Re: [PATCH v3 5/5] drm/fsl-dcu: only init fbdev if required

[Daniel Vetter]

On Tue, Oct 18, 2016 at 10:42:46AM -0700, Stefan Agner wrote:
> On 2016-10-18 00:44, Daniel Vetter wrote:
> > On Mon, Oct 17, 2016 at 02:33:21PM -0700, Stefan Agner wrote:
> >> There is no need to request a CMA backed framebuffer if fbdev
> >> emulation is not enabled.
> >>
> >> Signed-off-by: Stefan Agner <stefan@xxxxxxxx>
> >> ---
> >>  drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c | 3 ++-
> >>  1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c b/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c
> >> index e04efbe..3a5880c 100644
> >> --- a/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c
> >> +++ b/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c
> >> @@ -87,7 +87,8 @@ static int fsl_dcu_load(struct drm_device *dev, unsigned long flags)
> >>  		goto done;
> >>  	dev->irq_enabled = true;
> >>
> >> -	fsl_dcu_fbdev_init(dev);
> >> +	if (IS_ENABLED(CONFIG_DRM_FBDEV_EMULATION))
> >> +		fsl_dcu_fbdev_init(dev);
> > 
> > Totally not required, since this will all no-op out. Also, please nuke
> > that fsl_dcu_fbdv_init wrapper seems like pointless indirection.
> 
> Regarding fsl_dcu_fbdev_init wrapper: Fully agreed. That thing was there
> since inception of that driver, and I always was on the fence of doing
> it. Will do it for the next merge window!
> 
> I somehow remembered there was something more to it than just "no need",
> but I somehow failed to document it in the patch description... So I
> went back and tested some things without the patch, here is when it
> blows:
> 
> Unable to handle kernel NULL pointer dereference at virtual address
> 000002f0
> pgd = cc24c000
> [000002f0] *pgd=8c0df831, *pte=00000000, *ppte=00000000
> Internal error: Oops: 17 [#1] SMP ARM
> Modules linked in:
> CPU: 0 PID: 536 Comm: sh Not tainted 4.9.0-rc1-00001-g4b1532a-dirty #568
> Hardware name: Freescale Vybrid VF5xx/VF6xx (Device Tree)
> task: cc213000 task.stack: cc23e000
> PC is at drm_fbdev_cma_fini+0x14/0x5c
> LR is at fsl_dcu_unload+0x28/0x4c
> pc : [<c04cfb20>]    lr : [<c04fad74>]    psr: a0000013
> sp : cc23fd80  ip : cc23fd98  fp : cc23fd94
> r10: cc1d1e4c  r9 : cc23e000  r8 : 00000000
> r7 : c0e34888  r6 : 0000000d  r5 : 00000000  r4 : ce6ff100
> r3 : c0e13b18  r2 : c0e13b18  r1 : 00000110  r0 : ce6ff100
> Flags: NzCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
> Control: 10c5387d  Table: 8c24c04a  DAC: 00000051
> Process sh (pid: 536, stack limit = 0xcc23e210)
> Stack: (0xcc23fd80 to 0xcc240000)
> ...
> Backtrace:
> [<c04cfb0c>] (drm_fbdev_cma_fini) from [<c04fad74>]
> (fsl_dcu_unload+0x28/0x4c)
>  r5:ce61e810[  372.213609]  r4:ce61f000
> 
> [<c04fad4c>] (fsl_dcu_unload) from [<c04d9550>]
> (drm_dev_unregister+0x38/0xbc)
>  r5:ce61f000[  372.228535]  r4:ce61f000
> ...
> 
> > 
> > And if there really is an issue with the cma helpers allocating an fb when
> > they should, then the correct fix is to fix that in the helpers, not in
> > the drivers.
> 
> Hm, to safe memory, it would probably be best to do something like:
> 
> --- a/drivers/gpu/drm/drm_fb_cma_helper.c
> +++ b/drivers/gpu/drm/drm_fb_cma_helper.c
> @@ -492,6 +492,9 @@ struct drm_fbdev_cma
> *drm_fbdev_cma_init_with_funcs(struct drm_device *dev,
>         struct drm_fb_helper *helper;
>         int ret;
>  
> +       if (!drm_fbdev_emulation)
> +               return NULL;
> +
>         fbdev_cma = kzalloc(sizeof(*fbdev_cma), GFP_KERNEL);
>         if (!fbdev_cma) {
>                 dev_err(dev->dev, "Failed to allocate drm fbdev.\n");
> 
> But we don't have drm_fbdev_emulation emulation there.

Not just that, but you'd leak memory since cma_init always does the
kmalloc of the fbdev_cma struct.

> Maybe just add some NULL check in drm_fbdev_cma_fini?

Probably. I can't read arm-oopses, so no idea what exactly blew up. On a
hunch it's probably drm_fbdev_cma_defio_fini that got things wrong and
needs to check for !fbi || !fbi->defio.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch