Have taskq_create_synced() wait for threads to be created.

DataCoreSoftware · Feb 28, 2024 · 61ebd48 · 61ebd48
1 parent 3160047
commit 61ebd48
Show file tree

Hide file tree

Showing 4 changed files with 29 additions and 5 deletions.
diff --git a/include/os/windows/spl/sys/sysmacros.h b/include/os/windows/spl/sys/sysmacros.h
@@ -60,6 +60,8 @@ extern uint32_t cpu_number(void);
 #define	is_system_labeled()		0
 
 extern unsigned int max_ncpus;
+extern unsigned	int boot_ncpus;
+extern unsigned int num_ecores;
 
 #ifndef RLIM64_INFINITY
 #define	RLIM64_INFINITY			(~0ULL)
@@ -132,7 +134,6 @@ extern uint32_t zone_get_hostid(void *zone);
 extern void spl_setup(void);
 extern void spl_cleanup(void);
 
-#define	boot_ncpus max_ncpus
 #define	SET_ERROR(err) \
 	(__set_error(__FILE__, __func__, __LINE__, err), err)
 

diff --git a/include/os/windows/spl/sys/taskq.h b/include/os/windows/spl/sys/taskq.h
@@ -58,8 +58,9 @@ struct proc;
 #define	TASKQ_DC_BATCH		0x0010	/* Taskq uses SDC in batch mode */
 
 #ifdef _WIN32
-#define	TASKQ_TIMESHARE         0x0020  /* macOS dynamic thread priority */
-#define	TASKQ_REALLY_DYNAMIC    0x0040  /* don't filter out TASKQ_DYNAMIC */
+#define	TASKQ_TIMESHARE		0x0020  /* macOS dynamic thread priority */
+#define	TASKQ_REALLY_DYNAMIC	0x0040  /* don't filter out TASKQ_DYNAMIC */
+#define	TASKQ_CREATE_SYNCED	0x0080	/* don't deflate ncpus */
 #endif
 /*
  * Flags for taskq_dispatch. TQ_SLEEP/TQ_NOSLEEP should be same as

diff --git a/module/os/windows/spl/spl-taskq.c b/module/os/windows/spl/spl-taskq.c
@@ -2374,6 +2374,8 @@ taskq_create_common(const char *name, int instance, int nthreads, pri_t pri,
 	taskq_t *tq = kmem_cache_alloc(taskq_cache, KM_SLEEP);
 #ifdef _WIN32
 	uint_t ncpus = max_ncpus;
+	if (!(flags & TASKQ_CREATE_SYNCED))
+		ncpus = boot_ncpus; /* possibly deflated by num_ecores */
 #else
 	uint_t ncpus = ((boot_max_ncpus == -1) ? max_ncpus : boot_max_ncpus);
 #endif
@@ -2834,9 +2836,22 @@ taskq_create_synced(const char *name, int nthreads, pri_t pri,
 	flags &= ~(TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT | TASKQ_DC_BATCH);
 
 	tq = taskq_create(name, nthreads, minclsyspri, nthreads, INT_MAX,
-	    flags | TASKQ_PREPOPULATE);
+	    flags | TASKQ_PREPOPULATE | TASKQ_CREATE_SYNCED);
+
 	VERIFY(tq != NULL);
-	VERIFY(tq->tq_nthreads == nthreads);
+
+	/* wait until our minalloc (nthreads) threads are created */
+	mutex_enter(&tq->tq_lock);
+	for (int i = 1; tq->tq_nthreads != nthreads; i++) {
+		dprintf("SPL: %s:%d: waiting for tq_nthreads (%d)"
+		    " to be nthreads (%d), (target = %d, pass %d)\n",
+		    __func__, __LINE__,
+		    tq->tq_nthreads, tq->tq_nthreads_target,  nthreads, i);
+		cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
+	}
+	mutex_exit(&tq->tq_lock);
+
+	VERIFY3U(tq->tq_nthreads, ==, nthreads);
 
 	/* spawn all syncthreads */
 	for (int i = 0; i < nthreads; i++) {

diff --git a/module/os/windows/spl/spl-windows.c b/module/os/windows/spl/spl-windows.c
@@ -42,6 +42,7 @@
 static utsname_t utsname_static = { { 0 } };
 
 unsigned int max_ncpus = 0;
+unsigned int boot_ncpus = 0;
 uint64_t  total_memory = 0;
 uint64_t  real_total_memory = 0;
 
@@ -480,6 +481,12 @@ spl_start(PUNICODE_STRING RegistryPath)
 	max_ncpus = KeQueryActiveProcessorCountEx(ALL_PROCESSOR_GROUPS);
 	if (!max_ncpus) max_ncpus = 1;
 	dprintf("SPL: total ncpu %d\n", max_ncpus);
+#if defined(__arm64__)
+	num_ecores = (max_ncpus > 4) ? 4 : 0; // Apple has 4 eCores, fixme
+	boot_ncpus = MAX(1, (int)max_ncpus - (int)num_ecores);
+#else
+	boot_ncpus = max_ncpus;
+#endif
 
 	// Not sure how to get physical RAM size in a Windows Driver
 	// So until then, pull some numbers out of the aether. Next