IRC logs for #openrisc Saturday, 2014-12-20

--- Log opened Sat Dec 20 00:00:19 2014
stekernolofk: all the single stepping issues have been addressed, so shouldn't be related05:20
stekernthere could be new issues of course, there's been some changes going on in the control module05:22
* stekern thinks he's got the multiway tlbs ready for some testing07:16
stekernat least the code works with OPTION_xMMU_WAYS set to 107:16
poke53281stekern: Can you explain what this means. I looked up in the specification, but didnt find much.12:49
poke53281Do you need support under Linux?12:50
stekernbut I haven't done any of that yet13:08
stekernit means that you have x number of match and translation regs per each access13:10
stekernwhere x is 0-313:10
stekernso, the changes necessary for linux is to add some replacement mechanism to the tlb miss routines13:11
stekerni.e. lru, random or something like that13:12
olofkstekern: I ran the code to be single-stepped from ROM if that makes any difference14:32
poke53281Ahh, Ok. I think I understand15:22
poke53281so instead of writing always to way 0 you rotate through 0-315:23
-!- Netsplit *.net <-> *.split quits: trem, blueCmd, radii_, aou, Amadiro, ssvb16:30
-!- Netsplit *.net <-> *.split quits: Hesham, kiwichris16:42
-!- Netsplit over, joins: kiwichris16:45
-!- Netsplit *.net <-> *.split quits: atgreen16:45
-!- Netsplit over, joins: atgreen16:46
_franck_olofk: you can't singlestep code in ROM. You need hardware breakpoint for that17:33
olofk_franck_: Aha. I suspected there could be something like that18:57
olofkDoes the regular breakpoint just exchange the instruction with a trap?18:57
olofkAnd does the debug unit support hw breakpoints?18:57
olofkblueCmd_: Is freecores on github just a single snapshot or is it continously updated?18:58
olofkNot that it would make any difference for most of the repos :)19:01
_franck_olofk: right, software breakpoint replace the instruction by a trap19:14
_franck_and no there is no hardware brakepoint19:14
_franck_AFAIR, in the adv_debug_if svn repo, there is a patch for HW break point19:15
stekernyeah, but that's for or1200, no?19:26
stekernsupport for hw breakpoints (and watchpoints) would be cool19:27
olofkstekern, _franck_ : Where does the extra bits need to be added? In mor1kx, adv_debug_sys, both?19:43
olofkHmm.. why doesn't my USB serial thingamob show up in /dev?20:14
olofkI can see that it's found in dmesg, and I think I loaded the correct kernel module20:15
stekernnot sure, I would assume that adv_debug_sys has the support for it already20:18
stekernbut in mor1kx you'd need to add support for it20:18
olofkI looked at the code in mor1k_ctrl_cappuccino, and I would say thta the debug stuff in there is begging to be moved to a separate module20:35
stekernI'm not going to argue against that20:35
stekernconsider that the tick timer and pic was in there at some point as well ;)20:36
stekernjuliusb did a nice job factoring those out20:36
olofkI guess there would be extra benefits if the stuff could be shared between the espressos as well20:36
stekernyes, definitely20:37
olofkI took a first shot at it, but I would need more time to dive into the functionality and preferably some regression tests as well20:43
olofkhaha. Just noticed a weird bug/feature when using screen to connect to my USB serial adapter. I killed the screen from another terminal, but when the screen exited, the keystrokes were still redirected to the serial adapter20:51
stekernfirst shot at refactoring out the debug stuff, or hw bp?21:01
stekernok, cool21:01
stekernI can't decide if I like or dislike the fact that verilator ignores the difference between reg and wire21:05
olofkI think it's a bad thing. Verilator is usually quite picky with those things21:06
stekernI think I'm inclined to liking it21:06
olofkFor system verilog there's the 'logic' type which does ecactly that21:07
stekernall other tools will tell you about it right away, so it's going to be caught early anyways21:07
stekernand I bet you can set verilator to complain about it as well21:07
olofkYes, so you lose an early warning :)21:07
olofkYeah probably. It has tons of options21:07
stekernbut in early development of changes, it's nice to not have to bother with changing the type of the signal when you do a quick test of turning a assign sig = val; to always @(*) sig = val;21:08
olofkHuh? Why does gdb suddenly give me signed ints instead of hex when I read the memory?21:08
stekernbecause it remembers the setting, and I bet you have read something as a signed int previously21:09
olofkIt stopped doing it now21:09
olofkFuck knows why21:09
olofkHow do I reset the CPU from gdb?21:10
_franck_olofk: try "mon reset", or set npc to 0x10021:12
stekerncool, linux 3.18 merged without pain, and boots too21:14
_franck_this argument order to remember in fusesoc is anoying21:14
_franck_"verbose" before "sim", "force" after "sim"21:15
olofk_franck_: Yeah I know. I thought it was a clever way to do it, but it's annoying when you need to insert something in the middle21:15
olofkBut it's pretty cool that you can get help on different topics depending on where you put --help21:16
stekernreminds me... I should dig in and get autotools to automagically install that tab-completion script21:16
_franck_stekern: when will you push your linux upstream. It's about time the vanilla Kernel boots nicely (AFAIR it does not now)21:16
olofkstekern: Yeah, that would be great. Do you need any support in fusesoc for that?21:16
stekern_franck_: well, the patches that's needed for that has all been posted on lkml afaik21:17
stekernit's up to Jonas to pick them up...21:17
_franck_stekern: ah ok21:18
stekernI've offered to step in as a co-maintainer if he's too busy to handle it, but he never responded on that21:19
olofkYeah, it would be great to get some attention on that21:19
_franck_you sould insist ;) That would be a great benefit for our community21:19
stekernyeah, I should, I know. otoh, I'm also afraid of getting what I'm asking for and all that ;)21:23
--- Log closed Sun Dec 21 00:00:20 2014

Generated by 2.15.2 by Marius Gedminas - find it at!