xtensa: add support for TLS

The Xtensa architecture provides a global register called THREADPTR
for the purpose of Thread Local Storage (TLS) support. This allows us
to use a fairly simple implementation, keeping the thread pointer in
the regset and simply saving and restoring it upon entering/exiting
the from user space.
Signed-off-by: NChris Zankel <chris@zankel.net>

arch/xtensa/include/asm/elf.h

@@ -84,7 +84,8 @@ typedef struct {
 	elf_greg_t sar;
 	elf_greg_t windowstart;
 	elf_greg_t windowbase;
-	elf_greg_t reserved[8+48];
+	elf_greg_t threadptr;
+	elf_greg_t reserved[7+48];
 	elf_greg_t a[64];
 } xtensa_gregset_t;
 

arch/xtensa/include/asm/ptrace.h

@@ -38,6 +38,7 @@ struct pt_regs {
 	unsigned long syscall;		/*  56 */
 	unsigned long icountlevel;	/*  60 */
 	unsigned long scompare1;	/*  64 */
+	unsigned long threadptr;	/*  68 */
 
 	/* Additional configurable registers that are used by the compiler. */
 	xtregs_opt_t xtregs_opt;
@@ -48,7 +49,7 @@ struct pt_regs {
 	/* current register frame.
 	 * Note: The ESF for kernel exceptions ends after 16 registers!
 	 */
-	unsigned long areg[16];		/* 128 (64) */
+	unsigned long areg[16];
 };
 
 #include <variant/core.h>

arch/xtensa/kernel/asm-offsets.c

@@ -42,6 +42,7 @@ int main(void)
 	DEFINE(PT_ICOUNTLEVEL, offsetof (struct pt_regs, icountlevel));
 	DEFINE(PT_SYSCALL, offsetof (struct pt_regs, syscall));
 	DEFINE(PT_SCOMPARE1, offsetof(struct pt_regs, scompare1));
+	DEFINE(PT_THREADPTR, offsetof(struct pt_regs, threadptr));
 	DEFINE(PT_AREG, offsetof (struct pt_regs, areg[0]));
 	DEFINE(PT_AREG0, offsetof (struct pt_regs, areg[0]));
 	DEFINE(PT_AREG1, offsetof (struct pt_regs, areg[1]));

arch/xtensa/kernel/entry.S

@@ -130,6 +130,11 @@ _user_exception:
 	s32i	a3, a1, PT_SAR
 	s32i	a2, a1, PT_ICOUNTLEVEL
 
+#if XCHAL_HAVE_THREADPTR
+	rur	a2, threadptr
+	s32i	a2, a1, PT_THREADPTR
+#endif
+
 	/* Rotate ws so that the current windowbase is at bit0. */
 	/* Assume ws = xxwww1yyyy. Rotate ws right, so that a2 = yyyyxxwww1 */
 
@@ -510,6 +515,11 @@ user_exception_exit:
 	 *	 (if we have restored WSBITS-1 frames).
 	 */
 
+#if XCHAL_HAVE_THREADPTR
+	l32i	a3, a1, PT_THREADPTR
+	wur	a3, threadptr
+#endif
+
 2:	j	common_exception_exit
 
 	/* This is the kernel exception exit.
@@ -1955,7 +1965,7 @@ ENTRY(_switch_to)
 	s32i	a6, a3, EXC_TABLE_FIXUP
 	s32i	a7, a3, EXC_TABLE_KSTK
 
-	/* restore context of the task that 'next' addresses */
+	/* restore context of the task 'next' */
 
 	l32i	a0, a13, THREAD_RA	# restore return address
 	l32i	a1, a13, THREAD_SP	# restore stack pointer

arch/xtensa/kernel/process.c

@@ -259,9 +259,10 @@ int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
 			memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
 			       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
 		}
-// FIXME: we need to set THREADPTR in thread_info...
+
+		/* The thread pointer is passed in the '4th argument' (= a5) */
 		if (clone_flags & CLONE_SETTLS)
-			childregs->areg[2] = childregs->areg[6];
+			childregs->threadptr = childregs->areg[5];
 	} else {
 		p->thread.ra = MAKE_RA_FOR_CALL(
 				(unsigned long)ret_from_kernel_thread, 1);

arch/xtensa/kernel/ptrace.c

@@ -66,6 +66,7 @@ int ptrace_getregs(struct task_struct *child, void __user *uregs)
 	__put_user(regs->lcount, &gregset->lcount);
 	__put_user(regs->windowstart, &gregset->windowstart);
 	__put_user(regs->windowbase, &gregset->windowbase);
+	__put_user(regs->threadptr, &gregset->threadptr);
 
 	for (i = 0; i < XCHAL_NUM_AREGS; i++)
 		__put_user(regs->areg[i],
@@ -92,6 +93,7 @@ int ptrace_setregs(struct task_struct *child, void __user *uregs)
 	__get_user(regs->lcount, &gregset->lcount);
 	__get_user(ws, &gregset->windowstart);
 	__get_user(wb, &gregset->windowbase);
+	__get_user(regs->threadptr, &gregset->threadptr);
 
 	regs->ps = (regs->ps & ~ps_mask) | (ps & ps_mask) | (1 << PS_EXCM_BIT);
 

arch/xtensa/kernel/signal.c

@@ -337,7 +337,7 @@ static int setup_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
 	struct rt_sigframe *frame;
 	int err = 0;
 	int signal;
-	unsigned long sp, ra;
+	unsigned long sp, ra, tp;
 
 	sp = regs->areg[1];
 
@@ -395,7 +395,8 @@ static int setup_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
 	 * Return context not modified until this point.
 	 */
 
-	/* Set up registers for signal handler */
+	/* Set up registers for signal handler; preserve the threadptr */
+	tp = regs->threadptr;
 	start_thread(regs, (unsigned long) ka->sa.sa_handler,
 		     (unsigned long) frame);
 
@@ -406,6 +407,7 @@ static int setup_frame(int sig, struct k_sigaction *ka, siginfo_t *info,
 	regs->areg[6] = (unsigned long) signal;
 	regs->areg[7] = (unsigned long) &frame->info;
 	regs->areg[8] = (unsigned long) &frame->uc;
+	regs->threadptr = tp;
 
 	/* Set access mode to USER_DS.  Nomenclature is outdated, but
 	 * functionality is used in uaccess.h