| Message ID | 20201216012218.9205-1-laurent.pinchart+renesas@ideasonboard.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Kieran Bingham |
| Headers | show |
| Series | drm: rcar-du: Fix leak of CMM platform device reference | expand |
Hi Laurent, I wonder if 'leaked' is correct in subject. It probably is, un-balanced ref-counting will prevent the device from being released. It should however happen only at system tear-down, doesn't it ? On Wed, Dec 16, 2020 at 03:22:18AM +0200, Laurent Pinchart wrote: > The device references acquired by of_find_device_by_node() are not > released by the driver. Fix this by registering a cleanup action. > > Fixes: 8de707aeb452 ("drm: rcar-du: kms: Initialize CMM instances") > Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com> > --- > drivers/gpu/drm/rcar-du/rcar_du_kms.c | 20 ++++++++++++++++++-- > 1 file changed, 18 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/rcar-du/rcar_du_kms.c b/drivers/gpu/drm/rcar-du/rcar_du_kms.c > index 92dfa3d4c011..7070f3c9b529 100644 > --- a/drivers/gpu/drm/rcar-du/rcar_du_kms.c > +++ b/drivers/gpu/drm/rcar-du/rcar_du_kms.c > @@ -14,6 +14,7 @@ > #include <drm/drm_fb_cma_helper.h> > #include <drm/drm_gem_cma_helper.h> > #include <drm/drm_gem_framebuffer_helper.h> > +#include <drm/drm_managed.h> > #include <drm/drm_probe_helper.h> > #include <drm/drm_vblank.h> > > @@ -721,6 +722,8 @@ static int rcar_du_cmm_init(struct rcar_du_device *rcdu) > > of_node_put(cmm); > > + rcdu->cmms[i] = pdev; > + In the unfortunate event that the cmm intialization fails here below, rcdu->cmms[i] will stay assigned, causing the rcar_du_crtc_create() function which is called just after rcar_du_cmm_init() to access a non-valid cmm instance. I would either reset the rcdu->cmms[i] field to NULL in the below error paths, or maintain the cmms[i] = pdev assignement at the end of the function and put_device(pdev->dev) in the error paths. Thanks j > /* > * -ENODEV is used to report that the CMM config option is > * disabled: return 0 and let the DU continue probing. > @@ -739,13 +742,22 @@ static int rcar_du_cmm_init(struct rcar_du_device *rcdu) > "Failed to create device link to CMM%u\n", i); > return -EINVAL; > } > - > - rcdu->cmms[i] = pdev; > } > > return 0; > } > > +static void rcar_du_modeset_cleanup(struct drm_device *dev, void *res) > +{ > + struct rcar_du_device *rcdu = to_rcar_du_device(dev); > + unsigned int i; > + > + for (i = 0; i < ARRAY_SIZE(rcdu->cmms); ++i) { > + if (rcdu->cmms[i]) > + put_device(&rcdu->cmms[i]->dev); > + } > +} > + > int rcar_du_modeset_init(struct rcar_du_device *rcdu) > { > static const unsigned int mmio_offsets[] = { > @@ -766,6 +778,10 @@ int rcar_du_modeset_init(struct rcar_du_device *rcdu) > if (ret) > return ret; > > + ret = drmm_add_action(&rcdu->ddev, rcar_du_modeset_cleanup, NULL); > + if (ret) > + return ret; > + > dev->mode_config.min_width = 0; > dev->mode_config.min_height = 0; > dev->mode_config.normalize_zpos = true; > -- > Regards, > > Laurent Pinchart >
Hi Jacopo, On Wed, Dec 16, 2020 at 02:51:34PM +0100, Jacopo Mondi wrote: > Hi Laurent, > > I wonder if 'leaked' is correct in subject. It probably is, > un-balanced ref-counting will prevent the device from being released. > It should however happen only at system tear-down, doesn't it ? As the CMM shouldn't be hotplugged, yes. It's not really the device that is leaked here, but the reference. > On Wed, Dec 16, 2020 at 03:22:18AM +0200, Laurent Pinchart wrote: > > The device references acquired by of_find_device_by_node() are not > > released by the driver. Fix this by registering a cleanup action. > > > > Fixes: 8de707aeb452 ("drm: rcar-du: kms: Initialize CMM instances") > > Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com> > > --- > > drivers/gpu/drm/rcar-du/rcar_du_kms.c | 20 ++++++++++++++++++-- > > 1 file changed, 18 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/gpu/drm/rcar-du/rcar_du_kms.c b/drivers/gpu/drm/rcar-du/rcar_du_kms.c > > index 92dfa3d4c011..7070f3c9b529 100644 > > --- a/drivers/gpu/drm/rcar-du/rcar_du_kms.c > > +++ b/drivers/gpu/drm/rcar-du/rcar_du_kms.c > > @@ -14,6 +14,7 @@ > > #include <drm/drm_fb_cma_helper.h> > > #include <drm/drm_gem_cma_helper.h> > > #include <drm/drm_gem_framebuffer_helper.h> > > +#include <drm/drm_managed.h> > > #include <drm/drm_probe_helper.h> > > #include <drm/drm_vblank.h> > > > > @@ -721,6 +722,8 @@ static int rcar_du_cmm_init(struct rcar_du_device *rcdu) > > > > of_node_put(cmm); > > > > + rcdu->cmms[i] = pdev; > > + > > In the unfortunate event that the cmm intialization fails here below, > rcdu->cmms[i] will stay assigned, causing the rcar_du_crtc_create() > function which is called just after rcar_du_cmm_init() to access a > non-valid cmm instance. > > I would either reset the rcdu->cmms[i] field to NULL in the below error > paths, or maintain the cmms[i] = pdev assignement at the end of the > function and put_device(pdev->dev) in the error paths. You're right. I'll fix that. > > /* > > * -ENODEV is used to report that the CMM config option is > > * disabled: return 0 and let the DU continue probing. > > @@ -739,13 +742,22 @@ static int rcar_du_cmm_init(struct rcar_du_device *rcdu) > > "Failed to create device link to CMM%u\n", i); > > return -EINVAL; > > } > > - > > - rcdu->cmms[i] = pdev; > > } > > > > return 0; > > } > > > > +static void rcar_du_modeset_cleanup(struct drm_device *dev, void *res) > > +{ > > + struct rcar_du_device *rcdu = to_rcar_du_device(dev); > > + unsigned int i; > > + > > + for (i = 0; i < ARRAY_SIZE(rcdu->cmms); ++i) { > > + if (rcdu->cmms[i]) > > + put_device(&rcdu->cmms[i]->dev); > > + } > > +} > > + > > int rcar_du_modeset_init(struct rcar_du_device *rcdu) > > { > > static const unsigned int mmio_offsets[] = { > > @@ -766,6 +778,10 @@ int rcar_du_modeset_init(struct rcar_du_device *rcdu) > > if (ret) > > return ret; > > > > + ret = drmm_add_action(&rcdu->ddev, rcar_du_modeset_cleanup, NULL); > > + if (ret) > > + return ret; > > + > > dev->mode_config.min_width = 0; > > dev->mode_config.min_height = 0; > > dev->mode_config.normalize_zpos = true;
diff --git a/drivers/gpu/drm/rcar-du/rcar_du_kms.c b/drivers/gpu/drm/rcar-du/rcar_du_kms.c index 92dfa3d4c011..7070f3c9b529 100644 --- a/drivers/gpu/drm/rcar-du/rcar_du_kms.c +++ b/drivers/gpu/drm/rcar-du/rcar_du_kms.c @@ -14,6 +14,7 @@ #include <drm/drm_fb_cma_helper.h> #include <drm/drm_gem_cma_helper.h> #include <drm/drm_gem_framebuffer_helper.h> +#include <drm/drm_managed.h> #include <drm/drm_probe_helper.h> #include <drm/drm_vblank.h> @@ -721,6 +722,8 @@ static int rcar_du_cmm_init(struct rcar_du_device *rcdu) of_node_put(cmm); + rcdu->cmms[i] = pdev; + /* * -ENODEV is used to report that the CMM config option is * disabled: return 0 and let the DU continue probing. @@ -739,13 +742,22 @@ static int rcar_du_cmm_init(struct rcar_du_device *rcdu) "Failed to create device link to CMM%u\n", i); return -EINVAL; } - - rcdu->cmms[i] = pdev; } return 0; } +static void rcar_du_modeset_cleanup(struct drm_device *dev, void *res) +{ + struct rcar_du_device *rcdu = to_rcar_du_device(dev); + unsigned int i; + + for (i = 0; i < ARRAY_SIZE(rcdu->cmms); ++i) { + if (rcdu->cmms[i]) + put_device(&rcdu->cmms[i]->dev); + } +} + int rcar_du_modeset_init(struct rcar_du_device *rcdu) { static const unsigned int mmio_offsets[] = { @@ -766,6 +778,10 @@ int rcar_du_modeset_init(struct rcar_du_device *rcdu) if (ret) return ret; + ret = drmm_add_action(&rcdu->ddev, rcar_du_modeset_cleanup, NULL); + if (ret) + return ret; + dev->mode_config.min_width = 0; dev->mode_config.min_height = 0; dev->mode_config.normalize_zpos = true;
The device references acquired by of_find_device_by_node() are not released by the driver. Fix this by registering a cleanup action. Fixes: 8de707aeb452 ("drm: rcar-du: kms: Initialize CMM instances") Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com> --- drivers/gpu/drm/rcar-du/rcar_du_kms.c | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-)