[PATCH 2/2] i386 TIF flags for debug regs and io bitmap in ctxsw

[PATCH 2/2] i386 TIF flags for debug regs and io bitmap in ctxsw

[Stephane Eranian]

Hello,

Following my discussion with Andi. Here is a patch that introduces
two new TIF flags to simplify the context switch code in __switch_to().
The idea is to minimize the number of cache lines accessed in the common
case, i.e., when neither the debug registers nor the I/O bitmap are used.

This patch covers the i386 modifications.

Changelog:
	- add TIF_DEBUG to track when debug registers are active
	- add TIF_IO_BITMAP to track when I/O bitmap is used
	- modify __switch_to() to use the new TIF flags

<signed-off-by>: eranian@hpl.hp.com

diff -urNp linux-2.6.17.2.orig/arch/i386/kernel/ioport.c linux-2.6.17.2-tif/arch/i386/kernel/ioport.c
--- linux-2.6.17.2.orig/arch/i386/kernel/ioport.c	2006-06-17 18:49:35.000000000 -0700
+++ linux-2.6.17.2-tif/arch/i386/kernel/ioport.c	2006-07-03 09:28:07.000000000 -0700
@@ -79,6 +79,7 @@ asmlinkage long sys_ioperm(unsigned long
 
 		memset(bitmap, 0xff, IO_BITMAP_BYTES);
 		t->io_bitmap_ptr = bitmap;
+		set_thread_flag(TIF_IO_BITMAP);
 	}
 
 	/*
diff -urNp linux-2.6.17.2.orig/arch/i386/kernel/process.c linux-2.6.17.2-tif/arch/i386/kernel/process.c
--- linux-2.6.17.2.orig/arch/i386/kernel/process.c	2006-06-17 18:49:35.000000000 -0700
+++ linux-2.6.17.2-tif/arch/i386/kernel/process.c	2006-07-04 00:06:16.000000000 -0700
@@ -370,6 +370,7 @@ void exit_thread(void)
 
 		kfree(t->io_bitmap_ptr);
 		t->io_bitmap_ptr = NULL;
+		clear_thread_flag(TIF_IO_BITMAP);
 		/*
 		 * Careful, clear this in the TSS too:
 		 */
@@ -388,6 +389,7 @@ void flush_thread(void)
 
 	memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));	
+	clear_tsk_thread_flag(tsk, TIF_DEBUG);
 	/*
 	 * Forget coprocessor state..
 	 */
@@ -432,7 +434,7 @@ int copy_thread(int nr, unsigned long cl
 	savesegment(gs,p->thread.gs);
 
 	tsk = current;
-	if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
+	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 		p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 		if (!p->thread.io_bitmap_ptr) {
 			p->thread.io_bitmap_max = 0;
@@ -440,6 +442,7 @@ int copy_thread(int nr, unsigned long cl
 		}
 		memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 			IO_BITMAP_BYTES);
+		set_tsk_thread_flag(p, TIF_IO_BITMAP);
 	}
 
 	/*
@@ -534,10 +537,24 @@ int dump_task_regs(struct task_struct *t
 	return 1;
 }
 
-static inline void
-handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
+static inline void __switch_to_xtra(struct task_struct *next_p,
+				    struct tss_struct *tss)
 {
-	if (!next->io_bitmap_ptr) {
+	struct thread_struct *next;
+
+	next = &next_p->thread;
+
+	if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
+		set_debugreg(next->debugreg[0], 0);
+		set_debugreg(next->debugreg[1], 1);
+		set_debugreg(next->debugreg[2], 2);
+		set_debugreg(next->debugreg[3], 3);
+		/* no 4 and 5 */
+		set_debugreg(next->debugreg[6], 6);
+		set_debugreg(next->debugreg[7], 7);
+	}
+
+	if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP) == 0) {
 		/*
 		 * Disable the bitmap via an invalid offset. We still cache
 		 * the previous bitmap owner and the IO bitmap contents:
@@ -545,6 +562,7 @@ handle_io_bitmap(struct thread_struct *n
 		tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 		return;
 	}
+
 	if (likely(next == tss->io_bitmap_owner)) {
 		/*
 		 * Previous owner of the bitmap (hence the bitmap content)
@@ -674,18 +692,9 @@ struct task_struct fastcall * __switch_t
 	/*
 	 * Now maybe reload the debug registers
 	 */
-	if (unlikely(next->debugreg[7])) {
-		set_debugreg(next->debugreg[0], 0);
-		set_debugreg(next->debugreg[1], 1);
-		set_debugreg(next->debugreg[2], 2);
-		set_debugreg(next->debugreg[3], 3);
-		/* no 4 and 5 */
-		set_debugreg(next->debugreg[6], 6);
-		set_debugreg(next->debugreg[7], 7);
-	}
-
-	if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
-		handle_io_bitmap(next, tss);
+	if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
+	    || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
+		__switch_to_xtra(next_p, tss);
 
 	disable_tsc(prev_p, next_p);
 
diff -urNp linux-2.6.17.2.orig/arch/i386/kernel/ptrace.c linux-2.6.17.2-tif/arch/i386/kernel/ptrace.c
--- linux-2.6.17.2.orig/arch/i386/kernel/ptrace.c	2006-06-17 18:49:35.000000000 -0700
+++ linux-2.6.17.2-tif/arch/i386/kernel/ptrace.c	2006-07-04 00:19:59.000000000 -0700
@@ -468,8 +468,11 @@ long arch_ptrace(struct task_struct *chi
 				  for(i=0; i<4; i++)
 					  if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 						  goto out_tsk;
+				  if (data)
+					  set_tsk_thread_flag(child, TIF_DEBUG);
+				  else
+					  clear_tsk_thread_flag(child, TIF_DEBUG);
 			  }
-
 			  addr -= (long) &dummy->u_debugreg;
 			  addr = addr >> 2;
 			  child->thread.debugreg[addr] = data;
diff -urNp linux-2.6.17.2.orig/include/asm-i386/thread_info.h linux-2.6.17.2-tif/include/asm-i386/thread_info.h
--- linux-2.6.17.2.orig/include/asm-i386/thread_info.h	2006-06-17 18:49:35.000000000 -0700
+++ linux-2.6.17.2-tif/include/asm-i386/thread_info.h	2006-07-03 05:51:25.000000000 -0700
@@ -143,6 +143,8 @@ register unsigned long current_stack_poi
 #define TIF_RESTORE_SIGMASK	9	/* restore signal mask in do_signal() */
 #define TIF_POLLING_NRFLAG	16	/* true if poll_idle() is polling TIF_NEED_RESCHED */
 #define TIF_MEMDIE		17
+#define TIF_DEBUG		18	/* uses debug registers */
+#define TIF_IO_BITMAP		19	/* uses I/O bitmap */
 
 #define _TIF_SYSCALL_TRACE	(1<<TIF_SYSCALL_TRACE)
 #define _TIF_NOTIFY_RESUME	(1<<TIF_NOTIFY_RESUME)
@@ -155,6 +157,8 @@ register unsigned long current_stack_poi
 #define _TIF_SECCOMP		(1<<TIF_SECCOMP)
 #define _TIF_RESTORE_SIGMASK	(1<<TIF_RESTORE_SIGMASK)
 #define _TIF_POLLING_NRFLAG	(1<<TIF_POLLING_NRFLAG)
+#define _TIF_DEBUG		(1<<TIF_DEBUG)
+#define _TIF_IO_BITMAP		(1<<TIF_IO_BITMAP)
 
 /* work to do on interrupt/exception return */
 #define _TIF_WORK_MASK \
@@ -163,6 +167,9 @@ register unsigned long current_stack_poi
 /* work to do on any return to u-space */
 #define _TIF_ALLWORK_MASK	(0x0000FFFF & ~_TIF_SECCOMP)
 
+/* flags to check in __switch_to() */
+#define _TIF_WORK_CTXSW (_TIF_DEBUG|_TIF_IO_BITMAP)
+
 /*
  * Thread-synchronous status.
  *

[PATCH] i386 TIF flags for debug regs and io bitmap in ctxsw (v2)

[Stephane Eranian]

Hello,

Here is the second version of my TIF flag patch for i386.
This version incoporates the changes suggested by Chuck. 

This patch is against mainline.

This patch covers the i386 modifications.
 
Changelog:
	- add TIF_DEBUG to track when debug registers are active
 	- add TIF_IO_BITMAP to track when I/O bitmap is used
 	- modify __switch_to() to use the new TIF flags
 
<signed-off-by>: eranian@hpl.hp.com

diff --git a/arch/i386/kernel/ioport.c b/arch/i386/kernel/ioport.c
--- a/arch/i386/kernel/ioport.c
+++ b/arch/i386/kernel/ioport.c
@@ -79,6 +79,7 @@ asmlinkage long sys_ioperm(unsigned long
 
 		memset(bitmap, 0xff, IO_BITMAP_BYTES);
 		t->io_bitmap_ptr = bitmap;
+		set_thread_flag(TIF_IO_BITMAP);
 	}
 
 	/*
diff --git a/arch/i386/kernel/process.c b/arch/i386/kernel/process.c
--- a/arch/i386/kernel/process.c
+++ b/arch/i386/kernel/process.c
@@ -359,16 +359,16 @@ EXPORT_SYMBOL(kernel_thread);
  */
 void exit_thread(void)
 {
-	struct task_struct *tsk = current;
-	struct thread_struct *t = &tsk->thread;
-
 	/* The process may have allocated an io port bitmap... nuke it. */
-	if (unlikely(NULL != t->io_bitmap_ptr)) {
+	if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
+		struct task_struct *tsk = current;
+		struct thread_struct *t = &tsk->thread;
 		int cpu = get_cpu();
 		struct tss_struct *tss = &per_cpu(init_tss, cpu);
 
 		kfree(t->io_bitmap_ptr);
 		t->io_bitmap_ptr = NULL;
+		clear_thread_flag(TIF_IO_BITMAP);
 		/*
 		 * Careful, clear this in the TSS too:
 		 */
@@ -387,6 +387,7 @@ void flush_thread(void)
 
 	memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));	
+	clear_tsk_thread_flag(tsk, TIF_DEBUG);
 	/*
 	 * Forget coprocessor state..
 	 */
@@ -431,7 +432,7 @@ int copy_thread(int nr, unsigned long cl
 	savesegment(gs,p->thread.gs);
 
 	tsk = current;
-	if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
+	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 		p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 		if (!p->thread.io_bitmap_ptr) {
 			p->thread.io_bitmap_max = 0;
@@ -439,6 +440,7 @@ int copy_thread(int nr, unsigned long cl
 		}
 		memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 			IO_BITMAP_BYTES);
+		set_tsk_thread_flag(p, TIF_IO_BITMAP);
 	}
 
 	/*
@@ -533,10 +535,24 @@ int dump_task_regs(struct task_struct *t
 	return 1;
 }
 
-static inline void
-handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
+static inline void __switch_to_xtra(struct task_struct *next_p,
+				    struct tss_struct *tss)
 {
-	if (!next->io_bitmap_ptr) {
+	struct thread_struct *next;
+
+	next = &next_p->thread;
+
+	if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
+		set_debugreg(next->debugreg[0], 0);
+		set_debugreg(next->debugreg[1], 1);
+		set_debugreg(next->debugreg[2], 2);
+		set_debugreg(next->debugreg[3], 3);
+		/* no 4 and 5 */
+		set_debugreg(next->debugreg[6], 6);
+		set_debugreg(next->debugreg[7], 7);
+	}
+
+	if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 		/*
 		 * Disable the bitmap via an invalid offset. We still cache
 		 * the previous bitmap owner and the IO bitmap contents:
@@ -544,6 +560,7 @@ handle_io_bitmap(struct thread_struct *n
 		tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 		return;
 	}
+
 	if (likely(next == tss->io_bitmap_owner)) {
 		/*
 		 * Previous owner of the bitmap (hence the bitmap content)
@@ -673,18 +690,9 @@ struct task_struct fastcall * __switch_t
 	/*
 	 * Now maybe reload the debug registers
 	 */
-	if (unlikely(next->debugreg[7])) {
-		set_debugreg(next->debugreg[0], 0);
-		set_debugreg(next->debugreg[1], 1);
-		set_debugreg(next->debugreg[2], 2);
-		set_debugreg(next->debugreg[3], 3);
-		/* no 4 and 5 */
-		set_debugreg(next->debugreg[6], 6);
-		set_debugreg(next->debugreg[7], 7);
-	}
-
-	if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
-		handle_io_bitmap(next, tss);
+	if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
+	    || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
+		__switch_to_xtra(next_p, tss);
 
 	disable_tsc(prev_p, next_p);
 
diff --git a/arch/i386/kernel/ptrace.c b/arch/i386/kernel/ptrace.c
--- a/arch/i386/kernel/ptrace.c
+++ b/arch/i386/kernel/ptrace.c
@@ -468,8 +468,11 @@ long arch_ptrace(struct task_struct *chi
 				  for(i=0; i<4; i++)
 					  if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 						  goto out_tsk;
+				  if (data)
+					  set_tsk_thread_flag(child, TIF_DEBUG);
+				  else
+					  clear_tsk_thread_flag(child, TIF_DEBUG);
 			  }
-
 			  addr -= (long) &dummy->u_debugreg;
 			  addr = addr >> 2;
 			  child->thread.debugreg[addr] = data;
diff --git a/include/asm-i386/thread_info.h b/include/asm-i386/thread_info.h
--- a/include/asm-i386/thread_info.h
+++ b/include/asm-i386/thread_info.h
@@ -140,6 +140,8 @@ static inline struct thread_info *curren
 #define TIF_SECCOMP		8	/* secure computing */
 #define TIF_RESTORE_SIGMASK	9	/* restore signal mask in do_signal() */
 #define TIF_MEMDIE		16
+#define TIF_DEBUG		17	/* uses debug registers */
+#define TIF_IO_BITMAP		18	/* uses I/O bitmap */
 
 #define _TIF_SYSCALL_TRACE	(1<<TIF_SYSCALL_TRACE)
 #define _TIF_NOTIFY_RESUME	(1<<TIF_NOTIFY_RESUME)
@@ -151,6 +153,8 @@ static inline struct thread_info *curren
 #define _TIF_SYSCALL_AUDIT	(1<<TIF_SYSCALL_AUDIT)
 #define _TIF_SECCOMP		(1<<TIF_SECCOMP)
 #define _TIF_RESTORE_SIGMASK	(1<<TIF_RESTORE_SIGMASK)
+#define _TIF_DEBUG		(1<<TIF_DEBUG)
+#define _TIF_IO_BITMAP		(1<<TIF_IO_BITMAP)
 
 /* work to do on interrupt/exception return */
 #define _TIF_WORK_MASK \
@@ -159,6 +163,9 @@ static inline struct thread_info *curren
 /* work to do on any return to u-space */
 #define _TIF_ALLWORK_MASK	(0x0000FFFF & ~_TIF_SECCOMP)
 
+/* flags to check in __switch_to() */
+#define _TIF_WORK_CTXSW (_TIF_DEBUG|_TIF_IO_BITMAP)
+
 /*
  * Thread-synchronous status.
  *

Re: [PATCH] i386 TIF flags for debug regs and io bitmap in ctxsw (v2)

[Stephane Eranian]

Hello,

I would like to follow-up the TIF flags and especially on the
rules of inheritance. It appears the TIF flags are copied from
parent to child task systematically on copy_process.
Then they are adjusted in copy_threads() or sub-functions.

The TIF_IO_BITMAP is checked in copy_threads() with the following code:

int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
	unsigned long unused,
	struct task_struct * p, struct pt_regs * regs)
{
	....
	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
		p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
		if (!p->thread.io_bitmap_ptr) {
			p->thread.io_bitmap_max = 0;
			return -ENOMEM;
		}
		memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
			IO_BITMAP_BYTES);
		set_tsk_thread_flag(p, TIF_IO_BITMAP);
	}

I would think that the set_tsk_thread_flag() is extraneous.

As for TIF_DEBUG, my patch is not clearing it. I don't think you can
have HW breakpoints be inherited from one task to the other.

Any comments on this?


On Wed, Jul 05, 2006 at 04:54:06AM -0700, Stephane Eranian wrote:
> Hello,
> 
> Here is the second version of my TIF flag patch for i386.
> This version incoporates the changes suggested by Chuck. 
> 
> This patch is against mainline.
> 
> This patch covers the i386 modifications.
>  
> Changelog:
> 	- add TIF_DEBUG to track when debug registers are active
>  	- add TIF_IO_BITMAP to track when I/O bitmap is used
>  	- modify __switch_to() to use the new TIF flags
>  
> <signed-off-by>: eranian@hpl.hp.com
> 
> diff --git a/arch/i386/kernel/ioport.c b/arch/i386/kernel/ioport.c
> --- a/arch/i386/kernel/ioport.c
> +++ b/arch/i386/kernel/ioport.c
> @@ -79,6 +79,7 @@ asmlinkage long sys_ioperm(unsigned long
>  
>  		memset(bitmap, 0xff, IO_BITMAP_BYTES);
>  		t->io_bitmap_ptr = bitmap;
> +		set_thread_flag(TIF_IO_BITMAP);
>  	}
>  
>  	/*
> diff --git a/arch/i386/kernel/process.c b/arch/i386/kernel/process.c
> --- a/arch/i386/kernel/process.c
> +++ b/arch/i386/kernel/process.c
> @@ -359,16 +359,16 @@ EXPORT_SYMBOL(kernel_thread);
>   */
>  void exit_thread(void)
>  {
> -	struct task_struct *tsk = current;
> -	struct thread_struct *t = &tsk->thread;
> -
>  	/* The process may have allocated an io port bitmap... nuke it. */
> -	if (unlikely(NULL != t->io_bitmap_ptr)) {
> +	if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
> +		struct task_struct *tsk = current;
> +		struct thread_struct *t = &tsk->thread;
>  		int cpu = get_cpu();
>  		struct tss_struct *tss = &per_cpu(init_tss, cpu);
>  
>  		kfree(t->io_bitmap_ptr);
>  		t->io_bitmap_ptr = NULL;
> +		clear_thread_flag(TIF_IO_BITMAP);
>  		/*
>  		 * Careful, clear this in the TSS too:
>  		 */
> @@ -387,6 +387,7 @@ void flush_thread(void)
>  
>  	memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
>  	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));	
> +	clear_tsk_thread_flag(tsk, TIF_DEBUG);
>  	/*
>  	 * Forget coprocessor state..
>  	 */
> @@ -431,7 +432,7 @@ int copy_thread(int nr, unsigned long cl
>  	savesegment(gs,p->thread.gs);
>  
>  	tsk = current;
> -	if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
> +	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
>  		p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
>  		if (!p->thread.io_bitmap_ptr) {
>  			p->thread.io_bitmap_max = 0;
> @@ -439,6 +440,7 @@ int copy_thread(int nr, unsigned long cl
>  		}
>  		memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
>  			IO_BITMAP_BYTES);
> +		set_tsk_thread_flag(p, TIF_IO_BITMAP);
>  	}
>  
>  	/*
> @@ -533,10 +535,24 @@ int dump_task_regs(struct task_struct *t
>  	return 1;
>  }
>  
> -static inline void
> -handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
> +static inline void __switch_to_xtra(struct task_struct *next_p,
> +				    struct tss_struct *tss)
>  {
> -	if (!next->io_bitmap_ptr) {
> +	struct thread_struct *next;
> +
> +	next = &next_p->thread;
> +
> +	if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
> +		set_debugreg(next->debugreg[0], 0);
> +		set_debugreg(next->debugreg[1], 1);
> +		set_debugreg(next->debugreg[2], 2);
> +		set_debugreg(next->debugreg[3], 3);
> +		/* no 4 and 5 */
> +		set_debugreg(next->debugreg[6], 6);
> +		set_debugreg(next->debugreg[7], 7);
> +	}
> +
> +	if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
>  		/*
>  		 * Disable the bitmap via an invalid offset. We still cache
>  		 * the previous bitmap owner and the IO bitmap contents:
> @@ -544,6 +560,7 @@ handle_io_bitmap(struct thread_struct *n
>  		tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
>  		return;
>  	}
> +
>  	if (likely(next == tss->io_bitmap_owner)) {
>  		/*
>  		 * Previous owner of the bitmap (hence the bitmap content)
> @@ -673,18 +690,9 @@ struct task_struct fastcall * __switch_t
>  	/*
>  	 * Now maybe reload the debug registers
>  	 */
> -	if (unlikely(next->debugreg[7])) {
> -		set_debugreg(next->debugreg[0], 0);
> -		set_debugreg(next->debugreg[1], 1);
> -		set_debugreg(next->debugreg[2], 2);
> -		set_debugreg(next->debugreg[3], 3);
> -		/* no 4 and 5 */
> -		set_debugreg(next->debugreg[6], 6);
> -		set_debugreg(next->debugreg[7], 7);
> -	}
> -
> -	if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
> -		handle_io_bitmap(next, tss);
> +	if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
> +	    || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
> +		__switch_to_xtra(next_p, tss);
>  
>  	disable_tsc(prev_p, next_p);
>  
> diff --git a/arch/i386/kernel/ptrace.c b/arch/i386/kernel/ptrace.c
> --- a/arch/i386/kernel/ptrace.c
> +++ b/arch/i386/kernel/ptrace.c
> @@ -468,8 +468,11 @@ long arch_ptrace(struct task_struct *chi
>  				  for(i=0; i<4; i++)
>  					  if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
>  						  goto out_tsk;
> +				  if (data)
> +					  set_tsk_thread_flag(child, TIF_DEBUG);
> +				  else
> +					  clear_tsk_thread_flag(child, TIF_DEBUG);
>  			  }
> -
>  			  addr -= (long) &dummy->u_debugreg;
>  			  addr = addr >> 2;
>  			  child->thread.debugreg[addr] = data;
> diff --git a/include/asm-i386/thread_info.h b/include/asm-i386/thread_info.h
> --- a/include/asm-i386/thread_info.h
> +++ b/include/asm-i386/thread_info.h
> @@ -140,6 +140,8 @@ static inline struct thread_info *curren
>  #define TIF_SECCOMP		8	/* secure computing */
>  #define TIF_RESTORE_SIGMASK	9	/* restore signal mask in do_signal() */
>  #define TIF_MEMDIE		16
> +#define TIF_DEBUG		17	/* uses debug registers */
> +#define TIF_IO_BITMAP		18	/* uses I/O bitmap */
>  
>  #define _TIF_SYSCALL_TRACE	(1<<TIF_SYSCALL_TRACE)
>  #define _TIF_NOTIFY_RESUME	(1<<TIF_NOTIFY_RESUME)
> @@ -151,6 +153,8 @@ static inline struct thread_info *curren
>  #define _TIF_SYSCALL_AUDIT	(1<<TIF_SYSCALL_AUDIT)
>  #define _TIF_SECCOMP		(1<<TIF_SECCOMP)
>  #define _TIF_RESTORE_SIGMASK	(1<<TIF_RESTORE_SIGMASK)
> +#define _TIF_DEBUG		(1<<TIF_DEBUG)
> +#define _TIF_IO_BITMAP		(1<<TIF_IO_BITMAP)
>  
>  /* work to do on interrupt/exception return */
>  #define _TIF_WORK_MASK \
> @@ -159,6 +163,9 @@ static inline struct thread_info *curren
>  /* work to do on any return to u-space */
>  #define _TIF_ALLWORK_MASK	(0x0000FFFF & ~_TIF_SECCOMP)
>  
> +/* flags to check in __switch_to() */
> +#define _TIF_WORK_CTXSW (_TIF_DEBUG|_TIF_IO_BITMAP)
> +
>  /*
>   * Thread-synchronous status.
>   *

-- 

-Stephane

Re: [PATCH 2/2] i386 TIF flags for debug regs and io bitmap in ctxsw

[Chuck Ebbert]

In-Reply-To: <20060704072939.GC5902@frankl.hpl.hp.com>

On Tue, 4 Jul 2006 00:29:39 -0700, Stephane Eranian wrote:

> Following my discussion with Andi. Here is a patch that introduces
> two new TIF flags to simplify the context switch code in __switch_to().
> The idea is to minimize the number of cache lines accessed in the common
> case, i.e., when neither the debug registers nor the I/O bitmap are used.

I get a 5-10% speedup in task switch times with this patch.
Some very minor comments:


> <signed-off-by>: eranian@hpl.hp.com

Should be: Signed-off-by: Stephane Eranian <eranian@hpl.hp.com>


> +	if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP) == 0) {

preferred:

	if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {


> @@ -674,18 +692,9 @@ struct task_struct fastcall * __switch_t
>  	/*
>  	 * Now maybe reload the debug registers
>  	 */

 	/*
 	 * Now maybe reload the debug registers and/or IO bitmap
 	 */


And this should be added to the patch:

--- 2.6.17-nb.orig/arch/i386/kernel/process.c
+++ 2.6.17-nb/arch/i386/kernel/process.c
@@ -360,13 +360,12 @@ EXPORT_SYMBOL(kernel_thread);
  */
 void exit_thread(void)
 {
-	struct task_struct *tsk = current;
-	struct thread_struct *t = &tsk->thread;
-
 	/* The process may have allocated an io port bitmap... nuke it. */
-	if (unlikely(NULL != t->io_bitmap_ptr)) {
+	if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
 		int cpu = get_cpu();
 		struct tss_struct *tss = &per_cpu(init_tss, cpu);
+		struct task_struct *tsk = current;
+		struct thread_struct *t = &tsk->thread;
 
 		kfree(t->io_bitmap_ptr);
 		t->io_bitmap_ptr = NULL;
-- 
Chuck
 "You can't read a newspaper if you can't read."  --George W. Bush

Re: [PATCH 2/2] i386 TIF flags for debug regs and io bitmap in ctxsw

[Andi Kleen]

On Tuesday 04 July 2006 23:19, Chuck Ebbert wrote:
> In-Reply-To: <20060704072939.GC5902@frankl.hpl.hp.com>
>
> On Tue, 4 Jul 2006 00:29:39 -0700, Stephane Eranian wrote:
> > Following my discussion with Andi. Here is a patch that introduces
> > two new TIF flags to simplify the context switch code in __switch_to().
> > The idea is to minimize the number of cache lines accessed in the common
> > case, i.e., when neither the debug registers nor the I/O bitmap are used.
>
> I get a 5-10% speedup in task switch times with this patch.

That sounds too much. How did you measure it?

Note that lmbench tends to be unstable for this.

-Andi

Re: [PATCH 2/2] i386 TIF flags for debug regs and io bitmap in ctxsw

[Stephane Eranian]

On Tue, Jul 04, 2006 at 11:47:00PM +0200, Andi Kleen wrote:
> On Tuesday 04 July 2006 23:19, Chuck Ebbert wrote:
> > In-Reply-To: <20060704072939.GC5902@frankl.hpl.hp.com>
> >
> > On Tue, 4 Jul 2006 00:29:39 -0700, Stephane Eranian wrote:
> > > Following my discussion with Andi. Here is a patch that introduces
> > > two new TIF flags to simplify the context switch code in __switch_to().
> > > The idea is to minimize the number of cache lines accessed in the common
> > > case, i.e., when neither the debug registers nor the I/O bitmap are used.
> >
> > I get a 5-10% speedup in task switch times with this patch.
> 
> That sounds too much. How did you measure it?
> 
> Note that lmbench tends to be unstable for this.
> 
Yes, that is my observation as well. Anybody knows of a better
benchmark for ctxsw?

-- 

-Stephane

Re: [PATCH 2/2] i386 TIF flags for debug regs and io bitmap in ctxsw

[Stephane Eranian]

Chuck,

On Tue, Jul 04, 2006 at 05:19:02PM -0400, Chuck Ebbert wrote:
> 
> > +	if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP) == 0) {
> 
> preferred:
> 
> 	if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
> 
Ok, I will fix that.

> And this should be added to the patch:
> 
> --- 2.6.17-nb.orig/arch/i386/kernel/process.c
> +++ 2.6.17-nb/arch/i386/kernel/process.c
> @@ -360,13 +360,12 @@ EXPORT_SYMBOL(kernel_thread);
>   */
>  void exit_thread(void)
>  {
> -	struct task_struct *tsk = current;
> -	struct thread_struct *t = &tsk->thread;
> -
>  	/* The process may have allocated an io port bitmap... nuke it. */
> -	if (unlikely(NULL != t->io_bitmap_ptr)) {
> +	if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
>  		int cpu = get_cpu();
>  		struct tss_struct *tss = &per_cpu(init_tss, cpu);
> +		struct task_struct *tsk = current;
> +		struct thread_struct *t = &tsk->thread;
>  
>  		kfree(t->io_bitmap_ptr);
>  		t->io_bitmap_ptr = NULL;

Yes, I missed that test. Thanks for catching it.

-- 

-Stephane

Re: [PATCH 2/2] i386 TIF flags for debug regs and io bitmap in ctxsw

[Chuck Ebbert]

In-Reply-To: <200607042347.00598.ak@suse.de>

On Tue, 4 Jul 2006 23:47:00 +0200, Andi Kleen wrote:

> > I get a 5-10% speedup in task switch times with this patch.
>
> That sounds too much. How did you measure it?
>
> Note that lmbench tends to be unstable for this.

I used lmbench's lat_ctx(8).  According to its manpage, you can make
multiple runs and use the minimum, not the average.

And the gain may have been high because I tested on an old PII notebook
with small cache and slow memory.

-- 
Chuck
 "You can't read a newspaper if you can't read."  --George W. Bush