Skip to contentSkip to conversation

Submissions for 2.4.19-pre [x86 Syscall Optimizations (Alexander Khripin)]

PATCH Michael Cohen 3 participants 6 messages Feb 26, 2002

lore
Share & tools
lore
#0 lore raw

This is the sixth of several mails containing patches to be included in 2.4.19. Some are worthy of dicussion prior to inclusion and have been marked as such. The majority of these patches were found on lkml; the remaining ones have URLs listed.

This one originated on lkml.

------
Michael Cohen
OhDarn.net

arch/i386/kernel/entry.S +5 -5
@@ -111,9 +111,9 @@
 	pushl %edi; \
 	pushl %esi; \
 	pushl %edx; \
+	movl $(__KERNEL_DS),%edx; \
 	pushl %ecx; \
 	pushl %ebx; \
-	movl $(__KERNEL_DS),%edx; \
 	movl %edx,%ds; \
 	movl %edx,%es;
 
@@ -161,13 +161,13 @@
 	movl EFLAGS(%esp),%ecx	# and this is cs..
 	movl %eax,EFLAGS(%esp)	#
 	movl %edx,EIP(%esp)	# Now we move them to their "normal" places
-	movl %ecx,CS(%esp)	#
 	movl %esp,%ebx
+	movl %ecx,CS(%esp) 	#
 	pushl %ebx
 	andl $-8192,%ebx	# GET_CURRENT
 	movl exec_domain(%ebx),%edx	# Get the execution domain
-	movl 4(%edx),%edx	# Get the lcall7 handler for the domain
 	pushl $0x7
+	movl 4(%edx),%edx # Get the lcall7 handler for the domain
 	call *%edx
 	addl $4, %esp
 	popl %eax
@@ -182,13 +182,13 @@
 	movl EFLAGS(%esp),%ecx	# and this is cs..
 	movl %eax,EFLAGS(%esp)	#
 	movl %edx,EIP(%esp)	# Now we move them to their "normal" places
-	movl %ecx,CS(%esp)	#
 	movl %esp,%ebx
+	movl %ecx,CS(%esp) 	#
 	pushl %ebx
 	andl $-8192,%ebx	# GET_CURRENT
 	movl exec_domain(%ebx),%edx	# Get the execution domain
-	movl 4(%edx),%edx	# Get the lcall7 handler for the domain
 	pushl $0x27
+	movl 4(%edx),%edx 	# Get the lcall7 handler for the domain
 	call *%edx
 	addl $4, %esp
 	popl %eax
#1 lore raw
Quoted 2 lines

This is the sixth of several mails containing patches to be included in 2.4.19. Some are worthy of dicussion prior to inclusion and have been marked as such. The majority of these patches were found on lkml; the remaining ones have URLs listed. This one originated on lkml.

Credit for the originator and an explanation would be helpful

#2 lore raw
Quoted 4 lines · On Tue, 26 Feb 2002 02:34:46 +0000 (GMT) Alan Cox <[email protected]>…
Quoted 2 lines

This is the sixth of several mails containing patches to be included in 2.4.19. Some are worthy of dicussion prior to inclusion and have been marked as such. The majority of these patches were found on lkml; the remaining ones have URLs listed. This one originated on lkml.

Credit for the originator and an explanation would be helpful

See Subject ^^ Alexander Khripin. Looks to me like it improves latency quite a bit during syscalls. I'm unable to find the exact mail but I believe I was referred to this particular patch by someone on lkml.

------
Michael Cohen
OhDarn.net

#3 lore raw
Quoted 8 lines · Michael Cohen
Quoted 1 line · Alan Cox

Credit for the originator and an explanation would be helpful

See Subject ^^ Alexander Khripin. Looks to me like it improves latency quite a bit during syscalls. I'm unable to find the exact mail but I believe I was referred to this particular patch by someone on lkml.

Eh? The patch doesn't do anything but move movl ops around. I suspect the intention may to eliminate datapath stalls, which may be a fine micro-optimization, but I think we need some numbers first ...

Robert Love

#4 lore raw
Quoted 7 lines · On 25 Feb 2002 22:19:17 -0500 Robert Love <[email protected]> wrote:
Quoted 3 lines

See Subject ^^ Alexander Khripin. Looks to me like it improves latency quite a bit during syscalls. I'm unable to find the exact mail but I believe I was referred to this particular patch by someone on lkml.

Eh? The patch doesn't do anything but move movl ops around. I suspect the intention may to eliminate datapath stalls, which may be a fine micro-optimization, but I think we need some numbers first ...

UNIXbench numbers on the way :)

------
Michael Cohen

#5 lore raw
Quoted 5 lines
Quoted 3 lines

Eh? The patch doesn't do anything but move movl ops around. I suspect the intention may to eliminate datapath stalls, which may be a fine micro-optimization, but I think we need some numbers first ...

UNIXbench numbers on the way :)

Remember to get them with more than one vendor and family of processors. Quite a lot of proposed micro-optimisations turn out to be pessimisations except for the processor of the original proposer 8(