[v2,2/3] drm/i915: noop forcewake get/put when vgpu activated

Submitted by Weinan Li on Jan. 25, 2017, 1:44 p.m.

Details

Message ID 1485351892-20012-1-git-send-email-weinan.z.li@intel.com
State New
Headers show
Series "Series without cover letter" ( rev: 1 ) in Intel GFX

Not browsing as part of any series.

Commit Message

Weinan Li Jan. 25, 2017, 1:44 p.m.
Host maintian the hardware's forcewake state, guest don't need and also
can't control it. Although vgpu_read/write bypass forcewake_get/put in MMIO
read/write, but still have separate path called by
"intel_uncore_forcewake_get/put" and
"intel_uncore_forcewake_get/put__locked". Unnecessary MMIO access in guest
waste much CPU cost. Since we full virtualize the MMIO, just noop the
forcewake get/put.

Signed-off-by: Weinan Li <weinan.z.li@intel.com>
---
 drivers/gpu/drm/i915/intel_uncore.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/i915/intel_uncore.c b/drivers/gpu/drm/i915/intel_uncore.c
index abe0888..9fad4de 100644
--- a/drivers/gpu/drm/i915/intel_uncore.c
+++ b/drivers/gpu/drm/i915/intel_uncore.c
@@ -133,6 +133,13 @@ 
 }
 
 static void
+vgpu_fw_domains_nop(struct drm_i915_private *dev_priv,
+		    enum forcewake_domains fw_domains)
+{
+	/* Guest driver doesn't need to takes care forcewake. */
+}
+
+static void
 fw_domains_posting_read(struct drm_i915_private *dev_priv)
 {
 	struct intel_uncore_forcewake_domain *d;
@@ -1374,6 +1381,12 @@  static void intel_uncore_fw_domains_init(struct drm_i915_private *dev_priv)
 		fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
 			       FORCEWAKE, FORCEWAKE_ACK);
 	}
+	if (intel_vgpu_active(dev_priv)) {
+		dev_priv->uncore.funcs.force_wake_get =
+			vgpu_fw_domains_nop;
+		dev_priv->uncore.funcs.force_wake_put =
+			vgpu_fw_domains_nop;
+	}
 
 	/* All future platforms are expected to require complex power gating */
 	WARN_ON(dev_priv->uncore.fw_domains == 0);

Comments

On Wed, Jan 25, 2017 at 09:44:51PM +0800, Weinan Li wrote:
> Host maintian the hardware's forcewake state, guest don't need and also
> can't control it. Although vgpu_read/write bypass forcewake_get/put in MMIO
> read/write, but still have separate path called by
> "intel_uncore_forcewake_get/put" and
> "intel_uncore_forcewake_get/put__locked". Unnecessary MMIO access in guest
> waste much CPU cost. Since we full virtualize the MMIO, just noop the
> forcewake get/put.

For a virtualised GPU, the host maintains the forcewake state on the
real device. As we don't control forcewake ourselves, we can simply
set force_wake_get() and force_wake_put() to be no-ops. By setting the
vfuncs, we adjust both the manual control of forcewake and around the
mmio accessors (making our vgpu specific mmio routines redundant and
to be removed in the next patch).

> Signed-off-by: Weinan Li <weinan.z.li@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris