| *** tpb <[email protected]> has joined #openrisc | 00:00 | |
| zx2c4 | shorne: oh thanks ill fix that re:virt | 11:18 |
|---|---|---|
| zx2c4 | and yea i also tried the same with irqsaverestore in those places and found the same things | 11:18 |
| shorne | zx2c4: I was reading the code a bit more (probably just forgot) but the on_each_cpu_mask framework already takes care of preempt_disable/enable and most of what my patch does | 11:35 |
| shorne | so really what my patch did doesnt do much at all | 11:35 |
| shorne | so somewhere else we are getting locked up | 11:36 |
| shorne | I'll try lockdep and other debug options to see if something comes out | 11:36 |
| shorne | I found one place (update_mmu_cache) where the or1k port calls on_each_cpu, but should just execute on the local cpu, so fixing that would be good | 11:37 |
| shorne | that fix still doesnt solve the lockups | 11:40 |
| shorne | zx2c4: thanks for the virt patch probably that one is better for your needs | 11:51 |
| zx2c4 | shorne: btw do you plan to do goldfish_tty in the virt machine like m68k does? | 11:58 |
| shorne | no, I was planning to continue with the 8250 serial, any reason to use goldfish? | 12:09 |
| shorne | no one replied to my patches to convert goldfish-rtc to configure it as big-endian | 12:10 |
| shorne | zx2c4: why use goldfish_tty? | 12:10 |
| zx2c4 | annoying nobody replied to that, hm | 12:10 |
| zx2c4 | i dunno, i just noticed its what other things are using. it's a "virtualization tty" | 12:11 |
| zx2c4 | dunno if that counts for much | 12:11 |
| zx2c4 | seems to have an early serial mode | 12:11 |
| shorne | the current serial has early serial | 12:11 |
| shorne | if you run qemu with '-append earlycon' it works | 12:12 |
| shorne | I just added it your Makefile, it seems to work for me even with the current .config | 12:13 |
| shorne | I am not sure if '-append earlycon' works for all platforms though | 12:13 |
| zx2c4 | huh | 12:30 |
| zx2c4 | btw m68k (smp=1 platform) finishes over a minute faster than or1k NR_CPU=1 | 12:30 |
| shorne | interesting, ill have to compare instructions | 12:31 |
| *** wxwisiasdf <[email protected]> has joined #openrisc | 20:53 | |
| wxwisiasdf | hello ? | 20:53 |
| wxwisiasdf | any1 here? | 20:53 |
| shorne | hello | 20:58 |
| wxwisiasdf | shorne: oh hello | 21:00 |
| wxwisiasdf | so uhm, i'm quite interested in OR1K i find it fascinating - in contrast to RISC-V which gets all the acclaim | 21:00 |
| shorne | wxwisiasdf: thats good, risc-v is also nice, but or1k I feel has low barrier to entry | 21:02 |
| wxwisiasdf | really? huh, neat | 21:02 |
| shorne | for example we just have one spec document, not 10s of extensions | 21:02 |
| wxwisiasdf | sorry for not knowing, but where can i find said document? | 21:03 |
| shorne | http://openrisc.io/architecture <-- latest 1.4 version pdf is here | 21:03 |
| tpb | Title: Architecture - OpenRISC (at openrisc.io) | 21:03 |
| wxwisiasdf | incredible, the vector extensions and stuff are out-of-the box right, not extensions like rv? | 21:04 |
| shorne | well, yes if the core implements it | 21:04 |
| shorne | in openrisc there are a few instructions that are optional and so not implemented in every cpu | 21:05 |
| wxwisiasdf | oh, | 21:05 |
| wxwisiasdf | like which? | 21:05 |
| shorne | I can't recall I would have to search, we have or1200, mork1x and marocchino implementations | 21:06 |
| shorne | would need to look at verilog code | 21:06 |
| wxwisiasdf | ah, well, for economical reasons i don't have the possibility of buying an fpga | 21:07 |
| wxwisiasdf | but qemu should work right, right? | 21:07 |
| shorne | I have been doing mostly qemu for recent development... | 21:07 |
| shorne | qemu doesn't have the vector extensions yet, it would not be too hard to add | 21:09 |
| wxwisiasdf | fair, | 21:09 |
| shorne | wxwisiasdf: I need to head off | 21:15 |
| shorne | zx2c4: I think these smp lockups may be qemu issues, the failures/deadlocks are always when running IPIs on cpus | 21:15 |
| shorne | my guess is the IPI hardware in qemu is not working right, i.e. if we get a timer and ipi at the same time, only the timer is run, then the ipi gets lost | 21:15 |
| *** wxwisiasdf <[email protected]> has quit IRC (Ping timeout: 240 seconds) | 21:58 | |
| *** wxwisias1f <[email protected]> has joined #openrisc | 22:01 | |
| *** wxwisias1f <[email protected]> has quit IRC (Ping timeout: 248 seconds) | 22:38 | |
| *** wxwisias1f <[email protected]> has joined #openrisc | 23:05 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!