#asahi · 2021-03-25 18:25
kettenisbut I think the PCIe controller does some latching#asahi · 2021-03-25 18:23
kettenisand my emulation was written with edge triggered interrupts in mind#asahi · 2021-03-25 18:22
kettenishowever since OpenBSD uses interrupt priorities, I had to emulate those#asahi · 2021-03-25 18:16
kettenismaz: so far I'm not treating the MSIs in a special way#asahi · 2021-03-08 09:32
kettenisthat's when I switched to looking at pcie#asahi · 2021-03-08 09:31
kettenisno port status changes#asahi · 2021-03-08 09:31
kettenisyeah, my attempts to use host mode failed to detect any devices#asahi · 2021-03-08 09:30
kettenisyou don't have to muck around with the i2c type-c controller?#asahi · 2021-03-08 09:24
kettenisso not everything in that document will apply#asahi · 2021-03-08 09:22
kettenissven: TI probably has slightly different glue logic for the dwc3 controller#asahi · 2021-03-05 17:15
kettenisso not too shabby indeed#asahi · 2021-03-05 17:15
kettenissame thing takes around 24 minutes on a Cortex-A57 core in the overdrive 1000 (AMD Seattle SoC)#asahi · 2021-03-05 16:51
kettenisaround 11 minutes on a single "efficiency" core ;)#asahi · 2021-03-05 16:51
kettenisheh, I'm building my OpenBSD kernels natively now#asahi · 2021-03-04 10:14
kettenisthere is no architected way to find out how big the TLBs are is it?#asahi · 2021-03-04 10:09
kettenisnot good if you have more than 127 active processes#asahi · 2021-03-04 10:08
kettenisyeah, I do the same on OpenBSD#asahi · 2021-03-04 10:07
kettenishmm, there is no note in revision F of the ARM ARM to indicate that 16-bit ASID support is mandatory#asahi · 2021-03-04 10:06
kettenisbut now I have to deal with running out...#asahi · 2021-03-04 10:06
kettenisrestricting ASIDs to 8-bit fixed the issue#asahi · 2021-03-04 10:05
ketteniswas assuming nobody would do an implementation without 16-bit ASIDs#asahi · 2021-03-04 10:04
kettenis8-bit, which was why I had my userland processes crash occasionally#asahi · 2021-03-04 10:03
kettenisnope#asahi · 2021-03-04 10:02
kettenisI do hope Apple implements support for 16-bit ASIDs if they're going to build high corecount versions of their SoCs#asahi · 2021-03-03 22:33
ketteniscertainly explains why I was seeing crashes with OpenBSD as our kernel assumed 16-bit ASIDs were available#asahi · 2021-03-03 22:06
ketteniswhat? the M1 only supports 8-bit ASIDs?#asahi · 2021-03-01 11:02
kettenisnever got things quite right for PA-RISC#asahi · 2021-03-01 11:02
kettenisaggressive caching of TLB entries always leads to interesting bugs#asahi · 2021-03-01 11:01
kettenisbut maybe it's not all fixed#asahi · 2021-03-01 11:01
kettenisfell into that trap at some point#asahi · 2021-03-01 11:00
kettenisyup#asahi · 2021-03-01 10:59
kettenisit tends to be shell processes that crash so this maybe an issue with what happens during fork()#asahi · 2021-03-01 10:58
ketteniswill take a look at that#asahi · 2021-03-01 10:58
kettenisI'm not sure...#asahi · 2021-03-01 10:52
kettenisflushing the ASID when a reallocate it doesn't work#asahi · 2021-03-01 10:52
kettenisflushing the ASID during a context switch makes it work#asahi · 2021-03-01 10:50
kettenisI think that's the case#asahi · 2021-03-01 10:48
kettenismight very well be a case where I'm not following the right "break then make" procedure#asahi · 2021-03-01 10:47
kettenisbut unlike Linux I flush the ASIDs when the process exits#asahi · 2021-03-01 10:46
kettenismore or less the same way as Linux does, reserving two ASIDs per process to hide the kernel mappings#asahi · 2021-03-01 10:45
kettenisusing ASIDs#asahi · 2021-03-01 10:42
kettenisprobably an OpenBSD bug revealed by th Apple cores doing more speculation#asahi · 2021-03-01 10:42
kettenisI'm currently chasing an issue with crashing userland binaries that seems to be related to TLB flushing#asahi · 2021-03-01 09:04
kettenismarcan: ah good I couldn't find that text#asahi · 2021-02-28 19:39
kettenisLinux obviously assumes that the CPUs are all in the inner-sharable domain#asahi · 2021-02-28 19:39
kettenisthe architecture doesn't really specify what the inner-sharable and outer-sharable domains really consist of#asahi · 2021-02-28 11:59
kettenismarcan: yes another dmb should be enough to guarantee that the device sees the register write after it sees the earlier memory writes#asahi · 2021-02-27 18:00
kettenisso if you need to write to a register to tell the hardware that you updated the descriptor, you need another barrier#asahi · 2021-02-27 17:59
kettenisbut it doesn't guarantee that the memory access has completed#asahi · 2021-02-27 17:58
kettenisso if you want to guarantee that a flag in a descriptor gets updated after all the other fields, dmb would be enough