#milkymist IRC log for Thursday, 2012-06-14

kristianpaulgcc 4.7.1 freezed my machine when trying to compile lm32...19:26
wpwrakout of memory ?19:26
wpwrakor maybe a file system corruption19:27
kristianpaulhard disk got too much active..19:27
kristianpaulperhaps because out of memory19:28
kristianpaulwpwrak: how i can now the cause if the system got freeze?19:55
mischief6maybe try storing the source/objects in ramfs19:58
sh4rm4kristianpaul, running it with ulimit on should give you a clue20:19
sh4rm4LDFLAGS="-Wl,--no-keep-memory" might help to keep memory usage *a lot* smaller20:20
lekernelkristianpaul: use clang :)21:05
lekernelchances are the gcc bugs didn't get fixed, and in fact probably got worse21:06
kristianpaulclang filled my home dir with 10Gb of garbage...21:10
kristianpaulor whatever21:10
kristianpaulif got worse i'll last best working21:10
lekernelworksforme... and it's ~300MB21:10
kristianpaulbad experience :)21:10
lekernelwhat "garbage" did it produce?21:11
kristianpaulwhen running configure ;)21:11
kristianpaulhavent able to get it with cmake21:11
lekernelah, autocrap. haven't tried it - only cmake...21:11
lekernelI think it's better to fix this cmake problem rather than gcc bugs... most probably easier too21:12
wpwrakkristianpaul: if the disk gets very busy, this sounds like a lack of memory21:12
lekerneleither way gcc will segfault during bootstrap (or fail sooner)21:16
GitHub36[milkymist-mmu] fallen pushed 1 new commit to mmu-bios: http://git.io/Y2GvRw21:26
GitHub36[milkymist-mmu/mmu-bios] Fix incorrect return to physical address upon ITLB page fault exception - Yann Sionneau21:26
wpwrakimportant detail :)21:28
Fallenouit's not correct yet, I just noticed I modify ea *after* calling save_all , it will be overwritten by the call to restore_all_and_eret21:31
wpwrakon a dtlb fault, is the itlb deactivated ? or does it keep its mode ?21:31
Fallenouupon any exception, both dtlb and itlb are desactivated21:32
Fallenouupon eret (return from exception) both itlb and dtlb are activated back21:32
wpwrakgood. a dtlb miss then pushes the virtual address ?21:33
FallenouI hope :)21:37
FallenouI have not checked, but it should21:37
FallenouI hope it's not pushing the physical address21:37
Fallenouwill check tomorrow, I am running out of battery right now =)21:37
Fallenouthe thing is upon dtlb_miss you have access to the virtual address which caused the exeption21:42
Fallenoubut it's the loaded or stored address21:42
Fallenounot the address of the instruction itself21:42
Fallenounot sure if I will have easy access to the virtual address of the instruction causing the dtlb miss21:42
wpwrakyou'll need it for stack traces ... and also to return after resolving the fault21:43
Fallenouin theory PC should contain virtual address21:43
Fallenouso EA should be correctly set by the pipeline :o21:43
wpwrakgood. that's what i would have expected :) now, why is the PC physical (and not virtual)  in the case of an itlb miss ?21:44
Fallenouwell it's a good point21:44
FallenouI'm now feeling that my previous commit is total bullshit21:44
Fallenoudamn it21:44
FallenouI was completely sure about it this morning in the subway :)21:44
Fallenoulet think about it again tomorrow with hopefully a more clear/healthy brain21:45
Fallenouenough scary commits for tonight, gn8 :)21:46
wpwrakfaultless dreams ! :)21:47
Fallenouahah thanks, you too :p21:47
--- Fri Jun 15 201200:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!