[PATCH] Remove unnecessary #GP after flush_thread()

[Arun Sharma]

During execve(), we seem to have a window between flush_thread() and start_thread() where %gs = 0x33, but cpu_gdt_table[cpu][6] == 0. This causes an unnecessary #GP fault on __switch_to() if there is a context switch during the window. But most people are not noticing it because of the fixup in loadsegment().

Specifically, this BUG() was triggered during normal execution.

--- arch/i386/kernel/process.c- 2005-01-05 15:13:51.450321632 -0800
+++ arch/i386/kernel/process.c  2005-01-05 15:13:55.129762272 -0800
@@ -764,6 +764,12 @@
        * Restore %fs and %gs if needed.
        */
       if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
+#if 1
+                if ((next->gs == 0x33)
+                        && (cpu_gdt_table[cpu][6].a == 0)
+                        && (cpu_gdt_table[cpu][6].b == 0))
+                        BUG();
+#endif
               loadsegment(fs, next->fs);
               loadsegment(gs, next->gs);


This patch sets %gs = 0 _before_ the thread.tls_array is zeroed.

Signed-off-by: Arun Sharma <arun.sharma@xxxxxxxxx>

--- linux-2.6.10/arch/i386/kernel/process.c-	2005-01-05 15:15:25.469541136 -0800
+++ linux-2.6.10/arch/i386/kernel/process.c	2005-01-05 15:15:26.939317696 -0800
@@ -329,6 +329,7 @@
 {
 	struct task_struct *tsk = current;
 
+	__asm__("movl %0,%%gs": :"r" (0));
 	memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));	
 	/*