1 /* 2 * sun9i dram controller initialisation 3 * 4 * (C) Copyright 2007-2015 5 * Allwinner Technology Co., Ltd. <www.allwinnertech.com> 6 * Jerry Wang <wangflord@allwinnertech.com> 7 * 8 * (C) Copyright 2016 Theobroma Systems Design und Consulting GmbH 9 * Philipp Tomsich <philipp.tomsich@theobroma-systems.com> 10 * 11 * SPDX-License-Identifier: GPL-2.0+ 12 */ 13 14 #include <common.h> 15 #include <dm.h> 16 #include <errno.h> 17 #include <ram.h> 18 #include <asm/io.h> 19 #include <asm/arch/clock.h> 20 #include <asm/arch/dram.h> 21 #include <asm/arch/sys_proto.h> 22 23 #define DRAM_CLK (CONFIG_DRAM_CLK * 1000000) 24 25 /* 26 * The following amounts to an extensive rewrite of the code received from 27 * Allwinner as part of the open-source bootloader release (refer to 28 * https://github.com/allwinner-zh/bootloader.git) and augments the upstream 29 * sources (which act as the primary reference point for the inner workings 30 * of the 'underdocumented' DRAM controller in the A80) using the following 31 * documentation for other memory controllers based on the (Synopsys) 32 * Designware IP (DDR memory protocol controller and DDR PHY) 33 * * TI Keystone II Architecture: DDR3 Memory Controller, User's Guide 34 * Document 'SPRUHN7C', Oct 2013 (revised March 2015) 35 * * Xilinx Zynq UltraScale+ MPSoC Register Reference 36 * document ug1087 (v1.0) 37 * Note that the Zynq-documentation provides a very close match for the DDR 38 * memory protocol controller (and provides a very good guide to the rounding 39 * rules for various timings), whereas the TI Keystone II document should be 40 * referred to for DDR PHY specifics only. 41 * 42 * The DRAM controller in the A80 runs at half the frequency of the DDR PHY 43 * (i.e. the rules for MEMC_FREQ_RATIO=2 from the Zynq-documentation apply). 44 * 45 * Known limitations 46 * ================= 47 * In the current state, the following features are not fully supported and 48 * a number of simplifying assumptions have been made: 49 * 1) Only DDR3 support is implemented, as our test platform (the A80-Q7 50 * module) is designed to accomodate DDR3/DDR3L. 51 * 2) Only 2T-mode has been implemented and tested. 52 * 3) The controller supports two different clocking strategies (PLL6 can 53 * either be 2*CK or CK/2)... we only support the 2*CK clock at this 54 * time and haven't verified whether the alternative clocking strategy 55 * works. If you are interested in porting this over/testing this, 56 * please refer to cases where bit 0 of 'dram_tpr8' is tested in the 57 * original code from Allwinner. 58 * 4) Support for 2 ranks per controller is not implemented (as we don't 59 * the hardware to test it). 60 * 61 * Future directions 62 * ================= 63 * The driver should be driven from a device-tree based configuration that 64 * can dynamically provide the necessary timing parameters (i.e. target 65 * frequency and speed-bin information)---the data structures used in the 66 * calculation of the timing parameters are already designed to capture 67 * similar information as the device tree would provide. 68 * 69 * To enable a device-tree based configuration of the sun9i platform, we 70 * will need to enable CONFIG_TPL and bootstrap in 3 stages: initially 71 * into SRAM A1 (40KB) and next into SRAM A2 (160KB)---which would be the 72 * stage to initialise the platform via the device-tree---before having 73 * the full U-Boot run from DDR. 74 */ 75 76 /* 77 * A number of DDR3 timings are given as "the greater of a fixed number of 78 * clock cycles (CK) or nanoseconds. We express these using a structure 79 * that holds a cycle count and a duration in picoseconds (so we can model 80 * sub-ns timings, such as 7.5ns without losing precision or resorting to 81 * rounding up early. 82 */ 83 struct dram_sun9i_timing { 84 u32 ck; 85 u32 ps; 86 }; 87 88 /* */ 89 struct dram_sun9i_cl_cwl_timing { 90 u32 CL; 91 u32 CWL; 92 u32 tCKmin; /* in ps */ 93 u32 tCKmax; /* in ps */ 94 }; 95 96 struct dram_sun9i_para { 97 u32 dram_type; 98 99 u8 bus_width; 100 u8 chan; 101 u8 rank; 102 u8 rows; 103 u16 page_size; 104 105 /* Timing information for each speed-bin */ 106 struct dram_sun9i_cl_cwl_timing *cl_cwl_table; 107 u32 cl_cwl_numentries; 108 109 /* 110 * For the timings, we try to keep the order and grouping used in 111 * JEDEC Standard No. 79-3F 112 */ 113 114 /* timings */ 115 u32 tREFI; /* in ns */ 116 u32 tRFC; /* in ns */ 117 118 u32 tRAS; /* in ps */ 119 120 /* command and address timing */ 121 u32 tDLLK; /* in nCK */ 122 struct dram_sun9i_timing tRTP; 123 struct dram_sun9i_timing tWTR; 124 u32 tWR; /* in nCK */ 125 u32 tMRD; /* in nCK */ 126 struct dram_sun9i_timing tMOD; 127 u32 tRCD; /* in ps */ 128 u32 tRP; /* in ps */ 129 u32 tRC; /* in ps */ 130 u32 tCCD; /* in nCK */ 131 struct dram_sun9i_timing tRRD; 132 u32 tFAW; /* in ps */ 133 134 /* calibration timing */ 135 /* struct dram_sun9i_timing tZQinit; */ 136 struct dram_sun9i_timing tZQoper; 137 struct dram_sun9i_timing tZQCS; 138 139 /* reset timing */ 140 /* struct dram_sun9i_timing tXPR; */ 141 142 /* self-refresh timings */ 143 struct dram_sun9i_timing tXS; 144 u32 tXSDLL; /* in nCK */ 145 /* struct dram_sun9i_timing tCKESR; */ 146 struct dram_sun9i_timing tCKSRE; 147 struct dram_sun9i_timing tCKSRX; 148 149 /* power-down timings */ 150 struct dram_sun9i_timing tXP; 151 struct dram_sun9i_timing tXPDLL; 152 struct dram_sun9i_timing tCKE; 153 154 /* write leveling timings */ 155 u32 tWLMRD; /* min, in nCK */ 156 /* u32 tWLDQSEN; min, in nCK */ 157 u32 tWLO; /* max, in ns */ 158 /* u32 tWLOE; max, in ns */ 159 160 /* u32 tCKDPX; in nCK */ 161 /* u32 tCKCSX; in nCK */ 162 }; 163 164 static void mctl_sys_init(void); 165 166 #define SCHED_RDWR_IDLE_GAP(n) ((n & 0xff) << 24) 167 #define SCHED_GO2CRITICAL_HYSTERESIS(n) ((n & 0xff) << 16) 168 #define SCHED_LPR_NUM_ENTRIES(n) ((n & 0xff) << 8) 169 #define SCHED_PAGECLOSE (1 << 2) 170 #define SCHED_PREFER_WRITE (1 << 1) 171 #define SCHED_FORCE_LOW_PRI_N (1 << 0) 172 173 #define SCHED_CONFIG (SCHED_RDWR_IDLE_GAP(0xf) | \ 174 SCHED_GO2CRITICAL_HYSTERESIS(0x80) | \ 175 SCHED_LPR_NUM_ENTRIES(0x20) | \ 176 SCHED_FORCE_LOW_PRI_N) 177 #define PERFHPR0_CONFIG 0x0000001f 178 #define PERFHPR1_CONFIG 0x1f00001f 179 #define PERFLPR0_CONFIG 0x000000ff 180 #define PERFLPR1_CONFIG 0x0f0000ff 181 #define PERFWR0_CONFIG 0x000000ff 182 #define PERFWR1_CONFIG 0x0f0001ff 183 184 static void mctl_ctl_sched_init(unsigned long base) 185 { 186 struct sunxi_mctl_ctl_reg *mctl_ctl = 187 (struct sunxi_mctl_ctl_reg *)base; 188 189 /* Needs to be done before the global clk enable... */ 190 writel(SCHED_CONFIG, &mctl_ctl->sched); 191 writel(PERFHPR0_CONFIG, &mctl_ctl->perfhpr0); 192 writel(PERFHPR1_CONFIG, &mctl_ctl->perfhpr1); 193 writel(PERFLPR0_CONFIG, &mctl_ctl->perflpr0); 194 writel(PERFLPR1_CONFIG, &mctl_ctl->perflpr1); 195 writel(PERFWR0_CONFIG, &mctl_ctl->perfwr0); 196 writel(PERFWR1_CONFIG, &mctl_ctl->perfwr1); 197 } 198 199 static void mctl_sys_init(void) 200 { 201 struct sunxi_ccm_reg * const ccm = 202 (struct sunxi_ccm_reg *)SUNXI_CCM_BASE; 203 struct sunxi_mctl_com_reg * const mctl_com = 204 (struct sunxi_mctl_com_reg *)SUNXI_DRAM_COM_BASE; 205 206 debug("Setting PLL6 to %d\n", DRAM_CLK * 2); 207 clock_set_pll6(DRAM_CLK * 2); 208 209 /* Original dram init code which may come in handy later 210 ******************************************************** 211 clock_set_pll6(use_2channelPLL ? (DRAM_CLK * 2) : 212 (DRAM_CLK / 2), false); 213 214 if ((para->dram_clk <= 400)|((para->dram_tpr8 & 0x1)==0)) { 215 * PLL6 should be 2*CK * 216 * ccm_setup_pll6_ddr_clk(PLL6_DDR_CLK); * 217 ccm_setup_pll6_ddr_clk((1000000 * (para->dram_clk) * 2), 0); 218 } else { 219 * PLL6 should be CK/2 * 220 ccm_setup_pll6_ddr_clk((1000000 * (para->dram_clk) / 2), 1); 221 } 222 223 if (para->dram_tpr13 & (0xf<<18)) { 224 * 225 * bit21:bit18=0001:pll swing 0.4 226 * bit21:bit18=0010:pll swing 0.3 227 * bit21:bit18=0100:pll swing 0.2 228 * bit21:bit18=1000:pll swing 0.1 229 * 230 dram_dbg("DRAM fre extend open !\n"); 231 reg_val=mctl_read_w(CCM_PLL6_DDR_REG); 232 reg_val&=(0x1<<16); 233 reg_val=reg_val>>16; 234 235 if(para->dram_tpr13 & (0x1<<18)) 236 { 237 mctl_write_w(CCM_PLL_BASE + 0x114, 238 (0x3333U|(0x3<<17)|(reg_val<<19)|(0x120U<<20)| 239 (0x2U<<29)|(0x1U<<31))); 240 } 241 else if(para->dram_tpr13 & (0x1<<19)) 242 { 243 mctl_write_w(CCM_PLL_BASE + 0x114, 244 (0x6666U|(0x3U<<17)|(reg_val<<19)|(0xD8U<<20)| 245 (0x2U<<29)|(0x1U<<31))); 246 } 247 else if(para->dram_tpr13 & (0x1<<20)) 248 { 249 mctl_write_w(CCM_PLL_BASE + 0x114, 250 (0x9999U|(0x3U<<17)|(reg_val<<19)|(0x90U<<20)| 251 (0x2U<<29)|(0x1U<<31))); 252 } 253 else if(para->dram_tpr13 & (0x1<<21)) 254 { 255 mctl_write_w(CCM_PLL_BASE + 0x114, 256 (0xccccU|(0x3U<<17)|(reg_val<<19)|(0x48U<<20)| 257 (0x2U<<29)|(0x1U<<31))); 258 } 259 260 //frequency extend open 261 reg_val = mctl_read_w(CCM_PLL6_DDR_REG); 262 reg_val |= ((0x1<<24)|(0x1<<30)); 263 mctl_write_w(CCM_PLL6_DDR_REG, reg_val); 264 265 266 while(mctl_read_w(CCM_PLL6_DDR_REG) & (0x1<<30)); 267 } 268 269 aw_delay(0x20000); //make some delay 270 ******************************************************** 271 */ 272 273 /* assert mctl reset */ 274 clrbits_le32(&ccm->ahb_reset0_cfg, 1 << AHB_RESET_OFFSET_MCTL); 275 /* stop mctl clock */ 276 clrbits_le32(&ccm->ahb_gate0, 1 << AHB_GATE_OFFSET_MCTL); 277 278 sdelay(2000); 279 280 /* deassert mctl reset */ 281 setbits_le32(&ccm->ahb_reset0_cfg, 1 << AHB_RESET_OFFSET_MCTL); 282 /* enable mctl clock */ 283 setbits_le32(&ccm->ahb_gate0, 1 << AHB_GATE_OFFSET_MCTL); 284 285 /* set up the transactions scheduling before enabling the global clk */ 286 mctl_ctl_sched_init(SUNXI_DRAM_CTL0_BASE); 287 mctl_ctl_sched_init(SUNXI_DRAM_CTL1_BASE); 288 sdelay(1000); 289 290 debug("2\n"); 291 292 /* (3 << 12): PLL_DDR */ 293 writel((3 << 12) | (1 << 16), &ccm->dram_clk_cfg); 294 do { 295 debug("Waiting for DRAM_CLK_CFG\n"); 296 sdelay(10000); 297 } while (readl(&ccm->dram_clk_cfg) & (1 << 16)); 298 setbits_le32(&ccm->dram_clk_cfg, (1 << 31)); 299 300 /* TODO: we only support the common case ... i.e. 2*CK */ 301 setbits_le32(&mctl_com->ccr, (1 << 14) | (1 << 30)); 302 writel(2, &mctl_com->rmcr); /* controller clock is PLL6/4 */ 303 304 sdelay(2000); 305 306 /* Original dram init code which may come in handy later 307 ******************************************************** 308 if ((para->dram_clk <= 400) | ((para->dram_tpr8 & 0x1) == 0)) { 309 * PLL6 should be 2*CK * 310 * gating 2 channel pll * 311 reg_val = mctl_read_w(MC_CCR); 312 reg_val |= ((0x1 << 14) | (0x1U << 30)); 313 mctl_write_w(MC_CCR, reg_val); 314 mctl_write_w(MC_RMCR, 0x2); * controller clock use pll6/4 * 315 } else { 316 * enable 2 channel pll * 317 reg_val = mctl_read_w(MC_CCR); 318 reg_val &= ~((0x1 << 14) | (0x1U << 30)); 319 mctl_write_w(MC_CCR, reg_val); 320 mctl_write_w(MC_RMCR, 0x0); * controller clock use pll6 * 321 } 322 323 reg_val = mctl_read_w(MC_CCR); 324 reg_val &= ~((0x1<<15)|(0x1U<<31)); 325 mctl_write_w(MC_CCR, reg_val); 326 aw_delay(20); 327 //aw_delay(0x10); 328 ******************************************************** 329 */ 330 331 clrbits_le32(&mctl_com->ccr, MCTL_CCR_CH0_CLK_EN | MCTL_CCR_CH1_CLK_EN); 332 sdelay(1000); 333 334 setbits_le32(&mctl_com->ccr, MCTL_CCR_CH0_CLK_EN); 335 /* TODO if (para->chan == 2) */ 336 setbits_le32(&mctl_com->ccr, MCTL_CCR_CH1_CLK_EN); 337 } 338 339 static void mctl_com_init(struct dram_sun9i_para *para) 340 { 341 struct sunxi_mctl_com_reg * const mctl_com = 342 (struct sunxi_mctl_com_reg *)SUNXI_DRAM_COM_BASE; 343 344 /* TODO: hard-wired for DDR3 now */ 345 writel(((para->chan == 2) ? MCTL_CR_CHANNEL_DUAL : 346 MCTL_CR_CHANNEL_SINGLE) 347 | MCTL_CR_DRAMTYPE_DDR3 | MCTL_CR_BANK(1) 348 | MCTL_CR_ROW(para->rows) 349 | ((para->bus_width == 32) ? MCTL_CR_BUSW32 : MCTL_CR_BUSW16) 350 | MCTL_CR_PAGE_SIZE(para->page_size) | MCTL_CR_RANK(para->rank), 351 &mctl_com->cr); 352 353 debug("CR: %d\n", readl(&mctl_com->cr)); 354 } 355 356 static u32 mctl_channel_init(u32 ch_index, struct dram_sun9i_para *para) 357 { 358 struct sunxi_mctl_ctl_reg *mctl_ctl; 359 struct sunxi_mctl_phy_reg *mctl_phy; 360 361 u32 CL = 0; 362 u32 CWL = 0; 363 u16 mr[4] = { 0, }; 364 365 #define PS2CYCLES_FLOOR(n) ((n * CONFIG_DRAM_CLK) / 1000000) 366 #define PS2CYCLES_ROUNDUP(n) ((n * CONFIG_DRAM_CLK + 999999) / 1000000) 367 #define NS2CYCLES_FLOOR(n) ((n * CONFIG_DRAM_CLK) / 1000) 368 #define NS2CYCLES_ROUNDUP(n) ((n * CONFIG_DRAM_CLK + 999) / 1000) 369 #define MAX(a, b) ((a) > (b) ? (a) : (b)) 370 371 /* 372 * Convert the values to cycle counts (nCK) from what is provided 373 * by the definition of each speed bin. 374 */ 375 /* const u32 tREFI = NS2CYCLES_FLOOR(para->tREFI); */ 376 const u32 tREFI = NS2CYCLES_FLOOR(para->tREFI); 377 const u32 tRFC = NS2CYCLES_ROUNDUP(para->tRFC); 378 const u32 tRCD = PS2CYCLES_ROUNDUP(para->tRCD); 379 const u32 tRP = PS2CYCLES_ROUNDUP(para->tRP); 380 const u32 tRC = PS2CYCLES_ROUNDUP(para->tRC); 381 const u32 tRAS = PS2CYCLES_ROUNDUP(para->tRAS); 382 383 /* command and address timing */ 384 const u32 tDLLK = para->tDLLK; 385 const u32 tRTP = MAX(para->tRTP.ck, PS2CYCLES_ROUNDUP(para->tRTP.ps)); 386 const u32 tWTR = MAX(para->tWTR.ck, PS2CYCLES_ROUNDUP(para->tWTR.ps)); 387 const u32 tWR = NS2CYCLES_FLOOR(para->tWR); 388 const u32 tMRD = para->tMRD; 389 const u32 tMOD = MAX(para->tMOD.ck, PS2CYCLES_ROUNDUP(para->tMOD.ps)); 390 const u32 tCCD = para->tCCD; 391 const u32 tRRD = MAX(para->tRRD.ck, PS2CYCLES_ROUNDUP(para->tRRD.ps)); 392 const u32 tFAW = PS2CYCLES_ROUNDUP(para->tFAW); 393 394 /* calibration timings */ 395 /* const u32 tZQinit = MAX(para->tZQinit.ck, 396 PS2CYCLES_ROUNDUP(para->tZQinit.ps)); */ 397 const u32 tZQoper = MAX(para->tZQoper.ck, 398 PS2CYCLES_ROUNDUP(para->tZQoper.ps)); 399 const u32 tZQCS = MAX(para->tZQCS.ck, 400 PS2CYCLES_ROUNDUP(para->tZQCS.ps)); 401 402 /* reset timing */ 403 /* const u32 tXPR = MAX(para->tXPR.ck, 404 PS2CYCLES_ROUNDUP(para->tXPR.ps)); */ 405 406 /* power-down timings */ 407 const u32 tXP = MAX(para->tXP.ck, PS2CYCLES_ROUNDUP(para->tXP.ps)); 408 const u32 tXPDLL = MAX(para->tXPDLL.ck, 409 PS2CYCLES_ROUNDUP(para->tXPDLL.ps)); 410 const u32 tCKE = MAX(para->tCKE.ck, PS2CYCLES_ROUNDUP(para->tCKE.ps)); 411 412 /* 413 * self-refresh timings (keep below power-down timings, as tCKESR 414 * needs to be calculated based on the nCK value of tCKE) 415 */ 416 const u32 tXS = MAX(para->tXS.ck, PS2CYCLES_ROUNDUP(para->tXS.ps)); 417 const u32 tXSDLL = para->tXSDLL; 418 const u32 tCKSRE = MAX(para->tCKSRE.ck, 419 PS2CYCLES_ROUNDUP(para->tCKSRE.ps)); 420 const u32 tCKESR = tCKE + 1; 421 const u32 tCKSRX = MAX(para->tCKSRX.ck, 422 PS2CYCLES_ROUNDUP(para->tCKSRX.ps)); 423 424 /* write leveling timings */ 425 const u32 tWLMRD = para->tWLMRD; 426 /* const u32 tWLDQSEN = para->tWLDQSEN; */ 427 const u32 tWLO = PS2CYCLES_FLOOR(para->tWLO); 428 /* const u32 tWLOE = PS2CYCLES_FLOOR(para->tWLOE); */ 429 430 const u32 tRASmax = tREFI * 9; 431 int i; 432 433 for (i = 0; i < para->cl_cwl_numentries; ++i) { 434 const u32 tCK = 1000000 / CONFIG_DRAM_CLK; 435 436 if ((para->cl_cwl_table[i].tCKmin <= tCK) && 437 (tCK < para->cl_cwl_table[i].tCKmax)) { 438 CL = para->cl_cwl_table[i].CL; 439 CWL = para->cl_cwl_table[i].CWL; 440 441 debug("found CL/CWL: CL = %d, CWL = %d\n", CL, CWL); 442 break; 443 } 444 } 445 446 if ((CL == 0) && (CWL == 0)) { 447 printf("failed to find valid CL/CWL for operating point %d MHz\n", 448 CONFIG_DRAM_CLK); 449 return 0; 450 } 451 452 if (ch_index == 0) { 453 mctl_ctl = (struct sunxi_mctl_ctl_reg *)SUNXI_DRAM_CTL0_BASE; 454 mctl_phy = (struct sunxi_mctl_phy_reg *)SUNXI_DRAM_PHY0_BASE; 455 } else { 456 mctl_ctl = (struct sunxi_mctl_ctl_reg *)SUNXI_DRAM_CTL1_BASE; 457 mctl_phy = (struct sunxi_mctl_phy_reg *)SUNXI_DRAM_PHY1_BASE; 458 } 459 460 if (para->dram_type == DRAM_TYPE_DDR3) { 461 mr[0] = DDR3_MR0_PPD_FAST_EXIT | DDR3_MR0_WR(tWR) | 462 DDR3_MR0_CL(CL); 463 mr[1] = DDR3_MR1_RTT120OHM; 464 mr[2] = DDR3_MR2_TWL(CWL); 465 mr[3] = 0; 466 467 /* 468 * DRAM3 initialisation requires holding CKE LOW for 469 * at least 500us prior to starting the initialisation 470 * sequence and at least 10ns after driving CKE HIGH 471 * before the initialisation sequence may be started). 472 * 473 * Refer to Micron document "TN-41-07: DDR3 Power-Up, 474 * Initialization, and Reset DDR3 Initialization 475 * Routine" for details). 476 */ 477 writel(MCTL_INIT0_POST_CKE_x1024(1) | 478 MCTL_INIT0_PRE_CKE_x1024( 479 (500 * CONFIG_DRAM_CLK + 1023) / 1024), /* 500us */ 480 &mctl_ctl->init[0]); 481 writel(MCTL_INIT1_DRAM_RSTN_x1024(1), 482 &mctl_ctl->init[1]); 483 /* INIT2 is not used for DDR3 */ 484 writel(MCTL_INIT3_MR(mr[0]) | MCTL_INIT3_EMR(mr[1]), 485 &mctl_ctl->init[3]); 486 writel(MCTL_INIT4_EMR2(mr[2]) | MCTL_INIT4_EMR3(mr[3]), 487 &mctl_ctl->init[4]); 488 writel(MCTL_INIT5_DEV_ZQINIT_x32(512 / 32), /* 512 cycles */ 489 &mctl_ctl->init[5]); 490 } else { 491 /* !!! UNTESTED !!! */ 492 /* 493 * LPDDR2 and/or LPDDR3 require a 200us minimum delay 494 * after driving CKE HIGH in the initialisation sequence. 495 */ 496 writel(MCTL_INIT0_POST_CKE_x1024( 497 (200 * CONFIG_DRAM_CLK + 1023) / 1024), 498 &mctl_ctl->init[0]); 499 writel(MCTL_INIT1_DRAM_RSTN_x1024(1), 500 &mctl_ctl->init[1]); 501 writel(MCTL_INIT2_IDLE_AFTER_RESET_x32( 502 (CONFIG_DRAM_CLK + 31) / 32) /* 1us */ 503 | MCTL_INIT2_MIN_STABLE_CLOCK_x1(5), /* 5 cycles */ 504 &mctl_ctl->init[2]); 505 writel(MCTL_INIT3_MR(mr[1]) | MCTL_INIT3_EMR(mr[2]), 506 &mctl_ctl->init[3]); 507 writel(MCTL_INIT4_EMR2(mr[3]), 508 &mctl_ctl->init[4]); 509 writel(MCTL_INIT5_DEV_ZQINIT_x32( 510 (CONFIG_DRAM_CLK + 31) / 32) /* 1us */ 511 | MCTL_INIT5_MAX_AUTO_INIT_x1024( 512 (10 * CONFIG_DRAM_CLK + 1023) / 1024), 513 &mctl_ctl->init[5]); 514 } 515 516 /* (DDR3) We always use a burst-length of 8. */ 517 #define MCTL_BL 8 518 /* wr2pre: WL + BL/2 + tWR */ 519 #define WR2PRE (MCTL_BL/2 + CWL + tWTR) 520 /* wr2rd = CWL + BL/2 + tWTR */ 521 #define WR2RD (MCTL_BL/2 + CWL + tWTR) 522 /* 523 * rd2wr = RL + BL/2 + 2 - WL (for DDR3) 524 * rd2wr = RL + BL/2 + RU(tDQSCKmax/tCK) + 1 - WL (for LPDDR2/LPDDR3) 525 */ 526 #define RD2WR (CL + MCTL_BL/2 + 2 - CWL) 527 #define MCTL_PHY_TRTW 0 528 #define MCTL_PHY_TRTODT 0 529 530 #define MCTL_DIV2(n) ((n + 1)/2) 531 #define MCTL_DIV32(n) (n/32) 532 #define MCTL_DIV1024(n) (n/1024) 533 534 writel((MCTL_DIV2(WR2PRE) << 24) | (MCTL_DIV2(tFAW) << 16) | 535 (MCTL_DIV1024(tRASmax) << 8) | (MCTL_DIV2(tRAS) << 0), 536 &mctl_ctl->dramtmg[0]); 537 writel((MCTL_DIV2(tXP) << 16) | (MCTL_DIV2(tRTP) << 8) | 538 (MCTL_DIV2(tRC) << 0), 539 &mctl_ctl->dramtmg[1]); 540 writel((MCTL_DIV2(CWL) << 24) | (MCTL_DIV2(CL) << 16) | 541 (MCTL_DIV2(RD2WR) << 8) | (MCTL_DIV2(WR2RD) << 0), 542 &mctl_ctl->dramtmg[2]); 543 /* 544 * Note: tMRW is located at bit 16 (and up) in DRAMTMG3... 545 * this is only relevant for LPDDR2/LPDDR3 546 */ 547 writel((MCTL_DIV2(tMRD) << 12) | (MCTL_DIV2(tMOD) << 0), 548 &mctl_ctl->dramtmg[3]); 549 writel((MCTL_DIV2(tRCD) << 24) | (MCTL_DIV2(tCCD) << 16) | 550 (MCTL_DIV2(tRRD) << 8) | (MCTL_DIV2(tRP) << 0), 551 &mctl_ctl->dramtmg[4]); 552 writel((MCTL_DIV2(tCKSRX) << 24) | (MCTL_DIV2(tCKSRE) << 16) | 553 (MCTL_DIV2(tCKESR) << 8) | (MCTL_DIV2(tCKE) << 0), 554 &mctl_ctl->dramtmg[5]); 555 556 /* These timings are relevant for LPDDR2/LPDDR3 only */ 557 /* writel((MCTL_TCKDPDE << 24) | (MCTL_TCKDPX << 16) | 558 (MCTL_TCKCSX << 0), &mctl_ctl->dramtmg[6]); */ 559 560 /* printf("DRAMTMG7 reset value: 0x%x\n", 561 readl(&mctl_ctl->dramtmg[7])); */ 562 /* DRAMTMG7 reset value: 0x202 */ 563 /* DRAMTMG7 should contain t_ckpde and t_ckpdx: check reset values!!! */ 564 /* printf("DRAMTMG8 reset value: 0x%x\n", 565 readl(&mctl_ctl->dramtmg[8])); */ 566 /* DRAMTMG8 reset value: 0x44 */ 567 568 writel((MCTL_DIV32(tXSDLL) << 0), &mctl_ctl->dramtmg[8]); 569 570 writel((MCTL_DIV32(tREFI) << 16) | (MCTL_DIV2(tRFC) << 0), 571 &mctl_ctl->rfshtmg); 572 573 if (para->dram_type == DRAM_TYPE_DDR3) { 574 writel((2 << 24) | ((MCTL_DIV2(CL) - 2) << 16) | 575 (1 << 8) | ((MCTL_DIV2(CWL) - 2) << 0), 576 &mctl_ctl->dfitmg[0]); 577 } else { 578 /* TODO */ 579 } 580 581 /* TODO: handle the case of the write latency domain going to 0 ... */ 582 583 /* 584 * Disable dfi_init_complete_en (the triggering of the SDRAM 585 * initialisation when the PHY initialisation completes). 586 */ 587 clrbits_le32(&mctl_ctl->dfimisc, MCTL_DFIMISC_DFI_INIT_COMPLETE_EN); 588 /* Disable the automatic generation of DLL calibration requests */ 589 setbits_le32(&mctl_ctl->dfiupd[0], MCTL_DFIUPD0_DIS_AUTO_CTRLUPD); 590 591 /* A80-Q7: 2T, 1 rank, DDR3, full-32bit-DQ */ 592 /* TODO: make 2T and BUSWIDTH configurable */ 593 writel(MCTL_MSTR_DEVICETYPE(para->dram_type) | 594 MCTL_MSTR_BURSTLENGTH(para->dram_type) | 595 MCTL_MSTR_ACTIVERANKS(para->rank) | 596 MCTL_MSTR_2TMODE | MCTL_MSTR_BUSWIDTH32, 597 &mctl_ctl->mstr); 598 599 if (para->dram_type == DRAM_TYPE_DDR3) { 600 writel(MCTL_ZQCTRL0_TZQCL(MCTL_DIV2(tZQoper)) | 601 (MCTL_DIV2(tZQCS)), &mctl_ctl->zqctrl[0]); 602 /* 603 * TODO: is the following really necessary as the bottom 604 * half should already be 0x100 and the upper half should 605 * be ignored for a DDR3 device??? 606 */ 607 writel(MCTL_ZQCTRL1_TZQSI_x1024(0x100), 608 &mctl_ctl->zqctrl[1]); 609 } else { 610 writel(MCTL_ZQCTRL0_TZQCL(0x200) | MCTL_ZQCTRL0_TZQCS(0x40), 611 &mctl_ctl->zqctrl[0]); 612 writel(MCTL_ZQCTRL1_TZQRESET(0x28) | 613 MCTL_ZQCTRL1_TZQSI_x1024(0x100), 614 &mctl_ctl->zqctrl[1]); 615 } 616 617 /* Assert dfi_init_complete signal */ 618 setbits_le32(&mctl_ctl->dfimisc, MCTL_DFIMISC_DFI_INIT_COMPLETE_EN); 619 /* Disable auto-refresh */ 620 setbits_le32(&mctl_ctl->rfshctl3, MCTL_RFSHCTL3_DIS_AUTO_REFRESH); 621 622 /* PHY initialisation */ 623 624 /* TODO: make 2T and 8-bank mode configurable */ 625 writel(MCTL_PHY_DCR_BYTEMASK | MCTL_PHY_DCR_2TMODE | 626 MCTL_PHY_DCR_DDR8BNK | MCTL_PHY_DRAMMODE_DDR3, 627 &mctl_phy->dcr); 628 629 /* For LPDDR2 or LPDDR3, set DQSGX to 0 before training. */ 630 if (para->dram_type != DRAM_TYPE_DDR3) 631 clrbits_le32(&mctl_phy->dsgcr, (3 << 6)); 632 633 writel(mr[0], &mctl_phy->mr0); 634 writel(mr[1], &mctl_phy->mr1); 635 writel(mr[2], &mctl_phy->mr2); 636 writel(mr[3], &mctl_phy->mr3); 637 638 /* 639 * The DFI PHY is running at full rate. We thus use the actual 640 * timings in clock cycles here. 641 */ 642 writel((tRC << 26) | (tRRD << 22) | (tRAS << 16) | 643 (tRCD << 12) | (tRP << 8) | (tWTR << 4) | (tRTP << 0), 644 &mctl_phy->dtpr[0]); 645 writel((tMRD << 0) | ((tMOD - 12) << 2) | (tFAW << 5) | 646 (tRFC << 11) | (tWLMRD << 20) | (tWLO << 26), 647 &mctl_phy->dtpr[1]); 648 writel((tXS << 0) | (MAX(tXP, tXPDLL) << 10) | 649 (tCKE << 15) | (tDLLK << 19) | 650 (MCTL_PHY_TRTODT << 29) | (MCTL_PHY_TRTW << 30) | 651 (((tCCD - 4) & 0x1) << 31), 652 &mctl_phy->dtpr[2]); 653 654 /* tDQSCK and tDQSCKmax are used LPDDR2/LPDDR3 */ 655 /* writel((tDQSCK << 0) | (tDQSCKMAX << 3), &mctl_phy->dtpr[3]); */ 656 657 /* 658 * We use the same values used by Allwinner's Boot0 for the PTR 659 * (PHY timing register) configuration that is tied to the PHY 660 * implementation. 661 */ 662 writel(0x42C21590, &mctl_phy->ptr[0]); 663 writel(0xD05612C0, &mctl_phy->ptr[1]); 664 if (para->dram_type == DRAM_TYPE_DDR3) { 665 const unsigned int tdinit0 = 500 * CONFIG_DRAM_CLK; /* 500us */ 666 const unsigned int tdinit1 = (360 * CONFIG_DRAM_CLK + 999) / 667 1000; /* 360ns */ 668 const unsigned int tdinit2 = 200 * CONFIG_DRAM_CLK; /* 200us */ 669 const unsigned int tdinit3 = CONFIG_DRAM_CLK; /* 1us */ 670 671 writel((tdinit1 << 20) | tdinit0, &mctl_phy->ptr[3]); 672 writel((tdinit3 << 18) | tdinit2, &mctl_phy->ptr[4]); 673 } else { 674 /* LPDDR2 or LPDDR3 */ 675 const unsigned int tdinit0 = (100 * CONFIG_DRAM_CLK + 999) / 676 1000; /* 100ns */ 677 const unsigned int tdinit1 = 200 * CONFIG_DRAM_CLK; /* 200us */ 678 const unsigned int tdinit2 = 22 * CONFIG_DRAM_CLK; /* 11us */ 679 const unsigned int tdinit3 = 2 * CONFIG_DRAM_CLK; /* 2us */ 680 681 writel((tdinit1 << 20) | tdinit0, &mctl_phy->ptr[3]); 682 writel((tdinit3 << 18) | tdinit2, &mctl_phy->ptr[4]); 683 } 684 685 /* TEST ME */ 686 writel(0x00203131, &mctl_phy->acmdlr); 687 688 /* TODO: can we enable this for 2 ranks, even when we don't know yet */ 689 writel(MCTL_DTCR_DEFAULT | MCTL_DTCR_RANKEN(para->rank), 690 &mctl_phy->dtcr); 691 692 /* TODO: half width */ 693 debug("DX2GCR0 reset: 0x%x\n", readl(&mctl_phy->dx[2].gcr[0])); 694 writel(0x7C000285, &mctl_phy->dx[2].gcr[0]); 695 writel(0x7C000285, &mctl_phy->dx[3].gcr[0]); 696 697 clrsetbits_le32(&mctl_phy->zq[0].pr, 0xff, 698 (CONFIG_DRAM_ZQ >> 0) & 0xff); /* CK/CA */ 699 clrsetbits_le32(&mctl_phy->zq[1].pr, 0xff, 700 (CONFIG_DRAM_ZQ >> 8) & 0xff); /* DX0/DX1 */ 701 clrsetbits_le32(&mctl_phy->zq[2].pr, 0xff, 702 (CONFIG_DRAM_ZQ >> 16) & 0xff); /* DX2/DX3 */ 703 704 /* TODO: make configurable & implement non-ODT path */ 705 if (1) { 706 int lane; 707 for (lane = 0; lane < 4; ++lane) { 708 clrbits_le32(&mctl_phy->dx[lane].gcr[2], 0xffff); 709 clrbits_le32(&mctl_phy->dx[lane].gcr[3], 710 (0x3<<12) | (0x3<<4)); 711 } 712 } else { 713 /* TODO: check */ 714 int lane; 715 for (lane = 0; lane < 4; ++lane) { 716 clrsetbits_le32(&mctl_phy->dx[lane].gcr[2], 0xffff, 717 0xaaaa); 718 if (para->dram_type == DRAM_TYPE_DDR3) 719 setbits_le32(&mctl_phy->dx[lane].gcr[3], 720 (0x3<<12) | (0x3<<4)); 721 else 722 setbits_le32(&mctl_phy->dx[lane].gcr[3], 723 0x00000012); 724 } 725 } 726 727 writel(0x04058D02, &mctl_phy->zq[0].cr); /* CK/CA */ 728 writel(0x04058D02, &mctl_phy->zq[1].cr); /* DX0/DX1 */ 729 writel(0x04058D02, &mctl_phy->zq[2].cr); /* DX2/DX3 */ 730 731 /* Disable auto-refresh prior to data training */ 732 setbits_le32(&mctl_ctl->rfshctl3, MCTL_RFSHCTL3_DIS_AUTO_REFRESH); 733 734 setbits_le32(&mctl_phy->dsgcr, 0xf << 24); /* unclear what this is... */ 735 /* TODO: IODDRM (IO DDR-MODE) for DDR3L */ 736 clrsetbits_le32(&mctl_phy->pgcr[1], 737 MCTL_PGCR1_ZCKSEL_MASK, 738 MCTL_PGCR1_IODDRM_DDR3 | MCTL_PGCR1_INHVT_EN); 739 740 setbits_le32(&mctl_phy->pllcr, 0x3 << 19); /* PLL frequency select */ 741 /* TODO: single-channel PLL mode??? missing */ 742 setbits_le32(&mctl_phy->pllcr, 743 MCTL_PLLGCR_PLL_BYPASS | MCTL_PLLGCR_PLL_POWERDOWN); 744 /* setbits_le32(&mctl_phy->pir, MCTL_PIR_PLL_BYPASS); included below */ 745 746 /* Disable VT compensation */ 747 clrbits_le32(&mctl_phy->pgcr[0], 0x3f); 748 749 /* TODO: "other" PLL mode ... 0x20000 seems to be the PLL Bypass */ 750 if (para->dram_type == DRAM_TYPE_DDR3) 751 clrsetbits_le32(&mctl_phy->pir, MCTL_PIR_MASK, 0x20df3); 752 else 753 clrsetbits_le32(&mctl_phy->pir, MCTL_PIR_MASK, 0x2c573); 754 755 sdelay(10000); /* XXX necessary? */ 756 757 /* Wait for the INIT bit to clear itself... */ 758 while ((readl(&mctl_phy->pir) & MCTL_PIR_INIT) != MCTL_PIR_INIT) { 759 /* not done yet -- keep spinning */ 760 debug("MCTL_PIR_INIT not set\n"); 761 sdelay(1000); 762 /* TODO: implement timeout */ 763 } 764 765 /* TODO: not used --- there's a "2rank debug" section here */ 766 767 /* Original dram init code which may come in handy later 768 ******************************************************** 769 * LPDDR2 and LPDDR3 * 770 if ((para->dram_type) == 6 || (para->dram_type) == 7) { 771 reg_val = mctl_read_w(P0_DSGCR + ch_offset); 772 reg_val &= (~(0x3<<6)); * set DQSGX to 1 * 773 reg_val |= (0x1<<6); * dqs gate extend * 774 mctl_write_w(P0_DSGCR + ch_offset, reg_val); 775 dram_dbg("DQS Gate Extend Enable!\n", ch_index); 776 } 777 778 * Disable ZCAL after initial--for nand dma debug--20140330 by YSZ * 779 if (para->dram_tpr13 & (0x1<<31)) { 780 reg_val = mctl_read_w(P0_ZQ0CR + ch_offset); 781 reg_val |= (0x7<<11); 782 mctl_write_w(P0_ZQ0CR + ch_offset, reg_val); 783 } 784 ******************************************************** 785 */ 786 787 /* 788 * TODO: more 2-rank support 789 * (setting the "dqs gate delay to average between 2 rank") 790 */ 791 792 /* check if any errors are set */ 793 if (readl(&mctl_phy->pgsr[0]) & MCTL_PGSR0_ERRORS) { 794 debug("Channel %d unavailable!\n", ch_index); 795 return 0; 796 } else{ 797 /* initial OK */ 798 debug("Channel %d OK!\n", ch_index); 799 /* return 1; */ 800 } 801 802 while ((readl(&mctl_ctl->stat) & 0x1) != 0x1) { 803 debug("Waiting for INIT to be done (controller to come up into 'normal operating' mode\n"); 804 sdelay(100000); 805 /* init not done */ 806 /* TODO: implement time-out */ 807 } 808 debug("done\n"); 809 810 /* "DDR is controller by contoller" */ 811 clrbits_le32(&mctl_phy->pgcr[3], (1 << 25)); 812 813 /* TODO: is the following necessary? */ 814 debug("DFIMISC before writing 0: 0x%x\n", readl(&mctl_ctl->dfimisc)); 815 writel(0, &mctl_ctl->dfimisc); 816 817 /* Enable auto-refresh */ 818 clrbits_le32(&mctl_ctl->rfshctl3, MCTL_RFSHCTL3_DIS_AUTO_REFRESH); 819 820 debug("channel_init complete\n"); 821 return 1; 822 } 823 824 signed int DRAMC_get_dram_size(void) 825 { 826 struct sunxi_mctl_com_reg * const mctl_com = 827 (struct sunxi_mctl_com_reg *)SUNXI_DRAM_COM_BASE; 828 829 unsigned int reg_val; 830 unsigned int dram_size; 831 unsigned int temp; 832 833 reg_val = readl(&mctl_com->cr); 834 835 temp = (reg_val >> 8) & 0xf; /* page size code */ 836 dram_size = (temp - 6); /* (1 << dram_size) * 512Bytes */ 837 838 temp = (reg_val >> 4) & 0xf; /* row width code */ 839 dram_size += (temp + 1); /* (1 << dram_size) * 512Bytes */ 840 841 temp = (reg_val >> 2) & 0x3; /* bank number code */ 842 dram_size += (temp + 2); /* (1 << dram_size) * 512Bytes */ 843 844 temp = reg_val & 0x3; /* rank number code */ 845 dram_size += temp; /* (1 << dram_size) * 512Bytes */ 846 847 temp = (reg_val >> 19) & 0x1; /* channel number code */ 848 dram_size += temp; /* (1 << dram_size) * 512Bytes */ 849 850 dram_size = dram_size - 11; /* (1 << dram_size) MBytes */ 851 852 return 1 << dram_size; 853 } 854 855 unsigned long sunxi_dram_init(void) 856 { 857 struct sunxi_mctl_com_reg * const mctl_com = 858 (struct sunxi_mctl_com_reg *)SUNXI_DRAM_COM_BASE; 859 860 struct dram_sun9i_cl_cwl_timing cl_cwl[] = { 861 { .CL = 5, .CWL = 5, .tCKmin = 3000, .tCKmax = 3300 }, 862 { .CL = 6, .CWL = 5, .tCKmin = 2500, .tCKmax = 3300 }, 863 { .CL = 8, .CWL = 6, .tCKmin = 1875, .tCKmax = 2500 }, 864 { .CL = 10, .CWL = 7, .tCKmin = 1500, .tCKmax = 1875 }, 865 { .CL = 11, .CWL = 8, .tCKmin = 1250, .tCKmax = 1500 } 866 }; 867 868 /* Set initial parameters, these get modified by the autodetect code */ 869 struct dram_sun9i_para para = { 870 .dram_type = DRAM_TYPE_DDR3, 871 .bus_width = 32, 872 .chan = 2, 873 .rank = 1, 874 /* .rank = 2, */ 875 .page_size = 4096, 876 /* .rows = 16, */ 877 .rows = 15, 878 879 /* CL/CWL table for the speed bin */ 880 .cl_cwl_table = cl_cwl, 881 .cl_cwl_numentries = sizeof(cl_cwl) / 882 sizeof(struct dram_sun9i_cl_cwl_timing), 883 884 /* timings */ 885 .tREFI = 7800, /* 7.8us (up to 85 degC) */ 886 .tRFC = 260, /* 260ns for 4GBit devices */ 887 /* 350ns @ 8GBit */ 888 889 .tRCD = 13750, 890 .tRP = 13750, 891 .tRC = 48750, 892 .tRAS = 35000, 893 894 .tDLLK = 512, 895 .tRTP = { .ck = 4, .ps = 7500 }, 896 .tWTR = { .ck = 4, .ps = 7500 }, 897 .tWR = 15, 898 .tMRD = 4, 899 .tMOD = { .ck = 12, .ps = 15000 }, 900 .tCCD = 4, 901 .tRRD = { .ck = 4, .ps = 7500 }, 902 .tFAW = 40, 903 904 /* calibration timing */ 905 /* .tZQinit = { .ck = 512, .ps = 640000 }, */ 906 .tZQoper = { .ck = 256, .ps = 320000 }, 907 .tZQCS = { .ck = 64, .ps = 80000 }, 908 909 /* reset timing */ 910 /* .tXPR = { .ck = 5, .ps = 10000 }, */ 911 912 /* self-refresh timings */ 913 .tXS = { .ck = 5, .ps = 10000 }, 914 .tXSDLL = 512, 915 .tCKSRE = { .ck = 5, .ps = 10000 }, 916 .tCKSRX = { .ck = 5, .ps = 10000 }, 917 918 /* power-down timings */ 919 .tXP = { .ck = 3, .ps = 6000 }, 920 .tXPDLL = { .ck = 10, .ps = 24000 }, 921 .tCKE = { .ck = 3, .ps = 5000 }, 922 923 /* write leveling timings */ 924 .tWLMRD = 40, 925 /* .tWLDQSEN = 25, */ 926 .tWLO = 7500, 927 /* .tWLOE = 2000, */ 928 }; 929 930 /* 931 * Disable A80 internal 240 ohm resistor. 932 * 933 * This code sequence is adapated from Allwinner's Boot0 (see 934 * https://github.com/allwinner-zh/bootloader.git), as there 935 * is no documentation for these two registers in the R_PRCM 936 * block. 937 */ 938 setbits_le32(SUNXI_PRCM_BASE + 0x1e0, (0x3 << 8)); 939 writel(0, SUNXI_PRCM_BASE + 0x1e8); 940 941 mctl_sys_init(); 942 943 if (!mctl_channel_init(0, ¶)) 944 return 0; 945 946 /* dual-channel */ 947 if (!mctl_channel_init(1, ¶)) { 948 /* disable channel 1 */ 949 clrsetbits_le32(&mctl_com->cr, MCTL_CR_CHANNEL_MASK, 950 MCTL_CR_CHANNEL_SINGLE); 951 /* disable channel 1 global clock */ 952 clrbits_le32(&mctl_com->cr, MCTL_CCR_CH1_CLK_EN); 953 } 954 955 mctl_com_init(¶); 956 957 /* return the proper RAM size */ 958 return DRAMC_get_dram_size() << 20; 959 } 960