Serial VS Diagnostic interface

Serial VS Diagnostic interface

[andrew wiggin]

Hi all,

I am working with eCos on an AT91 platform for several month now, and
I have always used the serial port as a diagnostic interface for
debugging purpose. However, now I need to use it as a standard serial
in order to be able to use the O_NONBLOCK flag but I couldn't find a
way to use the 16x55 serial driver instead of the haldiag interface:
whatever I do, only the haldiag interface is selected, and removing
the haldiag seems to make it impossible to compile.

I know I can use the solution consisting in replacing the GETC
function with a nonblocking one, and patching whatever function may
need blocking calls on it, but I can't imagine there is no way in eCos
to select whether my only serial port shall be considered as a
diagnostic interface or a _real_ serial port.

Any clue will be welcome.

Thanks,

-- 
Andrew W.

RE: Serial VS Diagnostic interface

[Chris Zimman]

> I am working with eCos on an AT91 platform for several month now, and
> I have always used the serial port as a diagnostic interface for
> debugging purpose. However, now I need to use it as a standard serial
> in order to be able to use the O_NONBLOCK flag but I couldn't find a
> way to use the 16x55 serial driver instead of the haldiag interface:
> whatever I do, only the haldiag interface is selected, and removing
> the haldiag seems to make it impossible to compile.
> 
> I know I can use the solution consisting in replacing the GETC
> function with a nonblocking one, and patching whatever function may
> need blocking calls on it, but I can't imagine there is no way in eCos
> to select whether my only serial port shall be considered as a
> diagnostic interface or a _real_ serial port.
> 
> Any clue will be welcome.

Make sure you have the hardware serial port drivers compiled in.  With that
you can use "/dev/ser0" etc. instead of "/dev/haldiag" and use non-blocking
interfaces.

--Chris

Re: Serial VS Diagnostic interface

[andrew wiggin]

On Wed, Jan 14, 2009 at 1:20 AM, Chris Zimman <czimman@bloomberg.com> wrote:
>> I am working with eCos on an AT91 platform for several month now, and
>> I have always used the serial port as a diagnostic interface for
>> debugging purpose. However, now I need to use it as a standard serial
>> in order to be able to use the O_NONBLOCK flag but I couldn't find a
>> way to use the 16x55 serial driver instead of the haldiag interface:
>> whatever I do, only the haldiag interface is selected, and removing
>> the haldiag seems to make it impossible to compile.
>>
>> I know I can use the solution consisting in replacing the GETC
>> function with a nonblocking one, and patching whatever function may
>> need blocking calls on it, but I can't imagine there is no way in eCos
>> to select whether my only serial port shall be considered as a
>> diagnostic interface or a _real_ serial port.
>>
>> Any clue will be welcome.
>
> Make sure you have the hardware serial port drivers compiled in.  With that
> you can use "/dev/ser0" etc. instead of "/dev/haldiag" and use non-blocking
> interfaces.
>
> --Chris
>

Actually as it just inline code. I have already checked it is
correctly included. Maybe I should have been more explicit: Is there
anything to ensure the HW serial driver is correctly initialized ?
Where should pc_serial_lookup() be called in a normal behavior ?

I would have another question then, I have noticed that with GCC 4.2.2
(which is the one I use), the DEVTAB_ENTRY may be removed if the
TABLE_END is declared before adding any additional entry. Is anybody
else having the same issue ?

-- 
Andrew W.

RE: Serial VS Diagnostic interface

[Chris Zimman]

> Actually as it just inline code. I have already checked it is
> correctly included. Maybe I should have been more explicit: Is there
> anything to ensure the HW serial driver is correctly initialized ?
> Where should pc_serial_lookup() be called in a normal behavior ?

If you can build the serial tests and they pass, you should be in good shape.
Alternatively, if opening "/dev/ser0" succeeds, you're good to go.
 
> I would have another question then, I have noticed that with GCC 4.2.2
> (which is the one I use), the DEVTAB_ENTRY may be removed if the
> TABLE_END is declared before adding any additional entry. Is anybody
> else having the same issue ?

I don't use GCC 4.x having moved to EABI a while ago, so I'm afraid I can't
offer any advice here.

--Chris

Re: Serial VS Diagnostic interface

[damien millescamps]

On Wed, Jan 14, 2009 at 2:24 PM, Chris Zimman <czimman@bloomberg.com> wrote:
>> Actually as it just inline code. I have already checked it is
>> correctly included. Maybe I should have been more explicit: Is there
>> anything to ensure the HW serial driver is correctly initialized ?
>> Where should pc_serial_lookup() be called in a normal behavior ?
>
> If you can build the serial tests and they pass, you should be in good shape.
> Alternatively, if opening "/dev/ser0" succeeds, you're good to go.
>

opening /dev/ser0 results in an error. Now that I know it should
normally work, I can start to instrument a little bit to find out why
it doesn't work for me.

Thanks,

-- 
Andrew W.

>> I would have another question then, I have noticed that with GCC 4.2.2
>> (which is the one I use), the DEVTAB_ENTRY may be removed if the
>> TABLE_END is declared before adding any additional entry. Is anybody
>> else having the same issue ?
>
> I don't use GCC 4.x having moved to EABI a while ago, so I'm afraid I can't
> offer any advice here.
>
> --Chris
>
>

Re: Serial VS Diagnostic interface

[Nick Garnett]

"damien millescamps" <damien.millescamps@gmail.com> writes:

> On Wed, Jan 14, 2009 at 2:24 PM, Chris Zimman <czimman@bloomberg.com> wrote:
> >> Actually as it just inline code. I have already checked it is
> >> correctly included. Maybe I should have been more explicit: Is there
> >> anything to ensure the HW serial driver is correctly initialized ?
> >> Where should pc_serial_lookup() be called in a normal behavior ?
> >
> > If you can build the serial tests and they pass, you should be in good shape.
> > Alternatively, if opening "/dev/ser0" succeeds, you're good to go.
> >
> 
> opening /dev/ser0 results in an error. Now that I know it should
> normally work, I can start to instrument a little bit to find out why
> it doesn't work for me.

The most common problem with serial devices is not enabling
CYGPKG_IO_SERIAL_DEVICES. Obviously you also have to include the
CYGPKG_IO_SERIAL package too.

Note that AT91 serial devices cannot be shared between diagnostics and
serial drivers. So if you want to continue using diagnostics on a
particular UART, you need to disable it for the drivers.


-- 
Nick Garnett                                      eCos Kernel Architect
eCosCentric Limited    http://www.eCosCentric.com      The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.     Tel: +44 1223 245571
Registered in England and Wales:                        Reg No: 4422071

Re: Serial VS Diagnostic interface

[damien millescamps]

On Wed, Jan 14, 2009 at 3:19 PM, Nick Garnett <nickg@ecoscentric.com> wrote:
> "damien millescamps" <damien.millescamps@gmail.com> writes:
>
>> On Wed, Jan 14, 2009 at 2:24 PM, Chris Zimman <czimman@bloomberg.com> wrote:
>> >> Actually as it just inline code. I have already checked it is
>> >> correctly included. Maybe I should have been more explicit: Is there
>> >> anything to ensure the HW serial driver is correctly initialized ?
>> >> Where should pc_serial_lookup() be called in a normal behavior ?
>> >
>> > If you can build the serial tests and they pass, you should be in good shape.
>> > Alternatively, if opening "/dev/ser0" succeeds, you're good to go.
>> >
>>
>> opening /dev/ser0 results in an error. Now that I know it should
>> normally work, I can start to instrument a little bit to find out why
>> it doesn't work for me.
>
> The most common problem with serial devices is not enabling
> CYGPKG_IO_SERIAL_DEVICES. Obviously you also have to include the
> CYGPKG_IO_SERIAL package too.
>
> Note that AT91 serial devices cannot be shared between diagnostics and
> serial drivers. So if you want to continue using diagnostics on a
> particular UART, you need to disable it for the drivers.
>

I imagine you mean CYGPKG_IO_SERIAL_<platform> instead of
CYGPKG_IO_SERIAL_DEVICES which doesn't exist ?

When you say it is not possible to use the Diagnostic interface and
the serial drivers, is it just because of the ISR declaration ? If
this is the case, I already know how to handle it, or are you talking
about another issue ?

-- 
Andrew W.

>
> --
> Nick Garnett                                      eCos Kernel Architect
> eCosCentric Limited    http://www.eCosCentric.com      The eCos experts
> Barnwell House, Barnwell Drive, Cambridge, UK.     Tel: +44 1223 245571
> Registered in England and Wales:                        Reg No: 4422071
>
>

Re: Serial VS Diagnostic interface

[Nick Garnett]

"damien millescamps" <damien.millescamps@gmail.com> writes:

> On Wed, Jan 14, 2009 at 3:19 PM, Nick Garnett <nickg@ecoscentric.com> wrote:
> > The most common problem with serial devices is not enabling
> > CYGPKG_IO_SERIAL_DEVICES. Obviously you also have to include the
> > CYGPKG_IO_SERIAL package too.
> >
> > Note that AT91 serial devices cannot be shared between diagnostics and
> > serial drivers. So if you want to continue using diagnostics on a
> > particular UART, you need to disable it for the drivers.
> >
> 
> I imagine you mean CYGPKG_IO_SERIAL_<platform> instead of
> CYGPKG_IO_SERIAL_DEVICES which doesn't exist ?

No, I really do mean CYGPKG_IO_SERIAL_DEVICES, which is an option in
the CYGPKG_IO_SERIAL package. It causes the serial devices to be
enabled. If you cannot see it, then that probably means that
CYGPKG_IO_SERIAL is not included or enabled.

> 
> When you say it is not possible to use the Diagnostic interface and
> the serial drivers, is it just because of the ISR declaration ? If
> this is the case, I already know how to handle it, or are you talking
> about another issue ?

It has more to do with the DMA engines. The PDC is used to overcome
the UART's lack of FIFOs. Once the drivers are loaded they point the
DMA engines at their own buffers, not those previously used by the
diagnostic channel. As a result the diagnostic channel code will no
longer work.


-- 
Nick Garnett                                      eCos Kernel Architect
eCosCentric Limited    http://www.eCosCentric.com      The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.     Tel: +44 1223 245571
Registered in England and Wales:                        Reg No: 4422071

Re: Serial VS Diagnostic interface

[andrew wiggin]

Thanks to all for your help.

Actually, my UART (which is a synthetisable IP eventually), has a
slight modification from the original 16X5X behavior. I have added a
DEFINE in order to declare an extra register and do some extra checks
in the Init and DSR functions.
This is quite an odd situation, because the documentation of the IP is
subject to non disclosure, but the fact that I modified some GPL code
means I should provide my modifications. I have thought of two
solutions: continuing with my old driver so that no GPL applies, or
propose a patch with the modifications without documenting for which
IP vendor it applies, and in which configuration...

How do you guys handle this kind of situation ?

-- 
Andrew W.

On Wed, Jan 14, 2009 at 4:09 PM, Nick Garnett <nickg@ecoscentric.com> wrote:
> "damien millescamps" <damien.millescamps@gmail.com> writes:
>
>> On Wed, Jan 14, 2009 at 3:19 PM, Nick Garnett <nickg@ecoscentric.com> wrote:
>> > The most common problem with serial devices is not enabling
>> > CYGPKG_IO_SERIAL_DEVICES. Obviously you also have to include the
>> > CYGPKG_IO_SERIAL package too.
>> >
>> > Note that AT91 serial devices cannot be shared between diagnostics and
>> > serial drivers. So if you want to continue using diagnostics on a
>> > particular UART, you need to disable it for the drivers.
>> >
>>
>> I imagine you mean CYGPKG_IO_SERIAL_<platform> instead of
>> CYGPKG_IO_SERIAL_DEVICES which doesn't exist ?
>
> No, I really do mean CYGPKG_IO_SERIAL_DEVICES, which is an option in
> the CYGPKG_IO_SERIAL package. It causes the serial devices to be
> enabled. If you cannot see it, then that probably means that
> CYGPKG_IO_SERIAL is not included or enabled.
>
>>
>> When you say it is not possible to use the Diagnostic interface and
>> the serial drivers, is it just because of the ISR declaration ? If
>> this is the case, I already know how to handle it, or are you talking
>> about another issue ?
>
> It has more to do with the DMA engines. The PDC is used to overcome
> the UART's lack of FIFOs. Once the drivers are loaded they point the
> DMA engines at their own buffers, not those previously used by the
> diagnostic channel. As a result the diagnostic channel code will no
> longer work.
>
>
> --
> Nick Garnett                                      eCos Kernel Architect
> eCosCentric Limited    http://www.eCosCentric.com      The eCos experts
> Barnwell House, Barnwell Drive, Cambridge, UK.     Tel: +44 1223 245571
> Registered in England and Wales:                        Reg No: 4422071
>
>

RE: Serial VS Diagnostic interface

[Chris Zimman]

> Actually, my UART (which is a synthetisable IP eventually), has a
> slight modification from the original 16X5X behavior. I have added a
> DEFINE in order to declare an extra register and do some extra checks
> in the Init and DSR functions.
> This is quite an odd situation, because the documentation of the IP is
> subject to non disclosure, but the fact that I modified some GPL code
> means I should provide my modifications. I have thought of two
> solutions: continuing with my old driver so that no GPL applies, or
> propose a patch with the modifications without documenting for which
> IP vendor it applies, and in which configuration...

If this is IP that was customized by you (eg. presumably specific to a given
part), why would a patch need to go back into the mainline tree?

WRT what the NDA covers in terms of the documents and how the GPL affects all
of that, you'd probably better get clarification from a lawyer.

Re: Serial VS Diagnostic interface

[Jonathan Larmour]

Chris Zimman wrote:
>> Actually, my UART (which is a synthetisable IP eventually), has a
>> slight modification from the original 16X5X behavior. I have added a
>> DEFINE in order to declare an extra register and do some extra checks
>> in the Init and DSR functions.
>> This is quite an odd situation, because the documentation of the IP is
>> subject to non disclosure, but the fact that I modified some GPL code
>> means I should provide my modifications. I have thought of two
>> solutions: continuing with my old driver so that no GPL applies, or
>> propose a patch with the modifications without documenting for which
>> IP vendor it applies, and in which configuration...
> 
> If this is IP that was customized by you (eg. presumably specific to a given
> part), why would a patch need to go back into the mainline tree?

I think he means the driver was customised, to adapt to the IP.

Andrew: in case it's any help, the GPL clause 3(a) releases you from any
obligation to make public any source code *provided* that you accompany any
binary image (or product containing the binary image) with the relevant
source code. This means that if you are only distributing it to people who
have also already signed the NDA, then as long as you include the source
code with it at the same time you should be ok (I should add a disclaimer:
I am not a lawyer!). If they want to distribute it further, it's up to them
to comply with the licence.

I believe the potential conflict with NDAs is dealt with under GPL clause
7, essentially: "If you cannot distribute so as to satisfy simultaneously
your obligations under this License and any other pertinent obligations,
then as a consequence you may not distribute the Program at all. "

So your own driver might be the better solution. I would doubt that having
a patch but not documenting the IP vendor would be sufficient to avoid the
letter of your NDA.

Alternatively it may be easiest to ping whoever you have the NDA with to
ask if its ok to make public code using that register. It doesn't sound to
me like one particular register in a UART is something that is really the
focus of the private IP they are trying to conceal by NDA. A lot of
companies just habitually have NDAs for everything that they aren't
actually making public themselves.

Of course I Am Not A Lawyer.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["Si fractum non sit, noli id reficere"]------       Opinions==mine

Re: Serial VS Diagnostic interface

[andrew wiggin]

> Andrew: in case it's any help, the GPL clause 3(a) releases you from any
> obligation to make public any source code *provided* that you accompany any
> binary image (or product containing the binary image) with the relevant
> source code. This means that if you are only distributing it to people who
> have also already signed the NDA, then as long as you include the source
> code with it at the same time you should be ok (I should add a disclaimer:
> I am not a lawyer!). If they want to distribute it further, it's up to them
> to comply with the licence.
>

Thanks for the clarification. Considering the code is only distributed
to people sharing the NDA, and that it is always distributed in source
format with all modifications clearly mentioned, everything should be
OK.

-- 
Andrew W.


> I believe the potential conflict with NDAs is dealt with under GPL clause
> 7, essentially: "If you cannot distribute so as to satisfy simultaneously
> your obligations under this License and any other pertinent obligations,
> then as a consequence you may not distribute the Program at all. "
>
> So your own driver might be the better solution. I would doubt that having
> a patch but not documenting the IP vendor would be sufficient to avoid the
> letter of your NDA.
>
> Alternatively it may be easiest to ping whoever you have the NDA with to
> ask if its ok to make public code using that register. It doesn't sound to
> me like one particular register in a UART is something that is really the
> focus of the private IP they are trying to conceal by NDA. A lot of
> companies just habitually have NDAs for everything that they aren't
> actually making public themselves.
>
> Of course I Am Not A Lawyer.
>
> Jifl
> --
> eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
> Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
> Registered in England and Wales: Reg No 4422071.
> ------["Si fractum non sit, noli id reficere"]------       Opinions==mine
>

RE: Serial VS Diagnostic interface

[Chris Zimman]

>> I would have another question then, I have noticed that with GCC 4.2.2 
>> (which is the one I use), the DEVTAB_ENTRY may be removed if the 
>> TABLE_END is declared before adding any additional entry. Is anybody 
>> else having the same issue ?

> I don't use GCC 4.x having moved to EABI a while ago, so I'm afraid I
> can't offer any advice here.

Just for the purposes of clarification:

The EABI toolchain is 4.x based, but I assumed what you were talking about is
the standard GNU arm-elf toolchain.  
The linker script is modified under EABI though, and I personally haven't
seen any issues with the tables.

--Chris