Skip to content

Commit

Permalink
ARM: vfp: fix saving d16-d31 vfp registers on v6+ kernels
Browse files Browse the repository at this point in the history
Michael Olbrich reported that his test program fails when built with
-O2 -mcpu=cortex-a8 -mfpu=neon, and a kernel which supports v6 and v7
CPUs:

volatile int x = 2;
volatile int64_t y = 2;

int main() {
	volatile int a = 0;
	volatile int64_t b = 0;
	while (1) {
		a = (a + x) % (1 << 30);
		b = (b + y) % (1 << 30);
		assert(a == b);
	}
}

and two instances are run.  When built for just v7 CPUs, this program
works fine.  It uses the "vadd.i64 d19, d18, d16" VFP instruction.

It appears that we do not save the high-16 double VFP registers across
context switches when the kernel is built for v6 CPUs.  Fix that.

Cc: <stable@vger.kernel.org>
Tested-By: Michael Olbrich <m.olbrich@pengutronix.de>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
  • Loading branch information
Russell King committed Oct 9, 2012
1 parent f5a246e commit 846a136
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions arch/arm/include/asm/vfpmacros.h
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,7 @@
ldr \tmp, =elf_hwcap @ may not have MVFR regs
ldr \tmp, [\tmp, #0]
tst \tmp, #HWCAP_VFPv3D16
ldceq p11, cr0, [\base],#32*4 @ FLDMIAD \base!, {d16-d31}
ldceql p11, cr0, [\base],#32*4 @ FLDMIAD \base!, {d16-d31}
addne \base, \base, #32*4 @ step over unused register space
#else
VFPFMRX \tmp, MVFR0 @ Media and VFP Feature Register 0
Expand All @@ -52,7 +52,7 @@
ldr \tmp, =elf_hwcap @ may not have MVFR regs
ldr \tmp, [\tmp, #0]
tst \tmp, #HWCAP_VFPv3D16
stceq p11, cr0, [\base],#32*4 @ FSTMIAD \base!, {d16-d31}
stceql p11, cr0, [\base],#32*4 @ FSTMIAD \base!, {d16-d31}
addne \base, \base, #32*4 @ step over unused register space
#else
VFPFMRX \tmp, MVFR0 @ Media and VFP Feature Register 0
Expand Down

0 comments on commit 846a136

Please sign in to comment.