[PATCH] purgatory: compile sha256.c only on ia64 with -O0

Discussion:

Richard Weinberger

2013-03-31 20:59:09 UTC

If we have to compile sha256.c on ia64 with -O0 to produce sane code
do it only on ia64.

This change makes executing a new kernel on my AMD geode CPU much faster.
Without the patch "kexec -e" takes 27 seconds, with only 16.

Signed-off-by: Richard Weinberger <***@nod.at>
---
purgatory/Makefile | 15 +--------------
purgatory/arch/ia64/Makefile | 3 +++
2 files changed, 4 insertions(+), 14 deletions(-)

diff --git a/purgatory/Makefile b/purgatory/Makefile
index e39adec..6be8362 100644
--- a/purgatory/Makefile
+++ b/purgatory/Makefile
@@ -16,16 +16,7 @@ PURGATORY_SRCS += purgatory/string.c
dist += purgatory/Makefile $(PURGATORY_SRCS) \
purgatory/include/purgatory.h purgatory/include/string.h

-include $(srcdir)/purgatory/arch/alpha/Makefile
-include $(srcdir)/purgatory/arch/arm/Makefile
-include $(srcdir)/purgatory/arch/i386/Makefile
-include $(srcdir)/purgatory/arch/ia64/Makefile
-include $(srcdir)/purgatory/arch/mips/Makefile
-include $(srcdir)/purgatory/arch/ppc/Makefile
-include $(srcdir)/purgatory/arch/ppc64/Makefile
-include $(srcdir)/purgatory/arch/s390/Makefile
-include $(srcdir)/purgatory/arch/sh/Makefile
-include $(srcdir)/purgatory/arch/x86_64/Makefile
+include $(srcdir)/purgatory/arch/$(ARCH)/Makefile

PURGATORY_SRCS+=$($(ARCH)_PURGATORY_SRCS)

@@ -36,10 +27,6 @@ clean += $(PURGATORY_OBJS) $(PURGATORY_DEPS) $(PURGATORY)

-include $(PURGATORY_DEPS)

-# sha256.c needs to be compiled without optimization, else
-# purgatory fails to execute on ia64.
-purgatory/sha256.o: CFLAGS += -O0
-
purgatory/sha256.o: $(srcdir)/util_lib/sha256.c
mkdir -p $(@D)
$(COMPILE.c) -o $@ $^
diff --git a/purgatory/arch/ia64/Makefile b/purgatory/arch/ia64/Makefile
index 32c3d97..a8a5c62 100644
--- a/purgatory/arch/ia64/Makefile
+++ b/purgatory/arch/ia64/Makefile
@@ -11,3 +11,6 @@ ia64_PURGATORY_EXTRA_CFLAGS = -ffixed-r28
dist += purgatory/arch/ia64/Makefile $(ia64_PURGATORY_SRCS) \
purgatory/arch/ia64/io.h purgatory/arch/ia64/purgatory-ia64.h

+# sha256.c needs to be compiled without optimization, else
+# purgatory fails to execute on ia64.
+purgatory/sha256.o: ia64_PURGATORY_EXTRA_CFLAGS += -O0

--
1.8.1.4

Zhang Yanfei

2013-04-01 01:59:42 UTC

Permalink

The fix is sound.

Reviewed-by: Zhang Yanfei <***@cn.fujitsu.com>

Simon Horman

2013-04-01 05:04:31 UTC

Permalink

Post by Richard Weinberger
If we have to compile sha256.c on ia64 with -O0 to produce sane code
do it only on ia64.
This change makes executing a new kernel on my AMD geode CPU much faster.
Without the patch "kexec -e" takes 27 seconds, with only 16.
---
purgatory/Makefile | 15 +--------------
purgatory/arch/ia64/Makefile | 3 +++
2 files changed, 4 insertions(+), 14 deletions(-)
diff --git a/purgatory/Makefile b/purgatory/Makefile
index e39adec..6be8362 100644
--- a/purgatory/Makefile
+++ b/purgatory/Makefile
@@ -16,16 +16,7 @@ PURGATORY_SRCS += purgatory/string.c
dist += purgatory/Makefile $(PURGATORY_SRCS) \
purgatory/include/purgatory.h purgatory/include/string.h
-include $(srcdir)/purgatory/arch/alpha/Makefile
-include $(srcdir)/purgatory/arch/arm/Makefile
-include $(srcdir)/purgatory/arch/i386/Makefile
-include $(srcdir)/purgatory/arch/ia64/Makefile
-include $(srcdir)/purgatory/arch/mips/Makefile
-include $(srcdir)/purgatory/arch/ppc/Makefile
-include $(srcdir)/purgatory/arch/ppc64/Makefile
-include $(srcdir)/purgatory/arch/s390/Makefile
-include $(srcdir)/purgatory/arch/sh/Makefile
-include $(srcdir)/purgatory/arch/x86_64/Makefile
+include $(srcdir)/purgatory/arch/$(ARCH)/Makefile

I believe the above portion will break make dist my
causing it to omit purgatory/arch/* other than
purgatory/arch/$(ARCH)

Post by Richard Weinberger
PURGATORY_SRCS+=$($(ARCH)_PURGATORY_SRCS)
@@ -36,10 +27,6 @@ clean += $(PURGATORY_OBJS) $(PURGATORY_DEPS) $(PURGATORY)
-include $(PURGATORY_DEPS)
-# sha256.c needs to be compiled without optimization, else
-# purgatory fails to execute on ia64.
-purgatory/sha256.o: CFLAGS += -O0
-
purgatory/sha256.o: $(srcdir)/util_lib/sha256.c
diff --git a/purgatory/arch/ia64/Makefile b/purgatory/arch/ia64/Makefile
index 32c3d97..a8a5c62 100644
--- a/purgatory/arch/ia64/Makefile
+++ b/purgatory/arch/ia64/Makefile
@@ -11,3 +11,6 @@ ia64_PURGATORY_EXTRA_CFLAGS = -ffixed-r28
dist += purgatory/arch/ia64/Makefile $(ia64_PURGATORY_SRCS) \
purgatory/arch/ia64/io.h purgatory/arch/ia64/purgatory-ia64.h
+# sha256.c needs to be compiled without optimization, else
+# purgatory fails to execute on ia64.
+purgatory/sha256.o: ia64_PURGATORY_EXTRA_CFLAGS += -O0
--
1.8.1.4
_______________________________________________
kexec mailing list
http://lists.infradead.org/mailman/listinfo/kexec

Zhang Yanfei

2013-04-01 05:16:57 UTC

Permalink

Post by Simon Horman

I believe the above portion will break make dist my
causing it to omit purgatory/arch/* other than
purgatory/arch/$(ARCH)

Ah, I've missed this in my review, too.

Other than this, the change about purgatory/sha256.o below looks sound,
right?

Thanks
Zhang

Post by Simon Horman

_______________________________________________
kexec mailing list
http://lists.infradead.org/mailman/listinfo/kexec

Richard Weinberger

2013-04-01 10:05:55 UTC

Permalink

Simon, Zhang,

Post by Zhang Yanfei

Post by Simon Horman
I believe the above portion will break make dist my
causing it to omit purgatory/arch/* other than
purgatory/arch/$(ARCH)

Ah, I've missed this in my review, too.
Other than this, the change about purgatory/sha256.o below looks sound,
right?

The dist mechanism is quite strange and error prone.
Why can't you just use git archive in the $(TARBALL) target?
I'll happily send a patch...

Back to the purgatory/sha256.o issue, I did some more tests.
First I thought memcpy() within purgatory is slow, but it isn't.
Does kexec reset the CPU so some kind of legacy mode during purgatory?
It looks like the same code in sha256.o is fast in normal user space
but very slow during purgatory.

Computing sha256 of a 9MiB blob takes less than a second in user space
and 16 seconds (with my patch). Otherwise it takes > 25 seconds.

Thanks,
//richard