Commit a69035eb authored by Daniel Vetter's avatar Daniel Vetter

Merge tag 'gvt-next-2017-03-30' of https://github.com/01org/gvt-linux into drm-intel-next-queued

gvt-next-2017-03-30

- Add mdev attribute group for per-vgpu info
- Time slice based vGPU scheduling QoS support (Gao Ping)
- Initial KBL support for E3 server (Han Xu)
- other misc.

Link: http://patchwork.freedesktop.org/patch/msgid/20170330100516.dkavi3rtlsmnoepi@zhen-hp.sh.intel.comSigned-off-by: default avatarDaniel Vetter <daniel.vetter@ffwll.ch>
parents 228ec87c b35f34d1
......@@ -1215,7 +1215,7 @@ static int gen8_check_mi_display_flip(struct parser_exec_state *s,
if (!info->async_flip)
return 0;
if (IS_SKYLAKE(dev_priv)) {
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
stride = vgpu_vreg(s->vgpu, info->stride_reg) & GENMASK(9, 0);
tile = (vgpu_vreg(s->vgpu, info->ctrl_reg) &
GENMASK(12, 10)) >> 10;
......@@ -1243,7 +1243,7 @@ static int gen8_update_plane_mmio_from_mi_display_flip(
set_mask_bits(&vgpu_vreg(vgpu, info->surf_reg), GENMASK(31, 12),
info->surf_val << 12);
if (IS_SKYLAKE(dev_priv)) {
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
set_mask_bits(&vgpu_vreg(vgpu, info->stride_reg), GENMASK(9, 0),
info->stride_val);
set_mask_bits(&vgpu_vreg(vgpu, info->ctrl_reg), GENMASK(12, 10),
......@@ -1267,7 +1267,7 @@ static int decode_mi_display_flip(struct parser_exec_state *s,
if (IS_BROADWELL(dev_priv))
return gen8_decode_mi_display_flip(s, info);
if (IS_SKYLAKE(dev_priv))
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
return skl_decode_mi_display_flip(s, info);
return -ENODEV;
......@@ -1278,7 +1278,9 @@ static int check_mi_display_flip(struct parser_exec_state *s,
{
struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv;
if (IS_BROADWELL(dev_priv) || IS_SKYLAKE(dev_priv))
if (IS_BROADWELL(dev_priv)
|| IS_SKYLAKE(dev_priv)
|| IS_KABYLAKE(dev_priv))
return gen8_check_mi_display_flip(s, info);
return -ENODEV;
}
......@@ -1289,7 +1291,9 @@ static int update_plane_mmio_from_mi_display_flip(
{
struct drm_i915_private *dev_priv = s->vgpu->gvt->dev_priv;
if (IS_BROADWELL(dev_priv) || IS_SKYLAKE(dev_priv))
if (IS_BROADWELL(dev_priv)
|| IS_SKYLAKE(dev_priv)
|| IS_KABYLAKE(dev_priv))
return gen8_update_plane_mmio_from_mi_display_flip(s, info);
return -ENODEV;
}
......@@ -1569,7 +1573,8 @@ static int batch_buffer_needs_scan(struct parser_exec_state *s)
{
struct intel_gvt *gvt = s->vgpu->gvt;
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) {
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)
|| IS_KABYLAKE(gvt->dev_priv)) {
/* BDW decides privilege based on address space */
if (cmd_val(s, 0) & (1 << 8))
return 0;
......@@ -2604,6 +2609,9 @@ static int scan_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
unsigned long gma_head, gma_tail, gma_bottom, ring_size, ring_tail;
struct parser_exec_state s;
int ret = 0;
struct intel_vgpu_workload *workload = container_of(wa_ctx,
struct intel_vgpu_workload,
wa_ctx);
/* ring base is page aligned */
if (WARN_ON(!IS_ALIGNED(wa_ctx->indirect_ctx.guest_gma, GTT_PAGE_SIZE)))
......@@ -2618,14 +2626,14 @@ static int scan_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
s.buf_type = RING_BUFFER_INSTRUCTION;
s.buf_addr_type = GTT_BUFFER;
s.vgpu = wa_ctx->workload->vgpu;
s.ring_id = wa_ctx->workload->ring_id;
s.vgpu = workload->vgpu;
s.ring_id = workload->ring_id;
s.ring_start = wa_ctx->indirect_ctx.guest_gma;
s.ring_size = ring_size;
s.ring_head = gma_head;
s.ring_tail = gma_tail;
s.rb_va = wa_ctx->indirect_ctx.shadow_va;
s.workload = wa_ctx->workload;
s.workload = workload;
ret = ip_gma_set(&s, gma_head);
if (ret)
......@@ -2708,12 +2716,15 @@ static int shadow_indirect_ctx(struct intel_shadow_wa_ctx *wa_ctx)
{
int ctx_size = wa_ctx->indirect_ctx.size;
unsigned long guest_gma = wa_ctx->indirect_ctx.guest_gma;
struct intel_vgpu *vgpu = wa_ctx->workload->vgpu;
struct intel_vgpu_workload *workload = container_of(wa_ctx,
struct intel_vgpu_workload,
wa_ctx);
struct intel_vgpu *vgpu = workload->vgpu;
struct drm_i915_gem_object *obj;
int ret = 0;
void *map;
obj = i915_gem_object_create(wa_ctx->workload->vgpu->gvt->dev_priv,
obj = i915_gem_object_create(workload->vgpu->gvt->dev_priv,
roundup(ctx_size + CACHELINE_BYTES,
PAGE_SIZE));
if (IS_ERR(obj))
......@@ -2733,8 +2744,8 @@ static int shadow_indirect_ctx(struct intel_shadow_wa_ctx *wa_ctx)
goto unmap_src;
}
ret = copy_gma_to_hva(wa_ctx->workload->vgpu,
wa_ctx->workload->vgpu->gtt.ggtt_mm,
ret = copy_gma_to_hva(workload->vgpu,
workload->vgpu->gtt.ggtt_mm,
guest_gma, guest_gma + ctx_size,
map);
if (ret < 0) {
......@@ -2772,7 +2783,10 @@ static int combine_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
int intel_gvt_scan_and_shadow_wa_ctx(struct intel_shadow_wa_ctx *wa_ctx)
{
int ret;
struct intel_vgpu *vgpu = wa_ctx->workload->vgpu;
struct intel_vgpu_workload *workload = container_of(wa_ctx,
struct intel_vgpu_workload,
wa_ctx);
struct intel_vgpu *vgpu = workload->vgpu;
if (wa_ctx->indirect_ctx.size == 0)
return 0;
......
......@@ -161,8 +161,9 @@ static unsigned char virtual_dp_monitor_edid[GVT_EDID_NUM][EDID_SIZE] = {
#define DPCD_HEADER_SIZE 0xb
/* let the virtual display supports DP1.2 */
static u8 dpcd_fix_data[DPCD_HEADER_SIZE] = {
0x11, 0x0a, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
0x12, 0x014, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
};
static void emulate_monitor_status_change(struct intel_vgpu *vgpu)
......@@ -172,9 +173,20 @@ static void emulate_monitor_status_change(struct intel_vgpu *vgpu)
SDE_PORTC_HOTPLUG_CPT |
SDE_PORTD_HOTPLUG_CPT);
if (IS_SKYLAKE(dev_priv))
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
vgpu_vreg(vgpu, SDEISR) &= ~(SDE_PORTA_HOTPLUG_SPT |
SDE_PORTE_HOTPLUG_SPT);
vgpu_vreg(vgpu, SKL_FUSE_STATUS) |=
SKL_FUSE_DOWNLOAD_STATUS |
SKL_FUSE_PG0_DIST_STATUS |
SKL_FUSE_PG1_DIST_STATUS |
SKL_FUSE_PG2_DIST_STATUS;
vgpu_vreg(vgpu, LCPLL1_CTL) |=
LCPLL_PLL_ENABLE |
LCPLL_PLL_LOCK;
vgpu_vreg(vgpu, LCPLL2_CTL) |= LCPLL_PLL_ENABLE;
}
if (intel_vgpu_has_monitor_on_port(vgpu, PORT_B)) {
vgpu_vreg(vgpu, SDEISR) |= SDE_PORTB_HOTPLUG_CPT;
......@@ -191,7 +203,7 @@ static void emulate_monitor_status_change(struct intel_vgpu *vgpu)
vgpu_vreg(vgpu, SFUSE_STRAP) |= SFUSE_STRAP_DDID_DETECTED;
}
if (IS_SKYLAKE(dev_priv) &&
if ((IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) &&
intel_vgpu_has_monitor_on_port(vgpu, PORT_E)) {
vgpu_vreg(vgpu, SDEISR) |= SDE_PORTE_HOTPLUG_SPT;
}
......@@ -353,7 +365,7 @@ void intel_vgpu_clean_display(struct intel_vgpu *vgpu)
{
struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
if (IS_SKYLAKE(dev_priv))
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
clean_virtual_dp_monitor(vgpu, PORT_D);
else
clean_virtual_dp_monitor(vgpu, PORT_B);
......@@ -375,7 +387,7 @@ int intel_vgpu_init_display(struct intel_vgpu *vgpu, u64 resolution)
intel_vgpu_init_i2c_edid(vgpu);
if (IS_SKYLAKE(dev_priv))
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
return setup_virtual_dp_monitor(vgpu, PORT_D, GVT_DP_D,
resolution);
else
......
......@@ -394,9 +394,11 @@ static void prepare_shadow_batch_buffer(struct intel_vgpu_workload *workload)
static int update_wa_ctx_2_shadow_ctx(struct intel_shadow_wa_ctx *wa_ctx)
{
int ring_id = wa_ctx->workload->ring_id;
struct i915_gem_context *shadow_ctx =
wa_ctx->workload->vgpu->shadow_ctx;
struct intel_vgpu_workload *workload = container_of(wa_ctx,
struct intel_vgpu_workload,
wa_ctx);
int ring_id = workload->ring_id;
struct i915_gem_context *shadow_ctx = workload->vgpu->shadow_ctx;
struct drm_i915_gem_object *ctx_obj =
shadow_ctx->engine[ring_id].state->obj;
struct execlist_ring_context *shadow_ring_context;
......@@ -680,7 +682,6 @@ static int submit_context(struct intel_vgpu *vgpu, int ring_id,
CACHELINE_BYTES;
workload->wa_ctx.per_ctx.guest_gma =
per_ctx & PER_CTX_ADDR_MASK;
workload->wa_ctx.workload = workload;
WARN_ON(workload->wa_ctx.indirect_ctx.size && !(per_ctx & 0x1));
}
......
......@@ -2220,7 +2220,8 @@ int intel_gvt_init_gtt(struct intel_gvt *gvt)
gvt_dbg_core("init gtt\n");
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) {
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)
|| 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;
......
......@@ -106,7 +106,8 @@ static void init_device_info(struct intel_gvt *gvt)
struct intel_gvt_device_info *info = &gvt->device_info;
struct pci_dev *pdev = gvt->dev_priv->drm.pdev;
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) {
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)
|| IS_KABYLAKE(gvt->dev_priv)) {
info->max_support_vgpus = 8;
info->cfg_space_size = 256;
info->mmio_size = 2 * 1024 * 1024;
......@@ -143,6 +144,11 @@ static int gvt_service_thread(void *data)
intel_gvt_emulate_vblank(gvt);
mutex_unlock(&gvt->lock);
}
if (test_and_clear_bit(INTEL_GVT_REQUEST_SCHED,
(void *)&gvt->service_request)) {
intel_gvt_schedule(gvt);
}
}
return 0;
......@@ -196,6 +202,8 @@ void intel_gvt_clean_device(struct drm_i915_private *dev_priv)
idr_destroy(&gvt->vgpu_idr);
intel_gvt_destroy_idle_vgpu(gvt->idle_vgpu);
kfree(dev_priv->gvt);
dev_priv->gvt = NULL;
}
......@@ -214,6 +222,7 @@ void intel_gvt_clean_device(struct drm_i915_private *dev_priv)
int intel_gvt_init_device(struct drm_i915_private *dev_priv)
{
struct intel_gvt *gvt;
struct intel_vgpu *vgpu;
int ret;
/*
......@@ -286,6 +295,14 @@ int intel_gvt_init_device(struct drm_i915_private *dev_priv)
goto out_clean_types;
}
vgpu = intel_gvt_create_idle_vgpu(gvt);
if (IS_ERR(vgpu)) {
ret = PTR_ERR(vgpu);
gvt_err("failed to create idle vgpu\n");
goto out_clean_types;
}
gvt->idle_vgpu = vgpu;
gvt_dbg_core("gvt device initialization is done\n");
dev_priv->gvt = gvt;
return 0;
......
......@@ -138,6 +138,10 @@ struct intel_vgpu_display {
struct intel_vgpu_sbi sbi;
};
struct vgpu_sched_ctl {
int weight;
};
struct intel_vgpu {
struct intel_gvt *gvt;
int id;
......@@ -147,6 +151,7 @@ struct intel_vgpu {
bool failsafe;
bool resetting;
void *sched_data;
struct vgpu_sched_ctl sched_ctl;
struct intel_vgpu_fence fence;
struct intel_vgpu_gm gm;
......@@ -160,6 +165,7 @@ struct intel_vgpu {
struct list_head workload_q_head[I915_NUM_ENGINES];
struct kmem_cache *workloads;
atomic_t running_workload_num;
ktime_t last_ctx_submit_time;
DECLARE_BITMAP(tlb_handle_pending, I915_NUM_ENGINES);
struct i915_gem_context *shadow_ctx;
......@@ -215,6 +221,7 @@ struct intel_vgpu_type {
unsigned int low_gm_size;
unsigned int high_gm_size;
unsigned int fence;
unsigned int weight;
enum intel_vgpu_edid resolution;
};
......@@ -236,6 +243,7 @@ struct intel_gvt {
DECLARE_HASHTABLE(cmd_table, GVT_CMD_HASH_BITS);
struct intel_vgpu_type *types;
unsigned int num_types;
struct intel_vgpu *idle_vgpu;
struct task_struct *service_thread;
wait_queue_head_t service_thread_wq;
......@@ -249,6 +257,7 @@ static inline struct intel_gvt *to_gvt(struct drm_i915_private *i915)
enum {
INTEL_GVT_REQUEST_EMULATE_VBLANK = 0,
INTEL_GVT_REQUEST_SCHED = 1,
};
static inline void intel_gvt_request_service(struct intel_gvt *gvt,
......@@ -322,6 +331,8 @@ struct intel_vgpu_creation_params {
__u64 resolution;
__s32 primary;
__u64 vgpu_id;
__u32 weight;
};
int intel_vgpu_alloc_resource(struct intel_vgpu *vgpu,
......@@ -376,6 +387,8 @@ static inline void intel_vgpu_write_pci_bar(struct intel_vgpu *vgpu,
int intel_gvt_init_vgpu_types(struct intel_gvt *gvt);
void intel_gvt_clean_vgpu_types(struct intel_gvt *gvt);
struct intel_vgpu *intel_gvt_create_idle_vgpu(struct intel_gvt *gvt);
void intel_gvt_destroy_idle_vgpu(struct intel_vgpu *vgpu);
struct intel_vgpu *intel_gvt_create_vgpu(struct intel_gvt *gvt,
struct intel_vgpu_type *type);
void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu);
......
This diff is collapsed.
......@@ -580,7 +580,7 @@ static void gen8_init_irq(
SET_BIT_INFO(irq, 4, PRIMARY_C_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_C);
SET_BIT_INFO(irq, 5, SPRITE_C_FLIP_DONE, INTEL_GVT_IRQ_INFO_DE_PIPE_C);
} else if (IS_SKYLAKE(gvt->dev_priv)) {
} else if (IS_SKYLAKE(gvt->dev_priv) || IS_KABYLAKE(gvt->dev_priv)) {
SET_BIT_INFO(irq, 25, AUX_CHANNEL_B, INTEL_GVT_IRQ_INFO_DE_PORT);
SET_BIT_INFO(irq, 26, AUX_CHANNEL_C, INTEL_GVT_IRQ_INFO_DE_PORT);
SET_BIT_INFO(irq, 27, AUX_CHANNEL_D, INTEL_GVT_IRQ_INFO_DE_PORT);
......@@ -690,7 +690,8 @@ int intel_gvt_init_irq(struct intel_gvt *gvt)
gvt_dbg_core("init irq framework\n");
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)) {
if (IS_BROADWELL(gvt->dev_priv) || IS_SKYLAKE(gvt->dev_priv)
|| IS_KABYLAKE(gvt->dev_priv)) {
irq->ops = &gen8_irq_ops;
irq->irq_map = gen8_irq_map;
} else {
......
......@@ -295,10 +295,12 @@ static ssize_t description_show(struct kobject *kobj, struct device *dev,
return 0;
return sprintf(buf, "low_gm_size: %dMB\nhigh_gm_size: %dMB\n"
"fence: %d\nresolution: %s\n",
"fence: %d\nresolution: %s\n"
"weight: %d\n",
BYTES_TO_MB(type->low_gm_size),
BYTES_TO_MB(type->high_gm_size),
type->fence, vgpu_edid_str(type->resolution));
type->fence, vgpu_edid_str(type->resolution),
type->weight);
}
static MDEV_TYPE_ATTR_RO(available_instances);
......@@ -1146,8 +1148,40 @@ static long intel_vgpu_ioctl(struct mdev_device *mdev, unsigned int cmd,
return 0;
}
static ssize_t
vgpu_id_show(struct device *dev, struct device_attribute *attr,
char *buf)
{
struct mdev_device *mdev = mdev_from_dev(dev);
if (mdev) {
struct intel_vgpu *vgpu = (struct intel_vgpu *)
mdev_get_drvdata(mdev);
return sprintf(buf, "%d\n", vgpu->id);
}
return sprintf(buf, "\n");
}
static DEVICE_ATTR_RO(vgpu_id);
static struct attribute *intel_vgpu_attrs[] = {
&dev_attr_vgpu_id.attr,
NULL
};
static const struct attribute_group intel_vgpu_group = {
.name = "intel_vgpu",
.attrs = intel_vgpu_attrs,
};
static const struct attribute_group *intel_vgpu_groups[] = {
&intel_vgpu_group,
NULL,
};
static const struct mdev_parent_ops intel_vgpu_ops = {
.supported_type_groups = intel_vgpu_type_groups,
.mdev_attr_groups = intel_vgpu_groups,
.create = intel_vgpu_create,
.remove = intel_vgpu_remove,
......@@ -1339,13 +1373,6 @@ static int kvmgt_guest_init(struct mdev_device *mdev)
static bool kvmgt_guest_exit(struct kvmgt_guest_info *info)
{
struct intel_vgpu *vgpu = info->vgpu;
if (!info) {
gvt_vgpu_err("kvmgt_guest_info invalid\n");
return false;
}
kvm_page_track_unregister_notifier(info->kvm, &info->track_node);
kvmgt_protect_table_destroy(info);
gvt_cache_destroy(info->vgpu);
......
......@@ -44,20 +44,21 @@ struct intel_vgpu;
#define D_HSW (1 << 2)
#define D_BDW (1 << 3)
#define D_SKL (1 << 4)
#define D_KBL (1 << 5)
#define D_GEN9PLUS (D_SKL)
#define D_GEN8PLUS (D_BDW | D_SKL)
#define D_GEN75PLUS (D_HSW | D_BDW | D_SKL)
#define D_GEN7PLUS (D_IVB | D_HSW | D_BDW | D_SKL)
#define D_GEN9PLUS (D_SKL | D_KBL)
#define D_GEN8PLUS (D_BDW | D_SKL | D_KBL)
#define D_GEN75PLUS (D_HSW | D_BDW | D_SKL | D_KBL)
#define D_GEN7PLUS (D_IVB | D_HSW | D_BDW | D_SKL | D_KBL)
#define D_SKL_PLUS (D_SKL)
#define D_BDW_PLUS (D_BDW | D_SKL)
#define D_HSW_PLUS (D_HSW | D_BDW | D_SKL)
#define D_IVB_PLUS (D_IVB | D_HSW | D_BDW | D_SKL)
#define D_SKL_PLUS (D_SKL | D_KBL)
#define D_BDW_PLUS (D_BDW | D_SKL | D_KBL)
#define D_HSW_PLUS (D_HSW | D_BDW | D_SKL | D_KBL)
#define D_IVB_PLUS (D_IVB | D_HSW | D_BDW | D_SKL | D_KBL)
#define D_PRE_BDW (D_SNB | D_IVB | D_HSW)
#define D_PRE_SKL (D_SNB | D_IVB | D_HSW | D_BDW)
#define D_ALL (D_SNB | D_IVB | D_HSW | D_BDW | D_SKL)
#define D_ALL (D_SNB | D_IVB | D_HSW | D_BDW | D_SKL | D_KBL)
struct intel_gvt_mmio_info {
u32 offset;
......
......@@ -126,6 +126,18 @@ static struct render_mmio gen9_render_mmio_list[] = {
{VCS2, _MMIO(0x1c028), 0xffff, false},
{VECS, _MMIO(0x1a028), 0xffff, false},
{RCS, _MMIO(0x7304), 0xffff, true},
{RCS, _MMIO(0x2248), 0x0, false},
{RCS, _MMIO(0x940c), 0x0, false},
{RCS, _MMIO(0x4ab8), 0x0, false},
{RCS, _MMIO(0x4ab0), 0x0, false},
{RCS, _MMIO(0x20d4), 0x0, false},
{RCS, _MMIO(0xb004), 0x0, false},
{RCS, _MMIO(0x20a0), 0x0, false},
{RCS, _MMIO(0x20e4), 0xffff, false},
};
static u32 gen9_render_mocs[I915_NUM_ENGINES][64];
......@@ -159,7 +171,7 @@ static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
*/
fw = intel_uncore_forcewake_for_reg(dev_priv, reg,
FW_REG_READ | FW_REG_WRITE);
if (ring_id == RCS && IS_SKYLAKE(dev_priv))
if (ring_id == RCS && (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)))
fw |= FORCEWAKE_RENDER;
intel_uncore_forcewake_get(dev_priv, fw);
......@@ -192,7 +204,7 @@ static void load_mocs(struct intel_vgpu *vgpu, int ring_id)
if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
return;
if (!IS_SKYLAKE(dev_priv))
if (!(IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)))
return;
offset.reg = regs[ring_id];
......@@ -230,7 +242,7 @@ static void restore_mocs(struct intel_vgpu *vgpu, int ring_id)
if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
return;
if (!IS_SKYLAKE(dev_priv))
if (!(IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)))
return;
offset.reg = regs[ring_id];
......@@ -265,7 +277,8 @@ void intel_gvt_load_render_mmio(struct intel_vgpu *vgpu, int ring_id)
u32 inhibit_mask =
_MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
if (IS_SKYLAKE(vgpu->gvt->dev_priv)) {
if (IS_SKYLAKE(vgpu->gvt->dev_priv)
|| IS_KABYLAKE(vgpu->gvt->dev_priv)) {
mmio = gen9_render_mmio_list;
array_size = ARRAY_SIZE(gen9_render_mmio_list);
load_mocs(vgpu, ring_id);
......@@ -312,7 +325,7 @@ void intel_gvt_restore_render_mmio(struct intel_vgpu *vgpu, int ring_id)
u32 v;
int i, array_size;
if (IS_SKYLAKE(dev_priv)) {
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv)) {
mmio = gen9_render_mmio_list;
array_size = ARRAY_SIZE(gen9_render_mmio_list);
restore_mocs(vgpu, ring_id);
......
......@@ -47,11 +47,87 @@ static bool vgpu_has_pending_workload(struct intel_vgpu *vgpu)
return false;
}
struct vgpu_sched_data {
struct list_head lru_list;
struct intel_vgpu *vgpu;
ktime_t sched_in_time;
ktime_t sched_out_time;
ktime_t sched_time;
ktime_t left_ts;
ktime_t allocated_ts;
struct vgpu_sched_ctl sched_ctl;
};
struct gvt_sched_data {
struct intel_gvt *gvt;
struct hrtimer timer;
unsigned long period;
struct list_head lru_runq_head;
};
static void vgpu_update_timeslice(struct intel_vgpu *pre_vgpu)
{
ktime_t delta_ts;
struct vgpu_sched_data *vgpu_data = pre_vgpu->sched_data;
delta_ts = vgpu_data->sched_out_time - vgpu_data->sched_in_time;
vgpu_data->sched_time += delta_ts;
vgpu_data->left_ts -= delta_ts;
}
#define GVT_TS_BALANCE_PERIOD_MS 100
#define GVT_TS_BALANCE_STAGE_NUM 10
static void gvt_balance_timeslice(struct gvt_sched_data *sched_data)
{
struct vgpu_sched_data *vgpu_data;
struct list_head *pos;
static uint64_t stage_check;
int stage = stage_check++ % GVT_TS_BALANCE_STAGE_NUM;
/* The timeslice accumulation reset at stage 0, which is
* allocated again without adding previous debt.
*/
if (stage == 0) {
int total_weight = 0;
ktime_t fair_timeslice;
list_for_each(pos, &sched_data->lru_runq_head) {
vgpu_data = container_of(pos, struct vgpu_sched_data, lru_list);
total_weight += vgpu_data->sched_ctl.weight;
}
list_for_each(pos, &sched_data->lru_runq_head) {
vgpu_data = container_of(pos, struct vgpu_sched_data, lru_list);
fair_timeslice = ms_to_ktime(GVT_TS_BALANCE_PERIOD_MS) *
vgpu_data->sched_ctl.weight /
total_weight;
vgpu_data->allocated_ts = fair_timeslice;
vgpu_data->left_ts = vgpu_data->allocated_ts;
}
} else {
list_for_each(pos, &sched_data->lru_runq_head) {
vgpu_data = container_of(pos, struct vgpu_sched_data, lru_list);
/* timeslice for next 100ms should add the left/debt
* slice of previous stages.
*/
vgpu_data->left_ts += vgpu_data->allocated_ts;
}
}
}
static void try_to_schedule_next_vgpu(struct intel_gvt *gvt)
{
struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
enum intel_engine_id i;
struct intel_engine_cs *engine;
struct vgpu_sched_data *vgpu_data;
ktime_t cur_time;
/* no target to schedule */
if (!scheduler->next_vgpu)
......@@ -77,6 +153,15 @@ static void try_to_schedule_next_vgpu(struct intel_gvt *gvt)
gvt_dbg_sched("switch to next vgpu %d\n",
scheduler->next_vgpu->id);
cur_time = ktime_get();
if (scheduler->current_vgpu) {
vgpu_data = scheduler->current_vgpu->sched_data;
vgpu_data->sched_out_time = cur_time;
vgpu_update_timeslice(scheduler->current_vgpu);
}
vgpu_data = scheduler->next_vgpu->sched_data;
vgpu_data->sched_in_time = cur_time;
/* switch current vgpu */
scheduler->current_vgpu = scheduler->next_vgpu;
scheduler->next_vgpu = NULL;
......@@ -88,62 +173,61 @@ static void try_to_schedule_next_vgpu(struct intel_gvt *gvt)
wake_up(&scheduler->waitq[i]);
}
struct tbs_vgpu_data {
struct list_head list;
struct intel_vgpu *vgpu;
/* put some per-vgpu sched stats here */
};
struct tbs_sched_data {
struct intel_gvt *gvt;
struct delayed_work work;
unsigned long period;
struct list_head runq_head;
};
#define GVT_DEFAULT_TIME_SLICE (msecs_to_jiffies(1))
static void tbs_sched_func(struct work_struct *work)
static struct intel_vgpu *find_busy_vgpu(struct gvt_sched_data *sched_data)
{
struct tbs_sched_data *sched_data = container_of(work,
struct tbs_sched_data, work.work);
struct tbs_vgpu_data *vgpu_data;
struct intel_gvt *gvt = sched_data->gvt;
struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
struct vgpu_sched_data *vgpu_data;
struct intel_vgpu *vgpu = NULL;
struct list_head *pos, *head;
mutex_lock(&gvt->lock);
/* no vgpu or has already had a target */
if (list_empty(&sched_data->runq_head) || scheduler->next_vgpu)
goto out;
if (scheduler->current_vgpu) {
vgpu_data = scheduler->current_vgpu->sched_data;
head = &vgpu_data->list;
} else {
head = &sched_data->runq_head;
}
struct list_head *head = &sched_data->lru_runq_head;
struct list_head *pos;
/* search a vgpu with pending workload */
list_for_each(pos, head) {
if (pos == &sched_data->runq_head)
continue;
vgpu_data = container_of(pos, struct tbs_vgpu_data, list);
vgpu_data = container_of(pos, struct vgpu_sched_data, lru_list);
if (!vgpu_has_pending_workload(vgpu_data->vgpu))
continue;
/* Return the vGPU only if it has time slice left */
if (vgpu_data->left_ts > 0) {
vgpu = vgpu_data->vgpu;
break;
}
}
return vgpu;
}
/* in nanosecond */
#define GVT_DEFAULT_TIME_SLICE 1000000
static void tbs_sched_func(struct gvt_sched_data *sched_data)
{
struct intel_gvt *gvt = sched_data->gvt;
struct intel_gvt_workload_scheduler *scheduler = &gvt->scheduler;
struct vgpu_sched_data *vgpu_data;
struct intel_vgpu *vgpu = NULL;
static uint64_t timer_check;
if (!(timer_check++ % GVT_TS_BALANCE_PERIOD_MS))
gvt_balance_timeslice(sched_data);
/* no active vgpu or has already had a target */
if (list_empty(&sched_data->lru_runq_head) || scheduler->next_vgpu)
goto out;
vgpu = find_busy_vgpu(sched_data);
if (vgpu) {
scheduler->next_vgpu = vgpu;
/* Move the last used vGPU to the tail of lru_list */
vgpu_data = vgpu->sched_data;
list_del_init(&vgpu_data->lru_list);
list_add_tail(&vgpu_data->lru_list,
&sched_data->lru_runq_head);
gvt_dbg_sched("pick next vgpu %d\n", vgpu->id);
} else {
scheduler->next_vgpu = gvt->idle_vgpu;
}
out:
if (scheduler->next_vgpu) {
......@@ -151,34 +235,49 @@ static void tbs_sched_func(struct work_struct *work)
scheduler->next_vgpu->id);
try_to_schedule_next_vgpu(gvt);
}
}
/*
* still have vgpu on runq
* or last schedule haven't finished due to running workload
*/
if (!list_empty(&sched_data->runq_head) || scheduler->next_vgpu)
schedule_delayed_work(&sched_data->work, sched_data->period);
void intel_gvt_schedule(struct intel_gvt *gvt)
{
struct gvt_sched_data *sched_data = gvt->scheduler.sched_data;
mutex_lock(&gvt->lock);
tbs_sched_func(sched_data);
mutex_unlock(&gvt->lock);
}
static enum hrtimer_restart tbs_timer_fn(struct hrtimer *timer_data)
{
struct gvt_sched_data *data;
data = container_of(timer_data, struct gvt_sched_data, timer);
intel_gvt_request_service(data->gvt, INTEL_GVT_REQUEST_SCHED);
hrtimer_add_expires_ns(&data->timer, data->period);
return HRTIMER_RESTART;
}
static int tbs_sched_init(struct intel_gvt *gvt)
{
struct intel_gvt_workload_scheduler *scheduler =
&gvt->scheduler;
struct tbs_sched_data *data;
struct gvt_sched_data *data;
data = kzalloc(sizeof(*data), GFP_KERNEL);
if (!data)
return -ENOMEM;
INIT_LIST_HEAD(&data->runq_head);
INIT_DELAYED_WORK(&data->work, tbs_sched_func);
INIT_LIST_HEAD(&data->lru_runq_head);
hrtimer_init(&data->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
data->timer.function = tbs_timer_fn;
data->period = GVT_DEFAULT_TIME_SLICE;
data->gvt = gvt;
scheduler->sched_data = data;
return 0;
}
......@@ -186,25 +285,28 @@ static void tbs_sched_clean(struct intel_gvt *gvt)
{
struct intel_gvt_workload_scheduler *scheduler =
&gvt->scheduler;
struct tbs_sched_data *data = scheduler->sched_data;
struct gvt_sched_data *data = scheduler->sched_data;
hrtimer_cancel(&data->timer);
cancel_delayed_work(&data->work);
kfree(data);
scheduler->sched_data = NULL;
}
static int tbs_sched_init_vgpu(struct intel_vgpu *vgpu)
{
struct tbs_vgpu_data *data;
struct vgpu_sched_data *data;
data = kzalloc(sizeof(*data), GFP_KERNEL);
if (!data)
return -ENOMEM;
data->sched_ctl.weight = vgpu->sched_ctl.weight;
data->vgpu = vgpu;
INIT_LIST_HEAD(&data->list);
INIT_LIST_HEAD(&data->lru_list);
vgpu->sched_data = data;
return 0;
}
......@@ -216,21 +318,24 @@ static void tbs_sched_clean_vgpu(struct intel_vgpu *vgpu)
static void tbs_sched_start_schedule(struct intel_vgpu *vgpu)
{
struct tbs_sched_data *sched_data = vgpu->gvt->scheduler.sched_data;
struct tbs_vgpu_data *vgpu_data = vgpu->sched_data;
struct gvt_sched_data *sched_data = vgpu->gvt->scheduler.sched_data;
struct vgpu_sched_data *vgpu_data = vgpu->sched_data;
if (!list_empty(&vgpu_data->list))
if (!list_empty(&vgpu_data->lru_list))
return;
list_add_tail(&vgpu_data->list, &sched_data->runq_head);
schedule_delayed_work(&sched_data->work, 0);
list_add_tail(&vgpu_data->lru_list, &sched_data->lru_runq_head);
if (!hrtimer_active(&sched_data->timer))
hrtimer_start(&sched_data->timer, ktime_add_ns(ktime_get(),
sched_data->period), HRTIMER_MODE_ABS);
}
static void tbs_sched_stop_schedule(struct intel_vgpu *vgpu)
{
struct tbs_vgpu_data *vgpu_data = vgpu->sched_data;
struct vgpu_sched_data *vgpu_data = vgpu->sched_data;
list_del_init(&vgpu_data->list);
list_del_init(&vgpu_data->lru_list);
}
static struct intel_gvt_sched_policy_ops tbs_schedule_ops = {
......
......@@ -43,6 +43,8 @@ struct intel_gvt_sched_policy_ops {
void (*stop_schedule)(struct intel_vgpu *vgpu);
};
void intel_gvt_schedule(struct intel_gvt *gvt);
int intel_gvt_init_sched_policy(struct intel_gvt *gvt);
void intel_gvt_clean_sched_policy(struct intel_gvt *gvt);
......
......@@ -448,7 +448,8 @@ static int workload_thread(void *priv)
struct intel_vgpu_workload *workload = NULL;
struct intel_vgpu *vgpu = NULL;
int ret;
bool need_force_wake = IS_SKYLAKE(gvt->dev_priv);
bool need_force_wake = IS_SKYLAKE(gvt->dev_priv)
|| IS_KABYLAKE(gvt->dev_priv);
DEFINE_WAIT_FUNC(wait, woken_wake_function);
kfree(p);
......
......@@ -67,7 +67,6 @@ struct shadow_per_ctx {
};
struct intel_shadow_wa_ctx {
struct intel_vgpu_workload *workload;
struct shadow_indirect_ctx indirect_ctx;
struct shadow_per_ctx per_ctx;
......
......@@ -64,18 +64,28 @@ void populate_pvinfo_page(struct intel_vgpu *vgpu)
WARN_ON(sizeof(struct vgt_if) != VGT_PVINFO_SIZE);
}
#define VGPU_MAX_WEIGHT 16
#define VGPU_WEIGHT(vgpu_num) \
(VGPU_MAX_WEIGHT / (vgpu_num))
static struct {
unsigned int low_mm;
unsigned int high_mm;
unsigned int fence;
/* A vGPU with a weight of 8 will get twice as much GPU as a vGPU
* with a weight of 4 on a contended host, different vGPU type has
* different weight set. Legal weights range from 1 to 16.
*/
unsigned int weight;
enum intel_vgpu_edid edid;
char *name;
} vgpu_types[] = {
/* Fixed vGPU type table */
{ MB_TO_BYTES(64), MB_TO_BYTES(512), 4, GVT_EDID_1024_768, "8" },
{ MB_TO_BYTES(128), MB_TO_BYTES(512), 4, GVT_EDID_1920_1200, "4" },
{ MB_TO_BYTES(256), MB_TO_BYTES(1024), 4, GVT_EDID_1920_1200, "2" },
{ MB_TO_BYTES(512), MB_TO_BYTES(2048), 4, GVT_EDID_1920_1200, "1" },
{ MB_TO_BYTES(64), MB_TO_BYTES(384), 4, VGPU_WEIGHT(8), GVT_EDID_1024_768, "8" },
{ MB_TO_BYTES(128), MB_TO_BYTES(512), 4, VGPU_WEIGHT(4), GVT_EDID_1920_1200, "4" },
{ MB_TO_BYTES(256), MB_TO_BYTES(1024), 4, VGPU_WEIGHT(2), GVT_EDID_1920_1200, "2" },
{ MB_TO_BYTES(512), MB_TO_BYTES(2048), 4, VGPU_WEIGHT(1), GVT_EDID_1920_1200, "1" },
};
/**
......@@ -120,6 +130,12 @@ int intel_gvt_init_vgpu_types(struct intel_gvt *gvt)
gvt->types[i].low_gm_size = vgpu_types[i].low_mm;
gvt->types[i].high_gm_size = vgpu_types[i].high_mm;
gvt->types[i].fence = vgpu_types[i].fence;
if (vgpu_types[i].weight < 1 ||
vgpu_types[i].weight > VGPU_MAX_WEIGHT)
return -EINVAL;
gvt->types[i].weight = vgpu_types[i].weight;
gvt->types[i].resolution = vgpu_types[i].edid;
gvt->types[i].avail_instance = min(low_avail / vgpu_types[i].low_mm,
high_avail / vgpu_types[i].high_mm);
......@@ -131,11 +147,12 @@ int intel_gvt_init_vgpu_types(struct intel_gvt *gvt)
sprintf(gvt->types[i].name, "GVTg_V5_%s",
vgpu_types[i].name);
gvt_dbg_core("type[%d]: %s avail %u low %u high %u fence %u res %s\n",
gvt_dbg_core("type[%d]: %s avail %u low %u high %u fence %u weight %u res %s\n",
i, gvt->types[i].name,
gvt->types[i].avail_instance,
gvt->types[i].low_gm_size,
gvt->types[i].high_gm_size, gvt->types[i].fence,
gvt->types[i].weight,
vgpu_edid_str(gvt->types[i].resolution));
}
......@@ -216,6 +233,59 @@ void intel_gvt_destroy_vgpu(struct intel_vgpu *vgpu)
mutex_unlock(&gvt->lock);
}
#define IDLE_VGPU_IDR 0
/**
* intel_gvt_create_idle_vgpu - create an idle virtual GPU
* @gvt: GVT device
*
* This function is called when user wants to create an idle virtual GPU.
*
* Returns:
* pointer to intel_vgpu, error pointer if failed.
*/
struct intel_vgpu *intel_gvt_create_idle_vgpu(struct intel_gvt *gvt)
{
struct intel_vgpu *vgpu;
enum intel_engine_id i;
int ret;
vgpu = vzalloc(sizeof(*vgpu));
if (!vgpu)
return ERR_PTR(-ENOMEM);
vgpu->id = IDLE_VGPU_IDR;
vgpu->gvt = gvt;
for (i = 0; i < I915_NUM_ENGINES; i++)
INIT_LIST_HEAD(&vgpu->workload_q_head[i]);
ret = intel_vgpu_init_sched_policy(vgpu);
if (ret)
goto out_free_vgpu;
vgpu->active = false;
return vgpu;
out_free_vgpu:
vfree(vgpu);
return ERR_PTR(ret);
}
/**
* intel_gvt_destroy_vgpu - destroy an idle virtual GPU
* @vgpu: virtual GPU
*
* This function is called when user wants to destroy an idle virtual GPU.
*
*/
void intel_gvt_destroy_idle_vgpu(struct intel_vgpu *vgpu)
{
intel_vgpu_clean_sched_policy(vgpu);
vfree(vgpu);
}
static struct intel_vgpu *__intel_gvt_create_vgpu(struct intel_gvt *gvt,
struct intel_vgpu_creation_params *param)
{
......@@ -232,13 +302,15 @@ static struct intel_vgpu *__intel_gvt_create_vgpu(struct intel_gvt *gvt,
mutex_lock(&gvt->lock);
ret = idr_alloc(&gvt->vgpu_idr, vgpu, 1, GVT_MAX_VGPU, GFP_KERNEL);
ret = idr_alloc(&gvt->vgpu_idr, vgpu, IDLE_VGPU_IDR + 1, GVT_MAX_VGPU,
GFP_KERNEL);
if (ret < 0)
goto out_free_vgpu;
vgpu->id = ret;
vgpu->handle = param->handle;
vgpu->gvt = gvt;
vgpu->sched_ctl.weight = param->weight;
bitmap_zero(vgpu->tlb_handle_pending, I915_NUM_ENGINES);
intel_vgpu_init_cfg_space(vgpu, param->primary);
......@@ -325,6 +397,7 @@ struct intel_vgpu *intel_gvt_create_vgpu(struct intel_gvt *gvt,
param.low_gm_sz = type->low_gm_size;
param.high_gm_sz = type->high_gm_size;
param.fence_sz = type->fence;
param.weight = type->weight;
param.resolution = type->resolution;
/* XXX current param based on MB */
......
......@@ -45,6 +45,8 @@ static bool is_supported_device(struct drm_i915_private *dev_priv)
return true;
if (IS_SKYLAKE(dev_priv))
return true;
if (IS_KABYLAKE(dev_priv) && INTEL_DEVID(dev_priv) == 0x591D)
return true;
return false;
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment