This raises CONFIG_RCU_CPU_STALL_TIMEOUT from 21s to 60s and
CONFIG_RCU_EXP_CPU_STALL_TIMEOUT from 20ms to 60ms (the value `0` means
that it is set to the regular stall timeout divided by 1000).
The original values caused frequent kernel logs of
"rcu_preempt detected expedited stalls" while running CPU-intensive stuff
like waydroid and caused such processes or others like Phosh to crash.
The kernel defaults are 21 / 0 (21s / 21ms), though the config docs note that
the latter is "normally 20ms on Android devices". For what it's worth,
Debian 12's amd64 kernel uses the kernel defaults of 21 / 0 (21s / 21ms),
while OpenSUSE TW's amd64 kernel uses 60 / 0 (60s / 60ms) just like
this commit.
Ref: #2980
PulseAudio crashes on startup and the fix [^1] has been sitting unmerged for
many weeks despite pings. PipeWire works fine, so this change forces it to
be used.
PW works fine for Phosh and SXMO at least (with one exception [^2]),
and long term we want to switch from PA to PW anyway, so this just
hastens that switch.
Fixes#2968
---
[^1]: https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/68586
[^2]: On Phosh, using pw-pulse causes the volume slider in
the top pulldown panel to not work sometimes. This will likely be fixed in
the systemd branch once pw-pulse is started via socket activation,
since it is likely a race between phosh connecting to the pw-pulse socket
and pw-pulse binding it.
This new daemon implements the same D-Bus API as callaudiod
in aports community/callaudiod, but works with the new audio profiles
generated by PulseAudio v17 and PipeWire v1.2 for pine64-pinephone
that the aports package's daemon does not.
Installs that use pipewire-pulse for audio (eg SXMO) will automatically start
having working call audio with this update. Installs that use pulseaudio
have to wait for pulseaudio's crash on startup to be fixed first
( https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/68586 ).
Ref: #2968
[ci:skip-build] already built successfully in CI
With the latest kernel bump the config was updated based on the output
of:
make ARCH="$_carch" CC="${CC:-gcc}" defconfig KBUILD_DEFCONFIG=librem5_defconfig
but there are other changes that pop up when running
pmbootstrap kconfig edit linux-purism-librem5
This commit updates the config so that future config empty edits don't
generate any changes.
This is still pretty in sync with the Purism config, besides passing the
`kconfig check` and disablling SELinux as that causes boot failures.
Signed-off-by: Alistair Francis <alistair@alistair23.me>
[ci:skip-build]: already built successfully in CI
This gets Bluetooth working for the Sparklan card.
btattach doesn't work on the Librem5 at the moment (see [1]) so even
though it's deprecated let's use hciattach to get things working.
1: https://source.puri.sm/Librem5/librem5-base/-/issues/76
Signed-off-by: Alistair Francis <alistair@alistair23.me>
[ci:skip-build] already built successfully in CI
This cmdline argument is now a no-op. Drop it everywhere and add a new
CI check to enforce this.
Adjust the deprecated "minimal" initramfs variant to use
pmos.debug-shell to enable logging instead. It doesn't /support/
dropping to a shell, so the variable takes on a different meaning. But
that's ok since it's at least more consistent.
[ci:ignore-count]
[ci:skip-vercheck]
Signed-off-by: Caleb Connolly <caleb@postmarketos.org>
Tweaked-by: Oliver Smith <ollieparanoid@postmarketos.org>
[ci:skip-build]: kernel builds are very likely to go through now
I'm not able to give this device the attention it deserves.
This also removes me from maintaining fw and kernel packages used by the
L5, and adds the current L5 maintainers to those packages.
[ci:skip-vercheck]
[ci:skip-build]: already built successfully in CI
Bump to the 6.6.32 release, including updating the config with the
latest changes from Purism.
Signed-off-by: Alistair Francis <alistair@alistair23.me>
[ci:skip-build]: already built successfully in CI
At some point the driver started using symlinks w/ machine IDs for
finding/loading fw, and for some reason I thought it was a good idea to
not install those symlinks.
Fixes this failure when loading the brcmfmac driver:
brcmfmac mmc1:0001:1: Direct firmware load for brcm/brcmfmac43752-sdio.purism,librem5r4.bin failed with error -2
[ci:skip-build]: already built successfully in CI
Copied from qemu-amd64. Enabling support for EFI boot allows me/others
to easily test pmOS EFI boot loader and config changes. Legacy BIOS boot
will still work, see 22c602096 for details.
[ci:skip-build]: already built successfully in CI
- 0003-dts-pinetab-add-missing-bma223-ohci1.patch
The original patch added a node for BMA223 and a node for ohci1.
The former is already present in upstream, so the new patch
0003-dts-pinetab-add-missing-ohci1.patch only adds the latter.
- 0004-arm64-dts-allwinner-Add-bluetooth-node-to-the-PineTa.patch
The patch is now present in upstream so it has been removed.
However it's worth noting that upstream's node has the wake GPIOs
reversed from the original patch. The original patch added:
device-wake-gpios = <&r_pio 0 6 GPIO_ACTIVE_LOW>; /* PL6 */
host-wake-gpios = <&r_pio 0 5 GPIO_ACTIVE_HIGH>; /* PL5 */
... while upstream has:
device-wake-gpios = <&r_pio 0 5 GPIO_ACTIVE_LOW>; /* PL5 */
host-wake-gpios = <&r_pio 0 6 GPIO_ACTIVE_HIGH>; /* PL6 */
This has not been tested on a PineTab.
- 0005-dts-pinetab-make-audio-routing-consistent-with-pinep.patch
The first hunk of the patch is in upstream and has been removed.
- 0013-Revert-usb-quirks-Add-USB_QUIRK_RESET-for-Quectel-EG25G-Modem.patch
This patch reverted upstream commit that was present until v6.7
( f9e3e7ea4e )
but is not present in v6.8+, so the patch has been removed.
- 0014-usb-serial-option-add-reset-resume-callback-for-WWAN.patch
The patch is now present in upstream so it has been removed.
- 0016-Define-orientation-and-rotation-for-PinePhone-camera.patch
The patch is now present in upstream so it has been removed.
---
The issues found with the previous upgrade to 6.8.4
( 646e0b50df ), ie reliably receiving calls
in suspend ( #2734 ) and the headset / headphones detection being broken
because of the jack detection changes ( #2738 ) have been fixed.
[ci:skip-build]: already built successfully in CI
This reverts commit 1e3d484a64
"device-pine64-pinephone: update UCM config for new headphone jack name (MR 5022)"
[ci:skip-vercheck]
[ci:skip-build]: already built successfully in CI
Kernel 6.8.4 has multiple issues on the pine64-pinephone:
- Incoming calls don't work reliably when the phone is suspended. ( #2734 )
- Audio routing has changed such that it's seemingly not possible to use
headphones with the builtin mic. ( #2738 )
Kernel 6.7.9 does not have these problems. This commit reverts to that for now
until they're figured out on the new kernel.
The previous name stopped being valid with the kernel 6.8 upgrade. This caused
ALSA to think a headset was connected, which in turn caused callaudiod to
select the Headphones port in the PA sink instead of Earpiece / Speaker.
I missed this in my own kernel upgrade testing because I use
a custom callaudiod that doesn't care about the headphones port.
I now confirmed the issue with regular callaudiod and that this change
fixes it.
Co-authored-by: Fiona Klute <fiona.klute@gmx.de>
I just noticed that the virt kernel doesn't have mouse and network
working anymore in QEMU and Alpine edge. It did work with v23.12.
Everything seems to work with the LTS kernel though, and it doesn't
change often, so recommend that for now.
[ci:skip-build]: already built successfully in CI
- 0003-dts-pinetab-add-missing-bma223-ohci1.patch
The original patch added a node for BMA223 and a node for ohci1.
The former is already present in upstream, so the new patch
0003-dts-pinetab-add-missing-ohci1.patch only adds the latter.
- 0004-arm64-dts-allwinner-Add-bluetooth-node-to-the-PineTa.patch
The patch is now present in upstream so it has been removed.
However it's worth noting that upstream's node has the wake GPIOs
reversed from the original patch. The original patch added:
device-wake-gpios = <&r_pio 0 6 GPIO_ACTIVE_LOW>; /* PL6 */
host-wake-gpios = <&r_pio 0 5 GPIO_ACTIVE_HIGH>; /* PL5 */
... while upstream has:
device-wake-gpios = <&r_pio 0 5 GPIO_ACTIVE_LOW>; /* PL5 */
host-wake-gpios = <&r_pio 0 6 GPIO_ACTIVE_HIGH>; /* PL6 */
This has not been tested on a PineTab.
- 0005-dts-pinetab-make-audio-routing-consistent-with-pinep.patch
The first hunk of the patch is in upstream and has been removed.
- 0013-Revert-usb-quirks-Add-USB_QUIRK_RESET-for-Quectel-EG25G-Modem.patch
This patch reverted upstream commit that was present until v6.7
( f9e3e7ea4e )
but is not present in v6.8, so the patch has been removed.
- 0014-usb-serial-option-add-reset-resume-callback-for-WWAN.patch
The patch is now present in upstream so it has been removed.
eg25-manager is supposed to watch for the modem going away and recover it
automatically, but it doesn't seem to actually do anything.
Manually restarting eg25-manager in this situation does recover the modem.
This upstream issue https://gitlab.com/mobian1/eg25-manager/-/issues/47
is probably the same, and as that issue says, manually restarting eg25-manager
in this situation does recover the modem.
So this change adds a udev rule to automatically restart eg25-manager when
the modem dies. Common triggers for the modem dying are:
1. When the phone wakes up from suspend.
2. When an incoming call is disconnected, especially if the disconnect happens
very quickly after the call started, such as what happens when
phosh-antispam identified the call as spam.
3. When plugging or unplugging the phone from USB power, and when plugging
or unplugging other USB devices like keyboard from a connected dock.
udev shouldn't be used for spawning long-lasting processes,
but `rc-service start` will parent the process to the service manager and exit,
so that's fine. It does take a few seconds during which udev is blocked and
can't process other events, which udev also recommends against,
but in this case it's not a big deal IMO.
Fixes#2641
[ci:skip-build]: already built successfully in CI
aports' community/crust package generates a crust-pinephone subpackage
that is equivalent to pmaports' current main/crust package, that is,
the crust-pinephone subpackage contains the SCP firmware binary needed for
the PinePhone. The aports crust package itself only contains a LICENSE file.
Furthermore, aports' package updated to v0.6 recently, so apk would pick it
over pmaports' v0.5 package, which would cause the build to fail because of
the missing SCP firmware binary.
This change fixes the build of u-boot-pinephone by switching the dependency to
the aports crust-pinephone package.
