[PATCH] drm/i915: Improve on suspend / resume time with VT-d enabled
Thomas Hellström
thomas.hellstrom at linux.intel.com
Wed Mar 30 08:19:35 UTC 2022
When DMAR / VT-d is enabled, the display engine uses overfetching, presumably
to deal with the increased latency. To avoid display engine errors and DMAR
faults, as a workaround the GGTT is populated with scatch PTEs when VT-d
is enabled. However starting with gen10, Write-combined writing of scratch
PTES is no longer possible and as a result, populating the full GGTT with
scratch PTEs like on resume becomes very slow as uncached access is needed.
Therefore, on integrated GPUs utilize the fact that the PTEs are stored in
stolen memory which retain content across S3 suspend. Don't clear the PTEs
on suspend and resume. This improves on resume time with around 100 ms.
While 100+ms might appear like a short time it's 10% to 20% of total resume
time and important in some applications.
One notable exception is Intel Rapid Start Technology which may cause stolen
memory to be lost across what the OS percieves as S3 suspend. If IRST is
enabled or if we can't detect whether IRST is enabled, retain the old
workaround, clearing and re-instating PTEs.
As an additional measure, if we detect that the guard page PTE and the end
of GGTT was lost when resuming, print a warning and re-populate the GGTT.
On discrete GPUs, the display engine scans out from LMEM which isn't
subject to DMAR, and presumably the workaround is therefore not needed,
but that needs to be verified and disabling the workaround for dGPU,
if possible, will be deferred to a follow-up patch.
v2:
- Rely on retained ptes to also speed up suspend and resume re-binding.
- Re-build GGTT ptes if Intel rst is enabled.
v3:
- Re-build GGTT ptes also if we can't detect whether Intel rst is enabled,
and if the guard page PTE and end of GGTT was lost.
Signed-off-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
---
drivers/gpu/drm/i915/gt/intel_ggtt.c | 53 ++++++++++++++++++++++++----
drivers/gpu/drm/i915/gt/intel_gtt.h | 17 +++++++++
drivers/gpu/drm/i915/i915_driver.c | 16 +++++++++
3 files changed, 80 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index 04191fe2ee34..efd366f66906 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -23,6 +23,13 @@
#include "intel_gtt.h"
#include "gen8_ppgtt.h"
+static inline bool suspend_retains_ptes(struct i915_address_space *vm)
+{
+ return GRAPHICS_VER(vm->i915) >= 8 &&
+ !HAS_LMEM(vm->i915) &&
+ vm->is_ggtt;
+}
+
static void i915_ggtt_color_adjust(const struct drm_mm_node *node,
unsigned long color,
u64 *start,
@@ -179,7 +186,8 @@ void i915_ggtt_suspend_vm(struct i915_address_space *vm)
i915_gem_object_unlock(obj);
}
- vm->clear_range(vm, 0, vm->total);
+ if (!suspend_retains_ptes(vm))
+ vm->clear_range(vm, 0, vm->total);
vm->skip_pte_rewrite = save_skip_rewrite;
@@ -251,6 +259,22 @@ static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)
writeq(pte, addr);
}
+/*
+ * Check whether the pte pointing to the guard page at end of GGTT
+ * is still present, and warn if it wasn't.
+ */
+static bool is_guard_lost(struct i915_address_space *vm)
+{
+ struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
+ gen8_pte_t __iomem *pte;
+
+ GEM_BUG_ON(GRAPHICS_VER(vm->i915) < 8);
+
+ pte = (typeof(pte))ggtt->gsm + (vm->total / I915_GTT_PAGE_SIZE - 1);
+
+ return GEM_WARN_ON(readq(pte) != vm->scratch[0]->encode);
+}
+
static void gen8_ggtt_insert_page(struct i915_address_space *vm,
dma_addr_t addr,
u64 offset,
@@ -1180,6 +1204,8 @@ static int i915_gmch_probe(struct i915_ggtt *ggtt)
ggtt->do_idle_maps = true;
}
+ ggtt->pte_lost = true;
+
ggtt->vm.insert_page = i915_ggtt_insert_page;
ggtt->vm.insert_entries = i915_ggtt_insert_entries;
ggtt->vm.clear_range = i915_ggtt_clear_range;
@@ -1309,11 +1335,20 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm)
{
struct i915_vma *vma;
bool write_domain_objs = false;
+ bool retained_ptes;
drm_WARN_ON(&vm->i915->drm, !vm->is_ggtt && !vm->is_dpt);
- /* First fill our portion of the GTT with scratch pages */
- vm->clear_range(vm, 0, vm->total);
+ /*
+ * First fill our portion of the GTT with scratch pages if
+ * they were not retained across suspend.
+ */
+ retained_ptes = suspend_retains_ptes(vm) &&
+ !i915_vm_to_ggtt(vm)->pte_lost &&
+ !is_guard_lost(vm);
+
+ if (!retained_ptes)
+ vm->clear_range(vm, 0, vm->total);
/* clflush objects bound into the GGTT and rebind them. */
list_for_each_entry(vma, &vm->bound_list, vm_link) {
@@ -1322,9 +1357,10 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm)
atomic_read(&vma->flags) & I915_VMA_BIND_MASK;
GEM_BUG_ON(!was_bound);
- vma->ops->bind_vma(vm, NULL, vma->resource,
- obj ? obj->cache_level : 0,
- was_bound);
+ if (!retained_ptes)
+ vma->ops->bind_vma(vm, NULL, vma->resource,
+ obj ? obj->cache_level : 0,
+ was_bound);
if (obj) { /* only used during resume => exclusive access */
write_domain_objs |= fetch_and_zero(&obj->write_domain);
obj->read_domains |= I915_GEM_DOMAIN_GTT;
@@ -1352,3 +1388,8 @@ void i915_ggtt_resume(struct i915_ggtt *ggtt)
intel_ggtt_restore_fences(ggtt);
}
+
+void i915_ggtt_mark_pte_lost(struct drm_i915_private *i915, bool val)
+{
+ to_gt(i915)->ggtt->pte_lost = val;
+}
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h b/drivers/gpu/drm/i915/gt/intel_gtt.h
index 4529b5e9f6e6..1459b70bdb5a 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.h
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
@@ -345,6 +345,12 @@ struct i915_ggtt {
bool do_idle_maps;
+ /**
+ * Whether the system was recently restored from hibernate and
+ * thus may have lost pte content.
+ */
+ bool pte_lost;
+
int mtrr;
/** Bit 6 swizzling required for X tiling */
@@ -571,6 +577,17 @@ bool i915_ggtt_resume_vm(struct i915_address_space *vm);
void i915_ggtt_suspend(struct i915_ggtt *gtt);
void i915_ggtt_resume(struct i915_ggtt *ggtt);
+/**
+ * i915_ggtt_mark_pte_lost - Mark ggtt ptes as lost or clear such a marking
+ * @i915 The device private.
+ * @val whether the ptes should be marked as lost.
+ *
+ * In some cases pte content is retained across suspend, but typically lost
+ * across hibernate. Typically they should be marked as lost on
+ * hibernation restore and such marking cleared on suspend.
+ */
+void i915_ggtt_mark_pte_lost(struct drm_i915_private *i915, bool val);
+
void
fill_page_dma(struct drm_i915_gem_object *p, const u64 val, unsigned int count);
diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c
index 64e6f76861f9..f50256e4c2d2 100644
--- a/drivers/gpu/drm/i915/i915_driver.c
+++ b/drivers/gpu/drm/i915/i915_driver.c
@@ -98,6 +98,9 @@
#include "intel_region_ttm.h"
#include "vlv_suspend.h"
+/* Intel Rapid Start Technology ACPI device name */
+static const char irst_name[] = "INT3392";
+
static const struct drm_driver i915_drm_driver;
static int i915_get_bridge_dev(struct drm_i915_private *dev_priv)
@@ -1425,6 +1428,8 @@ static int i915_pm_suspend(struct device *kdev)
return -ENODEV;
}
+ i915_ggtt_mark_pte_lost(i915, false);
+
if (i915->drm.switch_power_state == DRM_SWITCH_POWER_OFF)
return 0;
@@ -1477,6 +1482,14 @@ static int i915_pm_resume(struct device *kdev)
if (i915->drm.switch_power_state == DRM_SWITCH_POWER_OFF)
return 0;
+ /*
+ * If IRST is enabled, or if we can't detect whether it's enabled,
+ * then we must assume we lost the GGTT page table entries, since
+ * they are not retained if IRST decided to enter S4.
+ */
+ if (!IS_ENABLED(CONFIG_ACPI) || acpi_dev_present(irst_name, NULL, -1))
+ i915_ggtt_mark_pte_lost(i915, true);
+
return i915_drm_resume(&i915->drm);
}
@@ -1536,6 +1549,9 @@ static int i915_pm_restore_early(struct device *kdev)
static int i915_pm_restore(struct device *kdev)
{
+ struct drm_i915_private *i915 = kdev_to_i915(kdev);
+
+ i915_ggtt_mark_pte_lost(i915, true);
return i915_pm_resume(kdev);
}
--
2.34.1
More information about the Intel-gfx-trybot
mailing list