Re: [PATCH 3/6] cpuset: convert cpuset_attach() to use cpumask_var_t

From: Li Zefan
Date: Mon Jan 05 2009 - 03:48:31 EST


Andrew Morton wrote:
> On Wed, 31 Dec 2008 16:36:11 +0800 Li Zefan <lizf@xxxxxxxxxxxxxx> wrote:
>
>> Impact: reduce stack usage
>>
>> Allocate a cpumask_var_t in cpuset_can_attach() and then use it in
>> cpuset_attach(), so we won't fail cpuset_attach().
>>
>> Signed-off-by: Li Zefan <lizf@xxxxxxxxxxxxxx>
>> ---
>> kernel/cpuset.c | 31 +++++++++++++++++++++++++------
>> 1 files changed, 25 insertions(+), 6 deletions(-)
>>
>> diff --git a/kernel/cpuset.c b/kernel/cpuset.c
>> index afa29cf..b4c36d5 100644
>> --- a/kernel/cpuset.c
>> +++ b/kernel/cpuset.c
>> @@ -1306,6 +1306,14 @@ static int fmeter_getrate(struct fmeter *fmp)
>> return val;
>> }
>>
>> +/*
>> + * Since cpuset_attach() can't fail, we allocate a cpumask_var_t in
>> + * cpuset_can_attach() and then use it in cpuset_attach().
>> + *
>> + * Protected by cgroup_lock.
>
> OK, but it's a bit fragile.
>
> How are we to ensure that all calls to cpuset_can_attach() are followed
> by a call to cpuset_attach()

my fault :(

> and how are we to ensure that both calls
> are protected by the same taking of the lock? For all time?
>

It's insured by cgroup. The 2 functions are callbacks that will be called
by cgroup core only.

> In fact, it's broken right now, isn't it? If cgroup_attach_task()'s
> call to find_css_set() fails, we leak the result of cpuset_can_attach()'s
> alloc_cpumask_var()?
>

Below is a replacement:

=================

[PATCH 3/6] cpuset: convert cpuset_attach() to use cpumask_var_t

Allocate a global cpumask_var_t at boot, and use it in cpuset_attach(), so
we won't fail cpuset_attach().

Signed-off-by: Li Zefan <lizf@xxxxxxxxxxxxxx>
Acked-by: Mike Travis <travis@xxxxxxx>
---
kernel/cpuset.c | 14 +++++++++-----
1 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index afa29cf..1e32e6b 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -1306,6 +1306,9 @@ static int fmeter_getrate(struct fmeter *fmp)
return val;
}

+/* Protected by cgroup_lock */
+static cpumask_var_t cpus_attach;
+
/* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */
static int cpuset_can_attach(struct cgroup_subsys *ss,
struct cgroup *cont, struct task_struct *tsk)
@@ -1330,7 +1333,6 @@ static void cpuset_attach(struct cgroup_subsys *ss,
struct cgroup *cont, struct cgroup *oldcont,
struct task_struct *tsk)
{
- cpumask_t cpus;
nodemask_t from, to;
struct mm_struct *mm;
struct cpuset *cs = cgroup_cs(cont);
@@ -1338,13 +1340,13 @@ static void cpuset_attach(struct cgroup_subsys *ss,
int err;

if (cs == &top_cpuset) {
- cpus = cpu_possible_map;
+ cpumask_copy(cpus_attach, cpu_possible_mask);
} else {
mutex_lock(&callback_mutex);
- guarantee_online_cpus(cs, &cpus);
+ guarantee_online_cpus(cs, cpus_attach);
mutex_unlock(&callback_mutex);
}
- err = set_cpus_allowed_ptr(tsk, &cpus);
+ err = set_cpus_allowed_ptr(tsk, cpus_attach);
if (err)
return;

@@ -1357,7 +1359,6 @@ static void cpuset_attach(struct cgroup_subsys *ss,
cpuset_migrate_mm(mm, &from, &to);
mmput(mm);
}
-
}

/* The various types of files and directories in a cpuset file system */
@@ -1838,6 +1839,9 @@ int __init cpuset_init(void)
if (err < 0)
return err;

+ if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL))
+ BUG();
+
number_of_cpusets = 1;
return 0;
}
--
1.5.4.rc3


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/