On Sun, Jan 31, 2021 at 8:38 AM Martin Kaiser [off-list ref] wrote:

Dear all,

Thus wrote Saravana Kannan (saravanak@google.com):

quoted

Sending again because I messed up the To/Cc for the coverletter.

quoted

This series combines two patches [1] [2] that'd conflict.

quoted

Greg,

quoted

Can you please pull this into driver-core-next?

quoted

-Saravana

quoted

[1] - https://lore.kernel.org/lkml/20210115210159.3090203-1-saravanak@google.com/ (local)
[2] - https://lore.kernel.org/lkml/20201218210750.3455872-1-saravanak@google.com/ (local)

I'm running linux-next on my hardware which is based on the imx258
chipset by Freescale/NXP.

When those two patches appeared in linux-next, my system would not boot
any more. It was stuck right after

Uncompressing Linux... done, booting the kernel.

Reverting the irq-patch made the system boot again. Still, a number of
devices like usb or nand flash controller are not found any more.
If I revert the gpio patch as well, all devices are available again.

My system's device tree is based on arch/arm/boot/dts/imx25.dtsi with
very few adaptations for my board.

I tried to play around with the new parse_interrupts() function to
figure out which device causes the boot failure. If I skip the following
device, I can boot again:

-       return of_irq_find_parent(np);
+       np_ret = of_irq_find_parent(np);
+       if (!strcmp(np->full_name, "serial@50008000")) {
+           printk(KERN_ERR "skip serial@50008000\n");
+           return NULL;
+       }
+      return np_ret;

This is uart4 of the imx258 chip, which I use as my serial console. The
imx25.dtsi device tree seems ok, we find an interrupt parent. The
problem must be in the code that processes the result of
parse_interrupts().

I tried to boot the unmodified code with qemu, simulating the imx25-pdk
device. This wouldn't boot either.

Does this ring any bells with anyone?

This series [1] has a high chance of fixing it for you if
CONFIG_MODULES is disabled in your set up. Can you give it a shot?

The real problem is that arch/arm/mach-imx/avic.c doesn't set the
OF_POPULATED flag for the "fsl,avic" node. fw_devlink uses this
information to know that this device node will never have a struct
device created for it. The proper way to do this for root IRQCHIP
nodes is to use IRQCHIP_DECLARE(). I Cc'ed you on a clean up patch for
IMX [2], can you please give [2] a shot *without* [1] and with
CONFIG_MODULES enabled? Things should boot properly with this
combination too.

Btw, for future reference, you can try enabling the logs in
device_links_check_suppliers() to see what devices are being blocked
on what supplier nodes.

[1] - https://lore.kernel.org/lkml/20210130040344.2807439-1-saravanak@google.com/ (local)
[2] - https://lore.kernel.org/lkml/20210131205654.3379661-1-saravanak@google.com/T/#u (local)

Thanks,
Saravana