[PATCH 0/3] xtensa: conversion to generic kernel_thread and friends

[PATCH 0/3] xtensa: conversion to generic kernel_thread and friends

[Max Filippov]

Hi.

This series converts xtensa arch to use kernel_thread, kernel_execve and
generic sys_execve.

Max Filippov (3):
  xtensa: switch to generic kernel_thread()
  xtensa: switch to generic kernel_execve()
  xtensa: switch to generic sys_execve()

 arch/xtensa/Kconfig                   |    2 +
 arch/xtensa/include/asm/processor.h   |    3 -
 arch/xtensa/include/asm/syscall.h     |    2 +-
 arch/xtensa/include/asm/unistd.h      |    1 +
 arch/xtensa/include/uapi/asm/unistd.h |    2 +-
 arch/xtensa/kernel/entry.S            |   57 +++++---------------
 arch/xtensa/kernel/process.c          |   97 +++++++++++++++------------------
 arch/xtensa/kernel/xtensa_ksyms.c     |    1 -
 8 files changed, 62 insertions(+), 103 deletions(-)

-- 
1.7.7.6

[PATCH 1/3] xtensa: switch to generic kernel_thread()

[Max Filippov]

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
---
 arch/xtensa/Kconfig                 |    1 +
 arch/xtensa/include/asm/processor.h |    3 -
 arch/xtensa/kernel/entry.S          |   41 ++++++-------------
 arch/xtensa/kernel/process.c        |   72 +++++++++++++++++++++-------------
 arch/xtensa/kernel/xtensa_ksyms.c   |    1 -
 5 files changed, 58 insertions(+), 60 deletions(-)

diff --git a/arch/xtensa/Kconfig b/arch/xtensa/Kconfig
index cdcb48a..9525b70 100644
--- a/arch/xtensa/Kconfig
+++ b/arch/xtensa/Kconfig
@@ -13,6 +13,7 @@ config XTENSA
 	select GENERIC_CPU_DEVICES
 	select MODULES_USE_ELF_RELA
 	select GENERIC_PCI_IOMAP
+	select GENERIC_KERNEL_THREAD
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	help
 	  Xtensa processors are 32-bit RISC machines designed by Tensilica
diff --git a/arch/xtensa/include/asm/processor.h b/arch/xtensa/include/asm/processor.h
index 5c371d8..a8642da 100644
--- a/arch/xtensa/include/asm/processor.h
+++ b/arch/xtensa/include/asm/processor.h
@@ -168,9 +168,6 @@ struct mm_struct;
 /* Free all resources held by a thread. */
 #define release_thread(thread) do { } while(0)
 
-/* Create a kernel thread without removing it from tasklists */
-extern int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags);
-
 /* Copy and release all segment info associated with a VM */
 #define copy_segments(p, mm)	do { } while(0)
 #define release_segments(mm)	do { } while(0)
diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S
index 18453067..d4ec138 100644
--- a/arch/xtensa/kernel/entry.S
+++ b/arch/xtensa/kernel/entry.S
@@ -1833,34 +1833,6 @@ ENTRY(system_call)
 
 
 /*
- * Create a kernel thread
- *
- * int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
- * a2                    a2                 a3             a4
- */
-
-ENTRY(kernel_thread)
-	entry	a1, 16
-
-	mov	a5, a2			# preserve fn over syscall
-	mov	a7, a3			# preserve args over syscall
-
-	movi	a3, _CLONE_VM | _CLONE_UNTRACED
-	movi	a2, __NR_clone
-	or	a6, a4, a3		# arg0: flags
-	mov	a3, a1			# arg1: sp
-	syscall
-
-	beq	a3, a1, 1f		# branch if parent
-	mov	a6, a7			# args
-	callx4	a5			# fn(args)
-
-	movi	a2, __NR_exit
-	syscall				# return value of fn(args) still in a6
-
-1:	retw
-
-/*
  * Do a system call from kernel instead of calling sys_execve, so we end up
  * with proper pt_regs.
  *
@@ -1958,3 +1930,16 @@ ENTRY(ret_from_fork)
 
 	j	common_exception_return
 
+/*
+ * Kernel thread creation helper
+ * On entry, set up by copy_thread: a2 = thread_fn, a3 = thread_fn arg
+ *           left from _switch_to: a6 = prev
+ */
+ENTRY(ret_from_kernel_thread)
+
+	call4	schedule_tail
+	mov	a6, a3
+	callx4	a2
+	call4	do_exit
+
+ENDPROC(ret_from_kernel_thread)
diff --git a/arch/xtensa/kernel/process.c b/arch/xtensa/kernel/process.c
index 1908f66..fd538bb 100644
--- a/arch/xtensa/kernel/process.c
+++ b/arch/xtensa/kernel/process.c
@@ -45,6 +45,7 @@
 #include <asm/regs.h>
 
 extern void ret_from_fork(void);
+extern void ret_from_kernel_thread(void);
 
 struct task_struct *current_set[NR_CPUS] = {&init_task, };
 
@@ -158,18 +159,30 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 /*
  * Copy thread.
  *
+ * There are two modes in which this function is called:
+ * 1) Userspace thread creation,
+ *    regs != NULL, usp_thread_fn is userspace stack pointer.
+ *    It is expected to copy parent regs (in case CLONE_VM is not set
+ *    in the clone_flags) and set up passed usp in the childregs.
+ * 2) Kernel thread creation,
+ *    regs == NULL, usp_thread_fn is the function to run in the new thread
+ *    and thread_fn_arg is its parameter.
+ *    childregs are not used for the kernel threads.
+ *
  * The stack layout for the new thread looks like this:
  *
- *	+------------------------+ <- sp in childregs (= tos)
+ *	+------------------------+
  *	|       childregs        |
  *	+------------------------+ <- thread.sp = sp in dummy-frame
  *	|      dummy-frame       |    (saved in dummy-frame spill-area)
  *	+------------------------+
  *
- * We create a dummy frame to return to ret_from_fork:
- *   a0 points to ret_from_fork (simulating a call4)
+ * We create a dummy frame to return to either ret_from_fork or
+ *   ret_from_kernel_thread:
+ *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
  *   sp points to itself (thread.sp)
- *   a2, a3 are unused.
+ *   a2, a3 are unused for userspace threads,
+ *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
  *
  * Note: This is a pristine frame, so we don't need any spill region on top of
  *       childregs.
@@ -185,41 +198,35 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  * involved.  Much simpler to just not copy those live frames across.
  */
 
-int copy_thread(unsigned long clone_flags, unsigned long usp,
-		unsigned long unused,
+int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
+		unsigned long thread_fn_arg,
                 struct task_struct * p, struct pt_regs * regs)
 {
-	struct pt_regs *childregs;
-	unsigned long tos;
-	int user_mode = user_mode(regs);
+	struct pt_regs *childregs = task_pt_regs(p);
 
 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
 	struct thread_info *ti;
 #endif
 
-	/* Set up new TSS. */
-	tos = (unsigned long)task_stack_page(p) + THREAD_SIZE;
-	if (user_mode)
-		childregs = (struct pt_regs*)(tos - PT_USER_SIZE);
-	else
-		childregs = (struct pt_regs*)tos - 1;
-
-	/* This does not copy all the regs.  In a bout of brilliance or madness,
-	   ARs beyond a0-a15 exist past the end of the struct. */
-	*childregs = *regs;
-
 	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
 	*((int*)childregs - 3) = (unsigned long)childregs;
 	*((int*)childregs - 4) = 0;
 
-	childregs->areg[2] = 0;
 	p->set_child_tid = p->clear_child_tid = NULL;
-	p->thread.ra = MAKE_RA_FOR_CALL((unsigned long)ret_from_fork, 0x1);
 	p->thread.sp = (unsigned long)childregs;
 
-	if (user_mode(regs)) {
+	if (regs) {
+		p->thread.ra = MAKE_RA_FOR_CALL(
+				(unsigned long)ret_from_fork, 0x1);
+
+		/* This does not copy all the regs.
+		 * In a bout of brilliance or madness,
+		 * ARs beyond a0-a15 exist past the end of the struct.
+		 */
+		*childregs = *regs;
+		childregs->areg[1] = usp_thread_fn;
+		childregs->areg[2] = 0;
 
-		childregs->areg[1] = usp;
 		if (clone_flags & CLONE_VM) {
 			childregs->wmask = 1;	/* can't share live windows */
 		} else {
@@ -230,11 +237,20 @@ int copy_thread(unsigned long clone_flags, unsigned long usp,
 // FIXME: we need to set THREADPTR in thread_info...
 		if (clone_flags & CLONE_SETTLS)
 			childregs->areg[2] = childregs->areg[6];
-
 	} else {
-		/* In kernel space, we start a new thread with a new stack. */
-		childregs->wmask = 1;
-		childregs->areg[1] = tos;
+		p->thread.ra = MAKE_RA_FOR_CALL(
+				(unsigned long)ret_from_kernel_thread, 1);
+
+		/* pass parameters to ret_from_kernel_thread:
+		 * a2 = thread_fn, a3 = thread_fn arg
+		 */
+		*((int *)childregs - 1) = thread_fn_arg;
+		*((int *)childregs - 2) = usp_thread_fn;
+
+		/* Childregs are only used when we're going to userspace
+		 * in which case start_thread will set them up.
+		 */
+		memset(childregs, 0, sizeof(*childregs));
 	}
 
 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
diff --git a/arch/xtensa/kernel/xtensa_ksyms.c b/arch/xtensa/kernel/xtensa_ksyms.c
index a8b9f1f..afe058b 100644
--- a/arch/xtensa/kernel/xtensa_ksyms.c
+++ b/arch/xtensa/kernel/xtensa_ksyms.c
@@ -43,7 +43,6 @@ EXPORT_SYMBOL(__strncpy_user);
 EXPORT_SYMBOL(clear_page);
 EXPORT_SYMBOL(copy_page);
 
-EXPORT_SYMBOL(kernel_thread);
 EXPORT_SYMBOL(empty_zero_page);
 
 /*
-- 
1.7.7.6

[PATCH 2/3] xtensa: switch to generic kernel_execve()

[Max Filippov]

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
---
 arch/xtensa/Kconfig        |    1 +
 arch/xtensa/kernel/entry.S |   18 +-----------------
 2 files changed, 2 insertions(+), 17 deletions(-)

diff --git a/arch/xtensa/Kconfig b/arch/xtensa/Kconfig
index 9525b70..0d1f36a 100644
--- a/arch/xtensa/Kconfig
+++ b/arch/xtensa/Kconfig
@@ -14,6 +14,7 @@ config XTENSA
 	select MODULES_USE_ELF_RELA
 	select GENERIC_PCI_IOMAP
 	select GENERIC_KERNEL_THREAD
+	select GENERIC_KERNEL_EXECVE
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	help
 	  Xtensa processors are 32-bit RISC machines designed by Tensilica
diff --git a/arch/xtensa/kernel/entry.S b/arch/xtensa/kernel/entry.S
index d4ec138..90bfc1d 100644
--- a/arch/xtensa/kernel/entry.S
+++ b/arch/xtensa/kernel/entry.S
@@ -1833,22 +1833,6 @@ ENTRY(system_call)
 
 
 /*
- * Do a system call from kernel instead of calling sys_execve, so we end up
- * with proper pt_regs.
- *
- * int kernel_execve(const char *fname, char *const argv[], charg *const envp[])
- * a2                        a2               a3                  a4
- */
-
-ENTRY(kernel_execve)
-	entry	a1, 16
-	mov	a6, a2			# arg0 is in a6
-	movi	a2, __NR_execve
-	syscall
-
-	retw
-
-/*
  * Task switch.
  *
  * struct task*  _switch_to (struct task* prev, struct task* next)
@@ -1940,6 +1924,6 @@ ENTRY(ret_from_kernel_thread)
 	call4	schedule_tail
 	mov	a6, a3
 	callx4	a2
-	call4	do_exit
+	j	common_exception_return
 
 ENDPROC(ret_from_kernel_thread)
-- 
1.7.7.6

[PATCH 3/3] xtensa: switch to generic sys_execve()

[Max Filippov]

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
---
 arch/xtensa/include/asm/syscall.h     |    2 +-
 arch/xtensa/include/asm/unistd.h      |    1 +
 arch/xtensa/include/uapi/asm/unistd.h |    2 +-
 arch/xtensa/kernel/process.c          |   25 -------------------------
 4 files changed, 3 insertions(+), 27 deletions(-)

diff --git a/arch/xtensa/include/asm/syscall.h b/arch/xtensa/include/asm/syscall.h
index c1dacca..124aeee 100644
--- a/arch/xtensa/include/asm/syscall.h
+++ b/arch/xtensa/include/asm/syscall.h
@@ -10,7 +10,7 @@
 
 struct pt_regs;
 struct sigaction;
-asmlinkage long xtensa_execve(char*, char**, char**, struct pt_regs*);
+asmlinkage long sys_execve(char*, char**, char**, struct pt_regs*);
 asmlinkage long xtensa_clone(unsigned long, unsigned long, struct pt_regs*);
 asmlinkage long xtensa_ptrace(long, long, long, long);
 asmlinkage long xtensa_sigreturn(struct pt_regs*);
diff --git a/arch/xtensa/include/asm/unistd.h b/arch/xtensa/include/asm/unistd.h
index 9ef1c31..e83e966 100644
--- a/arch/xtensa/include/asm/unistd.h
+++ b/arch/xtensa/include/asm/unistd.h
@@ -8,6 +8,7 @@
  * Copyright (C) 2001 - 2005 Tensilica Inc.
  */
 
+#define __ARCH_WANT_SYS_EXECVE
 #include <uapi/asm/unistd.h>
 
 
diff --git a/arch/xtensa/include/uapi/asm/unistd.h b/arch/xtensa/include/uapi/asm/unistd.h
index ac37eb4..c5d736f 100644
--- a/arch/xtensa/include/uapi/asm/unistd.h
+++ b/arch/xtensa/include/uapi/asm/unistd.h
@@ -269,7 +269,7 @@ __SYSCALL(115, sys_sendmmsg, 4)
 #define __NR_clone 				116
 __SYSCALL(116, xtensa_clone, 5)
 #define __NR_execve 				117
-__SYSCALL(117, xtensa_execve, 3)
+__SYSCALL(117, sys_execve, 3)
 #define __NR_exit 				118
 __SYSCALL(118, sys_exit, 1)
 #define __NR_exit_group 			119
diff --git a/arch/xtensa/kernel/process.c b/arch/xtensa/kernel/process.c
index fd538bb..58e50ff 100644
--- a/arch/xtensa/kernel/process.c
+++ b/arch/xtensa/kernel/process.c
@@ -350,28 +350,3 @@ long xtensa_clone(unsigned long clone_flags, unsigned long newsp,
                 newsp = regs->areg[1];
         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 }
-
-/*
- * xtensa_execve() executes a new program.
- */
-
-asmlinkage
-long xtensa_execve(const char __user *name,
-		   const char __user *const __user *argv,
-                   const char __user *const __user *envp,
-                   long a3, long a4, long a5,
-                   struct pt_regs *regs)
-{
-	long error;
-	struct filename *filename;
-
-	filename = getname(name);
-	error = PTR_ERR(filename);
-	if (IS_ERR(filename))
-		goto out;
-	error = do_execve(filename->name, argv, envp, regs);
-	putname(filename);
-out:
-	return error;
-}
-
-- 
1.7.7.6

Re: [PATCH 3/3] xtensa: switch to generic sys_execve()

[Al Viro]

On Wed, Oct 24, 2012 at 05:48:37AM +0400, Max Filippov wrote:
> Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
> ---
>  arch/xtensa/include/asm/syscall.h     |    2 +-
>  arch/xtensa/include/asm/unistd.h      |    1 +
>  arch/xtensa/include/uapi/asm/unistd.h |    2 +-
>  arch/xtensa/kernel/process.c          |   25 -------------------------
>  4 files changed, 3 insertions(+), 27 deletions(-)

All three applied, but there's a couple of things I'd like to have on top
of that.  Are you OK with the following incremental?
	1) ->{set,clear}_child_tid is set by the caller; no need to set it
in copy_thread().
	2) for later work it would be easier if copy_thread() hadn't looked
at regs argument at all; telling kernel_thread(9) from clone(2) can be done
by checking p->flags & PF_KTHREAD and for clone(2) it's always going to get
current_pt_regs().
	3) interpretation of zero newsp is probably better off in copy_thread()
as well.  Eventually we are going to kill regs argument of do_fork() et.al.

diff --git a/arch/xtensa/kernel/process.c b/arch/xtensa/kernel/process.c
index 58e50ff..9d8f387 100644
--- a/arch/xtensa/kernel/process.c
+++ b/arch/xtensa/kernel/process.c
@@ -200,7 +200,7 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 
 int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
 		unsigned long thread_fn_arg,
-                struct task_struct * p, struct pt_regs * regs)
+                struct task_struct * p, struct pt_regs *unused)
 {
 	struct pt_regs *childregs = task_pt_regs(p);
 
@@ -212,10 +212,10 @@ int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
 	*((int*)childregs - 3) = (unsigned long)childregs;
 	*((int*)childregs - 4) = 0;
 
-	p->set_child_tid = p->clear_child_tid = NULL;
 	p->thread.sp = (unsigned long)childregs;
 
-	if (regs) {
+	if (!(p->flags & PF_KTHREAD)) {
+		struct pt_regs *regs = current_pt_regs();
 		p->thread.ra = MAKE_RA_FOR_CALL(
 				(unsigned long)ret_from_fork, 0x1);
 
@@ -224,7 +224,8 @@ int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
 		 * ARs beyond a0-a15 exist past the end of the struct.
 		 */
 		*childregs = *regs;
-		childregs->areg[1] = usp_thread_fn;
+		if (usp_thread_fn)
+			childregs->areg[1] = usp_thread_fn;
 		childregs->areg[2] = 0;
 
 		if (clone_flags & CLONE_VM) {
@@ -346,7 +347,5 @@ long xtensa_clone(unsigned long clone_flags, unsigned long newsp,
                   void __user *child_tid, long a5,
                   struct pt_regs *regs)
 {
-        if (!newsp)
-                newsp = regs->areg[1];
         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 }

Re: [PATCH 3/3] xtensa: switch to generic sys_execve()

[Max Filippov]

On Wed, Oct 24, 2012 at 6:41 AM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> On Wed, Oct 24, 2012 at 05:48:37AM +0400, Max Filippov wrote:
>> Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
>> ---
>>  arch/xtensa/include/asm/syscall.h     |    2 +-
>>  arch/xtensa/include/asm/unistd.h      |    1 +
>>  arch/xtensa/include/uapi/asm/unistd.h |    2 +-
>>  arch/xtensa/kernel/process.c          |   25 -------------------------
>>  4 files changed, 3 insertions(+), 27 deletions(-)
>
> All three applied, but there's a couple of things I'd like to have on top
> of that.  Are you OK with the following incremental?

Yes. Looks like I need to send v2 to address Chris' comments,
I can put these changes there as well.

>         1) ->{set,clear}_child_tid is set by the caller; no need to set it
> in copy_thread().
>         2) for later work it would be easier if copy_thread() hadn't looked
> at regs argument at all; telling kernel_thread(9) from clone(2) can be done
> by checking p->flags & PF_KTHREAD and for clone(2) it's always going to get
> current_pt_regs().
>         3) interpretation of zero newsp is probably better off in copy_thread()
> as well.  Eventually we are going to kill regs argument of do_fork() et.al.
>
> diff --git a/arch/xtensa/kernel/process.c b/arch/xtensa/kernel/process.c
> index 58e50ff..9d8f387 100644
> --- a/arch/xtensa/kernel/process.c
> +++ b/arch/xtensa/kernel/process.c
> @@ -200,7 +200,7 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
>
>  int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
>                 unsigned long thread_fn_arg,
> -                struct task_struct * p, struct pt_regs * regs)
> +                struct task_struct * p, struct pt_regs *unused)
>  {
>         struct pt_regs *childregs = task_pt_regs(p);
>
> @@ -212,10 +212,10 @@ int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
>         *((int*)childregs - 3) = (unsigned long)childregs;
>         *((int*)childregs - 4) = 0;
>
> -       p->set_child_tid = p->clear_child_tid = NULL;
>         p->thread.sp = (unsigned long)childregs;
>
> -       if (regs) {
> +       if (!(p->flags & PF_KTHREAD)) {
> +               struct pt_regs *regs = current_pt_regs();
>                 p->thread.ra = MAKE_RA_FOR_CALL(
>                                 (unsigned long)ret_from_fork, 0x1);
>
> @@ -224,7 +224,8 @@ int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
>                  * ARs beyond a0-a15 exist past the end of the struct.
>                  */
>                 *childregs = *regs;
> -               childregs->areg[1] = usp_thread_fn;
> +               if (usp_thread_fn)
> +                       childregs->areg[1] = usp_thread_fn;
>                 childregs->areg[2] = 0;
>
>                 if (clone_flags & CLONE_VM) {
> @@ -346,7 +347,5 @@ long xtensa_clone(unsigned long clone_flags, unsigned long newsp,
>                    void __user *child_tid, long a5,
>                    struct pt_regs *regs)
>  {
> -        if (!newsp)
> -                newsp = regs->areg[1];
>          return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
>  }



-- 
Thanks.
-- Max

Re: [PATCH 3/3] xtensa: switch to generic sys_execve()

[Al Viro]

On Wed, Oct 24, 2012 at 01:16:39PM +0400, Max Filippov wrote:
> On Wed, Oct 24, 2012 at 6:41 AM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> > On Wed, Oct 24, 2012 at 05:48:37AM +0400, Max Filippov wrote:
> >> Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
> >> ---
> >>  arch/xtensa/include/asm/syscall.h     |    2 +-
> >>  arch/xtensa/include/asm/unistd.h      |    1 +
> >>  arch/xtensa/include/uapi/asm/unistd.h |    2 +-
> >>  arch/xtensa/kernel/process.c          |   25 -------------------------
> >>  4 files changed, 3 insertions(+), 27 deletions(-)
> >
> > All three applied, but there's a couple of things I'd like to have on top
> > of that.  Are you OK with the following incremental?
> 
> Yes. Looks like I need to send v2 to address Chris' comments,
> I can put these changes there as well.

Please, do.