Re: 4.5-rc4 kernel is failed to bootup on CN6880

From: David Daney
Date: Mon Feb 22 2016 - 15:56:27 EST


On 02/22/2016 11:15 AM, Yang Shi wrote:
On 2/22/2016 4:43 AM, Aaro Koskinen wrote:
Hi,

On Fri, Feb 19, 2016 at 05:12:41PM -0800, Yang Shi wrote:
I tried to boot 4.5-rc4 kernel on my CN6880 board, but it is failed at
booting up secondary cores. The error is:
With v4.5-rc5, EBB6800 is booting fine:

[ 0.000000] CPU0 revision is: 000d9108 (Cavium Octeon II)
[...]
[ 2286.273935] SMP: Booting CPU01 (CoreId 1)...
[ 2286.278201] CPU1 revision is: 000d9108 (Cavium Octeon II)
[...]
[ 2287.214953] SMP: Booting CPU31 (CoreId 31)...
[ 2287.224668] CPU31 revision is: 000d9108 (Cavium Octeon II)
[ 2287.224865] Brought up 32 CPUs

CPU31 revision is: 000d9101 (Cavium Octeon II)
SMP: Booting CPU32 (CoreId 32)...
Secondary boot timeout

I passed "numcores=32" in kernel commandline since there are 32 cores
ion
CN6880.
You shouldn't have CPU32 in that case, the numbering starts from zero.
Also the coremask is 32-bit.

I can reproduce your issue with CONFIG_NR_CPUS=64. Possibly this code
is incorrect for NR_CPUS bigger than 32:

/* The present CPUs get the lowest CPU numbers. */
cpus = 1;
for (id = 0; id < NR_CPUS; id++) {
if ((id != coreid) && (core_mask & (1 << id))) {
set_cpu_possible(cpus, true);
set_cpu_present(cpus, true);

What CONFIG_NR_CPUS did you use?

Thanks. I did have 48 NR_CPUS set. It works when I changed it to 32.

I think the problem is core_mask is 32 bit. But when NR_CPUS > 32, in
"core_mask & (1 << id)" core_mask will be sign extended, then the
statement will return non-zero all the time.


That is correct, and perhaps not coincidentally, the reason I have already submitted patches to fix this.

David.


Yang


A.