Mike Travis wrote:
> KOSAKI Motohiro wrote:
>> ===
>> Subject: [PATCH] cpumask:cpumask_of_node-ia64 fix
>> Impact: bug fix
>>
>> commit 07c636deede53ff3e96d54df0ece2c838e61951f introduce new CPU mask API and change cpu_mask variable type.
>>
>> -----------------------------------------------
>> - cpumask_t cpu_mask;
>> + const struct cpumask *cpu_mask;
>> --------------------------------------------------
>>
>>
>> However, its change is incomplete. in build time, compiler output following type mismatch warnings.
>>
>> arch/ia64/kernel/iosapic.c:708: warning: passing argument 2 of '__cpu_clear' from incompatible pointer type
>> arch/ia64/kernel/iosapic.c:711: warning: passing argument 1 of '__cpus_weight' from incompatible pointer type
>> arch/ia64/kernel/iosapic.c:719: warning: passing argument 1 of '__first_cpu' from incompatible pointer type
>> arch/ia64/kernel/iosapic.c:720: warning: passing argument 2 of '__next_cpu' from incompatible pointer type
>
> Good catch, however ...
>>
>> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx>
>> CC: "Luck, Tony" <tony.luck@xxxxxxxxx>
>> CC: Rusty Russell <rusty@xxxxxxxxxxxxxxx>
>> ---
>> arch/ia64/kernel/iosapic.c | 10 +++++-----
>> 1 file changed, 5 insertions(+), 5 deletions(-)
>>
>> Index: b/arch/ia64/kernel/iosapic.c
>> ===================================================================
>> --- a/arch/ia64/kernel/iosapic.c
>> +++ b/arch/ia64/kernel/iosapic.c
>> @@ -695,7 +695,7 @@ get_target_cpu (unsigned int gsi, int ir
>> #ifdef CONFIG_NUMA
>> {
>> int num_cpus, cpu_index, iosapic_index, numa_cpu, i = 0;
>> - const struct cpumask *cpu_mask;
>> + struct cpumask *cpu_mask;
>>
>> iosapic_index = find_iosapic(gsi);
>> if (iosapic_index < 0 ||
>> @@ -705,10 +705,10 @@ get_target_cpu (unsigned int gsi, int ir
>> cpu_mask = cpumask_of_node(iosapic_lists[iosapic_index].node);
>> for_each_cpu_and(numa_cpu, cpu_mask, &domain) {
>> if (!cpu_online(numa_cpu))
>> - cpu_clear(numa_cpu, cpu_mask);
>> + cpu_clear(numa_cpu, *cpu_mask);
>
> ... you cannot modify the read-only cpumask_of_node map. Suggested alternative below.
>
>> }
>>
>> - num_cpus = cpus_weight(cpu_mask);
>> + num_cpus = cpus_weight(*cpu_mask);
>>
>> if (!num_cpus)
>> goto skip_numa_setup;
>> @@ -716,8 +716,8 @@ get_target_cpu (unsigned int gsi, int ir
>> /* Use irq assignment to distribute across cpus in node */
>> cpu_index = irq % num_cpus;
>>
>> - for (numa_cpu = first_cpu(cpu_mask) ; i < cpu_index ; i++)
>> - numa_cpu = next_cpu(numa_cpu, cpu_mask);
>> + for (numa_cpu = first_cpu(*cpu_mask) ; i < cpu_index ; i++)
>> + numa_cpu = next_cpu(numa_cpu, *cpu_mask);
>>
>> if (numa_cpu != NR_CPUS)
>> return cpu_physical_id(numa_cpu);
>>
>>
>> --
> ---
> arch/ia64/kernel/iosapic.c | 16 ++++++++--------
> 1 file changed, 8 insertions(+), 8 deletions(-)
>
> --- linux-2.6-next.orig/arch/ia64/kernel/iosapic.c
> +++ linux-2.6-next/arch/ia64/kernel/iosapic.c
> @@ -703,23 +703,23 @@ get_target_cpu (unsigned int gsi, int ir
> goto skip_numa_setup;
>
> cpu_mask = cpumask_of_node(iosapic_lists[iosapic_index].node);
> + num_cpus = 0;
> for_each_cpu_and(numa_cpu, cpu_mask, &domain) {
> - if (!cpu_online(numa_cpu))
> - cpu_clear(numa_cpu, cpu_mask);
> + if (cpu_online(numa_cpu))
> + num_cpus++;
> }
>
> - num_cpus = cpus_weight(cpu_mask);
> -
> if (!num_cpus)
> goto skip_numa_setup;
>
> /* Use irq assignment to distribute across cpus in node */
> cpu_index = irq % num_cpus;
>
> - for (numa_cpu = first_cpu(cpu_mask) ; i < cpu_index ; i++)
> - numa_cpu = next_cpu(numa_cpu, cpu_mask);
> + for_each_cpu_and(numa_cpu, cpu_mask, cpu_online_mask)
> + if (i++ >= cpu_index)
> + break;
Oops, looks like I over-simplified the above...
for_each_cpu_and(numa_cpu, cpu_mask, &domain)
if (cpu_online(numa_cpu) && i++ >= cpu_index)
break;
>
> - if (numa_cpu != NR_CPUS)
> + if (numa_cpu < nr_cpu_ids)
> return cpu_physical_id(numa_cpu);
> }
> skip_numa_setup:
> @@ -730,7 +730,7 @@ skip_numa_setup:
> * case of NUMA.)
> */
> do {
> - if (++cpu >= NR_CPUS)
> + if (++cpu >= nr_cpu_ids)
> cpu = 0;
> } while (!cpu_online(cpu) || !cpu_isset(cpu, domain));
>
> ---
>
> One aside, I'm not sure of ia64, but x86_64 keeps only online cpus in the cpumask_of_node_map.
> IOW, if a cpu goes offline, it's removed from the map. So the whole and'ing thing above may
> be unnecessary.
--
To unsubscribe from this list: send the line "unsubscribe linux-next" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
