Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

Conflicts:
	drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c

Minor conflict between the BCM_CNIC define removal in net-next
and a bug fix added to net.  Based upon a conflict resolution
patch posted by Stephen Rothwell.

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 90 insertions, 57 deletions

diff --git a/arch/xtensa/kernel/process.c b/arch/xtensa/kernel/process.c
index bc020825cce5..09ae7bfab9a7 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, };
 
@@ -158,64 +159,123 @@ 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.
+ *
+ * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
+ * not an entire process), we're normally given a new usp, and we CANNOT share
+ * any live address register windows.  If we just copy those live frames over,
+ * the two threads (parent and child) will overflow the same frames onto the
+ * parent stack at different times, likely corrupting the parent stack (esp.
+ * if the parent returns from functions that called clone() and calls new
+ * ones, before the child overflows its now old copies of its parent windows).
+ * One solution is to spill windows to the parent stack, but that's fairly
+ * involved.  Much simpler to just not copy those live frames across.
  */
 
-int copy_thread(unsigned long clone_flags, unsigned long usp,
-		unsigned long unused,
-                struct task_struct * p, struct pt_regs * regs)
+int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
+		unsigned long thread_fn_arg,
+		struct task_struct *p, struct pt_regs *unused)
 {
-	struct pt_regs *childregs;
-	struct thread_info *ti;
-	unsigned long tos;
-	int user_mode = user_mode(regs);
-
-	/* 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;
+	struct pt_regs *childregs = task_pt_regs(p);
 
-	*childregs = *regs;
+#if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
+	struct thread_info *ti;
+#endif
 
 	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
 	*((int*)childregs - 3) = (unsigned long)childregs;
 	*((int*)childregs - 4) = 0;
 
-	childregs->areg[1] = tos;
-	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 (!(p->flags & PF_KTHREAD)) {
+		struct pt_regs *regs = current_pt_regs();
+		unsigned long usp = usp_thread_fn ?
+			usp_thread_fn : regs->areg[1];
 
-		int len = childregs->wmask & ~0xf;
+		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;
-		memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
-		       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
+		childregs->areg[2] = 0;
+
+		/* When sharing memory with the parent thread, the child
+		   usually starts on a pristine stack, so we have to reset
+		   windowbase, windowstart and wmask.
+		   (Note that such a new thread is required to always create
+		   an initial call4 frame)
+		   The exception is vfork, where the new thread continues to
+		   run on the parent's stack until it calls execve. This could
+		   be a call8 or call12, which requires a legal stack frame
+		   of the previous caller for the overflow handlers to work.
+		   (Note that it's always legal to overflow live registers).
+		   In this case, ensure to spill at least the stack pointer
+		   of that frame. */
+
+		if (clone_flags & CLONE_VM) {
+			/* check that caller window is live and same stack */
+			int len = childregs->wmask & ~0xf;
+			if (regs->areg[1] == usp && len != 0) {
+				int callinc = (regs->areg[0] >> 30) & 3;
+				int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
+				put_user(regs->areg[caller_ars+1],
+					 (unsigned __user*)(usp - 12));
+			}
+			childregs->wmask = 1;
+			childregs->windowstart = 1;
+			childregs->windowbase = 0;
+		} else {
+			int len = childregs->wmask & ~0xf;
+			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...
 		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;
+		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.
+		 */
 	}
 
 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
@@ -311,32 +371,5 @@ 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);
 }
-
-/*
- * 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;
-}
-