In-band Context Switch on oob_ioctl() call

In-band Context Switch on oob_ioctl() call

[Russell Johnson]

Hello,

I have added the oob_ioctl file descriptor to my driver, and I have made sure that I call the necessary evl functions in the open() function in the driver as well. I have a test app in userspace that is simply opening the driver and issuing an oob_ioctl every second. What I am seeing in the output from "evl ps -l" is that the "ISW" and "CTXSW" fields are incrementing with every oob_ioctl call. I added in the O_OOB flag to my open call, and I am still seeing the same behavior. I don't get why it would be switching to in-band every time the oob_ioctl() is called. Any ideas on what I could be doing wrong?

Thanks,
Russell

Re: In-band Context Switch on oob_ioctl() call

[Philippe Gerum]

Russell Johnson via Xenomai <xenomai@xenomai.org> writes:

> Hello,
>
> I have added the oob_ioctl file descriptor to my driver, and I have
> made sure that I call the necessary evl functions in the open()
> function in the driver as well. I have a test app in userspace that is
> simply opening the driver and issuing an oob_ioctl every second. What
> I am seeing in the output from "evl ps -l" is that the "ISW" and
> "CTXSW" fields are incrementing with every oob_ioctl call. I added in
> the O_OOB flag to my open call, and I am still seeing the same
> behavior. I don't get why it would be switching to in-band every time
> the oob_ioctl() is called. Any ideas on what I could be doing wrong?
>
> Thanks,
> Russell

It looks like the calling thread is undergoing the so-called weak
scheduling policy, see [1].

[1] https://evlproject.org/core/user-api/scheduling/#SCHED_WEAK

-- 
Philippe.

RE: [External] - Re: In-band Context Switch on oob_ioctl() call

[Russell Johnson]

I verified this, and the thread is under the FIFO scheduling policy.

CPU   PID   SCHED   PRIO  ISW     CTXSW     SYS       RWA       STAT     TIMEOUT      %CPU   CPUTIME     NAME
  9   12344  fifo    98   14      14        14        0          X          -           0.0  00:000.289  evl_dma_thread:12100

- Russell

-----Original Message-----
From: Philippe Gerum <rpm@xenomai.org> 
Sent: Saturday, February 19, 2022 8:29 AM
To: Russell Johnson <russell.johnson@kratosdefense.com>
Cc: xenomai@xenomai.org
Subject: [External] - Re: In-band Context Switch on oob_ioctl() call

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.


Russell Johnson via Xenomai <xenomai@xenomai.org> writes:

> Hello,
>
> I have added the oob_ioctl file descriptor to my driver, and I have 
> made sure that I call the necessary evl functions in the open() 
> function in the driver as well. I have a test app in userspace that is 
> simply opening the driver and issuing an oob_ioctl every second. What 
> I am seeing in the output from "evl ps -l" is that the "ISW" and 
> "CTXSW" fields are incrementing with every oob_ioctl call. I added in 
> the O_OOB flag to my open call, and I am still seeing the same 
> behavior. I don't get why it would be switching to in-band every time 
> the oob_ioctl() is called. Any ideas on what I could be doing wrong?
>
> Thanks,
> Russell

It looks like the calling thread is undergoing the so-called weak scheduling policy, see [1].

[1] https://urldefense.proofpoint.com/v2/url?u=https-3A__evlproject.org_core_user-2Dapi_scheduling_-23SCHED-5FWEAK&d=DwIBAg&c=zeCCs5WLaN-HWPHrpXwbFoOqeS0G3NH2_2IQ_bzV13g&r=-WrlDvHLmZCY7pl-EOIRaJl9uZvKbG-G8KkkUcIDh2F7Jr6SYxS_pdf-orjkExnI&m=0Ytm_g4N1lnCdUHifknw-Mckrpzq2gJlG2oKvQ543DBHhDfXPgbp5TWf-00etMCB&s=HKJUpkhAExaAqj5CUY7bze1EqRKRy8x6RHRKCl4OWH4&e=

--
Philippe.

Re: [External] - Re: In-band Context Switch on oob_ioctl() call

[Philippe Gerum]

Russell Johnson <russell.johnson@kratosdefense.com> writes:

> I verified this, and the thread is under the FIFO scheduling policy.
>
> CPU   PID   SCHED   PRIO  ISW     CTXSW     SYS       RWA       STAT     TIMEOUT      %CPU   CPUTIME     NAME
>   9   12344  fifo    98   14      14        14        0          X          -           0.0  00:000.289  evl_dma_thread:12100
>
                                                                  ^^^

Means "relaxed", in-band mode. So the thread does switch out of oob mode
voluntarily for some reason.

> - Russell
>
> -----Original Message-----
> From: Philippe Gerum <rpm@xenomai.org> 
> Sent: Saturday, February 19, 2022 8:29 AM
> To: Russell Johnson <russell.johnson@kratosdefense.com>
> Cc: xenomai@xenomai.org
> Subject: [External] - Re: In-band Context Switch on oob_ioctl() call
>
> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
>
>
> Russell Johnson via Xenomai <xenomai@xenomai.org> writes:
>
>> Hello,
>>
>> I have added the oob_ioctl file descriptor to my driver, and I have 
>> made sure that I call the necessary evl functions in the open() 
>> function in the driver as well. I have a test app in userspace that is 
>> simply opening the driver and issuing an oob_ioctl every second. What 
>> I am seeing in the output from "evl ps -l" is that the "ISW" and 
>> "CTXSW" fields are incrementing with every oob_ioctl call. I added in 
>> the O_OOB flag to my open call, and I am still seeing the same 
>> behavior. I don't get why it would be switching to in-band every time 
>> the oob_ioctl() is called. Any ideas on what I could be doing wrong?
>>
>> Thanks,
>> Russell
>
> It looks like the calling thread is undergoing the so-called weak scheduling policy, see [1].
>
> [1] https://urldefense.proofpoint.com/v2/url?u=https-3A__evlproject.org_core_user-2Dapi_scheduling_-23SCHED-5FWEAK&d=DwIBAg&c=zeCCs5WLaN-HWPHrpXwbFoOqeS0G3NH2_2IQ_bzV13g&r=-WrlDvHLmZCY7pl-EOIRaJl9uZvKbG-G8KkkUcIDh2F7Jr6SYxS_pdf-orjkExnI&m=0Ytm_g4N1lnCdUHifknw-Mckrpzq2gJlG2oKvQ543DBHhDfXPgbp5TWf-00etMCB&s=HKJUpkhAExaAqj5CUY7bze1EqRKRy8x6RHRKCl4OWH4&e=


-- 
Philippe.