Linux 2.6.28 Porting to Oxnas

[ftcodes]

I had only added that module ported by lucianm. I personally do not bother about NAND access from SATA only boot. So, I did not check it throroughly. But, I expected some clean up on that module.

Seems that the Ethernet module needs some checking, too. But that will be some hard work without detailed hardware reference manuals about registers and so on. However, it should be doable.

[WarheadsSE]

The ethernet, yeah.. especially with the coprocessor & tx/rx offloading with the gmac.

[ftcodes]

It will be a more thorough analysis than the SATA stuff due to that offloading stuff.

[xenoxaos]

One thing I've noticed is SATA speed on the 3.1 is a bit slower than the 2.6.31. .31 I was getting around 60MB/s read, now its mid 30's on the 3.1.

[ftcodes]

PLX did a lot of bypassing throughout the kernel which bypasses the SATA layer as such. This code was not portable to the 3.1 kernel at all as far as I have followed it. That stuff required a function pointer entry which seems like not existing anymore. And, in addition a lot of filesystem hacks. In my oppinion, it would be better to track the bottlenecks in the SATA framework instead of hacking together a major bypass code like PLX did. Maybe the SATA driver itself has some bottlenecks that may not have been solved.

In fact, the SATA driver has got some busy waiting. Those things can result into performance drops.

[ftcodes]

$this->bbcode_second_pass_quote('xenoxaos', 'O')ne thing I've noticed is SATA speed on the 3.1 is a bit slower than the 2.6.31. .31 I was getting around 60MB/s read, now its mid 30's on the 3.1.

[WarheadsSE]

I have registered mach-type arch numbers:
- 3973 - Pogoplug V3
- 3974 - Medion Life P89626
- 3975 - iomega HMND CE
- 3976 - Pogoplug V3 PCI - Pro/Video
I pointed the Pogoplug here, the Medion to mikrocontroller.net's page on it, and the HMND CE to iomegas page for now.

[ftcodes]

I have not yet looked where those arch numbers go to.

But, I did a short check on the DMA pathes inside SATA driver and found that it takes to different pathes. It looks almost like it is using a different path to DMA controller than the direct sata stuff does. If that is so, it should actually be possible to port that stuff into the regular path. But, I have yet to check where the current pathes are ending.

[WarheadsSE]

They are used to differentiate between boards at compile & runtime, as passed via the bootloader. Thus, we can have 1 LED module that runs all 3 known boards, button code where appropriate, as well.

See mach-kirkwood directory in arch/arm, leming is taking a stab at it now.

[elazar]

Hello,
Been lurking for some time, some great progress being made here [beer goes here ]. That said, I built a 3.1.8 kernel off of WarheadsSE git repo for my PogoPlug Pro, and I noticed that the gmac driver does not work(interface comes up, no traffic in or out), after some lost sleep and hair I figured out what was wrong. Overall, the gmac driver needs some overhaul; the issue was cause by the interface speed never being set, gmac barely utilizes ethtool_cmd. Patch below:

$this->bbcode_second_pass_code('', '
--- ox820-linux-3.1/drivers/net/gmac/gmac-napi.c 2012-01-10 22:21:45.000000000 +0000
+++ linux-3.1.8/drivers/net/gmac/gmac-napi.c 2012-01-16 19:29:30.000000000 +0000
@@ -848,15 +848,25 @@ static void watchdog_timer_action(unsign
int duplex_changed;
int speed_changed;
int pause_changed;
+ struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };

// Interpret the PHY/link state.
if (priv->phy_force_negotiation || (priv->watchdog_timer_state == WDS_RESETTING)) {
mii_check_link(&priv->mii);
ready = 1;
} else {
- /*duplex_changed = mii_check_media_ex(&priv->mii, 1,
- priv->mii_init_media, &speed_changed, &pause_changed,
- link_state_change_callback, priv);*/
+ duplex_changed = mii_check_media(&priv->mii, 1, 1);
+ mii_ethtool_gset(&priv->mii, &ecmd);
+ if (ethtool_cmd_speed(&(priv->ethtool_cmd)) != ecmd.speed) {
+ ethtool_cmd_speed_set(&(priv->ethtool_cmd), ecmd.speed);
+ speed_changed = 1;
+ }
+ if ((duplex_changed || speed_changed) || (duplex_changed && speed_changed)) {
+ link_state_change_callback(1, priv);
+ } else {
+ link_state_change_callback(0, priv);
+ }
+
priv->mii_init_media = 0;
ready = netif_carrier_ok(priv->netdev);
}
')

Thats it for now...

elazar