[r8169] patch for RTL8102 (5 new MAC/PHY)

[r8169] patch for RTL8102 (5 new MAC/PHY)

[Леонид Юрьев]

Hi,

INTRO:
Few days ago I need to run Linux on Intel's D945GCLF board with
RTL8102. Driver from 2.6.26 oops on boot, but 2.6.27-rc2 seems
working. Unfortunately really it fails too. Not even, but in some
cases with influence of BIOS setting...

Late I found obsolete Realtek's driver
(http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&ProductID=2916&DwnldID=16242&strOSs=39&OSFullName=Linux*&lang=eng).
It is a poor (IMHO) driver based on 2.6.18-version of r8169.c

Actual difference from the current r8169.c is in detection and "magic"
handle of the 5 new versions of RTL's MACs and corresponded PHY.

On Sunday I just pull some "magic" lines from Realtek to r8169, now my
D945GCLF board work fine in any cases :)

Not all "magic" from Realtek was pulled, few of them is really mad
(e.g. watchdog for loosing a PCI-config).

SUBJ:
If all agree, I will prepare for review a set of patches:
1) cosmetic: remark about supported chips (8168/8110/8111/8101/8102)
into KConfig and driver. Elsewhere it is not easy to find required
driver thru "make *config";
2) simple: by RTL8169 datasheet 'rx-missed' counter is 24-bits width,
but not 32;
3) simple: account dev->stats.rx_over_errors on RxFIFOOver and RxOverflow;
4) workaround: RTL8102 seems (sure) not have "rx-missed" (PCI-reg
0x4C) counter. So we should not account dev->stats.rx_missed_errors
like RTL8169;
5) big: a "magic" from Realtek's driver for RTL8102 (no any datasheet
are accessible/available);

Re: [r8169] patch for RTL8102 (5 new MAC/PHY)

[Ivan Vecera]

> Actual difference from the current r8169.c is in detection and "magic"
> handle of the 5 new versions of RTL's MACs and corresponded PHY.
The support for RTL8102 was added 2 days ago. See Francois's git tree
in r8169-fixes branch.
http://git.kernel.org/?p=linux/kernel/git/romieu/netdev-2.6.git;a=shortlog;h=r8169-fixes

> 2) simple: by RTL8169 datasheet 'rx-missed' counter is 24-bits width,
> but not 32;
interesting :-)

> 4) workaround: RTL8102 seems (sure) not have "rx-missed" (PCI-reg
> 0x4C) counter. So we should not account dev->stats.rx_missed_errors
> like RTL8169;
It seems (according Realtek's r8101 and r8168 drivers) that not only
RTL810x class but also RTL8168 class hardware don't have rx-missed. r8168
doesn't contain anything regarding RxMissed and r8101's RxMissed stuff
is commented out.

Ivan

Re: [r8169] patch for RTL8102 (5 new MAC/PHY)

[Леонид Юрьев]

2008/8/19 Ivan Vecera <ivecera@redhat.com>:
> The support for RTL8102 was added 2 days ago. See Francois's git tree
> in r8169-fixes branch.
> http://git.kernel.org/?p=linux/kernel/git/romieu/netdev-2.6.git;a=shortlog;h=r8169-fixes
Oh, yes :)

>> 2) simple: by RTL8169 datasheet 'rx-missed' counter is 24-bits width,
>> but not 32;
> interesting :-)

http://wiki.osdev.org/RTL8169, seek bottom of page for the link to
http://www.magnesium.net/~wpaul/rt/RTL8110S_8169S_DataSheet_1.3.pdf,
Datasheet for the RTL8169S and RTL8110S chpsets.

So, on page #15, can see:

004Ch-004Fh R/W MPC
Missed Packet Counter: This 24-bit counter indicates the number of
packets discarded due to Rx FIFO overflow. After a s/w reset, MPC
is cleared. Only the lower 3 bytes are valid.
When any value is written to MPC, it will be reset.

--
Leo Yuriev