[Qemu-devel] Re: QEMU pcnet 6/28 patch against 6/30 CVS - not working?

[Qemu-devel] Re: QEMU pcnet 6/28 patch against 6/30 CVS - not working?

[Antony T Curtis]

[-- Attachment #1: Type: text/plain, Size: 855 bytes --]

On Thu, 2004-07-01 at 20:54, David Black wrote:
> Hello,
> 
> I applied your AMD pcnet NIC patch dated 6/28, to the QEMU CVS snapshot 
> from dad-answers.com dated 6/30.
> 
> The patch seems to apply well, with some manual patching needed on vl.c 
> and Makefile.target, but nothing that looked problematic.

The patch attached should apply more cleanly against current CVS.

I will try to see if it works any differently with running Linux. I
assume you are using the pcnet32 driver, not the lance driver. Looking
at the Linux driver source, it is able to use the DWORD I/O mode ... The
FreeBSD, WinXP and OS/2 drivers appear to use the WORD I/O mode. It is
possible that there is a bug in the simulation - particularly where the
Linux driver probes the card in both modes. I will have to test it out.

-- 
Antony T Curtis <antony.t.curtis@ntlworld.com>

[-- Attachment #2: qemu-pcnet.patch2.gz --]
[-- Type: application/x-gzip, Size: 15163 bytes --]

[Qemu-devel] Re: QEMU pcnet 6/28 patch against 6/30 CVS - not working?

[Antony T Curtis]

[-- Attachment #1: Type: text/plain, Size: 2105 bytes --]

Hi,

This new patch works for me with Mandrake 10 Official

It should apply cleanly against QEMU CVS and hopefully compile cleanly
on Linux as well as FreeBSD. I have also implemented the TX-ring poll so
in theory, it should be fully functional.

On Fri, 2004-07-02 at 20:15, David Black wrote:
> Thanks - yes, it's the 'pcnet' driver, and it comes up in 32 bit mode:
> 
> Jul  1 14:04:22 vs1a kernel: pcnet32.c:v1.27b 01.10.2002
> tsbogend@alpha.franken.de
> Jul  1 14:04:22 vs1a kernel: PCI: Found IRQ 9 for device 0000:00:03.0
> Jul  1 14:04:22 vs1a kernel: PCI: Setting latency timer of device
> 0000:00:03.0 to 64
> Jul  1 14:04:22 vs1a kernel: pcnet32: PCnet/PCI II 79C970A at 0xc020,
> warning: CSR address invalid,
> Jul  1 14:04:22 vs1a kernel:     using instead PROM address of 52 54
> 00 12 34 56 assigned IRQ 9.
> Jul  1 14:04:22 vs1a kernel: eth0: registered as PCnet/PCI II 79C970A
> Jul  1 14:04:22 vs1a kernel: pcnet32: 1 cards_found.
> 
> This driver is attractive to me because of the promise of decreased
> latency.  I get occasional data corruption with the ne2k-pci driver,
> too.
> 
> Dave
> 
> Antony T Curtis wrote: 
> > On Thu, 2004-07-01 at 20:54, David Black wrote:
> >   
> > > Hello,
> > > 
> > > I applied your AMD pcnet NIC patch dated 6/28, to the QEMU CVS snapshot 
> > > from dad-answers.com dated 6/30.
> > > 
> > > The patch seems to apply well, with some manual patching needed on vl.c 
> > > and Makefile.target, but nothing that looked problematic.
> > >     
> > The patch attached should apply more cleanly against current CVS.
> > 
> > I will try to see if it works any differently with running Linux. I
> > assume you are using the pcnet32 driver, not the lance driver. Looking
> > at the Linux driver source, it is able to use the DWORD I/O mode ... The
> > FreeBSD, WinXP and OS/2 drivers appear to use the WORD I/O mode. It is
> > possible that there is a bug in the simulation - particularly where the
> > Linux driver probes the card in both modes. I will have to test it out.
> > 
> >   
> FLAGS (\Seen))
-- 
Antony T Curtis <antony.t.curtis@ntlworld.com>

[-- Attachment #2: qemu-pcnet.patch3.gz --]
[-- Type: application/x-gzip, Size: 22464 bytes --]

[Qemu-devel] Re: QEMU pcnet 6/28 patch against 6/30 CVS - not working?

[Antony T Curtis]

[-- Attachment #1: Type: text/plain, Size: 772 bytes --]

Hi

I have updated the patch yet again... What this new patch provides:

Fixes a bug in MMIO: fixes NetBSD.
Less bugs - fewer structures to go wrong.
Fewer data-copying. But I don't think it is endian-safe, so may not work
on big-endian systems.

Hope more people could give this a go...

On Sun, 2004-07-04 at 14:05, Hetz Ben Hamo wrote:
> Antony T Curtis wrote:
> > Hi,
> > 
> > This new patch works for me with Mandrake 10 Official
> > 
> > It should apply cleanly against QEMU CVS and hopefully compile cleanly
> > on Linux as well as FreeBSD. I have also implemented the TX-ring poll so
> > in theory, it should be fully functional.
> 
> I updated your patch on my web site.
> 
> Thanks,
> Hetz
> 
>  FLAGS (\Seen))
-- 
Antony T Curtis <antony.t.curtis@ntlworld.com>

[-- Attachment #2: qemu-pcnet.patch4.gz --]
[-- Type: application/x-gzip, Size: 21614 bytes --]

[Qemu-devel] Re: QEMU pcnet 6/28 patch against 6/30 CVS - not working?

[Hetz Ben Hamo]

Antony T Curtis wrote:
> Hi,
> 
> This new patch works for me with Mandrake 10 Official
> 
> It should apply cleanly against QEMU CVS and hopefully compile cleanly
> on Linux as well as FreeBSD. I have also implemented the TX-ring poll so
> in theory, it should be fully functional.

I updated your patch on my web site.

Thanks,
Hetz

Re: [Qemu-devel] Re: QEMU pcnet 6/28 patch against 6/30 CVS - not working?

[Brad Watson]

Hi Anthony,

The patch installed and compiled cleanly on my
slackware based 9.x box. I experimented a little bit
with it and made the following "quick-n-dirty" cursory
comparison beteen the QEMU realtek-8029 and pcnet
"cards" via an scp of 10MB file between host & guest: 

openbsd x86 3.5:
  pcnet: 238kb/s "average."
  realtek-8029: 276kb/s "average."

freebsd x86 5.2.1:
  pcnet: 289kb/s "average."
  realtec-8029: 262kb/s "average"

netbsd x86 1.6.1:
  pcnet: no data (hangs). note: machid is getting set
now, ping works, but little else appears to....
  realtec-8029: 320kb/s "average" note: occasional
console message: ne2: remote transmit DMA failed to
complete

I'm interested in experimenting w/ adding an etherboot
rom to your pcnet code, and a tftp server into the
slirp code to pass the -kernel and -initrd parameter
arguments to a network boot client via tftp.  Can you
give me any quick pointers on hooking the rom into
your code ? 

--- Antony T Curtis <antony.t.curtis@ntlworld.com>
wrote:
> Hi,
> 
> This new patch works for me with Mandrake 10
> Official
> 
> It should apply cleanly against QEMU CVS and
> hopefully compile cleanly
> on Linux as well as FreeBSD. I have also implemented
> the TX-ring poll so
> in theory, it should be fully functional.
> 
> On Fri, 2004-07-02 at 20:15, David Black wrote:
> > Thanks - yes, it's the 'pcnet' driver, and it
> comes up in 32 bit mode:
> > 
> > Jul  1 14:04:22 vs1a kernel: pcnet32.c:v1.27b
> 01.10.2002
> > tsbogend@alpha.franken.de
> > Jul  1 14:04:22 vs1a kernel: PCI: Found IRQ 9 for
> device 0000:00:03.0
> > Jul  1 14:04:22 vs1a kernel: PCI: Setting latency
> timer of device
> > 0000:00:03.0 to 64
> > Jul  1 14:04:22 vs1a kernel: pcnet32: PCnet/PCI II
> 79C970A at 0xc020,
> > warning: CSR address invalid,
> > Jul  1 14:04:22 vs1a kernel:     using instead
> PROM address of 52 54
> > 00 12 34 56 assigned IRQ 9.
> > Jul  1 14:04:22 vs1a kernel: eth0: registered as
> PCnet/PCI II 79C970A
> > Jul  1 14:04:22 vs1a kernel: pcnet32: 1
> cards_found.
> > 
> > This driver is attractive to me because of the
> promise of decreased
> > latency.  I get occasional data corruption with
> the ne2k-pci driver,
> > too.
> > 
> > Dave
> > 
> > Antony T Curtis wrote: 
> > > On Thu, 2004-07-01 at 20:54, David Black wrote:
> > >   
> > > > Hello,
> > > > 
> > > > I applied your AMD pcnet NIC patch dated 6/28,
> to the QEMU CVS snapshot 
> > > > from dad-answers.com dated 6/30.
> > > > 
> > > > The patch seems to apply well, with some
> manual patching needed on vl.c 
> > > > and Makefile.target, but nothing that looked
> problematic.
> > > >     
> > > The patch attached should apply more cleanly
> against current CVS.
> > > 
> > > I will try to see if it works any differently
> with running Linux. I
> > > assume you are using the pcnet32 driver, not the
> lance driver. Looking
> > > at the Linux driver source, it is able to use
> the DWORD I/O mode ... The
> > > FreeBSD, WinXP and OS/2 drivers appear to use
> the WORD I/O mode. It is
> > > possible that there is a bug in the simulation -
> particularly where the
> > > Linux driver probes the card in both modes. I
> will have to test it out.
> > > 
> > >   
> > FLAGS (\Seen))
> -- 
> Antony T Curtis <antony.t.curtis@ntlworld.com>
> 

> ATTACHMENT part 2 application/x-gzip
name=qemu-pcnet.patch3.gz
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://lists.nongnu.org/mailman/listinfo/qemu-devel
> 


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com