Message ID | 1464993799-2098-1-git-send-email-stefan@agner.ch (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Fri, Jun 03, 2016 at 03:43:19PM -0700, Stefan Agner wrote: > Using flat regmap cache instead of RB-tree to avoid the following > lockdep warning on driver load: > WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0x15c/0x160() > DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) > > The RB-tree regmap cache needs to allocate new space on first > writes. However, allocations in an atomic context (e.g. when a > spinlock is held) are not allowed. The function regmap_write > calls map->lock, which acquires a spinlock in the fast_io case. > Since the FSL DCU driver uses MMIO, the regmap bus of type > regmap_mmio is being used which has fast_io set to true. > > Use flat regmap cache and specify max register to be large > enouth to cover all registers available in LS1021a and Vybrids > register space. > > Signed-off-by: Stefan Agner <stefan@agner.ch> > Cc: Mark Brown <broonie@kernel.org> > Cc: stable@vger.kernel.org > --- > While regmap cache is used for suspend/resume only (which is > broken in its current state) Mark noted that using the RB regmap > cache can also cause issues during initialization of the driver. > This patch migrates to flat regmap cache (which we can also use > to fix the issue in stable kernels), and yet another patchset > moves to the atomic suspend/resume helpers (which will not go > into stable...) While talking suspend/resume, I highly recommend drm_atomic_helper_suspend and drm_atomic_helper_resume. In my experience dumb safe/restore of registers for restoring modeset state leads to tears. -Daniel > > Dave, I saw that you just sent out the pull for rc2, will send > a pull request for rc3 early next week... > > -- > Stefan > > 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 0ec1ad9..dc723f7 100644 > --- a/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c > +++ b/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c > @@ -42,9 +42,10 @@ 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, > > .volatile_reg = fsl_dcu_drm_is_volatile_reg, > + .max_register = 0x11fc, > }; > > static int fsl_dcu_drm_irq_init(struct drm_device *dev) > -- > 2.8.2 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
On 2016-06-03 16:00, Daniel Vetter wrote: > On Fri, Jun 03, 2016 at 03:43:19PM -0700, Stefan Agner wrote: >> Using flat regmap cache instead of RB-tree to avoid the following >> lockdep warning on driver load: >> WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0x15c/0x160() >> DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) >> >> The RB-tree regmap cache needs to allocate new space on first >> writes. However, allocations in an atomic context (e.g. when a >> spinlock is held) are not allowed. The function regmap_write >> calls map->lock, which acquires a spinlock in the fast_io case. >> Since the FSL DCU driver uses MMIO, the regmap bus of type >> regmap_mmio is being used which has fast_io set to true. >> >> Use flat regmap cache and specify max register to be large >> enouth to cover all registers available in LS1021a and Vybrids >> register space. >> >> Signed-off-by: Stefan Agner <stefan@agner.ch> >> Cc: Mark Brown <broonie@kernel.org> >> Cc: stable@vger.kernel.org >> --- >> While regmap cache is used for suspend/resume only (which is >> broken in its current state) Mark noted that using the RB regmap >> cache can also cause issues during initialization of the driver. >> This patch migrates to flat regmap cache (which we can also use >> to fix the issue in stable kernels), and yet another patchset >> moves to the atomic suspend/resume helpers (which will not go >> into stable...) > > While talking suspend/resume, I highly recommend > drm_atomic_helper_suspend and drm_atomic_helper_resume. In my experience > dumb safe/restore of registers for restoring modeset state leads to tears. > -Daniel Since you commented to the patchset implementing exactly that I assume you have seen it. I still would like to have this patch applied to fix the regmap cache warning also for older kernels. -- Stefan > >> >> Dave, I saw that you just sent out the pull for rc2, will send >> a pull request for rc3 early next week... >> >> -- >> Stefan >> >> 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 0ec1ad9..dc723f7 100644 >> --- a/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c >> +++ b/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c >> @@ -42,9 +42,10 @@ 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, >> >> .volatile_reg = fsl_dcu_drm_is_volatile_reg, >> + .max_register = 0x11fc, >> }; >> >> static int fsl_dcu_drm_irq_init(struct drm_device *dev) >> -- >> 2.8.2 >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Sat, Jun 4, 2016 at 3:51 AM, Stefan Agner <stefan@agner.ch> wrote: > On 2016-06-03 16:00, Daniel Vetter wrote: >> On Fri, Jun 03, 2016 at 03:43:19PM -0700, Stefan Agner wrote: >>> Using flat regmap cache instead of RB-tree to avoid the following >>> lockdep warning on driver load: >>> WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0x15c/0x160() >>> DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) >>> >>> The RB-tree regmap cache needs to allocate new space on first >>> writes. However, allocations in an atomic context (e.g. when a >>> spinlock is held) are not allowed. The function regmap_write >>> calls map->lock, which acquires a spinlock in the fast_io case. >>> Since the FSL DCU driver uses MMIO, the regmap bus of type >>> regmap_mmio is being used which has fast_io set to true. >>> >>> Use flat regmap cache and specify max register to be large >>> enouth to cover all registers available in LS1021a and Vybrids >>> register space. >>> >>> Signed-off-by: Stefan Agner <stefan@agner.ch> >>> Cc: Mark Brown <broonie@kernel.org> >>> Cc: stable@vger.kernel.org >>> --- >>> While regmap cache is used for suspend/resume only (which is >>> broken in its current state) Mark noted that using the RB regmap >>> cache can also cause issues during initialization of the driver. >>> This patch migrates to flat regmap cache (which we can also use >>> to fix the issue in stable kernels), and yet another patchset >>> moves to the atomic suspend/resume helpers (which will not go >>> into stable...) >> >> While talking suspend/resume, I highly recommend >> drm_atomic_helper_suspend and drm_atomic_helper_resume. In my experience >> dumb safe/restore of registers for restoring modeset state leads to tears. >> -Daniel > > Since you commented to the patchset implementing exactly that I assume > you have seen it. Yup, that was pretty awesome coincidence ;-) > I still would like to have this patch applied to fix the regmap cache > warning also for older kernels. Sure, this was just a quick comment on your remark that there's still trouble left with suspend/resume. -Daniel
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 0ec1ad9..dc723f7 100644 --- a/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c +++ b/drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c @@ -42,9 +42,10 @@ 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, .volatile_reg = fsl_dcu_drm_is_volatile_reg, + .max_register = 0x11fc, }; static int fsl_dcu_drm_irq_init(struct drm_device *dev)
Using flat regmap cache instead of RB-tree to avoid the following lockdep warning on driver load: WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0x15c/0x160() DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)) The RB-tree regmap cache needs to allocate new space on first writes. However, allocations in an atomic context (e.g. when a spinlock is held) are not allowed. The function regmap_write calls map->lock, which acquires a spinlock in the fast_io case. Since the FSL DCU driver uses MMIO, the regmap bus of type regmap_mmio is being used which has fast_io set to true. Use flat regmap cache and specify max register to be large enouth to cover all registers available in LS1021a and Vybrids register space. Signed-off-by: Stefan Agner <stefan@agner.ch> Cc: Mark Brown <broonie@kernel.org> Cc: stable@vger.kernel.org --- While regmap cache is used for suspend/resume only (which is broken in its current state) Mark noted that using the RB regmap cache can also cause issues during initialization of the driver. This patch migrates to flat regmap cache (which we can also use to fix the issue in stable kernels), and yet another patchset moves to the atomic suspend/resume helpers (which will not go into stable...) Dave, I saw that you just sent out the pull for rc2, will send a pull request for rc3 early next week... -- Stefan drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)