David Gibson [da...@gibson.dropbear.id.au] wrote:
| > Should we walk the /proc/device-tree/cpus/ tree and count only dirs with
| > device-type "cpu" (rather than relying on the pattern PowerPC,POWER*)?
|
| Yes, I think you'll have to.
|
Ok. Are we ok with the xscom check I have (copied below):
+static int kvmppc_count_sockets_chips_dt(int *num_sockets, int *num_chips)
+{
+ const char *chip_pattern = "/proc/device-tree/xscom@*/ibm,chip-id";
+ const char *module_pattern = "/proc/device-tree/xscom@*/ibm,hw-module-id";
---
I am not sure of a reliable way to detect number of chips besides the
xscom. I do see that sysfs has node entries:
for i in `ls -d /sys/bus/cpu/devices/cpu*`;
> do
> basename $i/node*
> done | sort | uniq
node0
node1
node16
node17
which seems to correspond to
$ lsprop /proc/device-tree/xscom@*/ibm,chip-id
/proc/device-tree/xscom@3fc0000000000/ibm,chip-id
00000000
/proc/device-tree/xscom@3fc0800000000/ibm,chip-id
00000001
/proc/device-tree/xscom@3fc8000000000/ibm,chip-id
00000010 (16)
/proc/device-tree/xscom@3fc8800000000/ibm,chip-id
00000011 (17)
but is that reliable?
> +
| > |
| > | In a number of ways I'd actually prefer to move to /cpus/cpu@NNN in
| > | general, since that follows the OF generic names recommendation we
| > | follow for most other nodes.
| >
| > Do you mean rename '/proc/device-tree/cpus/PowerPC,POWER8@NNN' to
| > /proc/device-tree/cpus/cpu@NNN?
|
| Yes. This is a firmware matter, so it's not something that can simply
| be changed everywhere, but it's the approach that I'd prefer to
| encourage for people making future machines and firmwares.
Sure. ppc64_cpu also seems to rely on '/proc/device-tree/cpus/PowerPC'
to count cores present and maybe there are other utilities that would
need to be updated.
Sukadev
