On Sat, 2016-09-04 at 06:14:04 UTC, "Aneesh Kumar K.V" wrote:
We can depend on ibm,pa-features to enable/disable radix. This gives us
a nice way to test p9 hash config, by changing device tree property.
I think we might want to be more careful here.
You set MMU_FTR_RADIX in the cputable entry. So it's on by default on P9 cpus.
Then if there is an ibm,pa-features property *and* it is >= 41 bytes long, the
below feature entry will hit. In that case the firmware controls whether it's on
or off.
I think it would be clearer if we removed RADIX from the cputable, and the below
became the only way to turn it on. Would that break anything?
quoted hunk ↗ jump to hunk
diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c
index 7030b035905d..a4d1f44364b8 100644
--- a/arch/powerpc/kernel/prom.c
+++ b/arch/powerpc/kernel/prom.c
@@ -165,6 +165,7 @@ static struct ibm_pa_feature {
* which is 0 if the kernel doesn't support TM.
*/
{CPU_FTR_TM_COMP, 0, 0, 22, 0, 0},
+ {0, MMU_FTR_RADIX, 0, 40, 0, 0},
So that says bit 0 of byte 40 enables MMU_FTR_RADIX. Where is that documented?
cheers