summaryrefslogtreecommitdiff
path: root/config/coreboot/haswell/patches/0013-haswell-NRI-Add-read-MPR-training.patch
blob: 5f154bcc66804d4049b28b472cb1b8c6115cdcf5 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
From e263f0d2e9d6d016d603342651da261bbcb6af1f Mon Sep 17 00:00:00 2001
From: Angel Pons <th3fanbus@gmail.com>
Date: Sun, 8 May 2022 11:35:49 +0200
Subject: [PATCH 13/20] haswell NRI: Add read MPR training

Implement read training using DDR3 MPR (Multi-Purpose Register).

Change-Id: Id17cb2c4c399ac9bcc937b595b58f863c152461b
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
---
 .../intel/haswell/native_raminit/Makefile.mk  |   1 +
 .../haswell/native_raminit/raminit_main.c     |   1 +
 .../haswell/native_raminit/raminit_native.h   |   4 +
 .../haswell/native_raminit/train_read_mpr.c   | 241 ++++++++++++++++++
 .../intel/haswell/registers/mchbar.h          |   2 +-
 5 files changed, 248 insertions(+), 1 deletion(-)
 create mode 100644 src/northbridge/intel/haswell/native_raminit/train_read_mpr.c

diff --git a/src/northbridge/intel/haswell/native_raminit/Makefile.mk b/src/northbridge/intel/haswell/native_raminit/Makefile.mk
index e2fbfb4211..c442be0728 100644
--- a/src/northbridge/intel/haswell/native_raminit/Makefile.mk
+++ b/src/northbridge/intel/haswell/native_raminit/Makefile.mk
@@ -16,4 +16,5 @@ romstage-y += setup_wdb.c
 romstage-y += spd_bitmunching.c
 romstage-y += testing_io.c
 romstage-y += timings_refresh.c
+romstage-y += train_read_mpr.c
 romstage-y += train_receive_enable.c
diff --git a/src/northbridge/intel/haswell/native_raminit/raminit_main.c b/src/northbridge/intel/haswell/native_raminit/raminit_main.c
index 7d444659c3..264d1468f5 100644
--- a/src/northbridge/intel/haswell/native_raminit/raminit_main.c
+++ b/src/northbridge/intel/haswell/native_raminit/raminit_main.c
@@ -61,6 +61,7 @@ static const struct task_entry cold_boot[] = {
 	{ do_jedec_init,                                          true, "JEDECINIT",  },
 	{ pre_training,                                           true, "PRETRAIN",   },
 	{ train_receive_enable,                                   true, "RCVET",      },
+	{ train_read_mpr,                                         true, "RDMPRT",     },
 };
 
 /* Return a generic stepping value to make stepping checks simpler */
diff --git a/src/northbridge/intel/haswell/native_raminit/raminit_native.h b/src/northbridge/intel/haswell/native_raminit/raminit_native.h
index 5242b16f28..49e9214656 100644
--- a/src/northbridge/intel/haswell/native_raminit/raminit_native.h
+++ b/src/northbridge/intel/haswell/native_raminit/raminit_native.h
@@ -27,6 +27,8 @@
 /* Always use 12 legs for emphasis (not trained) */
 #define TXEQFULLDRV		(3 << 4)
 
+#define LOOPCOUNT_INFINITE	0xff
+
 /* DDR3 mode register bits */
 #define MR0_DLL_RESET		BIT(8)
 
@@ -212,6 +214,7 @@ enum raminit_status {
 	RAMINIT_STATUS_POLL_TIMEOUT,
 	RAMINIT_STATUS_REUT_ERROR,
 	RAMINIT_STATUS_RCVEN_FAILURE,
+	RAMINIT_STATUS_RMPR_FAILURE,
 	RAMINIT_STATUS_UNSPECIFIED_ERROR, /** TODO: Deprecated in favor of specific values **/
 };
 
@@ -433,6 +436,7 @@ enum raminit_status configure_mc(struct sysinfo *ctrl);
 enum raminit_status configure_memory_map(struct sysinfo *ctrl);
 enum raminit_status do_jedec_init(struct sysinfo *ctrl);
 enum raminit_status train_receive_enable(struct sysinfo *ctrl);
+enum raminit_status train_read_mpr(struct sysinfo *ctrl);
 
 void configure_timings(struct sysinfo *ctrl);
 void configure_refresh(struct sysinfo *ctrl);
diff --git a/src/northbridge/intel/haswell/native_raminit/train_read_mpr.c b/src/northbridge/intel/haswell/native_raminit/train_read_mpr.c
new file mode 100644
index 0000000000..ade1e36148
--- /dev/null
+++ b/src/northbridge/intel/haswell/native_raminit/train_read_mpr.c
@@ -0,0 +1,241 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include <commonlib/bsd/clamp.h>
+#include <console/console.h>
+#include <delay.h>
+#include <northbridge/intel/haswell/haswell.h>
+#include <types.h>
+
+#include "raminit_native.h"
+#include "ranges.h"
+
+#define RMPR_START	(-32)
+#define RMPR_STOP	(32)
+#define RMPR_STEP	1
+
+#define RMPR_MIN_WIDTH	12
+
+#define RMPR_PLOT	RAM_DEBUG
+
+/*
+ * Clear rx_training_mode. For LPDDR, we first need to disable odt_samp_extend_en,
+ * then disable rx_training_mode, and finally re-enable odt_samp_extend_en.
+ */
+static void clear_rx_training_mode(struct sysinfo *ctrl, const uint8_t channel)
+{
+	for (uint8_t byte = 0; byte < ctrl->lanes; byte++)
+		mchbar_write32(DQ_CONTROL_2(channel, byte), ctrl->dq_control_2[channel][byte]);
+
+	if (ctrl->lpddr) {
+		union ddr_data_control_0_reg data_control_0 = {
+			.raw = mchbar_read32(DDR_DATA_ch_CONTROL_0(channel)),
+		};
+		data_control_0.odt_samp_extend_en = 0;
+		mchbar_write32(DDR_DATA_ch_CONTROL_0(channel), data_control_0.raw);
+		tick_delay(1);
+		data_control_0.rx_training_mode = 0;
+		mchbar_write32(DDR_DATA_ch_CONTROL_0(channel), data_control_0.raw);
+		tick_delay(1);
+	}
+	mchbar_write32(DDR_DATA_ch_CONTROL_0(channel), ctrl->dq_control_0[channel]);
+}
+
+static void set_rxdqs_edges_to_midpoint(struct sysinfo *ctrl)
+{
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!does_ch_exist(ctrl, channel))
+			continue;
+
+		for (uint8_t rank = 0; rank < NUM_SLOTRANKS; rank++) {
+			if (!rank_in_ch(ctrl, rank, channel))
+				continue;
+
+			for (uint8_t byte = 0; byte < ctrl->lanes; byte++)
+				update_rxt(ctrl, channel, rank, byte, RXT_RXDQS_BOTH, 32);
+		}
+	}
+}
+
+static void enter_mpr_train_ddr_mode(struct sysinfo *ctrl, const uint8_t rank)
+{
+	/* Program MR3 and mask RAS/WE to prevent scheduler from issuing non-read commands */
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!rank_in_ch(ctrl, rank, channel))
+			continue;
+
+		if (!ctrl->lpddr)
+			reut_issue_mrs(ctrl, channel, BIT(rank), 3, 1 << 2);
+
+		union reut_misc_odt_ctrl_reg reut_misc_odt_ctrl = {
+			.raw = mchbar_read32(REUT_ch_MISC_ODT_CTRL(channel)),
+		};
+		reut_misc_odt_ctrl.mpr_train_ddr_on = 1;
+		mchbar_write32(REUT_ch_MISC_ODT_CTRL(channel), reut_misc_odt_ctrl.raw);
+	}
+}
+
+static void leave_mpr_train_ddr_mode(struct sysinfo *ctrl, const uint8_t rank)
+{
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!rank_in_ch(ctrl, rank, channel))
+			continue;
+
+		/*
+		 * The mpr_train_ddr_on bit will force a special command.
+		 * Therefore, clear it before issuing the MRS command.
+		 */
+		union reut_misc_odt_ctrl_reg reut_misc_odt_ctrl = {
+			.raw = mchbar_read32(REUT_ch_MISC_ODT_CTRL(channel)),
+		};
+		reut_misc_odt_ctrl.mpr_train_ddr_on = 0;
+		mchbar_write32(REUT_ch_MISC_ODT_CTRL(channel), reut_misc_odt_ctrl.raw);
+		if (!ctrl->lpddr)
+			reut_issue_mrs(ctrl, channel, BIT(rank), 3, 0 << 2);
+	}
+}
+
+enum raminit_status train_read_mpr(struct sysinfo *ctrl)
+{
+	set_rxdqs_edges_to_midpoint(ctrl);
+	clear_data_offset_train_all(ctrl);
+	setup_io_test_mpr(ctrl, ctrl->chanmap, LOOPCOUNT_INFINITE, NSOE);
+	enum raminit_status status = RAMINIT_STATUS_SUCCESS;
+	for (uint8_t rank = 0; rank < NUM_SLOTRANKS; rank++) {
+		if (!does_rank_exist(ctrl, rank))
+			continue;
+
+		printk(BIOS_DEBUG, "Rank %u\n", rank);
+		printk(RMPR_PLOT, "Channel");
+		for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+			if (!rank_in_ch(ctrl, rank, channel))
+				continue;
+
+			printk(RMPR_PLOT, "\t%u\t\t", channel);
+		}
+		printk(RMPR_PLOT, "\nByte");
+		for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+			if (!rank_in_ch(ctrl, rank, channel))
+				continue;
+
+			printk(RMPR_PLOT, "\t");
+			for (uint8_t byte = 0; byte < ctrl->lanes; byte++)
+				printk(RMPR_PLOT, "%u ", byte);
+		}
+		enter_mpr_train_ddr_mode(ctrl, rank);
+		struct linear_train_data region_data[NUM_CHANNELS][NUM_LANES] = { 0 };
+		for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++)
+			select_reut_ranks(ctrl, channel, BIT(rank));
+
+		printk(RMPR_PLOT, "\nDqsDelay\n");
+		int8_t dqs_delay;
+		for (dqs_delay = RMPR_START; dqs_delay < RMPR_STOP; dqs_delay += RMPR_STEP) {
+			printk(RMPR_PLOT, "% 5d", dqs_delay);
+			const enum regfile_mode regfile = REG_FILE_USE_START;
+			/* Looks like MRC uses rank 0 here, but it feels wrong */
+			change_1d_margin_multicast(ctrl, RdT, dqs_delay, rank, false, regfile);
+			for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+				if (!rank_in_ch(ctrl, rank, channel))
+					continue;
+
+				for (uint8_t byte = 0; byte < ctrl->lanes; byte++) {
+					union ddr_data_control_2_reg data_control_2 = {
+						.raw = ctrl->dq_control_2[channel][byte],
+					};
+					data_control_2.force_bias_on = 1;
+					data_control_2.force_rx_on   = 1;
+					data_control_2.leaker_comp   = 0;
+					mchbar_write32(DQ_CONTROL_2(channel, byte),
+						data_control_2.raw);
+				}
+				union ddr_data_control_0_reg data_control_0 = {
+					.raw = ctrl->dq_control_0[channel],
+				};
+				data_control_0.rx_training_mode   = 1;
+				data_control_0.force_odt_on       = !ctrl->lpddr;
+				data_control_0.en_read_preamble   = 0;
+				data_control_0.odt_samp_extend_en = ctrl->lpddr;
+				const uint32_t reg_offset = DDR_DATA_ch_CONTROL_0(channel);
+				mchbar_write32(reg_offset, data_control_0.raw);
+			}
+			run_mpr_io_test(false);
+			for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+				if (!rank_in_ch(ctrl, rank, channel))
+					continue;
+
+				printk(RMPR_PLOT, "\t");
+				for (uint8_t byte = 0; byte < ctrl->lanes; byte++) {
+					uint32_t fb = get_data_train_feedback(channel, byte);
+					const bool pass = fb == 1;
+					printk(RMPR_PLOT, pass ? ". " : "# ");
+					linear_record_pass(
+						&region_data[channel][byte],
+						pass,
+						dqs_delay,
+						RMPR_START,
+						RMPR_STEP);
+				}
+			}
+			printk(RMPR_PLOT, "\n");
+			for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+				if (!rank_in_ch(ctrl, rank, channel))
+					continue;
+
+				clear_rx_training_mode(ctrl, channel);
+			}
+			io_reset();
+		}
+		printk(RMPR_PLOT, "\n");
+		leave_mpr_train_ddr_mode(ctrl, rank);
+		clear_data_offset_train_all(ctrl);
+		for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+			if (!rank_in_ch(ctrl, rank, channel))
+				continue;
+
+			printk(BIOS_DEBUG, "C%u.R%u: \tLeft\tRight\tWidth\tCenter\tRxDqsPN\n",
+				channel, rank);
+			for (uint8_t byte = 0; byte < ctrl->lanes; byte++) {
+				struct linear_train_data *data = &region_data[channel][byte];
+				const int32_t lwidth = range_width(data->largest);
+				if (lwidth <= RMPR_MIN_WIDTH) {
+					printk(BIOS_ERR,
+						"Bad eye (lwidth %d <= min %d) for byte %u\n",
+						lwidth, RMPR_MIN_WIDTH, byte);
+					status = RAMINIT_STATUS_RMPR_FAILURE;
+				}
+				/*
+				 * The MPR center may not be ideal on certain platforms for
+				 * unknown reasons. If so, adjust it with a magical number.
+				 * For Haswell, the magical number is zero. Hell knows why.
+				 */
+				const int32_t center = range_center(data->largest);
+				ctrl->rxdqsp[channel][rank][byte] = center - RMPR_START;
+				ctrl->rxdqsn[channel][rank][byte] = center - RMPR_START;
+				printk(BIOS_DEBUG, "  B%u: \t%d\t%d\t%d\t%d\t%u\n", byte,
+					data->largest.start, data->largest.end, lwidth,
+					center, ctrl->rxdqsp[channel][rank][byte]);
+			}
+			printk(BIOS_DEBUG, "\n");
+		}
+	}
+
+	/*
+	 * Now program the DQS center values on populated ranks. data is taken from
+	 * the host struct. We need to do it after all ranks are trained, because we
+	 * need to keep the same DQS value on all ranks during the training procedure.
+	 */
+	for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) {
+		if (!does_ch_exist(ctrl, channel))
+			continue;
+
+		for (uint8_t rank = 0; rank < NUM_SLOTRANKS; rank++) {
+			if (!rank_in_ch(ctrl, rank, channel))
+				continue;
+
+			for (uint8_t byte = 0; byte < ctrl->lanes; byte++)
+				update_rxt(ctrl, channel, rank, byte, RXT_RESTORE, 0);
+		}
+	}
+	change_1d_margin_multicast(ctrl, RdT, 0, 0, false, REG_FILE_USE_CURRENT);
+	io_reset();
+	return status;
+}
diff --git a/src/northbridge/intel/haswell/registers/mchbar.h b/src/northbridge/intel/haswell/registers/mchbar.h
index 0acafbc826..6a31d3a32c 100644
--- a/src/northbridge/intel/haswell/registers/mchbar.h
+++ b/src/northbridge/intel/haswell/registers/mchbar.h
@@ -122,7 +122,7 @@
 #define REUT_ch_ERR_DATA_MASK(ch)		_MCMAIN_C(0x40d8, ch)
 
 #define REUT_ch_MISC_CKE_CTRL(ch)		_MCMAIN_C(0x4190, ch)
-
+#define REUT_ch_MISC_ODT_CTRL(ch)		_MCMAIN_C(0x4194, ch)
 #define REUT_ch_MISC_PAT_CADB_CTRL(ch)		_MCMAIN_C(0x4198, ch)
 #define REUT_ch_PAT_CADB_MRS(ch)		_MCMAIN_C(0x419c, ch)
 #define REUT_ch_PAT_CADB_MUX_CTRL(ch)		_MCMAIN_C(0x41a0, ch)
-- 
2.39.2