Merge tag 'pull-jan-omnibus-020223-1' of https://gitlab.com/stsquad/qemu into staging

Testing, docs, semihosting and plugin updates

  - update playbooks for custom runners
  - add section timing support to gitlab
  - upgrade fedora images to 37
  - purge perl from the build system and deps
  - disable unstable tests in CI
  - improve intro, emulation and semihosting docs
  - semihosting bug fix and O_BINARY default
  - add memory-sve test
  - fix some races in qht
  - improve plugin handling of memory helpers
  - optimise plugin hooks
  - fix some plugin deadlocks
  - reduce win64-cross build time by dropping some targets

# -----BEGIN PGP SIGNATURE-----
#
# iQEzBAABCgAdFiEEZoWumedRZ7yvyN81+9DbCVqeKkQFAmPb3fgACgkQ+9DbCVqe
# KkQbXAf9Eoc+PdNvafbqzH/blPjvd9ve8pJ+GcPDukNXwxP8OF/jFEJUQ1E7l9O7
# y0qV4akKCdIqVice4R5bK2CAq44Y3aut8SDf56C8E3Riha2zA2RbQWOv/zCvA3OP
# LFF+OaXZyg4JTR48HUKzh9ei2bd1+ccBSUe+xlRi59XaV5K8+5bmcZj10QKUR0lD
# 0HC5auEWWpayvd5D7Da15C7+oVY3LMCFxSdpHwbuIPPan/TRo5yqMI6ChYDKB8QD
# gdwMCL8znj2ADCTBftyBDYDAtjKVyLQidf7KdQHiSF+nmXYopS6SbsPCOMtJqCMH
# tXcKAIxs/MEntPrWTKTdtdnzotJVKw==
# =AtfN
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 02 Feb 2023 15:59:52 GMT
# gpg:                using RSA key 6685AE99E75167BCAFC8DF35FBD0DB095A9E2A44
# gpg: Good signature from "Alex Bennée (Master Work Key) <alex.bennee@linaro.org>" [full]
# Primary key fingerprint: 6685 AE99 E751 67BC AFC8  DF35 FBD0 DB09 5A9E 2A44

* tag 'pull-jan-omnibus-020223-1' of https://gitlab.com/stsquad/qemu: (36 commits)
  gitlab: cut even more from cross-win64-system build
  plugins: Iterate on cb_lists in qemu_plugin_user_exit
  cpu-exec: assert that plugin_mem_cbs is NULL after execution
  tcg: exclude non-memory effecting helpers from instrumentation
  translator: always pair plugin_gen_insn_{start, end} calls
  plugins: fix optimization in plugin_gen_disable_mem_helpers
  plugins: make qemu_plugin_user_exit's locking order consistent with fork_start's
  util/qht: use striped locks under TSAN
  thread: de-const qemu_spin_destroy
  util/qht: add missing atomic_set(hashes[i])
  cpu: free cpu->tb_jmp_cache with RCU
  tests/tcg: add memory-sve test for aarch64
  semihosting: add O_BINARY flag in host_open for NT compatibility
  semihosting: Write back semihosting data before completion callback
  docs: add an introduction to the system docs
  semihosting: add semihosting section to the docs
  docs: add a new section to outline emulation support
  docs: add hotlinks to about preface text
  MAINTAINERS: Fix the entry for tests/tcg/nios2
  gitlab: wrap up test results for custom runners
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

1 files changed, 82 insertions, 15 deletions

diff --git a/util/qht.c b/util/qht.c
index 065fc501f4..92c6b78759 100644
--- a/util/qht.c
+++ b/util/qht.c
@@ -151,6 +151,22 @@ struct qht_bucket {
 
 QEMU_BUILD_BUG_ON(sizeof(struct qht_bucket) > QHT_BUCKET_ALIGN);
 
+/*
+ * Under TSAN, we use striped locks instead of one lock per bucket chain.
+ * This avoids crashing under TSAN, since TSAN aborts the program if more than
+ * 64 locks are held (this is a hardcoded limit in TSAN).
+ * When resizing a QHT we grab all the buckets' locks, which can easily
+ * go over TSAN's limit. By using striped locks, we avoid this problem.
+ *
+ * Note: this number must be a power of two for easy index computation.
+ */
+#define QHT_TSAN_BUCKET_LOCKS_BITS 4
+#define QHT_TSAN_BUCKET_LOCKS (1 << QHT_TSAN_BUCKET_LOCKS_BITS)
+
+struct qht_tsan_lock {
+    QemuSpin lock;
+} QEMU_ALIGNED(QHT_BUCKET_ALIGN);
+
 /**
  * struct qht_map - structure to track an array of buckets
  * @rcu: used by RCU. Keep it as the top field in the struct to help valgrind
@@ -160,6 +176,7 @@ QEMU_BUILD_BUG_ON(sizeof(struct qht_bucket) > QHT_BUCKET_ALIGN);
  * @n_added_buckets: number of added (i.e. "non-head") buckets
  * @n_added_buckets_threshold: threshold to trigger an upward resize once the
  *                             number of added buckets surpasses it.
+ * @tsan_bucket_locks: Array of striped locks to be used only under TSAN.
  *
  * Buckets are tracked in what we call a "map", i.e. this structure.
  */
@@ -169,6 +186,9 @@ struct qht_map {
     size_t n_buckets;
     size_t n_added_buckets;
     size_t n_added_buckets_threshold;
+#ifdef CONFIG_TSAN
+    struct qht_tsan_lock tsan_bucket_locks[QHT_TSAN_BUCKET_LOCKS];
+#endif
 };
 
 /* trigger a resize when n_added_buckets > n_buckets / div */
@@ -229,10 +249,56 @@ static inline size_t qht_elems_to_buckets(size_t n_elems)
     return pow2ceil(n_elems / QHT_BUCKET_ENTRIES);
 }
 
-static inline void qht_head_init(struct qht_bucket *b)
+/*
+ * When using striped locks (i.e. under TSAN), we have to be careful not
+ * to operate on the same lock twice (e.g. when iterating through all buckets).
+ * We achieve this by operating only on each stripe's first matching lock.
+ */
+static inline void qht_do_if_first_in_stripe(struct qht_map *map,
+                                             struct qht_bucket *b,
+                                             void (*func)(QemuSpin *spin))
+{
+#ifdef CONFIG_TSAN
+    unsigned long bucket_idx = b - map->buckets;
+    bool is_first_in_stripe = (bucket_idx >> QHT_TSAN_BUCKET_LOCKS_BITS) == 0;
+    if (is_first_in_stripe) {
+        unsigned long lock_idx = bucket_idx & (QHT_TSAN_BUCKET_LOCKS - 1);
+        func(&map->tsan_bucket_locks[lock_idx].lock);
+    }
+#else
+    func(&b->lock);
+#endif
+}
+
+static inline void qht_bucket_lock_do(struct qht_map *map,
+                                      struct qht_bucket *b,
+                                      void (*func)(QemuSpin *lock))
+{
+#ifdef CONFIG_TSAN
+    unsigned long bucket_idx = b - map->buckets;
+    unsigned long lock_idx = bucket_idx & (QHT_TSAN_BUCKET_LOCKS - 1);
+    func(&map->tsan_bucket_locks[lock_idx].lock);
+#else
+    func(&b->lock);
+#endif
+}
+
+static inline void qht_bucket_lock(struct qht_map *map,
+                                   struct qht_bucket *b)
+{
+    qht_bucket_lock_do(map, b, qemu_spin_lock);
+}
+
+static inline void qht_bucket_unlock(struct qht_map *map,
+                                     struct qht_bucket *b)
+{
+    qht_bucket_lock_do(map, b, qemu_spin_unlock);
+}
+
+static inline void qht_head_init(struct qht_map *map, struct qht_bucket *b)
 {
     memset(b, 0, sizeof(*b));
-    qemu_spin_init(&b->lock);
+    qht_do_if_first_in_stripe(map, b, qemu_spin_init);
     seqlock_init(&b->sequence);
 }
 
@@ -250,7 +316,7 @@ static void qht_map_lock_buckets(struct qht_map *map)
     for (i = 0; i < map->n_buckets; i++) {
         struct qht_bucket *b = &map->buckets[i];
 
-        qemu_spin_lock(&b->lock);
+        qht_do_if_first_in_stripe(map, b, qemu_spin_lock);
     }
 }
 
@@ -261,7 +327,7 @@ static void qht_map_unlock_buckets(struct qht_map *map)
     for (i = 0; i < map->n_buckets; i++) {
         struct qht_bucket *b = &map->buckets[i];
 
-        qemu_spin_unlock(&b->lock);
+        qht_do_if_first_in_stripe(map, b, qemu_spin_unlock);
     }
 }
 
@@ -308,7 +374,7 @@ void qht_map_lock_buckets__no_stale(struct qht *ht, struct qht_map **pmap)
  * Get a head bucket and lock it, making sure its parent map is not stale.
  * @pmap is filled with a pointer to the bucket's parent map.
  *
- * Unlock with qemu_spin_unlock(&b->lock).
+ * Unlock with qht_bucket_unlock.
  *
  * Note: callers cannot have ht->lock held.
  */
@@ -322,18 +388,18 @@ struct qht_bucket *qht_bucket_lock__no_stale(struct qht *ht, uint32_t hash,
     map = qatomic_rcu_read(&ht->map);
     b = qht_map_to_bucket(map, hash);
 
-    qemu_spin_lock(&b->lock);
+    qht_bucket_lock(map, b);
     if (likely(!qht_map_is_stale__locked(ht, map))) {
         *pmap = map;
         return b;
     }
-    qemu_spin_unlock(&b->lock);
+    qht_bucket_unlock(map, b);
 
     /* we raced with a resize; acquire ht->lock to see the updated ht->map */
     qht_lock(ht);
     map = ht->map;
     b = qht_map_to_bucket(map, hash);
-    qemu_spin_lock(&b->lock);
+    qht_bucket_lock(map, b);
     qht_unlock(ht);
     *pmap = map;
     return b;
@@ -345,12 +411,13 @@ static inline bool qht_map_needs_resize(const struct qht_map *map)
            map->n_added_buckets_threshold;
 }
 
-static inline void qht_chain_destroy(const struct qht_bucket *head)
+static inline void qht_chain_destroy(struct qht_map *map,
+                                     struct qht_bucket *head)
 {
     struct qht_bucket *curr = head->next;
     struct qht_bucket *prev;
 
-    qemu_spin_destroy(&head->lock);
+    qht_do_if_first_in_stripe(map, head, qemu_spin_destroy);
     while (curr) {
         prev = curr;
         curr = curr->next;
@@ -364,7 +431,7 @@ static void qht_map_destroy(struct qht_map *map)
     size_t i;
 
     for (i = 0; i < map->n_buckets; i++) {
-        qht_chain_destroy(&map->buckets[i]);
+        qht_chain_destroy(map, &map->buckets[i]);
     }
     qemu_vfree(map->buckets);
     g_free(map);
@@ -390,7 +457,7 @@ static struct qht_map *qht_map_create(size_t n_buckets)
     map->buckets = qemu_memalign(QHT_BUCKET_ALIGN,
                                  sizeof(*map->buckets) * n_buckets);
     for (i = 0; i < n_buckets; i++) {
-        qht_head_init(&map->buckets[i]);
+        qht_head_init(map, &map->buckets[i]);
     }
     return map;
 }
@@ -638,7 +705,7 @@ bool qht_insert(struct qht *ht, void *p, uint32_t hash, void **existing)
     b = qht_bucket_lock__no_stale(ht, hash, &map);
     prev = qht_insert__locked(ht, map, b, p, hash, &needs_resize);
     qht_bucket_debug__locked(b);
-    qemu_spin_unlock(&b->lock);
+    qht_bucket_unlock(map, b);
 
     if (unlikely(needs_resize) && ht->mode & QHT_MODE_AUTO_RESIZE) {
         qht_grow_maybe(ht);
@@ -688,7 +755,7 @@ static inline void qht_bucket_remove_entry(struct qht_bucket *orig, int pos)
     int i;
 
     if (qht_entry_is_last(orig, pos)) {
-        orig->hashes[pos] = 0;
+        qatomic_set(&orig->hashes[pos], 0);
         qatomic_set(&orig->pointers[pos], NULL);
         return;
     }
@@ -749,7 +816,7 @@ bool qht_remove(struct qht *ht, const void *p, uint32_t hash)
     b = qht_bucket_lock__no_stale(ht, hash, &map);
     ret = qht_remove__locked(b, p, hash);
     qht_bucket_debug__locked(b);
-    qemu_spin_unlock(&b->lock);
+    qht_bucket_unlock(map, b);
     return ret;
 }