-----Original Message----- From: Jani Nikula jani.nikula@linux.intel.com Sent: Thursday, June 16, 2022 7:58 PM To: Gupta, Anshuman anshuman.gupta@intel.com; intel- gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org Cc: Vivi, Rodrigo rodrigo.vivi@intel.com Subject: Re: [Intel-gfx] [PATCH v2 9/9] drm/i915/rpm: d3cold Policy
On Thu, 16 Jun 2022, Anshuman Gupta anshuman.gupta@intel.com wrote:
Add d3cold_sr_lmem_threshold modparam to choose between d3cold-off zero watt and d3cold-VRAM Self Refresh. i915 requires to evict the lmem objects to smem in order to support d3cold-Off.
If gfx root port is not capable of sending PME from d3cold then i915 don't need to program d3cold-off/d3cold-vram_sr sequence.
FIXME: Eviction of lmem objects in case of D3Cold off is wip.
Cc: Rodrigo Vivi rodrigo.vivi@intel.com Signed-off-by: Anshuman Gupta anshuman.gupta@intel.com
drivers/gpu/drm/i915/i915_driver.c | 27 ++++++++++++++++++++++++--- drivers/gpu/drm/i915/i915_params.c | 4 ++++ drivers/gpu/drm/i915/i915_params.h | 3 ++- 3 files changed, 30 insertions(+), 4 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c index fcff5f3fe05e..aef4b17efdbe 100644 --- a/drivers/gpu/drm/i915/i915_driver.c +++ b/drivers/gpu/drm/i915/i915_driver.c @@ -1560,15 +1560,36 @@ static int i915_pm_restore(struct device *kdev) static int intel_runtime_idle(struct device *kdev) { struct drm_i915_private *dev_priv = kdev_to_i915(kdev);
- struct pci_dev *pdev = to_pci_dev(dev_priv->drm.dev);
- u64 lmem_total = to_gt(dev_priv)->lmem->total;
- u64 lmem_avail = to_gt(dev_priv)->lmem->avail;
- u64 lmem_used = lmem_total - lmem_avail;
- struct pci_dev *root_pdev; int ret = 1;
- if (!HAS_LMEM_SR(dev_priv)) {
/*TODO: Prepare for D3Cold-Off */
- root_pdev = pcie_find_root_port(pdev);
- if (!root_pdev)
goto out;
- if (!pci_pme_capable(root_pdev, PCI_D3cold)) goto out;
}
disable_rpm_wakeref_asserts(&dev_priv->runtime_pm);
- if (lmem_used < dev_priv->params.d3cold_sr_lmem_threshold * 1024 *
- {
drm_dbg(&dev_priv->drm, "Prepare for D3Cold off\n");
pci_d3cold_enable(root_pdev);
/* FIXME: Eviction of lmem objects and guc reset is wip */
intel_pm_vram_sr(dev_priv, false);
enable_rpm_wakeref_asserts(&dev_priv->runtime_pm);
goto out;
- } else if (!HAS_LMEM_SR(dev_priv)) {
/* Disable D3Cold to reduce the eviction latency */
pci_d3cold_disable(root_pdev);
enable_rpm_wakeref_asserts(&dev_priv->runtime_pm);
goto out;
- }
This is *way* too low level code for such high level function. This needs to be abstracted better.
- ret = intel_pm_vram_sr(dev_priv, true); if (!ret) drm_dbg(&dev_priv->drm, "VRAM Self Refresh enabled\n"); diff
--git
a/drivers/gpu/drm/i915/i915_params.c b/drivers/gpu/drm/i915/i915_params.c index 701fbc98afa0..6c6b3c372d4d 100644 --- a/drivers/gpu/drm/i915/i915_params.c +++ b/drivers/gpu/drm/i915/i915_params.c @@ -197,6 +197,10 @@ i915_param_named(enable_gvt, bool, 0400, "Enable support for Intel GVT-g graphics virtualization host support(default:false)"); #endif
+i915_param_named_unsafe(d3cold_sr_lmem_threshold, int, 0400,
- "Enable Vidoe RAM Self refresh when size of lmem is greater to this
threshold. "
- "It helps to optimize the suspend/resume latecy. (default: 300mb)");
#if CONFIG_DRM_I915_REQUEST_TIMEOUT i915_param_named_unsafe(request_timeout_ms, uint, 0600, "Default request/fence/batch buffer expiration
timeout."); diff
--git a/drivers/gpu/drm/i915/i915_params.h b/drivers/gpu/drm/i915/i915_params.h index b5e7ea45d191..28f20ebaf41f 100644 --- a/drivers/gpu/drm/i915/i915_params.h +++ b/drivers/gpu/drm/i915/i915_params.h @@ -83,7 +83,8 @@ struct drm_printer; param(bool, verbose_state_checks, true, 0) \ param(bool, nuclear_pageflip, false, 0400) \ param(bool, enable_dp_mst, true, 0600) \
- param(bool, enable_gvt, false, IS_ENABLED(CONFIG_DRM_I915_GVT) ?
0400 : 0)
- param(bool, enable_gvt, false, IS_ENABLED(CONFIG_DRM_I915_GVT) ?
0400 : 0) \
- param(int, d3cold_sr_lmem_threshold, 300, 0600) \
What's the point of the parameter?
We want a configurable option to choose an optimum lmem usages threshold on which, i915 can choose lmem self-refresh. This threshold value would require some profiling as well. That is the reason this threshold kept as module param.
Also, please read the comment /* leave bools at the end to not create holes */ above.
Thanks for review comment , I will re order the module param. Regards , Anshuman Gupta.
BR, Jani.
#define MEMBER(T, member, ...) T member; struct i915_params {
-- Jani Nikula, Intel Open Source Graphics Center