xref: /linux/drivers/gpu/drm/amd/display/dc/dml/dcn10/dcn10_fpu.c (revision 7ae9fb1b7ecbb5d85d07857943f677fd1a559b18)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright 2021 Advanced Micro Devices, Inc.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be included in
13  * all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21  * OTHER DEALINGS IN THE SOFTWARE.
22  *
23  * Authors: AMD
24  *
25  */
26 
27 #include "dcn10/dcn10_resource.h"
28 
29 #include "dcn10_fpu.h"
30 #include "resource.h"
31 #include "amdgpu_dm/dc_fpu.h"
32 
33 /**
34  * DOC: DCN10 FPU manipulation Overview
35  *
36  * The DCN architecture relies on FPU operations, which require special
37  * compilation flags and the use of kernel_fpu_begin/end functions; ideally, we
38  * want to avoid spreading FPU access across multiple files. With this idea in
39  * mind, this file aims to centralize DCN10 functions that require FPU access
40  * in a single place. Code in this file follows the following code pattern:
41  *
42  * 1. Functions that use FPU operations should be isolated in static functions.
43  * 2. The FPU functions should have the noinline attribute to ensure anything
44  *    that deals with FP register is contained within this call.
45  * 3. All function that needs to be accessed outside this file requires a
46  *    public interface that not uses any FPU reference.
47  * 4. Developers **must not** use DC_FP_START/END in this file, but they need
48  *    to ensure that the caller invokes it before access any function available
49  *    in this file. For this reason, public functions in this file must invoke
50  *    dc_assert_fp_enabled();
51  *
52  * Let's expand a little bit more the idea in the code pattern. To fully
53  * isolate FPU operations in a single place, we must avoid situations where
54  * compilers spill FP values to registers due to FP enable in a specific C
55  * file. Note that even if we isolate all FPU functions in a single file and
56  * call its interface from other files, the compiler might enable the use of
57  * FPU before we call DC_FP_START. Nevertheless, it is the programmer's
58  * responsibility to invoke DC_FP_START/END in the correct place. To highlight
59  * situations where developers forgot to use the FP protection before calling
60  * the DC FPU interface functions, we introduce a helper that checks if the
61  * function is invoked under FP protection. If not, it will trigger a kernel
62  * warning.
63  */
64 
65 struct _vcs_dpi_ip_params_st dcn1_0_ip = {
66 	.rob_buffer_size_kbytes = 64,
67 	.det_buffer_size_kbytes = 164,
68 	.dpte_buffer_size_in_pte_reqs_luma = 42,
69 	.dpp_output_buffer_pixels = 2560,
70 	.opp_output_buffer_lines = 1,
71 	.pixel_chunk_size_kbytes = 8,
72 	.pte_enable = 1,
73 	.pte_chunk_size_kbytes = 2,
74 	.meta_chunk_size_kbytes = 2,
75 	.writeback_chunk_size_kbytes = 2,
76 	.line_buffer_size_bits = 589824,
77 	.max_line_buffer_lines = 12,
78 	.IsLineBufferBppFixed = 0,
79 	.LineBufferFixedBpp = -1,
80 	.writeback_luma_buffer_size_kbytes = 12,
81 	.writeback_chroma_buffer_size_kbytes = 8,
82 	.max_num_dpp = 4,
83 	.max_num_wb = 2,
84 	.max_dchub_pscl_bw_pix_per_clk = 4,
85 	.max_pscl_lb_bw_pix_per_clk = 2,
86 	.max_lb_vscl_bw_pix_per_clk = 4,
87 	.max_vscl_hscl_bw_pix_per_clk = 4,
88 	.max_hscl_ratio = 4,
89 	.max_vscl_ratio = 4,
90 	.hscl_mults = 4,
91 	.vscl_mults = 4,
92 	.max_hscl_taps = 8,
93 	.max_vscl_taps = 8,
94 	.dispclk_ramp_margin_percent = 1,
95 	.underscan_factor = 1.10,
96 	.min_vblank_lines = 14,
97 	.dppclk_delay_subtotal = 90,
98 	.dispclk_delay_subtotal = 42,
99 	.dcfclk_cstate_latency = 10,
100 	.max_inter_dcn_tile_repeaters = 8,
101 	.can_vstartup_lines_exceed_vsync_plus_back_porch_lines_minus_one = 0,
102 	.bug_forcing_LC_req_same_size_fixed = 0,
103 };
104 
105 struct _vcs_dpi_soc_bounding_box_st dcn1_0_soc = {
106 	.sr_exit_time_us = 9.0,
107 	.sr_enter_plus_exit_time_us = 11.0,
108 	.urgent_latency_us = 4.0,
109 	.writeback_latency_us = 12.0,
110 	.ideal_dram_bw_after_urgent_percent = 80.0,
111 	.max_request_size_bytes = 256,
112 	.downspread_percent = 0.5,
113 	.dram_page_open_time_ns = 50.0,
114 	.dram_rw_turnaround_time_ns = 17.5,
115 	.dram_return_buffer_per_channel_bytes = 8192,
116 	.round_trip_ping_latency_dcfclk_cycles = 128,
117 	.urgent_out_of_order_return_per_channel_bytes = 256,
118 	.channel_interleave_bytes = 256,
119 	.num_banks = 8,
120 	.num_chans = 2,
121 	.vmm_page_size_bytes = 4096,
122 	.dram_clock_change_latency_us = 17.0,
123 	.writeback_dram_clock_change_latency_us = 23.0,
124 	.return_bus_width_bytes = 64,
125 };
126 
dcn10_resource_construct_fp(struct dc * dc)127 void dcn10_resource_construct_fp(struct dc *dc)
128 {
129 	dc_assert_fp_enabled();
130 	if (dc->ctx->dce_version == DCN_VERSION_1_01) {
131 		struct dcn_soc_bounding_box *dcn_soc = dc->dcn_soc;
132 		struct dcn_ip_params *dcn_ip = dc->dcn_ip;
133 		struct display_mode_lib *dml = &dc->dml;
134 
135 		dml->ip.max_num_dpp = 3;
136 		/* TODO how to handle 23.84? */
137 		dcn_soc->dram_clock_change_latency = 23;
138 		dcn_ip->max_num_dpp = 3;
139 	}
140 	if (ASICREV_IS_RV1_F0(dc->ctx->asic_id.hw_internal_rev)) {
141 		dc->dcn_soc->urgent_latency = 3;
142 		dc->debug.disable_dmcu = true;
143 		dc->dcn_soc->fabric_and_dram_bandwidth_vmax0p9 = 41.60f;
144 	}
145 
146 	dc->dcn_soc->number_of_channels = dc->ctx->asic_id.vram_width / ddr4_dram_width;
147 	ASSERT(dc->dcn_soc->number_of_channels < 3);
148 	if (dc->dcn_soc->number_of_channels == 0)/*old sbios bug*/
149 		dc->dcn_soc->number_of_channels = 2;
150 
151 	if (dc->dcn_soc->number_of_channels == 1) {
152 		dc->dcn_soc->fabric_and_dram_bandwidth_vmax0p9 = 19.2f;
153 		dc->dcn_soc->fabric_and_dram_bandwidth_vnom0p8 = 17.066f;
154 		dc->dcn_soc->fabric_and_dram_bandwidth_vmid0p72 = 14.933f;
155 		dc->dcn_soc->fabric_and_dram_bandwidth_vmin0p65 = 12.8f;
156 		if (ASICREV_IS_RV1_F0(dc->ctx->asic_id.hw_internal_rev))
157 			dc->dcn_soc->fabric_and_dram_bandwidth_vmax0p9 = 20.80f;
158 	}
159 }
160