness.

For all intents and purposes, no functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/x86.c       | 17 ++++----------
 include/linux/kvm_host.h | 48 +++++++++++++++++++++++++++++++---------
 virt/kvm/pfncache.c      | 36 ++++++++++++++++++++++++++++++
 3 files changed, 78 insertions(+), 23 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 87e99756de0a..ea10ed4ab06f 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3268,17 +3268,11 @@ static void kvm_setup_guest_pvclock(struct pvclock_vcpu_time_info *ref_hv_clock,
 
 	memcpy(&hv_clock, ref_hv_clock, sizeof(hv_clock));
 
-	read_lock(&gpc->lock);
-	while (!kvm_gpc_check(gpc, offset + sizeof(*guest_hv_clock))) {
-		read_unlock(&gpc->lock);
+	CLASS(gpc_map_local, clock_map)(gpc, offset + sizeof(*guest_hv_clock));
+	if (IS_ERR(clock_map))
+		return;
 
-		if (kvm_gpc_refresh(gpc, offset + sizeof(*guest_hv_clock)))
-			return;
-
-		read_lock(&gpc->lock);
-	}
-
-	guest_hv_clock = (void *)(gpc->khva + offset);
+	guest_hv_clock = *clock_map + offset;
 
 	/*
 	 * This VCPU is paused, but it's legal for a guest to read another
@@ -3299,9 +3293,6 @@ static void kvm_setup_guest_pvclock(struct pvclock_vcpu_time_info *ref_hv_clock,
 
 	guest_hv_clock->version = ++hv_clock.version;
 
-	kvm_gpc_mark_dirty_in_slot(gpc);
-	read_unlock(&gpc->lock);
-
 	trace_kvm_pvclock_update(vcpu->vcpu_id, &hv_clock);
 }
 
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index ffbae1e6e84e..d70fa91cda0c 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -1535,6 +1535,44 @@ static inline bool kvm_gpc_is_hva_active(struct gfn_to_pfn_cache *gpc)
 	return gpc->active && kvm_is_error_gpa(gpc->gpa);
 }
 
+static inline void kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache *gpc)
+{
+	lockdep_assert_held(&gpc->lock);
+
+	if (!gpc->memslot || gpc->never_dirty)
+		return;
+
+	mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpa_to_gfn(gpc->gpa));
+}
+
+void **gpc_map_local_lock(struct gfn_to_pfn_cache *gpc, unsigned long len);
+void **gpc_try_map_local_lock(struct gfn_to_pfn_cache *gpc, unsigned long len);
+
+static inline void gpc_map_local_unlock(void **khva)
+{
+	struct gfn_to_pfn_cache *gpc = container_of(khva, struct gfn_to_pfn_cache, khva);
+
+	kvm_gpc_mark_dirty_in_slot(gpc);
+
+	read_unlock(&gpc->lock);
+}
+
+static inline void gpc_map_local_unlock_ro(void **khva)
+{
+	read_unlock(&container_of(khva, struct gfn_to_pfn_cache, khva)->lock);
+}
+
+#define DEFINE_GPC_CLASS(try, ro)						\
+DEFINE_CLASS(gpc##try##_map_local##ro, void **,					\
+	     if (!IS_ERR(_T)) gpc_map_local_unlock##ro(_T),			\
+	     gpc##try##_map_local_lock(gpc, len),				\
+	     struct gfn_to_pfn_cache *gpc, unsigned long len)			\
+
+DEFINE_GPC_CLASS(,);
+DEFINE_GPC_CLASS(_try,);
+DEFINE_GPC_CLASS(, _ro);
+DEFINE_GPC_CLASS(_try, _ro);
+
 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
 
@@ -1930,16 +1968,6 @@ static inline bool kvm_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
 	return !kvm_is_error_hva(hva);
 }
 
-static inline void kvm_gpc_mark_dirty_in_slot(struct gfn_to_pfn_cache *gpc)
-{
-	lockdep_assert_held(&gpc->lock);
-
-	if (!gpc->memslot || gpc->never_dirty)
-		return;
-
-	mark_page_dirty_in_slot(gpc->kvm, gpc->memslot, gpa_to_gfn(gpc->gpa));
-}
-
 enum kvm_stat_kind {
 	KVM_STAT_VM,
 	KVM_STAT_VCPU,
diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c
index 9209f06c46b4..d3e02a2bac38 100644
--- a/virt/kvm/pfncache.c
+++ b/virt/kvm/pfncache.c
@@ -484,3 +484,39 @@ void kvm_gpc_deactivate(struct gfn_to_pfn_cache *gpc)
 		gpc_unmap(old_pfn, old_khva);
 	}
 }
+
+void **gpc_try_map_local_lock(struct gfn_to_pfn_cache *gpc, unsigned long len)
+{
+	if (!read_trylock(&gpc->lock))
+		return ERR_PTR(-EWOULDBLOCK);
+
+	if (!kvm_gpc_check(gpc, len)) {
+		read_unlock(&gpc->lock);
+		return ERR_PTR(-EWOULDBLOCK);
+	}
+
+	return &gpc->khva;
+}
+
+void **gpc_map_local_lock(struct gfn_to_pfn_cache *gpc, unsigned long len)
+{
+	/*
+	 * Yes, this is an open-coded loop. But that's just what put_user()
+	 * does anyway. Page it in and retry the instruction. We're just a
+	 * little more honest about it.
+	 */
+	for (;;) {
+		int r;
+
+		read_lock(&gpc->lock);
+
+		if (kvm_gpc_check(gpc, len))
+			return &gpc->khva;
+
+		read_unlock(&gpc->lock);
+
+		r = kvm_gpc_refresh(gpc, len);
+		if (r)
+			return ERR_PTR(r);
+	}
+}
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 12/20] KVM: x86/xen: Convert kvm_xen_shared_info_init() to gpc's CLASS() APIs

[Sean Christopherson]

Use the newfangled "map local" CLASS() APIs to access the Xen shared info
page via its gpc.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 34 +++++++++++++++++-----------------
 1 file changed, 17 insertions(+), 17 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 7b527a983cfc..065b4c92f7ed 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -35,27 +35,19 @@ static bool kvm_xen_hcall_evtchn_send(struct kvm_vcpu *vcpu, u64 param, u64 *r);
 
 DEFINE_STATIC_KEY_DEFERRED_FALSE(kvm_xen_enabled, HZ);
 
-static int kvm_xen_shared_info_init(struct kvm *kvm)
+static int __kvm_xen_shared_info_init(struct kvm *kvm)
 {
 	struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
 	struct pvclock_wall_clock *wc;
 	u32 *wc_sec_hi;
 	u32 wc_version;
 	u64 wall_nsec;
-	int ret;
 
 	guard(srcu)(&kvm->srcu);
 
-	read_lock(&gpc->lock);
-	while (!kvm_gpc_check(gpc, PAGE_SIZE)) {
-		read_unlock(&gpc->lock);
-
-		ret = kvm_gpc_refresh(gpc, PAGE_SIZE);
-		if (ret)
-			return ret;
-
-		read_lock(&gpc->lock);
-	}
+	CLASS(gpc_map_local, shinfo_map)(gpc, PAGE_SIZE);
+	if (IS_ERR(shinfo_map))
+		return PTR_ERR(shinfo_map);
 
 	/*
 	 * This code mirrors kvm_write_wall_clock() except that it writes
@@ -74,14 +66,14 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	BUILD_BUG_ON(offsetof(struct shared_info, wc_sec_hi) != 0xc0c);
 
 	if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
-		struct shared_info *shinfo = gpc->khva;
+		struct shared_info *shinfo = *shinfo_map;
 
 		wc_sec_hi = &shinfo->wc_sec_hi;
 		wc = &shinfo->wc;
 	} else
 #endif
 	{
-		struct compat_shared_info *shinfo = gpc->khva;
+		struct compat_shared_info *shinfo = *shinfo_map;
 
 		wc_sec_hi = &shinfo->arch.wc_sec_hi;
 		wc = &shinfo->wc;
@@ -97,12 +89,20 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	smp_wmb();
 
 	wc->version = wc_version + 1;
-	read_unlock(&gpc->lock);
-
-	kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
 	return 0;
 }
 
+static int kvm_xen_shared_info_init(struct kvm *kvm)
+{
+	int r;
+
+	r = __kvm_xen_shared_info_init(kvm);
+	if (!r)
+		kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
+
+	return r;
+}
+
 void kvm_xen_inject_timer_irqs(struct kvm_vcpu *vcpu)
 {
 	if (atomic_read(&vcpu->arch.xen.timer_pending) > 0) {
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 13/20] KVM: x86/xen: Don't bother waiting on gpc->lock in SCHEDOP_poll

[Sean Christopherson]

Use trylock instead of waiting on the gpc->lock when locking and checking
the shared info page for SCHEDOP_poll, as odds are very good that if the
lock is contended, then the check will fail.  This will allow using the new
CLASS() APIs for local gpc mappings, without having to add a version that
waits on the lock, but doesn't refresh on failure.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 065b4c92f7ed..d9b09809e243 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -1438,7 +1438,9 @@ static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
 
 	guard(srcu)(&kvm->srcu);
 
-	read_lock(&gpc->lock);
+	if (!read_trylock(&gpc->lock))
+		return ret;
+
 	if (!kvm_gpc_check(gpc, PAGE_SIZE))
 		goto out_rcu;
 
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 14/20] KVM: x86/xen: Convert wait_pending_event() to gpc's CLASS() APIs

[Sean Christopherson]

Use gpc's CLASS() interface to lock and check the shared info page when
processing a SCHEDOP_poll hypercall.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 24 ++++++++----------------
 1 file changed, 8 insertions(+), 16 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index d9b09809e243..8f822acb11a4 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -1433,36 +1433,28 @@ static bool wait_pending_event(struct kvm_vcpu *vcpu, int nr_ports,
 	struct kvm *kvm = vcpu->kvm;
 	struct gfn_to_pfn_cache *gpc = &kvm->arch.xen.shinfo_cache;
 	unsigned long *pending_bits;
-	bool ret = true;
 	int i;
 
 	guard(srcu)(&kvm->srcu);
 
-	if (!read_trylock(&gpc->lock))
-		return ret;
+	CLASS(gpc_try_map_local_ro, shinfo_map)(gpc, PAGE_SIZE);
+	if (IS_ERR(shinfo_map))
+		return true;
 
-	if (!kvm_gpc_check(gpc, PAGE_SIZE))
-		goto out_rcu;
-
-	ret = false;
 	if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
-		struct shared_info *shinfo = gpc->khva;
+		struct shared_info *shinfo = *shinfo_map;
 		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
 	} else {
-		struct compat_shared_info *shinfo = gpc->khva;
+		struct compat_shared_info *shinfo = *shinfo_map;
 		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
 	}
 
 	for (i = 0; i < nr_ports; i++) {
-		if (test_bit(ports[i], pending_bits)) {
-			ret = true;
-			break;
-		}
+		if (test_bit(ports[i], pending_bits))
+			return true;
 	}
 
- out_rcu:
-	read_unlock(&gpc->lock);
-	return ret;
+	return false;
 }
 
 static bool kvm_xen_schedop_poll(struct kvm_vcpu *vcpu, bool longmode,
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 15/20] KVM: x86/xen: Convert kvm_xen_set_evtchn_fast() to gpc's CLASS() APIs

[Sean Christopherson]

Convert the event channel fastpath to the "map local" CLASS() APIs, using
the "try" variants as the faspath can't block.

Note!  The vcpu_info mapping is read/write, even though there is no
existing call to mark the page dirty.  Like Xen's shared info page, the
vCPU info page is assumed to be dirty at all times, and so isn't marked
dirty after every write.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 42 ++++++++++++++++--------------------------
 1 file changed, 16 insertions(+), 26 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 8f822acb11a4..47750316f132 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -1774,29 +1774,28 @@ static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
 	bool kick_vcpu = false;
 
 	/* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
-	if (!read_trylock(&gpc->lock)) {
-		/*
-		 * Could not access the vcpu_info. Set the bit in-kernel and
-		 * prod the vCPU to deliver it for itself.
-		 */
+	CLASS(gpc_try_map_local, vcpu_info_map)(gpc, sizeof(struct vcpu_info));
+
+	/*
+	 * If the vcpu_info is inaccessible, set the bit in-kernel and prod the
+	 * vCPU to deliver it for itself.
+	 */
+	if (IS_ERR(vcpu_info_map)) {
 		if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
 			kick_vcpu = true;
 		goto out_kick;
 	}
-	if (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-		if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
-			kick_vcpu = true;
-		goto out_unlock;
-	}
 
 	if (IS_ENABLED(CONFIG_64BIT) && vcpu->kvm->arch.xen.long_mode) {
-		struct vcpu_info *vcpu_info = gpc->khva;
+		struct vcpu_info *vcpu_info = *vcpu_info_map;
+
 		if (!test_and_set_bit(port_word_bit, &vcpu_info->evtchn_pending_sel)) {
 			WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
 			kick_vcpu = true;
 		}
 	} else {
-		struct compat_vcpu_info *vcpu_info = gpc->khva;
+		struct compat_vcpu_info *vcpu_info = *vcpu_info_map;
+
 		if (!test_and_set_bit(port_word_bit,
 					(unsigned long *)&vcpu_info->evtchn_pending_sel)) {
 			WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
@@ -1810,8 +1809,6 @@ static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
 		kick_vcpu = false;
 	}
 
-out_unlock:
-	read_unlock(&gpc->lock);
 out_kick:
 	if (kick_vcpu) {
 		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
@@ -1850,23 +1847,19 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 	if (xe->port >= max_evtchn_port(kvm))
 		return -EINVAL;
 
-	rc = -EWOULDBLOCK;
-
 	guard(srcu)(&kvm->srcu);
 
-	if (!read_trylock(&gpc->lock))
-		return rc;
-
-	if (!kvm_gpc_check(gpc, PAGE_SIZE))
-		goto out_unlock;
+	CLASS(gpc_try_map_local, shinfo_map)(gpc, PAGE_SIZE);
+	if (IS_ERR(shinfo_map))
+		return -EWOULDBLOCK;
 
 	if (IS_ENABLED(CONFIG_64BIT) && kvm->arch.xen.long_mode) {
-		struct shared_info *shinfo = gpc->khva;
+		struct shared_info *shinfo = *shinfo_map;
 		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
 		mask_bits = (unsigned long *)&shinfo->evtchn_mask;
 		port_word_bit = xe->port / 64;
 	} else {
-		struct compat_shared_info *shinfo = gpc->khva;
+		struct compat_shared_info *shinfo = *shinfo_map;
 		pending_bits = (unsigned long *)&shinfo->evtchn_pending;
 		mask_bits = (unsigned long *)&shinfo->evtchn_mask;
 		port_word_bit = xe->port / 32;
@@ -1888,9 +1881,6 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
 		rc = 1; /* Delivered to the bitmap in shared_info. */
 	}
 
-out_unlock:
-	read_unlock(&gpc->lock);
-
 	if (rc == 1)
 		__kvm_xen_set_evtchn_fast(vcpu, port_word_bit);
 	return rc;
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 16/20] KVM: x86/xen: Convert xen_get_guest_pvclock() to gpc's CLASS() APIs

[Sean Christopherson]

Convert the Xen pvclock reads to the "map local" CLASS() APIs, using a
read-only variants as KVM simply copying data from the pvclock.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 18 ++++--------------
 1 file changed, 4 insertions(+), 14 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 47750316f132..89daad3fe712 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -153,21 +153,11 @@ static int xen_get_guest_pvclock(struct kvm_vcpu *vcpu,
 				 struct gfn_to_pfn_cache *gpc,
 				 unsigned int offset)
 {
-	int r;
+	CLASS(gpc_map_local_ro, pvclock_map)(gpc, offset + sizeof(*hv_clock));
+	if (IS_ERR(pvclock_map))
+		return PTR_ERR(pvclock_map);
 
-	read_lock(&gpc->lock);
-	while (!kvm_gpc_check(gpc, offset + sizeof(*hv_clock))) {
-		read_unlock(&gpc->lock);
-
-		r = kvm_gpc_refresh(gpc, offset + sizeof(*hv_clock));
-		if (r)
-			return r;
-
-		read_lock(&gpc->lock);
-	}
-
-	memcpy(hv_clock, gpc->khva + offset, sizeof(*hv_clock));
-	read_unlock(&gpc->lock);
+	memcpy(hv_clock, *pvclock_map + offset, sizeof(*hv_clock));
 
 	/*
 	 * Sanity check TSC shift+multiplier to verify the guest's view of time
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 17/20] KVM: x86/xen: Drop local "kick_vcpu" from __kvm_xen_set_evtchn_fast()

[Sean Christopherson]

Now that the CLASS()-based gpc mappings allow for early returns, drop the
local "kick_vcpu" from the event channel fastpath, and simply return early
if the vCPU doesn't need to be kicked.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 33 +++++++++++++++------------------
 1 file changed, 15 insertions(+), 18 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 89daad3fe712..2c776e475a4f 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -1761,7 +1761,6 @@ static void kvm_xen_check_poller(struct kvm_vcpu *vcpu, int port)
 static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
 {
 	struct gfn_to_pfn_cache *gpc = &vcpu->arch.xen.vcpu_info_cache;
-	bool kick_vcpu = false;
 
 	/* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
 	CLASS(gpc_try_map_local, vcpu_info_map)(gpc, sizeof(struct vcpu_info));
@@ -1771,39 +1770,37 @@ static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
 	 * vCPU to deliver it for itself.
 	 */
 	if (IS_ERR(vcpu_info_map)) {
-		if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
-			kick_vcpu = true;
+		if (test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))
+			return;
 		goto out_kick;
 	}
 
 	if (IS_ENABLED(CONFIG_64BIT) && vcpu->kvm->arch.xen.long_mode) {
 		struct vcpu_info *vcpu_info = *vcpu_info_map;
 
-		if (!test_and_set_bit(port_word_bit, &vcpu_info->evtchn_pending_sel)) {
-			WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
-			kick_vcpu = true;
-		}
+		if (test_and_set_bit(port_word_bit, &vcpu_info->evtchn_pending_sel))
+			return;
+
+		WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
 	} else {
 		struct compat_vcpu_info *vcpu_info = *vcpu_info_map;
 
-		if (!test_and_set_bit(port_word_bit,
-					(unsigned long *)&vcpu_info->evtchn_pending_sel)) {
-			WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
-			kick_vcpu = true;
-		}
+		if (test_and_set_bit(port_word_bit,
+				     (unsigned long *)&vcpu_info->evtchn_pending_sel))
+			return;
+
+		WRITE_ONCE(vcpu_info->evtchn_upcall_pending, 1);
 	}
 
 	/* For the per-vCPU lapic vector, deliver it as MSI. */
-	if (kick_vcpu && vcpu->arch.xen.upcall_vector) {
+	if (vcpu->arch.xen.upcall_vector) {
 		kvm_xen_inject_vcpu_vector(vcpu);
-		kick_vcpu = false;
+		return;
 	}
 
 out_kick:
-	if (kick_vcpu) {
-		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
-		kvm_vcpu_kick(vcpu);
-	}
+	kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
+	kvm_vcpu_kick(vcpu);
 }
 
 /*
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 18/20] KVM: x86/xen: Convert event injection to gpc's CLASS() APIs

[Sean Christopherson]

Convert Xen event injection to the "map local" CLASS() APIs.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c | 24 +++++-------------------
 1 file changed, 5 insertions(+), 19 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 2c776e475a4f..3ebde7ba5558 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -628,24 +628,12 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 	if (!evtchn_pending_sel)
 		return;
 
-	/*
-	 * Yes, this is an open-coded loop. But that's just what put_user()
-	 * does anyway. Page it in and retry the instruction. We're just a
-	 * little more honest about it.
-	 */
-	read_lock(&gpc->lock);
-	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-		read_unlock(&gpc->lock);
+	CLASS(gpc_map_local, vcpu_info_map)(gpc, sizeof(struct vcpu_info));
+	if (IS_ERR(vcpu_info_map))
+		return;
 
-		if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info)))
-			return;
-
-		read_lock(&gpc->lock);
-	}
-
-	/* Now gpc->khva is a valid kernel address for the vcpu_info */
 	if (IS_ENABLED(CONFIG_64BIT) && v->kvm->arch.xen.long_mode) {
-		struct vcpu_info *vi = gpc->khva;
+		struct vcpu_info *vi = *vcpu_info_map;
 
 		asm volatile(LOCK_PREFIX "orq %0, %1\n"
 			     "notq %0\n"
@@ -657,7 +645,7 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 		WRITE_ONCE(vi->evtchn_upcall_pending, 1);
 	} else {
 		u32 evtchn_pending_sel32 = evtchn_pending_sel;
-		struct compat_vcpu_info *vi = gpc->khva;
+		struct compat_vcpu_info *vi = *vcpu_info_map;
 
 		asm volatile(LOCK_PREFIX "orl %0, %1\n"
 			     "notl %0\n"
@@ -669,8 +657,6 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 		WRITE_ONCE(vi->evtchn_upcall_pending, 1);
 	}
 
-	read_unlock(&gpc->lock);
-
 	/* For the per-vCPU lapic vector, deliver it as MSI. */
 	if (v->arch.xen.upcall_vector)
 		kvm_xen_inject_vcpu_vector(v);
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 19/20] KVM: Add "extended" gpc CLASS() APIs for sometimes-atomic cases

[Sean Christopherson]

Extend the CLASS() APIs for gpcs to allow choosing between the "normal" and
the "try" versions at runtime, depending on whether or not the caller is
running in atomic context.  Convert the "has interrupt" helper as the first
user, as it is called from IRQ context, but also needs to wait when called
from non-atomic context, i.e. can't tolerate false negatives in that case.

No functional change intended.

Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/kvm/xen.c       | 52 +++++++++++++---------------------------
 include/linux/kvm_host.h | 10 ++++++++
 2 files changed, 26 insertions(+), 36 deletions(-)

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 3ebde7ba5558..a2e88a76e8d9 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -666,7 +666,6 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 {
 	struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
 	bool atomic = in_atomic() || !task_is_running(current);
-	u8 rc = 0;
 
 	/*
 	 * If the global upcall vector (HVMIRQ_callback_vector) is set and
@@ -676,44 +675,25 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 	/* No need for compat handling here */
 	BUILD_BUG_ON(offsetof(struct vcpu_info, evtchn_upcall_pending) !=
 		     offsetof(struct compat_vcpu_info, evtchn_upcall_pending));
-	BUILD_BUG_ON(sizeof(rc) !=
-		     sizeof_field(struct vcpu_info, evtchn_upcall_pending));
-	BUILD_BUG_ON(sizeof(rc) !=
+	BUILD_BUG_ON(sizeof_field(struct vcpu_info, evtchn_upcall_pending) !=
 		     sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
 
-	if (atomic) {
-		if (!read_trylock(&gpc->lock))
-			return 1;
-	} else {
-		read_lock(&gpc->lock);
-	}
-	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-		read_unlock(&gpc->lock);
+	/*
+	 * This function gets called from kvm_vcpu_block() after setting the
+	 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
+	 * from a HLT. So we really mustn't sleep. If the page ended up absent
+	 * at that point, just return 1 in order to trigger an immediate wake,
+	 * and we'll end up getting called again from a context where we *can*
+	 * fault in the page and wait for it.
+	 *
+	 * For normal, non-atomic usage, nothing can be done if userspace has
+	 * screwed up the vcpu_info mapping. No interrupts for you.
+	 */
+	CLASS(gpc_map_local_ro_ex, info_map)(gpc, sizeof(struct vcpu_info), atomic);
+	if (IS_ERR(info_map))
+		return atomic ? 1 : 0;
 
-		/*
-		 * This function gets called from kvm_vcpu_block() after setting the
-		 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
-		 * from a HLT. So we really mustn't sleep. If the page ended up absent
-		 * at that point, just return 1 in order to trigger an immediate wake,
-		 * and we'll end up getting called again from a context where we *can*
-		 * fault in the page and wait for it.
-		 */
-		if (atomic)
-			return 1;
-
-		if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info))) {
-			/*
-			 * If this failed, userspace has screwed up the
-			 * vcpu_info mapping. No interrupts for you.
-			 */
-			return 0;
-		}
-		read_lock(&gpc->lock);
-	}
-
-	rc = ((struct vcpu_info *)gpc->khva)->evtchn_upcall_pending;
-	read_unlock(&gpc->lock);
-	return rc;
+	return ((struct vcpu_info *)*info_map)->evtchn_upcall_pending;
 }
 
 int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index d70fa91cda0c..0602d0ca731c 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -1573,6 +1573,16 @@ DEFINE_GPC_CLASS(_try,);
 DEFINE_GPC_CLASS(, _ro);
 DEFINE_GPC_CLASS(_try, _ro);
 
+#define DEFINE_GPC_CLASS_EX(ro)							\
+DEFINE_CLASS(gpc_map_local##ro##_ex, void **,					\
+	     if (!IS_ERR(_T)) gpc_map_local_unlock##ro(_T),			\
+	     atomic ? gpc_try_map_local_lock(gpc, len) :			\
+		      gpc_map_local_lock(gpc, len),				\
+	     struct gfn_to_pfn_cache *gpc, unsigned long len, bool atomic)
+
+DEFINE_GPC_CLASS_EX();
+DEFINE_GPC_CLASS_EX(_ro);
+
 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
 
-- 
2.54.0.823.g6e5bcc1fc9-goog

[PATCH v2 20/20] KVM: x86: Use gfn_to_pfn_cache for record_steal_time

[Sean Christopherson]

From: Carsten Stollmaier <stollmc@amazon.com>

This largely reverts commit 7e2175ebd695 ("KVM: x86: Fix recording of
guest steal time / preempted status"), which dropped the use of the
gfn_to_pfn_cache because it was not integrated with the MMU notifiers
at the time. That shortcoming has long since been addressed, making
the GPC work correctly for this use case.

Aside from cleaning up the last	open-coded assembler access to user
addresses and associated explicit asm exception fixups, moving back
to the now-functional GPC also resolves an issue with contention on
the mmap_lock with userfaultfd. The contention issue is as follows:

On vcpu_run, before entering the guest, the update of the steal time
information causes a page-fault if the page is not present. In our
scenario, this gets handled by do_user_addr_fault() and successively
handle_userfault() because the region is registered to that.

Since handle_userfault() uses TASK_INTERRUPTIBLE, it is interruptible
by signals. But do_user_addr_fault() then busy-retries if the pending
signal is non-fatal, which leads to heavy contention of the mmap_lock.

By restoring the use of GPC for accessing the guest steal time, the
contention is avoided and refreshing the GPC happens when the vCPU is
next scheduled.

Since the gfn_to_pfn_cache gives a kernel mapping rather than a
userspace HVA, accesses are now plain C instead of unsafe_put_user()
et al. Use READ_ONCE()/WRITE_ONCE() to prevent the compiler from
reordering or tearing the accesses, and add an smp_wmb() before the
final version increment to ensure the data writes are ordered before
the seqcount update — the old unsafe_put_user() inline assembly acted
as an implicit compiler barrier.

In kvm_steal_time_set_preempted(), use read_trylock() instead of
read_lock_irqsave() since this is called from the scheduler path
where rwlock_t is not safe on PREEMPT_RT (it becomes sleepable).
Since we only trylock and bail on failure, there is no risk of
deadlock with an interrupt handler, so no need to disable interrupts
at all. Setting the preempted flag is best-effort anyway.

Signed-off-by: Carsten Stollmaier <stollmc@amazon.com>
Co-developed-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/include/asm/kvm_host.h |   2 +-
 arch/x86/kvm/x86.c              | 122 ++++++++++++++------------------
 2 files changed, 54 insertions(+), 70 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 6ae7d539af90..9f652dcdda93 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -983,7 +983,7 @@ struct kvm_vcpu_arch {
 		u8 preempted;
 		u64 msr_val;
 		u64 last_steal;
-		struct gfn_to_hva_cache cache;
+		struct gfn_to_pfn_cache cache;
 	} st;
 
 	u64 l1_tsc_offset;
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index ea10ed4ab06f..1b27dd9ba0aa 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3728,10 +3728,8 @@ EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_service_local_tlb_flush_requests);
 
 static void record_steal_time(struct kvm_vcpu *vcpu)
 {
-	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
-	struct kvm_steal_time __user *st;
-	struct kvm_memslots *slots;
-	gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
+	struct gfn_to_pfn_cache *gpc = &vcpu->arch.st.cache;
+	struct kvm_steal_time *st;
 	u64 steal;
 	u32 version;
 
@@ -3746,42 +3744,20 @@ static void record_steal_time(struct kvm_vcpu *vcpu)
 	if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
 		return;
 
-	slots = kvm_memslots(vcpu->kvm);
+	/* We rely on the fact that it fits in a single page. */
+	BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
 
-	if (unlikely(slots->generation != ghc->generation ||
-		     gpa != ghc->gpa ||
-		     kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
-		/* We rely on the fact that it fits in a single page. */
-		BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
+	CLASS(gpc_map_local, st_map)(gpc, sizeof(*st));
+	if (IS_ERR(st_map))
+		return;
 
-		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st)) ||
-		    kvm_is_error_hva(ghc->hva) || !ghc->memslot)
-			return;
-	}
-
-	st = (struct kvm_steal_time __user *)ghc->hva;
+	st = *st_map;
 	/*
 	 * Doing a TLB flush here, on the guest's behalf, can avoid
 	 * expensive IPIs.
 	 */
 	if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
-		u8 st_preempted = 0;
-		int err = -EFAULT;
-
-		if (!user_access_begin(st, sizeof(*st)))
-			return;
-
-		asm volatile("1: xchgb %0, %2\n"
-			     "xor %1, %1\n"
-			     "2:\n"
-			     _ASM_EXTABLE_UA(1b, 2b)
-			     : "+q" (st_preempted),
-			       "+&r" (err),
-			       "+m" (st->preempted));
-		if (err)
-			goto out;
-
-		user_access_end();
+		u8 st_preempted = xchg(&st->preempted, 0);
 
 		vcpu->arch.st.preempted = 0;
 
@@ -3789,39 +3765,30 @@ static void record_steal_time(struct kvm_vcpu *vcpu)
 				       st_preempted & KVM_VCPU_FLUSH_TLB);
 		if (st_preempted & KVM_VCPU_FLUSH_TLB)
 			kvm_vcpu_flush_tlb_guest(vcpu);
-
-		if (!user_access_begin(st, sizeof(*st)))
-			goto dirty;
 	} else {
-		if (!user_access_begin(st, sizeof(*st)))
-			return;
-
-		unsafe_put_user(0, &st->preempted, out);
+		WRITE_ONCE(st->preempted, 0);
 		vcpu->arch.st.preempted = 0;
 	}
 
-	unsafe_get_user(version, &st->version, out);
+	version = READ_ONCE(st->version);
 	if (version & 1)
 		version += 1;  /* first time write, random junk */
 
 	version += 1;
-	unsafe_put_user(version, &st->version, out);
+	WRITE_ONCE(st->version, version);
 
 	smp_wmb();
 
-	unsafe_get_user(steal, &st->steal, out);
+	steal = READ_ONCE(st->steal);
 	steal += current->sched_info.run_delay -
 		vcpu->arch.st.last_steal;
 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
-	unsafe_put_user(steal, &st->steal, out);
+	WRITE_ONCE(st->steal, steal);
+
+	smp_wmb();
 
 	version += 1;
-	unsafe_put_user(version, &st->version, out);
-
- out:
-	user_access_end();
- dirty:
-	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
+	WRITE_ONCE(st->version, version);
 }
 
 /*
@@ -4162,8 +4129,11 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 
 		vcpu->arch.st.msr_val = data;
 
-		if (!(data & KVM_MSR_ENABLED))
-			break;
+		if (data & KVM_MSR_ENABLED)
+			kvm_gpc_activate(&vcpu->arch.st.cache, data & ~KVM_MSR_ENABLED,
+					sizeof(struct kvm_steal_time));
+		else
+			kvm_gpc_deactivate(&vcpu->arch.st.cache);
 
 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
 
@@ -5231,11 +5201,8 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 
 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
 {
-	struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
-	struct kvm_steal_time __user *st;
-	struct kvm_memslots *slots;
-	static const u8 preempted = KVM_VCPU_PREEMPTED;
-	gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
+	struct gfn_to_pfn_cache *gpc = &vcpu->arch.st.cache;
+	struct kvm_steal_time *st;
 
 	/*
 	 * The vCPU can be marked preempted if and only if the VM-Exit was on
@@ -5260,20 +5227,32 @@ static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
 	if (unlikely(current->mm != vcpu->kvm->mm))
 		return;
 
-	slots = kvm_memslots(vcpu->kvm);
-
-	if (unlikely(slots->generation != ghc->generation ||
-		     gpa != ghc->gpa ||
-		     kvm_is_error_hva(ghc->hva) || !ghc->memslot))
+	/*
+	 * Use a trylock as this is called from the scheduler path (via
+	 * kvm_sched_out), where rwlock_t is not safe on PREEMPT_RT (it
+	 * becomes sleepable).  Setting preempted is best-effort anyway;
+	 * the old HVA-based code used copy_to_user_nofault() which could
+	 * also silently fail.
+	 *
+	 * Since we only trylock and bail on failure, there is no risk of
+	 * deadlock with an interrupt handler, so no need to disable
+	 * interrupts.
+	 */
+	CLASS(gpc_try_map_local, st_map)(gpc, sizeof(st->preempted));
+	if (IS_ERR(st_map))
 		return;
 
-	st = (struct kvm_steal_time __user *)ghc->hva;
-	BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
+	st = *st_map;
+	WRITE_ONCE(st->preempted, KVM_VCPU_PREEMPTED);
+	vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
+}
 
-	if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
-		vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
-
-	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
+static void kvm_steal_time_reset(struct kvm_vcpu *vcpu)
+{
+	kvm_gpc_deactivate(&vcpu->arch.st.cache);
+	vcpu->arch.st.preempted = 0;
+	vcpu->arch.st.msr_val = 0;
+	vcpu->arch.st.last_steal = 0;
 }
 
 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
@@ -12819,6 +12798,8 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
 
 	kvm_gpc_init(&vcpu->arch.pv_time, vcpu->kvm);
 
+	kvm_gpc_init(&vcpu->arch.st.cache, vcpu->kvm);
+
 	if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
 		kvm_set_mp_state(vcpu, KVM_MP_STATE_RUNNABLE);
 	else
@@ -12926,6 +12907,8 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 	kvm_clear_async_pf_completion_queue(vcpu);
 	kvm_mmu_unload(vcpu);
 
+	kvm_steal_time_reset(vcpu);
+
 	kvmclock_reset(vcpu);
 
 	for_each_possible_cpu(cpu)
@@ -13046,7 +13029,8 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
 	kvm_make_request(KVM_REQ_EVENT, vcpu);
 	vcpu->arch.apf.msr_en_val = 0;
 	vcpu->arch.apf.msr_int_val = 0;
-	vcpu->arch.st.msr_val = 0;
+
+	kvm_steal_time_reset(vcpu);
 
 	kvmclock_reset(vcpu);
 
-- 
2.54.0.823.g6e5bcc1fc9-goog

Re: [PATCH v2 02/20] KVM: x86/xen: Use read_trylock() for GPC locks in hardirq/atomic paths

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Unbalanced interrupt disable causes permanent hard IRQ disable on PREEMPT_RT
--

commit 13e357726c8cbadaf777f9052a529db65b824338
Author: David Woodhouse <dwmw@amazon.co.uk>
KVM: x86/xen: Use read_trylock() for GPC locks in hardirq/atomic paths

This patch modifies the GPC lock acquisition paths to use read_trylock()
instead of read_lock_irqsave() in atomic contexts. It introduces an explicit
local_irq_save() to prevent issues on PREEMPT_RT kernels where rwlock_t
cannot be used with read_lock_irqsave() from hardirq context.

Reported-by: syzbot+208f7f3e5f59c11aeb90@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=208f7f3e5f59c11aeb90
Fixes: 14243b387137 ("KVM: x86/xen: Add KVM_IRQ_ROUTING_XEN_EVTCHN and event channel delivery")

> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> index 91fd3673c09a2..9bdb8e3cad58b 100644
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c
> @@ -713,7 +714,15 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
>  	BUILD_BUG_ON(sizeof(rc) !=
>  		     sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
>  
> -	read_lock_irqsave(&gpc->lock, flags);
> +	if (atomic) {
> +		local_irq_save(flags);
> +		if (!read_trylock(&gpc->lock)) {
> +			local_irq_restore(flags);
> +			return 1;
> +		}
> +	} else {
> +		read_lock_irqsave(&gpc->lock, flags);
> +	}
>  	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
>  		read_unlock_irqrestore(&gpc->lock, flags);

[Severity: High]
Does this result in unbalanced interrupt disables on PREEMPT_RT?

When atomic is true, local_irq_save() explicitly disables hardware interrupts.
On standard kernels, read_unlock_irqrestore() restores the interrupt state
correctly.

However, on PREEMPT_RT, rwlock_t is a sleepable lock and the implementation
of read_unlock_irqrestore() ignores the flags parameter, merely dropping the
lock without re-enabling interrupts. The interrupt state saved by
local_irq_save() is never restored, which would lead to hardware interrupts
being permanently disabled on the CPU.

I noticed this issue is resolved later in the series by commit 625a2fc07daa6
("KVM: x86/xen: Remove unnecessary irqsave from GPC lock usage in xen.c").

Should this fix be folded into this patch to ensure the kernel remains fully
functional and bisectable at this step?

[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260529165114.748639-1-seanjc@google.com?part=2

Re: [PATCH v2 03/20] KVM: x86/xen: Remove unnecessary irqsave from GPC lock usage in xen.c

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Calling read_unlock() on an rwlock_t inside kvm_sched_out() can cause deadlocks on PREEMPT_RT because waking a waiting writer requires acquiring runqueue locks while the current runqueue lock is already held.
--

commit 625a2fc07daa67a8fcafaa64daaf9791f5012ecd
Author: David Woodhouse <dwmw@amazon.co.uk>
Subject: KVM: x86/xen: Remove unnecessary irqsave from GPC lock usage in xen.c

This commit converts read_lock_irqsave() and read_unlock_irqrestore() 
calls to plain read_lock() and read_unlock() for gpc->lock in several 
xen.c functions, as the hardirq path now safely uses read_trylock().

> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c

[ ... ]

> @@ -421,16 +419,14 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
>  	 * gfn_to_pfn caches that cover the region.
>  	 */
>  	if (atomic) {
> -		local_irq_save(flags);
>  		if (!read_trylock(&gpc1->lock)) {
> -			local_irq_restore(flags);
>  			return;
>  		}
>  	} else {
> -		read_lock_irqsave(&gpc1->lock, flags);
> +		read_lock(&gpc1->lock);
>  	}
>  	while (!kvm_gpc_check(gpc1, user_len1)) {
> -		read_unlock_irqrestore(&gpc1->lock, flags);
> +		read_unlock(&gpc1->lock);

[Severity: High]
This is a pre-existing issue, but can calling read_unlock() here lead to a
deadlock on PREEMPT_RT?

kvm_xen_update_runstate_guest() is called with atomic=true from
kvm_sched_out(), which is invoked via preempt notifiers while holding the
runqueue lock for the current CPU.

On CONFIG_PREEMPT_RT, rwlock_t is mapped to an rt_rwlock. If a writer is
waiting (for example, from a concurrent kvm_gpc_deactivate()),
read_unlock() calls __rwbase_read_unlock() to wake the blocked writer.

__rwbase_read_unlock() observes the waiting writer and calls
rt_mutex_wake_up_q(), which synchronously calls try_to_wake_up().

try_to_wake_up() will then attempt to acquire the pi_lock and the target
task's rq_lock. Since the current CPU's runqueue lock is already held by
the scheduler, acquiring another rq_lock without double_rq_lock() violates
lock ordering and can cause an ABBA deadlock.

Is there a plan to handle this rt_rwlock wake-up behavior from the preempt
notifier context?

>  
>  		/* When invoked from kvm_sched_out() we cannot sleep */
>  		if (atomic)
> @@ -439,7 +435,7 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)

[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260529165114.748639-1-seanjc@google.com?part=3

Re: [PATCH v2 07/20] KVM: x86/xen: Extract delivery of event to vCPU into a separate helper

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [Critical] Lockless reads of `kvm->arch.xen.long_mode` across multiple checks allow TOCTOU races, leading to out-of-bounds memory corruption in both host kernel state and guest state.
--

commit f4fd761232e8e0f394f8395cfeec34caad9f061d
Author: Sean Christopherson <seanjc@google.com>

KVM: x86/xen: Extract delivery of event to vCPU into a separate helper
    
This commit extracts the fastpath delivery of an event to a vCPU into a
separate helper function to allow CLASS()-based gpc locking and checking.

> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> index 0c6b74b974080..020ef0ddab013 100644
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c
[ ... ]
> @@ -1775,6 +1775,57 @@ static void kvm_xen_check_poller(struct kvm_vcpu *vcpu, int port)
>  	}
>  }
>  
> +static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
> +{
> +	struct gfn_to_pfn_cache *gpc = &vcpu->arch.xen.vcpu_info_cache;
> +	bool kick_vcpu = false;
> +
> +	/* Now switch to the vCPU's vcpu_info to set the index and pending_sel */
> +	if (!read_trylock(&gpc->lock)) {
> +		/*
> +		 * Could not access the vcpu_info. Set the bit in-kernel and
> +		 * prod the vCPU to deliver it for itself.
> +		 */
> +		if (!test_and_set_bit(port_word_bit, &vcpu->arch.xen.evtchn_pending_sel))

[Severity: Critical]
This is a pre-existing issue, but can the lockless evaluation of long_mode
lead to out-of-bounds memory corruption here?

If userspace races KVM_XEN_HVM_SET_ATTR (setting long_mode = 0) with event
channel delivery (e.g., via KVM_XEN_HVM_EVTCHN_SEND), could the initial
max_evtchn_port() check observe long_mode == 1, permitting xe->port up to
4095?

If the subsequent check inside kvm_xen_set_evtchn_fast() observes
long_mode == 0, it falls into the 32-bit compat block, computing
port_word_bit = xe->port / 32, which could yield a bit index up to 127.

When this out-of-bounds port_word_bit is passed to test_and_set_bit(), it
would write past the 8-byte evtchn_pending_sel field, potentially corrupting
the adjacent vcpu_id and timer_virq fields in struct kvm_vcpu_xen.

> +			kick_vcpu = true;
> +		goto out_kick;
> +	}
[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260529165114.748639-1-seanjc@google.com?part=7

Re: [PATCH v2 11/20] KVM: Add CLASS() constructs to automagically handle lock+check of gpc

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] The cleanup wrapper defined by `DEFINE_GPC_CLASS` uses `if (!IS_ERR(_T))` instead of `if (!IS_ERR_OR_NULL(_T))`, failing to guard against `NULL` pointers. This leads to a guaranteed kernel crash if ownership of the class variable is ever transferred or if it is initialized to `NULL`.
--

commit 88a61c5d744692bfbffaf3ccd332184029a7ffee
Author: Sean Christopherson <seanjc@google.com>

KVM: Add CLASS() constructs to automagically handle lock+check of gpc

This commit adds CLASS() definitions for mapping a PFN given a gfn_to_pfn
cache (gpc), deduplicating lock, check, and unlock loops. The pvclock code
is converted as the first user of this new construct.

> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h

[ ... ]

> +static inline void gpc_map_local_unlock(void **khva)
> +{
> +	struct gfn_to_pfn_cache *gpc = container_of(khva, struct gfn_to_pfn_cache, khva);
> +
> +	kvm_gpc_mark_dirty_in_slot(gpc);
> +
> +	read_unlock(&gpc->lock);
> +}
> +
> +static inline void gpc_map_local_unlock_ro(void **khva)
> +{
> +	read_unlock(&container_of(khva, struct gfn_to_pfn_cache, khva)->lock);
> +}
> +
> +#define DEFINE_GPC_CLASS(try, ro)						\
> +DEFINE_CLASS(gpc##try##_map_local##ro, void **,					\
> +	     if (!IS_ERR(_T)) gpc_map_local_unlock##ro(_T),			\

[Severity: High]
Does this guard expression properly handle NULL values, or could it lead to a
regression?

Since IS_ERR(NULL) evaluates to false, !IS_ERR(NULL) evaluates to true. If a
user of this class manually sets the variable to NULL or uses a primitive
like no_free_ptr() to inhibit cleanup, the destructor will run and
erroneously execute gpc_map_local_unlock(NULL).

Inside gpc_map_local_unlock(), container_of(NULL, ...) will produce a large
negative pointer. The subsequent calls to kvm_gpc_mark_dirty_in_slot() and
read_unlock() will dereference this invalid pointer and cause a regression.

Would it be safer to use if (!IS_ERR_OR_NULL(_T)) instead?

> +	     gpc##try##_map_local_lock(gpc, len),				\
> +	     struct gfn_to_pfn_cache *gpc, unsigned long len)			\
> +
> +DEFINE_GPC_CLASS(,);
> +DEFINE_GPC_CLASS(_try,);
> +DEFINE_GPC_CLASS(, _ro);
> +DEFINE_GPC_CLASS(_try, _ro);

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260529165114.748639-1-seanjc@google.com?part=11

Re: [PATCH v2 15/20] KVM: x86/xen: Convert kvm_xen_set_evtchn_fast() to gpc's CLASS() APIs

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Scope-based `CLASS()` conversion unintentionally extends the duration of the `gpc->lock` read locks across expensive cross-vCPU IPIs (`kvm_vcpu_kick`) and nested lock acquisitions.
--

commit f52c566cf9e4ecf5b818d43ac788f5009d85b7bc
Author: Sean Christopherson <seanjc@google.com>

KVM: x86/xen: Convert kvm_xen_set_evtchn_fast() to gpc's CLASS() APIs
    
Convert the event channel fastpath to the "map local" CLASS() APIs, using
the "try" variants as the faspath can't block.

> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> index 8f822acb11a46..47750316f1322 100644
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c
[ ... ]
> @@ -1810,8 +1809,6 @@ static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
>  		kick_vcpu = false;
>  	}
>  
> -out_unlock:
> -	read_unlock(&gpc->lock);
>  out_kick:
>  	if (kick_vcpu) {
>  		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);

[Severity: High]
Does this unintentionally hold the gpc lock across the IPI kick?

By replacing the explicit unlock with a function-scoped CLASS() variable, the
vcpu_info_map destructor won't run until the end of the function. This means
the lock is now held while calling kvm_vcpu_kick(), whereas the original
code explicitly dropped it first.

[ ... ]
> @@ -1888,9 +1881,6 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
>  		rc = 1; /* Delivered to the bitmap in shared_info. */
>  	}
>  
> -out_unlock:
> -	read_unlock(&gpc->lock);
> -
>  	if (rc == 1)
>  		__kvm_xen_set_evtchn_fast(vcpu, port_word_bit);
>  	return rc;

[Severity: High]
Does this now hold the VM-wide shinfo_cache lock while calling
__kvm_xen_set_evtchn_fast() and kicking the vCPU?

Since shinfo_map is a function-scoped CLASS() variable, its destructor
won't release the lock until after __kvm_xen_set_evtchn_fast() returns.
This creates a nested locking dependency and holds locks over expensive
cross-vCPU operations, potentially serializing event channel deliveries
across the entire VM on the fast path.

Would it make sense to constrain the mapping to an explicit scope block
to match the original release points and maintain the "No functional change
intended" guarantee?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260529165114.748639-1-seanjc@google.com?part=15

Re: [PATCH v2 15/20] KVM: x86/xen: Convert kvm_xen_set_evtchn_fast() to gpc's CLASS() APIs

[Sean Christopherson]

On Fri, May 29, 2026, sashiko-bot@kernel.org wrote:
> > diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> > index 8f822acb11a46..47750316f1322 100644
> > --- a/arch/x86/kvm/xen.c
> > +++ b/arch/x86/kvm/xen.c
> [ ... ]
> > @@ -1810,8 +1809,6 @@ static void __kvm_xen_set_evtchn_fast(struct kvm_vcpu *vcpu, int port_word_bit)
> >  		kick_vcpu = false;
> >  	}
> >  
> > -out_unlock:
> > -	read_unlock(&gpc->lock);
> >  out_kick:
> >  	if (kick_vcpu) {
> >  		kvm_make_request(KVM_REQ_UNBLOCK, vcpu);
> 
> [Severity: High]
> Does this unintentionally hold the gpc lock across the IPI kick?

No, it intentionally holds the gpc lock across _sending_ the IPI kick.

> By replacing the explicit unlock with a function-scoped CLASS() variable, the
> vcpu_info_map destructor won't run until the end of the function. This means
> the lock is now held while calling kvm_vcpu_kick(), whereas the original
> code explicitly dropped it first.
> 
> [ ... ]
> > @@ -1888,9 +1881,6 @@ int kvm_xen_set_evtchn_fast(struct kvm_xen_evtchn *xe, struct kvm *kvm)
> >  		rc = 1; /* Delivered to the bitmap in shared_info. */
> >  	}
> >  
> > -out_unlock:
> > -	read_unlock(&gpc->lock);
> > -
> >  	if (rc == 1)
> >  		__kvm_xen_set_evtchn_fast(vcpu, port_word_bit);
> >  	return rc;
> 
> [Severity: High]
> Does this now hold the VM-wide shinfo_cache lock while calling
> __kvm_xen_set_evtchn_fast() and kicking the vCPU?
> 
> Since shinfo_map is a function-scoped CLASS() variable, its destructor
> won't release the lock until after __kvm_xen_set_evtchn_fast() returns.
> This creates a nested locking dependency and holds locks over expensive
> cross-vCPU operations, potentially serializing event channel deliveries
> across the entire VM on the fast path.

__kvm_vcpu_kick() is neither expensive nor cross-vCPU.  In the wait=false case,
which is the behavior of kvm_vcpu_kick(), it sends IPIs via smp_send_reschedule(),
i.e. it's more or less just __apic_send_IPI(cpu, RESCHEDULE_VECTOR), which is a
single WRMSR on modern harware.

> Would it make sense to constrain the mapping to an explicit scope block
> to match the original release points and maintain the "No functional change
> intended" guarantee?

I'll just drop that claim.

Re: [PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Peter Zijlstra]

On Fri, May 29, 2026 at 09:50:55AM -0700, Sean Christopherson wrote:
> From: David Woodhouse <dwmw@amazon.co.uk>
> 
> __rwbase_read_unlock() uses raw_spin_lock_irq()/raw_spin_unlock_irq()
> which unconditionally disables and re-enables interrupts. When
> read_unlock() is called from hardirq context (e.g. after a successful
> read_trylock() in a timer callback), the raw_spin_unlock_irq()
> incorrectly re-enables interrupts within the hardirq handler.
> 
> This causes lockdep warnings ('hardirqs_on_prepare' from hardirq
> context) and can lead to IRQ state corruption.
> 
> Using read_trylock() in hardirq context on PREEMPT_RT is safe because
> it does not record the lock owner. The read_unlock() acquires the
> wait_lock which is hardirq safe. This change additionally allows
> rwlock_t during early boot.
> 
> Switch to raw_spin_lock_irqsave()/raw_spin_unlock_irqrestore() to
> preserve the caller's IRQ state.
> 
> Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
> Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> Signed-off-by: Sean Christopherson <seanjc@google.com>

We have very specifically not supported the: trylock+unlock from hardirq
(although typically this comes up for mutex). Specifically with PI this
can lead to trying to boost the idle thread.

Consider doing this from an interrupt that hits idle, then idle becomes
the 'owner' of a successful acquisition. This is absolutely broken.

Re: [PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Peter Zijlstra]

On Fri, May 29, 2026 at 09:32:14PM +0200, Peter Zijlstra wrote:
> On Fri, May 29, 2026 at 09:50:55AM -0700, Sean Christopherson wrote:
> > From: David Woodhouse <dwmw@amazon.co.uk>
> > 
> > __rwbase_read_unlock() uses raw_spin_lock_irq()/raw_spin_unlock_irq()
> > which unconditionally disables and re-enables interrupts. When
> > read_unlock() is called from hardirq context (e.g. after a successful
> > read_trylock() in a timer callback), the raw_spin_unlock_irq()
> > incorrectly re-enables interrupts within the hardirq handler.
> > 
> > This causes lockdep warnings ('hardirqs_on_prepare' from hardirq
> > context) and can lead to IRQ state corruption.
> > 
> > Using read_trylock() in hardirq context on PREEMPT_RT is safe because
> > it does not record the lock owner. The read_unlock() acquires the
> > wait_lock which is hardirq safe. This change additionally allows
> > rwlock_t during early boot.

Forgot to reply to this; it is safe with this implementation. If we were
to ever do reader owner tracking this goes sideways real fast.

I really think this is a very bad approach.

> > Switch to raw_spin_lock_irqsave()/raw_spin_unlock_irqrestore() to
> > preserve the caller's IRQ state.
> > 
> > Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
> > Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> > Signed-off-by: Sean Christopherson <seanjc@google.com>
> 
> We have very specifically not supported the: trylock+unlock from hardirq
> (although typically this comes up for mutex). Specifically with PI this
> can lead to trying to boost the idle thread.
> 
> Consider doing this from an interrupt that hits idle, then idle becomes
> the 'owner' of a successful acquisition. This is absolutely broken.

Re: [PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Sean Christopherson]

On Fri, May 29, 2026, Peter Zijlstra wrote:
> On Fri, May 29, 2026 at 09:32:14PM +0200, Peter Zijlstra wrote:
> > On Fri, May 29, 2026 at 09:50:55AM -0700, Sean Christopherson wrote:
> > > From: David Woodhouse <dwmw@amazon.co.uk>
> > > 
> > > __rwbase_read_unlock() uses raw_spin_lock_irq()/raw_spin_unlock_irq()
> > > which unconditionally disables and re-enables interrupts. When
> > > read_unlock() is called from hardirq context (e.g. after a successful
> > > read_trylock() in a timer callback), the raw_spin_unlock_irq()
> > > incorrectly re-enables interrupts within the hardirq handler.
> > > 
> > > This causes lockdep warnings ('hardirqs_on_prepare' from hardirq
> > > context) and can lead to IRQ state corruption.
> > > 
> > > Using read_trylock() in hardirq context on PREEMPT_RT is safe because
> > > it does not record the lock owner. The read_unlock() acquires the
> > > wait_lock which is hardirq safe. This change additionally allows
> > > rwlock_t during early boot.
> 
> Forgot to reply to this; it is safe with this implementation. If we were
> to ever do reader owner tracking this goes sideways real fast.
> 
> I really think this is a very bad approach.
> 
> > > Switch to raw_spin_lock_irqsave()/raw_spin_unlock_irqrestore() to
> > > preserve the caller's IRQ state.
> > > 
> > > Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
> > > Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> > > Signed-off-by: Sean Christopherson <seanjc@google.com>
> > 
> > We have very specifically not supported the: trylock+unlock from hardirq
> > (although typically this comes up for mutex). Specifically with PI this
> > can lead to trying to boost the idle thread.
> > 
> > Consider doing this from an interrupt that hits idle, then idle becomes
> > the 'owner' of a successful acquisition. This is absolutely broken.

I assume the only alternative is to implement raw versions of rwlock?  Or do I
understand all of this even less than I thought? :-)

Re: [PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Peter Zijlstra]

On Fri, May 29, 2026 at 01:05:16PM -0700, Sean Christopherson wrote:
> On Fri, May 29, 2026, Peter Zijlstra wrote:
> > On Fri, May 29, 2026 at 09:32:14PM +0200, Peter Zijlstra wrote:
> > > On Fri, May 29, 2026 at 09:50:55AM -0700, Sean Christopherson wrote:
> > > > From: David Woodhouse <dwmw@amazon.co.uk>
> > > > 
> > > > __rwbase_read_unlock() uses raw_spin_lock_irq()/raw_spin_unlock_irq()
> > > > which unconditionally disables and re-enables interrupts. When
> > > > read_unlock() is called from hardirq context (e.g. after a successful
> > > > read_trylock() in a timer callback), the raw_spin_unlock_irq()
> > > > incorrectly re-enables interrupts within the hardirq handler.
> > > > 
> > > > This causes lockdep warnings ('hardirqs_on_prepare' from hardirq
> > > > context) and can lead to IRQ state corruption.
> > > > 
> > > > Using read_trylock() in hardirq context on PREEMPT_RT is safe because
> > > > it does not record the lock owner. The read_unlock() acquires the
> > > > wait_lock which is hardirq safe. This change additionally allows
> > > > rwlock_t during early boot.
> > 
> > Forgot to reply to this; it is safe with this implementation. If we were
> > to ever do reader owner tracking this goes sideways real fast.
> > 
> > I really think this is a very bad approach.
> > 
> > > > Switch to raw_spin_lock_irqsave()/raw_spin_unlock_irqrestore() to
> > > > preserve the caller's IRQ state.
> > > > 
> > > > Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
> > > > Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> > > > Signed-off-by: Sean Christopherson <seanjc@google.com>
> > > 
> > > We have very specifically not supported the: trylock+unlock from hardirq
> > > (although typically this comes up for mutex). Specifically with PI this
> > > can lead to trying to boost the idle thread.
> > > 
> > > Consider doing this from an interrupt that hits idle, then idle becomes
> > > the 'owner' of a successful acquisition. This is absolutely broken.
> 
> I assume the only alternative is to implement raw versions of rwlock?  Or do I
> understand all of this even less than I thought? :-)

Well, the code needs to do something when trylock fails, can't you
pretend it fails unconditionally?

It is somewhat possible to do an RT aware read-write spinlock thing, but
it is definitely non-trivial and this would be the only user.

Also, why does it have to be a rwlock? Can't this be RCU-ified in some
way?

Re: [PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Peter Zijlstra]

On Fri, May 29, 2026 at 10:13:35PM +0200, Peter Zijlstra wrote:

> It is somewhat possible to do an RT aware read-write spinlock thing, but
> it is definitely non-trivial and this would be the only user.

Furthermore, it is fundamentally one of the worst possible lock types.

It really isn't something you *want* to have -- arguably even for !RT.

They scale like ass; per them being a spinlock type, the critical
sections must be short, but this means there is nothing to amortize the
cost of bouncing the shared lock around -- which is the 'saving' grace
of the rwsem.

For short sections the cost of the shared access will dominate. I'm sure
Paul has a bunch of graphs to illustrate this point :-)

Re: [PATCH v2 02/20] KVM: x86/xen: Use read_trylock() for GPC locks in hardirq/atomic paths

[Hillf Danton]

On Fri, 29 May 2026 09:50:56 -0700 Sean Christopherson wrote:
> From: David Woodhouse <dwmw@amazon.co.uk>
> 
> kvm_xen_set_evtchn_fast() is called from hardirq context (timer
> callback, kvm_arch_set_irq_inatomic()). On PREEMPT_RT, rwlock_t is a
> sleeping lock, so read_lock_irqsave() cannot be used in this context.
> 
> Switch to read_trylock() and return -EWOULDBLOCK on contention, which is
> the designed fallback — there is always a slow path for the case where
> the GPC is invalid and needs to be refreshed.
> 
> Reported-by: syzbot+208f7f3e5f59c11aeb90@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=208f7f3e5f59c11aeb90
> Fixes: 14243b387137 ("KVM: x86/xen: Add KVM_IRQ_ROUTING_XEN_EVTCHN and event channel delivery")
> Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
> Signed-off-by: Sean Christopherson <seanjc@google.com>
> ---
>  arch/x86/kvm/xen.c | 32 +++++++++++++++++++++++---------
>  1 file changed, 23 insertions(+), 9 deletions(-)
> 
> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> index 91fd3673c09a..9bdb8e3cad58 100644
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c
> @@ -697,6 +697,7 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
>  int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
>  {
>  	struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
> +	bool atomic = in_atomic() || !task_is_running(current);
>  	unsigned long flags;
>  	u8 rc = 0;
>  
> @@ -713,7 +714,15 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
>  	BUILD_BUG_ON(sizeof(rc) !=
>  		     sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
>  
> -	read_lock_irqsave(&gpc->lock, flags);
> +	if (atomic) {
> +		local_irq_save(flags);
> +		if (!read_trylock(&gpc->lock)) {
> +			local_irq_restore(flags);
> +			return 1;
> +		}
> +	} else {
> +		read_lock_irqsave(&gpc->lock, flags);
> +	}
>  	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
>  		read_unlock_irqrestore(&gpc->lock, flags);
>  
I suspect this works given

static __always_inline void read_unlock_irqrestore(rwlock_t *rwlock,
						   unsigned long flags)
	__releases_shared(rwlock)
{
	rt_read_unlock(rwlock);
}

Re: [PATCH v2 01/20] locking/rt: Use raw_spin_lock_irqsave() in __rwbase_read_unlock()

[Sean Christopherson]

On Fri, May 29, 2026, Peter Zijlstra wrote:
> On Fri, May 29, 2026 at 10:13:35PM +0200, Peter Zijlstra wrote:
> 
> > It is somewhat possible to do an RT aware read-write spinlock thing, but
> > it is definitely non-trivial and this would be the only user.
> 
> Furthermore, it is fundamentally one of the worst possible lock types.
> 
> It really isn't something you *want* to have -- arguably even for !RT.
> 
> They scale like ass; per them being a spinlock type, the critical
> sections must be short, but this means there is nothing to amortize the
> cost of bouncing the shared lock around -- which is the 'saving' grace
> of the rwsem.
> 
> For short sections the cost of the shared access will dominate. I'm sure
> Paul has a bunch of graphs to illustrate this point :-)

Hrm.  We might be able to do an SRCU-based implementation.  Full RCU would be
too unpredictable on the update side.

The big hiccup is when KVM needs to reclaim memory in response to mmu_notifier
events from the OOM killer, which don't allow sleeping.  I'll stare at this a
bit next week.  I'm not exactly thrilled about the idea of ripping apart this
code, but there's a known performance bottleneck with mmu_notifier events,
maybe we can kill two birds with one stone.

Thanks for the input!