| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20180422135114.320584004@linuxfoundation.org>
Date: Sun, 22 Apr 2018 15:53:34 +0200
From: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To: linux-kernel@...r.kernel.org
Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
stable@...r.kernel.org, Kevin Tian <kevin.tian@...el.com>,
Zhenyu Wang <zhenyuw@...ux.intel.com>,
Weinan Li <weinan.z.li@...el.com>,
Changbin Du <changbin.du@...el.com>
Subject: [PATCH 4.16 194/196] drm/i915/gvt: init mmio by lri command in vgpu inhibit context
4.16-stable review patch. If anyone has any objections, please let me know.
------------------
From: Weinan Li <weinan.z.li@...el.com>
commit cd7e61b93d068a80bfe6cb55bf00f17332d831a1 upstream.
There is one issue relates to Coarse Power Gating(CPG) on KBL NUC in GVT-g,
vgpu can't get the correct default context by updating the registers before
inhibit context submission. It always get back the hardware default value
unless the inhibit context submission happened before the 1st time
forcewake put. With this wrong default context, vgpu will run with
incorrect state and meet unknown issues.
The solution is initialize these mmios by adding lri command in ring buffer
of the inhibit context, then gpu hardware has no chance to go down RC6 when
lri commands are right being executed, and then vgpu can get correct
default context for further use.
v3:
- fix code fault, use 'for' to loop through mmio render list(Zhenyu)
v4:
- save the count of engine mmio need to be restored for inhibit context and
refine some comments. (Kevin)
v5:
- code rebase
Cc: Kevin Tian <kevin.tian@...el.com>
Cc: Zhenyu Wang <zhenyuw@...ux.intel.com>
Signed-off-by: Weinan Li <weinan.z.li@...el.com>
Signed-off-by: Zhenyu Wang <zhenyuw@...ux.intel.com>
Signed-off-by: Changbin Du <changbin.du@...el.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
---
drivers/gpu/drm/i915/gvt/gvt.h | 5
drivers/gpu/drm/i915/gvt/mmio_context.c | 210 +++++++++++++++++++++++++++++---
drivers/gpu/drm/i915/gvt/mmio_context.h | 5
drivers/gpu/drm/i915/gvt/scheduler.c | 5
4 files changed, 205 insertions(+), 20 deletions(-)
--- a/drivers/gpu/drm/i915/gvt/gvt.h
+++ b/drivers/gpu/drm/i915/gvt/gvt.h
@@ -308,7 +308,10 @@ struct intel_gvt {
wait_queue_head_t service_thread_wq;
unsigned long service_request;
- struct engine_mmio *engine_mmio_list;
+ struct {
+ struct engine_mmio *mmio;
+ int ctx_mmio_count[I915_NUM_ENGINES];
+ } engine_mmio_list;
struct dentry *debugfs_root;
};
--- a/drivers/gpu/drm/i915/gvt/mmio_context.c
+++ b/drivers/gpu/drm/i915/gvt/mmio_context.c
@@ -50,6 +50,8 @@
#define RING_GFX_MODE(base) _MMIO((base) + 0x29c)
#define VF_GUARDBAND _MMIO(0x83a4)
+#define GEN9_MOCS_SIZE 64
+
/* Raw offset is appened to each line for convenience. */
static struct engine_mmio gen8_engine_mmio_list[] __cacheline_aligned = {
{RCS, GFX_MODE_GEN7, 0xffff, false}, /* 0x229c */
@@ -152,8 +154,8 @@ static struct engine_mmio gen9_engine_mm
static struct {
bool initialized;
- u32 control_table[I915_NUM_ENGINES][64];
- u32 l3cc_table[32];
+ u32 control_table[I915_NUM_ENGINES][GEN9_MOCS_SIZE];
+ u32 l3cc_table[GEN9_MOCS_SIZE / 2];
} gen9_render_mocs;
static void load_render_mocs(struct drm_i915_private *dev_priv)
@@ -170,7 +172,7 @@ static void load_render_mocs(struct drm_
for (ring_id = 0; ring_id < ARRAY_SIZE(regs); ring_id++) {
offset.reg = regs[ring_id];
- for (i = 0; i < 64; i++) {
+ for (i = 0; i < GEN9_MOCS_SIZE; i++) {
gen9_render_mocs.control_table[ring_id][i] =
I915_READ_FW(offset);
offset.reg += 4;
@@ -178,7 +180,7 @@ static void load_render_mocs(struct drm_
}
offset.reg = 0xb020;
- for (i = 0; i < 32; i++) {
+ for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
gen9_render_mocs.l3cc_table[i] =
I915_READ_FW(offset);
offset.reg += 4;
@@ -186,6 +188,153 @@ static void load_render_mocs(struct drm_
gen9_render_mocs.initialized = true;
}
+static int
+restore_context_mmio_for_inhibit(struct intel_vgpu *vgpu,
+ struct drm_i915_gem_request *req)
+{
+ u32 *cs;
+ int ret;
+ struct engine_mmio *mmio;
+ struct intel_gvt *gvt = vgpu->gvt;
+ int ring_id = req->engine->id;
+ int count = gvt->engine_mmio_list.ctx_mmio_count[ring_id];
+
+ if (count == 0)
+ return 0;
+
+ ret = req->engine->emit_flush(req, EMIT_BARRIER);
+ if (ret)
+ return ret;
+
+ cs = intel_ring_begin(req, count * 2 + 2);
+ if (IS_ERR(cs))
+ return PTR_ERR(cs);
+
+ *cs++ = MI_LOAD_REGISTER_IMM(count);
+ for (mmio = gvt->engine_mmio_list.mmio;
+ i915_mmio_reg_valid(mmio->reg); mmio++) {
+ if (mmio->ring_id != ring_id ||
+ !mmio->in_context)
+ continue;
+
+ *cs++ = i915_mmio_reg_offset(mmio->reg);
+ *cs++ = vgpu_vreg_t(vgpu, mmio->reg) |
+ (mmio->mask << 16);
+ gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
+ *(cs-2), *(cs-1), vgpu->id, ring_id);
+ }
+
+ *cs++ = MI_NOOP;
+ intel_ring_advance(req, cs);
+
+ ret = req->engine->emit_flush(req, EMIT_BARRIER);
+ if (ret)
+ return ret;
+
+ return 0;
+}
+
+static int
+restore_render_mocs_control_for_inhibit(struct intel_vgpu *vgpu,
+ struct drm_i915_gem_request *req)
+{
+ unsigned int index;
+ u32 *cs;
+
+ cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE + 2);
+ if (IS_ERR(cs))
+ return PTR_ERR(cs);
+
+ *cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE);
+
+ for (index = 0; index < GEN9_MOCS_SIZE; index++) {
+ *cs++ = i915_mmio_reg_offset(GEN9_GFX_MOCS(index));
+ *cs++ = vgpu_vreg_t(vgpu, GEN9_GFX_MOCS(index));
+ gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
+ *(cs-2), *(cs-1), vgpu->id, req->engine->id);
+
+ }
+
+ *cs++ = MI_NOOP;
+ intel_ring_advance(req, cs);
+
+ return 0;
+}
+
+static int
+restore_render_mocs_l3cc_for_inhibit(struct intel_vgpu *vgpu,
+ struct drm_i915_gem_request *req)
+{
+ unsigned int index;
+ u32 *cs;
+
+ cs = intel_ring_begin(req, 2 * GEN9_MOCS_SIZE / 2 + 2);
+ if (IS_ERR(cs))
+ return PTR_ERR(cs);
+
+ *cs++ = MI_LOAD_REGISTER_IMM(GEN9_MOCS_SIZE / 2);
+
+ for (index = 0; index < GEN9_MOCS_SIZE / 2; index++) {
+ *cs++ = i915_mmio_reg_offset(GEN9_LNCFCMOCS(index));
+ *cs++ = vgpu_vreg_t(vgpu, GEN9_LNCFCMOCS(index));
+ gvt_dbg_core("add lri reg pair 0x%x:0x%x in inhibit ctx, vgpu:%d, rind_id:%d\n",
+ *(cs-2), *(cs-1), vgpu->id, req->engine->id);
+
+ }
+
+ *cs++ = MI_NOOP;
+ intel_ring_advance(req, cs);
+
+ return 0;
+}
+
+/*
+ * Use lri command to initialize the mmio which is in context state image for
+ * inhibit context, it contains tracked engine mmio, render_mocs and
+ * render_mocs_l3cc.
+ */
+int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
+ struct drm_i915_gem_request *req)
+{
+ int ret;
+ u32 *cs;
+
+ cs = intel_ring_begin(req, 2);
+ if (IS_ERR(cs))
+ return PTR_ERR(cs);
+
+ *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
+ *cs++ = MI_NOOP;
+ intel_ring_advance(req, cs);
+
+ ret = restore_context_mmio_for_inhibit(vgpu, req);
+ if (ret)
+ goto out;
+
+ /* no MOCS register in context except render engine */
+ if (req->engine->id != RCS)
+ goto out;
+
+ ret = restore_render_mocs_control_for_inhibit(vgpu, req);
+ if (ret)
+ goto out;
+
+ ret = restore_render_mocs_l3cc_for_inhibit(vgpu, req);
+ if (ret)
+ goto out;
+
+out:
+ cs = intel_ring_begin(req, 2);
+ if (IS_ERR(cs))
+ return PTR_ERR(cs);
+
+ *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+ *cs++ = MI_NOOP;
+ intel_ring_advance(req, cs);
+
+ return ret;
+}
+
static void handle_tlb_pending_event(struct intel_vgpu *vgpu, int ring_id)
{
struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
@@ -252,11 +401,14 @@ static void switch_mocs(struct intel_vgp
if (WARN_ON(ring_id >= ARRAY_SIZE(regs)))
return;
+ if (IS_KABYLAKE(dev_priv) && ring_id == RCS)
+ return;
+
if (!pre && !gen9_render_mocs.initialized)
load_render_mocs(dev_priv);
offset.reg = regs[ring_id];
- for (i = 0; i < 64; i++) {
+ for (i = 0; i < GEN9_MOCS_SIZE; i++) {
if (pre)
old_v = vgpu_vreg_t(pre, offset);
else
@@ -274,7 +426,7 @@ static void switch_mocs(struct intel_vgp
if (ring_id == RCS) {
l3_offset.reg = 0xb020;
- for (i = 0; i < 32; i++) {
+ for (i = 0; i < GEN9_MOCS_SIZE / 2; i++) {
if (pre)
old_v = vgpu_vreg_t(pre, l3_offset);
else
@@ -294,6 +446,16 @@ static void switch_mocs(struct intel_vgp
#define CTX_CONTEXT_CONTROL_VAL 0x03
+bool is_inhibit_context(struct i915_gem_context *ctx, int ring_id)
+{
+ u32 *reg_state = ctx->engine[ring_id].lrc_reg_state;
+ u32 inhibit_mask =
+ _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
+
+ return inhibit_mask ==
+ (reg_state[CTX_CONTEXT_CONTROL_VAL] & inhibit_mask);
+}
+
/* Switch ring mmio values (context). */
static void switch_mmio(struct intel_vgpu *pre,
struct intel_vgpu *next,
@@ -301,9 +463,6 @@ static void switch_mmio(struct intel_vgp
{
struct drm_i915_private *dev_priv;
struct intel_vgpu_submission *s;
- u32 *reg_state, ctx_ctrl;
- u32 inhibit_mask =
- _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT);
struct engine_mmio *mmio;
u32 old_v, new_v;
@@ -311,10 +470,18 @@ static void switch_mmio(struct intel_vgp
if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
switch_mocs(pre, next, ring_id);
- for (mmio = dev_priv->gvt->engine_mmio_list;
+ for (mmio = dev_priv->gvt->engine_mmio_list.mmio;
i915_mmio_reg_valid(mmio->reg); mmio++) {
if (mmio->ring_id != ring_id)
continue;
+ /*
+ * No need to do save or restore of the mmio which is in context
+ * state image on kabylake, it's initialized by lri command and
+ * save or restore with context together.
+ */
+ if (IS_KABYLAKE(dev_priv) && mmio->in_context)
+ continue;
+
// save
if (pre) {
vgpu_vreg_t(pre, mmio->reg) = I915_READ_FW(mmio->reg);
@@ -328,16 +495,13 @@ static void switch_mmio(struct intel_vgp
// restore
if (next) {
s = &next->submission;
- reg_state =
- s->shadow_ctx->engine[ring_id].lrc_reg_state;
- ctx_ctrl = reg_state[CTX_CONTEXT_CONTROL_VAL];
/*
- * if it is an inhibit context, load in_context mmio
- * into HW by mmio write. If it is not, skip this mmio
- * write.
+ * No need to restore the mmio which is in context state
+ * image if it's not inhibit context, it will restore
+ * itself.
*/
if (mmio->in_context &&
- (ctx_ctrl & inhibit_mask) != inhibit_mask)
+ !is_inhibit_context(s->shadow_ctx, ring_id))
continue;
if (mmio->mask)
@@ -408,8 +572,16 @@ void intel_gvt_switch_mmio(struct intel_
*/
void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt)
{
+ struct engine_mmio *mmio;
+
if (IS_SKYLAKE(gvt->dev_priv) || IS_KABYLAKE(gvt->dev_priv))
- gvt->engine_mmio_list = gen9_engine_mmio_list;
+ gvt->engine_mmio_list.mmio = gen9_engine_mmio_list;
else
- gvt->engine_mmio_list = gen8_engine_mmio_list;
+ gvt->engine_mmio_list.mmio = gen8_engine_mmio_list;
+
+ for (mmio = gvt->engine_mmio_list.mmio;
+ i915_mmio_reg_valid(mmio->reg); mmio++) {
+ if (mmio->in_context)
+ gvt->engine_mmio_list.ctx_mmio_count[mmio->ring_id]++;
+ }
}
--- a/drivers/gpu/drm/i915/gvt/mmio_context.h
+++ b/drivers/gpu/drm/i915/gvt/mmio_context.h
@@ -49,4 +49,9 @@ void intel_gvt_switch_mmio(struct intel_
void intel_gvt_init_engine_mmio_context(struct intel_gvt *gvt);
+bool is_inhibit_context(struct i915_gem_context *ctx, int ring_id);
+
+int intel_vgpu_restore_inhibit_context(struct intel_vgpu *vgpu,
+ struct drm_i915_gem_request *req);
+
#endif
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/drivers/gpu/drm/i915/gvt/scheduler.c
@@ -275,6 +275,11 @@ static int copy_workload_to_ring_buffer(
struct intel_vgpu *vgpu = workload->vgpu;
void *shadow_ring_buffer_va;
u32 *cs;
+ struct drm_i915_gem_request *req = workload->req;
+
+ if (IS_KABYLAKE(req->i915) &&
+ is_inhibit_context(req->ctx, req->engine->id))
+ intel_vgpu_restore_inhibit_context(vgpu, req);
/* allocate shadow ring buffer */
cs = intel_ring_begin(workload->req, workload->rb_len / sizeof(u32));
Powered by blists - more mailing lists