* Added optional spinlock contention measurement feature. Enabled when

B_DEBUG_SPINLOCK_CONTENTION is defined to 1. It typedefs spinlock to a
  structure (thus breaking BeOS binary compatibility), containing a
  counter which is incremented whenever a thread has to wait for the
  spinlock.
* Added macros for spinlock initialization and access and changed
  code using spinlocks accordingly. This breaks compilation for BeOS --
  the macros should be defined in the respective compatibility wrappers.
* Added generic syscall to get the spinlock counters for the thread and
  the team spinlocks.


git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25752 a95241bf-73f2-0310-859d-f6bbb57e9c96

headers/os/drivers/KernelExport.h

@@ -13,7 +13,32 @@
 /* interrupts and spinlocks */
 
 typedef ulong cpu_status;
-typedef vint32 spinlock;
+
+// WARNING: For Haiku debugging only! This changes the spinlock type in a
+// binary incompatible way!
+//#define B_DEBUG_SPINLOCK_CONTENTION	1
+
+#if B_DEBUG_SPINLOCK_CONTENTION
+	typedef struct {
+		vint32	lock;
+		vint32	count_low;
+		vint32	count_high;
+	} spinlock;
+
+#	define B_SPINLOCK_INITIALIZER { 0, 0, 0 }
+#	define B_INITIALIZE_SPINLOCK(spinlock)	do {	\
+			(spinlock)->lock = 0;					\
+			(spinlock)->count_low = 0;				\
+			(spinlock)->count_high = 0;				\
+		} while (false)
+#	define B_SPINLOCK_IS_LOCKED(spinlock)	((spinlock)->lock > 0)
+#else
+	typedef vint32 spinlock;
+
+#	define B_SPINLOCK_INITIALIZER 0
+#	define B_INITIALIZE_SPINLOCK(lock)	do { *(lock) = 0; } while (false)
+#	define B_SPINLOCK_IS_LOCKED(lock)	(*(lock) > 0)
+#endif
 
 /* interrupt handling support for device drivers */
 

headers/private/kernel/smp.h

@@ -40,6 +40,7 @@ extern "C" {
 
 status_t smp_init(struct kernel_args *args);
 status_t smp_per_cpu_init(struct kernel_args *args, int32 cpu);
+status_t smp_init_post_generic_syscalls(void);
 bool smp_trap_non_boot_cpus(int32 cpu);
 void smp_wake_up_non_boot_cpus(void);
 void smp_cpu_rendezvous(volatile uint32 *var, int current_cpu);

headers/private/system/spinlock_contention.h

@@ -0,0 +1,21 @@
+/*
+ * Copyright 2008, Ingo Weinhold, ingo_weinhold@gmx.de.
+ * Distributed under the terms of the MIT License.
+ */
+#ifndef _SYSTEM_SPINLOCK_CONTENTION_H
+#define _SYSTEM_SPINLOCK_CONTENTION_H
+
+#include <OS.h>
+
+
+#define SPINLOCK_CONTENTION				"spinlock contention"
+#define GET_SPINLOCK_CONTENTION_INFO	0x01
+
+
+typedef struct spinlock_contention_info {
+	uint64	thread_spinlock_counter;
+	uint64	team_spinlock_counter;
+} spinlock_contention_info;
+
+
+#endif	/* _SYSTEM_SPINLOCK_CONTENTION_H */

src/add-ons/kernel/bus_managers/ide/ide_sim.c

@@ -521,7 +521,7 @@ ide_sim_init_bus(device_node *node, void **cookie)
 
 	memset(bus, 0, sizeof(*bus));
 	bus->node = node;
-	bus->lock = 0;
+	B_INITIALIZE_SPINLOCK(&bus->lock);
 	bus->num_running_reqs = 0;
 	bus->active_qrequest = NULL;
 	bus->disconnected = false;

src/add-ons/kernel/bus_managers/pci/arch/x86/pci_controller.c

@@ -32,7 +32,7 @@
        restore_interrupts(cpu_status);		\
 }
 
-spinlock sConfigLock = 0;
+spinlock sConfigLock = B_SPINLOCK_INITIALIZER;
 
 static status_t
 pci_mech1_read_config(void *cookie, uint8 bus, uint8 device, uint8 function,

src/add-ons/kernel/bus_managers/ps2/packet_buffer.cpp

@@ -40,7 +40,7 @@ create_packet_buffer(size_t size)
 		free(buffer);
 		return NULL;
 	}
-	buffer->lock = 0;
+	B_INITIALIZE_SPINLOCK(&buffer->lock);
 
 	return buffer;
 }

src/add-ons/kernel/bus_managers/scsi/scsi_lock.h

@@ -29,7 +29,7 @@ typedef struct spinlock_irq {
 static inline void
 spinlock_irq_init(spinlock_irq *lock)
 {
-	lock->lock = 0;
+	B_INITIALIZE_SPINLOCK(&lock->lock);
 }
 
 static inline void

src/add-ons/kernel/busses/scsi/ahci/ahci_port.cpp

@@ -30,7 +30,6 @@ AHCIPort::AHCIPort(AHCIController *controller, int index)
 	, fIndex(index)
 	, fRegs(&controller->fRegs->port[index])
 	, fArea(-1)
-	, fSpinlock(0)
 	, fCommandsActive(0)
 	, fRequestSem(-1)
 	, fResponseSem(-1)
@@ -42,6 +41,7 @@ AHCIPort::AHCIPort(AHCIController *controller, int index)
 	, fResetPort(false)
 	, fError(false)
 {
+	B_INITIALIZE_SPINLOCK(&fSpinlock);
 	fRequestSem = create_sem(1, "ahci request");
 	fResponseSem = create_sem(0, "ahci response");
 }

src/add-ons/kernel/busses/usb/ehci.cpp

@@ -782,7 +782,7 @@ EHCI::InterruptHandler(void *data)
 int32
 EHCI::Interrupt()
 {
-	static spinlock lock = 0;
+	static spinlock lock = B_SPINLOCK_INITIALIZER;
 	acquire_spinlock(&lock);
 
 	// check if any interrupt was generated

src/add-ons/kernel/busses/usb/ohci.cpp

@@ -694,7 +694,7 @@ OHCI::_InterruptHandler(void *data)
 int32
 OHCI::_Interrupt()
 {
-	static spinlock lock = 0;
+	static spinlock lock = B_SPINLOCK_INITIALIZER;
 	acquire_spinlock(&lock);
 
 	uint32 status = 0;

src/add-ons/kernel/busses/usb/uhci.cpp

@@ -1531,7 +1531,7 @@ UHCI::InterruptHandler(void *data)
 int32
 UHCI::Interrupt()
 {
-	static spinlock lock = 0;
+	static spinlock lock = B_SPINLOCK_INITIALIZER;
 	acquire_spinlock(&lock);
 
 	// Check if we really had an interrupt

src/add-ons/kernel/drivers/audio/ac97/auich/util.c

@@ -34,7 +34,7 @@
 #include "debug.h"
 #include "util.h"
 
-spinlock slock = 0;
+spinlock slock = B_SPINLOCK_INITIALIZER;
 
 uint32 round_to_pagesize(uint32 size);
 

src/add-ons/kernel/drivers/audio/ac97/auvia/util.c

@@ -34,7 +34,7 @@
 #include "debug.h"
 #include "util.h"
 
-spinlock slock = 0;
+spinlock slock = B_SPINLOCK_INITIALIZER;
 
 uint32 round_to_pagesize(uint32 size);
 

src/add-ons/kernel/drivers/audio/emuxki/util.c

@@ -34,7 +34,7 @@
 #include "debug.h"
 #include "util.h"
 
-spinlock slock = 0;
+spinlock slock = B_SPINLOCK_INITIALIZER;
 
 uint32 round_to_pagesize(uint32 size);
 

src/add-ons/kernel/drivers/graphics/radeon/irq.c

@@ -237,7 +237,7 @@ Radeon_SetupIRQ(device_info *di, char *buffer)
 		goto err3;
 	}
 
-	di->cap_spinlock = 0;
+	B_INITIALIZE_SPINLOCK(&di->cap_spinlock);
 
 	sprintf(buffer, "%04X_%04X_%02X%02X%02X DMA I",
 		di->pcii.vendor_id, di->pcii.device_id,

src/add-ons/kernel/drivers/network/rtl8169/device.c

@@ -467,7 +467,7 @@ rtl8169_open(const char *name, uint32 flags, void** cookie)
 	read_settings(device);
 
 	device->rxBuf = (void **)malloc(sizeof(void *) * device->rxBufferCount);
-	device->rxSpinlock = 0;
+	B_INITIALIZE_SPINLOCK(&device->rxSpinlock);
 	device->rxNextIndex = 0;
 	device->rxIntIndex = 0;
 	device->rxFree = device->rxBufferCount;
@@ -475,7 +475,7 @@ rtl8169_open(const char *name, uint32 flags, void** cookie)
 	set_sem_owner(device->rxReadySem, B_SYSTEM_TEAM);
 	
 	device->txBuf = (void **)malloc(sizeof(void *) * device->txBufferCount);
-	device->txSpinlock = 0;
+	B_INITIALIZE_SPINLOCK(&device->txSpinlock);
 	device->txNextIndex = 0;
 	device->txIntIndex = 0;
 	device->txUsed = 0;

src/add-ons/kernel/drivers/network/rtl8169/timer.c

@@ -187,7 +187,7 @@ initialize_timer(void)
 {
 	sTimerCount = 0;
 	sTimerNextId = 1;
-	sTimerSpinlock = 0;
+	B_INITIALIZE_SPINLOCK(&sTimerSpinlock);
 	
 	sTimerThread = spawn_kernel_thread(timer_thread, "rtl8169 timer", 80, 0);
 	sTimerSem = create_sem(0, "rtl8169 timer");

src/add-ons/kernel/generic/dpc/dpc.c

@@ -91,7 +91,7 @@ new_dpc_queue(void **handle, const char *name, int32 priority)
 	queue->head = queue->tail = 0;
 	queue->size = DPC_QUEUE_SIZE;
 	queue->count = 0;
-	queue->lock = 0;	// Init the spinlock
+	B_INITIALIZE_SPINLOCK(&queue->lock);	// Init the spinlock
 
 #ifdef __HAIKU__
 	snprintf(str, sizeof(str), "%.*s_wakeup_sem", 

src/add-ons/kernel/generic/mpu401/mpu401.c

@@ -480,7 +480,7 @@ _EXPORT generic_mpu401_module *modules[] =
 	NULL
 };
 
-spinlock locked = 0;
+spinlock locked = B_SPINLOCK_INITIALIZER;
 cpu_status 
 lock(void)
 {

src/libs/compat/freebsd_network/taskqueue.c

@@ -20,7 +20,7 @@ struct taskqueue {
 	taskqueue_enqueue_fn tq_enqueue;
 	void *tq_arg;
 	int tq_fast;
-	int32 tq_spinlock;
+	spinlock tq_spinlock;
 	sem_id tq_sem;
 	thread_id *tq_threads;
 	thread_id tq_thread_storage;
@@ -43,7 +43,7 @@ _taskqueue_create(const char *name, int mflags, int fast,
 	tq->tq_fast = fast;
 
 	if (fast) {
-		tq->tq_spinlock = 0;
+		B_INITIALIZE_SPINLOCK(&tq->tq_spinlock);
 	} else {
 		mutex_init_etc(&tq->tq_mutex, name, MUTEX_FLAG_CLONE_NAME);
 	}

src/system/kernel/arch/m68k/arch_vm_translation_map_impl.cpp

@@ -1180,7 +1180,7 @@ arch_vm_translation_map_init(kernel_args *args)
 
 	sQueryDesc.type = DT_INVALID;
 
-	tmap_list_lock = 0;
+	B_INITIALIZE_SPINLOCK(&tmap_list_lock);
 	tmap_list = NULL;
 
 	// allocate some space to hold physical page mapping info

src/system/kernel/arch/x86/arch_debug_console.c

@@ -99,7 +99,7 @@ static bool sKeyboardHandlerInstalled = false;
 static bool sBochsOutput = false;
 #endif
 
-static spinlock sSerialOutputSpinlock = 0;
+static spinlock sSerialOutputSpinlock = B_SPINLOCK_INITIALIZER;
 
 
 static void

src/system/kernel/arch/x86/arch_vm_translation_map.cpp

@@ -852,7 +852,7 @@ arch_vm_translation_map_init(kernel_args *args)
 	sKernelPhysicalPageDirectory = (page_directory_entry *)args->arch_args.phys_pgdir;
 	sKernelVirtualPageDirectory = (page_directory_entry *)args->arch_args.vir_pgdir;
 
-	tmap_list_lock = 0;
+	B_INITIALIZE_SPINLOCK(&tmap_list_lock);
 	tmap_list = NULL;
 
 	// allocate some space to hold physical page mapping info

src/system/kernel/debug/debug.cpp

@@ -54,7 +54,7 @@ static bool sBlueScreenEnabled = false;
 	// must always be false on startup
 static bool sDebugScreenEnabled = false;
 static bool sBlueScreenOutput = true;
-static spinlock sSpinlock = 0;
+static spinlock sSpinlock = B_SPINLOCK_INITIALIZER;
 static int32 sDebuggerOnCPU = -1;
 
 static sem_id sSyslogNotify = -1;

src/system/kernel/debug/debug_commands.cpp

@@ -30,7 +30,7 @@
 static const int32 kMaxInvokeCommandDepth = 5;
 static const int32 kOutputBufferSize = 1024;
 
-static spinlock sSpinlock = 0;
+static spinlock sSpinlock = B_SPINLOCK_INITIALIZER;
 
 static struct debugger_command *sCommands;
 

src/system/kernel/debug/user_debugger.cpp

@@ -219,7 +219,7 @@ clear_team_debug_info(struct team_debug_info *info, bool initLock)
 		info->debugger_write_lock = -1;
 
 		if (initLock)
-			info->lock = 0;
+			B_INITIALIZE_SPINLOCK(&info->lock);
 	}
 }
 

src/system/kernel/int.c

@@ -62,7 +62,7 @@ dump_int_statistics(int argc, char **argv)
 	for (i = 0; i < NUM_IO_VECTORS; i++) {
 		struct io_handler *io;
 
-		if (sVectors[i].vector_lock == 0
+		if (!B_SPINLOCK_IS_LOCKED(&sVectors[i].vector_lock)
 			&& sVectors[i].enable_count == 0
 			&& sVectors[i].handled_count == 0
 			&& sVectors[i].unhandled_count == 0
@@ -72,7 +72,7 @@ dump_int_statistics(int argc, char **argv)
 		kprintf("int %3d, enabled %ld, handled %8lld, unhandled %8lld%s%s\n",
 			i, sVectors[i].enable_count, sVectors[i].handled_count, 
 			sVectors[i].unhandled_count,
-			sVectors[i].vector_lock != 0 ? ", ACTIVE" : "",
+			B_SPINLOCK_IS_LOCKED(&sVectors[i].vector_lock) ? ", ACTIVE" : "",
 			sVectors[i].handler_list.next == &sVectors[i].handler_list
 				? ", no handler" : "");
 
@@ -114,7 +114,7 @@ int_init_post_vm(kernel_args *args)
 
 	/* initialize the vector list */
 	for (i = 0; i < NUM_IO_VECTORS; i++) {
-		sVectors[i].vector_lock = 0;			/* initialize spinlock */
+		B_INITIALIZE_SPINLOCK(&sVectors[i].vector_lock);
 		sVectors[i].enable_count = 0;
 		sVectors[i].no_lock_vector = false;
 #ifdef DEBUG_INT

src/system/kernel/main.cpp

@@ -160,6 +160,7 @@ _start(kernel_args *bootKernelArgs, int currentCPU)
 		driver_settings_init_post_sem(&sKernelArgs);
 		TRACE("init generic syscall\n");
 		generic_syscall_init();
+		smp_init_post_generic_syscalls();
 		TRACE("init cbuf\n");
 		cbuf_init();
 		TRACE("init teams\n");

src/system/kernel/port.cpp

@@ -72,7 +72,7 @@ static bool sPortsActive = false;
 static port_id sNextPort = 1;
 static int32 sFirstFreeSlot = 1;
 
-static spinlock sPortSpinlock = 0;
+static spinlock sPortSpinlock = B_SPINLOCK_INITIALIZER;
 
 #define GRAB_PORT_LIST_LOCK() acquire_spinlock(&sPortSpinlock)
 #define RELEASE_PORT_LIST_LOCK() release_spinlock(&sPortSpinlock)

src/system/kernel/sem.cpp

@@ -108,7 +108,7 @@ static bool sSemsActive = false;
 static struct sem_entry	*sFreeSemsHead = NULL;
 static struct sem_entry	*sFreeSemsTail = NULL;
 
-static spinlock sem_spinlock = 0;
+static spinlock sem_spinlock = B_SPINLOCK_INITIALIZER;
 #define GRAB_SEM_LIST_LOCK()     acquire_spinlock(&sem_spinlock)
 #define RELEASE_SEM_LIST_LOCK()  release_spinlock(&sem_spinlock)
 #define GRAB_SEM_LOCK(s)         acquire_spinlock(&(s).lock)

src/system/kernel/smp.c

@@ -8,18 +8,22 @@
 
 /* Functionality for symetrical multi-processors */
 
-#include <thread.h>
-#include <int.h>
 #include <smp.h>
-#include <cpu.h>
-#include <arch/cpu.h>
-#include <arch/smp.h>
-#include <arch/int.h>
-#include <arch/debug.h>
 
 #include <stdlib.h>
 #include <string.h>
 
+#include <arch/cpu.h>
+#include <arch/debug.h>
+#include <arch/int.h>
+#include <arch/smp.h>
+#include <cpu.h>
+#include <generic_syscall.h>
+#include <int.h>
+#include <spinlock_contention.h>
+#include <thread.h>
+
+
 #define DEBUG_SPINLOCKS 1
 //#define TRACE_SMP
 
@@ -53,17 +57,17 @@ struct smp_msg {
 #define MAILBOX_LOCAL 1
 #define MAILBOX_BCAST 2
 
-static spinlock boot_cpu_spin[SMP_MAX_CPUS] = { 0, };
+static spinlock boot_cpu_spin[SMP_MAX_CPUS] = { };
 
 static struct smp_msg *free_msgs = NULL;
 static volatile int free_msg_count = 0;
-static spinlock free_msg_spinlock = 0;
+static spinlock free_msg_spinlock = B_SPINLOCK_INITIALIZER;
 
 static struct smp_msg *smp_msgs[SMP_MAX_CPUS] = { NULL, };
-static spinlock cpu_msg_spinlock[SMP_MAX_CPUS] = { 0, };
+static spinlock cpu_msg_spinlock[SMP_MAX_CPUS];
 
 static struct smp_msg *smp_broadcast_msgs = NULL;
-static spinlock broadcast_msg_spinlock = 0;
+static spinlock broadcast_msg_spinlock = B_SPINLOCK_INITIALIZER;
 
 static bool sICIEnabled = false;
 static int32 sNumCPUs = 1;
@@ -115,6 +119,10 @@ acquire_spinlock(spinlock *lock)
 		int currentCPU = smp_get_current_cpu();
 		if (are_interrupts_enabled())
 			panic("acquire_spinlock: attempt to acquire lock %p with interrupts enabled\n", lock);
+#if B_DEBUG_SPINLOCK_CONTENTION
+		while (atomic_add(&lock->lock, 1) != 0)
+			process_pending_ici(currentCPU);
+#else
 		while (1) {
 			while (*lock != 0) {
 				process_pending_ici(currentCPU);
@@ -123,6 +131,7 @@ acquire_spinlock(spinlock *lock)
 			if (atomic_set((int32 *)lock, 1) == 0)
 				break;
 		}
+#endif
 	} else {
 #if DEBUG_SPINLOCKS
 		int32 oldValue;
@@ -148,12 +157,17 @@ acquire_spinlock_nocheck(spinlock *lock)
 		if (are_interrupts_enabled())
 			panic("acquire_spinlock_nocheck: attempt to acquire lock %p with interrupts enabled\n", lock);
 #endif
+#if B_DEBUG_SPINLOCK_CONTENTION
+		while (atomic_add(&lock->lock, 1) != 0) {
+		}
+#else
 		while (1) {
 			while(*lock != 0)
 				PAUSE();
 			if (atomic_set((int32 *)lock, 1) == 0)
 				break;
 		}
+#endif
 	} else {
 #if DEBUG_SPINLOCKS
 		if (are_interrupts_enabled())
@@ -171,8 +185,23 @@ release_spinlock(spinlock *lock)
 	if (sNumCPUs > 1) {
 		if (are_interrupts_enabled())
 			panic("release_spinlock: attempt to release lock %p with interrupts enabled\n", lock);
+#if B_DEBUG_SPINLOCK_CONTENTION
+		{
+			int32 count = atomic_set(&lock->lock, 0) - 1;
+			if (count < 0) {
+				panic("release_spinlock: lock %p was already released\n", lock);
+			} else {
+				// add to the total count -- deal with carry manually
+				if ((uint32)atomic_add(&lock->count_low, count) + count
+						< (uint32)count) {
+					atomic_add(&lock->count_high, 1);
+				}
+			}
+		}
+#else
 		if (atomic_set((int32 *)lock, 0) != 1)
 			panic("release_spinlock: lock %p was already released\n", lock);
+#endif
 	} else {
 		#if DEBUG_SPINLOCKS
 			if (are_interrupts_enabled())
@@ -405,6 +434,48 @@ process_pending_ici(int32 currentCPU)
 }
 
 
+#if B_DEBUG_SPINLOCK_CONTENTION
+
+static uint64
+get_spinlock_counter(spinlock* lock)
+{
+	uint32 high;
+	uint32 low;
+	do {
+		high = (uint32)atomic_get(&lock->count_high);
+		low = (uint32)atomic_get(&lock->count_low);
+	} while (high != atomic_get(&lock->count_high));
+
+	return ((uint64)high << 32) | low;
+}
+
+
+static status_t
+spinlock_contention_syscall(const char* subsystem, uint32 function,
+	void* buffer, size_t bufferSize)
+{
+	spinlock_contention_info info;
+
+	if (function != GET_SPINLOCK_CONTENTION_INFO)
+		return B_BAD_VALUE;
+
+	if (bufferSize < sizeof(spinlock_contention_info))
+		return B_BAD_VALUE;
+
+	info.thread_spinlock_counter = get_spinlock_counter(&thread_spinlock);
+	info.team_spinlock_counter = get_spinlock_counter(&team_spinlock);
+
+	if (!IS_USER_ADDRESS(buffer)
+		|| user_memcpy(buffer, &info, sizeof(info)) != B_OK) {
+		return B_BAD_ADDRESS;
+	}
+
+	return B_OK;
+}
+
+#endif	// B_DEBUG_SPINLOCK_CONTENTION
+
+
 //	#pragma mark -
 
 
@@ -553,7 +624,11 @@ bool
 smp_trap_non_boot_cpus(int32 cpu)
 {
 	if (cpu > 0) {
+#if B_DEBUG_SPINLOCK_CONTENTION
+		boot_cpu_spin[cpu].lock = 1;
+#else
 		boot_cpu_spin[cpu] = 1;
+#endif
 		acquire_spinlock_nocheck(&boot_cpu_spin[cpu]);
 		return false;
 	}
@@ -624,6 +699,18 @@ smp_per_cpu_init(kernel_args *args, int32 cpu)
 }
 
 
+status_t
+smp_init_post_generic_syscalls(void)
+{
+#if B_DEBUG_SPINLOCK_CONTENTION
+	return register_generic_syscall(SPINLOCK_CONTENTION,
+		&spinlock_contention_syscall, 0, 0);
+#else
+	return B_OK;
+#endif
+}
+
+
 void
 smp_set_num_cpus(int32 numCPUs)
 {

src/system/kernel/team.cpp

@@ -89,7 +89,7 @@ static struct team *sKernelTeam = NULL;
 static int32 sMaxTeams = 2048;
 static int32 sUsedTeams = 1;
 
-spinlock team_spinlock = 0;
+spinlock team_spinlock = B_SPINLOCK_INITIALIZER;
 
 
 // #pragma mark - Tracing

src/system/kernel/thread.cpp

@@ -60,7 +60,7 @@ struct thread_key {
 };
 
 // global
-spinlock thread_spinlock = 0;
+spinlock thread_spinlock = B_SPINLOCK_INITIALIZER;
 
 // thread list
 static struct thread sIdleThreads[B_MAX_CPU_COUNT];