#asahi-dev · 2022-08-22 14:42
DmitrySharshakov[m]I guess MacBook Air's also need some sort of capping functions, but as those are not present now (at least weren't there yesterday) driver will refuse to probe. Quite a smart solution to protect everything :)#asahi-dev · 2022-08-22 10:18
DmitrySharshakov[m]And for high-value cases a config option can be set to store the key on external media which can be wiped better#asahi-dev · 2022-08-22 10:16
DmitrySharshakov[m]Well yes. But the risk is still low enough for casual users and attacker will have to desolder NAND to take a look at lost fragments of a broken cryptocontainer (which first has to have a compromised password or cryptography)#asahi-dev · 2022-08-22 10:14
DmitrySharshakov[m]and when a better approach is ready (SEP-backed wrapping), then you free up some space in the container, do a trip through recoveryOS and enroll the new m1n1 then wrap its key in a new fashion#asahi-dev · 2022-08-22 10:11
DmitrySharshakov[m]At least that 'first approach' I suggested gives the same degree of security as custom sb keys on PC platforms. Btw you could also store the key on external media or do a secure erasure of the key material when changing the password#asahi-dev · 2022-08-22 10:04
DmitrySharshakov[m]idk#asahi-dev · 2022-08-22 10:03
DmitrySharshakov[m]marcan: Couldn't we do some sort of secure boot (with a password-protected key) to improve quality now and then replace it by SEP-backed hardware security?#asahi-dev · 2022-08-22 10:02
DmitrySharshakov[m]`it just means we need to support APFS and SEP enough to mount FV-enabled volumes` are you sure SEP container stuff doesn't restrict that? Or it'll be like encryption of an external drive?#asahi-dev · 2022-08-22 10:01
DmitrySharshakov[m]and reduced security is for those who want custom builds, they have a password-protected key story#asahi-dev · 2022-08-22 10:00
DmitrySharshakov[m]full security just enforces use of distro-provided kernel+initrd, will require support from distro#asahi-dev · 2022-08-22 09:59
DmitrySharshakov[m]Hmm, so we can have separate profiles (like full and reduced security)#asahi-dev · 2022-08-22 09:56
DmitrySharshakov[m]marcan: What of those exactly?#asahi-dev · 2022-08-22 09:56
DmitrySharshakov[m]Can we run a custom Rust binary in recoveryOS for the sake of simplicity?#asahi-dev · 2022-08-22 09:55
DmitrySharshakov[m]Or some keyslot madness can be implemented, so in case multiple users are present they all can unlock the keyfile#asahi-dev · 2022-08-22 09:54
DmitrySharshakov[m]This way we get a machine-owner-password-encrypted private key for signing. On updates machine owner has to confirm the action by entering their password#asahi-dev · 2022-08-22 09:53
DmitrySharshakov[m]Or gpg can be bundled into the part 2 of the installer to accomplish the job#asahi-dev · 2022-08-22 09:52
DmitrySharshakov[m]We can use AES + password-derived key I guess. recoveryOS part of the installer generates the keypair, burns pubkey into m1n1-1, and encrypts private one to be stored in the ESP#asahi-dev · 2022-08-22 09:51
DmitrySharshakov[m]marcan: Was going to write my ideas#asahi-dev · 2022-08-22 09:51
DmitrySharshakov[m]So probably for now we need that bootpicker in m1n1 and a more generic package for handling all the userspace stuff. Well, an utility like sbsign or update-grub which will weld m1n1-2 and all the subsequent parts together, then go to ESP, ask user's password to decrypt the private key, sign and store things#asahi-dev · 2022-08-22 09:49
DmitrySharshakov[m]marcan: Well, like I expected. feel free to close or I can do that#asahi-dev · 2022-08-22 09:48
DmitrySharshakov[m]marcan: And then conventional bootloader is gone completely? So m1n1-1 -> choice -> m1n1-2 -> u-boot + kernel + initramfs bundle is the boot sequence?#asahi-dev · 2022-08-22 09:47
DmitrySharshakov[m]Also I've once opened a question issue on GH whether booting m1n1 test builds on emulators would be easy enough. That'd help interested people without hardware (like me huh) to do hw-independent stuff and maybe autotests and fuzzing#asahi-dev · 2022-08-22 09:46
DmitrySharshakov[m]But we want to retain the GRUB/sd-boot I guess, so not welding m1n1-2+u-boot+kernel+initrd, as if kernel update breaks something I'd like to be able to choose previous one#asahi-dev · 2022-08-22 09:45
DmitrySharshakov[m]Since emulating all the MOK dances can be quite demanding and it's easy to do something wrong implementing that spec#asahi-dev · 2022-08-22 09:44
DmitrySharshakov[m]marcan: Yes, it can be like that#asahi-dev · 2022-08-22 09:43
DmitrySharshakov[m]But quite a great security can be accomplished even without any SEP driver#asahi-dev · 2022-08-22 09:43
DmitrySharshakov[m]Well, but I guess reversing the SEP is less trivial and might take a long to be really secure and more important - reliable#asahi-dev · 2022-08-22 09:42
DmitrySharshakov[m]s/truct/trust/#asahi-dev · 2022-08-22 09:42
DmitrySharshakov[m]The root of truct is m1n1-1 baked into boot policy#asahi-dev · 2022-08-22 09:41
DmitrySharshakov[m]Probably with some U-Boot patches it can be made to support sort of true UEFI-standard secure boot#asahi-dev · 2022-08-22 09:41
DmitrySharshakov[m]Secure boot trust chain can have various implementation and that also makes a factor for distro choice: how breaking would out secure boot policy changes be#asahi-dev · 2022-08-22 09:38
DmitrySharshakov[m]marcan: I also wanted to chat a bit about some ideas about secure boot#asahi-dev · 2022-08-22 09:38
DmitrySharshakov[m]`macOS-like secureboot` - like setting up an immutable volume with dm-verity etc? Yes, that's quite a great concept#asahi-dev · 2022-08-22 09:32
DmitrySharshakov[m]So textures/other buffers have to be copied since they cannot be placed into a memory area usable for both SoC and GPU?#asahi-dev · 2022-08-22 09:30
DmitrySharshakov[m]Or what is lacking for it to work really?#asahi-dev · 2022-08-22 09:30
DmitrySharshakov[m]Can it be possible for at least compute workloads running on eGPU? I heard that the issue is like lack of GTT support in M1/M2 PCIe controller?#asahi-dev · 2022-08-22 09:28
DmitrySharshakov[m]Yes, I meant memory object related to GPU work#asahi-dev · 2022-08-22 09:28
DmitrySharshakov[m]If memory is unified then passing a GPU memory BO shouldn't be a problem#asahi-dev · 2022-08-22 09:27
DmitrySharshakov[m]marcan: it appears thunderbolt-net is a real RDMA-over-Thunderbolt. https://github.com/torvalds/linux/blob/master/drivers/net/thunderbolt.c#L593#asahi-dev · 2022-08-22 09:24
DmitrySharshakov[m]Well, yes, that's true#asahi-dev · 2022-08-22 09:24
DmitrySharshakov[m]`(dma-buf over thunderbolt anyone?)` I've seen network over TB in default Linux. I guess it used some sort of shared memory thing#asahi-dev · 2022-08-22 09:23
DmitrySharshakov[m]I have sometimes looked at it and it's great for custom builds (OBS), also pretty beginner-friendly. I use Fedora on my computers, but am looking at openSUSE MicroOS as a great distro for some future systems. Overall openSUSE Tumbleweed should be a decent alternative to Arch, as Fedora Rawhide is not a rolling distro but a testing platform#asahi-dev · 2022-08-22 09:20
DmitrySharshakov[m]Why did you drop openSUSE from the discussion?#asahi-dev · 2022-08-21 07:17
DmitrySharshakov[m]then I guess updated config from https://tpaste.us/gZ9v should work with my WP patches to do the trick with DSP with the new kernel driver. Now 100% safe on anything :)#asahi-dev · 2022-08-21 07:13
DmitrySharshakov[m]wasn't snd-soc-applesilicon named with hyphens in kernel and with underscores in PW?#asahi-dev · 2022-08-21 06:33
DmitrySharshakov[m]I'm not sure if PipeWire sees the module name as `snd_soc_macaudio` or `snd-soc-macaudio` though#asahi-dev · 2022-08-21 06:32
DmitrySharshakov[m] * Fixups for newer kernel drivers: https://tpaste.us/gZ9v#asahi-dev · 2022-08-21 06:31
DmitrySharshakov[m]Also corrected node.name for virtual sink as well as kernel driver settings#asahi-dev · 2022-08-21 06:31
DmitrySharshakov[m]Fixups for newer kernel drivers: https://gitlab.freedesktop.org/pipewire/pipewire/-/commit/3ceb3288ca1e700996e2c1817f45feeedba1f9bb#asahi-dev · 2022-08-20 19:41
DmitrySharshakov[m]Getting everything rebased on top of 6.0 etc?