#tyndur · 2010-07-20

330 Nachrichten · Seite 1 von 2 · Original-CSV herunterladen

  1. pcmattmanHmm, I'm seriously wondering now how this e1000 driver even began to work
  2. kwolfHeh, is it that bad?
  3. pcmattmanDepends
  4. kwolfI guess it is what happens when you write for emulators
  5. pcmattmanDo you judge quality by "WTFs/minute"?
  6. kwolflol
  7. kwolfWe all know that WTFs/minute is the only valid metric ;-)
  8. pcmattmanDid I write this code, or was it contributed from someone else? :P
  9. kwolfWell, there's git blame, but I'd rather not use it... ;-)
  10. pcmattmanHeh, the 8254xx spec explains the TCP/IP stack concept, as well as packet formats
  11. pcmattmanApparently programming the hardware isn't important
  12. kwolfThe hardware only makes trouble, so better leave it alone
  13. kwolfAh, right, that was the spec where you had to scroll down almost to the end to get to the interesting information
  14. eddyblol
  15. pcmattmanHmm... I do believe this driver (on Pedigree) actually maps in the MMIO region... to a virtual address
  16. pcmattmanHello, XanClic :)
  17. kwolfIt probably should, otherwise you couldn't access it, right?
  18. XanClicMoin
  19. eddybpcmattman: you could ask him of our USB problems.... but I'm not doing it :P
  20. eddyb(anymore)
  21. pcmattmankwolf: I mean, the resource list may actually be virtual addresses... so when it does cdi_mem_map, it tries to map in a virtual address instead of a physical one
  22. pcmattmanXanClic: if you have a moment, we wouldn't mind another set of eyes on a problem: http://stackoverflow.com/questions/3285363/usb-ehci-help-with-transaction-errors
  23. pcmattmankwolf: but it's very easy to find out if I'm right or not, and if I am right... well, now I know why pcnet fails on Pedigree as well
  24. kwolfpcmattman: Oh, this way round. Not good.
  25. pcmattmanv -> v mappings aren't valid :)
  26. kwolfThey generally produce funny effects
  27. eddybthey have a name...
  28. pcmattmanNot necessarily faults
  29. kwolfDepends on what data you hit
  30. eddybtoo bad I can't remember it :P. I think something like "forward mapping" or "direct mapping"...
  31. pcmattmanIt'd be nice if they did fault, but usually they just "somehow" work... especially if you're writing to reserved memory which flips back to zero after you write to it
  32. pcmattmanMeans stuff like resets appear to work even
  33. eddyblol
  34. kwolfAnyway, it's time for lunch.
  35. kwolfI'll be back later
  36. XanClicBreakfast, you mean
  37. XanCliceddyb: I'd be glad if my own EHCI driver would even disable the EHCI, so I can't help you (because there's nothing obviously wrong) :-/
  38. kwolfXanClic: It's not weekend ;-)
  39. XanClicMaybe you should try disabling all ports before enumerating them
  40. eddyblol, was thinking on that... took a look at your driver, nothing special you do there :|
  41. XanClickwolf: Well, holidays for me, that's nearly the same ;-)
  42. eddybdisabling?
  43. eddybwhat good would do that?
  44. XanClicMaybe all ports are reset and activated before you take control, I don't know
  45. pcmattmanSimple - no interference from other ports
  46. eddybthey aren't enabled when the driver takes control
  47. XanClicIt can't hurt, too *g*
  48. eddybfor matt, not even powered...
  49. pcmattmanIt'd be a perfect explanation... except that on the beagleboard there's only one port, and my Stack Overflow post is mainly the outcome from that test
  50. XanClicNo, it wouldn't be
  51. XanClicThe SETUP packet would receive more than one ACK
  52. XanClicI don't know if the EHC wouldn't consider that an error
  53. pcmattmankwolf: turns out it is a physical address... so more debugging for me :)
  54. pcmattmanXanClic: out of interest, does your EHCI driver work on real hardware?
  55. XanClic[13:26:10] <XanClic> eddyb: I'd be glad if my own EHCI driver would even disable the EHCI, so I can't help you (because there's nothing obviously wrong) :-/
  56. XanClicNo ;-)
  57. pcmattmanJust clarifying :)
  58. pcmattmanI think I understand why everyone says USB is so hard now... you get it working in an emulator, and that's hard enough
  59. pcmattmanAnd then you find out that's the first 10% of the job :(
  60. eddybat some point it'd be easier to hack into intel's database, take the silicon layer of ICH7 and understand it than trying to find out how to do it...
  61. pcmattmanAt least I know we're not alone now :)
  62. XanClicWell, PrettyOS has working EHCI support (afaik): http://prettyos.svn.sourceforge.net/viewvc/prettyos/trunk/Source/kernel/
  63. pcmattmanAnd as for the e1000 driver... I'm gonna look at doing a rewrite kind of thing, see how that goes
  64. XanClicThough it's not a very nice interface…
  65. eddyb"working" as in "in qemu"
  66. XanClicNope
  67. pcmattmanHmm, so that's where PrettyOS code is
  68. XanClicThey say it works on real hardware, too
  69. pcmattmanI read the forum discussions for implementing EHCI into PrettyOS... at least, the first 4 pages
  70. pcmattmanYou've been working on it too, I believe?
  71. eddybthey lie or their realHW is one o f the easy ones...
  72. XanClicNo, I helped them a bit about USB in general
  73. pcmattmaneddyb: download it and try it out, prove that it works or doesn't
  74. pcmattmanXanClic: ah, makes sense
  75. eddybhow do I prove it doesn't work :P ?
  76. pcmattmanBy trying the OS out with USB devices :)
  77. pcmattmanI saw a USB/CDI patch come in the other day for tyndur, but I think that was just LS/FS stuff
  78. pcmattmanNot ehci
  79. pcmattmanAnd holy crap, this is fugly code
  80. eddybnoticed that :P
  81. XanClicWell, I guess I'm the only one writing USB CDI patches for týndur, so is it my code which is that ugly? ^^
  82. pcmattmanI was talking about PrettyOS :P
  83. XanClicAh :-)
  84. pcmattmanAlso, I just connected "Max" with you, heh, oops :)
  85. eddyb******* textColor
  86. eddyboh, and they copied more than half of the specs into the comments
  87. pcmattmanI don't complain about the comments
  88. pcmattmanEither way, he's got it working on real hardware... and we don't
  89. eddybI doubt that very much
  90. eddybwhat does it have...?
  91. eddyba 250ms reset time
  92. eddyb?
  93. pcmattmanI'm still trying to figure it out
  94. pcmattmanAlso, ehenkes is on the forums, we might get lucky and he might respond
  95. pcmattmanXanClic can benefit too :)
  96. eddybwhat forum?
  97. pcmattmanosdev.org
  98. XanClicWell, I once asked how to disable the EHCI and nobody answered ;-)
  99. pcmattmanI honestly think maybe 3 or 4 users on the forum actually know these answers
  100. pcmattmanAnd they rarely come around
  101. eddybXanClic: wait, do you want to know how to forward every port to companion host controllers?
  102. XanClicYes
  103. XanClicI know theoretically how to do that…
  104. eddyband I know practically
  105. XanClicSounds nice :-)
  106. eddybtoo bad my UHCI driver is as working on realHW as the EHCI one :(
  107. XanClicWell, afair it did somehow work…
  108. XanClicThe disabling
  109. eddybI get same behavior with forwarding all ports as with disabling EHCI from BIOS
  110. XanClicBut all transactions failed
  111. pcmattmaneddyb: the only thing he does differently is chain the qTDs
  112. eddybwhich is "FAIL", btw
  113. eddybthat's funny
  114. pcmattmanSo the SETUP qTD's pNext == the handshake IN's qTD
  115. eddybI wonder what difference would that make...
  116. eddybbtw, don't try anything yet, I have to figure how to get a kernel and an initrd for this thing...
  117. pcmattmanI think that may be the answer, to be honest
  118. eddybah, bash script build system
  119. XanClic“But all my high speed devices obviously refuse to work when they're connected to a root hub. If I disable the legacy support on my computer, I get a CRC/time out error together with a stall (maybe STALL or NAK in response to my SETUP packet), if I enable it, I had the same problem at the beginning but now there are page faults (though I don't know the exact reason). So, in general: My high speed devices work behind a full speed hub. Fine. But
  120. XanClicthey refuse to work on a root port.”
  121. XanClicThat was my problem ;-)
  122. kwolfBack
  123. XanClicwb
  124. eddybdoes this ehkens guy know how chmod +x works?
  125. XanClicNo, he uses Windows
  126. eddyboh, how nice, he claims to have an OS in an advanced state but he has a binary kernel...
  127. eddybI'll have to use some floppy emulators for usb flash drives...
  128. eddybhmm, I don't suppose I can chainload a floppy file from grub, can I?
  129. kwolfPrettyOS is a "GUI without memory management" type OS, just with s/GUI/USB/
  130. kwolfBut GUI is the next thing they started, so...
  131. eddybbbs, food
  132. eddybI'll have a look at this "OS" when I come back
  133. kwolfAnyway, ehenkes is frequently on this IRC server, so I think we may /invite him as soon as he arrives
  134. eddybwarn me to change nick
  135. pcmattmanI must say, his code certainly makes a lot of sense
  136. eddybyou ran it to an AI parser :P ?
  137. eddyb*thru
  138. pcmattmanNo, I traced the code path
  139. eddybit's hard
  140. pcmattmanTaking what I know from the spec and applying it to what I see in his code, I see a lot of stuff that makes heaps of sense
  141. eddybno stable naming conventions, there's more empty spaces and comments that code...
  142. eddyb*than
  143. pcmattmanIf all you see is the problems, you'll never understand it
  144. eddybanyway, what about that chainload thing
  145. eddyb?
  146. pcmattmanCombines the two operations into one micro-frame, I believe
  147. eddyb[15:01] <eddyb> hmm, I don't suppose I can chainload a floppy file from grub, can I?
  148. pcmattmanOh right, that chainload
  149. pcmattmanDunno
  150. kwolfYou mean, a floppy image?
  151. kwolfDon't think that's possible
  152. kwolfehenkes just joined #lost, btw. So if you'd like to talk to him, I can invite him.
  153. kwolf*lol*
  154. pcmattmanDoes eddyb know something I don't? :)
  155. Love4Boobiesdammit
  156. Love4BoobiesI don't think I can change name with a live connection, can I?
  157. pcmattmanNope
  158. XanClicNo
  159. Love4Boobiesok
  160. pcmattmanWell, I know I'm going to be having some fun rewriting parts of our USB stack over the next couple of days
  161. pcmattmanMay as well invite ehenkes though, can't hurt - it benefits both tyndur, CDI and Pedigree at the same time
  162. Love4Boobieswhat's up?
  163. kwolfpcmattman: It's your choice. I can always talk to him in German-speaking channels.
  164. Love4Boobieswhere are those channel logs?
  165. kwolfirc.tyndur.org
  166. kwolfDid you have some trouble with him, or what's your problem?
  167. pcmattmanYeah, I'm trying to figure out the problem here before I make a decision
  168. Love4Boobieskwolf: grep your logs for boo
  169. kwolfHe didn't like that? ;-)
  170. kwolfpcmattman: The only problem I have with him is that he always thinks he's right when he definitely isn't
  171. pcmattmanMm, I know too many people like that
  172. Love4Boobiespcmattman: what do you intend on rewriting?
  173. kwolfI mean, he started to tell me how things work right when he had a Hello World kernel running...
  174. pcmattmanQH/qTD creation and integration to the async list, mainly
  175. Love4Boobiesah
  176. Love4Boobiesmay not need that
  177. pcmattmanMainly as a result of reading Linux's driver and now having a look at PrettyOS
  178. Love4BoobiesI soon as I can run this crap on realHW
  179. pcmattmanIf nothing else cleaning up QH and qTD creation
  180. pcmattmanMy point is, "as soon as" may not happen if this doesn't get redone
  181. Love4Boobiesso better not touch/do anything unless I can't prove him wrong
  182. pcmattmanNah, I'll still make the changes, they'll need to be done eventually and are one step towards a solution
  183. Love4Boobiesok, it has an exe for pendrive install
  184. Love4Boobiesplease
  185. pcmattmanThrow that discussion onto #pedigree, and let's get ehenkes in here and start talking about his experience
  186. pcmattmanIf y'all don't mind, of course
  187. Love4Boobiesnot at all
  188. kwolfOk, I'll ask him
  189. pcmattmanHopefully he can identify the problem and knows a solution
  190. kwolfHeh, I shouldn't have asked publicly, I guess ;-)
  191. pcmattmanHeh.
  192. Love4Boobiesfound it: http://menuet.2.forumer.com/index.php?showtopic=231 :)
  193. Love4Boobiesso it is possible :P
  194. ehenkeshallo
  195. pcmattmanEvening
  196. kwolfHi ehenkes
  197. Programm_Noobhi
  198. ehenkeshie rgehts um EHCI?
  199. kwolfYes, but the Pedigree people don't speak German ;-)
  200. ehenkesok, no problem