[Intel-gfx] [PATCH v2 04/12] drm/i915: vgpu ppgtt page table pv support
Xiaolin Zhang
xiaolin.zhang at intel.com
Fri Oct 9 00:04:35 UTC 2020
to improve efficiency and reduce the complexsity of vgpu ppgtt support,
vgpu ppgtt page table operations are implemented in pv fashion and
implemented pv version of bind/unbind for ppgtt vma ops.
The pv version of ppgtt vma ops use the CTB protocol to communicate
pv ppgtt command along with data struct pv_vma from guest to GVT
and then GVT will implement command handler of PV_CMD_BIND_PPGTT and
PV_CMD_UBIND_PPGTT to support vgpu PPGTT feature.
new PV_PPGTT pv_cap is used to control this level of pv support in
both guest and host side.
Signed-off-by: Xiaolin Zhang <xiaolin.zhang at intel.com>
---
drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 4 +-
drivers/gpu/drm/i915/i915_vgpu.c | 96 ++++++++++++++++++++++++++++++++++++
drivers/gpu/drm/i915/i915_vgpu.h | 17 +++++++
3 files changed, 116 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
index b236aa0..d5c6007 100644
--- a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
@@ -732,8 +732,10 @@ struct i915_ppgtt *gen8_ppgtt_create(struct intel_gt *gt)
ppgtt->vm.pte_encode = gen8_pte_encode;
- if (intel_vgpu_active(gt->i915))
+ if (intel_vgpu_active(gt->i915)) {
+ intel_vgpu_pv_config_caps(gt->i915, PV_PPGTT, ppgtt);
gen8_ppgtt_notify_vgt(ppgtt, true);
+ }
ppgtt->vm.cleanup = gen8_ppgtt_cleanup;
diff --git a/drivers/gpu/drm/i915/i915_vgpu.c b/drivers/gpu/drm/i915/i915_vgpu.c
index c833823..81be1db5 100644
--- a/drivers/gpu/drm/i915/i915_vgpu.c
+++ b/drivers/gpu/drm/i915/i915_vgpu.c
@@ -99,6 +99,9 @@ void intel_vgpu_detect(struct drm_i915_private *dev_priv)
dev_priv->vgpu.active = true;
mutex_init(&dev_priv->vgpu.lock);
+ /* guest driver PV capability */
+ dev_priv->vgpu.pv_caps = PV_PPGTT;
+
if (intel_vgpu_pv_detect_caps(dev_priv, shared_area)) {
drm_info(&dev_priv->drm,
"Virtual GPU for Intel GVT-g detected with PV Capabilities.\n");
@@ -373,6 +376,91 @@ int intel_vgt_balloon(struct i915_ggtt *ggtt)
* i915 vgpu PV support for Linux
*/
+static int intel_vgpu_pv_vma_action(struct i915_address_space *vm,
+ struct i915_vma *vma, u32 action,
+ u64 flags, u64 pte_flag)
+{
+ struct drm_i915_private *i915 = vma->vm->i915;
+ struct sgt_iter sgt_iter;
+ dma_addr_t addr;
+ struct intel_vgpu_pv_vma pvvma;
+ u32 num_pages;
+ u64 *gpas;
+ int i = 0;
+ u32 data[32];
+ int ret;
+ u32 size = sizeof(pvvma) / 4;
+
+ if (1 + size > ARRAY_SIZE(data))
+ return -EIO;
+
+ memset(&pvvma, 0, sizeof(pvvma));
+ num_pages = vma->node.size >> PAGE_SHIFT;
+ pvvma.size = num_pages;
+ pvvma.start = vma->node.start;
+ pvvma.flags = flags;
+
+ if (action == PV_CMD_BIND_PPGTT || action == PV_CMD_UNBIND_PPGTT)
+ pvvma.pml4 = px_dma(i915_vm_to_ppgtt(vm)->pd);
+
+ if (num_pages == 1) {
+ pvvma.dma_addrs = vma->pages->sgl->dma_address | pte_flag;
+ goto out;
+ }
+
+ gpas = kmalloc_array(num_pages, sizeof(u64), GFP_KERNEL);
+ if (!gpas)
+ return -ENOMEM;
+
+ pvvma.dma_addrs = virt_to_phys((void *)gpas);
+ for_each_sgt_daddr(addr, sgt_iter, vma->pages)
+ gpas[i++] = addr | pte_flag;
+
+ /* Fill the allocated but "unused" space beyond the end of the buffer */
+ while (i < num_pages)
+ gpas[i++] = vm->scratch[0]->encode;
+out:
+ data[0] = action;
+ memcpy(&data[1], &pvvma, sizeof(pvvma));
+ ret = i915->vgpu.pv->send(i915, data, 1 + size);
+
+ if (num_pages > 1)
+ kfree(gpas);
+
+ return ret;
+}
+
+static void intel_vgpu_pv_ppgtt_bind(struct i915_address_space *vm,
+ struct i915_vm_pt_stash *stash,
+ struct i915_vma *vma,
+ enum i915_cache_level cache_level,
+ u32 flags)
+{
+ u32 pte_flags;
+ u64 pte_encode;
+
+ if (!test_bit(I915_VMA_ALLOC_BIT, __i915_vma_flags(vma))) {
+ set_bit(I915_VMA_ALLOC_BIT, __i915_vma_flags(vma));
+ flags |= BIT(I915_VMA_ALLOC_BIT);
+ }
+
+ /* Applicable to VLV, and gen8+ */
+ pte_flags = 0;
+ if (i915_gem_object_is_readonly(vma->obj))
+ pte_flags |= PTE_READ_ONLY;
+
+ pte_encode = vma->vm->pte_encode(0, cache_level, pte_flags);
+
+ intel_vgpu_pv_vma_action(vm, vma, PV_CMD_BIND_PPGTT, flags, pte_encode);
+}
+
+static void intel_vgpu_pv_ppgtt_unbind(struct i915_address_space *vm,
+ struct i915_vma *vma)
+{
+ if (test_and_clear_bit(I915_VMA_ALLOC_BIT, __i915_vma_flags(vma)))
+ intel_vgpu_pv_vma_action(vm, vma, PV_CMD_UNBIND_PPGTT, 0, 0);
+}
+
/**
* intel_vgpu_pv_wait_desc_update - Wait the command buffer descriptor update.
* @desc: buffer descriptor
@@ -655,8 +743,16 @@ static int intel_vgpu_pv_setup_shared_page(struct drm_i915_private *i915,
void intel_vgpu_pv_config_caps(struct drm_i915_private *i915,
enum intel_vgpu_pv_caps cap, void *data)
{
+ struct i915_ppgtt *ppgtt;
+
if (!intel_vgpu_pv_check_cap(i915, cap))
return;
+
+ if (cap == PV_PPGTT) {
+ ppgtt = (struct i915_ppgtt *)data;
+ ppgtt->vm.vma_ops.bind_vma = intel_vgpu_pv_ppgtt_bind;
+ ppgtt->vm.vma_ops.unbind_vma = intel_vgpu_pv_ppgtt_unbind;
+ }
}
/*
diff --git a/drivers/gpu/drm/i915/i915_vgpu.h b/drivers/gpu/drm/i915/i915_vgpu.h
index 18f2dd0..49fc174 100644
--- a/drivers/gpu/drm/i915/i915_vgpu.h
+++ b/drivers/gpu/drm/i915/i915_vgpu.h
@@ -37,6 +37,14 @@ struct i915_ggtt;
/* intel vGPU PV capabilities */
enum intel_vgpu_pv_caps {
PV_NONE = 0,
+ PV_PPGTT = BIT(0),
+};
+
+/* vgpu PV commands */
+enum intel_vgpu_pv_cmd {
+ PV_CMD_DEFAULT = 0x0,
+ PV_CMD_BIND_PPGTT,
+ PV_CMD_UNBIND_PPGTT,
};
/* A shared memory(4KB) between GVTg and vgpu allocated by guest */
@@ -45,6 +53,15 @@ struct intel_vgpu_pv_shared_page {
u16 ver_minor;
};
+/* PV virtual memory address for GGTT/PPGTT */
+struct intel_vgpu_pv_vma {
+ u32 size; /* num of pages */
+ u32 flags; /* bind or unbind flags */
+ u64 start; /* start of virtual address */
+ u64 dma_addrs; /* BO's dma address list */
+ u64 pml4; /* ppgtt handler */
+};
+
/*
* Definition of the command transport message header (DW0)
*
--
2.7.4
More information about the Intel-gfx
mailing list