#asahi-dev · 2023-03-14 16:04
marcan 9 files changed, 534 insertions(+), 535 deletions(-)#asahi-dev · 2023-03-14 15:15
marcanokay fine I'll make it a real driver#asahi-dev · 2023-03-14 15:14
marcandammit.#asahi-dev · 2023-03-14 15:14
marcanargh.#asahi-dev · 2023-03-14 15:14
marcanoh wait this is the DT thing isn't it. ugh.#asahi-dev · 2023-03-14 15:13
marcanit doesn't?#asahi-dev · 2023-03-14 12:41
marcanrtkit already has a workqueue#asahi-dev · 2023-03-14 12:40
marcanoh I don't think we need that any more#asahi-dev · 2023-03-14 12:35
marcanOTOH if actually hitting the hardware is the bottleneck, then my dedup thing should help in normal mode once I implement it#asahi-dev · 2023-03-14 12:34
marcanmaybe it's just because it slightly multithreads where it otherwise wouldn't?#asahi-dev · 2023-03-14 12:34
marcanthis is dumb#asahi-dev · 2023-03-14 12:34
marcanwait no I got it backwards, it's faster with mailbox lol#asahi-dev · 2023-03-14 12:31
marcanthat's 13% faster#asahi-dev · 2023-03-14 12:31
marcanokaaay terrain 147 -> 166 in sync mode#asahi-dev · 2023-03-14 12:25
marcanactually I should've tested sync mode...#asahi-dev · 2023-03-14 12:24
marcanit's quite possible some benchmarks will notice more#asahi-dev · 2023-03-14 12:24
marcanwith the async one the CPU is less in the way#asahi-dev · 2023-03-14 12:24
marcanthe synchronous one#asahi-dev · 2023-03-14 12:24
marcanmaybe the old driver would've gotten more actually#asahi-dev · 2023-03-14 12:24
marcanyeah, oh well#asahi-dev · 2023-03-14 12:23
marcannow it's faster and has fewer bugs and nonsense#asahi-dev · 2023-03-14 12:23
marcanI literally took the mailbox driver, ripped out the parts that talk to the subsystem, and just made rtkit call into it directly#asahi-dev · 2023-03-14 12:23
marcanand it doesn't *do* anything for us, there is no useful common functionality whatsoever#asahi-dev · 2023-03-14 12:22
marcanmaz: re mailbox, it's just bad. atomic mode bugs, a fixed-size queue that errors out if full instead of blocking, missing a way to set up the mailbox and *then* enable reception once everything is set up, etc.#asahi-dev · 2023-03-14 12:20
marcanalso s3_1_c15_c11_0 is some kind of global enable, if that is 0 the PMCs don't count (but it's 0xfffff by default)#asahi-dev · 2023-03-14 12:20
marcanconfirmed, that seems to be all you need to get them to count in EL1#asahi-dev · 2023-03-14 12:17
marcanamarioguy: the PMCs definitely count for me in EL1 with those bits set, and possibly others (checking)#asahi-dev · 2023-03-14 12:05
marcanthere's probably a whole set of parallel impdef PMC registers for {E2H,TGE}={1,0}, as there is for most apple impdef stuff#asahi-dev · 2023-03-14 12:05
marcanamarioguy: hint: s3_1_c15_c7_2 has the same layout as PMCR1_EL1, it's probably PMCR1_EL12 (and then s3_1_c15_c6_2 is probably PMCR1_EL2)#asahi-dev · 2023-03-14 11:34
marcannot as much of a perf improvement as I'd hoped, but the diffstat is ~-50 lines of code vs upstream and might have fixed the reboot issues and obsoletes a pile of downstream mailbox patches we had trouble upstreaming that fixed various broken stuff#asahi-dev · 2023-03-14 11:32
marcanasahi-6.2-7 pushed de-mailboxed#asahi-dev · 2023-03-14 03:05
marcaneiln: saw that, might wait until the current crop of kernel issues/release is sorted before merging new stuff though#asahi-dev · 2023-03-13 13:53
marcanpushed another asahi-configs bump to disable blur by default in kwin#asahi-dev · 2023-03-13 13:49
marcanjannau: yes, it's supposed to be slow as heck with that#asahi-dev · 2023-03-13 12:16
marcanso we can prooobably punt on that as long as we at least disable blur by default because that is very broken#asahi-dev · 2023-03-13 12:16
marcanjannau: the really nasty kwin stuff goes away with plasmashell running, probably by accident#asahi-dev · 2023-03-13 12:14
marcan5.15.8-1.fc37 I assume#asahi-dev · 2023-03-13 12:11
marcanjannau: the fedora repro, was that on qtdeclarative-5.15.8?#asahi-dev · 2023-03-13 11:39
marcansee toot#asahi-dev · 2023-03-13 11:39
marcanjannau: the problem is x9 getting clobbered#asahi-dev · 2023-03-13 11:36
marcanthat makes it work and it's not that slow#asahi-dev · 2023-03-13 11:35
marcanok, I think we are shipping QV4_FORCE_INTERPRETER=1#asahi-dev · 2023-03-13 11:22
marcandid they get the ABI wrong?!#asahi-dev · 2023-03-13 11:22
marcanok, WTF, this JIT code is calling a function then immediately storing a caller-preserved register into the stack#asahi-dev · 2023-03-13 11:10
marcanlet me see if I can do this again in a less stupid way...#asahi-dev · 2023-03-13 11:10
marcanthe write of the bad value occurs in what pretty clearly is JIT code#asahi-dev · 2023-03-13 11:09
marcanI worked around it with a ridiculous number of automated breakpoints to skip over the problem areas#asahi-dev · 2023-03-13 11:09
marcanok, it's a JIT bug#asahi-dev · 2023-03-13 11:00
marcanand "stp" also...#asahi-dev · 2023-03-13 10:37
marcanI think if you set a watchpoint and it gets hit by "dc zva", it hangs