#asahi-dev · 2023-04-28 12:10
poviki only read the text above device_link_add#asahi-dev · 2023-04-28 12:10
povikaha.#asahi-dev · 2023-04-28 12:10
poviki am surprised it doesn't work as is, that the device link to a consumer not supporting runtime PM doesn't prevent the DART from suspending implicitly#asahi-dev · 2023-04-28 12:08
povikthat would fix it, but my understanding is DL_FLAG_RPM_ACTIVE would keep the DART up even for devices that do support runtime PM and are suspended#asahi-dev · 2023-04-28 11:51
povikand the same DART will get resumed because of ADMAC, which shares it#asahi-dev · 2023-04-28 11:51
povikhttps://github.com/AsahiLinux/linux/blob/2fb9510f953092b1cc94e0cb26713f2f152754f8/drivers/iommu/apple-dart.c#L1341#asahi-dev · 2023-04-28 11:51
povikit does because of this reset#asahi-dev · 2023-04-28 11:47
povik* so PM core doesn't keep DART up while SIO is up#asahi-dev · 2023-04-28 11:47
povikso PM core doesn#asahi-dev · 2023-04-28 11:47
povikand of course it gets hit by the dart resume reset like the pcie devices get#asahi-dev · 2023-04-28 11:47
povikthe dart gets suspended/resumed while sio is in probe#asahi-dev · 2023-04-28 11:47
poviki don't think this is the same issue you have in mind#asahi-dev · 2023-04-28 11:09
povikjannau: ^#asahi-dev · 2023-04-28 11:08
povikthat's at probe time, maybe pm_runtime_enabled is true, like for those PCIe devices, when the device link is being set up#asahi-dev · 2023-04-28 11:08
povikthis happens with SIO (yes, that was the SIO boot breakage...) for which pm_runtime_enabled is false though#asahi-dev · 2023-04-28 11:07
povikcould this be the same issue as the one being worked around for PCIe?#asahi-dev · 2023-04-28 11:07
povikthat goes away when a driver for the consumer runs pm_runtime_set_active explicitly#asahi-dev · 2023-04-28 11:07
povikseems like PM core makes DART go suspend even with active consumers, if those consumers don't support runtime PM#asahi-dev · 2023-04-28 11:06
povikhttps://github.com/AsahiLinux/linux/commit/7edfa962bd38fc79c38b845c7cc87cc2f762cb42#asahi-dev · 2023-04-27 13:41
poviknever mind i switched what INBOX and OUTBOX means, so i don't seem to have gotten the reply#asahi-dev · 2023-04-27 13:33
povikwhich is in line with interpreting BIT(2) as IRQ_NOT_PEND#asahi-dev · 2023-04-27 13:33
povikoh, i forgot after i read the read sensitive registers CPU_STATUS changes to 0xc#asahi-dev · 2023-04-27 13:32
povikwe will see#asahi-dev · 2023-04-27 13:32
povikor... i am confusing myself#asahi-dev · 2023-04-27 13:32
povikthen i finally get a reply in the mailbox waiting on which timed out before#asahi-dev · 2023-04-27 13:32
povikwhich maybe clears it because of read sensitive registers#asahi-dev · 2023-04-27 13:32
povikthen i do mon.add(0x236400000, 0x4000)#asahi-dev · 2023-04-27 13:32
povikCPU_STATUS is 0x8#asahi-dev · 2023-04-27 13:31
povikit really looks like it spins because of an IRQ?#asahi-dev · 2023-04-27 13:31
povikhuh#asahi-dev · 2023-04-27 13:00
povikwe don't manage those, as evinced by the recent discovery that t6000 had them wrong in DT#asahi-dev · 2023-04-27 13:00
poviklet's see, what about the MCA IRQs?#asahi-dev · 2023-04-27 12:51
povikyeah, another take :p#asahi-dev · 2023-04-27 12:51
povikif that's of any help to you#asahi-dev · 2023-04-27 12:50
povikthey shape the signal in frequency domain#asahi-dev · 2023-04-27 12:49
povikhow sure are we about those?#asahi-dev · 2023-04-27 12:49
povikhttps://github.com/AsahiLinux/m1n1/blob/254a2acf189691b2081dc13603665e10a4e3ac0a/proxyclient/m1n1/hw/asc.py#asahi-dev · 2023-04-27 12:49
povik IRQ_NOT_PEND = 2 # guess#asahi-dev · 2023-04-27 12:48
povik FIQ_NOT_PEND = 3 # guess#asahi-dev · 2023-04-27 12:45
povikthat's the current idea on the SIO front#asahi-dev · 2023-04-27 12:45
povikcould the SIO be getting some unexpected IRQs due to bad interaction with one of our drivers?#asahi-dev · 2023-04-27 12:44
poviki guess convolution kernels are images too, over in the image processing land :)#asahi-dev · 2023-04-27 12:42
povikalyssa: it's a waveform and has a sample rate and format exactly like any audio signal would have#asahi-dev · 2023-04-27 11:35
povikbut that was with u-boot, so i am making sure it's safe to conclude power domains (even being transitionally turned off) are not to be blamed#asahi-dev · 2023-04-27 11:34
povikbefore that i tried patching the in-kernel pwrstate to never turn off domains, did a bunch of reboots, and saw them too#asahi-dev · 2023-04-27 11:34
povikyeah, i ran the full-os install under hypervisor, skipping over u-boot, and still saw them#asahi-dev · 2023-04-27 11:23
povikthis is to tie the hypothetical loose ends of an experiment i did#asahi-dev · 2023-04-27 11:22
povikyeah, me too#asahi-dev · 2023-04-27 11:18
povikhm#asahi-dev · 2023-04-27 11:18
povikif explicitly invoked from a device driver, right?