Thread (4 messages) flat view 4 messages, 2 authors, 2020-08-11

Re: [RFC PATCH v1] power: don't manage floating point regs when no FPU

From: Michael Ellerman <mpe@ellerman.id.au>
Date: 2020-08-11 12:07:10
Also in: lkml

Christophe Leroy [off-list ref] writes:
There is no point in copying floating point regs when there
is no FPU and MATH_EMULATION is not selected.
Yeah I guess you're right. I've never touched a system with neither, but
if such a thing exists then it does seem silly to copy regs around that
can't be used.
Create a new CONFIG_PPC_FPU_REGS bool that is selected by
CONFIG_MATH_EMULATION and CONFIG_PPC_FPU, and use it to
opt out everything related to fp_state in thread_struct.

The following app runs in approx 10.50 seconds on an 8xx without
the patch, and in 9.45 seconds with the patch.

	void sigusr1(int sig) { }

	int main(int argc, char **argv)
	{
		int i = 100000;

		signal(SIGUSR1, sigusr1);
		for (;i--;)
			raise(SIGUSR1);
		exit(0);
	}

Signed-off-by: Christophe Leroy <redacted>
---
 arch/powerpc/Kconfig                      |  1 +
 arch/powerpc/include/asm/processor.h      |  2 ++
 arch/powerpc/kernel/asm-offsets.c         |  2 ++
 arch/powerpc/kernel/process.c             |  4 ++++
 arch/powerpc/kernel/ptrace/ptrace-novsx.c |  8 ++++++++
 arch/powerpc/kernel/ptrace/ptrace.c       |  4 ++++
 arch/powerpc/kernel/signal.c              | 12 +++++++++++-
 arch/powerpc/kernel/signal_32.c           |  4 ++++
 arch/powerpc/kernel/traps.c               |  4 ++++
 arch/powerpc/platforms/Kconfig.cputype    |  4 ++++
 10 files changed, 44 insertions(+), 1 deletion(-)
In general this looks fine.

It's a bit #ifdef heavy. Maybe some of those can be cleaned up a bit
with some wrapper inlines?
quoted hunk ↗ jump to hunk
diff --git a/arch/powerpc/kernel/ptrace/ptrace-novsx.c b/arch/powerpc/kernel/ptrace/ptrace-novsx.c
index b2dc4e92d11a..8f87a11f3f8c 100644
--- a/arch/powerpc/kernel/ptrace/ptrace-novsx.c
+++ b/arch/powerpc/kernel/ptrace/ptrace-novsx.c
@@ -28,6 +29,9 @@ int fpr_get(struct task_struct *target, const struct user_regset *regset,
 
 	return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 				   &target->thread.fp_state, 0, -1);
+#else
+	return 0;
+#endif
Should we return -ENODEV/EIO here? Wonder if another arch can give us a clue.

cheers
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help