summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/i915/gvt/gtt.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/i915/gvt/gtt.c')
-rw-r--r--drivers/gpu/drm/i915/gvt/gtt.c1479
1 files changed, 680 insertions, 799 deletions
diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c
index 8d5317d0122d..d29281231507 100644
--- a/drivers/gpu/drm/i915/gvt/gtt.c
+++ b/drivers/gpu/drm/i915/gvt/gtt.c
@@ -38,6 +38,12 @@
#include "i915_pvinfo.h"
#include "trace.h"
+#if defined(VERBOSE_DEBUG)
+#define gvt_vdbg_mm(fmt, args...) gvt_dbg_mm(fmt, ##args)
+#else
+#define gvt_vdbg_mm(fmt, args...)
+#endif
+
static bool enable_out_of_sync = false;
static int preallocated_oos_pages = 8192;
@@ -264,7 +270,7 @@ static u64 read_pte64(struct drm_i915_private *dev_priv, unsigned long index)
return readq(addr);
}
-static void gtt_invalidate(struct drm_i915_private *dev_priv)
+static void ggtt_invalidate(struct drm_i915_private *dev_priv)
{
mmio_hw_access_pre(dev_priv);
I915_WRITE(GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN);
@@ -331,20 +337,20 @@ static inline int gtt_set_entry64(void *pt,
#define GTT_HAW 46
-#define ADDR_1G_MASK (((1UL << (GTT_HAW - 30)) - 1) << 30)
-#define ADDR_2M_MASK (((1UL << (GTT_HAW - 21)) - 1) << 21)
-#define ADDR_4K_MASK (((1UL << (GTT_HAW - 12)) - 1) << 12)
+#define ADDR_1G_MASK GENMASK_ULL(GTT_HAW - 1, 30)
+#define ADDR_2M_MASK GENMASK_ULL(GTT_HAW - 1, 21)
+#define ADDR_4K_MASK GENMASK_ULL(GTT_HAW - 1, 12)
static unsigned long gen8_gtt_get_pfn(struct intel_gvt_gtt_entry *e)
{
unsigned long pfn;
if (e->type == GTT_TYPE_PPGTT_PTE_1G_ENTRY)
- pfn = (e->val64 & ADDR_1G_MASK) >> 12;
+ pfn = (e->val64 & ADDR_1G_MASK) >> PAGE_SHIFT;
else if (e->type == GTT_TYPE_PPGTT_PTE_2M_ENTRY)
- pfn = (e->val64 & ADDR_2M_MASK) >> 12;
+ pfn = (e->val64 & ADDR_2M_MASK) >> PAGE_SHIFT;
else
- pfn = (e->val64 & ADDR_4K_MASK) >> 12;
+ pfn = (e->val64 & ADDR_4K_MASK) >> PAGE_SHIFT;
return pfn;
}
@@ -352,16 +358,16 @@ static void gen8_gtt_set_pfn(struct intel_gvt_gtt_entry *e, unsigned long pfn)
{
if (e->type == GTT_TYPE_PPGTT_PTE_1G_ENTRY) {
e->val64 &= ~ADDR_1G_MASK;
- pfn &= (ADDR_1G_MASK >> 12);
+ pfn &= (ADDR_1G_MASK >> PAGE_SHIFT);
} else if (e->type == GTT_TYPE_PPGTT_PTE_2M_ENTRY) {
e->val64 &= ~ADDR_2M_MASK;
- pfn &= (ADDR_2M_MASK >> 12);
+ pfn &= (ADDR_2M_MASK >> PAGE_SHIFT);
} else {
e->val64 &= ~ADDR_4K_MASK;
- pfn &= (ADDR_4K_MASK >> 12);
+ pfn &= (ADDR_4K_MASK >> PAGE_SHIFT);
}
- e->val64 |= (pfn << 12);
+ e->val64 |= (pfn << PAGE_SHIFT);
}
static bool gen8_gtt_test_pse(struct intel_gvt_gtt_entry *e)
@@ -371,7 +377,7 @@ static bool gen8_gtt_test_pse(struct intel_gvt_gtt_entry *e)
return false;
e->type = get_entry_type(e->type);
- if (!(e->val64 & BIT(7)))
+ if (!(e->val64 & _PAGE_PSE))
return false;
e->type = get_pse_type(e->type);
@@ -389,17 +395,17 @@ static bool gen8_gtt_test_present(struct intel_gvt_gtt_entry *e)
|| e->type == GTT_TYPE_PPGTT_ROOT_L4_ENTRY)
return (e->val64 != 0);
else
- return (e->val64 & BIT(0));
+ return (e->val64 & _PAGE_PRESENT);
}
static void gtt_entry_clear_present(struct intel_gvt_gtt_entry *e)
{
- e->val64 &= ~BIT(0);
+ e->val64 &= ~_PAGE_PRESENT;
}
static void gtt_entry_set_present(struct intel_gvt_gtt_entry *e)
{
- e->val64 |= BIT(0);
+ e->val64 |= _PAGE_PRESENT;
}
/*
@@ -447,58 +453,91 @@ static struct intel_gvt_gtt_gma_ops gen8_gtt_gma_ops = {
.gma_to_pml4_index = gen8_gma_to_pml4_index,
};
-static int gtt_entry_p2m(struct intel_vgpu *vgpu, struct intel_gvt_gtt_entry *p,
- struct intel_gvt_gtt_entry *m)
+/*
+ * MM helpers.
+ */
+static void _ppgtt_get_root_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index,
+ bool guest)
{
- struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
- unsigned long gfn, mfn;
+ struct intel_gvt_gtt_pte_ops *pte_ops = mm->vgpu->gvt->gtt.pte_ops;
- *m = *p;
+ GEM_BUG_ON(mm->type != INTEL_GVT_MM_PPGTT);
- if (!ops->test_present(p))
- return 0;
+ entry->type = mm->ppgtt_mm.root_entry_type;
+ pte_ops->get_entry(guest ? mm->ppgtt_mm.guest_pdps :
+ mm->ppgtt_mm.shadow_pdps,
+ entry, index, false, 0, mm->vgpu);
- gfn = ops->get_pfn(p);
+ pte_ops->test_pse(entry);
+}
- mfn = intel_gvt_hypervisor_gfn_to_mfn(vgpu, gfn);
- if (mfn == INTEL_GVT_INVALID_ADDR) {
- gvt_vgpu_err("fail to translate gfn: 0x%lx\n", gfn);
- return -ENXIO;
- }
+static inline void ppgtt_get_guest_root_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index)
+{
+ _ppgtt_get_root_entry(mm, entry, index, true);
+}
- ops->set_pfn(m, mfn);
- return 0;
+static inline void ppgtt_get_shadow_root_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index)
+{
+ _ppgtt_get_root_entry(mm, entry, index, false);
}
-/*
- * MM helpers.
- */
-int intel_vgpu_mm_get_entry(struct intel_vgpu_mm *mm,
- void *page_table, struct intel_gvt_gtt_entry *e,
- unsigned long index)
+static void _ppgtt_set_root_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index,
+ bool guest)
{
- struct intel_gvt *gvt = mm->vgpu->gvt;
- struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops;
- int ret;
+ struct intel_gvt_gtt_pte_ops *pte_ops = mm->vgpu->gvt->gtt.pte_ops;
- e->type = mm->page_table_entry_type;
+ pte_ops->set_entry(guest ? mm->ppgtt_mm.guest_pdps :
+ mm->ppgtt_mm.shadow_pdps,
+ entry, index, false, 0, mm->vgpu);
+}
- ret = ops->get_entry(page_table, e, index, false, 0, mm->vgpu);
- if (ret)
- return ret;
+static inline void ppgtt_set_guest_root_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index)
+{
+ _ppgtt_set_root_entry(mm, entry, index, true);
+}
- ops->test_pse(e);
- return 0;
+static inline void ppgtt_set_shadow_root_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index)
+{
+ _ppgtt_set_root_entry(mm, entry, index, false);
}
-int intel_vgpu_mm_set_entry(struct intel_vgpu_mm *mm,
- void *page_table, struct intel_gvt_gtt_entry *e,
- unsigned long index)
+static void ggtt_get_guest_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index)
{
- struct intel_gvt *gvt = mm->vgpu->gvt;
- struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops;
+ struct intel_gvt_gtt_pte_ops *pte_ops = mm->vgpu->gvt->gtt.pte_ops;
+
+ GEM_BUG_ON(mm->type != INTEL_GVT_MM_GGTT);
+
+ entry->type = GTT_TYPE_GGTT_PTE;
+ pte_ops->get_entry(mm->ggtt_mm.virtual_ggtt, entry, index,
+ false, 0, mm->vgpu);
+}
+
+static void ggtt_set_guest_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index)
+{
+ struct intel_gvt_gtt_pte_ops *pte_ops = mm->vgpu->gvt->gtt.pte_ops;
- return ops->set_entry(page_table, e, index, false, 0, mm->vgpu);
+ GEM_BUG_ON(mm->type != INTEL_GVT_MM_GGTT);
+
+ pte_ops->set_entry(mm->ggtt_mm.virtual_ggtt, entry, index,
+ false, 0, mm->vgpu);
+}
+
+static void ggtt_set_host_entry(struct intel_vgpu_mm *mm,
+ struct intel_gvt_gtt_entry *entry, unsigned long index)
+{
+ struct intel_gvt_gtt_pte_ops *pte_ops = mm->vgpu->gvt->gtt.pte_ops;
+
+ GEM_BUG_ON(mm->type != INTEL_GVT_MM_GGTT);
+
+ pte_ops->set_entry(NULL, entry, index, false, 0, mm->vgpu);
}
/*
@@ -520,12 +559,15 @@ static inline int ppgtt_spt_get_entry(
return -EINVAL;
ret = ops->get_entry(page_table, e, index, guest,
- spt->guest_page.track.gfn << I915_GTT_PAGE_SHIFT,
+ spt->guest_page.gfn << I915_GTT_PAGE_SHIFT,
spt->vgpu);
if (ret)
return ret;
ops->test_pse(e);
+
+ gvt_vdbg_mm("read ppgtt entry, spt type %d, entry type %d, index %lu, value %llx\n",
+ type, e->type, index, e->val64);
return 0;
}
@@ -541,18 +583,21 @@ static inline int ppgtt_spt_set_entry(
if (WARN(!gtt_type_is_entry(e->type), "invalid entry type\n"))
return -EINVAL;
+ gvt_vdbg_mm("set ppgtt entry, spt type %d, entry type %d, index %lu, value %llx\n",
+ type, e->type, index, e->val64);
+
return ops->set_entry(page_table, e, index, guest,
- spt->guest_page.track.gfn << I915_GTT_PAGE_SHIFT,
+ spt->guest_page.gfn << I915_GTT_PAGE_SHIFT,
spt->vgpu);
}
#define ppgtt_get_guest_entry(spt, e, index) \
ppgtt_spt_get_entry(spt, NULL, \
- spt->guest_page_type, e, index, true)
+ spt->guest_page.type, e, index, true)
#define ppgtt_set_guest_entry(spt, e, index) \
ppgtt_spt_set_entry(spt, NULL, \
- spt->guest_page_type, e, index, true)
+ spt->guest_page.type, e, index, true)
#define ppgtt_get_shadow_entry(spt, e, index) \
ppgtt_spt_get_entry(spt, spt->shadow_page.vaddr, \
@@ -562,159 +607,6 @@ static inline int ppgtt_spt_set_entry(
ppgtt_spt_set_entry(spt, spt->shadow_page.vaddr, \
spt->shadow_page.type, e, index, false)
-/**
- * intel_vgpu_init_page_track - init a page track data structure
- * @vgpu: a vGPU
- * @t: a page track data structure
- * @gfn: guest memory page frame number
- * @handler: the function will be called when target guest memory page has
- * been modified.
- *
- * This function is called when a user wants to prepare a page track data
- * structure to track a guest memory page.
- *
- * Returns:
- * Zero on success, negative error code if failed.
- */
-int intel_vgpu_init_page_track(struct intel_vgpu *vgpu,
- struct intel_vgpu_page_track *t,
- unsigned long gfn,
- int (*handler)(void *, u64, void *, int),
- void *data)
-{
- INIT_HLIST_NODE(&t->node);
-
- t->tracked = false;
- t->gfn = gfn;
- t->handler = handler;
- t->data = data;
-
- hash_add(vgpu->gtt.tracked_guest_page_hash_table, &t->node, t->gfn);
- return 0;
-}
-
-/**
- * intel_vgpu_clean_page_track - release a page track data structure
- * @vgpu: a vGPU
- * @t: a page track data structure
- *
- * This function is called before a user frees a page track data structure.
- */
-void intel_vgpu_clean_page_track(struct intel_vgpu *vgpu,
- struct intel_vgpu_page_track *t)
-{
- if (!hlist_unhashed(&t->node))
- hash_del(&t->node);
-
- if (t->tracked)
- intel_gvt_hypervisor_disable_page_track(vgpu, t);
-}
-
-/**
- * intel_vgpu_find_tracked_page - find a tracked guest page
- * @vgpu: a vGPU
- * @gfn: guest memory page frame number
- *
- * This function is called when the emulation layer wants to figure out if a
- * trapped GFN is a tracked guest page.
- *
- * Returns:
- * Pointer to page track data structure, NULL if not found.
- */
-struct intel_vgpu_page_track *intel_vgpu_find_tracked_page(
- struct intel_vgpu *vgpu, unsigned long gfn)
-{
- struct intel_vgpu_page_track *t;
-
- hash_for_each_possible(vgpu->gtt.tracked_guest_page_hash_table,
- t, node, gfn) {
- if (t->gfn == gfn)
- return t;
- }
- return NULL;
-}
-
-static int init_guest_page(struct intel_vgpu *vgpu,
- struct intel_vgpu_guest_page *p,
- unsigned long gfn,
- int (*handler)(void *, u64, void *, int),
- void *data)
-{
- p->oos_page = NULL;
- p->write_cnt = 0;
-
- return intel_vgpu_init_page_track(vgpu, &p->track, gfn, handler, data);
-}
-
-static int detach_oos_page(struct intel_vgpu *vgpu,
- struct intel_vgpu_oos_page *oos_page);
-
-static void clean_guest_page(struct intel_vgpu *vgpu,
- struct intel_vgpu_guest_page *p)
-{
- if (p->oos_page)
- detach_oos_page(vgpu, p->oos_page);
-
- intel_vgpu_clean_page_track(vgpu, &p->track);
-}
-
-static inline int init_shadow_page(struct intel_vgpu *vgpu,
- struct intel_vgpu_shadow_page *p, int type, bool hash)
-{
- struct device *kdev = &vgpu->gvt->dev_priv->drm.pdev->dev;
- dma_addr_t daddr;
-
- daddr = dma_map_page(kdev, p->page, 0, 4096, PCI_DMA_BIDIRECTIONAL);
- if (dma_mapping_error(kdev, daddr)) {
- gvt_vgpu_err("fail to map dma addr\n");
- return -EINVAL;
- }
-
- p->vaddr = page_address(p->page);
- p->type = type;
-
- INIT_HLIST_NODE(&p->node);
-
- p->mfn = daddr >> I915_GTT_PAGE_SHIFT;
- if (hash)
- hash_add(vgpu->gtt.shadow_page_hash_table, &p->node, p->mfn);
- return 0;
-}
-
-static inline void clean_shadow_page(struct intel_vgpu *vgpu,
- struct intel_vgpu_shadow_page *p)
-{
- struct device *kdev = &vgpu->gvt->dev_priv->drm.pdev->dev;
-
- dma_unmap_page(kdev, p->mfn << I915_GTT_PAGE_SHIFT, 4096,
- PCI_DMA_BIDIRECTIONAL);
-
- if (!hlist_unhashed(&p->node))
- hash_del(&p->node);
-}
-
-static inline struct intel_vgpu_shadow_page *find_shadow_page(
- struct intel_vgpu *vgpu, unsigned long mfn)
-{
- struct intel_vgpu_shadow_page *p;
-
- hash_for_each_possible(vgpu->gtt.shadow_page_hash_table,
- p, node, mfn) {
- if (p->mfn == mfn)
- return p;
- }
- return NULL;
-}
-
-#define page_track_to_guest_page(ptr) \
- container_of(ptr, struct intel_vgpu_guest_page, track)
-
-#define guest_page_to_ppgtt_spt(ptr) \
- container_of(ptr, struct intel_vgpu_ppgtt_spt, guest_page)
-
-#define shadow_page_to_ppgtt_spt(ptr) \
- container_of(ptr, struct intel_vgpu_ppgtt_spt, shadow_page)
-
static void *alloc_spt(gfp_t gfp_mask)
{
struct intel_vgpu_ppgtt_spt *spt;
@@ -737,63 +629,96 @@ static void free_spt(struct intel_vgpu_ppgtt_spt *spt)
kfree(spt);
}
-static void ppgtt_free_shadow_page(struct intel_vgpu_ppgtt_spt *spt)
+static int detach_oos_page(struct intel_vgpu *vgpu,
+ struct intel_vgpu_oos_page *oos_page);
+
+static void ppgtt_free_spt(struct intel_vgpu_ppgtt_spt *spt)
{
- trace_spt_free(spt->vgpu->id, spt, spt->shadow_page.type);
+ struct device *kdev = &spt->vgpu->gvt->dev_priv->drm.pdev->dev;
- clean_shadow_page(spt->vgpu, &spt->shadow_page);
- clean_guest_page(spt->vgpu, &spt->guest_page);
- list_del_init(&spt->post_shadow_list);
+ trace_spt_free(spt->vgpu->id, spt, spt->guest_page.type);
+
+ dma_unmap_page(kdev, spt->shadow_page.mfn << I915_GTT_PAGE_SHIFT, 4096,
+ PCI_DMA_BIDIRECTIONAL);
+
+ radix_tree_delete(&spt->vgpu->gtt.spt_tree, spt->shadow_page.mfn);
+
+ if (spt->guest_page.oos_page)
+ detach_oos_page(spt->vgpu, spt->guest_page.oos_page);
+
+ intel_vgpu_unregister_page_track(spt->vgpu, spt->guest_page.gfn);
+ list_del_init(&spt->post_shadow_list);
free_spt(spt);
}
-static void ppgtt_free_all_shadow_page(struct intel_vgpu *vgpu)
+static void ppgtt_free_all_spt(struct intel_vgpu *vgpu)
{
- struct hlist_node *n;
- struct intel_vgpu_shadow_page *sp;
- int i;
+ struct intel_vgpu_ppgtt_spt *spt;
+ struct radix_tree_iter iter;
+ void **slot;
- hash_for_each_safe(vgpu->gtt.shadow_page_hash_table, i, n, sp, node)
- ppgtt_free_shadow_page(shadow_page_to_ppgtt_spt(sp));
+ radix_tree_for_each_slot(slot, &vgpu->gtt.spt_tree, &iter, 0) {
+ spt = radix_tree_deref_slot(slot);
+ ppgtt_free_spt(spt);
+ }
}
static int ppgtt_handle_guest_write_page_table_bytes(
- struct intel_vgpu_guest_page *gpt,
+ struct intel_vgpu_ppgtt_spt *spt,
u64 pa, void *p_data, int bytes);
-static int ppgtt_write_protection_handler(void *data, u64 pa,
- void *p_data, int bytes)
+static int ppgtt_write_protection_handler(
+ struct intel_vgpu_page_track *page_track,
+ u64 gpa, void *data, int bytes)
{
- struct intel_vgpu_page_track *t = data;
- struct intel_vgpu_guest_page *p = page_track_to_guest_page(t);
+ struct intel_vgpu_ppgtt_spt *spt = page_track->priv_data;
+
int ret;
if (bytes != 4 && bytes != 8)
return -EINVAL;
- if (!t->tracked)
- return -EINVAL;
-
- ret = ppgtt_handle_guest_write_page_table_bytes(p,
- pa, p_data, bytes);
+ ret = ppgtt_handle_guest_write_page_table_bytes(spt, gpa, data, bytes);
if (ret)
return ret;
return ret;
}
-static int reclaim_one_mm(struct intel_gvt *gvt);
+/* Find a spt by guest gfn. */
+static struct intel_vgpu_ppgtt_spt *intel_vgpu_find_spt_by_gfn(
+ struct intel_vgpu *vgpu, unsigned long gfn)
+{
+ struct intel_vgpu_page_track *track;
+
+ track = intel_vgpu_find_page_track(vgpu, gfn);
+ if (track && track->handler == ppgtt_write_protection_handler)
+ return track->priv_data;
+
+ return NULL;
+}
+
+/* Find the spt by shadow page mfn. */
+static inline struct intel_vgpu_ppgtt_spt *intel_vgpu_find_spt_by_mfn(
+ struct intel_vgpu *vgpu, unsigned long mfn)
+{
+ return radix_tree_lookup(&vgpu->gtt.spt_tree, mfn);
+}
-static struct intel_vgpu_ppgtt_spt *ppgtt_alloc_shadow_page(
+static int reclaim_one_ppgtt_mm(struct intel_gvt *gvt);
+
+static struct intel_vgpu_ppgtt_spt *ppgtt_alloc_spt(
struct intel_vgpu *vgpu, int type, unsigned long gfn)
{
+ struct device *kdev = &vgpu->gvt->dev_priv->drm.pdev->dev;
struct intel_vgpu_ppgtt_spt *spt = NULL;
+ dma_addr_t daddr;
int ret;
retry:
spt = alloc_spt(GFP_KERNEL | __GFP_ZERO);
if (!spt) {
- if (reclaim_one_mm(vgpu->gvt))
+ if (reclaim_one_ppgtt_mm(vgpu->gvt))
goto retry;
gvt_vgpu_err("fail to allocate ppgtt shadow page\n");
@@ -801,44 +726,48 @@ retry:
}
spt->vgpu = vgpu;
- spt->guest_page_type = type;
atomic_set(&spt->refcount, 1);
INIT_LIST_HEAD(&spt->post_shadow_list);
/*
- * TODO: guest page type may be different with shadow page type,
- * when we support PSE page in future.
+ * Init shadow_page.
*/
- ret = init_shadow_page(vgpu, &spt->shadow_page, type, true);
- if (ret) {
- gvt_vgpu_err("fail to initialize shadow page for spt\n");
- goto err;
+ spt->shadow_page.type = type;
+ daddr = dma_map_page(kdev, spt->shadow_page.page,
+ 0, 4096, PCI_DMA_BIDIRECTIONAL);
+ if (dma_mapping_error(kdev, daddr)) {
+ gvt_vgpu_err("fail to map dma addr\n");
+ ret = -EINVAL;
+ goto err_free_spt;
}
+ spt->shadow_page.vaddr = page_address(spt->shadow_page.page);
+ spt->shadow_page.mfn = daddr >> I915_GTT_PAGE_SHIFT;
- ret = init_guest_page(vgpu, &spt->guest_page,
- gfn, ppgtt_write_protection_handler, NULL);
- if (ret) {
- gvt_vgpu_err("fail to initialize guest page for spt\n");
- goto err;
- }
+ /*
+ * Init guest_page.
+ */
+ spt->guest_page.type = type;
+ spt->guest_page.gfn = gfn;
- trace_spt_alloc(vgpu->id, spt, type, spt->shadow_page.mfn, gfn);
- return spt;
-err:
- ppgtt_free_shadow_page(spt);
- return ERR_PTR(ret);
-}
+ ret = intel_vgpu_register_page_track(vgpu, spt->guest_page.gfn,
+ ppgtt_write_protection_handler, spt);
+ if (ret)
+ goto err_unmap_dma;
-static struct intel_vgpu_ppgtt_spt *ppgtt_find_shadow_page(
- struct intel_vgpu *vgpu, unsigned long mfn)
-{
- struct intel_vgpu_shadow_page *p = find_shadow_page(vgpu, mfn);
+ ret = radix_tree_insert(&vgpu->gtt.spt_tree, spt->shadow_page.mfn, spt);
+ if (ret)
+ goto err_unreg_page_track;
- if (p)
- return shadow_page_to_ppgtt_spt(p);
+ trace_spt_alloc(vgpu->id, spt, type, spt->shadow_page.mfn, gfn);
+ return spt;
- gvt_vgpu_err("fail to find ppgtt shadow page: 0x%lx\n", mfn);
- return NULL;
+err_unreg_page_track:
+ intel_vgpu_unregister_page_track(vgpu, spt->guest_page.gfn);
+err_unmap_dma:
+ dma_unmap_page(kdev, daddr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+err_free_spt:
+ free_spt(spt);
+ return ERR_PTR(ret);
}
#define pt_entry_size_shift(spt) \
@@ -857,7 +786,7 @@ static struct intel_vgpu_ppgtt_spt *ppgtt_find_shadow_page(
if (!ppgtt_get_shadow_entry(spt, e, i) && \
spt->vgpu->gvt->gtt.pte_ops->test_present(e))
-static void ppgtt_get_shadow_page(struct intel_vgpu_ppgtt_spt *spt)
+static void ppgtt_get_spt(struct intel_vgpu_ppgtt_spt *spt)
{
int v = atomic_read(&spt->refcount);
@@ -866,17 +795,16 @@ static void ppgtt_get_shadow_page(struct intel_vgpu_ppgtt_spt *spt)
atomic_inc(&spt->refcount);
}
-static int ppgtt_invalidate_shadow_page(struct intel_vgpu_ppgtt_spt *spt);
+static int ppgtt_invalidate_spt(struct intel_vgpu_ppgtt_spt *spt);
-static int ppgtt_invalidate_shadow_page_by_shadow_entry(struct intel_vgpu *vgpu,
+static int ppgtt_invalidate_spt_by_shadow_entry(struct intel_vgpu *vgpu,
struct intel_gvt_gtt_entry *e)
{
struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
struct intel_vgpu_ppgtt_spt *s;
intel_gvt_gtt_type_t cur_pt_type;
- if (WARN_ON(!gtt_type_is_pt(get_next_pt_type(e->type))))
- return -EINVAL;
+ GEM_BUG_ON(!gtt_type_is_pt(get_next_pt_type(e->type)));
if (e->type != GTT_TYPE_PPGTT_ROOT_L3_ENTRY
&& e->type != GTT_TYPE_PPGTT_ROOT_L4_ENTRY) {
@@ -885,16 +813,33 @@ static int ppgtt_invalidate_shadow_page_by_shadow_entry(struct intel_vgpu *vgpu,
vgpu->gtt.scratch_pt[cur_pt_type].page_mfn)
return 0;
}
- s = ppgtt_find_shadow_page(vgpu, ops->get_pfn(e));
+ s = intel_vgpu_find_spt_by_mfn(vgpu, ops->get_pfn(e));
if (!s) {
gvt_vgpu_err("fail to find shadow page: mfn: 0x%lx\n",
ops->get_pfn(e));
return -ENXIO;
}
- return ppgtt_invalidate_shadow_page(s);
+ return ppgtt_invalidate_spt(s);
+}
+
+static inline void ppgtt_invalidate_pte(struct intel_vgpu_ppgtt_spt *spt,
+ struct intel_gvt_gtt_entry *entry)
+{
+ struct intel_vgpu *vgpu = spt->vgpu;
+ struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
+ unsigned long pfn;
+ int type;
+
+ pfn = ops->get_pfn(entry);
+ type = spt->shadow_page.type;
+
+ if (pfn == vgpu->gtt.scratch_pt[type].page_mfn)
+ return;
+
+ intel_gvt_hypervisor_dma_unmap_guest_page(vgpu, pfn << PAGE_SHIFT);
}
-static int ppgtt_invalidate_shadow_page(struct intel_vgpu_ppgtt_spt *spt)
+static int ppgtt_invalidate_spt(struct intel_vgpu_ppgtt_spt *spt)
{
struct intel_vgpu *vgpu = spt->vgpu;
struct intel_gvt_gtt_entry e;
@@ -903,30 +848,40 @@ static int ppgtt_invalidate_shadow_page(struct intel_vgpu_ppgtt_spt *spt)
int v = atomic_read(&spt->refcount);
trace_spt_change(spt->vgpu->id, "die", spt,
- spt->guest_page.track.gfn, spt->shadow_page.type);
+ spt->guest_page.gfn, spt->shadow_page.type);
trace_spt_refcount(spt->vgpu->id, "dec", spt, v, (v - 1));
if (atomic_dec_return(&spt->refcount) > 0)
return 0;
- if (gtt_type_is_pte_pt(spt->shadow_page.type))
- goto release;
-
for_each_present_shadow_entry(spt, &e, index) {
- if (!gtt_type_is_pt(get_next_pt_type(e.type))) {
- gvt_vgpu_err("GVT doesn't support pse bit for now\n");
- return -EINVAL;
+ switch (e.type) {
+ case GTT_TYPE_PPGTT_PTE_4K_ENTRY:
+ gvt_vdbg_mm("invalidate 4K entry\n");
+ ppgtt_invalidate_pte(spt, &e);
+ break;
+ case GTT_TYPE_PPGTT_PTE_2M_ENTRY:
+ case GTT_TYPE_PPGTT_PTE_1G_ENTRY:
+ WARN(1, "GVT doesn't support 2M/1GB page\n");
+ continue;
+ case GTT_TYPE_PPGTT_PML4_ENTRY:
+ case GTT_TYPE_PPGTT_PDP_ENTRY:
+ case GTT_TYPE_PPGTT_PDE_ENTRY:
+ gvt_vdbg_mm("invalidate PMUL4/PDP/PDE entry\n");
+ ret = ppgtt_invalidate_spt_by_shadow_entry(
+ spt->vgpu, &e);
+ if (ret)
+ goto fail;
+ break;
+ default:
+ GEM_BUG_ON(1);
}
- ret = ppgtt_invalidate_shadow_page_by_shadow_entry(
- spt->vgpu, &e);
- if (ret)
- goto fail;
}
-release:
+
trace_spt_change(spt->vgpu->id, "release", spt,
- spt->guest_page.track.gfn, spt->shadow_page.type);
- ppgtt_free_shadow_page(spt);
+ spt->guest_page.gfn, spt->shadow_page.type);
+ ppgtt_free_spt(spt);
return 0;
fail:
gvt_vgpu_err("fail: shadow page %p shadow entry 0x%llx type %d\n",
@@ -934,52 +889,44 @@ fail:
return ret;
}
-static int ppgtt_populate_shadow_page(struct intel_vgpu_ppgtt_spt *spt);
+static int ppgtt_populate_spt(struct intel_vgpu_ppgtt_spt *spt);
-static struct intel_vgpu_ppgtt_spt *ppgtt_populate_shadow_page_by_guest_entry(
+static struct intel_vgpu_ppgtt_spt *ppgtt_populate_spt_by_guest_entry(
struct intel_vgpu *vgpu, struct intel_gvt_gtt_entry *we)
{
struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
- struct intel_vgpu_ppgtt_spt *s = NULL;
- struct intel_vgpu_guest_page *g;
- struct intel_vgpu_page_track *t;
+ struct intel_vgpu_ppgtt_spt *spt = NULL;
int ret;
- if (WARN_ON(!gtt_type_is_pt(get_next_pt_type(we->type)))) {
- ret = -EINVAL;
- goto fail;
- }
+ GEM_BUG_ON(!gtt_type_is_pt(get_next_pt_type(we->type)));
- t = intel_vgpu_find_tracked_page(vgpu, ops->get_pfn(we));
- if (t) {
- g = page_track_to_guest_page(t);
- s = guest_page_to_ppgtt_spt(g);
- ppgtt_get_shadow_page(s);
- } else {
+ spt = intel_vgpu_find_spt_by_gfn(vgpu, ops->get_pfn(we));
+ if (spt)
+ ppgtt_get_spt(spt);
+ else {
int type = get_next_pt_type(we->type);
- s = ppgtt_alloc_shadow_page(vgpu, type, ops->get_pfn(we));
- if (IS_ERR(s)) {
- ret = PTR_ERR(s);
+ spt = ppgtt_alloc_spt(vgpu, type, ops->get_pfn(we));
+ if (IS_ERR(spt)) {
+ ret = PTR_ERR(spt);
goto fail;
}
- ret = intel_gvt_hypervisor_enable_page_track(vgpu,
- &s->guest_page.track);
+ ret = intel_vgpu_enable_page_track(vgpu, spt->guest_page.gfn);
if (ret)
goto fail;
- ret = ppgtt_populate_shadow_page(s);
+ ret = ppgtt_populate_spt(spt);
if (ret)
goto fail;
- trace_spt_change(vgpu->id, "new", s, s->guest_page.track.gfn,
- s->shadow_page.type);
+ trace_spt_change(vgpu->id, "new", spt, spt->guest_page.gfn,
+ spt->shadow_page.type);
}
- return s;
+ return spt;
fail:
gvt_vgpu_err("fail: shadow page %p guest entry 0x%llx type %d\n",
- s, we->val64, we->type);
+ spt, we->val64, we->type);
return ERR_PTR(ret);
}
@@ -994,7 +941,44 @@ static inline void ppgtt_generate_shadow_entry(struct intel_gvt_gtt_entry *se,
ops->set_pfn(se, s->shadow_page.mfn);
}
-static int ppgtt_populate_shadow_page(struct intel_vgpu_ppgtt_spt *spt)
+static int ppgtt_populate_shadow_entry(struct intel_vgpu *vgpu,
+ struct intel_vgpu_ppgtt_spt *spt, unsigned long index,
+ struct intel_gvt_gtt_entry *ge)
+{
+ struct intel_gvt_gtt_pte_ops *pte_ops = vgpu->gvt->gtt.pte_ops;
+ struct intel_gvt_gtt_entry se = *ge;
+ unsigned long gfn;
+ dma_addr_t dma_addr;
+ int ret;
+
+ if (!pte_ops->test_present(ge))
+ return 0;
+
+ gfn = pte_ops->get_pfn(ge);
+
+ switch (ge->type) {
+ case GTT_TYPE_PPGTT_PTE_4K_ENTRY:
+ gvt_vdbg_mm("shadow 4K gtt entry\n");
+ break;
+ case GTT_TYPE_PPGTT_PTE_2M_ENTRY:
+ case GTT_TYPE_PPGTT_PTE_1G_ENTRY:
+ gvt_vgpu_err("GVT doesn't support 2M/1GB entry\n");
+ return -EINVAL;
+ default:
+ GEM_BUG_ON(1);
+ };
+
+ /* direct shadow */
+ ret = intel_gvt_hypervisor_dma_map_guest_page(vgpu, gfn, &dma_addr);
+ if (ret)
+ return -ENXIO;
+
+ pte_ops->set_pfn(&se, dma_addr >> PAGE_SHIFT);
+ ppgtt_set_shadow_entry(spt, &se, index);
+ return 0;
+}
+
+static int ppgtt_populate_spt(struct intel_vgpu_ppgtt_spt *spt)
{
struct intel_vgpu *vgpu = spt->vgpu;
struct intel_gvt *gvt = vgpu->gvt;
@@ -1005,34 +989,30 @@ static int ppgtt_populate_shadow_page(struct intel_vgpu_ppgtt_spt *spt)
int ret;
trace_spt_change(spt->vgpu->id, "born", spt,
- spt->guest_page.track.gfn, spt->shadow_page.type);
+ spt->guest_page.gfn, spt->shadow_page.type);
- if (gtt_type_is_pte_pt(spt->shadow_page.type)) {
- for_each_present_guest_entry(spt, &ge, i) {
+ for_each_present_guest_entry(spt, &ge, i) {
+ if (gtt_type_is_pt(get_next_pt_type(ge.type))) {
+ s = ppgtt_populate_spt_by_guest_entry(vgpu, &ge);
+ if (IS_ERR(s)) {
+ ret = PTR_ERR(s);
+ goto fail;
+ }
+ ppgtt_get_shadow_entry(spt, &se, i);
+ ppgtt_generate_shadow_entry(&se, s, &ge);
+ ppgtt_set_shadow_entry(spt, &se, i);
+ } else {
gfn = ops->get_pfn(&ge);
- if (!intel_gvt_hypervisor_is_valid_gfn(vgpu, gfn) ||
- gtt_entry_p2m(vgpu, &ge, &se))
+ if (!intel_gvt_hypervisor_is_valid_gfn(vgpu, gfn)) {
ops->set_pfn(&se, gvt->gtt.scratch_mfn);
- ppgtt_set_shadow_entry(spt, &se, i);
- }
- return 0;
- }
-
- for_each_present_guest_entry(spt, &ge, i) {
- if (!gtt_type_is_pt(get_next_pt_type(ge.type))) {
- gvt_vgpu_err("GVT doesn't support pse bit now\n");
- ret = -EINVAL;
- goto fail;
- }
+ ppgtt_set_shadow_entry(spt, &se, i);
+ continue;
+ }
- s = ppgtt_populate_shadow_page_by_guest_entry(vgpu, &ge);
- if (IS_ERR(s)) {
- ret = PTR_ERR(s);
- goto fail;
+ ret = ppgtt_populate_shadow_entry(vgpu, spt, i, &ge);
+ if (ret)
+ goto fail;
}
- ppgtt_get_shadow_entry(spt, &se, i);
- ppgtt_generate_shadow_entry(&se, s, &ge);
- ppgtt_set_shadow_entry(spt, &se, i);
}
return 0;
fail:
@@ -1041,36 +1021,40 @@ fail:
return ret;
}
-static int ppgtt_handle_guest_entry_removal(struct intel_vgpu_guest_page *gpt,
+static int ppgtt_handle_guest_entry_removal(struct intel_vgpu_ppgtt_spt *spt,
struct intel_gvt_gtt_entry *se, unsigned long index)
{
- struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt);
- struct intel_vgpu_shadow_page *sp = &spt->shadow_page;
struct intel_vgpu *vgpu = spt->vgpu;
struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
int ret;
- trace_gpt_change(spt->vgpu->id, "remove", spt, sp->type, se->val64,
- index);
+ trace_spt_guest_change(spt->vgpu->id, "remove", spt,
+ spt->shadow_page.type, se->val64, index);
+
+ gvt_vdbg_mm("destroy old shadow entry, type %d, index %lu, value %llx\n",
+ se->type, index, se->val64);
if (!ops->test_present(se))
return 0;
- if (ops->get_pfn(se) == vgpu->gtt.scratch_pt[sp->type].page_mfn)
+ if (ops->get_pfn(se) ==
+ vgpu->gtt.scratch_pt[spt->shadow_page.type].page_mfn)
return 0;
if (gtt_type_is_pt(get_next_pt_type(se->type))) {
struct intel_vgpu_ppgtt_spt *s =
- ppgtt_find_shadow_page(vgpu, ops->get_pfn(se));
+ intel_vgpu_find_spt_by_mfn(vgpu, ops->get_pfn(se));
if (!s) {
gvt_vgpu_err("fail to find guest page\n");
ret = -ENXIO;
goto fail;
}
- ret = ppgtt_invalidate_shadow_page(s);
+ ret = ppgtt_invalidate_spt(s);
if (ret)
goto fail;
- }
+ } else
+ ppgtt_invalidate_pte(spt, se);
+
return 0;
fail:
gvt_vgpu_err("fail: shadow page %p guest entry 0x%llx type %d\n",
@@ -1078,21 +1062,22 @@ fail:
return ret;
}
-static int ppgtt_handle_guest_entry_add(struct intel_vgpu_guest_page *gpt,
+static int ppgtt_handle_guest_entry_add(struct intel_vgpu_ppgtt_spt *spt,
struct intel_gvt_gtt_entry *we, unsigned long index)
{
- struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt);
- struct intel_vgpu_shadow_page *sp = &spt->shadow_page;
struct intel_vgpu *vgpu = spt->vgpu;
struct intel_gvt_gtt_entry m;
struct intel_vgpu_ppgtt_spt *s;
int ret;
- trace_gpt_change(spt->vgpu->id, "add", spt, sp->type,
- we->val64, index);
+ trace_spt_guest_change(spt->vgpu->id, "add", spt, spt->shadow_page.type,
+ we->val64, index);
+
+ gvt_vdbg_mm("add shadow entry: type %d, index %lu, value %llx\n",
+ we->type, index, we->val64);
if (gtt_type_is_pt(get_next_pt_type(we->type))) {
- s = ppgtt_populate_shadow_page_by_guest_entry(vgpu, we);
+ s = ppgtt_populate_spt_by_guest_entry(vgpu, we);
if (IS_ERR(s)) {
ret = PTR_ERR(s);
goto fail;
@@ -1101,10 +1086,9 @@ static int ppgtt_handle_guest_entry_add(struct intel_vgpu_guest_page *gpt,
ppgtt_generate_shadow_entry(&m, s, we);
ppgtt_set_shadow_entry(spt, &m, index);
} else {
- ret = gtt_entry_p2m(vgpu, we, &m);
+ ret = ppgtt_populate_shadow_entry(vgpu, spt, index, we);
if (ret)
goto fail;
- ppgtt_set_shadow_entry(spt, &m, index);
}
return 0;
fail:
@@ -1119,41 +1103,39 @@ static int sync_oos_page(struct intel_vgpu *vgpu,
const struct intel_gvt_device_info *info = &vgpu->gvt->device_info;
struct intel_gvt *gvt = vgpu->gvt;
struct intel_gvt_gtt_pte_ops *ops = gvt->gtt.pte_ops;
- struct intel_vgpu_ppgtt_spt *spt =
- guest_page_to_ppgtt_spt(oos_page->guest_page);
- struct intel_gvt_gtt_entry old, new, m;
+ struct intel_vgpu_ppgtt_spt *spt = oos_page->spt;
+ struct intel_gvt_gtt_entry old, new;
int index;
int ret;
trace_oos_change(vgpu->id, "sync", oos_page->id,
- oos_page->guest_page, spt->guest_page_type);
+ spt, spt->guest_page.type);
- old.type = new.type = get_entry_type(spt->guest_page_type);
+ old.type = new.type = get_entry_type(spt->guest_page.type);
old.val64 = new.val64 = 0;
for (index = 0; index < (I915_GTT_PAGE_SIZE >>
info->gtt_entry_size_shift); index++) {
ops->get_entry(oos_page->mem, &old, index, false, 0, vgpu);
ops->get_entry(NULL, &new, index, true,
- oos_page->guest_page->track.gfn << PAGE_SHIFT, vgpu);
+ spt->guest_page.gfn << PAGE_SHIFT, vgpu);
if (old.val64 == new.val64
&& !test_and_clear_bit(index, spt->post_shadow_bitmap))
continue;
trace_oos_sync(vgpu->id, oos_page->id,
- oos_page->guest_page, spt->guest_page_type,
+ spt, spt->guest_page.type,
new.val64, index);
- ret = gtt_entry_p2m(vgpu, &new, &m);
+ ret = ppgtt_populate_shadow_entry(vgpu, spt, index, &new);
if (ret)
return ret;
ops->set_entry(oos_page->mem, &new, index, false, 0, vgpu);
- ppgtt_set_shadow_entry(spt, &m, index);
}
- oos_page->guest_page->write_cnt = 0;
+ spt->guest_page.write_cnt = 0;
list_del_init(&spt->post_shadow_list);
return 0;
}
@@ -1162,15 +1144,14 @@ static int detach_oos_page(struct intel_vgpu *vgpu,
struct intel_vgpu_oos_page *oos_page)
{
struct intel_gvt *gvt = vgpu->gvt;
- struct intel_vgpu_ppgtt_spt *spt =
- guest_page_to_ppgtt_spt(oos_page->guest_page);
+ struct intel_vgpu_ppgtt_spt *spt = oos_page->spt;
trace_oos_change(vgpu->id, "detach", oos_page->id,
- oos_page->guest_page, spt->guest_page_type);
+ spt, spt->guest_page.type);
- oos_page->guest_page->write_cnt = 0;
- oos_page->guest_page->oos_page = NULL;
- oos_page->guest_page = NULL;
+ spt->guest_page.write_cnt = 0;
+ spt->guest_page.oos_page = NULL;
+ oos_page->spt = NULL;
list_del_init(&oos_page->vm_list);
list_move_tail(&oos_page->list, &gvt->gtt.oos_page_free_list_head);
@@ -1178,51 +1159,49 @@ static int detach_oos_page(struct intel_vgpu *vgpu,
return 0;
}
-static int attach_oos_page(struct intel_vgpu *vgpu,
- struct intel_vgpu_oos_page *oos_page,
- struct intel_vgpu_guest_page *gpt)
+static int attach_oos_page(struct intel_vgpu_oos_page *oos_page,
+ struct intel_vgpu_ppgtt_spt *spt)
{
- struct intel_gvt *gvt = vgpu->gvt;
+ struct intel_gvt *gvt = spt->vgpu->gvt;
int ret;
- ret = intel_gvt_hypervisor_read_gpa(vgpu,
- gpt->track.gfn << I915_GTT_PAGE_SHIFT,
+ ret = intel_gvt_hypervisor_read_gpa(spt->vgpu,
+ spt->guest_page.gfn << I915_GTT_PAGE_SHIFT,
oos_page->mem, I915_GTT_PAGE_SIZE);
if (ret)
return ret;
- oos_page->guest_page = gpt;
- gpt->oos_page = oos_page;
+ oos_page->spt = spt;
+ spt->guest_page.oos_page = oos_page;
list_move_tail(&oos_page->list, &gvt->gtt.oos_page_use_list_head);
- trace_oos_change(vgpu->id, "attach", gpt->oos_page->id,
- gpt, guest_page_to_ppgtt_spt(gpt)->guest_page_type);
+ trace_oos_change(spt->vgpu->id, "attach", oos_page->id,
+ spt, spt->guest_page.type);
return 0;
}
-static int ppgtt_set_guest_page_sync(struct intel_vgpu *vgpu,
- struct intel_vgpu_guest_page *gpt)
+static int ppgtt_set_guest_page_sync(struct intel_vgpu_ppgtt_spt *spt)
{
+ struct intel_vgpu_oos_page *oos_page = spt->guest_page.oos_page;
int ret;
- ret = intel_gvt_hypervisor_enable_page_track(vgpu, &gpt->track);
+ ret = intel_vgpu_enable_page_track(spt->vgpu, spt->guest_page.gfn);
if (ret)
return ret;
- trace_oos_change(vgpu->id, "set page sync", gpt->oos_page->id,
- gpt, guest_page_to_ppgtt_spt(gpt)->guest_page_type);
+ trace_oos_change(spt->vgpu->id, "set page sync", oos_page->id,
+ spt, spt->guest_page.type);
- list_del_init(&gpt->oos_page->vm_list);
- return sync_oos_page(vgpu, gpt->oos_page);
+ list_del_init(&oos_page->vm_list);
+ return sync_oos_page(spt->vgpu, oos_page);
}
-static int ppgtt_allocate_oos_page(struct intel_vgpu *vgpu,
- struct intel_vgpu_guest_page *gpt)
+static int ppgtt_allocate_oos_page(struct intel_vgpu_ppgtt_spt *spt)
{
- struct intel_gvt *gvt = vgpu->gvt;
+ struct intel_gvt *gvt = spt->vgpu->gvt;
struct intel_gvt_gtt *gtt = &gvt->gtt;
- struct intel_vgpu_oos_page *oos_page = gpt->oos_page;
+ struct intel_vgpu_oos_page *oos_page = spt->guest_page.oos_page;
int ret;
WARN(oos_page, "shadow PPGTT page has already has a oos page\n");
@@ -1230,31 +1209,30 @@ static int ppgtt_allocate_oos_page(struct intel_vgpu *vgpu,
if (list_empty(&gtt->oos_page_free_list_head)) {
oos_page = container_of(gtt->oos_page_use_list_head.next,
struct intel_vgpu_oos_page, list);
- ret = ppgtt_set_guest_page_sync(vgpu, oos_page->guest_page);
+ ret = ppgtt_set_guest_page_sync(oos_page->spt);
if (ret)
return ret;
- ret = detach_oos_page(vgpu, oos_page);
+ ret = detach_oos_page(spt->vgpu, oos_page);
if (ret)
return ret;
} else
oos_page = container_of(gtt->oos_page_free_list_head.next,
struct intel_vgpu_oos_page, list);
- return attach_oos_page(vgpu, oos_page, gpt);
+ return attach_oos_page(oos_page, spt);
}
-static int ppgtt_set_guest_page_oos(struct intel_vgpu *vgpu,
- struct intel_vgpu_guest_page *gpt)
+static int ppgtt_set_guest_page_oos(struct intel_vgpu_ppgtt_spt *spt)
{
- struct intel_vgpu_oos_page *oos_page = gpt->oos_page;
+ struct intel_vgpu_oos_page *oos_page = spt->guest_page.oos_page;
if (WARN(!oos_page, "shadow PPGTT page should have a oos page\n"))
return -EINVAL;
- trace_oos_change(vgpu->id, "set page out of sync", gpt->oos_page->id,
- gpt, guest_page_to_ppgtt_spt(gpt)->guest_page_type);
+ trace_oos_change(spt->vgpu->id, "set page out of sync", oos_page->id,
+ spt, spt->guest_page.type);
- list_add_tail(&oos_page->vm_list, &vgpu->gtt.oos_page_list_head);
- return intel_gvt_hypervisor_disable_page_track(vgpu, &gpt->track);
+ list_add_tail(&oos_page->vm_list, &spt->vgpu->gtt.oos_page_list_head);
+ return intel_vgpu_disable_page_track(spt->vgpu, spt->guest_page.gfn);
}
/**
@@ -1279,7 +1257,7 @@ int intel_vgpu_sync_oos_pages(struct intel_vgpu *vgpu)
list_for_each_safe(pos, n, &vgpu->gtt.oos_page_list_head) {
oos_page = container_of(pos,
struct intel_vgpu_oos_page, vm_list);
- ret = ppgtt_set_guest_page_sync(vgpu, oos_page->guest_page);
+ ret = ppgtt_set_guest_page_sync(oos_page->spt);
if (ret)
return ret;
}
@@ -1290,17 +1268,15 @@ int intel_vgpu_sync_oos_pages(struct intel_vgpu *vgpu)
* The heart of PPGTT shadow page table.
*/
static int ppgtt_handle_guest_write_page_table(
- struct intel_vgpu_guest_page *gpt,
+ struct intel_vgpu_ppgtt_spt *spt,
struct intel_gvt_gtt_entry *we, unsigned long index)
{
- struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt);
struct intel_vgpu *vgpu = spt->vgpu;
int type = spt->shadow_page.type;
struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
- struct intel_gvt_gtt_entry se;
-
- int ret;
+ struct intel_gvt_gtt_entry old_se;
int new_present;
+ int ret;
new_present = ops->test_present(we);
@@ -1309,21 +1285,21 @@ static int ppgtt_handle_guest_write_page_table(
* guarantee the ppgtt table is validated during the window between
* adding and removal.
*/
- ppgtt_get_shadow_entry(spt, &se, index);
+ ppgtt_get_shadow_entry(spt, &old_se, index);
if (new_present) {
- ret = ppgtt_handle_guest_entry_add(gpt, we, index);
+ ret = ppgtt_handle_guest_entry_add(spt, we, index);
if (ret)
goto fail;
}
- ret = ppgtt_handle_guest_entry_removal(gpt, &se, index);
+ ret = ppgtt_handle_guest_entry_removal(spt, &old_se, index);
if (ret)
goto fail;
if (!new_present) {
- ops->set_pfn(&se, vgpu->gtt.scratch_pt[type].page_mfn);
- ppgtt_set_shadow_entry(spt, &se, index);
+ ops->set_pfn(&old_se, vgpu->gtt.scratch_pt[type].page_mfn);
+ ppgtt_set_shadow_entry(spt, &old_se, index);
}
return 0;
@@ -1333,12 +1309,13 @@ fail:
return ret;
}
-static inline bool can_do_out_of_sync(struct intel_vgpu_guest_page *gpt)
+
+
+static inline bool can_do_out_of_sync(struct intel_vgpu_ppgtt_spt *spt)
{
return enable_out_of_sync
- && gtt_type_is_pte_pt(
- guest_page_to_ppgtt_spt(gpt)->guest_page_type)
- && gpt->write_cnt >= 2;
+ && gtt_type_is_pte_pt(spt->guest_page.type)
+ && spt->guest_page.write_cnt >= 2;
}
static void ppgtt_set_post_shadow(struct intel_vgpu_ppgtt_spt *spt,
@@ -1378,8 +1355,8 @@ int intel_vgpu_flush_post_shadow(struct intel_vgpu *vgpu)
GTT_ENTRY_NUM_IN_ONE_PAGE) {
ppgtt_get_guest_entry(spt, &ge, index);
- ret = ppgtt_handle_guest_write_page_table(
- &spt->guest_page, &ge, index);
+ ret = ppgtt_handle_guest_write_page_table(spt,
+ &ge, index);
if (ret)
return ret;
clear_bit(index, spt->post_shadow_bitmap);
@@ -1390,10 +1367,9 @@ int intel_vgpu_flush_post_shadow(struct intel_vgpu *vgpu)
}
static int ppgtt_handle_guest_write_page_table_bytes(
- struct intel_vgpu_guest_page *gpt,
+ struct intel_vgpu_ppgtt_spt *spt,
u64 pa, void *p_data, int bytes)
{
- struct intel_vgpu_ppgtt_spt *spt = guest_page_to_ppgtt_spt(gpt);
struct intel_vgpu *vgpu = spt->vgpu;
struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
const struct intel_gvt_device_info *info = &vgpu->gvt->device_info;
@@ -1408,7 +1384,7 @@ static int ppgtt_handle_guest_write_page_table_bytes(
ops->test_pse(&we);
if (bytes == info->gtt_entry_size) {
- ret = ppgtt_handle_guest_write_page_table(gpt, &we, index);
+ ret = ppgtt_handle_guest_write_page_table(spt, &we, index);
if (ret)
return ret;
} else {
@@ -1416,7 +1392,7 @@ static int ppgtt_handle_guest_write_page_table_bytes(
int type = spt->shadow_page.type;
ppgtt_get_shadow_entry(spt, &se, index);
- ret = ppgtt_handle_guest_entry_removal(gpt, &se, index);
+ ret = ppgtt_handle_guest_entry_removal(spt, &se, index);
if (ret)
return ret;
ops->set_pfn(&se, vgpu->gtt.scratch_pt[type].page_mfn);
@@ -1428,128 +1404,54 @@ static int ppgtt_handle_guest_write_page_table_bytes(
if (!enable_out_of_sync)
return 0;
- gpt->write_cnt++;
+ spt->guest_page.write_cnt++;
- if (gpt->oos_page)
- ops->set_entry(gpt->oos_page->mem, &we, index,
+ if (spt->guest_page.oos_page)
+ ops->set_entry(spt->guest_page.oos_page->mem, &we, index,
false, 0, vgpu);
- if (can_do_out_of_sync(gpt)) {
- if (!gpt->oos_page)
- ppgtt_allocate_oos_page(vgpu, gpt);
+ if (can_do_out_of_sync(spt)) {
+ if (!spt->guest_page.oos_page)
+ ppgtt_allocate_oos_page(spt);
- ret = ppgtt_set_guest_page_oos(vgpu, gpt);
+ ret = ppgtt_set_guest_page_oos(spt);
if (ret < 0)
return ret;
}
return 0;
}
-/*
- * mm page table allocation policy for bdw+
- * - for ggtt, only virtual page table will be allocated.
- * - for ppgtt, dedicated virtual/shadow page table will be allocated.
- */
-static int gen8_mm_alloc_page_table(struct intel_vgpu_mm *mm)
-{
- struct intel_vgpu *vgpu = mm->vgpu;
- struct intel_gvt *gvt = vgpu->gvt;
- const struct intel_gvt_device_info *info = &gvt->device_info;
- void *mem;
-
- if (mm->type == INTEL_GVT_MM_PPGTT) {
- mm->page_table_entry_cnt = 4;
- mm->page_table_entry_size = mm->page_table_entry_cnt *
- info->gtt_entry_size;
- mem = kzalloc(mm->has_shadow_page_table ?
- mm->page_table_entry_size * 2
- : mm->page_table_entry_size, GFP_KERNEL);
- if (!mem)
- return -ENOMEM;
- mm->virtual_page_table = mem;
- if (!mm->has_shadow_page_table)
- return 0;
- mm->shadow_page_table = mem + mm->page_table_entry_size;
- } else if (mm->type == INTEL_GVT_MM_GGTT) {
- mm->page_table_entry_cnt =
- (gvt_ggtt_gm_sz(gvt) >> I915_GTT_PAGE_SHIFT);
- mm->page_table_entry_size = mm->page_table_entry_cnt *
- info->gtt_entry_size;
- mem = vzalloc(mm->page_table_entry_size);
- if (!mem)
- return -ENOMEM;
- mm->virtual_page_table = mem;
- }
- return 0;
-}
-
-static void gen8_mm_free_page_table(struct intel_vgpu_mm *mm)
-{
- if (mm->type == INTEL_GVT_MM_PPGTT) {
- kfree(mm->virtual_page_table);
- } else if (mm->type == INTEL_GVT_MM_GGTT) {
- if (mm->virtual_page_table)
- vfree(mm->virtual_page_table);
- }
- mm->virtual_page_table = mm->shadow_page_table = NULL;
-}
-
-static void invalidate_mm(struct intel_vgpu_mm *mm)
+static void invalidate_ppgtt_mm(struct intel_vgpu_mm *mm)
{
struct intel_vgpu *vgpu = mm->vgpu;
struct intel_gvt *gvt = vgpu->gvt;
struct intel_gvt_gtt *gtt = &gvt->gtt;
struct intel_gvt_gtt_pte_ops *ops = gtt->pte_ops;
struct intel_gvt_gtt_entry se;
- int i;
+ int index;
- if (WARN_ON(!mm->has_shadow_page_table || !mm->shadowed))
+ if (!mm->ppgtt_mm.shadowed)
return;
- for (i = 0; i < mm->page_table_entry_cnt; i++) {
- ppgtt_get_shadow_root_entry(mm, &se, i);
+ for (index = 0; index < ARRAY_SIZE(mm->ppgtt_mm.shadow_pdps); index++) {
+ ppgtt_get_shadow_root_entry(mm, &se, index);
+
if (!ops->test_present(&se))
continue;
- ppgtt_invalidate_shadow_page_by_shadow_entry(
- vgpu, &se);
+
+ ppgtt_invalidate_spt_by_shadow_entry(vgpu, &se);
se.val64 = 0;
- ppgtt_set_shadow_root_entry(mm, &se, i);
+ ppgtt_set_shadow_root_entry(mm, &se, index);
- trace_gpt_change(vgpu->id, "destroy root pointer",
- NULL, se.type, se.val64, i);
+ trace_spt_guest_change(vgpu->id, "destroy root pointer",
+ NULL, se.type, se.val64, index);
}
- mm->shadowed = false;
-}
-/**
- * intel_vgpu_destroy_mm - destroy a mm object
- * @mm: a kref object
- *
- * This function is used to destroy a mm object for vGPU
- *
- */
-void intel_vgpu_destroy_mm(struct kref *mm_ref)
-{
- struct intel_vgpu_mm *mm = container_of(mm_ref, typeof(*mm), ref);
- struct intel_vgpu *vgpu = mm->vgpu;
- struct intel_gvt *gvt = vgpu->gvt;
- struct intel_gvt_gtt *gtt = &gvt->gtt;
-
- if (!mm->initialized)
- goto out;
-
- list_del(&mm->list);
- list_del(&mm->lru_list);
-
- if (mm->has_shadow_page_table)
- invalidate_mm(mm);
-
- gtt->mm_free_page_table(mm);
-out:
- kfree(mm);
+ mm->ppgtt_mm.shadowed = false;
}
-static int shadow_mm(struct intel_vgpu_mm *mm)
+
+static int shadow_ppgtt_mm(struct intel_vgpu_mm *mm)
{
struct intel_vgpu *vgpu = mm->vgpu;
struct intel_gvt *gvt = vgpu->gvt;
@@ -1557,119 +1459,155 @@ static int shadow_mm(struct intel_vgpu_mm *mm)
struct intel_gvt_gtt_pte_ops *ops = gtt->pte_ops;
struct intel_vgpu_ppgtt_spt *spt;
struct intel_gvt_gtt_entry ge, se;
- int i;
- int ret;
+ int index, ret;
- if (WARN_ON(!mm->has_shadow_page_table || mm->shadowed))
+ if (mm->ppgtt_mm.shadowed)
return 0;
- mm->shadowed = true;
+ mm->ppgtt_mm.shadowed = true;
+
+ for (index = 0; index < ARRAY_SIZE(mm->ppgtt_mm.guest_pdps); index++) {
+ ppgtt_get_guest_root_entry(mm, &ge, index);
- for (i = 0; i < mm->page_table_entry_cnt; i++) {
- ppgtt_get_guest_root_entry(mm, &ge, i);
if (!ops->test_present(&ge))
continue;
- trace_gpt_change(vgpu->id, __func__, NULL,
- ge.type, ge.val64, i);
+ trace_spt_guest_change(vgpu->id, __func__, NULL,
+ ge.type, ge.val64, index);
- spt = ppgtt_populate_shadow_page_by_guest_entry(vgpu, &ge);
+ spt = ppgtt_populate_spt_by_guest_entry(vgpu, &ge);
if (IS_ERR(spt)) {
gvt_vgpu_err("fail to populate guest root pointer\n");
ret = PTR_ERR(spt);
goto fail;
}
ppgtt_generate_shadow_entry(&se, spt, &ge);
- ppgtt_set_shadow_root_entry(mm, &se, i);
+ ppgtt_set_shadow_root_entry(mm, &se, index);
- trace_gpt_change(vgpu->id, "populate root pointer",
- NULL, se.type, se.val64, i);
+ trace_spt_guest_change(vgpu->id, "populate root pointer",
+ NULL, se.type, se.val64, index);
}
+
return 0;
fail:
- invalidate_mm(mm);
+ invalidate_ppgtt_mm(mm);
return ret;
}
+static struct intel_vgpu_mm *vgpu_alloc_mm(struct intel_vgpu *vgpu)
+{
+ struct intel_vgpu_mm *mm;
+
+ mm = kzalloc(sizeof(*mm), GFP_KERNEL);
+ if (!mm)
+ return NULL;
+
+ mm->vgpu = vgpu;
+ kref_init(&mm->ref);
+ atomic_set(&mm->pincount, 0);
+
+ return mm;
+}
+
+static void vgpu_free_mm(struct intel_vgpu_mm *mm)
+{
+ kfree(mm);
+}
+
/**
- * intel_vgpu_create_mm - create a mm object for a vGPU
+ * intel_vgpu_create_ppgtt_mm - create a ppgtt mm object for a vGPU
* @vgpu: a vGPU
- * @mm_type: mm object type, should be PPGTT or GGTT
- * @virtual_page_table: page table root pointers. Could be NULL if user wants
- * to populate shadow later.
- * @page_table_level: describe the page table level of the mm object
- * @pde_base_index: pde root pointer base in GGTT MMIO.
+ * @root_entry_type: ppgtt root entry type
+ * @pdps: guest pdps.
*
- * This function is used to create a mm object for a vGPU.
+ * This function is used to create a ppgtt mm object for a vGPU.
*
* Returns:
* Zero on success, negative error code in pointer if failed.
*/
-struct intel_vgpu_mm *intel_vgpu_create_mm(struct intel_vgpu *vgpu,
- int mm_type, void *virtual_page_table, int page_table_level,
- u32 pde_base_index)
+struct intel_vgpu_mm *intel_vgpu_create_ppgtt_mm(struct intel_vgpu *vgpu,
+ intel_gvt_gtt_type_t root_entry_type, u64 pdps[])
{
struct intel_gvt *gvt = vgpu->gvt;
- struct intel_gvt_gtt *gtt = &gvt->gtt;
struct intel_vgpu_mm *mm;
int ret;
- mm = kzalloc(sizeof(*mm), GFP_KERNEL);
- if (!mm) {
- ret = -ENOMEM;
- goto fail;
- }
-
- mm->type = mm_type;
+ mm = vgpu_alloc_mm(vgpu);
+ if (!mm)
+ return ERR_PTR(-ENOMEM);
- if (page_table_level == 1)
- mm->page_table_entry_type = GTT_TYPE_GGTT_PTE;
- else if (page_table_level == 3)
- mm->page_table_entry_type = GTT_TYPE_PPGTT_ROOT_L3_ENTRY;
- else if (page_table_level == 4)
- mm->page_table_entry_type = GTT_TYPE_PPGTT_ROOT_L4_ENTRY;
- else {
- WARN_ON(1);
- ret = -EINVAL;
- goto fail;
- }
+ mm->type = INTEL_GVT_MM_PPGTT;
- mm->page_table_level = page_table_level;
- mm->pde_base_index = pde_base_index;
+ GEM_BUG_ON(root_entry_type != GTT_TYPE_PPGTT_ROOT_L3_ENTRY &&
+ root_entry_type != GTT_TYPE_PPGTT_ROOT_L4_ENTRY);
+ mm->ppgtt_mm.root_entry_type = root_entry_type;
- mm->vgpu = vgpu;
- mm->has_shadow_page_table = !!(mm_type == INTEL_GVT_MM_PPGTT);
+ INIT_LIST_HEAD(&mm->ppgtt_mm.list);
+ INIT_LIST_HEAD(&mm->ppgtt_mm.lru_list);
- kref_init(&mm->ref);
- atomic_set(&mm->pincount, 0);
- INIT_LIST_HEAD(&mm->list);
- INIT_LIST_HEAD(&mm->lru_list);
- list_add_tail(&mm->list, &vgpu->gtt.mm_list_head);
+ if (root_entry_type == GTT_TYPE_PPGTT_ROOT_L4_ENTRY)
+ mm->ppgtt_mm.guest_pdps[0] = pdps[0];
+ else
+ memcpy(mm->ppgtt_mm.guest_pdps, pdps,
+ sizeof(mm->ppgtt_mm.guest_pdps));
- ret = gtt->mm_alloc_page_table(mm);
+ ret = shadow_ppgtt_mm(mm);
if (ret) {
- gvt_vgpu_err("fail to allocate page table for mm\n");
- goto fail;
+ gvt_vgpu_err("failed to shadow ppgtt mm\n");
+ vgpu_free_mm(mm);
+ return ERR_PTR(ret);
}
- mm->initialized = true;
+ list_add_tail(&mm->ppgtt_mm.list, &vgpu->gtt.ppgtt_mm_list_head);
+ list_add_tail(&mm->ppgtt_mm.lru_list, &gvt->gtt.ppgtt_mm_lru_list_head);
+ return mm;
+}
+
+static struct intel_vgpu_mm *intel_vgpu_create_ggtt_mm(struct intel_vgpu *vgpu)
+{
+ struct intel_vgpu_mm *mm;
+ unsigned long nr_entries;
- if (virtual_page_table)
- memcpy(mm->virtual_page_table, virtual_page_table,
- mm->page_table_entry_size);
+ mm = vgpu_alloc_mm(vgpu);
+ if (!mm)
+ return ERR_PTR(-ENOMEM);
- if (mm->has_shadow_page_table) {
- ret = shadow_mm(mm);
- if (ret)
- goto fail;
- list_add_tail(&mm->lru_list, &gvt->gtt.mm_lru_list_head);
+ mm->type = INTEL_GVT_MM_GGTT;
+
+ nr_entries = gvt_ggtt_gm_sz(vgpu->gvt) >> I915_GTT_PAGE_SHIFT;
+ mm->ggtt_mm.virtual_ggtt = vzalloc(nr_entries *
+ vgpu->gvt->device_info.gtt_entry_size);
+ if (!mm->ggtt_mm.virtual_ggtt) {
+ vgpu_free_mm(mm);
+ return ERR_PTR(-ENOMEM);
}
+
return mm;
-fail:
- gvt_vgpu_err("fail to create mm\n");
- if (mm)
- intel_gvt_mm_unreference(mm);
- return ERR_PTR(ret);
+}
+
+/**
+ * _intel_vgpu_mm_release - destroy a mm object
+ * @mm_ref: a kref object
+ *
+ * This function is used to destroy a mm object for vGPU
+ *
+ */
+void _intel_vgpu_mm_release(struct kref *mm_ref)
+{
+ struct intel_vgpu_mm *mm = container_of(mm_ref, typeof(*mm), ref);
+
+ if (GEM_WARN_ON(atomic_read(&mm->pincount)))
+ gvt_err("vgpu mm pin count bug detected\n");
+
+ if (mm->type == INTEL_GVT_MM_PPGTT) {
+ list_del(&mm->ppgtt_mm.list);
+ list_del(&mm->ppgtt_mm.lru_list);
+ invalidate_ppgtt_mm(mm);
+ } else {
+ vfree(mm->ggtt_mm.virtual_ggtt);
+ }
+
+ vgpu_free_mm(mm);
}
/**
@@ -1680,9 +1618,6 @@ fail:
*/
void intel_vgpu_unpin_mm(struct intel_vgpu_mm *mm)
{
- if (WARN_ON(mm->type != INTEL_GVT_MM_PPGTT))
- return;
-
atomic_dec(&mm->pincount);
}
@@ -1701,36 +1636,34 @@ int intel_vgpu_pin_mm(struct intel_vgpu_mm *mm)
{
int ret;
- if (WARN_ON(mm->type != INTEL_GVT_MM_PPGTT))
- return 0;
+ atomic_inc(&mm->pincount);
- if (!mm->shadowed) {
- ret = shadow_mm(mm);
+ if (mm->type == INTEL_GVT_MM_PPGTT) {
+ ret = shadow_ppgtt_mm(mm);
if (ret)
return ret;
+
+ list_move_tail(&mm->ppgtt_mm.lru_list,
+ &mm->vgpu->gvt->gtt.ppgtt_mm_lru_list_head);
+
}
- atomic_inc(&mm->pincount);
- list_del_init(&mm->lru_list);
- list_add_tail(&mm->lru_list, &mm->vgpu->gvt->gtt.mm_lru_list_head);
return 0;
}
-static int reclaim_one_mm(struct intel_gvt *gvt)
+static int reclaim_one_ppgtt_mm(struct intel_gvt *gvt)
{
struct intel_vgpu_mm *mm;
struct list_head *pos, *n;
- list_for_each_safe(pos, n, &gvt->gtt.mm_lru_list_head) {
- mm = container_of(pos, struct intel_vgpu_mm, lru_list);
+ list_for_each_safe(pos, n, &gvt->gtt.ppgtt_mm_lru_list_head) {
+ mm = container_of(pos, struct intel_vgpu_mm, ppgtt_mm.lru_list);
- if (mm->type != INTEL_GVT_MM_PPGTT)
- continue;
if (atomic_read(&mm->pincount))
continue;
- list_del_init(&mm->lru_list);
- invalidate_mm(mm);
+ list_del_init(&mm->ppgtt_mm.lru_list);
+ invalidate_ppgtt_mm(mm);
return 1;
}
return 0;
@@ -1746,10 +1679,7 @@ static inline int ppgtt_get_next_level_entry(struct intel_vgpu_mm *mm,
struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
struct intel_vgpu_ppgtt_spt *s;
- if (WARN_ON(!mm->has_shadow_page_table))
- return -EINVAL;
-
- s = ppgtt_find_shadow_page(vgpu, ops->get_pfn(e));
+ s = intel_vgpu_find_spt_by_mfn(vgpu, ops->get_pfn(e));
if (!s)
return -ENXIO;
@@ -1780,85 +1710,72 @@ unsigned long intel_vgpu_gma_to_gpa(struct intel_vgpu_mm *mm, unsigned long gma)
unsigned long gpa = INTEL_GVT_INVALID_ADDR;
unsigned long gma_index[4];
struct intel_gvt_gtt_entry e;
- int i, index;
+ int i, levels = 0;
int ret;
- if (mm->type != INTEL_GVT_MM_GGTT && mm->type != INTEL_GVT_MM_PPGTT)
- return INTEL_GVT_INVALID_ADDR;
+ GEM_BUG_ON(mm->type != INTEL_GVT_MM_GGTT &&
+ mm->type != INTEL_GVT_MM_PPGTT);
if (mm->type == INTEL_GVT_MM_GGTT) {
if (!vgpu_gmadr_is_valid(vgpu, gma))
goto err;
- ret = ggtt_get_guest_entry(mm, &e,
- gma_ops->gma_to_ggtt_pte_index(gma));
- if (ret)
- goto err;
+ ggtt_get_guest_entry(mm, &e,
+ gma_ops->gma_to_ggtt_pte_index(gma));
+
gpa = (pte_ops->get_pfn(&e) << I915_GTT_PAGE_SHIFT)
+ (gma & ~I915_GTT_PAGE_MASK);
trace_gma_translate(vgpu->id, "ggtt", 0, 0, gma, gpa);
- return gpa;
- }
-
- switch (mm->page_table_level) {
- case 4:
- ret = ppgtt_get_shadow_root_entry(mm, &e, 0);
- if (ret)
- goto err;
- gma_index[0] = gma_ops->gma_to_pml4_index(gma);
- gma_index[1] = gma_ops->gma_to_l4_pdp_index(gma);
- gma_index[2] = gma_ops->gma_to_pde_index(gma);
- gma_index[3] = gma_ops->gma_to_pte_index(gma);
- index = 4;
- break;
- case 3:
- ret = ppgtt_get_shadow_root_entry(mm, &e,
- gma_ops->gma_to_l3_pdp_index(gma));
- if (ret)
- goto err;
- gma_index[0] = gma_ops->gma_to_pde_index(gma);
- gma_index[1] = gma_ops->gma_to_pte_index(gma);
- index = 2;
- break;
- case 2:
- ret = ppgtt_get_shadow_root_entry(mm, &e,
- gma_ops->gma_to_pde_index(gma));
- if (ret)
- goto err;
- gma_index[0] = gma_ops->gma_to_pte_index(gma);
- index = 1;
- break;
- default:
- WARN_ON(1);
- goto err;
- }
+ } else {
+ switch (mm->ppgtt_mm.root_entry_type) {
+ case GTT_TYPE_PPGTT_ROOT_L4_ENTRY:
+ ppgtt_get_shadow_root_entry(mm, &e, 0);
+
+ gma_index[0] = gma_ops->gma_to_pml4_index(gma);
+ gma_index[1] = gma_ops->gma_to_l4_pdp_index(gma);
+ gma_index[2] = gma_ops->gma_to_pde_index(gma);
+ gma_index[3] = gma_ops->gma_to_pte_index(gma);
+ levels = 4;
+ break;
+ case GTT_TYPE_PPGTT_ROOT_L3_ENTRY:
+ ppgtt_get_shadow_root_entry(mm, &e,
+ gma_ops->gma_to_l3_pdp_index(gma));
+
+ gma_index[0] = gma_ops->gma_to_pde_index(gma);
+ gma_index[1] = gma_ops->gma_to_pte_index(gma);
+ levels = 2;
+ break;
+ default:
+ GEM_BUG_ON(1);
+ }
- /* walk into the shadow page table and get gpa from guest entry */
- for (i = 0; i < index; i++) {
- ret = ppgtt_get_next_level_entry(mm, &e, gma_index[i],
- (i == index - 1));
- if (ret)
- goto err;
+ /* walk the shadow page table and get gpa from guest entry */
+ for (i = 0; i < levels; i++) {
+ ret = ppgtt_get_next_level_entry(mm, &e, gma_index[i],
+ (i == levels - 1));
+ if (ret)
+ goto err;
- if (!pte_ops->test_present(&e)) {
- gvt_dbg_core("GMA 0x%lx is not present\n", gma);
- goto err;
+ if (!pte_ops->test_present(&e)) {
+ gvt_dbg_core("GMA 0x%lx is not present\n", gma);
+ goto err;
+ }
}
- }
- gpa = (pte_ops->get_pfn(&e) << I915_GTT_PAGE_SHIFT)
- + (gma & ~I915_GTT_PAGE_MASK);
+ gpa = (pte_ops->get_pfn(&e) << I915_GTT_PAGE_SHIFT) +
+ (gma & ~I915_GTT_PAGE_MASK);
+ trace_gma_translate(vgpu->id, "ppgtt", 0,
+ mm->ppgtt_mm.root_entry_type, gma, gpa);
+ }
- trace_gma_translate(vgpu->id, "ppgtt", 0,
- mm->page_table_level, gma, gpa);
return gpa;
err:
gvt_vgpu_err("invalid mm type: %d gma %lx\n", mm->type, gma);
return INTEL_GVT_INVALID_ADDR;
}
-static int emulate_gtt_mmio_read(struct intel_vgpu *vgpu,
+static int emulate_ggtt_mmio_read(struct intel_vgpu *vgpu,
unsigned int off, void *p_data, unsigned int bytes)
{
struct intel_vgpu_mm *ggtt_mm = vgpu->gtt.ggtt_mm;
@@ -1887,7 +1804,7 @@ static int emulate_gtt_mmio_read(struct intel_vgpu *vgpu,
* Returns:
* Zero on success, error code if failed.
*/
-int intel_vgpu_emulate_gtt_mmio_read(struct intel_vgpu *vgpu, unsigned int off,
+int intel_vgpu_emulate_ggtt_mmio_read(struct intel_vgpu *vgpu, unsigned int off,
void *p_data, unsigned int bytes)
{
const struct intel_gvt_device_info *info = &vgpu->gvt->device_info;
@@ -1897,11 +1814,11 @@ int intel_vgpu_emulate_gtt_mmio_read(struct intel_vgpu *vgpu, unsigned int off,
return -EINVAL;
off -= info->gtt_start_offset;
- ret = emulate_gtt_mmio_read(vgpu, off, p_data, bytes);
+ ret = emulate_ggtt_mmio_read(vgpu, off, p_data, bytes);
return ret;
}
-static int emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off,
+static int emulate_ggtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off,
void *p_data, unsigned int bytes)
{
struct intel_gvt *gvt = vgpu->gvt;
@@ -1911,6 +1828,7 @@ static int emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off,
unsigned long g_gtt_index = off >> info->gtt_entry_size_shift;
unsigned long gma, gfn;
struct intel_gvt_gtt_entry e, m;
+ dma_addr_t dma_addr;
int ret;
if (bytes != 4 && bytes != 8)
@@ -1926,6 +1844,7 @@ static int emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off,
memcpy((void *)&e.val64 + (off & (info->gtt_entry_size - 1)), p_data,
bytes);
+ m = e;
if (ops->test_present(&e)) {
gfn = ops->get_pfn(&e);
@@ -1938,29 +1857,29 @@ static int emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off,
goto out;
}
- ret = gtt_entry_p2m(vgpu, &e, &m);
+ ret = intel_gvt_hypervisor_dma_map_guest_page(vgpu, gfn,
+ &dma_addr);
if (ret) {
- gvt_vgpu_err("fail to translate guest gtt entry\n");
+ gvt_vgpu_err("fail to populate guest ggtt entry\n");
/* guest driver may read/write the entry when partial
* update the entry in this situation p2m will fail
* settting the shadow entry to point to a scratch page
*/
ops->set_pfn(&m, gvt->gtt.scratch_mfn);
- }
- } else {
- m = e;
+ } else
+ ops->set_pfn(&m, dma_addr >> PAGE_SHIFT);
+ } else
ops->set_pfn(&m, gvt->gtt.scratch_mfn);
- }
out:
- ggtt_set_shadow_entry(ggtt_mm, &m, g_gtt_index);
- gtt_invalidate(gvt->dev_priv);
+ ggtt_set_host_entry(ggtt_mm, &m, g_gtt_index);
+ ggtt_invalidate(gvt->dev_priv);
ggtt_set_guest_entry(ggtt_mm, &e, g_gtt_index);
return 0;
}
/*
- * intel_vgpu_emulate_gtt_mmio_write - emulate GTT MMIO register write
+ * intel_vgpu_emulate_ggtt_mmio_write - emulate GTT MMIO register write
* @vgpu: a vGPU
* @off: register offset
* @p_data: data from guest write
@@ -1971,8 +1890,8 @@ out:
* Returns:
* Zero on success, error code if failed.
*/
-int intel_vgpu_emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off,
- void *p_data, unsigned int bytes)
+int intel_vgpu_emulate_ggtt_mmio_write(struct intel_vgpu *vgpu,
+ unsigned int off, void *p_data, unsigned int bytes)
{
const struct intel_gvt_device_info *info = &vgpu->gvt->device_info;
int ret;
@@ -1981,43 +1900,10 @@ int intel_vgpu_emulate_gtt_mmio_write(struct intel_vgpu *vgpu, unsigned int off,
return -EINVAL;
off -= info->gtt_start_offset;
- ret = emulate_gtt_mmio_write(vgpu, off, p_data, bytes);
- return ret;
-}
-
-int intel_vgpu_write_protect_handler(struct intel_vgpu *vgpu, u64 pa,
- void *p_data, unsigned int bytes)
-{
- struct intel_gvt *gvt = vgpu->gvt;
- int ret = 0;
-
- if (atomic_read(&vgpu->gtt.n_tracked_guest_page)) {
- struct intel_vgpu_page_track *t;
-
- mutex_lock(&gvt->lock);
-
- t = intel_vgpu_find_tracked_page(vgpu, pa >> PAGE_SHIFT);
- if (t) {
- if (unlikely(vgpu->failsafe)) {
- /* remove write protection to prevent furture traps */
- intel_vgpu_clean_page_track(vgpu, t);
- } else {
- ret = t->handler(t, pa, p_data, bytes);
- if (ret) {
- gvt_err("guest page write error %d, "
- "gfn 0x%lx, pa 0x%llx, "
- "var 0x%x, len %d\n",
- ret, t->gfn, pa,
- *(u32 *)p_data, bytes);
- }
- }
- }
- mutex_unlock(&gvt->lock);
- }
+ ret = emulate_ggtt_mmio_write(vgpu, off, p_data, bytes);
return ret;
}
-
static int alloc_scratch_pages(struct intel_vgpu *vgpu,
intel_gvt_gtt_type_t type)
{
@@ -2131,43 +2017,47 @@ err:
int intel_vgpu_init_gtt(struct intel_vgpu *vgpu)
{
struct intel_vgpu_gtt *gtt = &vgpu->gtt;
- struct intel_vgpu_mm *ggtt_mm;
- hash_init(gtt->tracked_guest_page_hash_table);
- hash_init(gtt->shadow_page_hash_table);
+ INIT_RADIX_TREE(&gtt->spt_tree, GFP_KERNEL);
- INIT_LIST_HEAD(&gtt->mm_list_head);
+ INIT_LIST_HEAD(&gtt->ppgtt_mm_list_head);
INIT_LIST_HEAD(&gtt->oos_page_list_head);
INIT_LIST_HEAD(&gtt->post_shadow_list_head);
- intel_vgpu_reset_ggtt(vgpu);
-
- ggtt_mm = intel_vgpu_create_mm(vgpu, INTEL_GVT_MM_GGTT,
- NULL, 1, 0);
- if (IS_ERR(ggtt_mm)) {
+ gtt->ggtt_mm = intel_vgpu_create_ggtt_mm(vgpu);
+ if (IS_ERR(gtt->ggtt_mm)) {
gvt_vgpu_err("fail to create mm for ggtt.\n");
- return PTR_ERR(ggtt_mm);
+ return PTR_ERR(gtt->ggtt_mm);
}
- gtt->ggtt_mm = ggtt_mm;
+ intel_vgpu_reset_ggtt(vgpu);
return create_scratch_page_tree(vgpu);
}
-static void intel_vgpu_free_mm(struct intel_vgpu *vgpu, int type)
+static void intel_vgpu_destroy_all_ppgtt_mm(struct intel_vgpu *vgpu)
{
struct list_head *pos, *n;
struct intel_vgpu_mm *mm;
- list_for_each_safe(pos, n, &vgpu->gtt.mm_list_head) {
- mm = container_of(pos, struct intel_vgpu_mm, list);
- if (mm->type == type) {
- vgpu->gvt->gtt.mm_free_page_table(mm);
- list_del(&mm->list);
- list_del(&mm->lru_list);
- kfree(mm);
- }
+ list_for_each_safe(pos, n, &vgpu->gtt.ppgtt_mm_list_head) {
+ mm = container_of(pos, struct intel_vgpu_mm, ppgtt_mm.list);
+ intel_vgpu_destroy_mm(mm);
}
+
+ if (GEM_WARN_ON(!list_empty(&vgpu->gtt.ppgtt_mm_list_head)))
+ gvt_err("vgpu ppgtt mm is not fully destroyed\n");
+
+ if (GEM_WARN_ON(!radix_tree_empty(&vgpu->gtt.spt_tree))) {
+ gvt_err("Why we still has spt not freed?\n");
+ ppgtt_free_all_spt(vgpu);
+ }
+}
+
+static void intel_vgpu_destroy_ggtt_mm(struct intel_vgpu *vgpu)
+{
+ intel_vgpu_destroy_mm(vgpu->gtt.ggtt_mm);
+ vgpu->gtt.ggtt_mm = NULL;
}
/**
@@ -2182,11 +2072,9 @@ static void intel_vgpu_free_mm(struct intel_vgpu *vgpu, int type)
*/
void intel_vgpu_clean_gtt(struct intel_vgpu *vgpu)
{
- ppgtt_free_all_shadow_page(vgpu);
+ intel_vgpu_destroy_all_ppgtt_mm(vgpu);
+ intel_vgpu_destroy_ggtt_mm(vgpu);
release_scratch_page_tree(vgpu);
-
- intel_vgpu_free_mm(vgpu, INTEL_GVT_MM_PPGTT);
- intel_vgpu_free_mm(vgpu, INTEL_GVT_MM_GGTT);
}
static void clean_spt_oos(struct intel_gvt *gvt)
@@ -2248,99 +2136,78 @@ fail:
* pointer to mm object on success, NULL if failed.
*/
struct intel_vgpu_mm *intel_vgpu_find_ppgtt_mm(struct intel_vgpu *vgpu,
- int page_table_level, void *root_entry)
+ u64 pdps[])
{
- struct list_head *pos;
struct intel_vgpu_mm *mm;
- u64 *src, *dst;
-
- list_for_each(pos, &vgpu->gtt.mm_list_head) {
- mm = container_of(pos, struct intel_vgpu_mm, list);
- if (mm->type != INTEL_GVT_MM_PPGTT)
- continue;
-
- if (mm->page_table_level != page_table_level)
- continue;
+ struct list_head *pos;
- src = root_entry;
- dst = mm->virtual_page_table;
+ list_for_each(pos, &vgpu->gtt.ppgtt_mm_list_head) {
+ mm = container_of(pos, struct intel_vgpu_mm, ppgtt_mm.list);
- if (page_table_level == 3) {
- if (src[0] == dst[0]
- && src[1] == dst[1]
- && src[2] == dst[2]
- && src[3] == dst[3])
+ switch (mm->ppgtt_mm.root_entry_type) {
+ case GTT_TYPE_PPGTT_ROOT_L4_ENTRY:
+ if (pdps[0] == mm->ppgtt_mm.guest_pdps[0])
return mm;
- } else {
- if (src[0] == dst[0])
+ break;
+ case GTT_TYPE_PPGTT_ROOT_L3_ENTRY:
+ if (!memcmp(pdps, mm->ppgtt_mm.guest_pdps,
+ sizeof(mm->ppgtt_mm.guest_pdps)))
return mm;
+ break;
+ default:
+ GEM_BUG_ON(1);
}
}
return NULL;
}
/**
- * intel_vgpu_g2v_create_ppgtt_mm - create a PPGTT mm object from
- * g2v notification
+ * intel_vgpu_get_ppgtt_mm - get or create a PPGTT mm object.
* @vgpu: a vGPU
- * @page_table_level: PPGTT page table level
+ * @root_entry_type: ppgtt root entry type
+ * @pdps: guest pdps
*
- * This function is used to create a PPGTT mm object from a guest to GVT-g
- * notification.
+ * This function is used to find or create a PPGTT mm object from a guest.
*
* Returns:
* Zero on success, negative error code if failed.
*/
-int intel_vgpu_g2v_create_ppgtt_mm(struct intel_vgpu *vgpu,
- int page_table_level)
+struct intel_vgpu_mm *intel_vgpu_get_ppgtt_mm(struct intel_vgpu *vgpu,
+ intel_gvt_gtt_type_t root_entry_type, u64 pdps[])
{
- u64 *pdp = (u64 *)&vgpu_vreg64_t(vgpu, vgtif_reg(pdp[0]));
struct intel_vgpu_mm *mm;
- if (WARN_ON((page_table_level != 4) && (page_table_level != 3)))
- return -EINVAL;
-
- mm = intel_vgpu_find_ppgtt_mm(vgpu, page_table_level, pdp);
+ mm = intel_vgpu_find_ppgtt_mm(vgpu, pdps);
if (mm) {
- intel_gvt_mm_reference(mm);
+ intel_vgpu_mm_get(mm);
} else {
- mm = intel_vgpu_create_mm(vgpu, INTEL_GVT_MM_PPGTT,
- pdp, page_table_level, 0);
- if (IS_ERR(mm)) {
+ mm = intel_vgpu_create_ppgtt_mm(vgpu, root_entry_type, pdps);
+ if (IS_ERR(mm))
gvt_vgpu_err("fail to create mm\n");
- return PTR_ERR(mm);
- }
}
- return 0;
+ return mm;
}
/**
- * intel_vgpu_g2v_destroy_ppgtt_mm - destroy a PPGTT mm object from
- * g2v notification
+ * intel_vgpu_put_ppgtt_mm - find and put a PPGTT mm object.
* @vgpu: a vGPU
- * @page_table_level: PPGTT page table level
+ * @pdps: guest pdps
*
- * This function is used to create a PPGTT mm object from a guest to GVT-g
- * notification.
+ * This function is used to find a PPGTT mm object from a guest and destroy it.
*
* Returns:
* Zero on success, negative error code if failed.
*/
-int intel_vgpu_g2v_destroy_ppgtt_mm(struct intel_vgpu *vgpu,
- int page_table_level)
+int intel_vgpu_put_ppgtt_mm(struct intel_vgpu *vgpu, u64 pdps[])
{
- u64 *pdp = (u64 *)&vgpu_vreg64_t(vgpu, vgtif_reg(pdp[0]));
struct intel_vgpu_mm *mm;
- if (WARN_ON((page_table_level != 4) && (page_table_level != 3)))
- return -EINVAL;
-
- mm = intel_vgpu_find_ppgtt_mm(vgpu, page_table_level, pdp);
+ mm = intel_vgpu_find_ppgtt_mm(vgpu, pdps);
if (!mm) {
gvt_vgpu_err("fail to find ppgtt instance.\n");
return -EINVAL;
}
- intel_gvt_mm_unreference(mm);
+ intel_vgpu_mm_put(mm);
return 0;
}
@@ -2367,8 +2234,6 @@ int intel_gvt_init_gtt(struct intel_gvt *gvt)
|| IS_KABYLAKE(gvt->dev_priv)) {
gvt->gtt.pte_ops = &gen8_gtt_pte_ops;
gvt->gtt.gma_ops = &gen8_gtt_gma_ops;
- gvt->gtt.mm_alloc_page_table = gen8_mm_alloc_page_table;
- gvt->gtt.mm_free_page_table = gen8_mm_free_page_table;
} else {
return -ENODEV;
}
@@ -2399,7 +2264,7 @@ int intel_gvt_init_gtt(struct intel_gvt *gvt)
return ret;
}
}
- INIT_LIST_HEAD(&gvt->gtt.mm_lru_list_head);
+ INIT_LIST_HEAD(&gvt->gtt.ppgtt_mm_lru_list_head);
return 0;
}
@@ -2426,6 +2291,28 @@ void intel_gvt_clean_gtt(struct intel_gvt *gvt)
}
/**
+ * intel_vgpu_invalidate_ppgtt - invalidate PPGTT instances
+ * @vgpu: a vGPU
+ *
+ * This function is called when invalidate all PPGTT instances of a vGPU.
+ *
+ */
+void intel_vgpu_invalidate_ppgtt(struct intel_vgpu *vgpu)
+{
+ struct list_head *pos, *n;
+ struct intel_vgpu_mm *mm;
+
+ list_for_each_safe(pos, n, &vgpu->gtt.ppgtt_mm_list_head) {
+ mm = container_of(pos, struct intel_vgpu_mm, ppgtt_mm.list);
+ if (mm->type == INTEL_GVT_MM_PPGTT) {
+ list_del_init(&mm->ppgtt_mm.lru_list);
+ if (mm->ppgtt_mm.shadowed)
+ invalidate_ppgtt_mm(mm);
+ }
+ }
+}
+
+/**
* intel_vgpu_reset_ggtt - reset the GGTT entry
* @vgpu: a vGPU
*
@@ -2437,28 +2324,25 @@ void intel_vgpu_reset_ggtt(struct intel_vgpu *vgpu)
{
struct intel_gvt *gvt = vgpu->gvt;
struct drm_i915_private *dev_priv = gvt->dev_priv;
- struct intel_gvt_gtt_pte_ops *ops = vgpu->gvt->gtt.pte_ops;
+ struct intel_gvt_gtt_pte_ops *pte_ops = vgpu->gvt->gtt.pte_ops;
+ struct intel_gvt_gtt_entry entry = {.type = GTT_TYPE_GGTT_PTE};
u32 index;
- u32 offset;
u32 num_entries;
- struct intel_gvt_gtt_entry e;
- memset(&e, 0, sizeof(struct intel_gvt_gtt_entry));
- e.type = GTT_TYPE_GGTT_PTE;
- ops->set_pfn(&e, gvt->gtt.scratch_mfn);
- e.val64 |= _PAGE_PRESENT;
+ pte_ops->set_pfn(&entry, gvt->gtt.scratch_mfn);
+ pte_ops->set_present(&entry);
index = vgpu_aperture_gmadr_base(vgpu) >> PAGE_SHIFT;
num_entries = vgpu_aperture_sz(vgpu) >> PAGE_SHIFT;
- for (offset = 0; offset < num_entries; offset++)
- ops->set_entry(NULL, &e, index + offset, false, 0, vgpu);
+ while (num_entries--)
+ ggtt_set_host_entry(vgpu->gtt.ggtt_mm, &entry, index++);
index = vgpu_hidden_gmadr_base(vgpu) >> PAGE_SHIFT;
num_entries = vgpu_hidden_sz(vgpu) >> PAGE_SHIFT;
- for (offset = 0; offset < num_entries; offset++)
- ops->set_entry(NULL, &e, index + offset, false, 0, vgpu);
+ while (num_entries--)
+ ggtt_set_host_entry(vgpu->gtt.ggtt_mm, &entry, index++);
- gtt_invalidate(dev_priv);
+ ggtt_invalidate(dev_priv);
}
/**
@@ -2471,13 +2355,10 @@ void intel_vgpu_reset_ggtt(struct intel_vgpu *vgpu)
*/
void intel_vgpu_reset_gtt(struct intel_vgpu *vgpu)
{
- ppgtt_free_all_shadow_page(vgpu);
-
/* Shadow pages are only created when there is no page
* table tracking data, so remove page tracking data after
* removing the shadow pages.
*/
- intel_vgpu_free_mm(vgpu, INTEL_GVT_MM_PPGTT);
-
+ intel_vgpu_destroy_all_ppgtt_mm(vgpu);
intel_vgpu_reset_ggtt(vgpu);
}
generated by cgit v1.2.3 (git 2.39.0) at 2024-11-23 00:02:25 +0300