--- Log for 17.07.119 Server: hitchcock.freenode.net Channel: #rockbox --- Nick: logbot Version: Dancer V4.16 Started: 25 days and 5 hours ago 00.03.44 Quit GeekShadow (Ping timeout: 248 seconds) 00.14.36 Quit eingast (Remote host closed the connection) 00.16.04 Quit SammysHP (Quit: *wuff*) 00.16.43 Join SammysHP [0] (~SammysHP@faol.sammyshp.de) 00.22.33 # __builtin having alignment issues again with lua on arm I suspect it has something to do with the plugin buffer but it is hard af to diagnose since the player locks up when it happens 00.31.43 Join krabador [0] (~krabador@unaffiliated/krabador) 00.32.14 # <__builtin> What device? 00.33.45 # <__builtin> If it has a piezo you can beep out the fault address in binary 00.34.47 # <__builtin> ulmutul: I will look into the flashes next time 00.34.48 *** Saving seen data "./dancer.seen" 00.35.33 # it is the clip+ that I tested it on but the clipzip and the fuze+ do it too just not at the same time lol 00.36.18 # <__builtin> Bilgus: can you repro on a device with piezo? 00.36.33 # its not just lua though generally when it happens codecs do it as well 00.36.56 # I haven't any devices with a piezo lol 00.37.05 # <__builtin> What exactly is the issue? 00.37.50 Quit foolsh (Ping timeout: 272 seconds) 00.37.52 # play a file open lua and it crashes the device sound plays till buffer runs out then non responsive with no screen 00.38.27 # put alignment in or add an instruction retry works fine every time 00.39.10 # <__builtin> No crash dump? 00.40.09 # nothing 00.40.40 # <__builtin> I could try and reproduce it on my 6g and get you a fault address 00.41.48 # ok hold on a few and I'll build an image although your device probably will work fine as I don't see it across all devices at all times but they each do it time to time 00.48.48 Join noobineer [0] (~noobineer@2601:401:8200:26d8:4820:ec54:435a:87be) 00.55.56 # __builtin, https://www.mediafire.com/folder/uqu76fppilrqe/ 01.00.08 # wonder if I could use the same idea to flash the backlight lol 01.03.32 # <__builtin> I'll test it later 01.04.05 # thanks 01.09.59 Join foolsh [0] (~quassel@185.246.209.88) 01.11.40 # <__builtin> Bilgus: the backlight sounds like a good idea too 01.12.11 # <__builtin> Maybe short/long flashes for 0/1 01.12.40 Quit MrZeus (Ping timeout: 252 seconds) 01.12.46 # <__builtin> See g#2105 for how I did it 01.12.48 # Gerrit review #2105 at http://gerrit.rockbox.org/r/2105 : system-arm: log PC address via ipod6g piezo on fault (HACK) by Franklin Wei 01.15.07 # yeah I just tried it and either it doesn't reach the handler or the screen is too borked to do it 01.15.47 # <__builtin> You're using the UIE handler, correct? 01.16.46 # yes 01.17.02 # same as your patch 01.21.23 Quit krabador (Read error: Connection reset by peer) 01.22.06 Join krabador [0] (~krabador@unaffiliated/krabador) 01.22.15 # yeah it never reaches it I put system_reboot() at the top and nothing new I wonder if tlsf could get stuck trying to allocate on unaligned addresses 01.46.49 # __builtin, Interesting I put in code to align down the pluginbuf address to tlsf and it doesn't crash then I thought well yeah changed the size with my code so I wrote iut so gcc couldn't optimize it away and tried with % 1 and it crashed 01.47.34 # I know our tlsf is slightly old I wonder if its a known bug 01.53.37 # I think I have it figured out 01.56.29 # so lua uses get_new_area exclusively to grab the remainder of the pluginbuf 01.57.03 # but if you use init_memory_pool it checks to see if it is word aligned 01.57.10 # https://github.com/Rockbox/rockbox/blob/master/lib/tlsf/src/tlsf.c#L467 01.57.24 # if (((intptr_t) mem_pool & PTR_MASK)) { 01.57.24 # ERROR_MSG("init_memory_pool (): mem_pool must be aligned to a word\n"); 01.57.24 # return -1; 01.57.24 DBUG Enqueued KICK Bilgus 01.57.24 # } 01.57.35 Quit ZincAlloy (Quit: Leaving.) 01.58.38 # now the question is how do I force alignment while rounding the proper way on stacks that grow up or down 02.34.52 *** Saving seen data "./dancer.seen" 03.57.24 Join dandels [0] (~dandels@unaffiliated/dandels) 03.58.27 Quit krabador (Remote host closed the connection) 04.07.10 Quit dandels (Ping timeout: 268 seconds) 04.34.55 *** Saving seen data "./dancer.seen" 04.44.30 Join GeekShadow [0] (~antoine@reactos/tester/GeekShadow) 05.16.14 Quit TheSeven (Ping timeout: 264 seconds) 05.16.34 Join TheSeven [0] (~quassel@rockbox/developer/TheSeven) 05.17.44 Quit foolsh (Ping timeout: 244 seconds) 06.32.32 Join foolsh [0] (~quassel@89.187.181.158) 06.34.58 *** Saving seen data "./dancer.seen" 06.49.38 Quit JanC (Ping timeout: 252 seconds) 07.11.49 Quit foolsh (Ping timeout: 258 seconds) 07.43.26 # __builtin, g#2113 fixes it for me I see you had disabled memset in 2017 do you remember details about that> 07.43.28 # ? 07.43.29 # Gerrit review #2113 at http://gerrit.rockbox.org/r/2113 : ARM fix non aligned crashes with tlsf by William Wilgus 08.01.50 Join ZincAlloy [0] (~Adium@2a02:8108:9440:dfc:4dee:52d4:9017:f831) 08.35.00 *** Saving seen data "./dancer.seen" 08.36.02 Quit TheSeven (Ping timeout: 264 seconds) 08.37.50 Join TheSeven [0] (~quassel@rockbox/developer/TheSeven) 10.29.23 Quit Huntereb (Ping timeout: 272 seconds) 10.35.04 *** Saving seen data "./dancer.seen" 10.53.50 Join Huntereb [0] (~Huntereb@d-69-161-102-241.va.cpe.atlanticbb.net) 11.21.34 Join vmx [0] (~vmx@ip5f5bfdfb.dynamic.kabel-deutschland.de) 12.35.06 *** No seen item changed, no save performed. 12.50.59 Quit ZincAlloy (Quit: Leaving.) 13.57.49 Join massiveH [0] (~massiveH@ool-18e4e27c.dyn.optonline.net) 14.35.08 *** Saving seen data "./dancer.seen" 14.57.57 Join foolsh [0] (~quassel@185.246.209.88) 15.05.40 Join JanC [0] (~janc@lugwv/member/JanC) 15.05.42 Join JanC_ [0] (~janc@lugwv/member/JanC) 15.07.05 Quit JanC_ (Remote host closed the connection) 15.08.26 Nick JanC is now known as Guest31310 (~janc@lugwv/member/JanC) 15.16.28 Quit Guest31310 (Quit: 'k zien d'r mee weh zi) 15.16.53 Join JanC_ [0] (~janc@lugwv/member/JanC) 15.18.31 Nick JanC_ is now known as JanC (~janc@lugwv/member/JanC) 15.26.56 Quit massiveH (Quit: Leaving) 15.38.23 Quit Galois (Remote host closed the connection) 15.39.51 Join Galois [0] (djao@efnet-math.org) 15.40.14 Quit Galois (Remote host closed the connection) 15.40.29 Join Galois [0] (djao@efnet-math.org) 16.28.47 Join speachy [0] (40eebded@64.238.189.237) 16.29.06 Quit speachy (Remote host closed the connection) 16.29.50 Join speachy [0] (40eebded@64.238.189.237) 16.35.12 *** Saving seen data "./dancer.seen" 17.11.56 Join dys [0] (~dys@b2b-94-79-136-46.unitymedia.biz) 17.20.03 # Build Server message: New build round started. Revision 47639fb, 280 builds, 9 clients. 17.36.39 Join ZincAlloy [0] (~Adium@2a02:8108:9440:dfc:719a:509b:a51:16c8) 17.40.50 Quit deevious (Quit: deevious) 18.09.51 Quit vmx (Remote host closed the connection) 18.35.13 *** Saving seen data "./dancer.seen" 19.20.51 Join lebellium [0] (~lebellium@89-92-69-110.hfc.dyn.abo.bbox.fr) 20.26.25 Join MrZeus [0] (~MrZeus@2a02:c7f:70d4:cd00:640e:264:e9f4:b11b) 20.35.15 *** No seen item changed, no save performed. 21.43.41 Join Rower [0] (~husvagn@m83-185-86-146.cust.tele2.se) 21.47.10 Quit Rower (Client Quit) 21.48.44 Join Rower [0] (~husvagn@m83-185-86-146.cust.tele2.se) 22.20.52 Ctcp Ignored 1 channel CTCP requests in 0 seconds at the last flood 22.20.52 # * __builtin is dipping his toe in ARM assembly 22.23.37 Join Rower- [0] (husvagn@d83-183-134-99.cust.tele2.se) 22.23.57 # Be careful. Combining toes and arms can be confusing 22.25.42 # heh, that should be 'combing thumbs and arms' :P 22.25.51 Quit Rower (Ping timeout: 245 seconds) 22.26.13 # That would make more sense :) 22.27.27 # oh, there are so many jokes along those lines. 22.30.42 # __builtin can you see if puzzles still crashes with memset enabled and if so if 2113 fixes it? 22.32.10 # <__builtin> Bilgus: give me a moment 22.35.17 *** Saving seen data "./dancer.seen" 22.36.08 # <__builtin> it feels like we're missing the root problem with all this memset() crap 22.36.27 # <__builtin> could it just be that our ARM memset has an issue with unaligned addresses? 22.37.08 # well, the mips memset was broken in that way 22.38.36 # but it does seem a bit odd that memset would have been broken this long.. 22.39.13 # <__builtin> I remember ARM's couldn't handle memsetting a negative value 22.43.25 # <__builtin> so it's not entirely impossible 22.46.38 Quit lebellium (Quit: Leaving) 22.50.32 # <__builtin> Bilgus: memset-tlsf appears to work fine under puzzles 22.50.40 # <__builtin> though I haven't tested much 22.50.53 # <__builtin> if you feel like reverting my patch, I'm fine with it 22.52.26 # I think it actually has to do with the offset being aligned to 4 and I'm not entirely sure if its actually arm or tlsf but I did try a newer version of tlsf and had the same crash 22.53.17 # I think when you removed memset you probably just changed the alignment slightly same as I did in lua 22.53.20 # <__builtin> I didn't give it an unaligned buffer though 22.53.55 # me either it seems to happen when the buffer is ending with 4 22.54.21 # moving it to 8 16 32 64 128 all seem to work fine 22.54.45 # its almost enough to make it look random 22.55.50 # how often does your plugin end in 0x4 (rhetorical) 22.57.17 # <__builtin> I could test 22.57.29 # <__builtin> or purposely make it end in 4 22.59.25 # yeah I did the same adding buffers of chars and nops and was able to crash it pretty consistently if music were plasying so there might be something else going on there 23.00.05 # perhaps an overflow in tlsf since the buffers are right next to each other 23.23.09 Quit speachy (Ping timeout: 260 seconds) 23.43.45 Quit Rower- (Quit: Hmmm...)