3.16.42-rc1 review patch. If anyone has any objections, please let me know.
------------------
From: Darren Stevens <darren@xxxxxxxxxxxxxxxx>
commit af2b7fa17eb92e52b65f96604448ff7a2a89ee99 upstream.
prom_init.c calls 'instance-to-package' twice, but the return
is not checked during prom_find_boot_cpu(). The result is then
passed to prom_getprop(), which could be PROM_ERROR. Add a return check
to prevent this.
This was found on a pasemi system, where CFE doesn't have a working
'instance-to package' prom call.
Before Commit 5c0484e25ec0 ('powerpc: Endian safe trampoline') the area
around addr 0 was mostly 0's and this doesn't cause a problem. Once the
macro 'FIXUP_ENDIAN' has been added to head_64.S, the low memory area
now has non-zero values, which cause the prom_getprop() call
to hang.
mpe: Also confirmed that under SLOF if 'instance-to-package' did fail
with PROM_ERROR we would crash in SLOF. So the bug is not specific to
CFE, it's just that other open firmwares don't trigger it because they
have a working 'instance-to-package'.
Fixes: 5c0484e25ec0 ("powerpc: Endian safe trampoline")
Signed-off-by: Darren Stevens <darren@xxxxxxxxxxxxxxxx>
Signed-off-by: Michael Ellerman <mpe@xxxxxxxxxxxxxx>
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
---
arch/powerpc/kernel/prom_init.c | 3 +++
1 file changed, 3 insertions(+)
--- a/arch/powerpc/kernel/prom_init.c
+++ b/arch/powerpc/kernel/prom_init.c
@@ -2635,6 +2635,9 @@ static void __init prom_find_boot_cpu(vo
cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
+ if (!PHANDLE_VALID(cpu_pkg))
+ return;
+
prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
prom.cpu = be32_to_cpu(rval);