From: Geoff Steckel <gwes@oat.com>
Subject: Re: 7.7->7.8 2 regressions: rge0 & amd embedded com0
To: Kevin Lo <kevlo@kevlo.org>
Cc: tech <tech@openbsd.org>
Date: Sun, 2 Nov 2025 21:37:00 -0500

On 11/2/25 9:11 PM, Kevin Lo wrote:
> On Sat, Nov 01, 2025 at 02:30:16AM -0400, Geoff Steckel wrote:
>> Just upgraded 7.7 -> 7.8
>> Noticed 2 regressions on devices that worked immediately before the upgrade.
>>
>> com0 embedded uart worked -almost- ok in 7.7
>>
>> Oct 31 17:58:30 eight /bsd: com0: 3 silo overflows, 0 ibuf overflows
>>
>> now
>>
>> Nov 1 02:00:59 eight /bsd: com0: 51 silo overflows, 0 ibuf overflows
>>
>> The embedded uart is a little tricky
>> The port is connected to a GPS receiver @9600 baud sending 2 lines once a
>> second.
>>
>> rge0 worked perfectly in 7.7 (and was used to do sysupgrade)
>>
>> Oct 9 13:25:26 eight /bsd: rge0 at pci3 dev 0 function 0 "Realtek RTL8125"
>> rev
>> 0x00: msix, address 78:2d:7e:12:5a:d9
>>
>> now
>>
>> Nov 1 01:43:45 eight /bsd: rge0 at pci3 dev 0 function 0 "Realtek RTL8125"
>> rev
>> 0x00: msixrge0: failed to stop all requests
>>
>> If anyone suggests a patch or a test I'll gladly try it.
>> thanks
>> Geoff Steckel
> Hi Geoff,
>
> First of all, I sincerely apologize for breaking the RTL8125 and causing
> your network card to stop working.
> I probably tested using an RTL8125B card while assuming it was an RTL8125,
> which is why I didn't encounter this issue.  In other words, I didn't
> actually test on an RTL8125 :(
>
> In revision 1.38 of if_rge.c, I added a check for the RTL8125 to verify
> that the chip has finished the request.
> Like the RTL8125B and RTL8125D, it doesn't require an additional check.
>
> The vendor driver for Linux works fine because it doesn't check the return
> value of rtl8125_stop_all_request().  If it did, it would also fail.
>
> The patch below fixes the issue I mentioned.
>
> Tested:
> rge0 at pci4 dev 0 function 0 "Realtek RTL8125" rev 0x00: msix, address 00:13:3b:xx:xx:xx
> rge1 at pci6 dev 0 function 0 "Realtek RTL8125" rev 0x04: msix: RTL8125B, address 7c:b5:9b:xx:xx:xx
> rge2 at pci7 dev 0 function 0 "Realtek RTL8125" rev 0x0c: msix: RTL8125D, address 34:5a:60:xx:xx:xx
>
> Index: sys/dev/pci/if_rge.c
> ===================================================================
> RCS file: /cvs/src/sys/dev/pci/if_rge.c,v
> diff -u -p -u -p -r1.39 if_rge.c
> --- sys/dev/pci/if_rge.c	27 Oct 2025 03:18:36 -0000	1.39
> +++ sys/dev/pci/if_rge.c	3 Nov 2025 01:06:55 -0000
> @@ -1528,19 +1528,7 @@ rge_reset(struct rge_softc *sc)
>   	RGE_SETBIT_1(sc, RGE_PPSW, 0x08);
>   
>   	RGE_SETBIT_1(sc, RGE_CMD, RGE_CMD_STOPREQ);
> -	if (sc->rge_type == MAC_R25) {
> -		for (i = 0; i < 20; i++) {
> -			DELAY(10);
> -			if (!(RGE_READ_1(sc, RGE_CMD) & RGE_CMD_STOPREQ))
> -				break;
> -		}
> -		if (i == 20) {
> -			printf("%s: failed to stop all requests\n",
> -			    sc->sc_dev.dv_xname);
> -			return ETIMEDOUT;
> -		}
> -	} else
> -		DELAY(200);
> +	DELAY(200);
>   
>   	for (i = 0; i < 3000; i++) {
>   		DELAY(50);
>
Hi Kevin,
   Thanks very much for looking into this. Realtek doesn't make it easy with
their semi-random chip numbering!
I'll put the patch in early tomorrow. I expect no problems!
   Geoff Steckel