@@ -122,6 +122,11 @@ enum slpc_param_id {
SLPC_MAX_PARAM = 32,
};
+#define SLPC_OPTIMIZED_STRATEGIES_COMPUTE REG_BIT(0)
+#define SLPC_OPTIMIZED_STRATEGIES_ASYNC_FLIP REG_BIT(1)
+#define SLPC_OPTIMIZED_STRATEGIES_MEDIA REG_BIT(2)
+#define SLPC_OPTIMIZED_STRATEGIES_VSYNC_FLIP REG_BIT(3)
+
enum slpc_media_ratio_mode {
SLPC_MEDIA_RATIO_MODE_DYNAMIC_CONTROL = 0,
SLPC_MEDIA_RATIO_MODE_FIXED_ONE_TO_ONE = 1,
@@ -207,6 +212,60 @@ struct slpc_shared_data {
u8 reserved_mode_definition[4096];
} __packed;
+#define SLPC_MAX_PIPES 8
+#define SLPC_MAX_PLANES_PER_PIPE 8
+
+struct slpc_display_global_info {
+ u32 version:8;
+ u32 num_pipes:4;
+ u32 num_planes_per_pipe:4;
+ u32 reserved_1:16;
+ u32 refresh_count:16;
+ u32 vblank_count:16;
+ u32 flip_count:16;
+ u32 reserved_2:16;
+ u32 reserved_3[13];
+} __packed;
+
+struct slpc_display_refresh_info {
+ u32 refresh_interval:16;
+ u32 is_variable:1;
+ u32 reserved:15;
+} __packed;
+
+/*
+ * The host must update each 32-bit part with a single atomic write so
+ * that SLPC will read the contained bit fields together. The host must
+ * update the two parts in order - total flip count and timestamp first,
+ * vsync and async flip counts second.
+ * Hence, these items are not defined with individual bitfields.
+ */
+#define SLPC_FLIP_P1_LAST REG_GENMASK(31, 7)
+#define SLPC_FLIP_P1_TOTAL_COUNT REG_GENMASK(6, 0)
+#define SLPC_FLIP_P2_ASYNC_COUNT REG_GENMASK(31, 16)
+#define SLPC_FLIP_P2_VSYNC_COUNT REG_GENMASK(15, 0)
+
+struct slpc_display_flip_metrics {
+ u32 part1;
+ u32 part2;
+} __packed;
+
+/*
+ * The host must update this 32-bit structure with a single atomic write
+ * so that SLPC will read the count and timestamp together.
+ * Hence, this item is not defined with individual bitfields.
+ */
+#define SLPC_VBLANK_LAST REG_GENMASK(31, 7)
+#define SLPC_VBLANK_COUNT REG_GENMASK(6, 0)
+
+struct slpc_display_data {
+ struct slpc_display_global_info global_info;
+ struct slpc_display_refresh_info refresh_info[SLPC_MAX_PIPES];
+ u32 vblank_metrics[SLPC_MAX_PIPES];
+ struct slpc_display_flip_metrics
+ flip_metrics[SLPC_MAX_PIPES][SLPC_MAX_PLANES_PER_PIPE];
+} __packed;
+
/**
* DOC: SLPC H2G MESSAGE FORMAT
*
@@ -14,6 +14,7 @@
#include "xe_gt.h"
#include "xe_gt_mcr.h"
#include "xe_gt_topology.h"
+#include "xe_guc_pc.h"
#include "xe_hw_engine.h"
#include "xe_lrc.h"
#include "xe_macros.h"
@@ -103,6 +104,15 @@ static int ggtt(struct seq_file *m, void *data)
return xe_ggtt_dump(gt_to_tile(gt)->mem.ggtt, &p);
}
+static int guc_pc(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_guc_pc_print(>->uc.guc.pc, &p);
+ return 0;
+}
+
static int register_save_restore(struct seq_file *m, void *data)
{
struct xe_gt *gt = node_to_gt(m->private);
@@ -197,6 +207,7 @@ static const struct drm_info_list debugfs_list[] = {
{"topology", topology, 0},
{"steering", steering, 0},
{"ggtt", ggtt, 0},
+ {"guc_pc", guc_pc, 0},
{"register-save-restore", register_save_restore, 0},
{"workarounds", workarounds, 0},
{"pat", pat, 0},
@@ -11,6 +11,8 @@
#include "abi/guc_actions_abi.h"
#include "abi/guc_actions_slpc_abi.h"
+#include "intel_display_guc_metrics.h"
+#include "intel_display_guc_metrics_types.h"
#include "regs/xe_gt_regs.h"
#include "regs/xe_regs.h"
#include "xe_bo.h"
@@ -40,6 +42,10 @@
#define GT_FREQUENCY_MULTIPLIER 50
#define GT_FREQUENCY_SCALER 3
+#define MCHBAR_BCLK_COUNT XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x5984)
+#define MTL_BCLK_COUNT XE_REG(0xc28)
+#define TIMESTAMP_MASK REG_GENMASK(30, 6)
+
/**
* DOC: GuC Power Conservation (PC)
*
@@ -88,19 +94,36 @@ pc_to_gt(struct xe_guc_pc *pc)
}
static struct iosys_map *
-pc_to_maps(struct xe_guc_pc *pc)
+pc_to_maps(struct xe_guc_pc *pc, bool display)
{
- return &pc->bo->vmap;
+ if (display)
+ return &pc->display_bo->vmap;
+ else
+ return &pc->bo->vmap;
+}
+
+static struct slpc_display_data *
+pc_to_display_data(struct xe_guc_pc *pc)
+{
+ return pc->display_bo->vmap.vaddr;
}
#define slpc_shared_data_read(pc_, field_) \
- xe_map_rd_field(pc_to_xe(pc_), pc_to_maps(pc_), 0, \
+ xe_map_rd_field(pc_to_xe(pc_), pc_to_maps(pc_, false), 0, \
struct slpc_shared_data, field_)
#define slpc_shared_data_write(pc_, field_, val_) \
- xe_map_wr_field(pc_to_xe(pc_), pc_to_maps(pc_), 0, \
+ xe_map_wr_field(pc_to_xe(pc_), pc_to_maps(pc_, false), 0, \
struct slpc_shared_data, field_, val_)
+#define slpc_display_data_read(pc_, field_) \
+ xe_map_rd_field(pc_to_xe(pc_), pc_to_maps(pc_, true), 0, \
+ struct slpc_display_data, field_)
+
+#define slpc_display_data_write(pc_, field_, val_) \
+ xe_map_wr_field(pc_to_xe(pc_), pc_to_maps(pc_, true), 0, \
+ struct slpc_display_data, field_, val_)
+
#define SLPC_EVENT(id, count) \
(FIELD_PREP(HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ID, id) | \
FIELD_PREP(HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ARGC, count))
@@ -225,6 +248,144 @@ static int pc_action_setup_gucrc(struct xe_guc_pc *pc, u32 mode)
return ret;
}
+static void slpc_display_data_init(struct slpc_display_data *data, int version,
+ int num_pipes, int num_planes_per_pipe)
+{
+ data->global_info.version = version;
+ data->global_info.num_pipes = num_pipes;
+ data->global_info.num_planes_per_pipe = num_planes_per_pipe;
+}
+
+static void slpc_refresh_info(struct slpc_display_data *data, int pipe,
+ u32 refresh_interval, bool vrr_enabled)
+{
+ data->refresh_info[pipe].refresh_interval = refresh_interval;
+ data->refresh_info[pipe].is_variable = vrr_enabled;
+ data->global_info.refresh_count += 1;
+}
+
+static void slpc_vblank(struct slpc_display_data *data, int pipe,
+ u32 timestamp)
+{
+ u32 vblank;
+
+ vblank = data->vblank_metrics[pipe];
+
+ vblank = REG_FIELD_GET(SLPC_VBLANK_COUNT, vblank);
+ vblank = REG_FIELD_PREP(SLPC_VBLANK_COUNT, vblank + 1);
+ vblank |= REG_FIELD_PREP(SLPC_VBLANK_LAST, timestamp);
+
+ data->vblank_metrics[pipe] = vblank;
+ data->global_info.vblank_count += 1;
+}
+
+static void slpc_flip(struct slpc_display_data *data, int pipe, int plane,
+ bool async_flip, u32 timestamp)
+{
+ u32 part1, part2, count;
+
+ part1 = data->flip_metrics[pipe][plane].part1;
+ part2 = data->flip_metrics[pipe][plane].part2;
+
+ part1 = REG_FIELD_GET(SLPC_FLIP_P1_TOTAL_COUNT, part1);
+ part1 = REG_FIELD_PREP(SLPC_FLIP_P1_TOTAL_COUNT, part1 + 1);
+ part1 |= REG_FIELD_PREP(SLPC_FLIP_P1_LAST, timestamp);
+
+ if (async_flip) {
+ count = REG_FIELD_GET(SLPC_FLIP_P2_ASYNC_COUNT, part2);
+ part2 &= ~SLPC_FLIP_P2_ASYNC_COUNT;
+ part2 |= REG_FIELD_PREP(SLPC_FLIP_P2_ASYNC_COUNT, count + 1);
+ } else {
+ count = REG_FIELD_GET(SLPC_FLIP_P2_VSYNC_COUNT, part2);
+ part2 &= ~SLPC_FLIP_P2_VSYNC_COUNT;
+ part2 |= REG_FIELD_PREP(SLPC_FLIP_P2_VSYNC_COUNT, count + 1);
+ }
+
+ data->flip_metrics[pipe][plane].part1 = part1;
+ data->flip_metrics[pipe][plane].part2 = part2;
+
+ data->global_info.flip_count += 1;
+}
+
+static void xe_guc_pc_refresh_info_update(void *gfx_device, int pipe,
+ u32 refresh_rate, bool vrr_enabled)
+{
+ struct xe_device *xe = gfx_device;
+ struct xe_gt *gt;
+ int i;
+
+ if (pipe > SLPC_MAX_PIPES) {
+ drm_err(&xe->drm, "GuC PC Max display pipe exceeded\n");
+ return;
+ }
+
+ for_each_gt(gt, xe, i)
+ slpc_refresh_info(pc_to_display_data(>->uc.guc.pc),
+ pipe, refresh_rate, vrr_enabled);
+}
+
+static u32 bclk_read_timestamp(struct xe_gt *gt)
+{
+ u32 timestamp;
+
+ if (gt_to_xe(gt)->info.platform == XE_METEORLAKE)
+ timestamp = xe_mmio_read32(gt, MTL_BCLK_COUNT);
+ else
+ timestamp = xe_mmio_read32(gt, MCHBAR_BCLK_COUNT);
+
+ return REG_FIELD_GET(TIMESTAMP_MASK, timestamp);
+}
+
+static void xe_guc_pc_vblank_update(void *gfx_device, int pipe)
+{
+ struct xe_device *xe = gfx_device;
+ struct xe_gt *gt;
+ u32 timestamp;
+ int i;
+
+ if (pipe > SLPC_MAX_PIPES) {
+ drm_err(&xe->drm, "GuC PC Max display pipe exceeded\n");
+ return;
+ }
+
+ for_each_gt(gt, xe, i) {
+ timestamp = bclk_read_timestamp(gt);
+ slpc_vblank(pc_to_display_data(>->uc.guc.pc),
+ pipe, timestamp);
+ }
+}
+
+static void xe_guc_pc_flip_update(void *gfx_device, int pipe, int plane,
+ bool async_flip)
+{
+ struct xe_device *xe = gfx_device;
+ struct xe_gt *gt;
+ u32 timestamp;
+ int i;
+
+ if (pipe > SLPC_MAX_PIPES) {
+ drm_err(&xe->drm, "GuC PC Max display pipe exceeded\n");
+ return;
+ }
+
+ if (plane > SLPC_MAX_PLANES_PER_PIPE) {
+ drm_err(&xe->drm, "GuC PC Max display planes exceeded\n");
+ return;
+ }
+
+ for_each_gt(gt, xe, i) {
+ timestamp = bclk_read_timestamp(gt);
+ slpc_flip(pc_to_display_data(>->uc.guc.pc),
+ pipe, plane, async_flip, timestamp);
+ }
+}
+
+static struct intel_display_guc_metrics guc_metrics = {
+ .refresh_info_update = xe_guc_pc_refresh_info_update,
+ .vblank_update = xe_guc_pc_vblank_update,
+ .flip_update = xe_guc_pc_flip_update,
+};
+
static u32 decode_freq(u32 raw)
{
return DIV_ROUND_CLOSEST(raw * GT_FREQUENCY_MULTIPLIER,
@@ -857,6 +1018,17 @@ static int pc_init_freqs(struct xe_guc_pc *pc)
return ret;
}
+static void slpc_display_metrics_start(struct xe_guc_pc *pc)
+{
+ struct xe_device *xe = pc_to_xe(pc);
+
+ slpc_display_data_init(pc_to_display_data(pc), 1, SLPC_MAX_PIPES,
+ SLPC_MAX_PLANES_PER_PIPE);
+ slpc_shared_data_write(pc, header.display_data_addr,
+ xe_bo_ggtt_addr(pc->display_bo));
+ intel_display_guc_metrics_init(xe, &xe->display, &guc_metrics);
+}
+
/**
* xe_guc_pc_start - Start GuC's Power Conservation component
* @pc: Xe_GuC_PC instance
@@ -888,7 +1060,9 @@ int xe_guc_pc_start(struct xe_guc_pc *pc)
}
memset(pc->bo->vmap.vaddr, 0, size);
+ memset(pc->display_bo->vmap.vaddr, 0, size);
slpc_shared_data_write(pc, header.size, size);
+ slpc_display_metrics_start(pc);
ret = pc_action_reset(pc);
if (ret)
@@ -900,6 +1074,12 @@ int xe_guc_pc_start(struct xe_guc_pc *pc)
goto out;
}
+ if (pc_action_set_param(pc, SLPC_PARAM_STRATEGIES,
+ SLPC_OPTIMIZED_STRATEGIES_VSYNC_FLIP |
+ SLPC_OPTIMIZED_STRATEGIES_ASYNC_FLIP))
+ drm_info(&pc_to_xe(pc)->drm, "GuC PC couldn't init Optimized strategies: %pe\n",
+ ERR_PTR(ret));
+
ret = pc_init_freqs(pc);
if (ret)
goto out;
@@ -973,7 +1153,7 @@ void xe_guc_pc_fini(struct xe_guc_pc *pc)
XE_WARN_ON(xe_guc_pc_stop(pc));
mutex_destroy(&pc->freq_lock);
}
-
+#include <linux/iosys-map.h>
/**
* xe_guc_pc_init - Initialize GuC's Power Conservation component
* @pc: Xe_GuC_PC instance
@@ -984,13 +1164,14 @@ int xe_guc_pc_init(struct xe_guc_pc *pc)
struct xe_tile *tile = gt_to_tile(gt);
struct xe_device *xe = gt_to_xe(gt);
struct xe_bo *bo;
- u32 size = PAGE_ALIGN(sizeof(struct slpc_shared_data));
+ u32 size;
if (xe->info.skip_guc_pc)
return 0;
mutex_init(&pc->freq_lock);
+ size = PAGE_ALIGN(sizeof(struct slpc_shared_data));
bo = xe_managed_bo_create_pin_map(xe, tile, size,
XE_BO_CREATE_VRAM_IF_DGFX(tile) |
XE_BO_CREATE_GGTT_BIT);
@@ -998,5 +1179,155 @@ int xe_guc_pc_init(struct xe_guc_pc *pc)
return PTR_ERR(bo);
pc->bo = bo;
+
+ size = PAGE_ALIGN(sizeof(struct slpc_display_data));
+ bo = xe_managed_bo_create_pin_map(xe, tile, size,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ pc->display_bo = bo;
+
return 0;
}
+
+static const char *slpc_global_state_to_string(enum slpc_global_state state)
+{
+ switch (state) {
+ case SLPC_GLOBAL_STATE_NOT_RUNNING:
+ return "not running";
+ case SLPC_GLOBAL_STATE_INITIALIZING:
+ return "initializing";
+ case SLPC_GLOBAL_STATE_RESETTING:
+ return "resetting";
+ case SLPC_GLOBAL_STATE_RUNNING:
+ return "running";
+ case SLPC_GLOBAL_STATE_SHUTTING_DOWN:
+ return "shutting down";
+ case SLPC_GLOBAL_STATE_ERROR:
+ return "error";
+ default:
+ return "unknown";
+ }
+}
+
+static const char *pc_get_state_string(struct xe_guc_pc *pc)
+{
+ enum slpc_global_state state;
+
+ state = slpc_shared_data_read(pc, header.global_state);
+ return slpc_global_state_to_string(state);
+}
+
+static void pc_print_display_metrics(struct drm_printer *p,
+ struct slpc_display_data *data)
+{
+ int pipe, plane;
+ u32 val;
+
+ drm_printf(p, "\nSLPC Display Data - Globals:\n");
+ drm_printf(p, "\tVersion: %d\n", data->global_info.version);
+ drm_printf(p, "\tNum Pipes: %d\n", data->global_info.num_pipes);
+ drm_printf(p, "\tNum Planes per Pipe: %d\n",
+ data->global_info.num_planes_per_pipe);
+ drm_printf(p, "\tGlobal Refresh Info Count: %d\n",
+ data->global_info.refresh_count);
+ drm_printf(p, "\tGlobal Vblank Count: %d\n",
+ data->global_info.vblank_count);
+ drm_printf(p, "\tGlobal Flip Count: %d\n",
+ data->global_info.flip_count);
+
+ for (pipe = 0; pipe < SLPC_MAX_PIPES; pipe++) {
+
+ if (!data->refresh_info[pipe].refresh_interval)
+ continue;
+
+ drm_printf(p, "\nSLPC Display Data - Refresh Info - Pipe[%d]:\n",
+ pipe);
+ drm_printf(p, "\tRefresh Interval: %d\n",
+ data->refresh_info[pipe].refresh_interval);
+ drm_printf(p, "\tIS VRR: %d\n",
+ data->refresh_info[pipe].is_variable);
+
+ drm_printf(p, "SLPC Display Data - Vblank Info - Pipe[%d]:\n",
+ pipe);
+ val = data->vblank_metrics[pipe];
+ drm_printf(p, "\tVBlank Last Timestamp: %x\n",
+ REG_FIELD_GET(SLPC_VBLANK_LAST, val));
+ drm_printf(p, "\tVBlank Count: %d\n",
+ REG_FIELD_GET(SLPC_VBLANK_COUNT, val));
+
+ drm_printf(p, "SLPC Display Data - Flip Info - Pipe[%d]:\n", pipe);
+ for (plane = 0; plane < SLPC_MAX_PLANES_PER_PIPE; plane++) {
+
+ val = data->flip_metrics[pipe][plane].part1;
+ if (!val)
+ continue;
+
+ drm_printf(p, "\tFlip Info - Plane[%d]:\n", plane);
+ drm_printf(p, "\t\tFlip Last Timestamp: %x\n",
+ REG_FIELD_GET(SLPC_FLIP_P1_LAST, val));
+ drm_printf(p, "\t\tFlip Total Count: %d\n",
+ REG_FIELD_GET(SLPC_FLIP_P1_TOTAL_COUNT, val));
+
+ val = data->flip_metrics[pipe][plane].part2;
+
+ drm_printf(p, "\t\tFlip Async Count: %d\n",
+ REG_FIELD_GET(SLPC_FLIP_P2_ASYNC_COUNT, val));
+ drm_printf(p, "\t\tFlip Vsync Count: %d\n",
+ REG_FIELD_GET(SLPC_FLIP_P2_VSYNC_COUNT, val));
+ }
+ }
+}
+
+/**
+ * xe_guc_pc_print - Print GuC's Power Conservation information for debug
+ * @pc: Xe_GuC_PC instance
+ * @p: drm_printer
+ */
+void xe_guc_pc_print(struct xe_guc_pc *pc, struct drm_printer *p)
+{
+ struct slpc_display_data *data = pc_to_display_data(pc);
+ u32 display_addr;
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+
+ drm_printf(p, "SLPC Shared Data Header:\n");
+ drm_printf(p, "\tSize: %x\n", slpc_shared_data_read(pc, header.size));
+ drm_printf(p, "\tGlobal State: %s\n", pc_get_state_string(pc));
+ display_addr = slpc_shared_data_read(pc, header.display_data_addr);
+ drm_printf(p, "\tDisplay Data Addr: %x\n", display_addr);
+
+ if(pc_action_query_task_state(pc))
+ goto out;
+
+ drm_printf(p, "\nSLPC Tasks Status:\n");
+ drm_printf(p, "\tGTPERF enabled: %s\n",
+ str_yes_no(slpc_shared_data_read(pc, task_state_data.status) &
+ SLPC_GTPERF_TASK_ENABLED));
+ drm_printf(p, "\tDCC enabled: %s\n",
+ str_yes_no(slpc_shared_data_read(pc, task_state_data.status) &
+ SLPC_DCC_TASK_ENABLED));
+ drm_printf(p, "\tDCC in: %s\n",
+ str_yes_no(slpc_shared_data_read(pc, task_state_data.status) &
+ SLPC_IN_DCC));
+ drm_printf(p, "\tBalancer enabled: %s\n",
+ str_yes_no(slpc_shared_data_read(pc, task_state_data.status) &
+ SLPC_BALANCER_ENABLED));
+ drm_printf(p, "\tIBC enabled: %s\n",
+ str_yes_no(slpc_shared_data_read(pc, task_state_data.status) &
+ SLPC_IBC_TASK_ENABLED));
+ drm_printf(p, "\tBalancer IA LMT enabled: %s\n",
+ str_yes_no(slpc_shared_data_read(pc, task_state_data.status) &
+ SLPC_BALANCER_IA_LMT_ENABLED));
+ drm_printf(p, "\tBalancer IA LMT active: %s\n",
+ str_yes_no(slpc_shared_data_read(pc, task_state_data.status) &
+ SLPC_BALANCER_IA_LMT_ACTIVE));
+
+ if (display_addr)
+ pc_print_display_metrics(p, data);
+
+out:
+ xe_device_mem_access_put(pc_to_xe(pc));
+}
@@ -8,11 +8,14 @@
#include "xe_guc_pc_types.h"
+struct drm_printer;
+
int xe_guc_pc_init(struct xe_guc_pc *pc);
void xe_guc_pc_fini(struct xe_guc_pc *pc);
int xe_guc_pc_start(struct xe_guc_pc *pc);
int xe_guc_pc_stop(struct xe_guc_pc *pc);
int xe_guc_pc_gucrc_disable(struct xe_guc_pc *pc);
+void xe_guc_pc_print(struct xe_guc_pc *pc, struct drm_printer *p);
u32 xe_guc_pc_get_act_freq(struct xe_guc_pc *pc);
int xe_guc_pc_get_cur_freq(struct xe_guc_pc *pc, u32 *freq);
@@ -15,6 +15,8 @@
struct xe_guc_pc {
/** @bo: GGTT buffer object that is shared with GuC PC */
struct xe_bo *bo;
+ /** @display_bo: GGTT buffer object that is shared with GuC PC for display metricts */
+ struct xe_bo *display_bo;
/** @rp0_freq: HW RP0 frequency - The Maximum one */
u32 rp0_freq;
/** @rpe_freq: HW RPe frequency - The Efficient one */
Let's register with intel_display to receive display metrics that we can then send to GuC SLPC through the shared data and let it to select the optimal frequency. Cc: Jani Nikula <jani.nikula@intel.com> Cc: Lucas De Marchi <lucas.demarchi@intel.com> Cc: Vinay Belgaumkar <vinay.belgaumkar@intel.com> Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com> --- drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h | 59 +++ drivers/gpu/drm/xe/xe_gt_debugfs.c | 11 + drivers/gpu/drm/xe/xe_guc_pc.c | 343 +++++++++++++++++- drivers/gpu/drm/xe/xe_guc_pc.h | 3 + drivers/gpu/drm/xe/xe_guc_pc_types.h | 2 + 5 files changed, 412 insertions(+), 6 deletions(-)