hexagon: kernel_thread()/kernel_execve() conversion

introduce sane current_pt_regs(), use it in syscalls where needed.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

arch/hexagon/Kconfig

@@ -31,6 +31,8 @@ config HEXAGON
 	select GENERIC_CLOCKEVENTS
 	select GENERIC_CLOCKEVENTS_BROADCAST
 	select MODULES_USE_ELF_RELA
+	select GENERIC_KERNEL_THREAD
+	select GENERIC_KERNEL_EXECVE
 	---help---
 	  Qualcomm Hexagon is a processor architecture designed for high
 	  performance and low power across a wide variety of applications.

arch/hexagon/include/asm/processor.h

@@ -34,7 +34,6 @@
 struct task_struct;
 
 /*  this is defined in arch/process.c  */
-extern pid_t kernel_thread(int (*fn)(void *), void * arg, unsigned long flags);
 extern unsigned long thread_saved_pc(struct task_struct *tsk);
 
 extern void start_thread(struct pt_regs *, unsigned long, unsigned long);

arch/hexagon/include/asm/ptrace.h

@@ -32,4 +32,8 @@
 extern int regs_query_register_offset(const char *name);
 extern const char *regs_query_register_name(unsigned int offset);
 
+#define current_pt_regs() \
+	((struct pt_regs *) \
+	 ((unsigned long)current_thread_info() + THREAD_SIZE) - 1)
+
 #endif

arch/hexagon/kernel/process.c

@@ -25,33 +25,6 @@
 #include <linux/uaccess.h>
 #include <linux/slab.h>
 
-/*
- * Kernel thread creation.  The desired kernel function is "wrapped"
- * in the kernel_thread_helper function, which does cleanup
- * afterwards.
- */
-static void __noreturn kernel_thread_helper(void *arg, int (*fn)(void *))
-{
-	do_exit(fn(arg));
-}
-
-int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
-{
-	struct pt_regs regs;
-
-	memset(&regs, 0, sizeof(regs));
-	/*
-	 * Yes, we're exploting illicit knowledge of the ABI here.
-	 */
-	regs.r00 = (unsigned long) arg;
-	regs.r01 = (unsigned long) fn;
-	pt_set_elr(&regs, (unsigned long)kernel_thread_helper);
-	pt_set_kmode(&regs);
-
-	return do_fork(flags|CLONE_VM|CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
-}
-EXPORT_SYMBOL(kernel_thread);
-
 /*
  * Program thread launch.  Often defined as a macro in processor.h,
  * but we're shooting for a small footprint and it's not an inner-loop
@@ -114,7 +87,7 @@ unsigned long thread_saved_pc(struct task_struct *tsk)
  * Copy architecture-specific thread state
  */
 int copy_thread(unsigned long clone_flags, unsigned long usp,
-		unsigned long unused, struct task_struct *p,
+		unsigned long arg, struct task_struct *p,
 		struct pt_regs *regs)
 {
 	struct thread_info *ti = task_thread_info(p);
@@ -125,61 +98,50 @@ int copy_thread(unsigned long clone_flags, unsigned long usp,
 	childregs = (struct pt_regs *) (((unsigned long) ti + THREAD_SIZE) -
 					sizeof(*childregs));
 
-	memcpy(childregs, regs, sizeof(*childregs));
 	ti->regs = childregs;
 
 	/*
 	 * Establish kernel stack pointer and initial PC for new thread
+	 * Note that unlike the usual situation, we do not copy the
+	 * parent's callee-saved here; those are in pt_regs and whatever
+	 * we leave here will be overridden on return to userland.
 	 */
 	ss = (struct hexagon_switch_stack *) ((unsigned long) childregs -
 						    sizeof(*ss));
 	ss->lr = (unsigned long)ret_from_fork;
 	p->thread.switch_sp = ss;
-
-	/* If User mode thread, set pt_reg stack pointer as per parameter */
-	if (user_mode(childregs)) {
-		pt_set_rte_sp(childregs, usp);
-
-		/* Child sees zero return value */
-		childregs->r00 = 0;
-
-		/*
-		 * The clone syscall has the C signature:
-		 * int [r0] clone(int flags [r0],
-		 *           void *child_frame [r1],
-		 *           void *parent_tid [r2],
-		 *           void *child_tid [r3],
-		 *           void *thread_control_block [r4]);
-		 * ugp is used to provide TLS support.
-		 */
-		if (clone_flags & CLONE_SETTLS)
-			childregs->ugp = childregs->r04;
-
-		/*
-		 * Parent sees new pid -- not necessary, not even possible at
-		 * this point in the fork process
-		 * Might also want to set things like ti->addr_limit
-		 */
-	} else {
-		/*
-		 * If kernel thread, resume stack is kernel stack base.
-		 * Note that this is pointer arithmetic on pt_regs *
-		 */
-		pt_set_rte_sp(childregs, (unsigned long)(childregs + 1));
-		/*
-		 * We need the current thread_info fast path pointer
-		 * set up in pt_regs.  The register to be used is
-		 * parametric for assembler code, but the mechanism
-		 * doesn't drop neatly into C.  Needs to be fixed.
-		 */
-		childregs->THREADINFO_REG = (unsigned long) ti;
+	if (unlikely(p->flags & PF_KTHREAD)) {
+		memset(childregs, 0, sizeof(struct pt_regs));
+		/* r24 <- fn, r25 <- arg */
+		ss->r2524 = usp | ((u64)arg << 32);
+		pt_set_kmode(childregs);
+		return 0;
 	}
+	memcpy(childregs, regs, sizeof(*childregs));
+	ss->r2524 = 0;
+
+	pt_set_rte_sp(childregs, usp);
+
+	/* Child sees zero return value */
+	childregs->r00 = 0;
+
+	/*
+	 * The clone syscall has the C signature:
+	 * int [r0] clone(int flags [r0],
+	 *           void *child_frame [r1],
+	 *           void *parent_tid [r2],
+	 *           void *child_tid [r3],
+	 *           void *thread_control_block [r4]);
+	 * ugp is used to provide TLS support.
+	 */
+	if (clone_flags & CLONE_SETTLS)
+		childregs->ugp = childregs->r04;
 
 	/*
-	 * thread_info pointer is pulled out of task_struct "stack"
-	 * field on switch_to.
+	 * Parent sees new pid -- not necessary, not even possible at
+	 * this point in the fork process
+	 * Might also want to set things like ti->addr_limit
 	 */
-	p->stack = (void *)ti;
 
 	return 0;
 }

arch/hexagon/kernel/signal.c

@@ -249,14 +249,14 @@ void do_notify_resume(struct pt_regs *regs, unsigned long thread_info_flags)
  */
 asmlinkage int sys_sigaltstack(const stack_t __user *uss, stack_t __user *uoss)
 {
-	struct pt_regs *regs = current_thread_info()->regs;
+	struct pt_regs *regs = current_pt_regs();
 
 	return do_sigaltstack(uss, uoss, regs->r29);
 }
 
 asmlinkage int sys_rt_sigreturn(void)
 {
-	struct pt_regs *regs = current_thread_info()->regs;
+	struct pt_regs *regs = current_pt_regs();
 	struct rt_sigframe __user *frame;
 	sigset_t blocked;
 

arch/hexagon/kernel/syscall.c

@@ -39,7 +39,7 @@ asmlinkage int sys_execve(char __user *ufilename,
 			  const char __user *const __user *argv,
 			  const char __user *const __user *envp)
 {
-	struct pt_regs *pregs = current_thread_info()->regs;
+	struct pt_regs *pregs = current_pt_regs();
 	struct filename *filename;
 	int retval;
 
@@ -57,33 +57,10 @@ asmlinkage int sys_execve(char __user *ufilename,
 asmlinkage int sys_clone(unsigned long clone_flags, unsigned long newsp,
 			 unsigned long parent_tidp, unsigned long child_tidp)
 {
-	struct pt_regs *pregs = current_thread_info()->regs;
+	struct pt_regs *pregs = current_pt_regs();
 
 	if (!newsp)
 		newsp = pregs->SP;
 	return do_fork(clone_flags, newsp, pregs, 0, (int __user *)parent_tidp,
 		       (int __user *)child_tidp);
 }
-
-/*
- * Do a system call from the kernel, so as to have a proper pt_regs
- * and recycle the sys_execvpe infrustructure.
- */
-int kernel_execve(const char *filename,
-		  const char *const argv[], const char *const envp[])
-{
-	register unsigned long __a0 asm("r0") = (unsigned long) filename;
-	register unsigned long __a1 asm("r1") = (unsigned long) argv;
-	register unsigned long __a2 asm("r2") = (unsigned long) envp;
-	int retval;
-
-	__asm__ volatile(
-		"	R6 = #%4;\n"
-		"	trap0(#1);\n"
-		"	%0 = R0;\n"
-		: "=r" (retval)
-		: "r" (__a0), "r" (__a1), "r" (__a2), "i" (__NR_execve)
-	);
-
-	return retval;
-}

arch/hexagon/kernel/vm_entry.S

@@ -266,4 +266,8 @@ _K_enter_machcheck:
 	.globl ret_from_fork
 ret_from_fork:
 	call schedule_tail
+	P0 = cmp.eq(R24, #0);
+	if P0 jump return_from_syscall
+	R0 = R25;
+	callr R24
 	jump return_from_syscall