Re: [PATCH 1/2] x86: separating entry text section

From: Jiri Olsa
Date: Tue Feb 22 2011 - 07:52:49 EST


On Tue, Feb 22, 2011 at 09:09:34AM +0100, Ingo Molnar wrote:
>
> * Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
>
> > Putting x86 entry code to the separate section: .entry.text.
>
> Trying to apply your patch i noticed one detail:
>
> > before patch:
> > 26282174 L1-icache-load-misses ( +- 0.099% ) (scaled from 81.00%)
> > 0.206651959 seconds time elapsed ( +- 0.152% )
> >
> > after patch:
> > 24237651 L1-icache-load-misses ( +- 0.117% ) (scaled from 80.96%)
> > 0.210509948 seconds time elapsed ( +- 0.140% )
>
> So time elapsed actually went up.
>
> hackbench is notoriously unstable when it comes to runtime - and increasing the
> --repeat value only has limited effects on that.
>
> Dropping all system caches:
>
> echo 1 > /proc/sys/vm/drop_caches
>
> Seems to do a better job of 'resetting' system state, but if we put that into the
> measured workload then the results are all over the place (as we now depend on IO
> being done):
>
> # cat hb10
>
> echo 1 > /proc/sys/vm/drop_caches
> ./hackbench 10
>
> # perf stat --repeat 3 ./hb10
>
> Time: 0.097
> Time: 0.095
> Time: 0.101
>
> Performance counter stats for './hb10' (3 runs):
>
> 21.351257 task-clock-msecs # 0.044 CPUs ( +- 27.165% )
> 6 context-switches # 0.000 M/sec ( +- 34.694% )
> 1 CPU-migrations # 0.000 M/sec ( +- 25.000% )
> 410 page-faults # 0.019 M/sec ( +- 0.081% )
> 25,407,650 cycles # 1189.984 M/sec ( +- 49.154% )
> 25,407,650 instructions # 1.000 IPC ( +- 49.154% )
> 5,126,580 branches # 240.107 M/sec ( +- 46.012% )
> 192,272 branch-misses # 3.750 % ( +- 44.911% )
> 901,701 cache-references # 42.232 M/sec ( +- 12.857% )
> 802,767 cache-misses # 37.598 M/sec ( +- 9.282% )
>
> 0.483297792 seconds time elapsed ( +- 31.152% )
>
> So here's a perf stat feature suggestion to solve such measurement problems: a new
> 'pre-run' 'dry' command could be specified that is executed before the real 'hot'
> run is executed. Something like this:
>
> perf stat --pre-run-script ./hb10 --repeat 10 ./hackbench 10
>
> Would do the cache-clearing before each run, it would run hackbench once (dry run)
> and then would run hackbench 10 for real - and would repeat the whole thing 10
> times. Only the 'hot' portion of the run would be measured and displayed in the perf
> stat output event counts.
>
> Another observation:
>
> > 24237651 L1-icache-load-misses ( +- 0.117% ) (scaled from 80.96%)
>
> Could you please do runs that do not display 'scaled from' messages? Since we are
> measuring a relatively small effect here, and scaling adds noise, it would be nice
> to ensure that the effect persists with non-scaled events as well:
>
> You can do that by reducing the number of events that are measured. The PMU can not
> measure all those L1 cache events you listed - so only use the most important one
> and add cycles and instructions to make sure the measurements are comparable:
>
> -e L1-icache-load-misses -e instructions -e cycles
>
> Btw., there's another 'perf stat' feature suggestion: it would be nice if it was
> possible to 'record' a perf stat run, and do a 'perf diff' over it. That would
> compare the two runs all automatically, without you having to do the comparison
> manually.

hi,

I made another test with "reseting" the system state as suggested and
only for cache-misses together with instructions and cycles events.

I can see even bigger drop of icache load misses than before
from 19359739 to 16448709 (about 15%).

The instruction/cycles count is slightly bigger in the patched
kernel run though..

perf stat --repeat 100 -e L1-icache-load-misses -e instructions -e cycles ./hackbench/hackbench 10

-------------------------------------------------------------------------------
before patch:

Performance counter stats for './hackbench/hackbench 10' (100 runs):

19359739 L1-icache-load-misses ( +- 0.313% )
2667528936 instructions # 0.498 IPC ( +- 0.165% )
5352849800 cycles ( +- 0.303% )

0.205402048 seconds time elapsed ( +- 0.299% )

Performance counter stats for './hackbench/hackbench 10' (500 runs):

19417627 L1-icache-load-misses ( +- 0.147% )
2676914223 instructions # 0.497 IPC ( +- 0.079% )
5389516026 cycles ( +- 0.144% )

0.206267711 seconds time elapsed ( +- 0.138% )


-------------------------------------------------------------------------------
after patch:

Performance counter stats for './hackbench/hackbench 10' (100 runs):

16448709 L1-icache-load-misses ( +- 0.426% )
2698406306 instructions # 0.500 IPC ( +- 0.177% )
5393976267 cycles ( +- 0.321% )

0.206072845 seconds time elapsed ( +- 0.276% )

Performance counter stats for './hackbench/hackbench 10' (500 runs):

16490788 L1-icache-load-misses ( +- 0.180% )
2717734941 instructions # 0.502 IPC ( +- 0.079% )
5414756975 cycles ( +- 0.148% )

0.206747566 seconds time elapsed ( +- 0.137% )


Attaching patch with above numbers in comment.

thanks,
jirka


---
Putting x86 entry code to the separate section: .entry.text.

Separating the entry text section seems to have performance
benefits with regards to the instruction cache usage.

Running hackbench showed that the change compresses the icache
footprint. The icache load miss rate went down by about 15%:

before patch:
19417627 L1-icache-load-misses ( +- 0.147% )

after patch:
16490788 L1-icache-load-misses ( +- 0.180% )


Whole perf output follows.

- results for current tip tree:
Performance counter stats for './hackbench/hackbench 10' (500 runs):

19417627 L1-icache-load-misses ( +- 0.147% )
2676914223 instructions # 0.497 IPC ( +- 0.079% )
5389516026 cycles ( +- 0.144% )

0.206267711 seconds time elapsed ( +- 0.138% )

- results for current tip tree with the patch applied are:
Performance counter stats for './hackbench/hackbench 10' (500 runs):

16490788 L1-icache-load-misses ( +- 0.180% )
2717734941 instructions # 0.502 IPC ( +- 0.079% )
5414756975 cycles ( +- 0.148% )

0.206747566 seconds time elapsed ( +- 0.137% )


wbr,
jirka


Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
---
arch/x86/ia32/ia32entry.S | 2 ++
arch/x86/kernel/entry_32.S | 6 ++++--
arch/x86/kernel/entry_64.S | 6 ++++--
arch/x86/kernel/vmlinux.lds.S | 1 +
include/asm-generic/sections.h | 1 +
include/asm-generic/vmlinux.lds.h | 6 ++++++
6 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S
index 0ed7896..50f1630 100644
--- a/arch/x86/ia32/ia32entry.S
+++ b/arch/x86/ia32/ia32entry.S
@@ -25,6 +25,8 @@
#define sysretl_audit ia32_ret_from_sys_call
#endif

+ .section .entry.text, "ax"
+
#define IA32_NR_syscalls ((ia32_syscall_end - ia32_sys_call_table)/8)

.macro IA32_ARG_FIXUP noebp=0
diff --git a/arch/x86/kernel/entry_32.S b/arch/x86/kernel/entry_32.S
index c8b4efa..f5accf8 100644
--- a/arch/x86/kernel/entry_32.S
+++ b/arch/x86/kernel/entry_32.S
@@ -65,6 +65,8 @@
#define sysexit_audit syscall_exit_work
#endif

+ .section .entry.text, "ax"
+
/*
* We use macros for low-level operations which need to be overridden
* for paravirtualization. The following will never clobber any registers:
@@ -788,7 +790,7 @@ ENDPROC(ptregs_clone)
*/
.section .init.rodata,"a"
ENTRY(interrupt)
-.text
+.section .entry.text, "ax"
.p2align 5
.p2align CONFIG_X86_L1_CACHE_SHIFT
ENTRY(irq_entries_start)
@@ -807,7 +809,7 @@ vector=FIRST_EXTERNAL_VECTOR
.endif
.previous
.long 1b
- .text
+ .section .entry.text, "ax"
vector=vector+1
.endif
.endr
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index 891268c..39f8d21 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -61,6 +61,8 @@
#define __AUDIT_ARCH_LE 0x40000000

.code64
+ .section .entry.text, "ax"
+
#ifdef CONFIG_FUNCTION_TRACER
#ifdef CONFIG_DYNAMIC_FTRACE
ENTRY(mcount)
@@ -744,7 +746,7 @@ END(stub_rt_sigreturn)
*/
.section .init.rodata,"a"
ENTRY(interrupt)
- .text
+ .section .entry.text
.p2align 5
.p2align CONFIG_X86_L1_CACHE_SHIFT
ENTRY(irq_entries_start)
@@ -763,7 +765,7 @@ vector=FIRST_EXTERNAL_VECTOR
.endif
.previous
.quad 1b
- .text
+ .section .entry.text
vector=vector+1
.endif
.endr
diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S
index e70cc3d..459dce2 100644
--- a/arch/x86/kernel/vmlinux.lds.S
+++ b/arch/x86/kernel/vmlinux.lds.S
@@ -105,6 +105,7 @@ SECTIONS
SCHED_TEXT
LOCK_TEXT
KPROBES_TEXT
+ ENTRY_TEXT
IRQENTRY_TEXT
*(.fixup)
*(.gnu.warning)
diff --git a/include/asm-generic/sections.h b/include/asm-generic/sections.h
index b3bfabc..c1a1216 100644
--- a/include/asm-generic/sections.h
+++ b/include/asm-generic/sections.h
@@ -11,6 +11,7 @@ extern char _sinittext[], _einittext[];
extern char _end[];
extern char __per_cpu_load[], __per_cpu_start[], __per_cpu_end[];
extern char __kprobes_text_start[], __kprobes_text_end[];
+extern char __entry_text_start[], __entry_text_end[];
extern char __initdata_begin[], __initdata_end[];
extern char __start_rodata[], __end_rodata[];

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index fe77e33..906c3ce 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -424,6 +424,12 @@
*(.kprobes.text) \
VMLINUX_SYMBOL(__kprobes_text_end) = .;

+#define ENTRY_TEXT \
+ ALIGN_FUNCTION(); \
+ VMLINUX_SYMBOL(__entry_text_start) = .; \
+ *(.entry.text) \
+ VMLINUX_SYMBOL(__entry_text_end) = .;
+
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
#define IRQENTRY_TEXT \
ALIGN_FUNCTION(); \
--
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/