lkml.org 
[lkml]   [2019]   [Apr]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH v2 3/5] locking/qspinlock: Introduce CNA into the slow path of qspinlock
From
Date

> On Apr 1, 2019, at 5:33 AM, Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Mon, Apr 01, 2019 at 11:06:53AM +0200, Peter Zijlstra wrote:
>> On Fri, Mar 29, 2019 at 11:20:04AM -0400, Alex Kogan wrote:
>>> diff --git a/kernel/locking/mcs_spinlock.h b/kernel/locking/mcs_spinlock.h
>>> index bc6d3244e1af..71ee4b64c5d4 100644
>>> --- a/kernel/locking/mcs_spinlock.h
>>> +++ b/kernel/locking/mcs_spinlock.h
>>> @@ -17,8 +17,18 @@
>>>
>>> struct mcs_spinlock {
>>> struct mcs_spinlock *next;
>>> +#ifndef CONFIG_NUMA_AWARE_SPINLOCKS
>>> int locked; /* 1 if lock acquired */
>>> int count; /* nesting count, see qspinlock.c */
>>> +#else /* CONFIG_NUMA_AWARE_SPINLOCKS */
>>> + uintptr_t locked; /* 1 if lock acquired, 0 if not, other values */
>>> + /* represent a pointer to the secondary queue head */
>>> + u32 node_and_count; /* node id on which this thread is running */
>>> + /* with two lower bits reserved for nesting */
>>> + /* count, see qspinlock.c */
>>> + u32 encoded_tail; /* encoding of this node as the main queue tail */
>>> + struct mcs_spinlock *tail; /* points to the secondary queue tail */
>>> +#endif /* CONFIG_NUMA_AWARE_SPINLOCKS */
>>> };
>>
>> Please, have another look at the paravirt code, in particular at struct
>> pv_node and its usage. This is horrible.
>
> Thing is, this turns into a right mess when you also have PV spinlocks
> on.
>
> One thing we could maybe do is change locked and count to u8, then your
> overlay structure could be something like:
>
> struct mcs_spinlock {
> struct mcs_spinlock *next;
> u8 locked;
> u8 count;
> };
I was trying to keep the size of the mcs_spinlock structure for the non-NUMA variant unchanged.
If this is not a huge concern, changing the fields as above would indeed simplify a few things.

— Alex

>
> struct cna_node {
> /* struct mcs_spinlock overlay */
> struct mcs_spinlock *next;
> u8 locked;
> u8 count;
>
> /* our CNA bits, consuming the slack and PV space */
> u16 node;
> u32 encoded_tail;
> struct mcs_spinlock *head;
> struct mcs_spinlock *tail;
> };
>
> Then you also don't need the first two patches.

\
 
 \ /
  Last update: 2019-04-03 17:55    [W:0.059 / U:13.340 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site