On Mon, Sep 7, 2015 at 7:23 AM, Tomeu Vizoso tomeu.vizoso@collabora.com wrote:
Hello,
I have a problem with the panel on my Tegra Chromebook taking longer than expected to be ready during boot (Stéphane Marchesin reported what is basically the same issue in [0]), and have looked into ordered probing as a better way of solving this than moving nodes around in the DT or playing with initcall levels and linking order.
While reading the thread [1] that Alexander Holler started with his series to make probing order deterministic, it occurred to me that it should be possible to achieve the same by probing devices as they are referenced by other devices.
This basically reuses the information that is already implicit in the probe() implementations, saving us from refactoring existing drivers or adding information to DTBs.
During review of v1 of this series Linus Walleij suggested that it should be the device driver core to make sure that dependencies are ready before probing a device. I gave this idea a try [2] but Mark Brown pointed out to the logic duplication between the resource acquisition and dependency discovery code paths (though I think it's fairly minor).
To address that code duplication I experimented with Arnd's devm_probe [3] concept of having drivers declare their dependencies instead of acquiring them during probe, and while it worked [4], I don't think we end up winning anything when compared to just probing devices on-demand from resource getters.
One remaining objection is to the "sprinkling" of calls to of_device_probe() in the resource getters of each subsystem, but I think it's the right thing to do given that the storage of resources is currently subsystem-specific.
We could avoid the above by moving resource storage into the core, but I don't think there's a compelling case for that.
I have tested this on boards with Tegra, iMX.6, Exynos, Rockchip and OMAP SoCs, and these patches were enough to eliminate all the deferred probes (except one in PandaBoard because omap_dma_system doesn't have a firmware node as of yet).
Have submitted a branch [5] with only these patches on top of thursday's linux-next to kernelci.org and I don't see any issues that could be caused by them. For some reason it currently has more passes than the version of -next it's based on!
With this series I get the kernel to output to the panel in 0.5s, instead of 2.8s.
Regards,
Tomeu
[0] http://lists.freedesktop.org/archives/dri-devel/2014-August/066527.html
[1] https://lkml.org/lkml/2014/5/12/452
[2] https://lkml.org/lkml/2015/6/17/305
[3] http://article.gmane.org/gmane.linux.ports.arm.kernel/277689
[4] https://lkml.org/lkml/2015/7/21/441a
[5] https://git.collabora.com/cgit/user/tomeu/linux.git/log/?h=on-demand-probes-...
[6] http://kernelci.org/boot/all/job/collabora/kernel/v4.2-11902-g25d80c927f8b/
[7] http://kernelci.org/boot/all/job/next/kernel/next-20150903/
Changes in v4:
- Added bus.pre_probe callback so the probes of Primecell devices can be deferred if their device IDs cannot be yet read because of the clock driver not having probed when they are registered. Maybe this goes overboard and the matching information should be in the DT if there is one.
Seems overboard to me or at least a separate problem. Most clocks have to be setup before the driver model simply because timers depend on clocks usually.
Rob