Broadcom 4702?

Broadcom 4702?

[Charlie Brady]

I haven't found signs of it in the archives, but is anyone aware of any 
efforts to fold in Broadcom's support for their 4702 processor, as used in 
Wireless gateways such as the Linksys WRT54G? Source code for their kernel 
port can be found here:

  http://www.linksys.com/support/gpl.asp

Would someone from Broadcom prefer to provide the patches?

--
Charlie

Re: Broadcom 4702?

[Jun Sun]

On Tue, Jan 13, 2004 at 05:00:27PM -0500, Charlie Brady wrote:
> 
> I haven't found signs of it in the archives, but is anyone aware of any 
> efforts to fold in Broadcom's support for their 4702 processor, as used in 
> Wireless gateways such as the Linksys WRT54G? Source code for their kernel 
> port can be found here:
> 
>   http://www.linksys.com/support/gpl.asp
> 
> Would someone from Broadcom prefer to provide the patches?
> 

Mvista is in the process of supporting bcm4704, which should be close to
bcm4702.  If there is much interest, we can push the patch out to 
the linux-mips.org tree.

Jun

Re: Broadcom 4702?

[Charlie Brady]

On Tue, 13 Jan 2004, Jun Sun wrote:

> Mvista is in the process of supporting bcm4704, which should be close to
> bcm4702.  If there is much interest, we can push the patch out to 
> the linux-mips.org tree.

I'm certain that there'd be people make use of the code if you put it 
there.

I've noticed that Broadcom have patched binutils and gcc, to work around 
some hardware bugs. Have you found that necessary with the bcm4704?

--
Charlie

Re: Broadcom 4702?

[Jim Thompson]

My question is why these patches, which are subject to GPL (it is 
binutils and gcc, after all), haven't been released.

On Jan 14, 2004, at 12:49 PM, Charlie Brady wrote:

>
> On Tue, 13 Jan 2004, Jun Sun wrote:
>
>> Mvista is in the process of supporting bcm4704, which should be close 
>> to
>> bcm4702.  If there is much interest, we can push the patch out to
>> the linux-mips.org tree.
>
> I'm certain that there'd be people make use of the code if you put it
> there.
>
> I've noticed that Broadcom have patched binutils and gcc, to work 
> around
> some hardware bugs. Have you found that necessary with the bcm4704?
>
> --
> Charlie
>
>

Re: Broadcom 4702?

[Alan Cox]

On Mer, 2004-01-14 at 21:05, Jim Thompson wrote:
> My question is why these patches, which are subject to GPL (it is 
> binutils and gcc, after all), haven't been released.

There is no obligation for anyone to provide the source except to those
they provide the binaries. They just can't stop those people then 
redistributing it.

Re: Broadcom 4702?

[Jun Sun]

On Wed, Jan 14, 2004 at 03:49:09PM -0500, Charlie Brady wrote:
> 
> On Tue, 13 Jan 2004, Jun Sun wrote:
> 
> > Mvista is in the process of supporting bcm4704, which should be close to
> > bcm4702.  If there is much interest, we can push the patch out to 
> > the linux-mips.org tree.
> 
> I'm certain that there'd be people make use of the code if you put it 
> there.
> 
> I've noticed that Broadcom have patched binutils and gcc, to work around 
> some hardware bugs. Have you found that necessary with the bcm4704?
> 

Not to my knowledge.

Jun

Re: Broadcom 4702?

[Jim Thompson]

I have binaries.  I asked.  I was told "no".

On Jan 14, 2004, at 1:15 PM, Alan Cox wrote:

> On Mer, 2004-01-14 at 21:05, Jim Thompson wrote:
>> My question is why these patches, which are subject to GPL (it is
>> binutils and gcc, after all), haven't been released.
>
> There is no obligation for anyone to provide the source except to those
> they provide the binaries. They just can't stop those people then
> redistributing it.
>
>

Broadcom gcc/binutils changes (was Re: Broadcom 4702?)

[Charlie Brady]

On Wed, 14 Jan 2004, Jim Thompson wrote:

> My question is why these patches, which are subject to GPL (it is 
> binutils and gcc, after all), haven't been released.

They have, which is why I know about them, and what they do. See URL in my 
post which started this thread.

[What's released is unfortunately a tarball, not a patch. More work for 
us, I guess.]

--
Charlie

Broadcom gcc/binutils mods (Re: Broadcom 4702?)

[Charlie Brady]

On Wed, 14 Jan 2004, Jim Thompson wrote:

> I have binaries.  I asked.  I was told "no".

They got better. Or more accurately, Linksys/Cisco got better for them.

See WRT54G/tools-src/gnu-20010422/ in wrt54g.1.42.3.tar.gz

Trendware, Buffalo and Belkin from busybox's Hall of Shame all ship linux
based wireless routers using Broadcom CPUs, with no source available:

http://www.busybox.net/shame.html

I suspect that Broadcom should share a large part of the [bl,sh]ame.

--
Charlie

Re: Broadcom 4702?

[Alan Cox]

On Mer, 2004-01-14 at 21:42, Jim Thompson wrote:
> I have binaries.  I asked.  I was told "no".

Please report that to the Free Software Foundation. If Broadcom provided
you the binaries and refused to provide you the source they are in
violation of license and the FSF as copyright holder will be happy to
take a (*polite* initially) hammer to their toenails...

Also beware of stupid support people. Some of them come preprogrammed
with "you cant have the source" and you have to go via their managers
for GPL stuff.

Re: Broadcom 4702?

[Jun Sun]

On Wed, Jan 14, 2004 at 01:42:01PM -0800, Jim Thompson wrote:
> 
> I have binaries.  I asked.  I was told "no".
> 
> On Jan 14, 2004, at 1:15 PM, Alan Cox wrote:
> 
> > On Mer, 2004-01-14 at 21:05, Jim Thompson wrote:
> >> My question is why these patches, which are subject to GPL (it is
> >> binutils and gcc, after all), haven't been released.
> >
> > There is no obligation for anyone to provide the source except to those
> > they provide the binaries. They just can't stop those people then
> > redistributing it.
> >

Since we are on this subject, I am curious if I buy a Cisco's router
whether it is considered that Cisco distributs the binaries to me
and whether I can demand for the source code if they are GPL'ed software.

I can see arguments go either way.  Do open source community and
industry have some concensus on this issue?

Jun

Re: Broadcom 4702?

[Charlie Brady]

On Wed, 14 Jan 2004, Jun Sun wrote:

> Since we are on this subject, I am curious if I buy a Cisco's router
> whether it is considered that Cisco distributs the binaries to me
> and whether I can demand for the source code if they are GPL'ed software.

You don't need to demand the source code to Cisco/linksys's linux based 
wireless routers. It's available for download from the URL I provided at 
the start of the thread.

> I can see arguments go either way.  Do open source community and
> industry have some concensus on this issue?

Please read the various licenses (GPL and other), and consult your lawyer.
And if you want a definitive answer (in your jurisdiction) get the license 
tested in Court (and subsequent Appeals Courts). :-)

--
Charlie

Re: Broadcom 4702?

[Dominic Sweetman]

> On Wed, 14 Jan 2004, Jun Sun wrote:
> 
> > Since we are on this subject, I am curious if I buy a Cisco's router
> > whether it is considered that Cisco distributs the binaries to me
> > and whether I can demand for the source code if they are GPL'ed software.
> ...
> > I can see arguments go either way.  Do open source community and
> > industry have some concensus on this issue?

Charlie replied...

> Please read the various licenses (GPL and other), and consult your
> lawyer.  And if you want a definitive answer (in your jurisdiction)
> get the license tested in Court (and subsequent Appeals Courts). :-)

But Jun Sun asked whether distribution of a binary in a ROM inside a
black box might not really count as distribution.  I don't think you
need a lawyer to resolve that.  If the software was a computer game
(for example) I can't quite see its commercial owners smiling
indulgently and saying "it's only a ROM, carry on...", and I don't see
them having trouble over their position in court.

So yes, binary code distributed in a black box is still distributed,
and if it was GPL software you are entitled to the source code.  It's
sensible of Cisco to put it quietly on a web site somewhere.

--
Dominic Sweetman
(not necessarily the view of MIPS Technologies)

Re: Broadcom 4702?

[Charlie Brady]

On Tue, 13 Jan 2004, Charlie Brady wrote:

> I haven't found signs of it in the archives, but is anyone aware of any 
> efforts to fold in Broadcom's support for their 4702 processor, as used in 
> Wireless gateways such as the Linksys WRT54G?

FWIW, there was some mention of this on lkml.

http://testing.lkml.org/slashdot.php?mid=313689

Looks like it may have quickly been put in the "Too Hard" basket. The 
bulk of the 15Mb patch, however, is not a port per se, but addition of 
kdbg and XFS, so there won't be anywhere near that much real work.

Here's an important one, however, which I alluded to yesterday:

+ifdef CONFIG_BCM4710
+GCCFLAGS       += -m4710a0kern
 endif

I haven't tried building and running a kernel built without the gcc 
workarounds, so I don't know whether they are only required for early 
silicon. My guess would be not. Is there anyone from Broadcom here who 
knows or can find out?

--
Charlie

Re: Broadcom 4702?

[John W. Linville]

Charlie Brady wrote:

>+ifdef CONFIG_BCM4710
>+GCCFLAGS       += -m4710a0kern
> endif
>
>I haven't tried building and running a kernel built without the gcc 
>workarounds, so I don't know whether they are only required for early 
>  
>
I don't know about the 4710 or 4702 (I haven't got around to that yet), 
but the 4704 doesn't seem to need any special flags for building the 
kernel (or anything else).

Of course, YMMV...

John

-- 
John W. Linville
LVL7 Systems, Inc.

Broadcom gcc changes (was Re: Broadcom 4702?)

[Charlie Brady]

On Thu, 15 Jan 2004, John W. Linville wrote:

> Charlie Brady wrote:
> 
> >+ifdef CONFIG_BCM4710
> >+GCCFLAGS       += -m4710a0kern
> > endif
> >
> >I haven't tried building and running a kernel built without the gcc 
> >workarounds, so I don't know whether they are only required for early 
> >  
> >
> I don't know about the 4710 or 4702 (I haven't got around to that yet), 
> but the 4704 doesn't seem to need any special flags for building the 
> kernel (or anything else).

The acid test is not whether the kernel builds, but whether it runs 
correctly under all circumstances :-)

Here's some of the gcc changes, to give you all a feel for what changes 
they've made.

--- gcc-3.0/gcc/config/mips/mips.h      2001-06-14 16:42:18.000000000 
-0400
+++ WRT54G/tools-src/gnu-20010422/gcc/config/mips/mips.h        2003-10-10 
15:15:14.000000000 -0400
@@ -214,6 +214,7 @@
 #define MASK_UNINIT_CONST_IN_RODATA \
                           0x01000000   /* Store uninitialized
                                           consts in rodata */
+#define MASK_NO4710A0     0x02000000   /* WA_BCM4710A0: Don't work-around 
BCM4710A0 CPU bugs */
                                                                                                                                                             
                                        /* Debug switches, not documented 
*/
 #define MASK_DEBUG     0               /* unused */
@@ -310,6 +311,9 @@
 #define TARGET_NO_CHECK_ZERO_DIV (target_flags & MASK_NO_CHECK_ZERO_DIV)
 #define TARGET_CHECK_RANGE_DIV  (target_flags & MASK_CHECK_RANGE_DIV)
                                                                                                                                                             
+/* WA_BCM4710A0 */
+#define TARGET_4710A0          !(target_flags & MASK_NO4710A0)
+
 /* This is true if we must enable the assembly language file switching
    code.  */
                                                                                                                                                             
@@ -423,6 +427,12 @@
      N_("Work around early 4300 hardware bug")},                       \
   {"no-fix4300",         -MASK_4300_MUL_FIX,                           \
      N_("Don't work around early 4300 hardware bug")},                 \
+  {"4710a0",            -MASK_NO4710A0,                                \
+     N_("Work around BCM4710A0 hardware bugs")},                       \
+  {"no-4710a0",                  MASK_NO4710A0,                                
\
+     N_("Don't work around BCM4710A0 hardware bugs")},                 \
+  {"4710a0kern",         MASK_NO4710A0,                                \
+     N_("Don't work around BCM4710A0 hardware bugs")},                 \
   {"4650",               MASK_MAD | MASK_SINGLE_FLOAT,                 \
      N_("Optimize for 4650")},                                         \
   {"3900",               MASK_MIPS3900,                                \
@@ -590,7 +600,8 @@
                                 )
                                                                                                                                                             
 /* ISA has branch likely instructions (eg. mips2). */
-#define ISA_HAS_BRANCHLIKELY   (mips_isa != 1)
+/* WA_BCM4710A0 */
+#define ISA_HAS_BRANCHLIKELY   (mips_isa != 1 && !TARGET_4710A0)

 /* ISA has the conditional move instructions introduced in mips4. */
 #define ISA_HAS_CONDMOVE        (mips_isa == 4                         \
@@ -784,7 +795,7 @@
 /* GAS_ASM_SPEC is passed when using gas, rather than the MIPS
    assembler.  */
                                                                                                                                                             
-#define GAS_ASM_SPEC "%{mcpu=*} %{m4650} %{mmad:-m4650} %{m3900} %{v} 
%{mgp32} %{mgp64}"
+#define GAS_ASM_SPEC "%{mcpu=*} %{m4650} %{m4710a0} %{m4710a0kern} 
%{mmad:-m4650} %{m3900} %{v} %{mgp32} %{mgp64}"
                                                                                                                                                             
 /* TARGET_ASM_SPEC is used to select either MIPS_AS_ASM_SPEC or
                                                                                                                                                             
...

And so we all have some idea where they started from:

--- gcc-3.0/gcc/version.c       2001-06-17 15:44:25.000000000 -0400
+++ WRT54G/tools-src/gnu-20010422/gcc/version.c 2003-10-10 
15:15:11.000000000 -0400
@@ -1,4 +1,4 @@
 #include "gansidecl.h"
 #include "version.h"
                                                                                                                                                             
-const char *const version_string = "3.0";
+const char *const version_string = "3.0 20010422 (prerelease) with 
bcm4710a0 modifications";


--
Charlie

Re: Broadcom gcc changes (was Re: Broadcom 4702?)

[Steven J. Hill]

Charlie Brady wrote:
> 
> Here's some of the gcc changes, to give you all a feel for what changes 
> they've made.
> 
> --- gcc-3.0/gcc/config/mips/mips.h      2001-06-14 16:42:18.000000000 
> -0400
> +++ WRT54G/tools-src/gnu-20010422/gcc/config/mips/mips.h        2003-10-10 
> 15:15:14.000000000 -0400
> @@ -214,6 +214,7 @@

The change simply disables the compiler from emitting branch likely
instructions when compiler userspace code. Branch likely instructions
are allowed when compiling kernel code.

-Steve

Re: Broadcom gcc changes (was Re: Broadcom 4702?)

[John W. Linville]

Charlie Brady wrote:

>On Thu, 15 Jan 2004, John W. Linville wrote:
>  
>
>>I don't know about the 4710 or 4702 (I haven't got around to that yet), 
>>but the 4704 doesn't seem to need any special flags for building the 
>>kernel (or anything else).
>>    
>>
>
>The acid test is not whether the kernel builds, but whether it runs 
>correctly under all circumstances :-)
>
Of course.  But, "all circumstances" is difficult to test... :-)

The BCM4704 kernels that I've been building seem to be running find, so 
far.  Hopefully they will continue to do so... :-)

John

-- 
John W. Linville
LVL7 Systems, Inc.

Re: Broadcom 4702?

[Alan Cox]

On Iau, 2004-01-15 at 09:11, Dominic Sweetman wrote:
> So yes, binary code distributed in a black box is still distributed,
> and if it was GPL software you are entitled to the source code.  It's
> sensible of Cisco to put it quietly on a web site somewhere.

This is actually one reason some embedded people have concerns about the
GPL. It isnt providing the code, its how to handle the offer/provision
when the product in question is a tea maker or some appliance where it
isnt natural to ship say an accompanying CD of windows drivers.

Re: Broadcom 4702?

[ilya]

I don't really see any problem with that. I think providing it
on web site and listing URL somewhere in manual should be more then
enough.
IANAL, of course.

	Ilya.

On Thu, Jan 15, 2004 at 09:33:13PM +0000, Alan Cox wrote:
> This is actually one reason some embedded people have concerns about the
> GPL. It isnt providing the code, its how to handle the offer/provision
> when the product in question is a tea maker or some appliance where it
> isnt natural to ship say an accompanying CD of windows drivers.
> 
>