In the numa=off kernel command-line configuration init_chip_info() loops
around the number of chips and attempts to copy the cpumask of that node
which is NULL for all iterations after the first chip.
Hence adding a check to bail out after the first initialization if there
is only one node.
Fixes: 053819e0bf84 ("cpufreq: powernv: Handle throttling due to Pmax capping at chip level")
Signed-off-by: Pratik R. Sampat <[email protected]>
Reported-by: Shirisha Ganta <[email protected]>
---
drivers/cpufreq/powernv-cpufreq.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
index e439b43c19eb..663f9c4b5e3a 100644
--- a/drivers/cpufreq/powernv-cpufreq.c
+++ b/drivers/cpufreq/powernv-cpufreq.c
@@ -1078,6 +1078,8 @@ static int init_chip_info(void)
INIT_WORK(&chips[i].throttle, powernv_cpufreq_work_fn);
for_each_cpu(cpu, &chips[i].mask)
per_cpu(chip_info, cpu) = &chips[i];
+ if (num_possible_nodes() == 1)
+ break;
}
free_and_return:
--
2.30.2
Hello Pratik,
On Tue, Jun 15, 2021 at 10:39:49AM +0530, Pratik R. Sampat wrote:
> In the numa=off kernel command-line configuration init_chip_info() loops
> around the number of chips and attempts to copy the cpumask of that node
> which is NULL for all iterations after the first chip.
Thanks for taking a look into this. Indeed there is an issue here
because the code here assumes that node_mask as a proxy for the
chip_mask. This assumption breaks when run with numa=off, since there will only be a
single node, but multiple chips.
>
> Hence adding a check to bail out after the first initialization if there
> is only one node.
>
> Fixes: 053819e0bf84 ("cpufreq: powernv: Handle throttling due to Pmax capping at chip level")
> Signed-off-by: Pratik R. Sampat <[email protected]>
> Reported-by: Shirisha Ganta <[email protected]>
> ---
> drivers/cpufreq/powernv-cpufreq.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
> index e439b43c19eb..663f9c4b5e3a 100644
> --- a/drivers/cpufreq/powernv-cpufreq.c
> +++ b/drivers/cpufreq/powernv-cpufreq.c
> @@ -1078,6 +1078,8 @@ static int init_chip_info(void)
> INIT_WORK(&chips[i].throttle, powernv_cpufreq_work_fn);
> for_each_cpu(cpu, &chips[i].mask)
> per_cpu(chip_info, cpu) = &chips[i];
> + if (num_possible_nodes() == 1)
> + break;
With this we will only initialize the chip[0].throttle work function,
while for the rest of the chips chip[i].throttle will be
uninitialized. While we may be running in the numa=off mode, the fact
remains that those other chips do exist and they may experiencing
throttling, during which they will try to schedule work for chip[i] in
order to take corrective action, which will fail.
Hence a more correct approach may be to maintain a chip[i] mask
independent of the node mask.
> }
>
> free_and_return:
> --
> 2.30.2
>