Garrett D'Amore | 4 Jan 23:01 2006

review request repeated....

All,

Hope everyone had a fun and safe New Years, and a merry Christmas or
whatever other holiday(s) you may happen to celebrate.

I'm still looking for reviews for a few bits, and I need to get this
stuff integrated before I can submit the larger set of changes for PCI
(largely to resolve the dependency matrix).

If someone could review these and commit as appropriate, I'd be
grateful.  Here is the list:

port-evbmips/31992 alchemy ICU is too Au1000 specific
port-evbmips/32355 USB clocking incorrect for Au1550 boards
port-evbimps/32086 Improve baud rate selection of console

Granted, the last two aren't strictly required, but I'd have to rework
the code to split these out, as my current PCI code already has these
changes in it.

I'd also like it if someone could at least take a quick gander at

port-evbmips/32362 Alchemy aucom serial ports do not have /dev/ node

just to make sure that my analysis is correct on it.  (I'm still new
enough to NetBSD and since this the first time I've mucked about with
the /dev node logic in MAKEDEV, I'd like someone to double check it.) 
If it is good, I can submit a patch, but right now it isn't crucial for
my work.

(Continue reading)

Garrett D'Amore | 25 Jan 00:04 2006

Alchemy Au15XX PCI diffs

Everyone,

I've posted my latest diffs for Alchemy PCI support (and probably a few
other Alchemy fixes as well -- I didn't want to split them all out --
kind of painful to do that) into a diff file at

    http://garrett.damore.org/software/netbsd/aupci.diff

The diffs are about 7K lines, and are done against -current updated last
night.  They appear to work properly (at least I'm able to use a RealTek
gigE card without problem).

There appears to be at least two debugging lines still lurking in the
source, and I'll remove those before final commit to CVS.

I would love to have someone review and (for those with hardware) test
these changes.

These changes introduce three kernel configs in evbmips:

    * ALCHEMY - generic Au1550 support (no PCI support) -- same as the
old PB1000 file basically.

    * ZINFANDEL - Au1500 development board support

    * CABERNET - Au1550 development board support

I have removed the old PB1000 file.

Configs for other boards/hardware can be created if someone wants to
(Continue reading)

Matt Thomas | 25 Jan 00:33 2006

Re: Alchemy Au15XX PCI diffs

Garrett D'Amore wrote:
> Everyone,
> 
> I've posted my latest diffs for Alchemy PCI support (and probably a few
> other Alchemy fixes as well -- I didn't want to split them all out --
> kind of painful to do that) into a diff file at
> 
>     http://garrett.damore.org/software/netbsd/aupci.diff
>
> Constructive criticism is welcomed.  Flames to /dev/null.

Use const more.  For instance, obiodev_t arrays should be const.
In machdep.c, cpus[] should be const and make name [8] instead of
a pointer.

be consistent.  either use obiodev_t everywhere or struct obiodev
but don't mix them.

Add file-system TMPFS to the config files.

Are the interrupts evcnt'ed?
--

-- 
Matt Thomas                     email: matt <at> 3am-software.com
3am Software Foundry              www: http://3am-software.com/bio/matt/
Cupertino, CA              disclaimer: I avow all knowledge of this message.

Garrett D'Amore | 25 Jan 01:01 2006

Re: Alchemy Au15XX PCI diffs

Matt Thomas wrote:
> Garrett D'Amore wrote:
>   
>> Everyone,
>>
>> I've posted my latest diffs for Alchemy PCI support (and probably a few
>> other Alchemy fixes as well -- I didn't want to split them all out --
>> kind of painful to do that) into a diff file at
>>
>>     http://garrett.damore.org/software/netbsd/aupci.diff
>>
>> Constructive criticism is welcomed.  Flames to /dev/null.
>>     
>
> Use const more.  For instance, obiodev_t arrays should be const.
> In machdep.c, cpus[] should be const and make name [8] instead of
> a pointer.
>   
Good suggestions.

Out of curiosity, is there some benefit to declaring name[8] other than
the slight storage improvement -- perhaps 4 pointers in this case, but
I'd also think this would be offset slightly by the fact that the
strings have to be aligned and same sized whereas with a pointer and
strings stored off in text they can be packed.   (In this case it is a
win to use name[8], but if you had strings of very different sizes, the
other way with packed strings could be more space efficient.)

> be consistent.  either use obiodev_t everywhere or struct obiodev
> but don't mix them.
(Continue reading)

Garrett D'Amore | 26 Jan 02:46 2006

alchemy USB clocking problems....


Anyone have any problem with me committing the fix for the USB clocking issue 
raised in PR 32355?  (The fix basically removes the explicit programming of 
the USB clock registers and leaves things as YAMON has set them)?

Recall this only happens (the bug) on Au1550 boards.  The fix could impact 
other boards, but I've tested on both DBAu1500 and DBAu1550 boards.

--

-- 
Garrett D'Amore, Principal Software Engineer
Tadpole Computer / Computing Technologies Division,
General Dynamics C4 Systems
http://www.tadpolecomputer.com/
Phone: 951 325-2134  Fax: 951 325-2191
Garrett D'Amore | 26 Jan 19:05 2006

PCI diffs/patches & copyright

FYI, 

I have been advised that as Tadpole Computer is now a division of General 
Dynamics, there *may* be a need to have the copyright ownership notices in 
the diffs I sent out changed.  I'm still waiting legal approval and final 
verification of which entity should have ownership (there are at least 4 
possibilities -- gotta love corporate politics), but I don't think there is 
likely to be any reason that we won't still want it released under a 3-clause 
license. 

(At some level, if they tried to suppress that, I suppose folks could claim 
that the code had *already* been released by my earlier post, and the license 
could not be revoked.  I would personally defend such a claim, as I had prior 
permission to release it.  Anyway, I consider it a very remote possibility, 
but when lawyers get involved you can never be certain. :-)

Anyway, please do *not* commit the diffs I've posted yet (I have a CVS account 
now and can do so myself) until I find out if GD wants the copyright 
ownership changed or not.

For the curious, Tadpole Computer was purchased last year by General Dynamics, 
who also purchased Itronix.  Tadpole's operations have largely been merged 
with Itronix, so the four possible entities that could be named on a 
copyright are:

	Tadpole Computer
	Itronix
	General Dynamics C4 Systems
	General Dynamics

(Continue reading)

Garrett D'Amore | 27 Jan 02:42 2006

Re: Alchemy Au15XX PCI diffs

On Tuesday 24 January 2006 3:33 pm, Matt Thomas wrote:
> Garrett D'Amore wrote:
> > Everyone,
> >
> > I've posted my latest diffs for Alchemy PCI support (and probably a few
> > other Alchemy fixes as well -- I didn't want to split them all out --
> > kind of painful to do that) into a diff file at
> >
> >     http://garrett.damore.org/software/netbsd/aupci.diff
> >
> > Constructive criticism is welcomed.  Flames to /dev/null.
>
> Use const more.  For instance, obiodev_t arrays should be const.
> In machdep.c, cpus[] should be const and make name [8] instead of
> a pointer.

I've made this change.

>
> be consistent.  either use obiodev_t everywhere or struct obiodev
> but don't mix them.

Looks like I started to convert to obiodev_t, but didn't bother to change all 
of the orginal code from which this was derived.  I've completed the change 
now.

>
> Add file-system TMPFS to the config files.

Hmmm... upon inspection, a lot of ports do not include this by default (e.g. 
(Continue reading)

Garrett D'Amore | 27 Jan 19:25 2006

last call on USB clocking fix..

The silence on my inquiry regardng the USB clocking fix for Alchemy parts has 
been deafening.  Can I take this as a general consensus that it is fine for 
me to commit that (or total apathy because no one is impacted, so I should 
also just commit it?)
--

-- 
Garrett D'Amore, Principal Software Engineer
Tadpole Computer / Computing Technologies Division,
General Dynamics C4 Systems
http://www.tadpolecomputer.com/
Phone: 951 325-2134  Fax: 951 325-2191
Garrett D'Amore | 27 Jan 22:14 2006

converted web page to XML... added Alchemy notes

FYI:

I have converted the port-evbmips page to XML following the other standards I 
could find in the NetBSD tree.  I've also added a few notes about the Alchemy 
(basically just adding DBAu1500 and DBAu1550 board to the list of supported 
models.)

I want to add stuff about periphs and whatnot, but that can wait.

The old stuff is still in CVS in case you object to any of this.

Updates to the published page should happen in 30 mins or so.

	-- Garrett
Izumi Tsutsui | 28 Jan 17:57 2006
Picon

Re: Alchemy Au15XX PCI diffs

In article <43D6B26E.1070409 <at> tadpole.com>
garrett_damore <at> tadpole.com wrote:

>     http://garrett.damore.org/software/netbsd/aupci.diff
 :
> Constructive criticism is welcomed.  Flames to /dev/null.

Mostly it looks fine for me, but I'd like some more cosmetics:

> +extern void board_init(void);

I think function declarations don't need "extern".

> +	return (cabernet_obio);

No parentheses are needed around the return value, as share/misc/style.
(though many other sources have them)

> +evbmips_iointr(u_int32_t status, u_int32_t cause, u_int32_t pc,
> +    u_int32_t ipending)

uintXX_t is better rather than u_intXX_t.

> +# CPU support
> +options		ALCHEMY_AU1000

Use "options<space><tab>".

> +options 	DIAGNOSTIC	# extra kernel sanity checking

(Continue reading)


Gmane