Message ID | 20190318144758.21267-1-tzimmermann@suse.de (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [RESEND] drm/vmwgfx: Don't double-free the mode stored in par->set_mode | expand |
Hi Thomas, Thanks for doing this and somehow I missed the last patch, sorry about that. Have some questions below otherwise the patch looks good to me. Reviewed-by: Deepak Rawat <drawat@vmware.com> I will include your changes in vmwgfx-next and run tests. On Mon, 2019-03-18 at 15:47 +0100, Thomas Zimmermann wrote: > When calling vmw_fb_set_par(), the mode stored in par->set_mode gets > free'd > twice. The first free is in vmw_fb_kms_detach(), the second is near > the > end of vmw_fb_set_par() under the name of 'old_mode'. The mode- > setting code > only works correctly if the mode doesn't actually change. You mean to say that without your patch vmwgfx fb driver fail to change the mode? > Removing 'old_mode' > in favor of using par->set_mode directly fixes the problem. > > Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> > --- > drivers/gpu/drm/vmwgfx/vmwgfx_fb.c | 12 +++--------- > 1 file changed, 3 insertions(+), 9 deletions(-) > > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > index b913a56f3426..2a9112515f46 100644 > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > @@ -564,11 +564,9 @@ static int vmw_fb_set_par(struct fb_info *info) > 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, > DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) > }; > - struct drm_display_mode *old_mode; > struct drm_display_mode *mode; > int ret; > > - old_mode = par->set_mode; > mode = drm_mode_duplicate(vmw_priv->dev, &new_mode); > if (!mode) { > DRM_ERROR("Could not create new fb mode.\n"); > @@ -579,11 +577,7 @@ static int vmw_fb_set_par(struct fb_info *info) > mode->vdisplay = var->yres; > vmw_guess_mode_timing(mode); > > - if (old_mode && drm_mode_equal(old_mode, mode)) { > - drm_mode_destroy(vmw_priv->dev, mode); > - mode = old_mode; > - old_mode = NULL; I am having hard time understanding original intention for this piece of code. Was there a restriction to send pointer to old mode if mode were same and that restriction don't hold anymore. Sorry I am not familiar with this code area. > - } else if (!vmw_kms_validate_mode_vram(vmw_priv, > + if (!vmw_kms_validate_mode_vram(vmw_priv, > mode->hdisplay * > DIV_ROUND_UP(var- > >bits_per_pixel, 8), > mode->vdisplay)) { > @@ -620,8 +614,8 @@ static int vmw_fb_set_par(struct fb_info *info) > schedule_delayed_work(&par->local_work, 0); > > out_unlock: > - if (old_mode) > - drm_mode_destroy(vmw_priv->dev, old_mode); > + if (par->set_mode) > + drm_mode_destroy(vmw_priv->dev, par->set_mode); > par->set_mode = mode; > > mutex_unlock(&par->bo_mutex); > -- > 2.20.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Fdri-devel&data=02%7C01%7Cdrawat%40vmware.com%7Cb1508247a3954fb0b08b08d6abb0bcd0%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636885172908359973&sdata=AN6UTrMzxcVK7MC6bX3OxNbcyq0j4HdKt0dk1yyHOHc%3D&reserved=0
Hi Deepak Am 18.03.19 um 17:59 schrieb Deepak Singh Rawat: > Hi Thomas, > > Thanks for doing this and somehow I missed the last patch, sorry about > that. Have some questions below otherwise the patch looks good to me. > > Reviewed-by: Deepak Rawat <drawat@vmware.com> > > I will include your changes in vmwgfx-next and run tests. Thank you. > > On Mon, 2019-03-18 at 15:47 +0100, Thomas Zimmermann wrote: >> When calling vmw_fb_set_par(), the mode stored in par->set_mode gets >> free'd >> twice. The first free is in vmw_fb_kms_detach(), the second is near >> the >> end of vmw_fb_set_par() under the name of 'old_mode'. The mode- >> setting code >> only works correctly if the mode doesn't actually change. > > You mean to say that without your patch vmwgfx fb driver fail to change > the mode? Yes. It's hard to trigger as the usual resolution always appears to be 800x600. We (SUSE) have a customer with a custom X11 modeline setting, which sets the resolution to something different. In this case, the double-free bug happens. > >> Removing 'old_mode' >> in favor of using par->set_mode directly fixes the problem. >> >> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> >> --- >> drivers/gpu/drm/vmwgfx/vmwgfx_fb.c | 12 +++--------- >> 1 file changed, 3 insertions(+), 9 deletions(-) >> >> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c >> b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c >> index b913a56f3426..2a9112515f46 100644 >> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c >> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c >> @@ -564,11 +564,9 @@ static int vmw_fb_set_par(struct fb_info *info) >> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, >> DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) >> }; >> - struct drm_display_mode *old_mode; >> struct drm_display_mode *mode; >> int ret; >> >> - old_mode = par->set_mode; >> mode = drm_mode_duplicate(vmw_priv->dev, &new_mode); >> if (!mode) { >> DRM_ERROR("Could not create new fb mode.\n"); >> @@ -579,11 +577,7 @@ static int vmw_fb_set_par(struct fb_info *info) >> mode->vdisplay = var->yres; >> vmw_guess_mode_timing(mode); >> >> - if (old_mode && drm_mode_equal(old_mode, mode)) { >> - drm_mode_destroy(vmw_priv->dev, mode); >> - mode = old_mode; >> - old_mode = NULL; > > I am having hard time understanding original intention for this piece > of code. Was there a restriction to send pointer to old mode if mode > were same and that restriction don't hold anymore. Sorry I am not > familiar with this code area. Hmm, I can only guess about motivation: it looks like the author wanted to keep the original value of 'par->set_mode' unless the mode really changes. But it doesn't make a difference whether the pointer's value changes or not. Best regards Thomas > >> - } else if (!vmw_kms_validate_mode_vram(vmw_priv, >> + if (!vmw_kms_validate_mode_vram(vmw_priv, >> mode->hdisplay * >> DIV_ROUND_UP(var- >>> bits_per_pixel, 8), >> mode->vdisplay)) { >> @@ -620,8 +614,8 @@ static int vmw_fb_set_par(struct fb_info *info) >> schedule_delayed_work(&par->local_work, 0); >> >> out_unlock: >> - if (old_mode) >> - drm_mode_destroy(vmw_priv->dev, old_mode); >> + if (par->set_mode) >> + drm_mode_destroy(vmw_priv->dev, par->set_mode); >> par->set_mode = mode; >> >> mutex_unlock(&par->bo_mutex); >> -- >> 2.20.1 >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> > https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Fdri-devel&data=02%7C01%7Cdrawat%40vmware.com%7Cb1508247a3954fb0b08b08d6abb0bcd0%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636885172908359973&sdata=AN6UTrMzxcVK7MC6bX3OxNbcyq0j4HdKt0dk1yyHOHc%3D&reserved=0 >
Hi, Deepak, On Mon, 2019-03-18 at 09:59 -0700, Deepak Singh Rawat wrote: > Hi Thomas, > > Thanks for doing this and somehow I missed the last patch, sorry > about > that. Have some questions below otherwise the patch looks good to me. > > Reviewed-by: Deepak Rawat <drawat@vmware.com> > > I will include your changes in vmwgfx-next and run tests. I think we need to run this through vmwgfx-fixes and CC stable. I'll pick up the patch for that. > > On Mon, 2019-03-18 at 15:47 +0100, Thomas Zimmermann wrote: > > When calling vmw_fb_set_par(), the mode stored in par->set_mode > > gets > > free'd > > twice. The first free is in vmw_fb_kms_detach(), the second is near > > the > > end of vmw_fb_set_par() under the name of 'old_mode'. The mode- > > setting code > > only works correctly if the mode doesn't actually change. > > You mean to say that without your patch vmwgfx fb driver fail to > change > the mode? > > > Removing 'old_mode' > > in favor of using par->set_mode directly fixes the problem. > > > > Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> > > --- > > drivers/gpu/drm/vmwgfx/vmwgfx_fb.c | 12 +++--------- > > 1 file changed, 3 insertions(+), 9 deletions(-) > > > > diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > > b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > > index b913a56f3426..2a9112515f46 100644 > > --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > > +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c > > @@ -564,11 +564,9 @@ static int vmw_fb_set_par(struct fb_info > > *info) > > 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, > > DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) > > }; > > - struct drm_display_mode *old_mode; > > struct drm_display_mode *mode; > > int ret; > > > > - old_mode = par->set_mode; > > mode = drm_mode_duplicate(vmw_priv->dev, &new_mode); > > if (!mode) { > > DRM_ERROR("Could not create new fb mode.\n"); > > @@ -579,11 +577,7 @@ static int vmw_fb_set_par(struct fb_info > > *info) > > mode->vdisplay = var->yres; > > vmw_guess_mode_timing(mode); > > > > - if (old_mode && drm_mode_equal(old_mode, mode)) { > > - drm_mode_destroy(vmw_priv->dev, mode); > > - mode = old_mode; > > - old_mode = NULL; > > I am having hard time understanding original intention for this piece > of code. Was there a restriction to send pointer to old mode if mode > were same and that restriction don't hold anymore. Sorry I am not > familiar with this code area. It looks like that code is there to reuse the old mode if possible. In that case things work, but if a new mode is indeed created, the old mode will be double-freed :( /Thomas > > > - } else if (!vmw_kms_validate_mode_vram(vmw_priv, > > + if (!vmw_kms_validate_mode_vram(vmw_priv, > > mode->hdisplay * > > DIV_ROUND_UP(var- > > > bits_per_pixel, 8), > > mode->vdisplay)) { > > @@ -620,8 +614,8 @@ static int vmw_fb_set_par(struct fb_info *info) > > schedule_delayed_work(&par->local_work, 0); > > > > out_unlock: > > - if (old_mode) > > - drm_mode_destroy(vmw_priv->dev, old_mode); > > + if (par->set_mode) > > + drm_mode_destroy(vmw_priv->dev, par->set_mode); > > par->set_mode = mode; > > > > mutex_unlock(&par->bo_mutex); > > -- > > 2.20.1 > > > > _______________________________________________ > > dri-devel mailing list > > dri-devel@lists.freedesktop.org > > > https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Fdri-devel&data=02%7C01%7Cdrawat%40vmware.com%7Cb1508247a3954fb0b08b08d6abb0bcd0%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636885172908359973&sdata=AN6UTrMzxcVK7MC6bX3OxNbcyq0j4HdKt0dk1yyHOHc%3D&reserved=0 >
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c index b913a56f3426..2a9112515f46 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c @@ -564,11 +564,9 @@ static int vmw_fb_set_par(struct fb_info *info) 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_PVSYNC) }; - struct drm_display_mode *old_mode; struct drm_display_mode *mode; int ret; - old_mode = par->set_mode; mode = drm_mode_duplicate(vmw_priv->dev, &new_mode); if (!mode) { DRM_ERROR("Could not create new fb mode.\n"); @@ -579,11 +577,7 @@ static int vmw_fb_set_par(struct fb_info *info) mode->vdisplay = var->yres; vmw_guess_mode_timing(mode); - if (old_mode && drm_mode_equal(old_mode, mode)) { - drm_mode_destroy(vmw_priv->dev, mode); - mode = old_mode; - old_mode = NULL; - } else if (!vmw_kms_validate_mode_vram(vmw_priv, + if (!vmw_kms_validate_mode_vram(vmw_priv, mode->hdisplay * DIV_ROUND_UP(var->bits_per_pixel, 8), mode->vdisplay)) { @@ -620,8 +614,8 @@ static int vmw_fb_set_par(struct fb_info *info) schedule_delayed_work(&par->local_work, 0); out_unlock: - if (old_mode) - drm_mode_destroy(vmw_priv->dev, old_mode); + if (par->set_mode) + drm_mode_destroy(vmw_priv->dev, par->set_mode); par->set_mode = mode; mutex_unlock(&par->bo_mutex);
When calling vmw_fb_set_par(), the mode stored in par->set_mode gets free'd twice. The first free is in vmw_fb_kms_detach(), the second is near the end of vmw_fb_set_par() under the name of 'old_mode'. The mode-setting code only works correctly if the mode doesn't actually change. Removing 'old_mode' in favor of using par->set_mode directly fixes the problem. Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de> --- drivers/gpu/drm/vmwgfx/vmwgfx_fb.c | 12 +++--------- 1 file changed, 3 insertions(+), 9 deletions(-) -- 2.20.1