Commit graph

743 commits

Author SHA1 Message Date
Asahi Lina
5a389fe87b kboot_gpu: Add another pwcalc difference for T602x
This doesn't make any sense... M2 Ultra should be 2x M2 Max, so there's
no reason for this constant to be different.

I get the feeling Apple messed this one up too...

Also clean up some other constants.

Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-08-17 19:46:07 +09:00
Asahi Lina
db49377e5a kboot_gpu: Fixes for >1 die t602x
Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-08-17 19:46:07 +09:00
Hector Martin
8e3016e323 pcie: T602X APCIE-GE support
Maybe maybe?

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-08-07 21:05:25 +09:00
Hector Martin
721107aa0a kboot_gpu: Fixes for T6022
This is still blind. Lina send help.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-08-07 19:27:47 +09:00
Hector Martin
bb0e9ace57 display: Make M2+ desktop disable hack more generic
This should work on the Mac Pro.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-08-07 19:27:47 +09:00
Hector Martin
9b73944684 Initial t6022 support
GPU stuff is a wild guess.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-08-07 19:27:47 +09:00
Asahi Lina
28708461fc kboot_gpu: Fix core count for T8112
Oops...

Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-08-02 21:10:41 +09:00
Asahi Lina
e322312c94 kboot_gpu: Do not try to set cs/afr pstates on <G14X
Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-08-02 15:11:32 +09:00
Asahi Lina
86d75d4dba kboot: Add 13.5 firmware compats
Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-07-20 11:54:11 +09:00
Asahi Lina
f58e213529 kboot_gpu: Power calc fixes for t6020
Looks like this is different from t6021, and now I wonder if we need to
recheck t6000...

Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-07-20 11:54:11 +09:00
Asahi Lina
dfecc20114 firmware: Add 13.5b4 and 13.5 versions
Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-07-19 13:12:24 +09:00
Asahi Lina
b7389c59be kboot_gpu: Add T602x to the support list
Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-06-27 18:35:01 +09:00
Asahi Lina
b35f0d7cf2 kboot_gpu: Initial T602x support
Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-06-27 18:35:00 +09:00
Asahi Lina
af85202b58 math: Add powf()
Signed-off-by: Asahi Lina <lina@asahilina.net>
2023-06-27 18:35:00 +09:00
Hector Martin
88df8a6335 cpufreq: Initialize a bunch more stuff
At least some of this fixes hangs when overheating on j413. We still
don't know exactly what most of this is, but let's try to get closet to
the init that macOS does.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-06-04 20:10:39 +09:00
Hector Martin
5cd985b4db cpufreq: Redefine the dvmr flag as pcluster
We're going to use this later.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-06-04 17:03:58 +09:00
Hector Martin
7ce414224c pmgr: Add a helper to test for pmgr features in the ADT
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-06-04 17:02:48 +09:00
Hector Martin
1f154402a4 cpufreq: Drop the DVMR stuff
This is enabled by iBoot in all versions we support.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-06-04 17:02:44 +09:00
Hector Martin
a0d2ed20be payload: Do cpufreq init before smp_start_secondaries()
This also requires exporting this to the proxy so linux.py can use it.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-06-04 15:56:42 +09:00
Hector Martin
4124485265 hv: Make breadcrumbs per-CPU and introduce panic handler
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-06-03 12:47:39 +09:00
Hector Martin
2c531aa7ae hv_exc: Remove PMC0 CNT enable shadowing from fast sysreg path
This is outside the hv_exc_entry/hv_exc_exit bracketing now, so modify
the register directly. We only stop the PMCs when we go into the
slowpath.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-05-22 00:27:08 +09:00
Janne Grunau
076217a802 display: Skip init on j473/474s while it's not supported
Signed-off-by: Janne Grunau <j@jannau.net>
2023-05-16 18:23:44 +09:00
Hector Martin
48db941199 cpu_regs.h: Typo fix
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-05-05 19:20:48 +09:00
Hector Martin
50c808df2c pmgr: Power up parents before children
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-05-05 12:39:26 +09:00
Hector Martin
59e0032ba4 cpu_regs,apple_regs.json: More registers
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-05-05 04:46:11 +09:00
Hector Martin
895488aebb chickens: Change IRQ mode from 0 to 2
0 seems to misbehave when IRQs fire in EL1. 2 works as intended.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-05-05 00:32:42 +09:00
Hector Martin
abe4bb04c5 exception: Improve IRQ/FIQ log messages
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-05-05 00:31:05 +09:00
Hector Martin
ebd4613f61 proxy: Add smp_call[_sync]_el1 variants
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-05-05 00:31:05 +09:00
Janne Grunau
254a2acf18 dcp_iboot: Support macOS 13.3 DCP firmware
The macOS 13.3 DCP firmware extended swap_set_layer_cmd in an
incompatible way. Let dcp_ib_swap_set_layer() send the matching struct
based on the firmware version.

Signed-off-by: Janne Grunau <j@jannau.net>
2023-04-27 04:11:18 +09:00
Janne Grunau
01ea1ae8df afk: Accept message type TYPE_REPLY in start_interface
The macOS 13.3 DCP firmware uses TYPE_REPLY instead of TYPE_NOTIFY for
the disp0-service interface init message.

Signed-off-by: Janne Grunau <j@jannau.net>
2023-04-27 04:11:18 +09:00
Janne Grunau
cd8e9d6676 payload: Add a custom header with a payload size field for initramfs'
This add supports for concatenated (compressed) cpio archives and relies
on the kernel to decompress the initramfs.

Signed-off-by: Janne Grunau <j@jannau.net>
2023-04-27 04:09:49 +09:00
Hector Martin
54f44be7d6 hv_exc: Add hv_add_time() function & proxy thunk
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-26 00:51:51 +09:00
Hector Martin
9602fb1859 hv_exc: Add a fast path for sysregs that can be handled without locking
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-25 20:08:59 +09:00
Hector Martin
b8b76f34f4 cpufreq: Correct fixup for M2 devices
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-19 04:44:58 +09:00
Hector Martin
bd63b119ec startup: Direct primary CPU RVBAR entry to m1n1 main
This definitely isn't going to work yet, but it's better than ending up
in the secondary loop if it ever works.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-19 03:21:58 +09:00
Hector Martin
06b732e788 smp: Add smp_stop_secondaries()
Doesn't work properly yet since we can't restart them...

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-19 03:21:58 +09:00
Hector Martin
3288548faa utils: Add cpu_sleep() function
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-19 03:21:58 +09:00
Hector Martin
121d20cddf cpu_regs: Make fields usable from assembler
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-19 02:37:02 +09:00
Hector Martin
4638632657 hv: Implement emulated guest CPU shutdown
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-19 01:21:33 +09:00
Hector Martin
cae2059c06 hv: Add a rendezvous timeout and panic on failure
Prints the CPUs that failed rendezvous, which should help debug what
caused them to go away.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 23:50:11 +09:00
Hector Martin
587da648a3 hv_exc: Be a bit more paranoid handling IPIs
Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 23:38:40 +09:00
Hector Martin
b0c3f6d7ae hv: Implement ECV on M2 and later chips
This lets us eliminate HV timer ticks on secondary CPUs, which should
significantly improve the scalability of this whole thing.

(We leave them at 1Hz just in case, since that much won't hurt)

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 23:37:18 +09:00
Hector Martin
a75afff83b hv_exc: Pass through dangerous CYC_OVRD writes to Python
This makes it easier to experiment with the sleep/WFI stuff.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 23:36:55 +09:00
Hector Martin
bdf54ba7e1 usb_dwc3: Populate serial number with the actual serial number
This will be useful for CI and similar use cases, to distinguish
different machines. Also fix spurious NUL-termination of the descriptor
strings while I'm here.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 22:07:21 +09:00
Hector Martin
cf7da86eba cpufreq: Fix m1n1 stage 1 bugs
Old stage1 did cpufreq init and didn't do it right. Undo the damage in
stage2.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 21:55:12 +09:00
Hector Martin
9b944c5f56 payload: Pass m1n1 stage 1 version when chainloading
This might come in handy in the future.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 21:55:12 +09:00
Hector Martin
5a4e30a98d hv: Restore CYC_OVRD WFI mode and handle FIQ disable
Put the WFI mode back how macOS expects it on boot, since it won't set
it itself. Then, also trap attempts by macOS to disable FIQs. This
happens in the sleep/core off path. It is not sufficient to make all
that work properly, but it's a start.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 21:55:12 +09:00
Hector Martin
8cf0df0553 utils_asm: Save/restore CYC_OVRD register
This keeps the previous WFI setting instead of unconditionally flipping
to 3 then back to 2.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 21:55:12 +09:00
Hector Martin
b067f26971 cpu_regs.h: Add ACC_OVRD register
Not used yet, but might be in the future.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 21:55:12 +09:00
Hector Martin
cbd9b7b9e1 main, kboot: Move tunables/cpufreq/clk from main to kboot
I already screwed something up in cpufreq, and it makes it difficult to
understand what happened if it happens in stage1 before the proxy runs,
and difficult to undo in stage2.

Let's move all this stuff to kboot, since it isn't necessary until then
(and this will also avoid doing unnecessary init for macOS under the
HV). If we need it in proxy scripts, we can add proxy commands for this
later.

Signed-off-by: Hector Martin <marcan@marcan.st>
2023-04-18 21:55:12 +09:00