Update.

	* elf/rtld.c (_dl_start_final): Allocate TLS and initialize
	thread-pointer as soon as possible.
	* sysdeps/generic/ldsodefs.h: Include <tls.h>.  Define first TLS
	elements in rtld_global.
	* sysdeps/generic/tls.h: New file.
	* elf/Makefile (distribute): Add tls.h.
	* sysdeps/i386/dl-machine.h (elf_machine_rel): Add support for TLS
	relocations.  Not complete yet.

	* resolv/resolv.h: Allow user to define __need_res_state and only
	define __res_start structure then.
	* include/resolv.h: Only declare functions if _RESOLV_H_ is defined.

22 files changed, 513 insertions, 197 deletions

diff --git a/ChangeLog b/ChangeLog
index f6c2bac197..4931ba52c7 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,18 @@
 2002-02-04  Ulrich Drepper  <drepper@redhat.com>
 
+	* elf/rtld.c (_dl_start_final): Allocate TLS and initialize
+	thread-pointer as soon as possible.
+	* sysdeps/generic/ldsodefs.h: Include <tls.h>.  Define first TLS
+	elements in rtld_global.
+	* sysdeps/generic/tls.h: New file.
+	* elf/Makefile (distribute): Add tls.h.
+	* sysdeps/i386/dl-machine.h (elf_machine_rel): Add support for TLS
+	relocations.  Not complete yet.
+
+	* resolv/resolv.h: Allow user to define __need_res_state and only
+	define __res_start structure then.
+	* include/resolv.h: Only declare functions if _RESOLV_H_ is defined.
+
 	* sysdeps/generic/dl-sysdep.c (_dl_sysdep_start): Move
 	dl_cpuclock_offset initialization to _dl_start_final.
 	(_dl_show_auxv): Avoid unnecessary sign extension.
diff --git a/elf/Makefile b/elf/Makefile
index 12277741c4..e84c6d15cc 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -67,7 +67,7 @@ distribute	:= $(rtld-routines:=.c) dynamic-link.h do-rel.h dl-machine.h \
 		   reldep6mod4.c reldep6mod3.c reldep6mod2.c reldep6mod1.c \
 		   reldep6mod0.c \
 		   unwind-dw2.c unwind-dw2-fde.c unwind.h unwind-pe.h \
-		   unwind-dw2-fde.h dwarf2.h dl-procinfo.c
+		   unwind-dw2-fde.h dwarf2.h dl-procinfo.c tls.h
 
 include ../Makeconfig
 
diff --git a/elf/rtld.c b/elf/rtld.c
index e2360e3ae1..b7439b2e7b 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -218,6 +218,16 @@ _dl_start_final (void *arg, struct link_map *bootstrap_map_p,
      which use `alloca'.  */
   ElfW(Addr) *start_addr = alloca (sizeof (ElfW(Addr)));
   extern char _begin[], _end[];
+#ifdef USE_TLS
+  ElfW(Ehdr) *ehdr;
+  ElfW(Phdr) *phdr;
+  size_t cnt;
+  size_t tlssize = 0;
+  size_t tlsimagesize = 0;
+  const void *tlsimage = NULL;
+  void *tlsblock = NULL;
+  dtv_t initdtv[2];
+#endif
 
   if (HP_TIMING_AVAIL)
     {
@@ -244,11 +254,76 @@ _dl_start_final (void *arg, struct link_map *bootstrap_map_p,
   HP_TIMING_NOW (GL(dl_cpuclock_offset));
 #endif
 
+#if USE_TLS
+  /* Get the dynamic linkers program header.  */
+  ehdr = (ElfW(Ehdr) *) bootstrap_map_p->l_addr;
+  phdr = (ElfW(Phdr) *) (bootstrap_map_p->l_addr + ehdr->e_phoff);
+  for (cnt = 0; cnt < ehdr->e_phnum; ++cnt)
+    if (phdr[cnt].p_type == PT_TLS)
+      {
+	size_t align = MAX (TLS_INIT_TCB_ALIGN,  phdr[cnt].p_align);
+
+	tlssize = phdr[cnt].p_memsz;
+	tlsimagesize = phdr[cnt].p_filesz;
+	tlsimage = (void *) (bootstrap_map_p->l_addr + phdr[cnt].p_offset);
+
+	/* We can now allocate the initial TLS block.  This can happen
+	   on the stack.  We'll get the final memory later when we
+	   know all about the various objects loaded at startup
+	   time.  */
+# if TLS_TCB_AT_TP
+	tlsblock = alloca (roundup (tlssize, TLS_INIT_TCB_ALIGN)
+			   + TLS_INIT_TCB_SIZE
+			   + align);
+# elif TLS_DTV_AT_TP
+	tlsblock = alloca (roundup (TLS_INIT_TCB_SIZE, phdr[cnt].p_align)
+			   + tlssize
+			   + align);
+# else
+	/* In case a model with a different layout for the TCB and DTV
+	   is defined add another #elif here and in the following #ifs.  */
+#  error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
+# endif
+	/* Align the TLS block.  */
+	tlsblock = (void *) (((uintptr_t) tlsblock + align - 1)
+			     & ~(align - 1));
+
+	/* Initialize the dtv.  */
+	initdtv[0].counter = 1;
+
+	/* Initialize the TLS block.  */
+# if TLS_TCB_AT_TP
+	initdtv[1].pointer = tlsblock;
+# elif TLS_DTV_AT_TP
+	GL(rtld_tlsoffset) = roundup (TLS_INIT_TCB_SIZE, phdr[cnt].p_align);
+	initdtv[1].pointer = (char *) tlsblock + GL(rtld_tlsoffset);
+# else
+#  error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
+# endif
+	memset (__mempcpy (initdtv[1].pointer, tlsimage, tlsimagesize),
+		'\0', tlssize - tlsimagesize);
+
+	/* Initialize the thread pointer.  */
+# if TLS_TCB_AT_TP
+	GL(rtld_tlsoffset) = roundup (tlssize, TLS_INIT_TCB_ALIGN);
+	TLS_INIT_TP ((char *) tlsblock + GL(rtld_tlsoffset), initdtv);
+# elif TLS_DTV_AT_TP
+	TLS_INIT_TP (tlsblock, intidtv);
+# else
+#  error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined"
+# endif
+
+	/* There can only be one PT_TLS entry.  */
+	break;
+      }
+#endif	/* use TLS */
+
   /* Call the OS-dependent function to set up life so we can do things like
      file access.  It will call `dl_main' (below) to do all the real work
      of the dynamic linker, and then unwind our frame and run the user
      entry point on the same stack we entered on.  */
   *start_addr =  _dl_sysdep_start (arg, &dl_main);
+
 #ifndef HP_TIMING_NONAVAIL
   if (HP_TIMING_AVAIL)
     {
diff --git a/include/resolv.h b/include/resolv.h
index fe79c16ba3..d4a531f244 100644
--- a/include/resolv.h
+++ b/include/resolv.h
@@ -10,6 +10,7 @@
 
 #include <resolv/resolv.h>
 
+#ifdef _RESOLV_H_
 /* Now define the internal interfaces.  */
 extern int __res_vinit (res_state, int);
 extern void _sethtent (int);
@@ -26,5 +27,6 @@ extern void res_send_setqhook (res_send_qhook __hook);
 extern void res_send_setrhook (res_send_rhook __hook);
 extern int res_ourserver_p (const res_state __statp,
 			    const struct sockaddr_in6 *__inp);
+#endif
 
 #endif
diff --git a/linuxthreads/ChangeLog b/linuxthreads/ChangeLog
index 6f9c93687b..f55477ea41 100644
--- a/linuxthreads/ChangeLog
+++ b/linuxthreads/ChangeLog
@@ -1,3 +1,9 @@
+2002-02-04  Ulrich Drepper  <drepper@redhat.com>
+
+	* internals.h: Move thread descriptor definition...
+	* descr.h.: ...here.  New file.
+	* sysdeps/i386/tls.h: New file.
+
 2002-02-01  H.J. Lu  <hjl@gnu.org>
 
 	* sysdeps/mips/pspinlock.c (__pthread_spin_lock): Use a
diff --git a/linuxthreads/descr.h b/linuxthreads/descr.h
new file mode 100644
index 0000000000..4cfad89dcb
--- /dev/null
+++ b/linuxthreads/descr.h
@@ -0,0 +1,168 @@
+/* Linuxthreads - a simple clone()-based implementation of Posix        */
+/* threads for Linux.                                                   */
+/* Copyright (C) 1996 Xavier Leroy (Xavier.Leroy@inria.fr)              */
+/*                                                                      */
+/* This program is free software; you can redistribute it and/or        */
+/* modify it under the terms of the GNU Library General Public License  */
+/* as published by the Free Software Foundation; either version 2       */
+/* of the License, or (at your option) any later version.               */
+/*                                                                      */
+/* This program is distributed in the hope that it will be useful,      */
+/* but WITHOUT ANY WARRANTY; without even the implied warranty of       */
+/* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the        */
+/* GNU Library General Public License for more details.                 */
+
+#ifndef _DESCR_H
+#define _DESCR_H	1
+
+#define __need_res_state
+#include <resolv.h>
+#include <sched.h>
+#include <setjmp.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <hp-timing.h>
+#include <bits/libc-tsd.h> /* for _LIBC_TSD_KEY_N */
+
+
+/* The type of thread descriptors */
+typedef struct _pthread_descr_struct *pthread_descr;
+
+
+/* Some more includes.  */
+#include <pt-machine.h>
+#include <linuxthreads_db/thread_dbP.h>
+
+
+/* Arguments passed to thread creation routine */
+struct pthread_start_args {
+  void *(*start_routine)(void *); /* function to run */
+  void *arg;                      /* its argument */
+  sigset_t mask;                  /* initial signal mask for thread */
+  int schedpolicy;                /* initial scheduling policy (if any) */
+  struct sched_param schedparam;  /* initial scheduling parameters (if any) */
+};
+
+
+/* Callback interface for removing the thread from waiting on an
+   object if it is cancelled while waiting or about to wait.
+   This hold a pointer to the object, and a pointer to a function
+   which ``extricates'' the thread from its enqueued state.
+   The function takes two arguments: pointer to the wait object,
+   and a pointer to the thread. It returns 1 if an extrication
+   actually occured, and hence the thread must also be signalled.
+   It returns 0 if the thread had already been extricated. */
+typedef struct _pthread_extricate_struct {
+    void *pu_object;
+    int (*pu_extricate_func)(void *, pthread_descr);
+} pthread_extricate_if;
+
+
+/* Atomic counter made possible by compare_and_swap */
+struct pthread_atomic {
+  long p_count;
+  int p_spinlock;
+};
+
+
+/* Context info for read write locks. The pthread_rwlock_info structure
+   is information about a lock that has been read-locked by the thread
+   in whose list this structure appears. The pthread_rwlock_context
+   is embedded in the thread context and contains a pointer to the
+   head of the list of lock info structures, as well as a count of
+   read locks that are untracked, because no info structure could be
+   allocated for them. */
+struct _pthread_rwlock_t;
+typedef struct _pthread_rwlock_info {
+  struct _pthread_rwlock_info *pr_next;
+  struct _pthread_rwlock_t *pr_lock;
+  int pr_lock_count;
+} pthread_readlock_info;
+
+
+/* We keep thread specific data in a special data structure, a two-level
+   array.  The top-level array contains pointers to dynamically allocated
+   arrays of a certain number of data pointers.  So we can implement a
+   sparse array.  Each dynamic second-level array has
+	PTHREAD_KEY_2NDLEVEL_SIZE
+   entries.  This value shouldn't be too large.  */
+#define PTHREAD_KEY_2NDLEVEL_SIZE	32
+
+/* We need to address PTHREAD_KEYS_MAX key with PTHREAD_KEY_2NDLEVEL_SIZE
+   keys in each subarray.  */
+#define PTHREAD_KEY_1STLEVEL_SIZE \
+  ((PTHREAD_KEYS_MAX + PTHREAD_KEY_2NDLEVEL_SIZE - 1) \
+   / PTHREAD_KEY_2NDLEVEL_SIZE)
+
+
+union dtv;
+
+
+struct _pthread_descr_struct {
+  /* XXX Remove this union for IA-64 style TLS module */
+  union {
+    struct {
+      pthread_descr self;	/* Pointer to this structure */
+      union dtv *dtvp;
+    } data;
+    void *__padding[16];
+  } p_header;
+  pthread_descr p_nextlive, p_prevlive;
+                                /* Double chaining of active threads */
+  pthread_descr p_nextwaiting;  /* Next element in the queue holding the thr */
+  pthread_descr p_nextlock;	/* can be on a queue and waiting on a lock */
+  pthread_t p_tid;              /* Thread identifier */
+  int p_pid;                    /* PID of Unix process */
+  int p_priority;               /* Thread priority (== 0 if not realtime) */
+  struct _pthread_fastlock * p_lock; /* Spinlock for synchronized accesses */
+  int p_signal;                 /* last signal received */
+  sigjmp_buf * p_signal_jmp;    /* where to siglongjmp on a signal or NULL */
+  sigjmp_buf * p_cancel_jmp;    /* where to siglongjmp on a cancel or NULL */
+  char p_terminated;            /* true if terminated e.g. by pthread_exit */
+  char p_detached;              /* true if detached */
+  char p_exited;                /* true if the assoc. process terminated */
+  void * p_retval;              /* placeholder for return value */
+  int p_retcode;                /* placeholder for return code */
+  pthread_descr p_joining;      /* thread joining on that thread or NULL */
+  struct _pthread_cleanup_buffer * p_cleanup; /* cleanup functions */
+  char p_cancelstate;           /* cancellation state */
+  char p_canceltype;            /* cancellation type (deferred/async) */
+  char p_canceled;              /* cancellation request pending */
+  int * p_errnop;               /* pointer to used errno variable */
+  int p_errno;                  /* error returned by last system call */
+  int * p_h_errnop;             /* pointer to used h_errno variable */
+  int p_h_errno;                /* error returned by last netdb function */
+  char * p_in_sighandler;       /* stack address of sighandler, or NULL */
+  char p_sigwaiting;            /* true if a sigwait() is in progress */
+  struct pthread_start_args p_start_args; /* arguments for thread creation */
+  void ** p_specific[PTHREAD_KEY_1STLEVEL_SIZE]; /* thread-specific data */
+  void * p_libc_specific[_LIBC_TSD_KEY_N]; /* thread-specific data for libc */
+  int p_userstack;		/* nonzero if the user provided the stack */
+  void *p_guardaddr;		/* address of guard area or NULL */
+  size_t p_guardsize;		/* size of guard area */
+  int p_nr;                     /* Index of descriptor in __pthread_handles */
+  int p_report_events;		/* Nonzero if events must be reported.  */
+  td_eventbuf_t p_eventbuf;     /* Data for event.  */
+  struct pthread_atomic p_resume_count; /* number of times restart() was
+					   called on thread */
+  char p_woken_by_cancel;       /* cancellation performed wakeup */
+  char p_condvar_avail;		/* flag if conditional variable became avail */
+  char p_sem_avail;             /* flag if semaphore became available */
+  pthread_extricate_if *p_extricate; /* See above */
+  pthread_readlock_info *p_readlock_list;  /* List of readlock info structs */
+  pthread_readlock_info *p_readlock_free;  /* Free list of structs */
+  int p_untracked_readlock_count;	/* Readlocks not tracked by list */
+  struct __res_state *p_resp;	/* Pointer to resolver state */
+  struct __res_state p_res;	/* per-thread resolver state */
+  int p_inheritsched;           /* copied from the thread attribute */
+#if HP_TIMING_AVAIL
+  hp_timing_t p_cpuclock_offset; /* Initial CPU clock for thread.  */
+#endif
+  /* New elements must be added at the end.  */
+} __attribute__ ((aligned(32))); /* We need to align the structure so that
+				    doubles are aligned properly.  This is 8
+				    bytes on MIPS and 16 bytes on MIPS64.
+				    32 bytes might give better cache
+				    utilization.  */
+
+#endif	/* descr.h */
diff --git a/linuxthreads/internals.h b/linuxthreads/internals.h
index 8297c781c8..e1768b4a1f 100644
--- a/linuxthreads/internals.h
+++ b/linuxthreads/internals.h
@@ -20,21 +20,15 @@
 /* Includes */
 
 #include <limits.h>
-#include <resolv.h>
-#include <setjmp.h>
-#include <signal.h>
 #include <unistd.h>
 #include <stackinfo.h>
-#include <sys/types.h>
-#include <bits/libc-tsd.h> /* for _LIBC_TSD_KEY_N */
+
+#include "descr.h"
 
 extern long int testandset (int *spinlock);
 extern int __compare_and_swap (long int *p, long int oldval, long int newval);
 
-#include "pt-machine.h"
 #include "semaphore.h"
-#include "../linuxthreads_db/thread_dbP.h"
-#include <hp-timing.h>
 
 #ifndef THREAD_GETMEM
 # define THREAD_GETMEM(descr, member) descr->member
@@ -49,31 +43,6 @@ extern int __compare_and_swap (long int *p, long int oldval, long int newval);
 # define THREAD_SETMEM_NC(descr, member, value) descr->member = (value)
 #endif
 
-/* Arguments passed to thread creation routine */
-
-struct pthread_start_args {
-  void * (*start_routine)(void *); /* function to run */
-  void * arg;                   /* its argument */
-  sigset_t mask;                /* initial signal mask for thread */
-  int schedpolicy;              /* initial scheduling policy (if any) */
-  struct sched_param schedparam; /* initial scheduling parameters (if any) */
-};
-
-
-/* We keep thread specific data in a special data structure, a two-level
-   array.  The top-level array contains pointers to dynamically allocated
-   arrays of a certain number of data pointers.  So we can implement a
-   sparse array.  Each dynamic second-level array has
-	PTHREAD_KEY_2NDLEVEL_SIZE
-   entries.  This value shouldn't be too large.  */
-#define PTHREAD_KEY_2NDLEVEL_SIZE	32
-
-/* We need to address PTHREAD_KEYS_MAX key with PTHREAD_KEY_2NDLEVEL_SIZE
-   keys in each subarray.  */
-#define PTHREAD_KEY_1STLEVEL_SIZE \
-  ((PTHREAD_KEYS_MAX + PTHREAD_KEY_2NDLEVEL_SIZE - 1) \
-   / PTHREAD_KEY_2NDLEVEL_SIZE)
-
 typedef void (*destr_function)(void *);
 
 struct pthread_key_struct {
@@ -85,112 +54,6 @@ struct pthread_key_struct {
 #define PTHREAD_START_ARGS_INITIALIZER(fct) \
   { (void *(*) (void *)) fct, NULL, {{0, }}, 0, { 0 } }
 
-/* The type of thread descriptors */
-
-typedef struct _pthread_descr_struct * pthread_descr;
-
-/* Callback interface for removing the thread from waiting on an
-   object if it is cancelled while waiting or about to wait.
-   This hold a pointer to the object, and a pointer to a function
-   which ``extricates'' the thread from its enqueued state.
-   The function takes two arguments: pointer to the wait object,
-   and a pointer to the thread. It returns 1 if an extrication
-   actually occured, and hence the thread must also be signalled.
-   It returns 0 if the thread had already been extricated. */
-
-typedef struct _pthread_extricate_struct {
-    void *pu_object;
-    int (*pu_extricate_func)(void *, pthread_descr);
-} pthread_extricate_if;
-
-/* Atomic counter made possible by compare_and_swap */
-
-struct pthread_atomic {
-  long p_count;
-  int p_spinlock;
-};
-
-/* Context info for read write locks. The pthread_rwlock_info structure
-   is information about a lock that has been read-locked by the thread
-   in whose list this structure appears. The pthread_rwlock_context
-   is embedded in the thread context and contains a pointer to the
-   head of the list of lock info structures, as well as a count of
-   read locks that are untracked, because no info structure could be
-   allocated for them. */
-
-struct _pthread_rwlock_t;
-
-typedef struct _pthread_rwlock_info {
-  struct _pthread_rwlock_info *pr_next;
-  struct _pthread_rwlock_t *pr_lock;
-  int pr_lock_count;
-} pthread_readlock_info;
-
-struct _pthread_descr_struct {
-  union {
-    struct {
-      pthread_descr self;	/* Pointer to this structure */
-    } data;
-    void *__padding[16];
-  } p_header;
-  pthread_descr p_nextlive, p_prevlive;
-                                /* Double chaining of active threads */
-  pthread_descr p_nextwaiting;  /* Next element in the queue holding the thr */
-  pthread_descr p_nextlock;	/* can be on a queue and waiting on a lock */
-  pthread_t p_tid;              /* Thread identifier */
-  int p_pid;                    /* PID of Unix process */
-  int p_priority;               /* Thread priority (== 0 if not realtime) */
-  struct _pthread_fastlock * p_lock; /* Spinlock for synchronized accesses */
-  int p_signal;                 /* last signal received */
-  sigjmp_buf * p_signal_jmp;    /* where to siglongjmp on a signal or NULL */
-  sigjmp_buf * p_cancel_jmp;    /* where to siglongjmp on a cancel or NULL */
-  char p_terminated;            /* true if terminated e.g. by pthread_exit */
-  char p_detached;              /* true if detached */
-  char p_exited;                /* true if the assoc. process terminated */
-  void * p_retval;              /* placeholder for return value */
-  int p_retcode;                /* placeholder for return code */
-  pthread_descr p_joining;      /* thread joining on that thread or NULL */
-  struct _pthread_cleanup_buffer * p_cleanup; /* cleanup functions */
-  char p_cancelstate;           /* cancellation state */
-  char p_canceltype;            /* cancellation type (deferred/async) */
-  char p_canceled