]> git.baikalelectronics.ru Git - kernel.git/commit
base/platform: Fix platform drivers with no probe callback
authorMartin Wilck <Martin.Wilck@ts.fujitsu.com>
Mon, 30 Nov 2015 11:50:05 +0000 (12:50 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 26 Jan 2016 16:47:25 +0000 (08:47 -0800)
commitd6a0a1246999ac2320f9f4fbe246b0fd29d039a1
tree0bfc4b5140547e81b9d126b7e1573d4d619d662c
parentd7e15052ecc9ac4f938d1d7432c1d120a2711a7e
base/platform: Fix platform drivers with no probe callback

Since 0cf72c42601b, platform_drv_probe() is called for all platform
devices. If drv->probe is NULL, and dev_pm_domain_attach() fails,
platform_drv_probe() will return the error code from dev_pm_domain_attach().

This causes real_probe() to enter the "probe_failed" path and set
dev->driver to NULL. Before 0cf72c42601b, real_probe() would assume
success if both dev->bus->probe and drv->probe were missing. As a result,
a device and driver could be "bound" together just by matching their names;
this doesn't work any more after 0cf72c42601b.

This may cause problems later for certain usage of platform_driver_register()
and platform_device_register_simple(). I observed a panic while loading
the tpm_tis driver with parameter "force=1" (i.e. registering tpm_tis as
a platform driver), because tpm_tis_init's assumption that the device
returned by platform_device_register_simple() was bound didn't hold any more
(tpmm_chip_alloc() dereferences chip->pdev->driver, causing panic).

This patch restores the previous (4.3.0 and earlier) behavior of
platform_drv_probe() in the case when the associated platform driver has
no "probe" function.

Fixes: 0cf72c42601b ("base/platform: assert that dev_pm_domain callbacks are called unconditionally")
Signed-off-by: Martin Wilck <Martin.Wilck@ts.fujitsu.com>
Cc: stable <stable@vger.kernel.org> # 4.4
Cc: Martin Fuzzey <mfuzzey@parkeon.com>
Acked-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/platform.c