| Message ID | 1448884205-17989-1-git-send-email-martin.wilck@ts.fujitsu.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hello, On Mon, Nov 30, 2015 at 12:50:05PM +0100, martin.wilck@ts.fujitsu.com wrote: > From: Martin Wilck <Martin.Wilck@ts.fujitsu.com> > > Since b8b2c7d845d5, 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 b8b2c7d845d5, 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 b8b2c7d845d5. > > 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). I'm a bit uncertain if I should be happy with this wording or not. While b8b2c7d845d5 has a bug which made the tpm_tis driver provoke a panic and which needs fixing, this doesn't mean that the tpm_tis driver was right. So "This may cause problems later ..." isn't a correct justification because even in the presence of the bug introduced by b8b2c7d845d5, the tpm_tis driver must not assume that it's device is bound on return of platform_device_register_simple. It's just that the bug in b8b2c7d845d5 made a wrong assumption of the tpm_tis driver obvious, and now both need fixing. If this were my patch, I'd not talk about tpm_tis at all. b8b2c7d845d5 broke binding for platform drivers with no probe function in certain situations. This is corrected here. > 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. > > v2: fixed style issues, rephrased commit message. This must go after the triple-dash below to not be included in the commit. > Fixes: b8b2c7d845d5 ("base/platform: assert that dev_pm_domain callbacks are called unconditionally") > Signed-off-by: Martin Wilck <Martin.Wilck@ts.fujitsu.com> > --- Best regards Uwe
diff --git a/drivers/base/platform.c b/drivers/base/platform.c index 1dd6d3b..176b59f 100644 --- a/drivers/base/platform.c +++ b/drivers/base/platform.c @@ -513,10 +513,15 @@ static int platform_drv_probe(struct device *_dev) return ret; ret = dev_pm_domain_attach(_dev, true); - if (ret != -EPROBE_DEFER && drv->probe) { - ret = drv->probe(dev); - if (ret) - dev_pm_domain_detach(_dev, true); + if (ret != -EPROBE_DEFER) { + if (drv->probe) { + ret = drv->probe(dev); + if (ret) + dev_pm_domain_detach(_dev, true); + } else { + /* don't fail if just dev_pm_domain_attach failed */ + ret = 0; + } } if (drv->prevent_deferred_probe && ret == -EPROBE_DEFER) {