#milkymist IRC log for Tuesday, 2012-07-24

hellekinso, when I meet with the guys from CEITEC, the chip fab in Brazil, what should I tell them to convince them to build free hardware?00:03
wpwrakhellekin: "it's fun" ? :-)00:05
wpwrakhellekin: what kind of chips do you have in mind ?00:05
hellekinwpwrak: I think the best compromise would be some router stuff. They do have radio engineers, but mostly to produce RFID, so it's probably not the same as proper RF or WiFi00:06
hellekinbut in general, I'd like that they think about producing public infrastructure, so cheap mesh networking would be best00:07
wpwrakshall we move it to #qi-hardware ?00:08
hellekinas a fallback, PBX chips could be interesting. I don't see any of non-leaders building consumer nanoware anytime soon. A smartphone is very unlikely, although Brazil does have some local production (e.g., gradiente)00:09
hellekincould be00:09
wpwrakhmm. board-m1/r4/m1.net - why ?!? it's a generated file09:31
FallenouI think I know why adding the rcsr to the itlb miss handler made it work, and removing it made it go in endless loop19:04
Fallenousomehow the value (modulo 16) of the address of instruction doing the eret is important19:04
Fallenoubecause if you have a icache miss while doing eret, you are going to tlb miss during itlb handler19:06
Fallenoubecause icache miss is delaying the pipeline movement19:06
Fallenouso if eret address is a multiple of 0x10 then you lose (or something like that)19:07
mwalleFallenou: athough you can do ops with vaddr now in one operations, you still need one more instruction for the or ;)20:58
mwalleFallenou: how many bits are unused in the VADDR register?20:58
Fallenouhum I would say something like 1120:59
Fallenoudepends on your page size though :)20:59
Fallenoufor 4 kB pages, you have 12 unused bits, one you have to keep for TLB selection21:00
mwalleand we encode the operation in that free bits, now?21:04
mwalleexcept for update, which is implcitly done with writes to paddr21:04
Fallenouyes that's what we agreed on :)21:20
mwalleupdate is not atomic, but i guess thats ok21:30
Fallenouupdate should run only in miss handler with no exceptions21:45
mwalleFallenou: wheres the valid bit?21:54
mwalleeg if i want to read back a TLB entry21:55
GitHub36[linux-milkymist] sbourdeauducq pushed 1 new commit to ng: http://git.io/9z25hg22:35
GitHub36[linux-milkymist/ng] merge - Sebastien Bourdeauducq22:35
Fallenoumwalle: wy do you want to read back a tlb entry ?22:56
Fallenouyou have the page tables in the OS22:56
Fallenoutlb can be updated or invalidated22:57
Fallenoubut I didn't think the OS needed to be able to read the content22:57
Fallenouas of now there is no read back feature22:57
--- Wed Jul 25 201200:00

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