iio_buffer length

iio_buffer length

[Denis Ciocca]

Hi guys,

I finally managed to resolve the issues about iio_trigger and iio_buffer.
Now I ask please you if you can clarify to me about use of
iio_generic_buffer and how I can use the /dev/iio:deviceX char-event
to catch data from buffer.
I compiled the generic-buffer program and I don't understand how datas
come out. My driver has 4 channels:
- 3 for accelerometer raw data (x,y,z)
- 1 for timestamp

The application set up the length of buffer to 32, but when I run the
application I don't see 32 samples but only 16. What means 32?
It is possible the use of buffer to catch every sample? (buffer length = 1)?
When I set buffer length to 1 the application doesn't work!

Thanks,
Denis

Re: iio_buffer length

[Jonathan Cameron]

Denis Ciocca <denis.ciocca@gmail.com> wrote:

>Hi guys,
>
>I finally managed to resolve the issues about iio_trigger and
>iio_buffer.
>Now I ask please you if you can clarify to me about use of
>iio_generic_buffer and how I can use the /dev/iio:deviceX char-event
>to catch data from buffer.
>I compiled the generic-buffer program and I don't understand how datas
>come out. My driver has 4 channels:
>- 3 for accelerometer raw data (x,y,z)
>- 1 for timestamp
>
>The application set up the length of buffer to 32, but when I run the
>application I don't see 32 samples but only 16. What means 32?
>It is possible the use of buffer to catch every sample? (buffer length
>= 1)?
>When I set buffer length to 1 the application doesn't work!

Firstly please confirm you are using a kfifo buffer? Swring is deprecated but will do precisely what you are seeing I think. It uses a half full watershead to decide to signal userspace.  If you wait till the buffer is full, unless your sampling is very slow you will have data coming in before userspace has read the whole buffer out. Hence you will loose data.  Kfifo has a more conventional poll and will fire off userspace whenever any data is in the buffer. Hence kfifo will work for your usecase.  I would like to have a fill level capable buffer again in the future as they are very useful in high speed uses where latency doesn't matter. 

If you are using kfifo then something odd is going on!  Its trivial to switch a driver between them.
>
>Thanks,
>Denis
>--
>To unsubscribe from this list: send the line "unsubscribe linux-iio" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Sent from my Android phone with K-9 Mail. Please excuse my brevity.

Re: iio_buffer length

[Denis Ciocca]

[-- Attachment #1: Type: text/plain, Size: 787 bytes --]

Hi Jon,

> Firstly please confirm you are using a kfifo buffer?

No, I use sw-ring.

> Swring is deprecated but will do precisely what you are seeing I think.
It uses a half full watershead to decide to signal userspace.  If you wait
till the buffer is full, unless your sampling is very slow you will have
data coming in before userspace has read the whole buffer out. Hence you
will loose data.

You are right. I saw exatly what you said.

 Kfifo has a more conventional poll and will fire off userspace whenever
any data is in the buffer. Hence kfifo will work for your usecase.  I would
like to have a fill level capable buffer again in the future as they are
very useful in high speed uses where latency doesn't matter.

Ok, I will change my sw-ring buffer to kfifo!

Thanks,
Denis

[-- Attachment #2: Type: text/html, Size: 876 bytes --]

RE: iio_buffer length

[Ge Gao]

To use KFIFO, you need Jonanthan's KFIFO poll patch,  you probably also
need a patch from me that fixed the industrial-buffer.c. But that buffer
is not well tested and is only used by me. The original KFIFO does not
have poll functionality and has some other minor bug.

Ge


-----Original Message-----
From: linux-iio-owner@vger.kernel.org
[mailto:linux-iio-owner@vger.kernel.org] On Behalf Of Denis Ciocca
Sent: Tuesday, September 25, 2012 2:35 PM
To: jic23@jic23.retrosnub.co.uk
Cc: linux-iio@vger.kernel.org
Subject: iio_buffer length

Hi Jon,

> Firstly please confirm you are using a kfifo buffer?

No, I use sw-ring.


> Swring is deprecated but will do precisely what you are seeing I think.
It uses a half full watershead to decide to signal userspace.  If you wait
till the buffer is full, unless your sampling is very slow you will have
data coming in before userspace has read the whole buffer out. Hence you
will loose data.

You are right. I saw exatly what you said.


> Kfifo has a more conventional poll and will fire off userspace
> whenever
any data is in the buffer. Hence kfifo will work for your usecase.  I
would like to have a fill level capable buffer again in the future as they
are very useful in high speed uses where latency doesn't matter.

Ok, I will change my sw-ring buffer to kfifo!

Thanks,
Denis
--
To unsubscribe from this list: send the line "unsubscribe linux-iio" in
the body of a message to majordomo@vger.kernel.org More majordomo info at
http://vger.kernel.org/majordomo-info.html

Re: iio_buffer length

[Jonathan Cameron]

On 26/09/12 02:32, Ge Gao wrote:
> To use KFIFO, you need Jonanthan's KFIFO poll patch,
That's in staging-next now and will merge into mainline in the
next merge window (about 1 to 2 weeks time).
> you probably also
> need a patch from me that fixed the industrial-buffer.c. But that buffer
> is not well tested and is only used by me.
There are a lot of drivers using kfifo. It's hidden in many of them as 
they use it via the industrialio-triggered-buffer helper functions that
deal with the most common use cases.

With that I count 9 in drivers/iio and 6 in drivers/staging/iio vs 10
in total for sw_ring most of which will go away once someone has a spare
couple of hours.

Basically any driver that has come along since we moved to kfifo is the
preferred choice and any that anyone has touched since then.

> The original KFIFO does not
> have poll functionality and has some other minor bug.
If I've missed any patches from you on this (or just forgotten them ;)
please do keep pestering me.
>
> Ge
>
>
> -----Original Message-----
> From: linux-iio-owner@vger.kernel.org
> [mailto:linux-iio-owner@vger.kernel.org] On Behalf Of Denis Ciocca
> Sent: Tuesday, September 25, 2012 2:35 PM
> To: jic23@jic23.retrosnub.co.uk
> Cc: linux-iio@vger.kernel.org
> Subject: iio_buffer length
>
> Hi Jon,
>
>> Firstly please confirm you are using a kfifo buffer?
>
> No, I use sw-ring.
>
>
>> Swring is deprecated but will do precisely what you are seeing I think.
> It uses a half full watershead to decide to signal userspace.  If you wait
> till the buffer is full, unless your sampling is very slow you will have
> data coming in before userspace has read the whole buffer out. Hence you
> will loose data.
>
> You are right. I saw exatly what you said.
>
>
>> Kfifo has a more conventional poll and will fire off userspace
>> whenever
> any data is in the buffer. Hence kfifo will work for your usecase.  I
> would like to have a fill level capable buffer again in the future as they
> are very useful in high speed uses where latency doesn't matter.
>
> Ok, I will change my sw-ring buffer to kfifo!
>
> Thanks,
> Denis
> --
> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
> the body of a message to majordomo@vger.kernel.org More majordomo info at
> http://vger.kernel.org/majordomo-info.html
> --
> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Re: iio_buffer length

[Lars-Peter Clausen]

On 09/26/2012 09:27 AM, Jonathan Cameron wrote:
> On 26/09/12 02:32, Ge Gao wrote:
>> To use KFIFO, you need Jonanthan's KFIFO poll patch,
> That's in staging-next now and will merge into mainline in the
> next merge window (about 1 to 2 weeks time).
>> you probably also
>> need a patch from me that fixed the industrial-buffer.c. But that buffer
>> is not well tested and is only used by me.
> There are a lot of drivers using kfifo. It's hidden in many of them as they
> use it via the industrialio-triggered-buffer helper functions that
> deal with the most common use cases.
> 
> With that I count 9 in drivers/iio and 6 in drivers/staging/iio vs 10
> in total for sw_ring most of which will go away once someone has a spare
> couple of hours.

I've already converted all the adis* drivers locally, so it's basically down
to 4 remaining users.

- Lars

Re: iio_buffer length

[Denis Ciocca]

Hi Jon and Ge,

I switched my driver code by using kfifo (using code on linux-next),
now what I see is:

using generic-buffer application, I catch every sample independently
by the value of buffer length, is that what you tell me by "Kfifo has
a more conventional poll and will fire off userspace whenever any data
is in the buffer."?
To catch data only when the buffer is full, I must catch a particular
signal event in /dev/iio:deviceX?

Thanks

Denis




2012/9/26 Jonathan Cameron <jic23@jic23.retrosnub.co.uk>:
> On 26/09/12 02:32, Ge Gao wrote:
>>
>> To use KFIFO, you need Jonanthan's KFIFO poll patch,
>
> That's in staging-next now and will merge into mainline in the
> next merge window (about 1 to 2 weeks time).
>
>> you probably also
>> need a patch from me that fixed the industrial-buffer.c. But that buffer
>> is not well tested and is only used by me.
>
> There are a lot of drivers using kfifo. It's hidden in many of them as they
> use it via the industrialio-triggered-buffer helper functions that
> deal with the most common use cases.
>
> With that I count 9 in drivers/iio and 6 in drivers/staging/iio vs 10
> in total for sw_ring most of which will go away once someone has a spare
> couple of hours.
>
> Basically any driver that has come along since we moved to kfifo is the
> preferred choice and any that anyone has touched since then.
>
>
>> The original KFIFO does not
>> have poll functionality and has some other minor bug.
>
> If I've missed any patches from you on this (or just forgotten them ;)
> please do keep pestering me.
>
>>
>> Ge
>>
>>
>> -----Original Message-----
>> From: linux-iio-owner@vger.kernel.org
>> [mailto:linux-iio-owner@vger.kernel.org] On Behalf Of Denis Ciocca
>> Sent: Tuesday, September 25, 2012 2:35 PM
>> To: jic23@jic23.retrosnub.co.uk
>> Cc: linux-iio@vger.kernel.org
>> Subject: iio_buffer length
>>
>> Hi Jon,
>>
>>> Firstly please confirm you are using a kfifo buffer?
>>
>>
>> No, I use sw-ring.
>>
>>
>>> Swring is deprecated but will do precisely what you are seeing I think.
>>
>> It uses a half full watershead to decide to signal userspace.  If you wait
>> till the buffer is full, unless your sampling is very slow you will have
>> data coming in before userspace has read the whole buffer out. Hence you
>> will loose data.
>>
>> You are right. I saw exatly what you said.
>>
>>
>>> Kfifo has a more conventional poll and will fire off userspace
>>> whenever
>>
>> any data is in the buffer. Hence kfifo will work for your usecase.  I
>> would like to have a fill level capable buffer again in the future as they
>> are very useful in high speed uses where latency doesn't matter.
>>
>> Ok, I will change my sw-ring buffer to kfifo!
>>
>> Thanks,
>> Denis
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
>> the body of a message to majordomo@vger.kernel.org More majordomo info at
>> http://vger.kernel.org/majordomo-info.html
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>

Re: iio_buffer length

[Jonathan Cameron]

On 26/09/12 13:37, Denis Ciocca wrote:
> Hi Jon and Ge,
>
> I switched my driver code by using kfifo (using code on linux-next),
> now what I see is:
>
> using generic-buffer application, I catch every sample independently
> by the value of buffer length, is that what you tell me by "Kfifo has
> a more conventional poll and will fire off userspace whenever any data
> is in the buffer."?
> To catch data only when the buffer is full, I must catch a particular
> signal event in /dev/iio:deviceX?
>
You never want the buffer to be full.  If it is you WILL loose data
due to unlucky timing sooner or later.

Only real option if you want to grab data in chunks from that buffer
is to deadreckon the timing using a userspace timer.


> Thanks
>
> Denis
>
>
>
>
> 2012/9/26 Jonathan Cameron <jic23@jic23.retrosnub.co.uk>:
>> On 26/09/12 02:32, Ge Gao wrote:
>>>
>>> To use KFIFO, you need Jonanthan's KFIFO poll patch,
>>
>> That's in staging-next now and will merge into mainline in the
>> next merge window (about 1 to 2 weeks time).
>>
>>> you probably also
>>> need a patch from me that fixed the industrial-buffer.c. But that buffer
>>> is not well tested and is only used by me.
>>
>> There are a lot of drivers using kfifo. It's hidden in many of them as they
>> use it via the industrialio-triggered-buffer helper functions that
>> deal with the most common use cases.
>>
>> With that I count 9 in drivers/iio and 6 in drivers/staging/iio vs 10
>> in total for sw_ring most of which will go away once someone has a spare
>> couple of hours.
>>
>> Basically any driver that has come along since we moved to kfifo is the
>> preferred choice and any that anyone has touched since then.
>>
>>
>>> The original KFIFO does not
>>> have poll functionality and has some other minor bug.
>>
>> If I've missed any patches from you on this (or just forgotten them ;)
>> please do keep pestering me.
>>
>>>
>>> Ge
>>>
>>>
>>> -----Original Message-----
>>> From: linux-iio-owner@vger.kernel.org
>>> [mailto:linux-iio-owner@vger.kernel.org] On Behalf Of Denis Ciocca
>>> Sent: Tuesday, September 25, 2012 2:35 PM
>>> To: jic23@jic23.retrosnub.co.uk
>>> Cc: linux-iio@vger.kernel.org
>>> Subject: iio_buffer length
>>>
>>> Hi Jon,
>>>
>>>> Firstly please confirm you are using a kfifo buffer?
>>>
>>>
>>> No, I use sw-ring.
>>>
>>>
>>>> Swring is deprecated but will do precisely what you are seeing I think.
>>>
>>> It uses a half full watershead to decide to signal userspace.  If you wait
>>> till the buffer is full, unless your sampling is very slow you will have
>>> data coming in before userspace has read the whole buffer out. Hence you
>>> will loose data.
>>>
>>> You are right. I saw exatly what you said.
>>>
>>>
>>>> Kfifo has a more conventional poll and will fire off userspace
>>>> whenever
>>>
>>> any data is in the buffer. Hence kfifo will work for your usecase.  I
>>> would like to have a fill level capable buffer again in the future as they
>>> are very useful in high speed uses where latency doesn't matter.
>>>
>>> Ok, I will change my sw-ring buffer to kfifo!
>>>
>>> Thanks,
>>> Denis
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
>>> the body of a message to majordomo@vger.kernel.org More majordomo info at
>>> http://vger.kernel.org/majordomo-info.html
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

RE: iio_buffer length

[Ge Gao]

I updated the latest the staging next and noticed that your patch is
already in the kfifo along with other fixes. One thing I don't understand
why in the "iio_store_to_kfifo", only 1 byte is stored vs.
bytes_per_datum is stored before. Isn't KFIFO using bytes_per_datum when
reading and writing?
I send a patch for industrial-buffer to you.
Thanks.

Best Regards,

Ge GAO


-----Original Message-----
From: J.I. Cameron [mailto:jic23@hermes.cam.ac.uk] On Behalf Of Jonathan
Cameron
Sent: Wednesday, September 26, 2012 7:23 AM
To: Denis Ciocca
Cc: Ge Gao; linux-iio@vger.kernel.org
Subject: Re: iio_buffer length

On 26/09/12 13:37, Denis Ciocca wrote:
> Hi Jon and Ge,
>
> I switched my driver code by using kfifo (using code on linux-next),
> now what I see is:
>
> using generic-buffer application, I catch every sample independently
> by the value of buffer length, is that what you tell me by "Kfifo has
> a more conventional poll and will fire off userspace whenever any data
> is in the buffer."?
> To catch data only when the buffer is full, I must catch a particular
> signal event in /dev/iio:deviceX?
>
You never want the buffer to be full.  If it is you WILL loose data due to
unlucky timing sooner or later.

Only real option if you want to grab data in chunks from that buffer is to
deadreckon the timing using a userspace timer.


> Thanks
>
> Denis
>
>
>
>
> 2012/9/26 Jonathan Cameron <jic23@jic23.retrosnub.co.uk>:
>> On 26/09/12 02:32, Ge Gao wrote:
>>>
>>> To use KFIFO, you need Jonanthan's KFIFO poll patch,
>>
>> That's in staging-next now and will merge into mainline in the next
>> merge window (about 1 to 2 weeks time).
>>
>>> you probably also
>>> need a patch from me that fixed the industrial-buffer.c. But that
>>> buffer is not well tested and is only used by me.
>>
>> There are a lot of drivers using kfifo. It's hidden in many of them
>> as they use it via the industrialio-triggered-buffer helper functions
>> that deal with the most common use cases.
>>
>> With that I count 9 in drivers/iio and 6 in drivers/staging/iio vs 10
>> in total for sw_ring most of which will go away once someone has a
>> spare couple of hours.
>>
>> Basically any driver that has come along since we moved to kfifo is
>> the preferred choice and any that anyone has touched since then.
>>
>>
>>> The original KFIFO does not
>>> have poll functionality and has some other minor bug.
>>
>> If I've missed any patches from you on this (or just forgotten them
>> ;) please do keep pestering me.
>>
>>>
>>> Ge
>>>
>>>
>>> -----Original Message-----
>>> From: linux-iio-owner@vger.kernel.org
>>> [mailto:linux-iio-owner@vger.kernel.org] On Behalf Of Denis Ciocca
>>> Sent: Tuesday, September 25, 2012 2:35 PM
>>> To: jic23@jic23.retrosnub.co.uk
>>> Cc: linux-iio@vger.kernel.org
>>> Subject: iio_buffer length
>>>
>>> Hi Jon,
>>>
>>>> Firstly please confirm you are using a kfifo buffer?
>>>
>>>
>>> No, I use sw-ring.
>>>
>>>
>>>> Swring is deprecated but will do precisely what you are seeing I
think.
>>>
>>> It uses a half full watershead to decide to signal userspace.  If
>>> you wait till the buffer is full, unless your sampling is very slow
>>> you will have data coming in before userspace has read the whole
>>> buffer out. Hence you will loose data.
>>>
>>> You are right. I saw exatly what you said.
>>>
>>>
>>>> Kfifo has a more conventional poll and will fire off userspace
>>>> whenever
>>>
>>> any data is in the buffer. Hence kfifo will work for your usecase.
>>> I would like to have a fill level capable buffer again in the future
>>> as they are very useful in high speed uses where latency doesn't
matter.
>>>
>>> Ok, I will change my sw-ring buffer to kfifo!
>>>
>>> Thanks,
>>> Denis
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-iio"
>>> in the body of a message to majordomo@vger.kernel.org More majordomo
>>> info at http://vger.kernel.org/majordomo-info.html
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-iio"
>>> in the body of a message to majordomo@vger.kernel.org More majordomo
>>> info at  http://vger.kernel.org/majordomo-info.html
>>>
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-iio"
> in the body of a message to majordomo@vger.kernel.org More majordomo
> info at  http://vger.kernel.org/majordomo-info.html
>