uaenet.device problem with amitcp 2.2

[Leandro Jardim]

I am using uaenet.device and AmiTCP 2.2 and at startup the message is shown:
S2_CONFIG_INTERFACE: Inappropriate State (Attempt to config twice).

And Internet doesnt works.

Previously (with A2065 Z2) the message isnt shown.

It happens with the config generated by AmiTCP.

[Toni Wilen]

Create log with -netlog command line parameter.

[Leandro Jardim]

Done. Now Internet worked even message being shown...

[Toni Wilen]

Hmm.. Following messages are more serious (something confuses copper emulation), could you try to find out what causes them to appear?

error2: vpos=0 until_hpos=219
cvcmp=12 chcmp=0 chpos=162 cvpos=0 ci1=0C01 ci2=FFFE
cstate=7 ip=4bc SPCFLAGS=4

[Leandro Jardim]

Well, it happens everytime WinUAE is transmitting something with uaenet.device or bsdsocket.library and CPU is in 68EC020 mode, AGA chipset. I usually use a interlaced screen 640x512 16/64/256 color. The screen blinks many times and the problem is more visible when I run Copper-Demon. I didnt tested other modes. But I tested with other CPUs and no problem was seen with them. And the interesting is that it never happens with the Z2 CBM card emulation.

I browse the Internet in an AGA 16 color screen mode propositionally to stress a bit WinUAE. I love to see the emulation working and I loved doing this on my Amiga.

[Toni Wilen]

I can't duplicate. I used your config (only replaced harddrive) and no strange logging, even if I used miami + a2065 or uaenet.

Could you create minimal harddrive that still shows the problem? As minimal as possible to reduce the number of unknown variables? (But do not upload it, it most likely includes programs that shouldn't be uploaded)

[Leandro Jardim]

Originally Posted by Toni Wilen

I can't duplicate. I used your config (only replaced harddrive) and no strange logging, even if I used miami + a2065 or uaenet.

Could you create minimal harddrive that still shows the problem? As minimal as possible to reduce the number of unknown variables? (But do not upload it, it most likely includes programs that shouldn't be uploaded)

Interesting, because I have this problem not only with the computer I have actually, but with another completely different computer I used before... The CPU brand, the video card, all different... And with different hardfiles... This problem is very, very old.

Then I will do this: I will comment suspicious lines in my startup-sequence and remove all programs in my WBStartup until I find the problem.

One suspicious program I have installed is SystemPatch 2.9.

[Leandro Jardim]

As I said the problem is very old. It happens even before you put the Cycle-Exact feature into WinUAE.

What I want to say is that a similar problem appeared in the 0.8.x versions of WinUAE. I reported it to Brian King at that time but I stopped reporting the error because I had lost my Internet connection. I am unsure of what happened to WinUAE since then. But the side effects was the same. The only difference is that at that time instead of writing a log WinUAE showed a dialog with the copper registers.

But at that time Brian King didnt know about the reason of the bug. Maybe you should talk a bit with Brian King, to talk about this bug, maybe him remembers it well and could say to you something about it.

[Leandro Jardim]

Well guys... it appears that the problem is caused by FBlit, any configuration but I use its default. With it AmigaOS crashes near instantly when in AWeb. But SystemPatch seems to fix the bug a bit, what happens is that the screen blinks and flashes very much, but interesting is that AmigaOS gets usable. Dont forget to use a Copper backdrop, because the screen will blink/flash even more.

I use an AGA interlaced screen, with 16 colours. I installed SystemPatch 2.99, ClassAct 3.3, FBlit 3.79a, CopperDemon 3.2, and AWEB 3.5.09. With BSDSocket.library only.

I submitted a new configuration.

[Toni Wilen]

Originally Posted by Leandro Jardim

Well guys... it appears that the problem is caused by FBlit, any configuration but I use its default. With it AmigaOS crashes near instantly when in AWeb. But SystemPatch seems to fix the bug a bit, what happens is that the screen blinks and flashes very much, but interesting is that AmigaOS gets usable. Dont forget to use a Copper backdrop, because the screen will blink/flash even more.

I use an AGA interlaced screen, with 16 colours. I installed SystemPatch 2.99, ClassAct 3.3, FBlit 3.79a, CopperDemon 3.2, and AWEB 3.5.09. With BSDSocket.library only.

I submitted a new configuration.

If you run with -log command line parameter, when does above log messages appear? Immediately you run AWeb? Does it always happen when bsdsocket.library is opened? (Opened by any program, does even simple ping cause it?)

I assume it also happens without JIT?

[Leandro Jardim]

It happens only if I disable JIT. Thats because the bug needs the 24 bits addressing of 680EC20 to appear. Tested. I use a 68EC020 cycle-exact configuration with 4 MB and AGA (based on the A1200 quickstart configuration) always. I formatted a new hard-file and installed everything, from scratch.

The message appears on a ping or when AWeb is downloading a HTML page. For example the home page of Aminet or even better, EAB (because its much bigger and has more pictures).

If I use A2065 this is the message that I see (but the screen does not flash/blink):

38-869 [4304 002-000]: A2065: RECEIVE BUFFER ERROR

I am thinking that probably the problem may not be related directly to FBlit, etc., but only to the 24 bit addressing of the registers of the 68EC020.

[Toni Wilen]

Does it still happen (or does something change?) if you enable RTG in Zorro2 mode (set for example 1M VRAM size).

There is no need to actually use RTG in AOS, only need to enable it in emulation configuration.

EDIT: Must use latest 2.4.2 beta! (RTG enabled or 68020+ = different memory allocation)

[Leandro Jardim]

I downloaded the last beta and tested it...

Without RTG, I noticed the screen doesnt blinked/flashed anymore, but I got 3 recoverable alerts. But it dont crashed, before it crashed sometimes.

With RTG, the screen blinked/flashed 4 times and it didnt crashed. There was almost no difference in the logs.

Perhaps I would need to test it some time more, mainly because of the crashes, but my Internet connection is horrible.

[Toni Wilen]

Problem found (and it was ancient), hopefully this is fully fixed in next version.
(Thanks to OP for sending perfectly working test case)

It was caused by new host/m68k switching calltrap code which needed to store (and later restore) CPU state (=registers etc..) but accidentally it also stored variables that are used by chipset emulation (including flag that tells if next scanline ends copper wait)

It seems only some bsdsocket calls can trigger this problem (uses trap that does emulated m68k code -> native C code -> native C code calls AOS library function -> m68k code -> RTS from emulated m68k code -> back to native -> back to original call)

Problem was triggered only if chipset state changed during middle m68k code which is very rare unless CPU emulation is slow (non-JIT) and chipset was busy (for example copper background)

At least it was not completely my fault because this code was implemented years ago by (disappeared) E-UAE author (It is just a tiny mistake, new trap code was great improvement, old one was really horrible and non-portable)

[Leandro Jardim]

Wow, so its should be a very complex problem and you fixed it... Wow... Congratulations!

[cosmicfrog]

I like happy endings