On Fri, 07 Dec 2018, Gabriel F. T. Gomes wrote:
On Fri, 07 Dec 2018, Rogerio Alves wrote:
+/* This test is supported only on POWER 5 or higher. */
+#define PPC_CPU_SUPPORTED (PPC_FEATURE_POWER5 | PPC_FEATURE_POWER5_PLUS | \
+ PPC_FEATURE_ARCH_2_05 | PPC_FEATURE_ARCH_2_06 | \
+ PPC_FEATURE2_ARCH_2_07)
Is this actually needed? Glibc has code to fill all the bits for older
architectures in sysdeps/powerpc/hwcapinfo.c [1]. So, as far as I can
see, you only need to test for AT_HWCAP & PPC_FEATURE_POWER5.
[1] https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/powerpc/hwcapinfo.c;h=a09b18309324423d0cdf04e88367899a4396bab7;hb=HEAD#l47
Hrm, that's only true for the hwcap info that is copied into the TCB, not
when accessing it with getauxval, so my comment is wrong (as Rogerio
kindly pointed out to me in a private message... thanks).
So, nevermind this comment.
On the other hand, on powerpc64 builds configured with --with-cpu set to
power4, power5, and power6, (but not to power8 or power7), I got the
following error message:
../sysdeps/powerpc/powerpc64/tst-ucontext-ppc64-vscr.c: In function ‘do_test’:
../sysdeps/powerpc/powerpc64/tst-ucontext-ppc64-vscr.c:51:3: error: inconsistent operand constraints in an ‘asm’
asm volatile ("vspltisb %0,0\n"
^~~
I guess we didn't get this error message before, because of the
`#ifdef ARCH_PWR8' statement. So, unless you know a better way to fix
this, you should reintroduce the `#ifdef' statement (now, you could use
ARCH_PWR7).