* For each userland team the kernel creates an area in the userland

address space that is fully locked and marked B_KERNEL_AREA. It can
  thus be accessed by the kernel without additional checks.
* For each userland thread we do create a user_thread structure in that
  area. The structure is accessible from userland via TLS, using the
  private get_user_thread() function.
* Introduced private userland functions [un]defer_signals(). They can be
  used to cheaply disable/re-enable signal delivery. They use the
  user_thread::defer_signals/pending_signals fields which are
  checked/updated by the kernel.


git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@25451 a95241bf-73f2-0310-859d-f6bbb57e9c96

headers/private/kernel/arch/x86/arch_kernel.h

@@ -32,7 +32,8 @@
 #define USER_SIZE     (KERNEL_BASE - 0x10000)
 #define USER_TOP      (USER_BASE + USER_SIZE)
 
-#define USER_STACK_REGION 0x70000000
-#define USER_STACK_REGION_SIZE (USER_TOP - USER_STACK_REGION)
+#define KERNEL_USER_DATA_BASE	0x6fff0000
+#define USER_STACK_REGION		0x70000000
+#define USER_STACK_REGION_SIZE	(USER_TOP - USER_STACK_REGION)
 
 #endif	/* _KERNEL_ARCH_x86_KERNEL_H */

headers/private/kernel/team.h

@@ -47,6 +47,9 @@ status_t start_watching_team(team_id team, void (*hook)(team_id, void *),
 status_t stop_watching_team(team_id team, void (*hook)(team_id, void *),
 			void *data);
 
+struct user_thread* team_allocate_user_thread(struct team* team);
+void team_free_user_thread(struct thread* thread);
+
 // used in syscalls.c
 thread_id _user_load_image(int32 argCount, const char **args, int32 envCount,
 			const char **env, int32 priority, uint32 flags,

headers/private/kernel/thread_types.h

@@ -71,6 +71,7 @@ enum {
 struct image;					// defined in image.c
 struct realtime_sem_context;	// defined in realtime_sem.cpp
 struct select_info;
+struct user_thread;				// defined in libroot/user_thread.h
 
 struct death_entry {
 	struct list_link	link;
@@ -165,6 +166,11 @@ struct team_dead_children : team_job_control_children {
 #endif	// __cplusplus
 
 
+struct free_user_thread {
+	struct free_user_thread*	next;
+	struct user_thread*			thread;
+};
+
 struct team {
 	struct team		*next;			// next in hash
 	struct team		*siblings_next;
@@ -199,6 +205,12 @@ struct team {
 	struct list		watcher_list;
 	struct arch_team arch_info;
 
+	addr_t			user_data;
+	area_id			user_data_area;
+	size_t			user_data_size;
+	size_t			used_user_data;
+	struct free_user_thread* free_user_threads;
+
 	struct team_debug_info debug_info;
 
 	bigtime_t		dead_threads_kernel_time;
@@ -239,13 +251,15 @@ struct thread {
 	size_t			signal_stack_size;
 	bool			signal_stack_enabled;
 
+	bool			in_kernel;
+	bool			was_yielded;
+
+	struct user_thread*	user_thread;
+
 	struct {
 		uint8		parameters[32];
 	} syscall_restart;
 
-	bool			in_kernel;
-	bool			was_yielded;
-
 	struct {
 		status_t	status;				// current wait status
 		uint32		flags;				// interrupable flags

headers/private/kernel/tls.h

@@ -17,6 +17,7 @@ enum {
 	TLS_THREAD_ID_SLOT,
 	TLS_ERRNO_SLOT,
 	TLS_ON_EXIT_THREAD_SLOT,
+	TLS_USER_THREAD_SLOT,
 
 	// Note: these entries can safely be changed between
 	// releases; 3rd party code always calls tls_allocate()

headers/private/libroot/user_thread.h

@@ -0,0 +1,48 @@
+/*
+ * Copyright 2008, Ingo Weinhold, [email protected].
+ * Distributed under the terms of the MIT License.
+ */
+#ifndef _LIBROOT_USER_THREAD_H
+#define _LIBROOT_USER_THREAD_H
+
+#include <OS.h>
+#include <TLS.h>
+
+#include <tls.h>		/* kernel header */
+
+
+struct user_thread {
+	int32			defer_signals;		// counter; 0 == signals allowed
+	uint32			pending_signals;	// signals that are pending, when
+										// signals are deferred
+	status_t		wait_status;
+};
+
+
+static inline user_thread*
+get_user_thread()
+{
+	return (user_thread*)tls_get(TLS_USER_THREAD_SLOT);
+}
+
+
+static void inline
+defer_signals()
+{
+	get_user_thread()->defer_signals++;
+}
+
+
+static void inline
+undefer_signals()
+{
+	user_thread* thread = get_user_thread();
+	if (--thread->defer_signals == 0 && thread->pending_signals != 0) {
+		// signals shall no longer be deferred -- call a dummy syscall to handle
+		// the pending ones
+		is_computer_on();
+	}
+}
+
+
+#endif	/* _LIBROOT_USER_THREAD_H */

src/system/kernel/Jamfile

@@ -10,6 +10,7 @@ SubDir HAIKU_TOP src system kernel ;
 	SubDirC++Flags $(defines) ;
 }
 
+UsePrivateHeaders libroot ;
 UsePrivateHeaders shared ;
 UsePrivateHeaders runtime_loader ;
 

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

@@ -309,15 +309,17 @@ arch_thread_init_kthread_stack(struct thread *t, int (*start_func)(void),
 status_t
 arch_thread_init_tls(struct thread *thread)
 {
-	uint32 tls[TLS_THREAD_ID_SLOT + 1];
+	uint32 tls[TLS_USER_THREAD_SLOT + 1];
 	int32 i;
 
 	thread->user_local_storage = thread->user_stack_base
 		+ thread->user_stack_size;
 
 	// initialize default TLS fields
+	memset(tls, 0, sizeof(tls));
 	tls[TLS_BASE_ADDRESS_SLOT] = thread->user_local_storage;
 	tls[TLS_THREAD_ID_SLOT] = thread->id;
+	tls[TLS_USER_THREAD_SLOT] = (addr_t)thread->user_thread;
 
 	return user_memcpy((void *)thread->user_local_storage, tls, sizeof(tls));
 }

src/system/kernel/signal.cpp

@@ -24,6 +24,7 @@
 #include <thread.h>
 #include <tracing.h>
 #include <user_debugger.h>
+#include <user_thread.h>
 #include <util/AutoLock.h>
 
 
@@ -42,6 +43,11 @@
 #define DEFAULT_IGNORE_SIGNALS \
 	(SIGNAL_TO_MASK(SIGCHLD) | SIGNAL_TO_MASK(SIGWINCH) \
 	| SIGNAL_TO_MASK(SIGCONT))
+#define NON_DEFERRABLE_SIGNALS	\
+	(KILL_SIGNALS				\
+	| SIGNAL_TO_MASK(SIGILL)	\
+	| SIGNAL_TO_MASK(SIGFPE)	\
+	| SIGNAL_TO_MASK(SIGSEGV))
 
 
 const char * const sigstr[NSIG] = {
@@ -270,6 +276,14 @@ handle_signals(struct thread *thread)
 	if (signalMask == 0)
 		return 0;
 
+	if (thread->user_thread->defer_signals > 0
+		&& (signalMask & NON_DEFERRABLE_SIGNALS) == 0) {
+		thread->user_thread->pending_signals = signalMask;
+		return 0;
+	}
+
+	thread->user_thread->pending_signals = 0;
+
 	bool restart = (atomic_and(&thread->flags,
 			~THREAD_FLAGS_DONT_RESTART_SYSCALL)
 		& THREAD_FLAGS_DONT_RESTART_SYSCALL) == 0;

src/system/kernel/team.cpp

@@ -1,4 +1,5 @@
 /*
+ * Copyright 2008, Ingo Weinhold, [email protected].
  * Copyright 2002-2008, Axel Dörfler, [email protected].
  * Distributed under the terms of the MIT License.
  *
@@ -35,6 +36,7 @@
 #include <tls.h>
 #include <tracing.h>
 #include <user_runtime.h>
+#include <user_thread.h>
 #include <usergroup.h>
 #include <vfs.h>
 #include <vm.h>
@@ -69,6 +71,7 @@ struct fork_arg {
 	size_t		user_stack_size;
 	addr_t		user_local_storage;
 	sigset_t	sig_block_mask;
+	struct user_thread* user_thread;
 
 	struct arch_fork_arg arch_info;
 };
@@ -830,6 +833,11 @@ create_team_struct(const char *name, bool kernel)
 	team->state = TEAM_STATE_BIRTH;
 	team->flags = 0;
 	team->death_sem = -1;
+	team->user_data_area = -1;
+	team->user_data = 0;
+	team->used_user_data = 0;
+	team->user_data_size = 0;
+	team->free_user_threads = NULL;
 
 	team->supplementary_groups = NULL;
 	team->supplementary_group_count = 0;
@@ -911,6 +919,11 @@ delete_team_struct(struct team *team)
 	while (job_control_entry* entry = team->dead_children->entries.RemoveHead())
 		delete entry;
 
+	while (free_user_thread* entry = team->free_user_threads) {
+		team->free_user_threads = entry->next;
+		free(entry);
+	}
+
 	malloc_referenced_release(team->supplementary_groups);
 
 	delete team->job_control_entry;
@@ -935,6 +948,42 @@ get_arguments_data_size(char **args, int32 argc)
 }
 
 
+static status_t
+create_team_user_data(struct team* team)
+{
+	void* address = (void*)KERNEL_USER_DATA_BASE;
+	size_t size = 4 * B_PAGE_SIZE;
+	team->user_data_area = create_area_etc(team, "user area", &address,
+		B_BASE_ADDRESS, size, B_FULL_LOCK, B_READ_AREA | B_WRITE_AREA);
+	if (team->user_data_area < 0)
+		return team->user_data_area;
+
+	team->user_data = (addr_t)address;
+	team->used_user_data = 0;
+	team->user_data_size = size;
+	team->free_user_threads = NULL;
+
+	return B_OK;
+}
+
+
+static void
+delete_team_user_data(struct team* team)
+{
+	if (team->user_data_area >= 0) {
+		delete_area_etc(team, team->user_data_area);
+		team->user_data = 0;
+		team->used_user_data = 0;
+		team->user_data_size = 0;
+		team->user_data_area = -1;
+		while (free_user_thread* entry = team->free_user_threads) {
+			team->free_user_threads = entry->next;
+			free(entry);
+		}
+	}
+}
+
+
 static void
 free_team_arg(struct team_arg *teamArg)
 {
@@ -1004,6 +1053,9 @@ team_create_thread_start(void *args)
 
 	TRACE(("team_create_thread_start: entry thread %ld\n", t->id));
 
+	// get a user thread for the main thread
+	t->user_thread = team_allocate_user_thread(team);
+
 	// create an initial primary stack area
 
 	// Main stack area layout is currently as follows (starting from 0):
@@ -1216,13 +1268,18 @@ load_image_etc(int32 argCount, char * const *args, int32 envCount,
 	else
 		threadName = args[0];
 
+	// create the user data area
+	status = create_team_user_data(team);
+	if (status != B_OK)
+		goto err4;
+
 	// Create a kernel thread, but under the context of the new team
 	// The new thread will take over ownership of teamArgs
 	thread = spawn_kernel_thread_etc(team_create_thread_start, threadName,
 		B_NORMAL_PRIORITY, teamArgs, team->id, team->id);
 	if (thread < 0) {
 		status = thread;
-		goto err4;
+		goto err5;
 	}
 
 	// wait for the loader of the new team to finish its work
@@ -1264,6 +1321,8 @@ load_image_etc(int32 argCount, char * const *args, int32 envCount,
 
 	return thread;
 
+err5:
+	delete_team_user_data(team);
 err4:
 	vm_put_address_space(team->address_space);
 err3:
@@ -1362,6 +1421,7 @@ exec_team(const char *path, int32 argCount, char * const *args,
 
 	user_debug_prepare_for_exec();
 
+	delete_team_user_data(team);
 	vm_delete_areas(team->address_space);
 	delete_owned_ports(team->id);
 	sem_delete_owned_sems(team->id);
@@ -1370,6 +1430,14 @@ exec_team(const char *path, int32 argCount, char * const *args,
 	delete_realtime_sem_context(team->realtime_sem_context);
 	team->realtime_sem_context = NULL;
 
+	status = create_team_user_data(team);
+	if (status != B_OK) {
+		// creating the user data failed -- we're toast
+		// TODO: We should better keep the old user area in the first place.
+		exit_thread(status);
+		return status;
+	}
+
 	user_debug_finish_after_exec();
 
 	// rename the team
@@ -1422,6 +1490,7 @@ fork_team_thread_start(void *_args)
 	thread->user_stack_size = forkArgs->user_stack_size;
 	thread->user_local_storage = forkArgs->user_local_storage;
 	thread->sig_block_mask = forkArgs->sig_block_mask;
+	thread->user_thread = forkArgs->user_thread;
 
 	arch_thread_init_tls(thread);
 
@@ -1509,6 +1578,8 @@ fork_team(void)
 	// ToDo: should be able to handle stack areas differently (ie. don't have them copy-on-write)
 	// ToDo: all stacks of other threads than the current one could be left out
 
+	forkArgs->user_thread = NULL;
+
 	cookie = 0;
 	while (get_next_area_info(B_CURRENT_TEAM, &cookie, &info) == B_OK) {
 		void *address;
@@ -1519,13 +1590,30 @@ fork_team(void)
 			break;
 		}
 
-		if (info.area == parentThread->user_stack_area)
+		if (info.area == parentThread->user_stack_area) {
 			forkArgs->user_stack_area = area;
+		} else if (info.area == parentTeam->user_data_area) {
+			team->user_data = (addr_t)address;
+			team->used_user_data = 0;
+			team->user_data_size = info.size;
+			team->user_data_area = area;
+			team->free_user_threads = NULL;
+			forkArgs->user_thread = team_allocate_user_thread(team);
+		}
 	}
 
 	if (status < B_OK)
 		goto err4;
 
+	if (forkArgs->user_thread == NULL) {
+#if KDEBUG
+		panic("user data area not found, parent area is %ld",
+			parentTeam->user_data_area);
+#endif
+		status = B_ERROR;
+		goto err4;
+	}
+
 	forkArgs->user_stack_base = parentThread->user_stack_base;
 	forkArgs->user_stack_size = parentThread->user_stack_size;
 	forkArgs->user_local_storage = parentThread->user_local_storage;
@@ -2558,6 +2646,68 @@ stop_watching_team(team_id teamID, void (*hook)(team_id, void *), void *data)
 }
 
 
+/*!	The team lock must be held or the team must still be single threaded.
+*/
+struct user_thread*
+team_allocate_user_thread(struct team* team)
+{
+	if (team->user_data == 0)
+		return NULL;
+
+	user_thread* thread = NULL;
+
+	// take an entry from the free list, if any
+	if (struct free_user_thread* entry = team->free_user_threads) {
+		thread = entry->thread;
+		team->free_user_threads = entry->next;
+		deferred_free(entry);
+		return thread;
+	} else {
+		// enough space left?
+		size_t needed = _ALIGN(sizeof(user_thread));
+		if (team->user_data_size - team->used_user_data < needed)
+			return NULL;
+		// TODO: This imposes a per team thread limit! We should resize the
+		// area, if necessary. That's problematic at this point, though, since
+		// we've got the team lock.
+
+		thread = (user_thread*)(team->user_data + team->used_user_data);
+		team->used_user_data += needed;
+	}
+
+	thread->defer_signals = 0;
+	thread->pending_signals = 0;
+	thread->wait_status = B_OK;
+
+	return thread;
+}
+
+
+/*!	The team lock must not be held. \a thread must be the current thread.
+*/
+void
+team_free_user_thread(struct thread* thread)
+{
+	user_thread* userThread = thread->user_thread;
+	if (userThread == NULL)
+		return;
+
+	// create a free list entry
+	free_user_thread* entry
+		= (free_user_thread*)malloc(sizeof(free_user_thread));
+	if (entry == NULL) {
+		// we have to leak the user thread :-/
+		return;
+	}
+
+	InterruptsSpinLocker _(team_spinlock);
+
+	entry->thread = userThread;
+	entry->next = thread->team->free_user_threads;
+	thread->team->free_user_threads = entry;
+}
+
+
 //	#pragma mark - Public kernel API
 
 

src/system/kernel/thread.cpp

@@ -444,7 +444,16 @@ create_thread(thread_creation_attributes& attributes, bool kernel)
 	GRAB_TEAM_LOCK();
 	// look at the team, make sure it's not being deleted
 	team = team_get_team_struct_locked(attributes.team);
-	if (team != NULL && team->state != TEAM_STATE_DEATH) {
+
+	if (team == NULL || team->state == TEAM_STATE_DEATH)
+		abort = true;
+
+	if (!abort && !kernel) {
+		thread->user_thread = team_allocate_user_thread(team);
+		abort = thread->user_thread == NULL;
+	}
+
+	if (!abort) {
 		// Debug the new thread, if the parent thread required that (see above),
 		// or the respective global team debug flag is set. But only, if a
 		// debugger is installed for the team.
@@ -457,8 +466,7 @@ create_thread(thread_creation_attributes& attributes, bool kernel)
 		}
 
 		insert_thread_into_team(team, thread);
-	} else
-		abort = true;
+	}
 
 	RELEASE_TEAM_LOCK();
 	if (abort) {
@@ -1407,8 +1415,10 @@ thread_exit(void)
 		if (team->main_thread == thread) {
 			// this was the main thread in this team, so we will delete that as well
 			deleteTeam = true;
-		} else
+		} else {
 			threadDeathEntry = (death_entry*)malloc(sizeof(death_entry));
+			team_free_user_thread(thread);
+		}
 
 		// remove this thread from the current team and add it to the kernel
 		// put the thread into the kernel team until it dies
@@ -2937,4 +2947,3 @@ _user_setrlimit(int resource, const struct rlimit *userResourceLimit)
 
 	return common_setrlimit(resource, &resourceLimit);
 }
-