| Message ID | 24acf224-e792-6648-fcdb-8729ded6df84@rock-chips.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, Jun 14, 2016 at 6:14 PM, Frank Wang <frank.wang@rock-chips.com> wrote: > Hi Heiko & Guenter, > > > On 2016/6/14 22:00, Heiko Stübner wrote: >> >> Am Dienstag, 14. Juni 2016, 06:50:31 schrieb Guenter Roeck: >>> >>> On Tue, Jun 14, 2016 at 6:27 AM, Heiko Stübner <heiko@sntech.de> wrote: >>>> >>>> Am Montag, 13. Juni 2016, 10:10:10 schrieb Frank Wang: >>>>> >>>>> The newer SoCs (rk3366, rk3399) take a different usb-phy IP block >>>>> than rk3288 and before, and most of phy-related registers are also >>>>> different from the past, so a new phy driver is required necessarily. >>>>> >>>>> Signed-off-by: Frank Wang <frank.wang@rock-chips.com> >>>>> --- >> >> [...] >> >>>>> +static int rockchip_usb2phy_init(struct phy *phy) >>>>> +{ >>>>> + struct rockchip_usb2phy_port *rport = phy_get_drvdata(phy); >>>>> + struct rockchip_usb2phy *rphy = dev_get_drvdata(phy->dev.parent); >>>>> + int ret; >>>>> + >>>>> >>>> if (!rport->port_cfg) >>>> return 0; >>>> >>>> Otherwise the currently empty otg-port will cause null-pointer >>>> dereferences >>>> when it gets assigned in the devicetree already. >>> >>> Not really, at least not here - that port should not have port_id set >>> to USB2PHY_PORT_HOST. >>> >>> Does it even make sense to instantiate the otg port ? Is it going to >>> do anything without port configuration ? >> >> Ok, that would be the other option - not creating the phy in the driver. > > > Well, I will put this conditional inside *_host_port_init(), if it is an > empty, the phy-device should not be created. > Something like the following: > > --- a/drivers/phy/phy-rockchip-inno-usb2.c > +++ b/drivers/phy/phy-rockchip-inno-usb2.c > @@ -483,9 +483,13 @@ static int rockchip_usb2phy_host_port_init(struct > rockchip_usb2phy *rphy, > { > int ret; > > - rport->port_id = USB2PHY_PORT_HOST; > rport->port_cfg = &rphy->phy_cfg->port_cfgs[USB2PHY_PORT_HOST]; > + if (!rport->port_cfg) { > + dev_err(rphy->dev, "no host port-config provided.\n"); > + return -EINVAL; > + } This would never be NULL. At issue is that you don't assign port_cfg if the port is _not_ a host port. Guenter > > + rport->port_id = USB2PHY_PORT_HOST; > >> Or from what I've seen, handling it as similar to the host-port should >> work >> initially as well most likely, supplying the additional otg-parts later >> on. > > > @Guenter, just as Heiko said, the otg-parts is not ready now, it will be > supplied later. > > > BR. > Frank >
Hi Guenter & Heiko, On 2016/6/15 23:47, Guenter Roeck wrote: > On Tue, Jun 14, 2016 at 6:14 PM, Frank Wang <frank.wang@rock-chips.com> wrote: >> Hi Heiko & Guenter, >> >> >> On 2016/6/14 22:00, Heiko Stübner wrote: >>> Am Dienstag, 14. Juni 2016, 06:50:31 schrieb Guenter Roeck: >>>> On Tue, Jun 14, 2016 at 6:27 AM, Heiko Stübner <heiko@sntech.de> wrote: >>>>> Am Montag, 13. Juni 2016, 10:10:10 schrieb Frank Wang: >>>>>> The newer SoCs (rk3366, rk3399) take a different usb-phy IP block >>>>>> than rk3288 and before, and most of phy-related registers are also >>>>>> different from the past, so a new phy driver is required necessarily. >>>>>> >>>>>> Signed-off-by: Frank Wang <frank.wang@rock-chips.com> >>>>>> --- >>> [...] >>> >>>>>> +static int rockchip_usb2phy_init(struct phy *phy) >>>>>> +{ >>>>>> + struct rockchip_usb2phy_port *rport = phy_get_drvdata(phy); >>>>>> + struct rockchip_usb2phy *rphy = dev_get_drvdata(phy->dev.parent); >>>>>> + int ret; >>>>>> + >>>>>> >>>>> if (!rport->port_cfg) >>>>> return 0; >>>>> >>>>> Otherwise the currently empty otg-port will cause null-pointer >>>>> dereferences >>>>> when it gets assigned in the devicetree already. >>>> Not really, at least not here - that port should not have port_id set >>>> to USB2PHY_PORT_HOST. >>>> >>>> Does it even make sense to instantiate the otg port ? Is it going to >>>> do anything without port configuration ? >>> Ok, that would be the other option - not creating the phy in the driver. >> >> Well, I will put this conditional inside *_host_port_init(), if it is an >> empty, the phy-device should not be created. >> Something like the following: >> >> --- a/drivers/phy/phy-rockchip-inno-usb2.c >> +++ b/drivers/phy/phy-rockchip-inno-usb2.c >> @@ -483,9 +483,13 @@ static int rockchip_usb2phy_host_port_init(struct >> rockchip_usb2phy *rphy, >> { >> int ret; >> >> - rport->port_id = USB2PHY_PORT_HOST; >> rport->port_cfg = &rphy->phy_cfg->port_cfgs[USB2PHY_PORT_HOST]; >> + if (!rport->port_cfg) { >> + dev_err(rphy->dev, "no host port-config provided.\n"); >> + return -EINVAL; >> + } > This would never be NULL. At issue is that you don't assign port_cfg > if the port is _not_ a host port. Sorry, I made a mistake. How about something like the following: @@ -574,6 +579,15 @@ static int rockchip_usb2phy_probe(struct platform_device *pdev) struct rockchip_usb2phy_port *rport = &rphy->ports[index]; struct phy *phy; + /* + * This driver aim to support both otg-port and host-port, + * but unfortunately, the otg part is not ready in current, + * so this comments and below codes are interim, which should + * be removed after otg-port is supplied soon. + */ + if (of_node_cmp(child_np->name, "host-port")) + goto next_child; + phy = devm_phy_create(dev, child_np, &rockchip_usb2phy_ops); if (IS_ERR(phy)) { dev_err(dev, "failed to create phy\n"); @@ -582,17 +596,13 @@ static int rockchip_usb2phy_probe(struct platform_device *pdev) } rport->phy = phy; - - /* initialize otg/host port separately */ - if (!of_node_cmp(child_np->name, "host-port")) { - ret = rockchip_usb2phy_host_port_init(rphy, rport, - child_np); - if (ret) - goto put_child; - } - phy_set_drvdata(rport->phy, rport); + ret = rockchip_usb2phy_host_port_init(rphy, rport, child_np); + if (ret) + goto put_child; + +next_child: /* to prevent out of boundary */ if (++index >= rphy->phy_cfg->num_ports) break; BR. Frank
On 06/15/2016 06:47 PM, Frank Wang wrote: > Hi Guenter & Heiko, > > On 2016/6/15 23:47, Guenter Roeck wrote: >> On Tue, Jun 14, 2016 at 6:14 PM, Frank Wang <frank.wang@rock-chips.com> wrote: >>> Hi Heiko & Guenter, >>> >>> >>> On 2016/6/14 22:00, Heiko Stübner wrote: >>>> Am Dienstag, 14. Juni 2016, 06:50:31 schrieb Guenter Roeck: >>>>> On Tue, Jun 14, 2016 at 6:27 AM, Heiko Stübner <heiko@sntech.de> wrote: >>>>>> Am Montag, 13. Juni 2016, 10:10:10 schrieb Frank Wang: >>>>>>> The newer SoCs (rk3366, rk3399) take a different usb-phy IP block >>>>>>> than rk3288 and before, and most of phy-related registers are also >>>>>>> different from the past, so a new phy driver is required necessarily. >>>>>>> >>>>>>> Signed-off-by: Frank Wang <frank.wang@rock-chips.com> >>>>>>> --- >>>> [...] >>>> >>>>>>> +static int rockchip_usb2phy_init(struct phy *phy) >>>>>>> +{ >>>>>>> + struct rockchip_usb2phy_port *rport = phy_get_drvdata(phy); >>>>>>> + struct rockchip_usb2phy *rphy = dev_get_drvdata(phy->dev.parent); >>>>>>> + int ret; >>>>>>> + >>>>>>> >>>>>> if (!rport->port_cfg) >>>>>> return 0; >>>>>> >>>>>> Otherwise the currently empty otg-port will cause null-pointer >>>>>> dereferences >>>>>> when it gets assigned in the devicetree already. >>>>> Not really, at least not here - that port should not have port_id set >>>>> to USB2PHY_PORT_HOST. >>>>> >>>>> Does it even make sense to instantiate the otg port ? Is it going to >>>>> do anything without port configuration ? >>>> Ok, that would be the other option - not creating the phy in the driver. >>> >>> Well, I will put this conditional inside *_host_port_init(), if it is an >>> empty, the phy-device should not be created. >>> Something like the following: >>> >>> --- a/drivers/phy/phy-rockchip-inno-usb2.c >>> +++ b/drivers/phy/phy-rockchip-inno-usb2.c >>> @@ -483,9 +483,13 @@ static int rockchip_usb2phy_host_port_init(struct >>> rockchip_usb2phy *rphy, >>> { >>> int ret; >>> >>> - rport->port_id = USB2PHY_PORT_HOST; >>> rport->port_cfg = &rphy->phy_cfg->port_cfgs[USB2PHY_PORT_HOST]; >>> + if (!rport->port_cfg) { >>> + dev_err(rphy->dev, "no host port-config provided.\n"); >>> + return -EINVAL; >>> + } >> This would never be NULL. At issue is that you don't assign port_cfg >> if the port is _not_ a host port. > > Sorry, I made a mistake. How about something like the following: > Yes, that should work. Just keep in mind that there could always be a port named "something-port", so you'll always need some kind of check (and possibly return an error if a port with a wrong name is provided). Thanks, Guenter > @@ -574,6 +579,15 @@ static int rockchip_usb2phy_probe(struct platform_device *pdev) > struct rockchip_usb2phy_port *rport = &rphy->ports[index]; > struct phy *phy; > > + /* > + * This driver aim to support both otg-port and host-port, > + * but unfortunately, the otg part is not ready in current, > + * so this comments and below codes are interim, which should > + * be removed after otg-port is supplied soon. > + */ > + if (of_node_cmp(child_np->name, "host-port")) > + goto next_child; > + > phy = devm_phy_create(dev, child_np, &rockchip_usb2phy_ops); > if (IS_ERR(phy)) { > dev_err(dev, "failed to create phy\n"); > @@ -582,17 +596,13 @@ static int rockchip_usb2phy_probe(struct platform_device *pdev) > } > > rport->phy = phy; > - > - /* initialize otg/host port separately */ > - if (!of_node_cmp(child_np->name, "host-port")) { > - ret = rockchip_usb2phy_host_port_init(rphy, rport, > - child_np); > - if (ret) > - goto put_child; > - } > - > phy_set_drvdata(rport->phy, rport); > > + ret = rockchip_usb2phy_host_port_init(rphy, rport, child_np); > + if (ret) > + goto put_child; > + > +next_child: > /* to prevent out of boundary */ > if (++index >= rphy->phy_cfg->num_ports) > break; > > > BR. > Frank > >
Hi Guenter, On 2016/6/16 21:12, Guenter Roeck wrote: > On 06/15/2016 06:47 PM, Frank Wang wrote: >> Hi Guenter & Heiko, >> >> On 2016/6/15 23:47, Guenter Roeck wrote: >>> On Tue, Jun 14, 2016 at 6:14 PM, Frank Wang >>> <frank.wang@rock-chips.com> wrote: >>>> Hi Heiko & Guenter, >>>> >>>> >>>> On 2016/6/14 22:00, Heiko Stübner wrote: >>>>> Am Dienstag, 14. Juni 2016, 06:50:31 schrieb Guenter Roeck: >>>>>> On Tue, Jun 14, 2016 at 6:27 AM, Heiko Stübner <heiko@sntech.de> >>>>>> wrote: >>>>>>> Am Montag, 13. Juni 2016, 10:10:10 schrieb Frank Wang: >>>>>>>> The newer SoCs (rk3366, rk3399) take a different usb-phy IP block >>>>>>>> than rk3288 and before, and most of phy-related registers are also >>>>>>>> different from the past, so a new phy driver is required >>>>>>>> necessarily. >>>>>>>> >>>>>>>> Signed-off-by: Frank Wang <frank.wang@rock-chips.com> >>>>>>>> --- >>>>> [...] >>>>> >>>>>>>> +static int rockchip_usb2phy_init(struct phy *phy) >>>>>>>> +{ >>>>>>>> + struct rockchip_usb2phy_port *rport = phy_get_drvdata(phy); >>>>>>>> + struct rockchip_usb2phy *rphy = >>>>>>>> dev_get_drvdata(phy->dev.parent); >>>>>>>> + int ret; >>>>>>>> + >>>>>>>> >>>>>>> if (!rport->port_cfg) >>>>>>> return 0; >>>>>>> >>>>>>> Otherwise the currently empty otg-port will cause null-pointer >>>>>>> dereferences >>>>>>> when it gets assigned in the devicetree already. >>>>>> Not really, at least not here - that port should not have port_id >>>>>> set >>>>>> to USB2PHY_PORT_HOST. >>>>>> >>>>>> Does it even make sense to instantiate the otg port ? Is it going to >>>>>> do anything without port configuration ? >>>>> Ok, that would be the other option - not creating the phy in the >>>>> driver. >>>> >>>> Well, I will put this conditional inside *_host_port_init(), if it >>>> is an >>>> empty, the phy-device should not be created. >>>> Something like the following: >>>> >>>> --- a/drivers/phy/phy-rockchip-inno-usb2.c >>>> +++ b/drivers/phy/phy-rockchip-inno-usb2.c >>>> @@ -483,9 +483,13 @@ static int rockchip_usb2phy_host_port_init(struct >>>> rockchip_usb2phy *rphy, >>>> { >>>> int ret; >>>> >>>> - rport->port_id = USB2PHY_PORT_HOST; >>>> rport->port_cfg = >>>> &rphy->phy_cfg->port_cfgs[USB2PHY_PORT_HOST]; >>>> + if (!rport->port_cfg) { >>>> + dev_err(rphy->dev, "no host port-config provided.\n"); >>>> + return -EINVAL; >>>> + } >>> This would never be NULL. At issue is that you don't assign port_cfg >>> if the port is _not_ a host port. >> >> Sorry, I made a mistake. How about something like the following: >> > Yes, that should work. Just keep in mind that there could always be > a port named "something-port", so you'll always need some kind of check > (and possibly return an error if a port with a wrong name is provided). > > OK, thanks for your reminding, I am going to send out it with a new version later. BR. Frank > >> @@ -574,6 +579,15 @@ static int rockchip_usb2phy_probe(struct >> platform_device *pdev) >> struct rockchip_usb2phy_port *rport = >> &rphy->ports[index]; >> struct phy *phy; >> >> + /* >> + * This driver aim to support both otg-port and >> host-port, >> + * but unfortunately, the otg part is not ready in >> current, >> + * so this comments and below codes are interim, >> which should >> + * be removed after otg-port is supplied soon. >> + */ >> + if (of_node_cmp(child_np->name, "host-port")) >> + goto next_child; >> + >> phy = devm_phy_create(dev, child_np, >> &rockchip_usb2phy_ops); >> if (IS_ERR(phy)) { >> dev_err(dev, "failed to create phy\n"); >> @@ -582,17 +596,13 @@ static int rockchip_usb2phy_probe(struct >> platform_device *pdev) >> } >> >> rport->phy = phy; >> - >> - /* initialize otg/host port separately */ >> - if (!of_node_cmp(child_np->name, "host-port")) { >> - ret = rockchip_usb2phy_host_port_init(rphy, >> rport, >> - child_np); >> - if (ret) >> - goto put_child; >> - } >> - >> phy_set_drvdata(rport->phy, rport); >> >> + ret = rockchip_usb2phy_host_port_init(rphy, rport, >> child_np); >> + if (ret) >> + goto put_child; >> + >> +next_child: >> /* to prevent out of boundary */ >> if (++index >= rphy->phy_cfg->num_ports) >> break; >> >> >> >
--- a/drivers/phy/phy-rockchip-inno-usb2.c +++ b/drivers/phy/phy-rockchip-inno-usb2.c @@ -483,9 +483,13 @@ static int rockchip_usb2phy_host_port_init(struct rockchip_usb2phy *rphy, { int ret; - rport->port_id = USB2PHY_PORT_HOST; rport->port_cfg = &rphy->phy_cfg->port_cfgs[USB2PHY_PORT_HOST]; + if (!rport->port_cfg) { + dev_err(rphy->dev, "no host port-config provided.\n"); + return -EINVAL; + } + rport->port_id = USB2PHY_PORT_HOST; > Or from what I've seen, handling it as similar to the host-port should work