Commit e502a2af authored by Changbin Du's avatar Changbin Du Committed by Zhenyu Wang
Browse files

drm/i915/gvt: Provide generic page_track infrastructure for write-protected page 



This patch provide generic page_track infrastructure for write-protected
guest page. The old page_track logic gets rewrote and now stays in a new
standalone page_track.c. This page track infrastructure can be both used
by vGUC and GTT shadowing.

The important change is that it uses radix tree instead of hash table.
We don't have a predictable number of pages that will be tracked.

Here is some performance data (duration in us) of looking up a element:
Before: (aka. intel_vgpu_find_tracked_page)
 0.091 0.089 0.090 ... 0.093 0.091 0.087 ... 0.292 0.285 0.292 0.291
After: (aka. intel_vgpu_find_page_track)
 0.104 0.105 0.100 0.102 0.102 0.100 ... 0.101 0.101 0.105 0.105

The hash table has good performance at beginning, but turns bad with
more pages being tracked even no 3D applications are running. As
expected, radix tree has stable duration and very quick.

The overall benchmark (tested with Heaven Benchmark) marginally improved
since this is not the bottleneck. What we benefit more from this change
is scalability.

Signed-off-by: default avatarChangbin Du <changbin.du@intel.com>
Signed-off-by: default avatarZhenyu Wang <zhenyuw@linux.intel.com>
parent 09475728

drivers/gpu/drm/i915/gvt/Makefile

+1 −1
Original line number Diff line number Diff line
@@ -3,7 +3,7 @@ GVT_DIR := gvt 
GVT_SOURCE := gvt.o aperture_gm.o handlers.o vgpu.o trace_points.o firmware.o \

	interrupt.o gtt.o cfg_space.o opregion.o mmio.o display.o edid.o \

	execlist.o scheduler.o sched_policy.o mmio_context.o cmd_parser.o debugfs.o \

	fb_decoder.o dmabuf.o

	fb_decoder.o dmabuf.o page_track.o



ccflags-y				+= -I$(src) -I$(src)/$(GVT_DIR)

i915-y					+= $(addprefix $(GVT_DIR)/, $(GVT_SOURCE))

drivers/gpu/drm/i915/gvt/gtt.c

+24 −95
Original line number Diff line number Diff line
@@ -559,7 +559,7 @@ 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;
@@ -587,7 +587,7 @@ static inline int ppgtt_spt_set_entry( 
		    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);

}
@@ -607,9 +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)



#define page_track_to_ppgtt_spt(ptr) \

	container_of(ptr, struct intel_vgpu_ppgtt_spt, guest_page.track)



static void *alloc_spt(gfp_t gfp_mask)

{

	struct intel_vgpu_ppgtt_spt *spt;
@@ -632,30 +629,6 @@ static void free_spt(struct intel_vgpu_ppgtt_spt *spt) 
	kfree(spt);

}



/**

 * 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 detach_oos_page(struct intel_vgpu *vgpu,

		struct intel_vgpu_oos_page *oos_page);
@@ -673,12 +646,7 @@ static void ppgtt_free_spt(struct intel_vgpu_ppgtt_spt *spt) 
	if (spt->guest_page.oos_page)

		detach_oos_page(spt->vgpu, spt->guest_page.oos_page);



	if (!hlist_unhashed(&spt->guest_page.track.node))

		hash_del(&spt->guest_page.track.node);



	if (spt->guest_page.track.tracked)

		intel_gvt_hypervisor_disable_page_track(spt->vgpu,

				spt->guest_page.track.gfn);

	intel_vgpu_unregister_page_track(spt->vgpu, spt->guest_page.gfn);



	list_del_init(&spt->post_shadow_list);

	free_spt(spt);
@@ -698,21 +666,18 @@ static int ppgtt_handle_guest_write_page_table_bytes( 
		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_ppgtt_spt *spt = page_track_to_ppgtt_spt(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(spt,

		pa, p_data, bytes);

	ret = ppgtt_handle_guest_write_page_table_bytes(spt, gpa, data, bytes);

	if (ret)

		return ret;

	return ret;
@@ -724,9 +689,9 @@ static struct intel_vgpu_ppgtt_spt *intel_vgpu_find_spt_by_gfn( 
{

	struct intel_vgpu_page_track *track;



	track = intel_vgpu_find_tracked_page(vgpu, gfn);

	if (track)

		return page_track_to_ppgtt_spt(track);

	track = intel_vgpu_find_page_track(vgpu, gfn);

	if (track && track->handler == ppgtt_write_protection_handler)

		return track->priv_data;



	return NULL;

}
@@ -752,6 +717,7 @@ static struct intel_vgpu_ppgtt_spt *ppgtt_alloc_spt( 
	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);
@@ -787,10 +753,13 @@ retry: 
	spt->guest_page.type = type;

	spt->guest_page.gfn = gfn;



	spt->guest_page.track.gfn = gfn;

	spt->guest_page.track.handler = ppgtt_write_protection_handler;

	hash_add(vgpu->gtt.tracked_guest_page_hash_table,

		&spt->guest_page.track.node, gfn);

	ret = intel_vgpu_register_page_track(vgpu, spt->guest_page.gfn,

					ppgtt_write_protection_handler, spt);

	if (ret) {

		free_spt(spt);

		dma_unmap_page(kdev, daddr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);

		return ERR_PTR(ret);

	}



	INIT_HLIST_NODE(&spt->node);

	hash_add(vgpu->gtt.spt_hash_table, &spt->node, spt->shadow_page.mfn);
@@ -926,11 +895,10 @@ static struct intel_vgpu_ppgtt_spt *ppgtt_populate_spt_by_guest_entry( 
			goto fail;

		}



		ret = intel_gvt_hypervisor_enable_page_track(vgpu, spt->guest_page.track.gfn);

		ret = intel_vgpu_enable_page_track(vgpu, spt->guest_page.gfn);

		if (ret)

			goto fail;



		spt->guest_page.track.tracked = true;

		ret = ppgtt_populate_spt(spt);

		if (ret)

			goto fail;
@@ -1002,7 +970,7 @@ static int ppgtt_populate_spt(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);



	for_each_present_guest_entry(spt, &ge, i) {

		if (gtt_type_is_pt(get_next_pt_type(ge.type))) {
@@ -1197,10 +1165,9 @@ 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(spt->vgpu, spt->guest_page.track.gfn);

	ret = intel_vgpu_enable_page_track(spt->vgpu, spt->guest_page.gfn);

	if (ret)

		return ret;

	spt->guest_page.track.tracked = true;



	trace_oos_change(spt->vgpu->id, "set page sync", oos_page->id,

			 spt, spt->guest_page.type);
@@ -1236,7 +1203,6 @@ static int ppgtt_allocate_oos_page(struct intel_vgpu_ppgtt_spt *spt) 
static int ppgtt_set_guest_page_oos(struct intel_vgpu_ppgtt_spt *spt)

{

	struct intel_vgpu_oos_page *oos_page = spt->guest_page.oos_page;

	int ret;



	if (WARN(!oos_page, "shadow PPGTT page should have a oos page\n"))

		return -EINVAL;
@@ -1245,11 +1211,7 @@ static int ppgtt_set_guest_page_oos(struct intel_vgpu_ppgtt_spt *spt) 
			 spt, spt->guest_page.type);



	list_add_tail(&oos_page->vm_list, &spt->vgpu->gtt.oos_page_list_head);

	ret = intel_gvt_hypervisor_disable_page_track(spt->vgpu, spt->guest_page.track.gfn);

	if (ret)

		return ret;

	spt->guest_page.track.tracked = false;

	return 0;

	return intel_vgpu_disable_page_track(spt->vgpu, spt->guest_page.gfn);

}



/**
@@ -1918,38 +1880,6 @@ int intel_vgpu_emulate_ggtt_mmio_write(struct intel_vgpu *vgpu, 
	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;



	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_gvt_hypervisor_disable_page_track(vgpu, t->gfn);

		} 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);



	return ret;

}





static int alloc_scratch_pages(struct intel_vgpu *vgpu,

		intel_gvt_gtt_type_t type)

{
@@ -2064,7 +1994,6 @@ int intel_vgpu_init_gtt(struct intel_vgpu *vgpu) 
{

	struct intel_vgpu_gtt *gtt = &vgpu->gtt;



	hash_init(gtt->tracked_guest_page_hash_table);

	hash_init(gtt->spt_hash_table);



	INIT_LIST_HEAD(&gtt->ppgtt_mm_list_head);

drivers/gpu/drm/i915/gvt/gtt.h

+0 −14
Original line number Diff line number Diff line
@@ -187,8 +187,6 @@ struct intel_vgpu_gtt { 
	unsigned long active_ppgtt_mm_bitmap;

	struct list_head ppgtt_mm_list_head;

	DECLARE_HASHTABLE(spt_hash_table, INTEL_GVT_GTT_HASH_BITS);

	DECLARE_HASHTABLE(tracked_guest_page_hash_table, INTEL_GVT_GTT_HASH_BITS);

	atomic_t n_tracked_guest_page;

	struct list_head oos_page_list_head;

	struct list_head post_shadow_list_head;

	struct intel_vgpu_scratch_pt scratch_pt[GTT_TYPE_MAX];
@@ -205,14 +203,6 @@ extern void intel_gvt_clean_gtt(struct intel_gvt *gvt); 
extern struct intel_vgpu_mm *intel_gvt_find_ppgtt_mm(struct intel_vgpu *vgpu,

		int page_table_level, void *root_entry);



struct intel_vgpu_page_track {

	struct hlist_node node;

	bool tracked;

	unsigned long gfn;

	int (*handler)(void *, u64, void *, int);

	void *data;

};



struct intel_vgpu_oos_page {

	struct intel_vgpu_ppgtt_spt *spt;

	struct list_head list;
@@ -240,7 +230,6 @@ struct intel_vgpu_ppgtt_spt { 
		intel_gvt_gtt_type_t type;

		unsigned long gfn;

		unsigned long write_cnt;

		struct intel_vgpu_page_track track;

		struct intel_vgpu_oos_page *oos_page;

	} guest_page;
@@ -273,7 +262,4 @@ int intel_vgpu_emulate_ggtt_mmio_read(struct intel_vgpu *vgpu, 
int intel_vgpu_emulate_ggtt_mmio_write(struct intel_vgpu *vgpu,

	unsigned int off, void *p_data, unsigned int bytes);



int intel_vgpu_write_protect_handler(struct intel_vgpu *vgpu, u64 pa,

				     void *p_data, unsigned int bytes);



#endif /* _GVT_GTT_H_ */

drivers/gpu/drm/i915/gvt/gvt.c

+1 −1
Original line number Diff line number Diff line
@@ -183,7 +183,7 @@ static const struct intel_gvt_ops intel_gvt_ops = { 
	.get_gvt_attrs = intel_get_gvt_attrs,

	.vgpu_query_plane = intel_vgpu_query_plane,

	.vgpu_get_dmabuf = intel_vgpu_get_dmabuf,

	.write_protect_handler = intel_vgpu_write_protect_handler,

	.write_protect_handler = intel_vgpu_page_track_handler,

};



/**

drivers/gpu/drm/i915/gvt/gvt.h

+2 −0
Original line number Diff line number Diff line
@@ -48,6 +48,7 @@ 
#include "cmd_parser.h"

#include "fb_decoder.h"

#include "dmabuf.h"

#include "page_track.h"



#define GVT_MAX_VGPU 8
@@ -190,6 +191,7 @@ struct intel_vgpu { 
	struct intel_vgpu_opregion opregion;

	struct intel_vgpu_display display;

	struct intel_vgpu_submission submission;

	struct radix_tree_root page_track_tree;

	u32 hws_pga[I915_NUM_ENGINES];



	struct dentry *debugfs;

CRA Git | Maintained and supported by SUSTech CRA and CCSE