linux-kernel - Disabling MFD sub-devices through the device tree

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190822182624.GA2640@gerhold.net>
Date:   Thu, 22 Aug 2019 20:26:24 +0200
From:   Stephan Gerhold <stephan@...hold.net>
To:     Lee Jones <lee.jones@...aro.org>
Cc:     Linus Walleij <linus.walleij@...aro.org>,
        linux-kernel@...r.kernel.org
Subject: Disabling MFD sub-devices through the device tree

Hi,

I am looking for a way to disable a MFD sub-device through the device
tree. Setting status = "disabled" for the device node does not seem to
have any effect when mfd_add_devices() is used.

For MFD sub-devices, this was discussed before in [1].
However, as far as I can tell it was never actually fixed.
I was thinking about simply skipping creation of the platform device if
the device node is set to disabled, e.g.:

--- a/drivers/mfd/mfd-core.c
+++ b/drivers/mfd/mfd-core.c
@@ -174,6 +174,9 @@ static int mfd_add_device(struct device *parent, int id,
 	if (parent->of_node && cell->of_compatible) {
 		for_each_child_of_node(parent->of_node, np) {
 			if (of_device_is_compatible(np, cell->of_compatible)) {
+				if (!of_device_is_available(np))
+					goto fail_alias;
+
 				pdev->dev.of_node = np;
 				pdev->dev.fwnode = &np->fwnode;
 				break;

But I believe this would introduce a rather ugly bug in
mfd_remove_devices() if the first sub-device is set to disabled:
It iterates over the children devices to find the base address of the
allocated "usage count" array, which is then used to free it.
If the first sub-device is missing, it would free the wrong address.

(At the moment, the MFD core seems to be built on the assumption that
all the children devices are actually created...)

A different approach I have seen in the kernel is to add a check to
of_device_is_available() in the device drivers of the MFD sub-devices.
e.g. drivers/power/supply/axp20x_*.c all check of_device_is_available()
as first thing in their probe() method, and abort probing with -ENODEV
otherwise.

On the other hand, duplicating that check in each and every driver
that you may want to disable eventually doesn't sound like a great idea.
Especially because this is not necessary if the devices are registered
directly through the device tree.

What do you think?

Thanks,
Stephan

[1]: https://www.spinics.net/lists/arm-kernel/msg366309.html