Re: [Xenomai] access SPI and DMA hardware from application ("external abort...")

Re: [Xenomai] access SPI and DMA hardware from application ("external abort...")

[Reitinger, Peter]

Hello,

meanwhile I have solved the issue on my own. I tried to access the SPI1 HW with traditional mmap approach from user space and had the same problem until I saw that I had to start the whole SPI1 module by setting a specific control bit to enable its clock. Sorry for posting this, but I was really puzzled then. I guess it will also be possible to access the hw registers of the SPI with the xenomai rtdm io user mapping now...

Kind regards,
Peter Reitinger

Re: [Xenomai] access SPI and DMA hardware from application ("external abort...")

[Gilles Chanteperdrix]

On 08/15/2014 08:47 AM, Reitinger, Peter wrote:
> Hello,
> 
> meanwhile I have solved the issue on my own. I tried to access the
> SPI1 HW with traditional mmap approach from user space and had the
> same problem until I saw that I had to start the whole SPI1 module by
> setting a specific control bit to enable its clock. Sorry for posting
> this, but I was really puzzled then. I guess it will also be possible
> to access the hw registers of the SPI with the xenomai rtdm io user
> mapping now...

It is normally recommended to implement drivers in kernel-space using
the RTDM API: this forces you to define a standard interface
(open/read/write/ioctl) between application and driver, which makes it
easy to re-use the same driver in a different application, or to
implement a driver for a differrent hardware without changing your
application.

-- 
                                                                Gilles.

Re: [Xenomai] access SPI and DMA hardware from application ("external abort...")

[Gilles Chanteperdrix]

On 08/19/2014 02:13 PM, Reitinger, Peter wrote:
> Hello,
> 
> so I am not up-to-date that it is no longer recommended to do as much
> in user space as possible?

It has never been recommended. Maybe by people developing micro-kernels,
but Linux is not a micro-kernel.

> I see here a relation to the fact that
> interrupt api is going to be removed (or is already removed in 3.0)?

The user-space interrupt API were removed from 3.0 (but replaced by the
UDD framework, see recent commits in the xenomai-forge repository),
because they were a sure way to shoot one-self in the foot. But also
note that the kernel-space API were removed from xenomai-forge, because
moving everything to kernel-space is not recommended either.

> 
> But, it is correct that all kernel-space sources have to be GPL?

Yes, but normally, a driver is dumb (if you read the excellent "Linux
device drivers", I believe the advice is "no policy in kernel-space"),
so there is nothing to hide. Do you really think that an SPI controller
driver is worth hiding?

Please no top-posting, nor private mails (I usually ignore private mails).

-- 
                                                                Gilles.