Re: [PATCH 4.4 16/30] [media] pvrusb2: reduce stack usage pvr2_eeprom_analyze()

From: Arnd Bergmann
Date: Mon Jul 03 2017 - 03:48:03 EST

On Mon, Jul 3, 2017 at 9:31 AM, Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> On Thu, Jun 29, 2017 at 05:15:17PM +0100, Ben Hutchings wrote:
>> On Mon, 2017-06-19 at 23:20 +0800, Greg Kroah-Hartman wrote:
>> > 4.4-stable review patch. If anyone has any objections, please let me know.
>> >
>> > From: Arnd Bergmann <arnd@xxxxxxxx>
>> >
>> > commit 6830733d53a4517588e56227b9c8538633f0c496 upstream.
>> >
>> > The driver uses a relatively large data structure on the stack, which
>> > showed up on my radar as we get a warning with the "latent entropy"
>> > GCC plugin:
>> >
>> > drivers/media/usb/pvrusb2/pvrusb2-eeprom.c:153:1: error: the frame size of 1376 bytes is larger than 1152 bytes [-Werror=frame-larger-than=]
>> >
>> > The warning is usually hidden as we raise the warning limit to 2048
>> > when the plugin is enabled, but I'd like to lower that again in the
>> > future, and making this function smaller helps to do that without
>> > build regressions.
>> >
>> > Further analysis shows that putting an 'i2c_client' structure on
>> > the stack is not really supported, as the embedded 'struct device'
>> > is not initialized here, and we are only saved by the fact that
>> > the function that is called here does not use the pointer at all.
>> [...]
>> That is not true in 4.4-stable. This commit depends on:
>> commit 6037b3ca28f4258d913dbe77248fd77827702ae3
>> Author: Mauro Carvalho Chehab <mchehab@xxxxxxxxxxxxxxxx>
>> Date: Wed Nov 16 14:21:48 2016 -0200
>> [media] tveeprom: print log messages using pr_foo()
> It does? I don't understand how the two are connected. Removing
> i2c_client off of the stack is a good thing. Ah, I see how the pointer
> is used in tveeprom_hauppauge_analog(), but this shouldn't matter here,
> right?

My reading of the two patches is that we actually need at least one
of them to avoid interpreting uninitialized dev->class/bus: With just my
6830733d53a patch, we replace the uninitialized data with a NULL
pointer, which is handled gracefully in __dev_printk(), while the
6037b3ca28 patch by itself will avoid using the 'dev' pointer completely,
and give a saner output (no "(NULL device)" string or worse).

I think we probably want both of them backported to 4.4, but I don't see a
dependency between them.