path: root/0051-locking-attempt-to-ensure-lock-wrappers-are-always-i.patch

From 2cc5e57be680a516aa5cdef4281856d09b9d0ea6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Roger=20Pau=20Monn=C3=A9?= <roger.pau@citrix.com>
Date: Mon, 4 Mar 2024 14:29:36 +0100
Subject: [PATCH 51/67] locking: attempt to ensure lock wrappers are always
 inline
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

In order to prevent the locking speculation barriers from being inside of
`call`ed functions that could be speculatively bypassed.

While there also add an extra locking barrier to _mm_write_lock() in the branch
taken when the lock is already held.

Note some functions are switched to use the unsafe variants (without speculation
barrier) of the locking primitives, but a speculation barrier is always added
to the exposed public lock wrapping helper.  That's the case with
sched_spin_lock_double() or pcidevs_lock() for example.

This is part of XSA-453 / CVE-2024-2193

Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
(cherry picked from commit 197ecd838a2aaf959a469df3696d4559c4f8b762)
---
 xen/arch/x86/hvm/vpt.c         | 10 +++++++---
 xen/arch/x86/include/asm/irq.h |  1 +
 xen/arch/x86/mm/mm-locks.h     | 28 +++++++++++++++-------------
 xen/arch/x86/mm/p2m-pod.c      |  2 +-
 xen/common/event_channel.c     |  5 +++--
 xen/common/grant_table.c       |  6 +++---
 xen/common/sched/core.c        | 19 ++++++++++++-------
 xen/common/sched/private.h     | 26 ++++++++++++++++++++++++--
 xen/common/timer.c             |  8 +++++---
 xen/drivers/passthrough/pci.c  |  5 +++--
 xen/include/xen/event.h        |  4 ++--
 xen/include/xen/pci.h          |  8 ++++++--
 12 files changed, 82 insertions(+), 40 deletions(-)

diff --git a/xen/arch/x86/hvm/vpt.c b/xen/arch/x86/hvm/vpt.c
index cb1d81bf9e..66f1095245 100644
--- a/xen/arch/x86/hvm/vpt.c
+++ b/xen/arch/x86/hvm/vpt.c
@@ -161,7 +161,7 @@ static int pt_irq_masked(struct periodic_time *pt)
  * pt->vcpu field, because another thread holding the pt_migrate lock
  * may already be spinning waiting for your vcpu lock.
  */
-static void pt_vcpu_lock(struct vcpu *v)
+static always_inline void pt_vcpu_lock(struct vcpu *v)
 {
     spin_lock(&v->arch.hvm.tm_lock);
 }
@@ -180,9 +180,13 @@ static void pt_vcpu_unlock(struct vcpu *v)
  * need to take an additional lock that protects against pt->vcpu
  * changing.
  */
-static void pt_lock(struct periodic_time *pt)
+static always_inline void pt_lock(struct periodic_time *pt)
 {
-    read_lock(&pt->vcpu->domain->arch.hvm.pl_time->pt_migrate);
+    /*
+     * Use the speculation unsafe variant for the first lock, as the following
+     * lock taking helper already includes a speculation barrier.
+     */
+    _read_lock(&pt->vcpu->domain->arch.hvm.pl_time->pt_migrate);
     spin_lock(&pt->vcpu->arch.hvm.tm_lock);
 }
 
diff --git a/xen/arch/x86/include/asm/irq.h b/xen/arch/x86/include/asm/irq.h
index f6a0207a80..823d627fd0 100644
--- a/xen/arch/x86/include/asm/irq.h
+++ b/xen/arch/x86/include/asm/irq.h
@@ -178,6 +178,7 @@ void cf_check irq_complete_move(struct irq_desc *);
 
 extern struct irq_desc *irq_desc;
 
+/* Not speculation safe, only used for AP bringup. */
 void lock_vector_lock(void);
 void unlock_vector_lock(void);
 
diff --git a/xen/arch/x86/mm/mm-locks.h b/xen/arch/x86/mm/mm-locks.h
index c1523aeccf..265239c49f 100644
--- a/xen/arch/x86/mm/mm-locks.h
+++ b/xen/arch/x86/mm/mm-locks.h
@@ -86,8 +86,8 @@ static inline void _set_lock_level(int l)
     this_cpu(mm_lock_level) = l;
 }
 
-static inline void _mm_lock(const struct domain *d, mm_lock_t *l,
-                            const char *func, int level, int rec)
+static always_inline void _mm_lock(const struct domain *d, mm_lock_t *l,
+                                   const char *func, int level, int rec)
 {
     if ( !((mm_locked_by_me(l)) && rec) )
         _check_lock_level(d, level);
@@ -137,8 +137,8 @@ static inline int mm_write_locked_by_me(mm_rwlock_t *l)
     return (l->locker == get_processor_id());
 }
 
-static inline void _mm_write_lock(const struct domain *d, mm_rwlock_t *l,
-                                  const char *func, int level)
+static always_inline void _mm_write_lock(const struct domain *d, mm_rwlock_t *l,
+                                         const char *func, int level)
 {
     if ( !mm_write_locked_by_me(l) )
     {
@@ -149,6 +149,8 @@ static inline void _mm_write_lock(const struct domain *d, mm_rwlock_t *l,
         l->unlock_level = _get_lock_level();
         _set_lock_level(_lock_level(d, level));
     }
+    else
+        block_speculation();
     l->recurse_count++;
 }
 
@@ -162,8 +164,8 @@ static inline void mm_write_unlock(mm_rwlock_t *l)
     percpu_write_unlock(p2m_percpu_rwlock, &l->lock);
 }
 
-static inline void _mm_read_lock(const struct domain *d, mm_rwlock_t *l,
-                                 int level)
+static always_inline void _mm_read_lock(const struct domain *d, mm_rwlock_t *l,
+                                        int level)
 {
     _check_lock_level(d, level);
     percpu_read_lock(p2m_percpu_rwlock, &l->lock);
@@ -178,15 +180,15 @@ static inline void mm_read_unlock(mm_rwlock_t *l)
 
 /* This wrapper uses the line number to express the locking order below */
 #define declare_mm_lock(name)                                                 \
-    static inline void mm_lock_##name(const struct domain *d, mm_lock_t *l,   \
-                                      const char *func, int rec)              \
+    static always_inline void mm_lock_##name(                                 \
+        const struct domain *d, mm_lock_t *l, const char *func, int rec)      \
     { _mm_lock(d, l, func, MM_LOCK_ORDER_##name, rec); }
 #define declare_mm_rwlock(name)                                               \
-    static inline void mm_write_lock_##name(const struct domain *d,           \
-                                            mm_rwlock_t *l, const char *func) \
+    static always_inline void mm_write_lock_##name(                           \
+        const struct domain *d, mm_rwlock_t *l, const char *func)             \
     { _mm_write_lock(d, l, func, MM_LOCK_ORDER_##name); }                     \
-    static inline void mm_read_lock_##name(const struct domain *d,            \
-                                           mm_rwlock_t *l)                    \
+    static always_inline void mm_read_lock_##name(const struct domain *d,     \
+                                                  mm_rwlock_t *l)             \
     { _mm_read_lock(d, l, MM_LOCK_ORDER_##name); }
 /* These capture the name of the calling function */
 #define mm_lock(name, d, l) mm_lock_##name(d, l, __func__, 0)
@@ -321,7 +323,7 @@ declare_mm_lock(altp2mlist)
 #define MM_LOCK_ORDER_altp2m                 40
 declare_mm_rwlock(altp2m);
 
-static inline void p2m_lock(struct p2m_domain *p)
+static always_inline void p2m_lock(struct p2m_domain *p)
 {
     if ( p2m_is_altp2m(p) )
         mm_write_lock(altp2m, p->domain, &p->lock);
diff --git a/xen/arch/x86/mm/p2m-pod.c b/xen/arch/x86/mm/p2m-pod.c
index fc110506dc..99dbcb3101 100644
--- a/xen/arch/x86/mm/p2m-pod.c
+++ b/xen/arch/x86/mm/p2m-pod.c
@@ -36,7 +36,7 @@
 #define superpage_aligned(_x)  (((_x)&(SUPERPAGE_PAGES-1))==0)
 
 /* Enforce lock ordering when grabbing the "external" page_alloc lock */
-static inline void lock_page_alloc(struct p2m_domain *p2m)
+static always_inline void lock_page_alloc(struct p2m_domain *p2m)
 {
     page_alloc_mm_pre_lock(p2m->domain);
     spin_lock(&(p2m->domain->page_alloc_lock));
diff --git a/xen/common/event_channel.c b/xen/common/event_channel.c
index f5e0b12d15..dada9f15f5 100644
--- a/xen/common/event_channel.c
+++ b/xen/common/event_channel.c
@@ -62,7 +62,7 @@
  * just assume the event channel is free or unbound at the moment when the
  * evtchn_read_trylock() returns false.
  */
-static inline void evtchn_write_lock(struct evtchn *evtchn)
+static always_inline void evtchn_write_lock(struct evtchn *evtchn)
 {
     write_lock(&evtchn->lock);
 
@@ -364,7 +364,8 @@ int evtchn_alloc_unbound(evtchn_alloc_unbound_t *alloc, evtchn_port_t port)
     return rc;
 }
 
-static void double_evtchn_lock(struct evtchn *lchn, struct evtchn *rchn)
+static always_inline void double_evtchn_lock(struct evtchn *lchn,
+                                             struct evtchn *rchn)
 {
     ASSERT(lchn != rchn);
 
diff --git a/xen/common/grant_table.c b/xen/common/grant_table.c
index ee7cc496b8..62a8685cd5 100644
--- a/xen/common/grant_table.c
+++ b/xen/common/grant_table.c
@@ -410,7 +410,7 @@ static inline void act_set_gfn(struct active_grant_entry *act, gfn_t gfn)
 
 static DEFINE_PERCPU_RWLOCK_GLOBAL(grant_rwlock);
 
-static inline void grant_read_lock(struct grant_table *gt)
+static always_inline void grant_read_lock(struct grant_table *gt)
 {
     percpu_read_lock(grant_rwlock, &gt->lock);
 }
@@ -420,7 +420,7 @@ static inline void grant_read_unlock(struct grant_table *gt)
     percpu_read_unlock(grant_rwlock, &gt->lock);
 }
 
-static inline void grant_write_lock(struct grant_table *gt)
+static always_inline void grant_write_lock(struct grant_table *gt)
 {
     percpu_write_lock(grant_rwlock, &gt->lock);
 }
@@ -457,7 +457,7 @@ nr_active_grant_frames(struct grant_table *gt)
     return num_act_frames_from_sha_frames(nr_grant_frames(gt));
 }
 
-static inline struct active_grant_entry *
+static always_inline struct active_grant_entry *
 active_entry_acquire(struct grant_table *t, grant_ref_t e)
 {
     struct active_grant_entry *act;
diff --git a/xen/common/sched/core.c b/xen/common/sched/core.c
index 078beb1adb..29bbab5ac6 100644
--- a/xen/common/sched/core.c
+++ b/xen/common/sched/core.c
@@ -348,23 +348,28 @@ uint64_t get_cpu_idle_time(unsigned int cpu)
  * This avoids dead- or live-locks when this code is running on both
  * cpus at the same time.
  */
-static void sched_spin_lock_double(spinlock_t *lock1, spinlock_t *lock2,
-                                   unsigned long *flags)
+static always_inline void sched_spin_lock_double(
+    spinlock_t *lock1, spinlock_t *lock2, unsigned long *flags)
 {
+    /*
+     * In order to avoid extra overhead, use the locking primitives without the
+     * speculation barrier, and introduce a single barrier here.
+     */
     if ( lock1 == lock2 )
     {
-        spin_lock_irqsave(lock1, *flags);
+        *flags = _spin_lock_irqsave(lock1);
     }
     else if ( lock1 < lock2 )
     {
-        spin_lock_irqsave(lock1, *flags);
-        spin_lock(lock2);
+        *flags = _spin_lock_irqsave(lock1);
+        _spin_lock(lock2);
     }
     else
     {
-        spin_lock_irqsave(lock2, *flags);
-        spin_lock(lock1);
+        *flags = _spin_lock_irqsave(lock2);
+        _spin_lock(lock1);
     }
+    block_lock_speculation();
 }
 
 static void sched_spin_unlock_double(spinlock_t *lock1, spinlock_t *lock2,
diff --git a/xen/common/sched/private.h b/xen/common/sched/private.h
index 0527a8c70d..24a93dd0c1 100644
--- a/xen/common/sched/private.h
+++ b/xen/common/sched/private.h
@@ -207,8 +207,24 @@ DECLARE_PER_CPU(cpumask_t, cpumask_scratch);
 #define cpumask_scratch        (&this_cpu(cpumask_scratch))
 #define cpumask_scratch_cpu(c) (&per_cpu(cpumask_scratch, c))
 
+/*
+ * Deal with _spin_lock_irqsave() returning the flags value instead of storing
+ * it in a passed parameter.
+ */
+#define _sched_spinlock0(lock, irq) _spin_lock##irq(lock)
+#define _sched_spinlock1(lock, irq, arg) ({ \
+    BUILD_BUG_ON(sizeof(arg) != sizeof(unsigned long)); \
+    (arg) = _spin_lock##irq(lock); \
+})
+
+#define _sched_spinlock__(nr) _sched_spinlock ## nr
+#define _sched_spinlock_(nr)  _sched_spinlock__(nr)
+#define _sched_spinlock(lock, irq, args...) \
+    _sched_spinlock_(count_args(args))(lock, irq, ## args)
+
 #define sched_lock(kind, param, cpu, irq, arg...) \
-static inline spinlock_t *kind##_schedule_lock##irq(param EXTRA_TYPE(arg)) \
+static always_inline spinlock_t \
+*kind##_schedule_lock##irq(param EXTRA_TYPE(arg)) \
 { \
     for ( ; ; ) \
     { \
@@ -220,10 +236,16 @@ static inline spinlock_t *kind##_schedule_lock##irq(param EXTRA_TYPE(arg)) \
          * \
          * It may also be the case that v->processor may change but the \
          * lock may be the same; this will succeed in that case. \
+         * \
+         * Use the speculation unsafe locking helper, there's a speculation \
+         * barrier before returning to the caller. \
          */ \
-        spin_lock##irq(lock, ## arg); \
+        _sched_spinlock(lock, irq, ## arg); \
         if ( likely(lock == get_sched_res(cpu)->schedule_lock) ) \
+        { \
+            block_lock_speculation(); \
             return lock; \
+        } \
         spin_unlock##irq(lock, ## arg); \
     } \
 }
diff --git a/xen/common/timer.c b/xen/common/timer.c
index 9b5016d5ed..459668d417 100644
--- a/xen/common/timer.c
+++ b/xen/common/timer.c
@@ -240,7 +240,7 @@ static inline void deactivate_timer(struct timer *timer)
     list_add(&timer->inactive, &per_cpu(timers, timer->cpu).inactive);
 }
 
-static inline bool_t timer_lock(struct timer *timer)
+static inline bool_t timer_lock_unsafe(struct timer *timer)
 {
     unsigned int cpu;
 
@@ -254,7 +254,8 @@ static inline bool_t timer_lock(struct timer *timer)
             rcu_read_unlock(&timer_cpu_read_lock);
             return 0;
         }
-        spin_lock(&per_cpu(timers, cpu).lock);
+        /* Use the speculation unsafe variant, the wrapper has the barrier. */
+        _spin_lock(&per_cpu(timers, cpu).lock);
         if ( likely(timer->cpu == cpu) )
             break;
         spin_unlock(&per_cpu(timers, cpu).lock);
@@ -267,8 +268,9 @@ static inline bool_t timer_lock(struct timer *timer)
 #define timer_lock_irqsave(t, flags) ({         \
     bool_t __x;                                 \
     local_irq_save(flags);                      \
-    if ( !(__x = timer_lock(t)) )               \
+    if ( !(__x = timer_lock_unsafe(t)) )        \
         local_irq_restore(flags);               \
+    block_lock_speculation();                   \
     __x;                                        \
 })
 
diff --git a/xen/drivers/passthrough/pci.c b/xen/drivers/passthrough/pci.c
index 8c62b14d19..1b3d285166 100644
--- a/xen/drivers/passthrough/pci.c
+++ b/xen/drivers/passthrough/pci.c
@@ -52,9 +52,10 @@ struct pci_seg {
 
 static spinlock_t _pcidevs_lock = SPIN_LOCK_UNLOCKED;
 
-void pcidevs_lock(void)
+/* Do not use, as it has no speculation barrier, use pcidevs_lock() instead. */
+void pcidevs_lock_unsafe(void)
 {
-    spin_lock_recursive(&_pcidevs_lock);
+    _spin_lock_recursive(&_pcidevs_lock);
 }
 
 void pcidevs_unlock(void)
diff --git a/xen/include/xen/event.h b/xen/include/xen/event.h
index 8eae9984a9..dd96e84c69 100644
--- a/xen/include/xen/event.h
+++ b/xen/include/xen/event.h
@@ -114,12 +114,12 @@ void notify_via_xen_event_channel(struct domain *ld, int lport);
 #define bucket_from_port(d, p) \
     ((group_from_port(d, p))[((p) % EVTCHNS_PER_GROUP) / EVTCHNS_PER_BUCKET])
 
-static inline void evtchn_read_lock(struct evtchn *evtchn)
+static always_inline void evtchn_read_lock(struct evtchn *evtchn)
 {
     read_lock(&evtchn->lock);
 }
 
-static inline bool evtchn_read_trylock(struct evtchn *evtchn)
+static always_inline bool evtchn_read_trylock(struct evtchn *evtchn)
 {
     return read_trylock(&evtchn->lock);
 }
diff --git a/xen/include/xen/pci.h b/xen/include/xen/pci.h
index 5975ca2f30..b373f139d1 100644
--- a/xen/include/xen/pci.h
+++ b/xen/include/xen/pci.h
@@ -155,8 +155,12 @@ struct pci_dev {
  * devices, it also sync the access to the msi capability that is not
  * interrupt handling related (the mask bit register).
  */
-
-void pcidevs_lock(void);
+void pcidevs_lock_unsafe(void);
+static always_inline void pcidevs_lock(void)
+{
+    pcidevs_lock_unsafe();
+    block_lock_speculation();
+}
 void pcidevs_unlock(void);
 bool_t __must_check pcidevs_locked(void);
 
-- 
2.44.0