Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [RFC PATCH 1/2] pipe: introduce busy wait for pipe
To:     Subhra Mazumdar <subhra.mazumdar@oracle.com>,
        linux-kernel@vger.kernel.org
Cc:     peterz@infradead.org, dhaval.giani@oracle.com
References: <20180830202458.32579-1-subhra.mazumdar@oracle.com>
 <20180830202458.32579-2-subhra.mazumdar@oracle.com>
 <db00b844-055e-2024-2334-8e4b0e0ac5c5@oracle.com>
 <575e7d13-98e2-b5bf-e241-3f72a28b8c8a@oracle.com>
From:   Steven Sistare <steven.sistare@oracle.com>
Organization: Oracle Corporation
Message-ID: <5c6ca51b-9128-a6f3-a74c-67abb53a37ae@oracle.com>
Date:   Wed, 5 Sep 2018 09:45:36 -0400
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:52.0) Gecko/20100101
 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: <575e7d13-98e2-b5bf-e241-3f72a28b8c8a@oracle.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 9/4/2018 8:50 PM, Subhra Mazumdar wrote:
> On 08/31/2018 09:09 AM, Steven Sistare wrote:
>> On 8/30/2018 4:24 PM, subhra mazumdar wrote:
>>> Introduce pipe_ll_usec field for pipes that indicates the amount of micro
>>> seconds a thread should spin if pipe is empty or full before sleeping. This
>>> is similar to network sockets. Workloads like hackbench in pipe mode
>>> benefits significantly from this by avoiding the sleep and wakeup overhead.
>>> Other similar usecases can benefit. pipe_wait_flag is used to signal any
>>> thread busy waiting. pipe_busy_loop_timeout checks if spin time is over.
>>>
>>> Signed-off-by: subhra mazumdar <subhra.mazumdar@oracle.com>
>>> ---
>>>   include/linux/pipe_fs_i.h | 19 +++++++++++++++++++
>>>   1 file changed, 19 insertions(+)
>>>
>>> diff --git a/include/linux/pipe_fs_i.h b/include/linux/pipe_fs_i.h
>>> index e7497c9..fdfd2a2 100644
>>> --- a/include/linux/pipe_fs_i.h
>>> +++ b/include/linux/pipe_fs_i.h
>>> @@ -1,6 +1,8 @@
>>>   #ifndef _LINUX_PIPE_FS_I_H
>>>   #define _LINUX_PIPE_FS_I_H
>>>   +#include <linux/sched/clock.h>
>>> +
>>>   #define PIPE_DEF_BUFFERS    16
>>>     #define PIPE_BUF_FLAG_LRU    0x01    /* page is on the LRU */
>>> @@ -54,6 +56,8 @@ struct pipe_inode_info {
>>>       unsigned int waiting_writers;
>>>       unsigned int r_counter;
>>>       unsigned int w_counter;
>>> +    unsigned int pipe_ll_usec;
>>> +    unsigned long pipe_wait_flag;
>>>       struct page *tmp_page;
>>>       struct fasync_struct *fasync_readers;
>>>       struct fasync_struct *fasync_writers;
>>> @@ -157,6 +161,21 @@ static inline int pipe_buf_steal(struct pipe_inode_info *pipe,
>>>       return buf->ops->steal(pipe, buf);
>>>   }
>>>   +static inline unsigned long pipe_busy_loop_current_time(void)
>>> +{
>>> +    return (unsigned long)(local_clock() >> 10);
>> Why ">> 10" ? local_lock() has nanosec units, and you compare to the tunable
>> pipe_llc_sec which has microsec units.  Should be ">> 3".  Better yet, redefine
>> the tunable to have nanosec units.  I suspect you will need very large values
>> of the tunable to show similar results.
> It's 2^10. I don't think using nanosec units is necessary. It is unlikely
> data will be read or written in nano seconds. sk_busy_loop_timeout for
> sockets uses micro seconds too.

Ah, you are using 2^10 as an approximation of 1000.  OK.

- Steve

>>
>> Also, since this type of optimization consumes CPU extra cycles that could
>> be used by other tasks, show the overall CPU utilization before and after
>> the optimization, such as by using "time hackbench ...".
> OK.
> 
> Thanks,
> Subhra
>>
>> - Steve
>>
>>> +}
>>> +
>>> +static inline bool pipe_busy_loop_timeout(struct pipe_inode_info *pipe,
>>> +                      unsigned long start_time)
>>> +{
>>> +    unsigned long bp_usec = READ_ONCE(pipe->pipe_ll_usec);
>>> +    unsigned long end_time = start_time + bp_usec;
>>> +    unsigned long now = pipe_busy_loop_current_time();
>>> +
>>> +    return time_after(now, end_time);
>>> +}
>>> +
>>>   /* Differs from PIPE_BUF in that PIPE_SIZE is the length of the actual
>>>      memory allocation, whereas PIPE_BUF makes atomicity guarantees.  */
>>>   #define PIPE_SIZE        PAGE_SIZE
>>>
>