2017-12-18 08:09:04

by Jia-Ju Bai

Subject: [PATCH] hv: Fix unnecessary sleeping in hv_synic_alloc

The kzalloc function is called with GFP_ATOMIC.
But according to driver call graph, it is not in atomic context,
namely no spinlock is held nor in an interrupt handler.

This GFP_ATOMIC is unnecessary, and replace with GFP_KERNEL.

Signed-off-by: Jia-Ju Bai <[email protected]>
---
drivers/hv/hv.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
index 8267439..b0d025a 100644
--- a/drivers/hv/hv.c
+++ b/drivers/hv/hv.c
@@ -150,7 +150,7 @@ int hv_synic_alloc(void)
int cpu;

hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
- GFP_ATOMIC);
+ GFP_KERNEL);
if (hv_context.hv_numa_map == NULL) {
pr_err("Unable to allocate NUMA map\n");
goto err;
--
1.7.9.5

2017-12-18 08:54:33

by Vitaly Kuznetsov

Subject: Re: [PATCH] hv: Fix unnecessary sleeping in hv_synic_alloc

Jia-Ju Bai <[email protected]> writes:

> The kzalloc function is called with GFP_ATOMIC.
> But according to driver call graph, it is not in atomic context,
> namely no spinlock is held nor in an interrupt handler.
>
> This GFP_ATOMIC is unnecessary, and replace with GFP_KERNEL.
>
> Signed-off-by: Jia-Ju Bai <[email protected]>
> ---
> drivers/hv/hv.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
> index 8267439..b0d025a 100644
> --- a/drivers/hv/hv.c
> +++ b/drivers/hv/hv.c
> @@ -150,7 +150,7 @@ int hv_synic_alloc(void)
> int cpu;
>
> hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
> - GFP_ATOMIC);
> + GFP_KERNEL);
> if (hv_context.hv_numa_map == NULL) {
> pr_err("Unable to allocate NUMA map\n");
> goto err;

Reviewed-by: Vitaly Kuznetsov <[email protected]>

The subject line is not very accurate: GFP_KERNEL you switch too is the
one supporting 'sleep' when there's not enough memory, not
GFP_ATOMIC so you don't actually "Fix unnecessary sleeping". I'd suggest
you use something like "hyper-v: use GFP_KERNEL for hv_context.hv_numa_map"

--
Vitaly

2017-12-18 08:58:37

by Jia-Ju Bai

Subject: Re: [PATCH] hv: Fix unnecessary sleeping in hv_synic_alloc

On 2017/12/18 16:54, Vitaly Kuznetsov wrote:
> Jia-Ju Bai <[email protected]> writes:
>
>> The kzalloc function is called with GFP_ATOMIC.
>> But according to driver call graph, it is not in atomic context,
>> namely no spinlock is held nor in an interrupt handler.
>>
>> This GFP_ATOMIC is unnecessary, and replace with GFP_KERNEL.
>>
>> Signed-off-by: Jia-Ju Bai <[email protected]>
>> ---
>> drivers/hv/hv.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
>> index 8267439..b0d025a 100644
>> --- a/drivers/hv/hv.c
>> +++ b/drivers/hv/hv.c
>> @@ -150,7 +150,7 @@ int hv_synic_alloc(void)
>> int cpu;
>>
>> hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
>> - GFP_ATOMIC);
>> + GFP_KERNEL);
>> if (hv_context.hv_numa_map == NULL) {
>> pr_err("Unable to allocate NUMA map\n");
>> goto err;
> Reviewed-by: Vitaly Kuznetsov <[email protected]>
>
> The subject line is not very accurate: GFP_KERNEL you switch too is the
> one supporting 'sleep' when there's not enough memory, not
> GFP_ATOMIC so you don't actually "Fix unnecessary sleeping". I'd suggest
> you use something like "hyper-v: use GFP_KERNEL for hv_context.hv_numa_map"
>

Thanks for you suggestion :)
Okay, I found my description is not correct, too, sorry.
I will revise it and resubmit the patch.


Thanks,
Jia-Ju Bai