CFT: vr(4)

[Pyun YongHyeon]

On Mon, Mar 03, 2008 at 10:10:25AM +0100, Milan Obuch wrote:
 > On Monday 03 March 2008, Pyun YongHyeon wrote:
 > > On Sat, Mar 01, 2008 at 11:53:41AM -0500, Mike Tancsa wrote:
 > >
 > > Sorry for late handling. I wanted to solve Milan Obuch's issue first
 > > before committing vr(4). But it seems that it's not easy to fix
 > > Milan's issue. :-(
 > >
 > 
 > Well, I see some progress there... Today I was able to do some tests again, 
 > and I was able to ping -f another box on the same network for some time. I 
 > tried then csup sources and I got hard hang, again, this time with following 
 > lines on console:
 > 
 > vr0: PCI bus error -- resetting
 > vr0: restarting
 > 

Hmm, this is interesting. 6105M datasheet said nothing what can be
done for this case. I guess this kind of error can come from
improperly seated NICs or broken hardware. Would you re-seat the NIC
or change PCI slot and try again with attached patch?

 > And no ability to enter kdb, either.
 > Just for record, I am getting following when kldload'ing if_vr:
 > 
 > vr0: <VIA VT6105M Rhine III 10/100BaseTX> port 0x9c00-0x9cff mem 
 > 0xfceff000-0xfceff0ff irq 18 at device 8.0 on pci3
 > vr0: Quirks: 0x6
 > vr0: Revision: 0x96
 > miibus1: <MII bus> on vr0
 > ukphy0: <Generic IEEE 802.3u media interface> PHY 1 on miibus1
 > ukphy0:  10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto
 > 
 > (one out of four devices shown)
 > 
 > >  > At 07:30 PM 2/27/2008, Pyun YongHyeon wrote:
 > >  > >I never thought this kind of testing. It's good to hear vr(4)
 > >  > >recovers from the abrupt link change events. I guess this also
 > >  > >indicates the overhauled vr(4) can close lots of PR for vr(4).
 > >  >
 > >  > BTW, any chance of these fixes being backported to RELENG_7 and
 > >  > RELENG_6 ? Its not just media speed changes that causes the nic to
 > >
 > > I'm sure I'll MFC the change to RELENG_7 but not sure it could be
 > > done on RELENG_6 due to lack of spare time.
 > >
 > 
 > In my eyes, if new vr works for others and no regression was found, it should 
 > go in. I did not encountered a regression - it did not work with old driver, 
 > it does not work (yet) with the new... but I hope we can get this one 
 > working, too...
 > 

Yes, I really like to fix it too.

 > Milan

-- 
Regards,
Pyun YongHyeon
-------------- next part --------------
--- if_vr.c.orig	2008-02-18 10:46:10.000000000 +0900
+++ if_vr.c	2008-03-03 19:27:15.000000000 +0900
@@ -1733,8 +1733,10 @@
 		CSR_WRITE_2(sc, VR_ISR, status);
 		if ((status & (VR_ISR_BUSERR | VR_ISR_LINKSTAT2 |
 		    VR_ISR_STATSOFLOW)) != 0) {
-			if (vr_error(sc, status) != 0)
-				goto done_locked;
+			if (vr_error(sc, status) != 0) {
+				VR_UNLOCK(sc);
+				return;
+			}
 		}
 		vr_rxeof(sc);
 		if ((status & (VR_ISR_RX_NOBUF | VR_ISR_RX_OFLOW)) != 0) {
@@ -1762,12 +1764,18 @@
 static int
 vr_error(struct vr_softc *sc, uint16_t status)
 {
+	uint16_t pcis;
 
 	status &= VR_ISR_BUSERR | VR_ISR_LINKSTAT2 | VR_ISR_STATSOFLOW;
 	if ((status & VR_ISR_BUSERR) != 0) {
 		status &= ~VR_ISR_BUSERR;
 		sc->vr_stat.bus_errors++;
-		device_printf(sc->vr_dev, "PCI bus error -- resetting\n");
+		/* Disable further interrupts. */
+		CSR_WRITE_2(sc, VR_IMR, 0);
+		pcis = pci_read_config(sc->vr_dev, PCIR_STATUS, 2);
+		device_printf(sc->vr_dev, "PCI bus error(0x%04x) -- "
+		    "resetting\n", pcis);
+		pci_write_config(sc->vr_dev, PCIR_STATUS, pcis, 2);
 		sc->vr_flags |= VR_F_RESTART;
 		return (EAGAIN);
 	}