Please consider reverting commit bb1860efc817 ("coresight: Make sysfs functional...") from v5.4.y

Please consider reverting commit bb1860efc817 ("coresight: Make sysfs functional...") from v5.4.y

[Guenter Roeck]

Hi,

I get the following build warning in v5.4.75.

drivers/hwtracing/coresight/coresight-etm-perf.c: In function 'etm_setup_aux':
drivers/hwtracing/coresight/coresight-etm-perf.c:226:37: warning:
			passing argument 1 of 'coresight_get_enabled_sink' makes pointer from integer without a cast

Actually, the warning is fatal, since the call is
	sink = coresight_get_enabled_sink(true);
However, the argument to coresight_get_enabled_sink() is now a pointer.
The parameter change was introduced with commit 8fd52a21ab57
("coresight: Make sysfs functional on topologies with per core sink").

In the upstream kernel, the call is removed with commit bb1860efc817
("coresight: etm: perf: Sink selection using sysfs is deprecated").
That commit alone would, however, likely not solve the problem.
It looks like at least two more commits would be needed.

716f5652a131 coresight: etm: perf: Fix warning caused by etm_setup_aux failure
8e264c52e1da coresight: core: Allow the coresight core driver to be built as a module
39a7661dcf65 coresight: Fix uninitialised pointer bug in etm_setup_aux()

Looking into the coresight code, I see several additional commits affecting
the sysfs interface since v5.4. I have no idea what would actually be needed
for stable code in v5.4.y, short of applying them all.

With all this in mind, I would suggest to revert commit 8fd52a21ab57
("coresight: Make sysfs functional on topologies with per core sink")
from v5.4.y, especially since it is not marked as bug fix or for stable.

Thanks,
Guenter

Re: Please consider reverting commit bb1860efc817 ("coresight: Make sysfs functional...") from v5.4.y

[Sasha Levin]

On Sat, Nov 07, 2020 at 07:05:11PM -0800, Guenter Roeck wrote:
>Hi,
>
>I get the following build warning in v5.4.75.
>
>drivers/hwtracing/coresight/coresight-etm-perf.c: In function 'etm_setup_aux':
>drivers/hwtracing/coresight/coresight-etm-perf.c:226:37: warning:
>			passing argument 1 of 'coresight_get_enabled_sink' makes pointer from integer without a cast
>
>Actually, the warning is fatal, since the call is
>	sink = coresight_get_enabled_sink(true);
>However, the argument to coresight_get_enabled_sink() is now a pointer.
>The parameter change was introduced with commit 8fd52a21ab57
>("coresight: Make sysfs functional on topologies with per core sink").
>
>In the upstream kernel, the call is removed with commit bb1860efc817
>("coresight: etm: perf: Sink selection using sysfs is deprecated").
>That commit alone would, however, likely not solve the problem.
>It looks like at least two more commits would be needed.
>
>716f5652a131 coresight: etm: perf: Fix warning caused by etm_setup_aux failure
>8e264c52e1da coresight: core: Allow the coresight core driver to be built as a module
>39a7661dcf65 coresight: Fix uninitialised pointer bug in etm_setup_aux()
>
>Looking into the coresight code, I see several additional commits affecting
>the sysfs interface since v5.4. I have no idea what would actually be needed
>for stable code in v5.4.y, short of applying them all.
>
>With all this in mind, I would suggest to revert commit 8fd52a21ab57
>("coresight: Make sysfs functional on topologies with per core sink")
>from v5.4.y, especially since it is not marked as bug fix or for stable.

I'll revert it, thanks!

-- 
Thanks,
Sasha