[PATCH v3] powerpc/vphn: NUMA node code expects big-endian

Greg Kurz gkurz at linux.vnet.ibm.com
Wed Oct 15 21:42:58 AEDT 2014

The associativity domain numbers are obtained from the hypervisor through
registers and written into memory by the guest: the packed array passed to
vphn_unpack_associativity() is then native-endian, unlike what was assumed
in the following commit:

commit b08a2a12e44eaec5024b2b969f4fcb98169d1ca3
Author: Alistair Popple <alistair at popple.id.au>
Date:   Wed Aug 7 02:01:44 2013 +1000

    powerpc: Make NUMA device node code endian safe

This issue fills the topology with bogus data and makes it unusable. It may
lead to severe performance breakdowns.

We should ideally patch the vphn_unpack_associativity() function to do the
64-bit loads, but this requires some more brain storming.

In the meantime, let's go for a suboptimal and temporary bug fix: this patch
converts each 64-bit value of the packed array to big endian, as expected by
the current parsing code in vphn_unpack_associativity().

Signed-off-by: Greg Kurz <gkurz at linux.vnet.ibm.com>

Changes in v3:
- dropped the complicated bit-shift based solution as requested by Michael

 arch/powerpc/mm/numa.c |    3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
index 649666d..b81bfe3 100644
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -1442,8 +1442,11 @@ static long hcall_vphn(unsigned long cpu, __be32 *associativity)
 	long retbuf[PLPAR_HCALL9_BUFSIZE] = {0};
 	u64 flags = 1;
 	int hwcpu = get_hard_smp_processor_id(cpu);
+	int i;
 	rc = plpar_hcall9(H_HOME_NODE_ASSOCIATIVITY, retbuf, flags, hwcpu);
+	for (i = 0; i < 6; i++)
+		retbuf[i] = cpu_to_be64(retbuf[i]);
 	vphn_unpack_associativity(retbuf, associativity);
 	return rc;

More information about the Linuxppc-dev mailing list