28/11 : avec le code ARM7 de Tobias et le code ARM9 de l'exemple httpget.c, je parviens à avoir une transmission WiFi fonctionnelle. C'est donc (contre toute attente) le code ARM9 qui foire.
29/11 : la présence d'IrqInit() dans le code de GuiEngine::prepare() était clairement une des causes d'erreur, mais je ne sais pas encore dire si c'était la seule. Je vais devoir continuer à importer petit à petit les fonctionnalités de runme sur cette nouvelle base pour mettre le doigt sur les éléments perturbateurs.
Seven's log, stardate 201011.27 -- I am trapped in my ARM7 cortical implants. Although I could potentially still communicate with the WiFi collective -- if any was at short range -- I have lost connection with the ARM9 behaviour controller that would allow me to communicate with the crew of the Voyager. I have analysed the defect starfleet fifosystem.c that the Doctor installed to replace my damaged borg implants that is the root cause of my malfunction. The only tool I have to operate is a standard DESMuMe tricorder from the federation filled with irrelevant features, and that reaches only 67.3% of the efficience of a corresponding Borg nanoprobe.
Seven's log, supplemental -- A double initialisation of my interrupt mask might be the cause of this FIFO dysfunction. Deeper inspection of the starfleet FIFO protocol revealed a rigid structure with 16 virtual channels and imperfect typing of messages into one-word addresses, one-word value and external larger dataquads. I will resume my investigation when my refresh cycle is complete.
PS: curieux qu'avec le couple de processeurs ARM7 et ARM9, personne (que je connaisse) n'ai fait le rapprochement entre la DS de nintendo et le borg converti "Seven of Nine" de Star Trek : Voyager, vous ne trouvez pas ?
Note: j'aimerais bien également en savoir plus sur le support des fonctions FIFO dans desmume ... je note par exemple dans la fonction
_MMU_ARM9_write32
case REG_IPCFIFOSEND : IPC_FIFOsend(ARMCPU_ARM9, val); return;
Mais aussi (dans le même fichier FIFO.cpp) des fonctions GFX_FIFOsend, apparemment utilisées par l'implémentation des fonctions 3d -- et qui n'a donc probablement rien à voir dans le cas qui m'intéresse.
PS: to the devkitpro and desmume teams : no offense intended. That's Seven's "personality". Watch season 4 of Star Trek Voyager if you haven't yet: you'll know what I mean.
PPS: here's a patch against svn revision 3601 of desmume that enhance debugging of FIFO by
1) enabling per-direction logging of writes to FIFO registers through the DESMUME_FIFOLOG environment variable
2) uses location 0x04042000 as REG_DEBUG : everytime you write some 32-bit word there, it will be dumped -- together with processor no and PC value -- on STDERR.
I don't think I have "fixed" anything, so it looks like r3601 of desmume-cli already supported the full FIFO features (even though generation of interrupts are a bit obscured and could use some more documentation), but it could certainly help people understanding or validating the fifosystem.c of libnds, or other ARM7 initialisation code.
In its latest SVN version, desmume has FIFO support, and FIFO interrupt generation is handled by a gen_IF() function in MMU.cpp ... I got it working in _some_ situations.
ReplyDeleteWifi initialisation proceed with exchanging of a table. I see table &0x20518B8, WifiData=0x28518B8 (afaik, the same location, but uncached) and 0x480518B8 exchanged on the FIFO (likely channel = 4, type=8, value = 200000+518b8)
What is likely to work poorly with desmume is the fact that you install fifo handlers *after* you activated FIFO interrupts as a whole...