From mboxrd@z Thu Jan 1 00:00:00 1970 From: Scott Wood Subject: Re: RFC: (re-)binding the VFIO platform driver to a platform device Date: Wed, 9 Oct 2013 15:03:19 -0500 Message-ID: <1381348999.7979.360.camel@snotra.buserror.net> References: <1380738758.12932.43.camel@snotra.buserror.net> <20131002184330.GC5108@cbox> <20131002203735.GA10871@kroah.com> <1380748121.12932.89.camel@snotra.buserror.net> <20131002211631.GA11914@kroah.com> <1380749715.12932.109.camel@snotra.buserror.net> <20131002234009.GA27714@kroah.com> <1380825207.12932.151.camel@snotra.buserror.net> <20131003185434.GA26123@kroah.com> <1380827494.12932.161.camel@snotra.buserror.net> <20131003203226.GB27336@kroah.com> <9F6FE96B71CF29479FF1CDC8046E15036DDA62@039-SN1MPN1-002.039d.mgd.msft.net> <1381346507.7979.334.camel@snotra.buserror.net> <9F6FE96B71CF29479FF1CDC8046E15036DDAB4@039-SN1MPN1-002.039d.mgd.msft.net> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: Wood Scott-B07421 , Kim Phillips , Christoffer Dall , Alex Williamson , "linux-kernel@vger.kernel.org" , "a.motakis@virtualopensystems.com" , "agraf@suse.de" , Sethi Varun-B16395 , Bhushan Bharat-R65777 , "peter.maydell@linaro.org" , "santosh.shukla@linaro.org" , "kvm@vger.kernel.org" , "gregkh@linuxfoundation.org" To: Yoder Stuart-B08248 Return-path: In-Reply-To: <9F6FE96B71CF29479FF1CDC8046E15036DDAB4@039-SN1MPN1-002.039d.mgd.msft.net> Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On Wed, 2013-10-09 at 14:44 -0500, Yoder Stuart-B08248 wrote: > > > -----Original Message----- > > From: Wood Scott-B07421 > > Sent: Wednesday, October 09, 2013 2:22 PM > > To: Yoder Stuart-B08248 > > Cc: Wood Scott-B07421; Kim Phillips; Christoffer Dall; Alex Williamson; > > linux-kernel@vger.kernel.org; a.motakis@virtualopensystems.com; > > agraf@suse.de; Sethi Varun-B16395; Bhushan Bharat-R65777; > > peter.maydell@linaro.org; santosh.shukla@linaro.org; kvm@vger.kernel.org; > > gregkh@linuxfoundation.org > > Subject: Re: RFC: (re-)binding the VFIO platform driver to a platform > > device > > > > On Wed, 2013-10-09 at 14:02 -0500, Yoder Stuart-B08248 wrote: > > > Have been thinking about this issue some more. As Scott mentioned, > > > 'wildcard' matching for a driver can be fairly done in the platform > > > bus driver. We could add a new flag to the platform driver struct: > > > > > > diff --git a/drivers/base/platform.c b/drivers/base/platform.c > > > index 4f8bef3..4d6cf14 100644 > > > --- a/drivers/base/platform.c > > > +++ b/drivers/base/platform.c > > > @@ -727,6 +727,10 @@ static int platform_match(struct device *dev, > > struct device_driver *drv) > > > struct platform_device *pdev = to_platform_device(dev); > > > struct platform_driver *pdrv = to_platform_driver(drv); > > > > > > + /* the driver matches any device */ > > > + if (pdrv->match_any) > > > + return 1; > > > + > > > /* Attempt an OF style match first */ > > > if (of_driver_match_device(dev, drv)) > > > return 1; > > > > > > However, the more problematic issue is that a bus driver has no way to > > > differentiate from an explicit bind request via sysfs and a bind that > > > happened through bus probing. > > > > Again, I think the wildcard match should be orthogonal to "don't bind by > > default" as far as the mechanism goes. > > > > There's already a "bool suppress_bind_attrs" to prevent sysfs > > bind/unbind. I suggested a similar flag to mean the oppsosite -- bind > > *only* through sysfs. Greg KH was skeptical and wanted to see a patch > > before any further discussion. > > Ah, think I understand now...yes that works as well, and would be > less intrustive. So are you writing a patch? :) I've been meaning to since the previous round of discussion, but I've been busy. Would someone else be able to test it in the context of using it for VFIO? > It would be something like this, right? > > diff --git a/drivers/base/dd.c b/drivers/base/dd.c > index 35fa368..c9a61ea 100644 > --- a/drivers/base/dd.c > +++ b/drivers/base/dd.c > @@ -389,7 +389,7 @@ static int __device_attach(struct device_driver *drv, void *data) > { > struct device *dev = data; > > - if (!driver_match_device(drv, dev)) > + if (!drv->explicit_bind_only && !driver_match_device(drv, dev)) > return 0; if (drv->explicit_bind_only || !driver_match_device(drv, dev)) return 0; > return driver_probe_device(drv, dev); > @@ -450,7 +450,7 @@ static int __driver_attach(struct device *dev, void *data) > * is an error. > */ > > - if (!driver_match_device(drv, dev)) > + if (!drv->explicit_bind_only && !driver_match_device(drv, dev)) > return 0; Likewise -- or error out earlier in driver_attach(). Otherwise, that looks about right, for the driver side (though driver_attach could error out earlier rather than testing it inside the loop). The other half of fixing the raciness is to ensure that the device doesn't get bound back to a non-VFIO driver (e.g. due to a module load or new_id). The solution I proposed for that was a similar explicit-bind-only flag for a device, that the user sets through sysfs prior to unbinding. This would also be useful in non-VFIO contexts to simply say "I don't want to use this device at all". -Scott