oslib-posix: initialize backend memory objects in parallel

QEMU initializes preallocated backend memory as the objects are parsed from
the command line. This is not optimal in some cases (e.g. memory spanning
multiple NUMA nodes) because the memory objects are initialized in series.

Allow the initialization to occur in parallel (asynchronously). In order to
ensure optimal thread placement, asynchronous initialization requires prealloc
context threads to be in use.

Signed-off-by: Mark Kanda <mark.kanda@oracle.com>
Message-ID: <20240131165327.3154970-2-mark.kanda@oracle.com>
Tested-by: Mario Casquero <mcasquer@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>

1 files changed, 100 insertions, 31 deletions

diff --git a/util/oslib-posix.c b/util/oslib-posix.c
index 7c297003b9..3c379f96c2 100644
--- a/util/oslib-posix.c
+++ b/util/oslib-posix.c
@@ -42,6 +42,7 @@
 #include "qemu/cutils.h"
 #include "qemu/units.h"
 #include "qemu/thread-context.h"
+#include "qemu/main-loop.h"
 
 #ifdef CONFIG_LINUX
 #include <sys/syscall.h>
@@ -63,11 +64,15 @@
 
 struct MemsetThread;
 
+static QLIST_HEAD(, MemsetContext) memset_contexts =
+    QLIST_HEAD_INITIALIZER(memset_contexts);
+
 typedef struct MemsetContext {
     bool all_threads_created;
     bool any_thread_failed;
     struct MemsetThread *threads;
     int num_threads;
+    QLIST_ENTRY(MemsetContext) next;
 } MemsetContext;
 
 struct MemsetThread {
@@ -412,19 +417,44 @@ static inline int get_memset_num_threads(size_t hpagesize, size_t numpages,
     return ret;
 }
 
+static int wait_and_free_mem_prealloc_context(MemsetContext *context)
+{
+    int i, ret = 0, tmp;
+
+    for (i = 0; i < context->num_threads; i++) {
+        tmp = (uintptr_t)qemu_thread_join(&context->threads[i].pgthread);
+
+        if (tmp) {
+            ret = tmp;
+        }
+    }
+    g_free(context->threads);
+    g_free(context);
+    return ret;
+}
+
 static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
-                           int max_threads, ThreadContext *tc,
+                           int max_threads, ThreadContext *tc, bool async,
                            bool use_madv_populate_write)
 {
     static gsize initialized = 0;
-    MemsetContext context = {
-        .num_threads = get_memset_num_threads(hpagesize, numpages, max_threads),
-    };
+    MemsetContext *context = g_malloc0(sizeof(MemsetContext));
     size_t numpages_per_thread, leftover;
     void *(*touch_fn)(void *);
-    int ret = 0, i = 0;
+    int ret, i = 0;
     char *addr = area;
 
+    /*
+     * Asynchronous preallocation is only allowed when using MADV_POPULATE_WRITE
+     * and prealloc context for thread placement.
+     */
+    if (!use_madv_populate_write || !tc) {
+        async = false;
+    }
+
+    context->num_threads =
+        get_memset_num_threads(hpagesize, numpages, max_threads);
+
     if (g_once_init_enter(&initialized)) {
         qemu_mutex_init(&page_mutex);
         qemu_cond_init(&page_cond);
@@ -432,8 +462,11 @@ static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
     }
 
     if (use_madv_populate_write) {
-        /* Avoid creating a single thread for MADV_POPULATE_WRITE */
-        if (context.num_threads == 1) {
+        /*
+         * Avoid creating a single thread for MADV_POPULATE_WRITE when
+         * preallocating synchronously.
+         */
+        if (context->num_threads == 1 && !async) {
             if (qemu_madvise(area, hpagesize * numpages,
                              QEMU_MADV_POPULATE_WRITE)) {
                 return -errno;
@@ -445,50 +478,86 @@ static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
         touch_fn = do_touch_pages;
     }
 
-    context.threads = g_new0(MemsetThread, context.num_threads);
-    numpages_per_thread = numpages / context.num_threads;
-    leftover = numpages % context.num_threads;
-    for (i = 0; i < context.num_threads; i++) {
-        context.threads[i].addr = addr;
-        context.threads[i].numpages = numpages_per_thread + (i < leftover);
-        context.threads[i].hpagesize = hpagesize;
-        context.threads[i].context = &context;
+    context->threads = g_new0(MemsetThread, context->num_threads);
+    numpages_per_thread = numpages / context->num_threads;
+    leftover = numpages % context->num_threads;
+    for (i = 0; i < context->num_threads; i++) {
+        context->threads[i].addr = addr;
+        context->threads[i].numpages = numpages_per_thread + (i < leftover);
+        context->threads[i].hpagesize = hpagesize;
+        context->threads[i].context = context;
         if (tc) {
-            thread_context_create_thread(tc, &context.threads[i].pgthread,
+            thread_context_create_thread(tc, &context->threads[i].pgthread,
                                          "touch_pages",
-                                         touch_fn, &context.threads[i],
+                                         touch_fn, &context->threads[i],
                                          QEMU_THREAD_JOINABLE);
         } else {
-            qemu_thread_create(&context.threads[i].pgthread, "touch_pages",
-                               touch_fn, &context.threads[i],
+            qemu_thread_create(&context->threads[i].pgthread, "touch_pages",
+                               touch_fn, &context->threads[i],
                                QEMU_THREAD_JOINABLE);
         }
-        addr += context.threads[i].numpages * hpagesize;
+        addr += context->threads[i].numpages * hpagesize;
+    }
+
+    if (async) {
+        /*
+         * async requests currently require the BQL. Add it to the list and kick
+         * preallocation off during qemu_finish_async_prealloc_mem().
+         */
+        assert(bql_locked());
+        QLIST_INSERT_HEAD(&memset_contexts, context, next);
+        return 0;
     }
 
     if (!use_madv_populate_write) {
-        sigbus_memset_context = &context;
+        sigbus_memset_context = context;
     }
 
     qemu_mutex_lock(&page_mutex);
-    context.all_threads_created = true;
+    context->all_threads_created = true;
     qemu_cond_broadcast(&page_cond);
     qemu_mutex_unlock(&page_mutex);
 
-    for (i = 0; i < context.num_threads; i++) {
-        int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread);
+    ret = wait_and_free_mem_prealloc_context(context);
 
+    if (!use_madv_populate_write) {
+        sigbus_memset_context = NULL;
+    }
+    return ret;
+}
+
+bool qemu_finish_async_prealloc_mem(Error **errp)
+{
+    int ret = 0, tmp;
+    MemsetContext *context, *next_context;
+
+    /* Waiting for preallocation requires the BQL. */
+    assert(bql_locked());
+    if (QLIST_EMPTY(&memset_contexts)) {
+        return true;
+    }
+
+    qemu_mutex_lock(&page_mutex);
+    QLIST_FOREACH(context, &memset_contexts, next) {
+        context->all_threads_created = true;
+    }
+    qemu_cond_broadcast(&page_cond);
+    qemu_mutex_unlock(&page_mutex);
+
+    QLIST_FOREACH_SAFE(context, &memset_contexts, next, next_context) {
+        QLIST_REMOVE(context, next);
+        tmp = wait_and_free_mem_prealloc_context(context);
         if (tmp) {
             ret = tmp;
         }
     }
 
-    if (!use_madv_populate_write) {
-        sigbus_memset_context = NULL;
+    if (ret) {
+        error_setg_errno(errp, -ret,
+                         "qemu_prealloc_mem: preallocating memory failed");
+        return false;
     }
-    g_free(context.threads);
-
-    return ret;
+    return true;
 }
 
 static bool madv_populate_write_possible(char *area, size_t pagesize)
@@ -498,7 +567,7 @@ static bool madv_populate_write_possible(char *area, size_t pagesize)
 }
 
 bool qemu_prealloc_mem(int fd, char *area, size_t sz, int max_threads,
-                       ThreadContext *tc, Error **errp)
+                       ThreadContext *tc, bool async, Error **errp)
 {
     static gsize initialized;
     int ret;
@@ -540,7 +609,7 @@ bool qemu_prealloc_mem(int fd, char *area, size_t sz, int max_threads,
     }
 
     /* touch pages simultaneously */
-    ret = touch_all_pages(area, hpagesize, numpages, max_threads, tc,
+    ret = touch_all_pages(area, hpagesize, numpages, max_threads, tc, async,
                           use_madv_populate_write);
     if (ret) {
         error_setg_errno(errp, -ret,