On Friday 25 March 2016 09:24:28, Stefan Agner wrote:

Hi Alexander,

On 2016-03-24 06:33, Alexander Stein wrote:

...

Using REGCACHE_RBTREE for MMIO regmap is not valid as spinlock's will be used during cache allocation.

This fixes the following bug: BUG: sleeping function called from invalid context at mm/slab.h:388 in_atomic(): 1, irqs_disabled(): 128, pid: 192, name: udevd [...]

Signed-off-by: Alexander Stein alexander.stein@systec-electronic.com

Please refer also to the discussion at https://lists.freedesktop.org/archives/dri-devel/2016-January/098696.html

drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c | 2 +- 1 file changed, 1 insertion(+), 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 e8d9337..ea65140 100644 --- a/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c +++ b/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c @@ -40,7 +40,7 @@ static const struct regmap_config fsl_dcu_regmap_config = {> .reg_bits = 32, .reg_stride = 4, .val_bits = 32,

• .cache_type = REGCACHE_RBTREE,

• .cache_type = REGCACHE_FLAT,

To use flat regcache you also need to set the max_register.

Ah, thanks for that hint, I was not aware of that. I wondered why my kernel crashed using flat cache. Seems cache allocation is too small when max_register=0. Just sent a patch enforcing setting max_register.

I already have such a patch, see: https://lists.freedesktop.org/archives/dri-devel/2016-January/099121.html

Thanks, meanwhile this will do for me. I think.

However, the current suspend/resume implementation (which uses the regcache) is anyway flawed. I currently prepare a patchset to remove the regmap cache entirely and use Thierry's new generic suspend/resume functions.

Feel free to cc me. Best regards, Alexander