| Message ID | 1243752113.3425.12.camel@myhost (mailing list archive) |
|---|---|
| State | Rejected |
| Headers | show |
Em Sun, 31 May 2009 14:41:52 +0800 "Figo.zhang" <figo1802@gmail.com> escreveu: > video_register_device() return zero on success, it would not return a positive integer. > > Signed-off-by: Figo.zhang <figo1802@gmail.com> > --- > drivers/media/video/ov511.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/drivers/media/video/ov511.c b/drivers/media/video/ov511.c > index 9af5532..816427e 100644 > --- a/drivers/media/video/ov511.c > +++ b/drivers/media/video/ov511.c > @@ -5851,7 +5851,7 @@ ov51x_probe(struct usb_interface *intf, const struct usb_device_id *id) > break; > > if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, > - unit_video[i]) >= 0) { > + unit_video[i]) == 0) { > break; > } > } Nack. Errors are always negative. So, any zero or positive value indicates that no error occurred. Yet, the logic for forcing ov51x to specific minor number seems broken: it will end by registering the device twice, if used. So, that part of the function needs a rewrite. I'll fix it.
Em Wed, 10 Jun 2009 22:39:51 -0300 Mauro Carvalho Chehab <mchehab@redhat.com> escreveu: > Em Sun, 31 May 2009 14:41:52 +0800 > "Figo.zhang" <figo1802@gmail.com> escreveu: > > > video_register_device() return zero on success, it would not return a positive integer. > > > > Signed-off-by: Figo.zhang <figo1802@gmail.com> > > --- > > drivers/media/video/ov511.c | 2 +- > > 1 files changed, 1 insertions(+), 1 deletions(-) > > > > diff --git a/drivers/media/video/ov511.c b/drivers/media/video/ov511.c > > index 9af5532..816427e 100644 > > --- a/drivers/media/video/ov511.c > > +++ b/drivers/media/video/ov511.c > > @@ -5851,7 +5851,7 @@ ov51x_probe(struct usb_interface *intf, const struct usb_device_id *id) > > break; > > > > if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, > > - unit_video[i]) >= 0) { > > + unit_video[i]) == 0) { > > break; > > } > > } > > Nack. > > Errors are always negative. So, any zero or positive value indicates that no error occurred. > > Yet, the logic for forcing ov51x to specific minor number seems broken: it will > end by registering the device twice, if used. > > So, that part of the function needs a rewrite. I'll fix it. > Hmm... the issue seems more complex than I've imagined... When ov511 were written, it was assumed that video_register_device(dev, v, nr) would have the following behavior: if nr = -1, it would do dynamic minor allocation; if nr >= 0, it would allocate to 'nr' minor or fail. With the original behavior. The ov511_probe registering loop is doing something like this (I did a cleanup to allow a better understand of the logic): <snip> #define OV511_MAX_UNIT_VIDEO 16 ... static int unit_video[OV511_MAX_UNIT_VIDEO]; ... module_param_array(unit_video, int, &num_uv, 0); MODULE_PARM_DESC(unit_video, "Force use of specific minor number(s). 0 is not allowed."); ... static int ov51x_probe(struct usb_interface *intf, const struct usb_device_id *id) { ... for (i = 0; i < OV511_MAX_UNIT_VIDEO; i++) { /* Minor 0 cannot be specified; assume user wants autodetect */ if (unit_video[i] == 0) break; if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, unit_video[i]) >= 0) goto register_succeded; } /* Use the next available one */ if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, -1) < 0) { err("video_register_device failed"); goto error; register_succeeded: dev_info(&intf->dev, "Device at %s registered to minor %d\n", ov->usb_path, ov->vdev->minor); </snip> So, if you probe ov511 with a list of device numbers, for example: modprobe ov511 4,1,3 And assuming that you have 5 ov511 devices, and connect they one by one, they'll gain the following device numbers (at the connection order): /dev/video4 /dev/video1 /dev/video3 /dev/video0 /dev/video2 However, changeset 9133: changeset: 9133:64aed7485a43 parent: 9073:4db9722caf4f user: Hans Verkuil <hverkuil@xs4all.nl> date: Sat Oct 04 13:36:54 2008 +0200 summary: v4l: disconnect kernel number from minor Changed the behavior video_register_device(dev, v, nr): + nr = find_next_zero_bit(video_nums[type], minor_cnt, nr == -1 ? 0 : nr); So, instead of accepting just -1 or nr, it will now do: if nr = -1, it will do dynamic minor allocation (as before); if nr >= 0, it will also do dynamic minor allocation, but starting from nr. So, the new code won't fail if nr is already allocated, but it will return the next unallocated nr. With the ov511 code, this means that you'll have, instead: /dev/video5 /dev/video6 /dev/video7 /dev/video8 /dev/video9 So, changeset 9133 actually broke the ov511 probe original behavior. In order to restore the original behavior, the probe logic should be replaced by something else (like the approach taken by em28xx). Also, changeset 9133 may also cause other side effects on other drivers that were expecting the original behavior. To make things worse, the function comment doesn't properly explain the current behavior. --- Figo, Since we are in the middle of a merge window, it is unlikely that I'll have enough time to properly fix it right now (since my priority right now is to merge the existing patches). As you started proposing a patch for it, maybe you could try to fix it and check about similar troubles on other drivers. Cheers, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thursday 11 June 2009 06:40:14 Mauro Carvalho Chehab wrote: > Em Wed, 10 Jun 2009 22:39:51 -0300 > Mauro Carvalho Chehab <mchehab@redhat.com> escreveu: > > > Em Sun, 31 May 2009 14:41:52 +0800 > > "Figo.zhang" <figo1802@gmail.com> escreveu: > > > > > video_register_device() return zero on success, it would not return a positive integer. > > > > > > Signed-off-by: Figo.zhang <figo1802@gmail.com> > > > --- > > > drivers/media/video/ov511.c | 2 +- > > > 1 files changed, 1 insertions(+), 1 deletions(-) > > > > > > diff --git a/drivers/media/video/ov511.c b/drivers/media/video/ov511.c > > > index 9af5532..816427e 100644 > > > --- a/drivers/media/video/ov511.c > > > +++ b/drivers/media/video/ov511.c > > > @@ -5851,7 +5851,7 @@ ov51x_probe(struct usb_interface *intf, const struct usb_device_id *id) > > > break; > > > > > > if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, > > > - unit_video[i]) >= 0) { > > > + unit_video[i]) == 0) { > > > break; > > > } > > > } > > > > Nack. > > > > Errors are always negative. So, any zero or positive value indicates that no error occurred. > > > > Yet, the logic for forcing ov51x to specific minor number seems broken: it will > > end by registering the device twice, if used. > > > > So, that part of the function needs a rewrite. I'll fix it. > > > > Hmm... the issue seems more complex than I've imagined... > > When ov511 were written, it was assumed that video_register_device(dev, v, nr) > would have the following behavior: > > if nr = -1, it would do dynamic minor allocation; > if nr >= 0, it would allocate to 'nr' minor or fail. > > With the original behavior. > > The ov511_probe registering loop is doing something like this (I did a cleanup > to allow a better understand of the logic): > > <snip> > #define OV511_MAX_UNIT_VIDEO 16 > ... > static int unit_video[OV511_MAX_UNIT_VIDEO]; > ... > module_param_array(unit_video, int, &num_uv, 0); > MODULE_PARM_DESC(unit_video, > "Force use of specific minor number(s). 0 is not allowed."); > ... > static int > ov51x_probe(struct usb_interface *intf, const struct usb_device_id *id) > { > ... > for (i = 0; i < OV511_MAX_UNIT_VIDEO; i++) { > /* Minor 0 cannot be specified; assume user wants autodetect */ > if (unit_video[i] == 0) > break; > > if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, > unit_video[i]) >= 0) > goto register_succeded; > } > > /* Use the next available one */ > if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, -1) < 0) { > err("video_register_device failed"); > goto error; > > register_succeeded: > dev_info(&intf->dev, "Device at %s registered to minor %d\n", > ov->usb_path, ov->vdev->minor); > </snip> > > So, if you probe ov511 with a list of device numbers, for example: > > modprobe ov511 4,1,3 > > And assuming that you have 5 ov511 devices, and connect they one by one, > they'll gain the following device numbers (at the connection order): > /dev/video4 > /dev/video1 > /dev/video3 > /dev/video0 > /dev/video2 > > However, changeset 9133: > > changeset: 9133:64aed7485a43 > parent: 9073:4db9722caf4f > user: Hans Verkuil <hverkuil@xs4all.nl> > date: Sat Oct 04 13:36:54 2008 +0200 > summary: v4l: disconnect kernel number from minor > > Changed the behavior video_register_device(dev, v, nr): > > + nr = find_next_zero_bit(video_nums[type], minor_cnt, nr == -1 ? 0 : nr); > > So, instead of accepting just -1 or nr, it will now do: > if nr = -1, it will do dynamic minor allocation (as before); > if nr >= 0, it will also do dynamic minor allocation, but starting from nr. > > So, the new code won't fail if nr is already allocated, but it will return the > next unallocated nr. > > With the ov511 code, this means that you'll have, instead: > > /dev/video5 > /dev/video6 > /dev/video7 > /dev/video8 > /dev/video9 > > So, changeset 9133 actually broke the ov511 probe original behavior. > > In order to restore the original behavior, the probe logic should be replaced > by something else (like the approach taken by em28xx). > > Also, changeset 9133 may also cause other side effects on other drivers that > were expecting the original behavior. > > To make things worse, the function comment doesn't properly explain the current > behavior. > > --- > > Figo, > > Since we are in the middle of a merge window, it is unlikely that I'll have > enough time to properly fix it right now (since my priority right now is to > merge the existing patches). > > As you started proposing a patch for it, maybe you could try to fix it and > check about similar troubles on other drivers. > > Cheers, > Mauro > Since I made that change I'm willing to look at this. Some comments definitely need improving at the least. Also ivtv and cx18 rely on the current behavior, so any changes need to be done carefully. But one question first: isn't the current approach not better anyway than the old approach? In the past device creation would fail if you specified an explicit device kernel number that was already in use. Now it will attempt to fulfill the request and skip to the next free number otherwise. Seems a pretty good approach to me. There haven't been any complaints about this (probably also because nobody is using it). Let me know what you want and I can implement it. It's not that hard. Regards, Hans
Em Thu, 11 Jun 2009 08:36:00 +0200 Hans Verkuil <hverkuil@xs4all.nl> escreveu: > Since I made that change I'm willing to look at this. Some comments definitely > need improving at the least. Thanks! Since the behavior changed, it is important to better document it. > Also ivtv and cx18 rely on the current behavior, > so any changes need to be done carefully. > > But one question first: isn't the current approach not better anyway than the > old approach? In the past device creation would fail if you specified an > explicit device kernel number that was already in use. Now it will attempt > to fulfill the request and skip to the next free number otherwise. Seems a > pretty good approach to me. Well, the idea of not failing due to that is interesting. Yet, those force parameters are provided to avoid that a different modprobe order would change the device nodes. By doing something different than requested by the users without even warning them about that doesn't sound nice. At least a KERN_ERR log should be printed if the selected nr is different than the requested one. In the specific case of ov511, however, it caused a regression, since the used logic to get the next value of the array were based on the failure of video_register_device(). As it doesn't fail anymore, the current logic is broken. > There haven't been any complaints about this (probably also because nobody is > using it). > > Let me know what you want and I can implement it. It's not that hard. IMO, what should be done: 1) better comment the function; 2) generate a KERN_ERR at v4l2-dev, if the requested nr is not available; 3) replace ov511 logic to restore the old behavior (or improve it a little bit); 4) double check if similar regressions are present on other drivers. Since ov511 is an old driver, I don't doubt that its logic is duplicated on other devices. Since ov511 is used for an usb device, extra care should be taken, since it should be considered the possibility of successive hot plug/unplug. I wrote an interesting logic for this at em28xx driver, that can be used as an alternative to the current logic Cheers, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/media/video/ov511.c b/drivers/media/video/ov511.c index 9af5532..816427e 100644 --- a/drivers/media/video/ov511.c +++ b/drivers/media/video/ov511.c @@ -5851,7 +5851,7 @@ ov51x_probe(struct usb_interface *intf, const struct usb_device_id *id) break; if (video_register_device(ov->vdev, VFL_TYPE_GRABBER, - unit_video[i]) >= 0) { + unit_video[i]) == 0) { break; } }
video_register_device() return zero on success, it would not return a positive integer. Signed-off-by: Figo.zhang <figo1802@gmail.com> --- drivers/media/video/ov511.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html