#
✕ nick:eiln
#asahi-dev · 2023-10-06 08:43
eilnchadmed: agreed
#asahi-dev · 2023-10-06 08:43
eilnjanneg: if it's "noisy" as in grainy it's very likely a temporal/motion filter issue (constants are off, graph order isn't right). lina should have an m2 fork of the tracer, which I do not see in any branch
#asahi-dev · 2023-10-06 08:43
eilnsounds like m2 SET_FILE_LOAD abi changed between 12.4 & 13.5? maybe that's where iova became u32 -> u64? if so we'll need the dcp os version checks
#asahi-dev · 2023-10-05 23:32
eilncalib file id is in big endian btw
#asahi-dev · 2023-10-05 23:06
eilnSET_FILE_LOAD works even when I send garbage, looks like the struct issue
#asahi-dev · 2023-10-05 23:05
eilnif that's not the issue we can add a hook in the tracer to fetch the first 4 bytes of SET_FILE_LOAD iova to get the calib file id
#asahi-dev · 2023-10-05 23:04
eilnmarcan: can you check if m2 calib file send struct is right?
#asahi-dev · 2023-10-05 23:04
eilnjanneg: just in case, can you editing the sensor switch case id to "ISP_IMX248_1822_02" / "apple/isp_1822_02XX.dat"? I doubt this is the issue but just in case
#asahi-dev · 2023-10-05 09:05
eiln248v2 might be M2 but I doubt it. also I tested it on mine and v1 vs v2 looks the same
#asahi-dev · 2023-10-05 08:59
eilnmarcan: 12.4 doesn't even have the IMX248/558 V2 setfiles, so no way they're the correct ones for us. it looks like an iPad thing. sorry I meant to PR with that.
#asahi-dev · 2023-10-04 22:09
eilnmarcan: we'll have to clean it up a bit for upstream but it looks good to me :)
#asahi-dev · 2023-10-01 06:58
eilnthat happens every run
#asahi-dev · 2023-10-01 06:57
eiln> AppleH13CamIn::ISP_Suspend - Firmware timeout: 0, type: 3, cmdID: 0xffff, res: 0, isISPCPUNotInWFI: 1 indicate dirty shutdown to DART driver
#asahi-dev · 2023-10-01 06:57
eiln> AppleH13CamIn::ISP_Suspend - ISPCPU not in WFI after CISP_CMD_POWER_DOWN
#asahi-dev · 2023-10-01 06:57
eiln> AppleH13CamIn::ISP_Suspend - retries=20 ASCWRAP_IDLE_STATUS = 0x0
#asahi-dev · 2023-10-01 06:57
eilnpower
#asahi-dev · 2023-10-01 06:57
eilnlina: sounds like we're not cleaning up properly? j293ap they call CMD_STOP then CMD_POWER_DOWN. commands obviously don't send after CMD_STOP, so CMD_POWER_DOWN never succeeds. but CMD_POWER_DOWN doesn't succeed, ever, actually (same in facetimehd repo). I used CMD_SUSPEND instead because CMD_STOP actually causes a DART fault and makes ASC unrecoverable, I think they mess with
#asahi-dev · 2023-10-01 06:34
eilnlina: "[ERROR] PIODMA[0] Irq Error! ErrorCount: 1" in https://gist.github.com/jannau/139b7599dbae683e41b5dc12aa3ad889, right? PIODMA is DART->MMIO
#asahi-dev · 2023-10-01 05:21
eilnlina: is it still DART/DAPF? I will tune in if you do stream
#asahi-dev · 2023-09-30 01:54
eilnlina: that's awesome news. seems like we're all good now..?
#asahi-dev · 2023-09-29 23:23
eilnjanneg: cyan/magenta tinted? uh oh. that doesn't sound like the grey/grainy + rainbow spec err I used to get, mine was a filter graph ordering err
#asahi-dev · 2023-09-29 23:22
eilnI am wary of indexing "bt_surf" by strings though. "MISC" is everything they couldn't be bothered to name, and I'm surprised 0xc000 is a filter. maybe we can narrow it down by tracking state? also we shouldn't need to vmap every surface now
#asahi-dev · 2023-09-29 23:22
eilnlina: thanks for the fixes, studying them
#asahi-dev · 2023-09-28 21:38
eilnah
#asahi-dev · 2023-09-28 21:36
eilnI just realized the m2 macbook air is 1080p but the pro regressed to 720p. that is weird
#asahi-dev · 2023-09-28 21:06
eilneven better, CH print overflows for non-CH ioctls. I am surely on the 91012th sensor.
#asahi-dev · 2023-09-28 21:05
eiln> ISPASC: CH = 0x16384, CMD = 0x0011 [CISP_CMD_SET_ISP_PMU_BASE], receive at 105531.734 ms, delta: 1.0234 ms.
#asahi-dev · 2023-09-28 21:05
eiln> ISPASC: PMU base is 0x9587146752
#asahi-dev · 2023-09-28 21:05
eilnlina: 16 looks like count in BUFFER_POOL_CONFIG_SET
#asahi-dev · 2023-09-28 08:51
eilnjannau: t6001 should work though. do you have a trace?
#asahi-dev · 2023-09-28 08:51
eilnMLVNR just won't die. it's 248 (J413AP 720p) with SCL1, but 248 only has one camera config. either we bring back the platform_id hack or I RE and port ANE-ISP IPC (probably even more platform problems).
#asahi-dev · 2023-09-28 08:51
eilnmarcan: sigh. why is the isp this uncooperative
#asahi-dev · 2023-09-25 21:17
eilnupdate we can decode with the firmware. trigger "mailbox" by literally writing to nvic with the physaddr it jumps to. I just really, really don't want to do it this way. I'm gonna fuzz the params to try to get the intermediate passes
#asahi-dev · 2023-09-25 18:39
eilnwe should be able to trace meta though
#asahi-dev · 2023-09-25 16:41
eilnas a last resort I think I can kick up the ane for this sake. we're not calling face detection, we just need the client up. not sure how the hell that's gonna work in v4l2 though
#asahi-dev · 2023-09-25 16:28
eilnwill look at the buffer logic in a bit. that bit has always been hacky, but got even hackier with the single planar hack
#asahi-dev · 2023-09-25 16:28
eilnlina: the platform_id clobbering is a disgusting hack, I love it. don't we install 13.5 as default though?
#asahi-dev · 2023-09-25 16:28
eilnam slightly sick too. was whining about the heat a week ago and now the cold has gotten me. woke up at midnight rn
#asahi-dev · 2023-09-24 01:06
eilnlina: cleaned it up and PRed/pushed to isp-m1n1
#asahi-dev · 2023-09-23 16:09
eilnyes!
#asahi-dev · 2023-09-23 16:09
eilnI'll actually clean that up rn
#asahi-dev · 2023-09-23 16:08
eilnsounds great :)
#asahi-dev · 2023-09-23 16:08
eilnI'm free all of tomorrow, and usually past 6pm KST on weekdays
#asahi-dev · 2023-09-23 16:07
eilnsorry I wasn't clear, argh. should've bundled it with the isp pr too
#asahi-dev · 2023-09-23 16:02
eilnI'm at 2:32:18 in the stream rn. I sent the patched tracer and opcodes a couple days ago. It should work OOTB adjusting the mailbox offsets to get the interrupt hooks. It dumps all the structs :(
#asahi-dev · 2023-09-23 16:02
eilnlina: boss kept me around extra late today, so sorry
#asahi-dev · 2023-09-22 22:16
eilnignore that
#asahi-dev · 2023-09-22 22:15
eilndw
#asahi-dev · 2023-09-22 11:00
eilnlina: estimated time of the sekrit work stream (if any)? I really don't wanna miss that one ;)
#asahi-dev · 2023-09-21 08:58
eilnone of the greatest computer experts in the world not is familiar with operating systems :( https://github.com/tinygrad/tinygrad/blob/cb5d6576cbf47a145676ff6f1b067439d5a50775/examples/llama.py#L459-L464

plus ›