[Intel-gfx] [PATCH] drm/i915: enable memory self refresh on 945GM
Li Peng
peng.li at linux.intel.com
Wed Dec 2 14:29:50 CET 2009
On Wed, 2009-12-02 at 11:11 +0800, Fu Michael wrote:
> Li Peng wrote:
> > I did some test and found that memory self refresh on 945GM
> > isn't really enabled in commit 652c39. On 945, we need to use
> > bit 31 of FW_BLC_SELF to enable the write to self refresh bit
> > and bit 16 to enable the write of self refresh watermark.
> >
> > This patch enables memory self refresh on 945GM when graphics
> > is idle, and disable self refresh when it is busy. It can save
> > about 0.2W idle power on the netbook with 945GSE chipset.
> >
> > Signed-off-by: Li Peng <peng.li at intel.com>
> > ---
> > drivers/gpu/drm/i915/i915_reg.h | 4 +++-
> > drivers/gpu/drm/i915/intel_display.c | 23 ++++++++++++++++++++++-
> > 2 files changed, 25 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> > index 1687edf..93b5077 100644
> > --- a/drivers/gpu/drm/i915/i915_reg.h
> > +++ b/drivers/gpu/drm/i915/i915_reg.h
> > @@ -301,7 +301,9 @@
> > #define FW_BLC 0x020d8
> > #define FW_BLC2 0x020dc
> > #define FW_BLC_SELF 0x020e0 /* 915+ only */
> > -#define FW_BLC_SELF_EN (1<<15)
> > +#define FW_BLC_SELF_EN_MASK (1<<31)
> > +#define FW_BLC_SELF_FIFO_MASK (1<<16)
> > +#define FW_BLC_SELF_EN (1<<15)
> > #define MM_BURST_LENGTH 0x00700000
> > #define MM_FIFO_WATERMARK 0x0001F000
> > #define LM_BURST_LENGTH 0x00000700
> > diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> > index 3ba6546..9b96439 100644
> > --- a/drivers/gpu/drm/i915/intel_display.c
> > +++ b/drivers/gpu/drm/i915/intel_display.c
> > @@ -2580,7 +2580,10 @@ static void i9xx_update_wm(struct drm_device *dev, int planea_clock,
> > srwm = total_size - sr_entries;
> > if (srwm < 0)
> > srwm = 1;
> > - I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN | (srwm & 0x3f));
> > + if (IS_I945GM(dev))
> > + I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
> > + else
> > + I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN | (srwm & 0x3f));
> > }
> >
> > DRM_DEBUG("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
> >
> 915GM's SR_EN bit is in INSTPM:12 rather than in FW_BLC_SELF reg.
>
Thanks for pointing out, I will fix it.
> There seems no need to explicitly disable SR bit. system should be able
> to use it automatically when condition is met. Do you see any issue
> without below hunk?
I have tried that, just enabling SR bit and setting srwm in
i9xx_update_wm(), but X got frozen after boot, so I add the
code to disable SR explicitly. Maybe H/W isn't that smart :)
>
> > @@ -3671,11 +3674,23 @@ static void intel_gpu_idle_timer(unsigned long arg)
> > void intel_increase_renderclock(struct drm_device *dev, bool schedule)
> > {
> > drm_i915_private_t *dev_priv = dev->dev_private;
> > + u32 fw_blc_self;
> > + int need_schedule = 0;
> >
> > if (IS_IGDNG(dev))
> > return;
> >
> > + if (IS_I945GM(dev)) {
> > + DRM_DEBUG("disable memory self refresh on 945\n");
> > + fw_blc_self = I915_READ(FW_BLC_SELF);
> > + fw_blc_self &= ~FW_BLC_SELF_EN;
> > + I915_WRITE(FW_BLC_SELF, fw_blc_self | FW_BLC_SELF_EN_MASK);
> > + need_schedule = 1;
> > + }
> > +
> > if (!dev_priv->render_reclock_avail) {
> > + if (need_schedule)
> > + goto schedule_timer;
> > DRM_DEBUG("not reclocking render clock\n");
> > return;
> > }
> > @@ -3688,6 +3703,7 @@ void intel_increase_renderclock(struct drm_device *dev, bool schedule)
> > DRM_DEBUG("increasing render clock frequency\n");
> >
> > /* Schedule downclock */
> > +schedule_timer:
> > if (schedule)
> > mod_timer(&dev_priv->idle_timer, jiffies +
> > msecs_to_jiffies(GPU_IDLE_TIMEOUT));
> > @@ -3700,6 +3716,11 @@ void intel_decrease_renderclock(struct drm_device *dev)
> > if (IS_IGDNG(dev))
> > return;
> >
> > + if (IS_I945GM(dev)) {
> > + DRM_DEBUG("enable memory self refresh on 945\n");
> > + I915_WRITE(FW_BLC_SELF, FW_BLC_SELF_EN_MASK | FW_BLC_SELF_EN);
> > + }
> > +
> > if (!dev_priv->render_reclock_avail) {
> > DRM_DEBUG("not reclocking render clock\n");
> > return;
> >
>
More information about the Intel-gfx
mailing list