hurd: add TLS support

	* sysdeps/generic/thread_state.h (MACHINE_NEW_THREAD_STATE_FLAVOR):
	Define macro.
	* sysdeps/mach/thread_state.h (MACHINE_THREAD_STATE_FIX_NEW): New macro.
	* sysdeps/mach/i386/thread_state.h
	(MACHINE_NEW_THREAD_STATE_FLAVOR): New macro, defined to
	i386_THREAD_STATE.
	(MACHINE_THREAD_STATE_FLAVOR): Define to i386_REGS_SEGS_STATE instead of
	i386_THREAD_STATE.
	(MACHINE_THREAD_STATE_FIX_NEW): New macro, reads segments.

	* sysdeps/mach/hurd/i386/trampoline.c (_hurd_setup_sighandler): Use
	i386_REGS_SEGS_STATE instead of i386_THREAD_STATE.

	* sysdeps/mach/hurd/i386/tls.h (TCB_ALIGNMENT, HURD_SEL_LDT): New
	macros.
	(_hurd_tls_fork): Add original thread parameter, Duplicate existing LDT
	descriptor instead of creating a new one.
	(_hurd_tls_new): New function, creates a new descriptor and updates tcb.

	* mach/setup-thread.c: Include <ldsodefs.h>.
	(__mach_setup_thread): Call _dl_allocate_tls, pass
	MACHINE_NEW_THREAD_STATE_FLAVOR to __thread_set_state instead of
	MACHINE_THREAD_STATE_FLAVOR, before getting
	MACHINE_THREAD_STATE_FLAVOR, calling _hurd_tls_new, and setting
	MACHINE_THREAD_STATE_FLAVOR with the result.
	* hurd/hurdfault.c (_hurdsig_fault_init): Call
	MACHINE_THREAD_STATE_FIX_NEW.
	* sysdeps/mach/hurd/fork.c (__fork): Call _hurd_tls_fork for sigthread
	too.  Add original thread parameter.

9 files changed, 132 insertions, 10 deletions

diff --git a/ChangeLog b/ChangeLog
index c2468c8..175bcaf 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,35 @@
+2018-03-16  Samuel Thibault  <samuel.thibault@ens-lyon.org>
+
+	* sysdeps/generic/thread_state.h (MACHINE_NEW_THREAD_STATE_FLAVOR):
+	Define macro.
+	* sysdeps/mach/thread_state.h (MACHINE_THREAD_STATE_FIX_NEW): New macro.
+	* sysdeps/mach/i386/thread_state.h
+	(MACHINE_NEW_THREAD_STATE_FLAVOR): New macro, defined to
+	i386_THREAD_STATE.
+	(MACHINE_THREAD_STATE_FLAVOR): Define to i386_REGS_SEGS_STATE instead of
+	i386_THREAD_STATE.
+	(MACHINE_THREAD_STATE_FIX_NEW): New macro, reads segments.
+
+	* sysdeps/mach/hurd/i386/trampoline.c (_hurd_setup_sighandler): Use
+	i386_REGS_SEGS_STATE instead of i386_THREAD_STATE.
+
+	* sysdeps/mach/hurd/i386/tls.h (TCB_ALIGNMENT, HURD_SEL_LDT): New
+	macros.
+	(_hurd_tls_fork): Add original thread parameter, Duplicate existing LDT
+	descriptor instead of creating a new one.
+	(_hurd_tls_new): New function, creates a new descriptor and updates tcb.
+
+	* mach/setup-thread.c: Include <ldsodefs.h>.
+	(__mach_setup_thread): Call _dl_allocate_tls, pass
+	MACHINE_NEW_THREAD_STATE_FLAVOR to __thread_set_state instead of
+	MACHINE_THREAD_STATE_FLAVOR, before getting
+	MACHINE_THREAD_STATE_FLAVOR, calling _hurd_tls_new, and setting
+	MACHINE_THREAD_STATE_FLAVOR with the result.
+	* hurd/hurdfault.c (_hurdsig_fault_init): Call
+	MACHINE_THREAD_STATE_FIX_NEW.
+	* sysdeps/mach/hurd/fork.c (__fork): Call _hurd_tls_fork for sigthread
+	too.  Add original thread parameter.
+
 2018-03-16  Joseph Myers  <joseph@codesourcery.com>
 
 	* sysdeps/x86/fpu/bits/mathinline.h [__USE_MISC] (__finite):
diff --git a/hurd/hurdfault.c b/hurd/hurdfault.c
index b5a4056..39a4522 100644
--- a/hurd/hurdfault.c
+++ b/hurd/hurdfault.c
@@ -204,6 +204,7 @@ _hurdsig_fault_init (void)
   /* This state will be restored when we fault.
      It runs the function above.  */
   memset (&state, 0, sizeof state);
+  MACHINE_THREAD_STATE_FIX_NEW (&state);
   MACHINE_THREAD_STATE_SET_PC (&state, faulted);
   MACHINE_THREAD_STATE_SET_SP (&state, faultstack, sizeof faultstack);
 
diff --git a/mach/setup-thread.c b/mach/setup-thread.c
index 2eeacc2..6716813 100644
--- a/mach/setup-thread.c
+++ b/mach/setup-thread.c
@@ -19,6 +19,7 @@
 #include <thread_state.h>
 #include <string.h>
 #include <mach/machine/vm_param.h>
+#include <ldsodefs.h>
 #include "sysdep.h"		/* Defines stack direction.  */
 
 #define	STACK_SIZE	(16 * 1024 * 1024) /* 16MB, arbitrary.  */
@@ -41,6 +42,7 @@ __mach_setup_thread (task_t task, thread_t thread, void *pc,
   vm_address_t stack;
   vm_size_t size;
   int anywhere;
+  tcbhead_t *tcb;
 
   size = stack_size ? *stack_size ? : STACK_SIZE : STACK_SIZE;
   stack = stack_base ? *stack_base ? : 0 : 0;
@@ -50,6 +52,10 @@ __mach_setup_thread (task_t task, thread_t thread, void *pc,
   if (error)
     return error;
 
+  tcb = _dl_allocate_tls(NULL);
+  if (tcb == NULL)
+    return KERN_RESOURCE_SHORTAGE;
+
   if (stack_size)
     *stack_size = size;
 
@@ -72,8 +78,21 @@ __mach_setup_thread (task_t task, thread_t thread, void *pc,
   if (error = __vm_protect (task, stack, __vm_page_size, 0, VM_PROT_NONE))
     return error;
 
-  return __thread_set_state (thread, MACHINE_THREAD_STATE_FLAVOR,
-			     (natural_t *) &ts, tssize);
+  if (error = __thread_set_state (thread, MACHINE_NEW_THREAD_STATE_FLAVOR,
+			     (natural_t *) &ts, tssize))
+    return error;
+
+  if (error = __thread_get_state (thread, MACHINE_THREAD_STATE_FLAVOR,
+			     (natural_t *) &ts, &tssize))
+    return error;
+  assert (tssize == MACHINE_THREAD_STATE_COUNT);
+
+  _hurd_tls_new(thread, &ts, tcb);
+
+  error = __thread_set_state (thread, MACHINE_THREAD_STATE_FLAVOR,
+			      (natural_t *) &ts, tssize);
+
+  return error;
 }
 
 weak_alias (__mach_setup_thread, mach_setup_thread)
diff --git a/sysdeps/generic/thread_state.h b/sysdeps/generic/thread_state.h
index 32994c3..99b7c92 100644
--- a/sysdeps/generic/thread_state.h
+++ b/sysdeps/generic/thread_state.h
@@ -22,6 +22,11 @@
 
 /* Replace <machine> with "i386" or "mips" or whatever.  */
 
+/* This lets the kernel define architecture-specific registers for a new
+   thread.  */
+#define MACHINE_NEW_THREAD_STATE_FLAVOR	<machine>_NEW_THREAD_STATE
+/* This makes the kernel load all architectures-specific registers for the
+   thread.  */
 #define MACHINE_THREAD_STATE_FLAVOR	<machine>_THREAD_STATE
 #define MACHINE_THREAD_STATE_COUNT	<machine>_THREAD_STATE_COUNT
 
diff --git a/sysdeps/mach/hurd/fork.c b/sysdeps/mach/hurd/fork.c
index b55eecf..cc322eb 100644
--- a/sysdeps/mach/hurd/fork.c
+++ b/sysdeps/mach/hurd/fork.c
@@ -507,6 +507,11 @@ __fork (void)
 #endif
       MACHINE_THREAD_STATE_SET_PC (&state,
 				   (unsigned long int) _hurd_msgport_receive);
+
+      /* Do special signal thread setup for TLS if needed.  */
+      if (err = _hurd_tls_fork (sigthread, _hurd_msgport_thread, &state))
+	LOSE;
+
       if (err = __thread_set_state (sigthread, MACHINE_THREAD_STATE_FLAVOR,
 				    (natural_t *) &state, statecount))
 	LOSE;
@@ -517,7 +522,7 @@ __fork (void)
       _hurd_longjmp_thread_state (&state, env, 1);
 
       /* Do special thread setup for TLS if needed.  */
-      if (err = _hurd_tls_fork (thread, &state))
+      if (err = _hurd_tls_fork (thread, ss->thread, &state))
 	LOSE;
 
       if (err = __thread_set_state (thread, MACHINE_THREAD_STATE_FLAVOR,
diff --git a/sysdeps/mach/hurd/i386/tls.h b/sysdeps/mach/hurd/i386/tls.h
index b194a49..4c78685 100644
--- a/sysdeps/mach/hurd/i386/tls.h
+++ b/sysdeps/mach/hurd/i386/tls.h
@@ -52,6 +52,15 @@ typedef struct
 #define TLS_TCB_AT_TP	1
 #define TLS_DTV_AT_TP	0
 
+/* Alignment requirement for TCB.
+
+   Some processors such as Intel Atom pay a big penalty on every
+   access using a segment override if that segment's base is not
+   aligned to the size of a cache line.  (See Intel 64 and IA-32
+   Architectures Optimization Reference Manual, section 13.3.3.3,
+   "Segment Base".)  On such machines, a cache line is 64 bytes.  */
+#define TCB_ALIGNMENT		64
+
 #ifndef __ASSEMBLER__
 
 /* Use i386-specific RPCs to arrange that %gs segment register prefix
@@ -78,6 +87,7 @@ typedef struct
       | (((unsigned int) (tcb)) & 0xff000000) /* base 24..31 */		      \
     }
 
+# define HURD_SEL_LDT(sel) (__builtin_expect((sel) & 4, 0))
 
 static inline const char * __attribute__ ((unused))
 _hurd_tls_init (tcbhead_t *tcb)
@@ -141,9 +151,40 @@ _hurd_tls_init (tcbhead_t *tcb)
 
 # include <mach/machine/thread_status.h>
 
-/* Set up TLS in the new thread of a fork child, copying from our own.  */
-static inline error_t __attribute__ ((unused))
-_hurd_tls_fork (thread_t child, struct i386_thread_state *state)
+/* Set up TLS in the new thread of a fork child, copying from the original.  */
+static inline kern_return_t __attribute__ ((unused))
+_hurd_tls_fork (thread_t child, thread_t orig, struct i386_thread_state *state)
+{
+  /* Fetch the selector set by _hurd_tls_init.  */
+  int sel;
+  asm ("mov %%gs, %w0" : "=q" (sel) : "0" (0));
+  if (sel == state->ds)		/* _hurd_tls_init was never called.  */
+    return 0;
+
+  struct descriptor desc, *_desc = &desc;
+  error_t err;
+  unsigned int count = 1;
+
+  if (HURD_SEL_LDT(sel))
+    err = __i386_get_ldt (orig, sel, 1, &_desc, &count);
+  else
+    err = __i386_get_gdt (orig, sel, &desc);
+
+  assert_perror (err);
+  if (err)
+    return err;
+
+  if (HURD_SEL_LDT(sel))
+    err = __i386_set_ldt (child, sel, &desc, 1);
+  else
+    err = __i386_set_gdt (child, &sel, desc);
+
+  state->gs = sel;
+  return err;
+}
+
+static inline kern_return_t __attribute__ ((unused))
+_hurd_tls_new (thread_t child, struct i386_thread_state *state, tcbhead_t *tcb)
 {
   /* Fetch the selector set by _hurd_tls_init.  */
   int sel;
@@ -151,11 +192,13 @@ _hurd_tls_fork (thread_t child, struct i386_thread_state *state)
   if (sel == state->ds)		/* _hurd_tls_init was never called.  */
     return 0;
 
-  tcbhead_t *const tcb = THREAD_SELF;
   HURD_TLS_DESC_DECL (desc, tcb);
   error_t err;
 
-  if (__builtin_expect (sel, 0x50) & 4) /* LDT selector */
+  tcb->tcb = tcb;
+  tcb->self = child;
+
+  if (HURD_SEL_LDT(sel))
     err = __i386_set_ldt (child, sel, &desc, 1);
   else
     err = __i386_set_gdt (child, &sel, desc);
diff --git a/sysdeps/mach/hurd/i386/trampoline.c b/sysdeps/mach/hurd/i386/trampoline.c
index d240ca7..0c68759 100644
--- a/sysdeps/mach/hurd/i386/trampoline.c
+++ b/sysdeps/mach/hurd/i386/trampoline.c
@@ -62,7 +62,7 @@ _hurd_setup_sighandler (struct hurd_sigstate *ss, __sighandler_t handler,
 		  sizeof (state->basic));
 	  memcpy (&state->fpu, &ss->context->sc_i386_float_state,
 		  sizeof (state->fpu));
-	  state->set |= (1 << i386_THREAD_STATE) | (1 << i386_FLOAT_STATE);
+	  state->set |= (1 << i386_REGS_SEGS_STATE) | (1 << i386_FLOAT_STATE);
 	}
     }
 
diff --git a/sysdeps/mach/i386/thread_state.h b/sysdeps/mach/i386/thread_state.h
index 56d91df..be3c8a9 100644
--- a/sysdeps/mach/i386/thread_state.h
+++ b/sysdeps/mach/i386/thread_state.h
@@ -21,7 +21,10 @@
 
 #include <mach/machine/thread_status.h>
 
-#define MACHINE_THREAD_STATE_FLAVOR	i386_THREAD_STATE
+/* This lets the kernel define segments for a new thread.  */
+#define MACHINE_NEW_THREAD_STATE_FLAVOR	i386_THREAD_STATE
+/* This makes the kernel load our segments descriptors.  */
+#define MACHINE_THREAD_STATE_FLAVOR	i386_REGS_SEGS_STATE
 #define MACHINE_THREAD_STATE_COUNT	i386_THREAD_STATE_COUNT
 
 #define machine_thread_state i386_thread_state
@@ -30,6 +33,14 @@
 #define SP uesp
 #define SYSRETURN eax
 
+#define MACHINE_THREAD_STATE_FIX_NEW(ts) do { \
+	asm ("mov %%cs, %w0" : "=q" ((ts)->cs)); \
+	asm ("mov %%ds, %w0" : "=q" ((ts)->ds)); \
+	asm ("mov %%es, %w0" : "=q" ((ts)->es)); \
+	asm ("mov %%fs, %w0" : "=q" ((ts)->fs)); \
+	asm ("mov %%gs, %w0" : "=q" ((ts)->gs)); \
+} while(0)
+
 struct machine_thread_all_state
   {
     int set;			/* Mask of bits (1 << FLAVOR).  */
diff --git a/sysdeps/mach/thread_state.h b/sysdeps/mach/thread_state.h
index bc4feef..2e3a10c 100644
--- a/sysdeps/mach/thread_state.h
+++ b/sysdeps/mach/thread_state.h
@@ -38,6 +38,12 @@
 #endif
 #endif
 
+/* This copies architecture-specific bits from the current thread to the new
+   thread state.  */
+#ifndef MACHINE_THREAD_STATE_FIX_NEW
+# define MACHINE_THREAD_STATE_FIX_NEW(ts)
+#endif
+
 /* These functions are of use in machine-dependent signal trampoline
    implementations.  */