Date:	Mon, 30 Nov 2009 16:00:44 +0100
From:	sjur.brandeland@...ricsson.com
To:	netdev@...r.kernel.org, stefano.babic@...ic.homelinux.org
Cc:	randy.dunlap@...cle.com, kim.xx.lilliestierna@...ricsson.com,
	christian.bejram@...ricsson.com, daniel.martensson@...ricsson.com,
	Sjur Braendeland <sjur.brandeland@...ricsson.com>
Subject: [RFC PATCH v3 0/8] CAIF Protocol Stack

From: Sjur Braendeland <sjur.brandeland@...ricsson.com>

This is the third RFC on the CAIF patch set. CAIF is the primary protocol
used to communicate between to ST-Ericsson modems and the external host system.

Feedback on the patch set is welcomed!

This time I have reworked the architecture a bit.
The Link Layers in CAIF are now proper NET Devices of type ETH_P_CAIF,
and Character Devices are replaced by CAIF Sockets.

This patch set is compiled on x86 for 2.6.32.

Issues:
* Line Discipline implementation is working on 2.6.28,
  but not on 2.6.32.
* A few known stability issues when running high load over TTY.
* Different physical layers types (e.g. SPI) are currently identified in
  the "CAIF Protocol layer". CAIF Protocol layer should be ignorant
  of the kind of physical connection. This will be fixed later on.
* Too many log statements scattered throughout the code.
* Checks on LINUX_VERSION not yet removed.
* Global defines such as PF_CAIF, AF_CAIF, SOL_CAIF, N_CAIF, ETH_P_CAIF
  are defined locally as well. This duplication will be fixed  later on.
* CAIF Socket interface not all sockopt are implemented.
* RTNL configuration of CAIF is not yet tested.



Change Log:

General:
* Sockets has replaced the character device implementation,
  when accessing the modem. AT channels has to open sockets.
* CAIF Link layer is now implemented as net-devices.
* Cleanup of asserts no using caif_assert everywhere.
* Style cleanups, spacing, typedefs etc.

Documentation/CAIF
* Removed tool chardevconfig for configuration.
  Instead  CAIF NET device can be configured with RTNL or socket IOCTL.
* Removed replaced user space tool ldiscd to configure TTY.
  TTY is now opened from kernel when loading caif_serial.ko
* Added caif socket MAN page to Documentation (Will be moved later on).
* Added example on CAIF NET device configuration.
* Removed reference to  Doc No: 155 19-CRH 109 913.

include/linux
* Added caif_socket.h defining CAIF Socket interface.
* Added if_caif.h define CAIF IP Net Device configuration

include/net
* Added caif_dev.h defining interface between driver and caif-core.

include/net/generic
* Updated assert statement
* Removed typedefs

net/caif
* Added socket implementation and removed character device.
* General removed typedefs and fixed CHECKPATCH warnings
* Minor change on error handling in cfcnfg.c
* chnl_net.c Added RTNL support and IOCTL
* Fixed typos in Kconfig


I have merged feedback on previous patchset:

RE: [PATCH] [CAIF-RFC 0/8-v2] CAIF Protocol Stack
David Miller wrote:
> There is a lot of coding style issues here I'd like you to address in
> your next submission.
> Comments are formatted in several places like this:
>
> /*
> *
> *
> */
>
> It should be: ....
>
> /*
>  *
>  *
>  */
>
> Also often there are tons of extraneous empty lines in the source
> files.  Particularly right after the comment at the top of the source
> file.  For example include/net/caif/generic/cfcnfg.h there is the top
> comment (which needs to be fixed as described above) and then 8 empty
> lines before the #ifndef CFCNFG_H_  
>
> That's rediculious, just have one empty line there seperating things.
>
> Seriously, you should be able to just scan over your patch and see
> all of these oddities.  They jumped right out at me.  Please clean
> them up. 
I think I have fixed up License statements and got rid of empty lines. 
I'm afraid there are still style issues left, but I have tried to fix
those you pointed out and have ran checkpatch. Complaints on 
LINUX_VERSION_CODE will be fixed in a later patch-set.


RE: [PATCH] [CAIF-RFC 1/8-v2] CAIF Protocol Stack
Stefano Babic wrote:
>> + * \b Documentation see STE Doc No: 155 19-CRH 109 913.
> Probably you should remove this line as you already did in other
> places.
Yes, agree - I have done so.

RE: [PATCH] [CAIF-RFC 2/8-v2] CAIF Protocol Stack
Stefano Babic wrote:
> You actually removed the shared memory driver. Do you plan to insert
> it again ?
Well,  yes and no. The shared memory driver is actually back on the chart,
and we  have support for more drivers internally,
but I want to start contributing the main CAIF protocol before releasing more
drivers.

On the other hand, the CAIF protocol itself should not need to know the 
type of physical link protocol. I intend to remove the references to physical 
layer types (e.g. SPI) later on.

> There are references to pictures that are not provided.
Fixed: I have removed the references to the gifs.

>> +/* ASSERT */
>> +#define cfglu_assert(exp) BUG_ON(!(exp))
>
> I do not why, but even GENERIC_BUG is not defined for all
> architectures.
> This means that BUG_ON is simply removed by the compiler and we get
> no track if the assert fails.
> It should be better to add at least an internal trace (adding for
> example a CFLOG_FATAL call).
Fixed: We have updated caif_assert with BUG_ON and printout.

> Is this file probably obsolete ?
Yes, but I'm not sure I'm able to remove all legacies right now.

>
> Not sure I have understood. There is an abstraction layer for SPI in
> kernel and some generic purpose functions are provided (spy_sync,
> spi_async, etc.) that are HW independent. I know they provide only
> the methods for the data transfer, but I have imagined that we need
> at this point some kind of "caif_spi_device" that is able to talk
> with the caif stack on one side and uses the functions in the SPI
> framework. In any case, HW independent. Have I missed something ?   

As mentioned above we have internally support for physical link layers 
used for CAIF like SPI and others.
I hope we'll be able to contribute this in a later stage.
This is why the source code indicates support for things distributed.
But as previously mentioned, I'm planning to remove all references to 
physical devices (except the Serial Driver) in the general CAIF part in 
our next patch-set

RE: [PATCH] [CAIF-RFC 4/8-v2] CAIF Protocol Stack
Stefano Babic wrote:
> sjur.brandeland@...ricsson.com wrote:
> s/happends/happens/, there are other occurencies.
Done, thanks
>
>> +
>> +bool cfcnfg_del_adapt_layer(struct _cfcnfg_t *cnfg, layer_t
>> +*adap_layer) { + uint8 channel_id = 0;
>> + struct cfcnfg_phyinfo *phyinfo = NULL;
>> + uint8 phyid = 0;
>> + CFLOG_TRACE(("cfcnfg: enter del_adaptation_layer\n")); +
>> + cfglu_assert(adap_layer != NULL);
>> + channel_id = adap_layer->id;
>> + cfglu_assert(channel_id != 0);
>
> My two cents about using assert in the code, but I prefer to get some
> info from system when something goes wrong as to call in some way
> panic (the assert calls BUG_ON) and blocks forever. This line is not
> different as checking the adapt_layer->dn pointer some lines after
> and I think an error is better recognized in that case. So IMHO
> should be better something like:   
>
> if (channel_id == 0) {
> CFLOG_ERROR(("cfcnfg:adap_layer->dn is NULL\n"));
> return CFGLU_EINVAL;
> }
Ok, I have fixed this particular case as you have suggested.
I have also updated cfglu_assert to do printout before calling BUG_ON.
My thinking is to use asserts on internal in-variants. But error checking
on APIs e.g. between physical interface drivers and CAIF Stack should at
least be proper checks as you suggested.
Please check what you think of the asserts in this release and get back to be!

>
>> diff --git a/net/caif/generic/cfpkt_plain.c
> Probably you plan to use always the magic number in your buffer
> management. Then should be better to remove all the #if CHECK_MEM
> stuff. 
The file cfpkt_plain.c is removed, the file cfpkt_pain.c should be
used instead.

RE: [PATCH] [CAIF-RFC 5/8-v2] CAIF Protocol Stack
Stefano Babic wrote:
> sjur.brandeland@...ricsson.com wrote:
>> From: Sjur Braendeland <sjur.brandeland@...ricsson.com>
>
> Hi Sjur,
>
>> diff --git a/net/caif/caif_chr.c b/net/caif/caif_chr.c
>
>> +#define caif_assert(assert) BUG_ON(!(assert))
>
> Do we need special assert for each module (cfglu_assert,
> caif_assert,...) ? They are all defined in the same way.
> At this point I have already set a comment about using BUG_ON in a
> previous patch.
OK, Fixed we should have 'caif_assert' all over.

>
> I see a mixed policy in this patch, using sometimes _assert and
> sometimes directly BUG_ON, too.
Fixed.

>
>> + mutex_lock_interruptible(&dev->mutex);
> The return value is not checked and it must be, as in all other cases.
OK Fixed.


RE: [PATCH] [CAIF-RFC 5/8-v2] CAIF Protocol Stack
Randy Dunlap wrote:
>> diff --git a/net/caif/Kconfig b/net/caif/Kconfig new file mode 100644
>> + Say Y here if you need to use a phone modem that uses CAIF as
>> +transport
>
> end above with period ('.').
OK, thanks.
>
>> + You will also need to say yes to any caif physical devices that
>> your platform + supports. + This can be either built-in or as a
>> loadable module, if you select +to build it as module
>
> s/,/;/
OK, thanks.
>
>> + the other CAIF also needs to built as modules
>
> the other CAIF {options or drivers or some other word here} also
> need  ... modules. (end with period)
>
>
...
>> +
>> +config  CAIF_USE_PLAIN
>> + bool  "Use plain buffers instead of SKB in caif"
>> + default n
>> + ---help---
>> + Use plain buffer to transport data,
>
> s/,/./
OK, thanks.
>
...
>> + Enable the inclusion of debug code in the caif stack,
>> + be aware that doing this will impact performance. + If unsure say
>> N here. +
>> +# Include physical drivers
>> +# source "drivers/net/caif/Kconfig"
>
> Drop the above line.
OK, thanks.


RE: [PATCH] [CAIF-RFC 6/8-v2] CAIF Protocol Stack
Stefano Babic wrote:
> s/disicpline/discipline/
OK, Removed the line discipline completely.
>
>> diff --git a/drivers/net/caif/phyif_loop.c
>
> Is there a reason why do you prefer to implement your own ring buffer
> management else to use the circ_buf already implemented in kernel ?
I've re-written and simplified this loopback device completely.
>
>> +#define phyif_assert(assert) BUG_ON(!(assert))
>
> Specialized assert function for this module, really needed ?
Nop, this has also gone.
>
>> +
>> + result = tty_register_ldisc(N_MOUSE, &phyif_ldisc);
>
> I think it is time to set up your own discipline include/linux/tty.h,
> inserting a N_CAIF line discipline. Reusing other discipline
> conflicts with other drivers. 

OK, I will look into this.

RE: [PATCH] [CAIF-RFC 7/8-v2] CAIF Protocol Stack
Stefano Babic wrote:
> sjur.brandeland@...ricsson.com wrote:
>> +ST-Ericsson modems support a number of transports between modem and
>> +host, currently Uart and Shared Memory are available for Linux.
>
> Shared Memory was removed in this patchset.
OK - s/Shared Memory/Loopback
>
>> +== CAIF structure ==
>> +
>> +The goal is to have caif as system independent as possible.
>> +All caif code can be found under GenCaif/src and GenCaif/inc.
>
> The path are wrong, I think you mean net/caif/generic and
> include/net/caif.
OK - corrected path.
>
>> +The actual linux module implementation is under src/kernel.
>> +There is also a user space program that is not up to date to run the
>> +stack in user space for testing.
>
> I think you can remove these phrases.
OK, done.
>
>> +
>> +We have tested the kernel implementation on the emulator with a
>> modem +and we are able to enumerate and make a link setup.
>
> Only as comment: I tested this patchset again on an ARM platform and
> I am able to send successfully AT commands to the Ericsson Test
> Device "B26". 
Great! It is reassuring to hear that you're able to run CAIF :-)

>
>> +      - CFSHML CAIF Shared Memory layer.
>
> Again, this layer was removed and there is no track about SPI Layer.
Done, removed.
>
>> +To achieve this we need the help of a daemon program called ldiscd.
>> +The benefit is that we can hook up to any TTY, the downside is that
>> +we need an extra operation in order to install the line discipline.
>
> I understand the reason, it looks only a quite odd that we need to
> start only for this purpose a user space program. And there is no
> hint if the daemon is not started, simply caif does not work. What do
> you think to set up the line discipline directly in kernel ?  

OK, I have removed the line discipline now and moved this code
into kernel. The tty to be used is now a module argument.

Ref. "Re: [PATCH] [CAIF-RFC 6/8-v2] CAIF Protocol Stack".

>> +Install the VEI channel (this will enumerate and do the linksetup
>> of the first VEI channel. If this goes well you should see
>> /dev/chn*) (There are printks logging all buffers that can be
>> checked with dmesg): +$ modprobe chnl_chr + +The AT (VEI) channel is
>> ready to use (you can now send AT commands on it): 
>
> Not really. at this point, the channels are not configured if we do
> not use chardevconfig (or we do the same in kernel).

Agree, I have removed this paragraph.

>
>> diff --git a/Documentation/CAIF/caif_user_config.dox
>> b/Documentation/CAIF/caif_user_config.dox
>
> This file seems obsolete. There is no track about ACTIVATE/DEACTIVATE.
Yes, Agree I have removed this file completely.
This patch set removes the use of character devices for access and configuration,
and introduces Sockets and RTNL instead.

>> diff --git a/Documentation/CAIF/ldiscd/ldiscd.c
>> b/Documentation/CAIF/ldiscd/ldiscd.c
>> +#define CAIF_LDISC_TTY "/dev/ttyS0"
>
> I think it should not be bad to avoid a fixed device name and take it
> as program parameter.
Yes, agree. It is now a module paramter to caif_serial.c

BR/Sjur Brændeland
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists