[PATCH 1/2] drm/amd/pm: correct gpu metrics related data structures V2

[Lazar, Lijo]

[AMD Public Use]



-----Original Message-----
From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Evan Quan
Sent: Monday, March 1, 2021 11:20 AM
To: amd-gfx at lists.freedesktop.org
Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Quan, Evan <Evan.Quan at amd.com>
Subject: [PATCH 1/2] drm/amd/pm: correct gpu metrics related data structures V2

To make sure they are naturally aligned. Also updating the data type for link_speed/width for future PCIE5 support.

Change-Id: I0a139e1e1f09fe27deffdce1cec6ea4594947625
Signed-off-by: Evan Quan <evan.quan at amd.com>
---
 .../gpu/drm/amd/include/kgd_pp_interface.h    | 114 ++++++++++++++++++
 1 file changed, 114 insertions(+)

diff --git a/drivers/gpu/drm/amd/include/kgd_pp_interface.h b/drivers/gpu/drm/amd/include/kgd_pp_interface.h
index 41c89f7d6412..dbb066a39fdc 100644
--- a/drivers/gpu/drm/amd/include/kgd_pp_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_pp_interface.h
@@ -341,6 +341,10 @@ struct metrics_table_header {
 	uint8_t				content_revision;
 };
 
+/*
+ * gpu_metrics_v1_0 is not recommended as it's not naturally aligned.
+ * Use gpu_metrics_v1_1 or later instead.
+ */
 struct gpu_metrics_v1_0 {
 	struct metrics_table_header	common_header;
 
@@ -393,6 +397,64 @@ struct gpu_metrics_v1_0 {
 	uint8_t				pcie_link_speed; // in 0.1 GT/s
 };
 
+struct gpu_metrics_v1_1 {
+	struct metrics_table_header	common_header;
+
+	/* Temperature */
+	uint16_t			temperature_edge;
+	uint16_t			temperature_hotspot;
+	uint16_t			temperature_mem;
+	uint16_t			temperature_vrgfx;
+	uint16_t			temperature_vrsoc;
+	uint16_t			temperature_vrmem;
+
+	/* Utilization */
+	uint16_t			average_gfx_activity;
+	uint16_t			average_umc_activity; // memory controller
+	uint16_t			average_mm_activity; // UVD or VCN
+
+	/* Power/Energy */
+	uint16_t			average_socket_power;
+	uint32_t			energy_accumulator;

<> Energy accumulator is 64-bit in newer FW interface. Once change is made, this will need realignment.

+	/* Average clocks */
+	uint16_t			average_gfxclk_frequency;
+	uint16_t			average_socclk_frequency;
+	uint16_t			average_uclk_frequency;
+	uint16_t			average_vclk0_frequency;
+	uint16_t			average_dclk0_frequency;
+	uint16_t			average_vclk1_frequency;
+	uint16_t			average_dclk1_frequency;
+
+	/* Current clocks */
+	uint16_t			current_gfxclk;
+	uint16_t			current_socclk;
+	uint16_t			current_uclk;
+	uint16_t			current_vclk0;
+	uint16_t			current_dclk0;
+	uint16_t			current_vclk1;
+	uint16_t			current_dclk1;
+
+	/* Driver attached timestamp (in ns) */
+	uint64_t			system_clock_counter;
+
+	/* Throttle status */
+	uint32_t			throttle_status;
+
+	/* Fans */
+	uint16_t			current_fan_speed;
+
+	/* Link width/speed */
+	uint16_t			pcie_link_width;
+	uint16_t			pcie_link_speed; // in 0.1 GT/s
+
+	uint16_t			padding[3];
+};
+
+/*
+ * gpu_metrics_v2_0 is not recommended as it's not naturally aligned.
+ * Use gpu_metrics_v2_1 or later instead.
+ */
 struct gpu_metrics_v2_0 {
 	struct metrics_table_header	common_header;
 
@@ -443,4 +505,56 @@ struct gpu_metrics_v2_0 {
 	uint16_t			padding;
 };
 
+struct gpu_metrics_v2_1 {
+	struct metrics_table_header	common_header;
+
+	/* Temperature */
+	uint16_t			temperature_gfx; // gfx temperature on APUs
+	uint16_t			temperature_soc; // soc temperature on APUs
+	uint16_t			temperature_core[8]; // CPU core temperature on APUs
+	uint16_t			temperature_l3[2];
+
+	/* Utilization */
+	uint16_t			average_gfx_activity;
+	uint16_t			average_mm_activity; // UVD or VCN
+
+	/* Power/Energy */
+	uint16_t			average_socket_power; // dGPU + APU power on A + A platform
+	uint16_t			average_cpu_power;
+	uint16_t			average_soc_power;
+	uint16_t			average_gfx_power;
+	uint16_t			average_core_power[8]; // CPU core power on APUs
+
+	/* Average clocks */
+	uint16_t			average_gfxclk_frequency;
+	uint16_t			average_socclk_frequency;
+	uint16_t			average_uclk_frequency;
+	uint16_t			average_fclk_frequency;
+	uint16_t			average_vclk_frequency;
+	uint16_t			average_dclk_frequency;
+
+	/* Current clocks */
+	uint16_t			current_gfxclk;
+	uint16_t			current_socclk;
+	uint16_t			current_uclk;
+	uint16_t			current_fclk;
+	uint16_t			current_vclk;
+	uint16_t			current_dclk;
+	uint16_t			current_coreclk[8]; // CPU core clocks
+	uint16_t			current_l3clk[2];
+
+	uint16_t			padding[2];

<> This padding can be avoided if throttle status is moved up.

Thanks,
Lijo

+	/* Driver attached timestamp (in ns) */
+	uint64_t			system_clock_counter;
+
+	/* Throttle status */
+	uint32_t			throttle_status;
+
+	/* Fans */
+	uint16_t			fan_pwm;
+
+	uint16_t			padding_1;
+};
+
 #endif
--
2.29.0

_______________________________________________
amd-gfx mailing list
amd-gfx at lists.freedesktop.org
https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Famd-gfx&data=04%7C01%7Clijo.lazar%40amd.com%7Cb8068a2b79e64889dc5408d8dc75f3a0%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637501746480810976%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=QzggLryhGEI%2FtBWhlu0RfsIX3C%2F7rK21F9VJPF0vNt0%3D&reserved=0