macbook thunderbolt issue

[mfauvain]

Hi, having an issue with macbookpro14,1 running linux (kernel 4.16) and thunderbolt display (connected via TB3 to 2 apple adapter), if not right forum let me know. Was using a thunderbolt display with some connected devices (usb and firewire) and everything was working well (facetime camera, sound, ethernet, and all connected devices) up until a week ago when only display was recognized. Nothing was changed on the laptop, no update, nada. I've been searching a lot but no answer so far. Only display is recognized (and working, even 2 thunderbolt display daisy chained are working fine, except with no devices, no facetime, no ethernet, no sound).
only clue in dmesg is [ 27.895901] thunderbolt 0000:06:00.0: 4:c: disabled by eeprom etc..

tbtadm shows tbdisplay is authorized. tried bolt as well and tbdisplay is authorized. tried reset SMC, NVRAM no luck.
Any clue?

EDIT : for completness here is the tbtadm topology output :
Controller 0
├─ Details:
│ ├─ Name: Macintosh, Apple Inc.
│ └─ Security level: SL0 (none)
│
└─ Thunderbolt Display, Apple, Inc.
└─ Details:
├─ Route-string: 0-4
├─ Authorized: Yes
├─ In ACL: No
└─ UUID: 90ad5500-0100-0100-ffff-ffffffffffff

EDIT2 : also, booting OSX : all devices (ethernet, sound, facetime, firewire and USB) and displays shown and working

[westeri]

So on Apple system you don't need tbtadm or bolt because the firmware connects all devices automatically.

You are saying that display port works but PCIe does not? There might be some issue with the PCIe hotplug. Can you share your full dmesg when you have plugged in a device and output of 'sudo lspci -vv'?

[mfauvain]

correct. thunderbolt display (connected to tb3 to tb2 adapter) works, even 2 daisy chained display works, however none of the devices connected (or integrated) to the displays are recognized anymore. Here are the lspci -vv and dmesg
dmesg.txt
lspci.txt

[westeri]

Can you remove pci=hpbussize=15,hpmemsize=10M from the kernel command line (you should never need to pass any custom command line options to the kernel) and then retry so that you boot the system without the device connected, then plug it in once the system is up and running? Please provide dmesg from that and sudo lspci -vv (note sudo).

[mfauvain]

sure, here are the files
dmesg.txt
sudolspci.txt

[westeri]

Hmm, I don't see any errors in the dmesg that could be related. The device itself is detected just fine but it looks like PCIe is not tunneled for some reason. If you check /sys/bus/thunderbolt/devices/0-4/authorized does it contain 1?

[mfauvain]

indeed, /sys/bus/thunderbolt/devices/0-4/authorized is 1
in dmesg, lines saying disabled by eeprom are fine?

[westeri]

Yes, those are fine.

[westeri]

When it stopped working, are you sure nothing was changed like for example kernel got upgraded by the distro?

[mfauvain]

positive. I use an arch repo from archive (to have kernel 4.16, later kernels dont boot on this macbook) and everything is fixed. what I don't get is, usb key used to install arch on this machine now present the same problem. When I initialy booted form this usb key, thunderbolt display and all devices were present and working, after install (booting the installed arch) tbdisplay and devices connected to it were working. Now even when booting the USB key, tbdisplay is working but none of the connected devices are. Was something written to mac bios? if it was, how come booting OSX get the tbdisplay AND connected devices working? is bios detecting when non-mac os is running and change security level of thunderbolt?

[westeri]

macOS uses different approach to connected devices. It creates tunnels via software but when you boot Linux we start the firmware which does the tunnel creation. To me it looks like the firmware got the tunnels up and running but something prevents the PCIe hotplug event from propagating. One thing to try is to run manually (as root) echo 1 > /sys/bus/pci/rescan. That rescans the whole PCI bus. If it finds the devices then at least the PCIe tunnel is up.

[mfauvain]

I did try this before, and just did again for mental sanity and no luck. Just to make sure, sudo echo 1 > /sys/bus/pci/rescan gives permission denied so I did sudo -i and then echo 1 > /sys/bus/pci/rescan which doen't resturn anything right?

[westeri]

Right, it does not return anything but you should see something new in the dmesg and lspci output if it finds anything.

[mfauvain]

was actually looking at dmesg and have this (right after echo 1 > ...)
[ 3781.086186] pcieport 0000:05:02.0: bridge window [mem 0x00100000-0x000fffff 64bit pref] to [bus 07] add_size 200000 add_align 100000
[ 3781.086241] pcieport 0000:05:02.0: BAR 15: no space for [mem size 0x00200000 64bit pref]
[ 3781.086243] pcieport 0000:05:02.0: BAR 15: failed to assign [mem size 0x00200000 64bit pref]
[ 3781.086247] pcieport 0000:05:02.0: BAR 15: no space for [mem size 0x00200000 64bit pref]
[ 3781.086248] pcieport 0000:05:02.0: BAR 15: failed to assign [mem size 0x00200000 64bit pref]

[mfauvain]

not familiar at all with this but, the kernel params I had set after the issue appeared the first time was pci=hpbussize=10,hpmemsize=2M and looking at the dmesg I am wondering if the part add_size 200000 add_align 100000 correspond to this? is kernel param stuck somewhere?

[westeri]

It is the default the kernel uses for MMIO if nothing is specified. Those messages in dmesg are fine.

However, for some reason the PCIe tunnel is not established properly and there is no indication in the log what might be wrong :(