From 78b25eb96baef7da2f5481572a6df2b88ee2b3d4 Mon Sep 17 00:00:00 2001 From: Angel Pons <th3fanbus@gmail.com> Date: Sun, 8 May 2022 00:11:29 +0200 Subject: [PATCH 50/65] haswell NRI: Add REUT I/O test library Implement a library to run I/O tests using the REUT hardware. Change-Id: Id7b207cd0a3989ddd23c88c6b1f0cfa79d2c861f Signed-off-by: Angel Pons <th3fanbus@gmail.com> --- .../intel/haswell/native_raminit/Makefile.mk | 1 + .../haswell/native_raminit/raminit_native.h | 110 +++ .../haswell/native_raminit/reg_structs.h | 121 +++ .../intel/haswell/native_raminit/testing_io.c | 744 ++++++++++++++++++ .../intel/haswell/registers/mchbar.h | 30 + 5 files changed, 1006 insertions(+) create mode 100644 src/northbridge/intel/haswell/native_raminit/testing_io.c diff --git a/src/northbridge/intel/haswell/native_raminit/Makefile.mk b/src/northbridge/intel/haswell/native_raminit/Makefile.mk index 8d7d4e4db0..6e1b365602 100644 --- a/src/northbridge/intel/haswell/native_raminit/Makefile.mk +++ b/src/northbridge/intel/haswell/native_raminit/Makefile.mk @@ -12,4 +12,5 @@ romstage-y += raminit_native.c romstage-y += reut.c romstage-y += setup_wdb.c romstage-y += spd_bitmunching.c +romstage-y += testing_io.c romstage-y += timings_refresh.c diff --git a/src/northbridge/intel/haswell/native_raminit/raminit_native.h b/src/northbridge/intel/haswell/native_raminit/raminit_native.h index 4d9487d79c..f029e7f076 100644 --- a/src/northbridge/intel/haswell/native_raminit/raminit_native.h +++ b/src/northbridge/intel/haswell/native_raminit/raminit_native.h @@ -59,6 +59,88 @@ enum { REUT_MODE_NOP = 3, /* Normal operation mode */ }; +/* REUT error counter control */ +enum { + COUNT_ERRORS_PER_CHANNEL = 0, + COUNT_ERRORS_PER_LANE = 1, + COUNT_ERRORS_PER_BYTE_GROUP = 2, + COUNT_ERRORS_PER_CHUNK = 3, +}; + +enum wdb_dq_pattern { + BASIC_VA = 0, + SEGMENT_WDB, + CADB, + TURN_AROUND, + LMN_VA, + TURN_AROUND_WR, + TURN_AROUND_ODT, + RD_RD_TA, + RD_RD_TA_ALL, +}; + +enum reut_cmd_pat { + PAT_WR_RD, + PAT_WR, + PAT_RD, + PAT_RD_WR_TA, + PAT_WR_RD_TA, + PAT_ODT_TA, +}; + +/* REUT subsequence types (B = Base, O = Offset) */ +enum { + SUBSEQ_B_RD = 0 << 22, + SUBSEQ_B_WR = 1 << 22, + SUBSEQ_B_RD_WR = 2 << 22, + SUBSEQ_B_WR_RD = 3 << 22, + SUBSEQ_O_RD = 4 << 22, + SUBSEQ_O_WR = 5 << 22, +}; + +/* REUT mux control */ +enum { + REUT_MUX_LMN = 0, + REUT_MUX_BTBUFFER = 1, + REUT_MUX_LFSR = 2, +}; + +/* Increment scale */ +enum { + SCALE_LOGARITHM = 0, + SCALE_LINEAR = 1, +}; + +enum test_stop { + NSOE = 0, /* Never stop on error */ + NTHSOE = 1, /* Stop on the nth error (we use n = 1) */ + ABGSOE = 2, /* Stop on all byte groups error */ + ALSOE = 3, /* Stop on all lanes error */ +}; + +struct wdb_pat { + uint32_t start_ptr; /* Starting pointer in WDB */ + uint32_t stop_ptr; /* Stopping pointer in WDB */ + uint16_t inc_rate; /* How quickly the WDB walks through cachelines */ + uint8_t dq_pattern; /* DQ pattern to use (see enum wdb_dq_pattern above) */ +}; + +struct reut_pole { + uint16_t start; + uint16_t stop; + uint16_t order; + uint32_t inc_rate; + uint16_t inc_val; + bool wrap_trigger; +}; + +struct reut_box { + struct reut_pole rank; + struct reut_pole bank; + struct reut_pole row; + struct reut_pole col; +}; + enum command_training_iteration { CT_ITERATION_CLOCK = 0, CT_ITERATION_CMD_NORTH, @@ -200,6 +282,10 @@ struct sysinfo { uint16_t mr1[NUM_CHANNELS][NUM_SLOTS]; uint16_t mr2[NUM_CHANNELS][NUM_SLOTS]; uint16_t mr3[NUM_CHANNELS][NUM_SLOTS]; + + uint8_t dq_pat; + + uint8_t dq_pat_lc; }; static inline bool is_hsw_ult(void) @@ -341,6 +427,30 @@ void write_wdb_va_pat( void program_wdb_lfsr(const struct sysinfo *ctrl, bool cleanup); void setup_wdb(const struct sysinfo *ctrl); +void program_seq_addr(uint8_t channel, const struct reut_box *reut_addr, bool log_seq_addr); +void program_loop_count(const struct sysinfo *ctrl, uint8_t channel, uint8_t lc_exp); + +void setup_io_test( + struct sysinfo *ctrl, + uint8_t chanmask, + enum reut_cmd_pat cmd_pat, + uint16_t num_cl, + uint8_t lc, + const struct reut_box *reut_addr, + enum test_stop soe, + const struct wdb_pat *pat, + uint8_t en_cadb, + uint8_t subseq_wait); + +void setup_io_test_cadb(struct sysinfo *ctrl, uint8_t chanmask, uint8_t lc, enum test_stop soe); +void setup_io_test_basic_va(struct sysinfo *ctrl, uint8_t chm, uint8_t lc, enum test_stop soe); +void setup_io_test_mpr(struct sysinfo *ctrl, uint8_t chanmask, uint8_t lc, enum test_stop soe); + +uint8_t select_reut_ranks(struct sysinfo *ctrl, uint8_t channel, uint8_t rankmask); + +void run_mpr_io_test(bool clear_errors); +uint8_t run_io_test(struct sysinfo *ctrl, uint8_t chanmask, uint8_t dq_pat, bool clear_errors); + uint8_t get_rx_bias(const struct sysinfo *ctrl); uint8_t get_tCWL(uint32_t mem_clock_mhz); diff --git a/src/northbridge/intel/haswell/native_raminit/reg_structs.h b/src/northbridge/intel/haswell/native_raminit/reg_structs.h index 7aa8d8c8b2..b943259b91 100644 --- a/src/northbridge/intel/haswell/native_raminit/reg_structs.h +++ b/src/northbridge/intel/haswell/native_raminit/reg_structs.h @@ -347,6 +347,54 @@ union reut_pat_cl_mux_lmn_reg { uint32_t raw; }; +union reut_err_ctl_reg { + struct __packed { + uint32_t stop_on_nth_error : 6; // Bits 5:0 + uint32_t : 6; // Bits 11:6 + uint32_t stop_on_error_control : 2; // Bits 13:12 + uint32_t : 2; // Bits 15:14 + uint32_t selective_err_enable_chunk : 8; // Bits 23:16 + uint32_t selective_err_enable_cacheline : 8; // Bits 31:24 + }; + uint32_t raw; +}; + +union reut_pat_cadb_mux_ctrl_reg { + struct __packed { + uint32_t mux_0_ctrl : 2; // Bits 1:0 + uint32_t : 2; // Bits 3:2 + uint32_t mux_1_ctrl : 2; // Bits 5:4 + uint32_t : 2; // Bits 7:6 + uint32_t mux_2_ctrl : 2; // Bits 9:8 + uint32_t : 6; // Bits 15:10 + uint32_t sel_mux_0_ctrl : 2; // Bits 17:16 + uint32_t : 2; // Bits 19:18 + uint32_t sel_mux_1_ctrl : 2; // Bits 21:20 + uint32_t : 2; // Bits 23:22 + uint32_t sel_mux_2_ctrl : 2; // Bits 25:24 + uint32_t : 6; // Bits 31:26 + }; + uint32_t raw; +}; + +union reut_pat_wdb_cl_mux_cfg_reg { + struct __packed { + uint32_t mux_0_control : 2; // Bits 1:0 + uint32_t : 1; // Bits 2:2 + uint32_t mux_1_control : 2; // Bits 4:3 + uint32_t : 1; // Bits 5:5 + uint32_t mux_2_control : 2; // Bits 7:6 + uint32_t : 6; // Bits 13:8 + uint32_t ecc_replace_byte_ctl : 1; // Bits 14:14 + uint32_t ecc_data_source_sel : 1; // Bits 15:15 + uint32_t save_lfsr_seed_rate : 6; // Bits 21:16 + uint32_t : 2; // Bits 23:22 + uint32_t reload_lfsr_seed_rate : 3; // Bits 26:24 + uint32_t : 5; // Bits 31:27 + }; + uint32_t raw; +}; + union reut_pat_cadb_prog_reg { struct __packed { uint32_t addr : 16; // Bits 15:0 @@ -366,6 +414,19 @@ union reut_pat_cadb_prog_reg { uint32_t raw32[2]; }; +union reut_pat_wdb_cl_ctrl_reg { + struct __packed { + uint32_t inc_rate : 5; // Bits 4:0 + uint32_t inc_scale : 1; // Bits 5:5 + uint32_t : 2; // Bits 7:6 + uint32_t start_ptr : 6; // Bits 13:8 + uint32_t : 2; // Bits 15:14 + uint32_t end_ptr : 6; // Bits 21:16 + uint32_t : 10; // Bits 31:22 + }; + uint32_t raw; +}; + union reut_pat_cadb_mrs_reg { struct __packed { uint32_t delay_gap : 3; // Bits 2:0 @@ -406,6 +467,66 @@ union reut_seq_cfg_reg { uint32_t raw32[2]; }; +union reut_seq_base_addr_reg { + struct __packed { + uint32_t : 3; // Bits 2:0 + uint32_t col_addr : 8; // Bits 10:3 + uint32_t : 13; // Bits 23:11 + uint32_t row_addr : 16; // Bits 39:24 + uint32_t : 8; // Bits 47:40 + uint32_t bank_addr : 3; // Bits 50:48 + uint32_t : 5; // Bits 55:51 + uint32_t rank_addr : 3; // Bits 58:56 + uint32_t : 5; // Bits 63:59 + }; + uint32_t raw32[2]; + uint64_t raw; +}; + +union reut_seq_misc_ctl_reg { + struct __packed { + uint32_t col_addr_order : 2; // Bits 1:0 + uint32_t row_addr_order : 2; // Bits 3:2 + uint32_t bank_addr_order : 2; // Bits 5:4 + uint32_t rank_addr_order : 2; // Bits 7:6 + uint32_t : 5; // Bits 12:8 + uint32_t addr_invert_rate : 3; // Bits 15:13 + uint32_t : 4; // Bits 19:16 + uint32_t col_addr_invert_en : 1; // Bits 20:20 + uint32_t row_addr_invert_en : 1; // Bits 21:21 + uint32_t bank_addr_invert_en : 1; // Bits 22:22 + uint32_t rank_addr_invert_en : 1; // Bits 23:23 + uint32_t col_wrap_trigger_en : 1; // Bits 24:24 + uint32_t row_wrap_trigger_en : 1; // Bits 25:25 + uint32_t bank_wrap_trigger_en : 1; // Bits 26:26 + uint32_t rank_wrap_trigger_en : 1; // Bits 27:27 + uint32_t col_wrap_carry_en : 1; // Bits 28:28 + uint32_t row_wrap_carry_en : 1; // Bits 29:29 + uint32_t bank_wrap_carry_en : 1; // Bits 30:30 + uint32_t rank_wrap_carry_en : 1; // Bits 31:31 + }; + uint32_t raw; +}; + +union reut_seq_addr_inc_ctl_reg { + struct __packed { + uint32_t : 3; // Bits 2:0 + uint32_t col_addr_increment : 8; // Bits 10:3 + uint32_t : 1; // Bits 11:11 + uint32_t col_addr_update : 8; // Bits 19:12 + uint32_t row_addr_increment : 12; // Bits 31:20 + uint32_t row_addr_update : 6; // Bits 37:32 + uint32_t bank_addr_increment : 3; // Bits 40:38 + uint32_t : 3; // Bits 43:41 + uint32_t bank_addr_update : 8; // Bits 53:44 + uint32_t rank_addr_increment : 3; // Bits 54:52 + uint32_t : 1; // Bits 55:55 + uint32_t rank_addr_update : 8; // Bits 63:56 + }; + uint64_t raw; + uint32_t raw32[2]; +}; + union reut_seq_ctl_reg { struct __packed { uint32_t start_test : 1; // Bits 0:0 diff --git a/src/northbridge/intel/haswell/native_raminit/testing_io.c b/src/northbridge/intel/haswell/native_raminit/testing_io.c new file mode 100644 index 0000000000..2632c238f8 --- /dev/null +++ b/src/northbridge/intel/haswell/native_raminit/testing_io.c @@ -0,0 +1,744 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ + +#include <console/console.h> +#include <delay.h> +#include <lib.h> +#include <northbridge/intel/haswell/haswell.h> +#include <timer.h> +#include <types.h> + +#include "raminit_native.h" + +static void set_cadb_patterns(const uint8_t channel, const uint16_t seeds[NUM_CADB_MUX_SEEDS]) +{ + for (uint8_t i = 0; i < NUM_CADB_MUX_SEEDS; i++) + mchbar_write32(REUT_ch_PAT_CADB_MUX_x(channel, i), seeds[i]); +} + +static void setup_cadb( + struct sysinfo *ctrl, + const uint8_t channel, + const uint8_t vic_spread, + const uint8_t vic_bit) +{ + const bool lmn_en = false; + + /* + * Currently, always start writing at CADB row 0. + * Could add a start point parameter in the future. + */ + mchbar_write8(REUT_ch_PAT_CADB_WRITE_PTR(channel), 0); + const uint8_t num_cadb_rows = 8; + for (uint8_t row = 0; row < num_cadb_rows; row++) { + const uint8_t lfsr0 = (row >> 0) & 1; + const uint8_t lfsr1 = (row >> 1) & 1; + uint64_t reg64 = 0; + for (uint8_t bit = 0; bit < 22; bit++) { + uint8_t bremap; + if (bit >= 19) { + /* (bremap in 40 .. 42) => CADB data control */ + bremap = bit + 21; + } else if (bit >= 16) { + /* (bremap in 24 .. 26) => CADB data bank */ + bremap = bit + 8; + } else { + /* (bremap in 0 .. 15) => CADB data address */ + bremap = bit; + } + const uint8_t fine = bit % vic_spread; + reg64 |= ((uint64_t)(fine == vic_bit ? lfsr0 : lfsr1)) << bremap; + } + /* + * Write row. CADB pointer is auto incremented after every write. This must be + * a single 64-bit write, otherwise the CADB pointer will auto-increment twice. + */ + mchbar_write64(REUT_ch_PAT_CADB_PROG(channel), reg64); + } + const union reut_pat_cadb_mux_ctrl_reg cadb_mux_ctrl = { + .mux_0_ctrl = lmn_en ? REUT_MUX_LMN : REUT_MUX_LFSR, + .mux_1_ctrl = REUT_MUX_LFSR, + .mux_2_ctrl = REUT_MUX_LFSR, + }; + mchbar_write32(REUT_ch_PAT_CADB_MUX_CTRL(channel), cadb_mux_ctrl.raw); + const union reut_pat_cl_mux_lmn_reg cadb_cl_mux_lmn = { + .en_sweep_freq = 1, + .l_counter = 1, + .m_counter = 1, + .n_counter = 6, + }; + mchbar_write32(REUT_ch_PAT_CADB_CL_MUX_LMN(channel), cadb_cl_mux_lmn.raw); + const uint16_t cadb_mux_seeds[NUM_CADB_MUX_SEEDS] = { 0x0ea1, 0xbeef, 0xdead }; + set_cadb_patterns(channel, cadb_mux_seeds); +} + +static uint32_t calc_rate(const uint32_t rate, const uint32_t lim, const uint8_t scale_bit) +{ + return rate > lim ? log2_ceil(rate - 1) : BIT(scale_bit) | rate; +} + +void program_seq_addr( + const uint8_t channel, + const struct reut_box *reut_addr, + const bool log_seq_addr) +{ + const int loglevel = log_seq_addr ? BIOS_ERR : BIOS_NEVER; + const uint32_t div = 8; + union reut_seq_base_addr_reg reut_seq_addr_start = { + .col_addr = reut_addr->col.start / div, + .row_addr = reut_addr->row.start, + .bank_addr = reut_addr->bank.start, + .rank_addr = reut_addr->rank.start, + }; + mchbar_write64(REUT_ch_SEQ_ADDR_START(channel), reut_seq_addr_start.raw); + reut_seq_addr_start.raw = mchbar_read64(REUT_ch_SEQ_ADDR_START(channel)); + printk(loglevel, "\tStart column: %u\n", reut_seq_addr_start.col_addr); + printk(loglevel, "\tStart row: %u\n", reut_seq_addr_start.row_addr); + printk(loglevel, "\tStart bank: %u\n", reut_seq_addr_start.bank_addr); + printk(loglevel, "\tStart rank: %u\n", reut_seq_addr_start.rank_addr); + printk(loglevel, "\n"); + + union reut_seq_base_addr_reg reut_seq_addr_stop = { + .col_addr = reut_addr->col.stop / div, + .row_addr = reut_addr->row.stop, + .bank_addr = reut_addr->bank.stop, + .rank_addr = reut_addr->rank.stop, + }; + mchbar_write64(REUT_ch_SEQ_ADDR_WRAP(channel), reut_seq_addr_stop.raw); + reut_seq_addr_stop.raw = mchbar_read64(REUT_ch_SEQ_ADDR_WRAP(channel)); + printk(loglevel, "\tStop column: %u\n", reut_seq_addr_stop.col_addr); + printk(loglevel, "\tStop row: %u\n", reut_seq_addr_stop.row_addr); + printk(loglevel, "\tStop bank: %u\n", reut_seq_addr_stop.bank_addr); + printk(loglevel, "\tStop rank: %u\n", reut_seq_addr_stop.rank_addr); + printk(loglevel, "\n"); + + union reut_seq_misc_ctl_reg reut_seq_misc_ctl = { + .col_wrap_trigger_en = reut_addr->col.wrap_trigger, + .row_wrap_trigger_en = reut_addr->row.wrap_trigger, + .bank_wrap_trigger_en = reut_addr->bank.wrap_trigger, + .rank_wrap_trigger_en = reut_addr->rank.wrap_trigger, + }; + mchbar_write32(REUT_ch_SEQ_MISC_CTL(channel), reut_seq_misc_ctl.raw); + printk(loglevel, "\tWrap column: %u\n", reut_addr->col.wrap_trigger); + printk(loglevel, "\tWrap row: %u\n", reut_addr->row.wrap_trigger); + printk(loglevel, "\tWrap bank: %u\n", reut_addr->bank.wrap_trigger); + printk(loglevel, "\tWrap rank: %u\n", reut_addr->rank.wrap_trigger); + printk(loglevel, "\n"); + + union reut_seq_addr_inc_ctl_reg reut_seq_addr_inc_ctl = { + .col_addr_update = calc_rate(reut_addr->col.inc_rate, 31, 7), + .row_addr_update = calc_rate(reut_addr->row.inc_rate, 15, 5), + .bank_addr_update = calc_rate(reut_addr->bank.inc_rate, 31, 7), + .rank_addr_update = calc_rate(reut_addr->rank.inc_rate, 31, 7), + .col_addr_increment = reut_addr->col.inc_val, + .row_addr_increment = reut_addr->row.inc_val, + .bank_addr_increment = reut_addr->bank.inc_val, + .rank_addr_increment = reut_addr->rank.inc_val, + }; + printk(loglevel, "\tUpdRate column: %u\n", reut_addr->col.inc_rate); + printk(loglevel, "\tUpdRate row: %u\n", reut_addr->row.inc_rate); + printk(loglevel, "\tUpdRate bank: %u\n", reut_addr->bank.inc_rate); + printk(loglevel, "\tUpdRate rank: %u\n", reut_addr->rank.inc_rate); + printk(loglevel, "\n"); + printk(loglevel, "\tUpdRateCR column: %u\n", reut_seq_addr_inc_ctl.col_addr_update); + printk(loglevel, "\tUpdRateCR row: %u\n", reut_seq_addr_inc_ctl.row_addr_update); + printk(loglevel, "\tUpdRateCR bank: %u\n", reut_seq_addr_inc_ctl.bank_addr_update); + printk(loglevel, "\tUpdRateCR rank: %u\n", reut_seq_addr_inc_ctl.rank_addr_update); + printk(loglevel, "\n"); + printk(loglevel, "\tUpdInc column: %u\n", reut_seq_addr_inc_ctl.col_addr_increment); + printk(loglevel, "\tUpdInc row: %u\n", reut_seq_addr_inc_ctl.row_addr_increment); + printk(loglevel, "\tUpdInc bank: %u\n", reut_seq_addr_inc_ctl.bank_addr_increment); + printk(loglevel, "\tUpdInc rank: %u\n", reut_seq_addr_inc_ctl.rank_addr_increment); + printk(loglevel, "\n"); + mchbar_write64(REUT_ch_SEQ_ADDR_INC_CTL(channel), reut_seq_addr_inc_ctl.raw); +} + +/* + * Early steppings take exponential (base 2) loopcount values, + * but later steppings take linear loopcount values elsewhere. + * Address the differences in register offset and format here. + */ +void program_loop_count(const struct sysinfo *ctrl, const uint8_t channel, const uint8_t lc_exp) +{ + if (ctrl->stepping >= STEPPING_C0) { + const uint32_t loopcount = lc_exp >= 32 ? 0 : BIT(lc_exp); + mchbar_write32(HSW_REUT_ch_SEQ_LOOP_COUNT(channel), loopcount); + } else { + const uint8_t loopcount = lc_exp >= 32 ? 0 : lc_exp + 1; + union reut_seq_cfg_reg reut_seq_cfg = { + .raw = mchbar_read64(REUT_ch_SEQ_CFG(channel)), + }; + reut_seq_cfg.early_steppings_loop_count = loopcount; + mchbar_write64(REUT_ch_SEQ_CFG(channel), reut_seq_cfg.raw); + } +} + +static inline void write_subseq(const uint8_t channel, const uint8_t idx, const uint32_t ssq) +{ + mchbar_write32(REUT_ch_SUBSEQ_x_CTL(channel, idx), ssq); +} + +static void program_subseq( + struct sysinfo *const ctrl, + const uint8_t channel, + const enum reut_cmd_pat cmd_pat, + const uint32_t ss_a, + const uint32_t ss_b) +{ + switch (cmd_pat) { + case PAT_WR_RD_TA: + write_subseq(channel, 0, ss_a | SUBSEQ_B_WR); + for (uint8_t i = 1; i < 7; i++) + write_subseq(channel, i, ss_b | SUBSEQ_B_RD_WR); + + write_subseq(channel, 7, ss_a | SUBSEQ_B_RD); + break; + case PAT_RD_WR_TA: + write_subseq(channel, 0, ss_b | SUBSEQ_B_WR_RD); + break; + case PAT_ODT_TA: + write_subseq(channel, 0, ss_a | SUBSEQ_B_WR); + write_subseq(channel, 1, ss_b | SUBSEQ_B_RD_WR); + write_subseq(channel, 2, ss_a | SUBSEQ_B_RD); + write_subseq(channel, 3, ss_b | SUBSEQ_B_WR_RD); + break; + default: + write_subseq(channel, 0, ss_a | SUBSEQ_B_WR); + write_subseq(channel, 1, ss_a | SUBSEQ_B_RD); + break; + } +} + +void setup_io_test( + struct sysinfo *ctrl, + const uint8_t chanmask, + const enum reut_cmd_pat cmd_pat, + const uint16_t num_cl, + const uint8_t lc, + const struct reut_box *const reut_addr, + const enum test_stop soe, + const struct wdb_pat *const pat, + const uint8_t en_cadb, + const uint8_t subseq_wait) +{ + if (!chanmask) { + printk(BIOS_ERR, "\n%s: chanmask is invalid\n", __func__); + return; + } + + /* + * Prepare variables needed for both channels. + * Check for the cases where this MUST be 1: when + * we manually walk through subseq ODT and TA Wr. + */ + uint8_t lc_exp = MAX(lc - log2_ceil(num_cl), 0); + if (cmd_pat == PAT_WR_RD_TA || cmd_pat == PAT_ODT_TA) + lc_exp = 0; + + uint8_t num_clcr; + if (num_cl > 127) { + /* Assume exponential number */ + num_clcr = log2_ceil(num_cl); + } else { + /* Set number of cache lines as linear number */ + num_clcr = num_cl | BIT(7); + } + + const uint16_t num_cl2 = 2 * num_cl; + uint8_t num_cl2cr; + if (num_cl2 > 127) { + /* Assume exponential number */ + num_cl2cr = log2_ceil(num_cl2); + } else { + /* Set number of cache lines as linear number */ + num_cl2cr = num_cl2 | BIT(7); + } + + for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) { + if (!(chanmask & BIT(channel))) { + union reut_seq_cfg_reg reut_seq_cfg = { + .raw = mchbar_read64(REUT_ch_SEQ_CFG(channel)), + }; + reut_seq_cfg.global_control = 0; + mchbar_write64(REUT_ch_SEQ_CFG(channel), reut_seq_cfg.raw); + continue; + } + + /* + * Program CADB + */ + mchbar_write8(REUT_ch_MISC_PAT_CADB_CTRL(channel), !!en_cadb); + if (en_cadb) + setup_cadb(ctrl, channel, 7, 8); + + /* + * Program sequence + */ + uint8_t subseq_start = 0; + uint8_t subseq_end = 0; + switch (cmd_pat) { + case PAT_WR_RD: + subseq_end = 1; + break; + case PAT_WR: + break; + case PAT_RD: + subseq_start = 1; + subseq_end = 1; + break; + case PAT_RD_WR_TA: + break; + case PAT_WR_RD_TA: + subseq_end = 7; + break; + case PAT_ODT_TA: + subseq_end = 3; + break; + default: + die("\n%s: Pattern type %u is invalid\n", __func__, cmd_pat); + } + const union reut_seq_cfg_reg reut_seq_cfg = { + .global_control = 1, + .initialization_mode = REUT_MODE_TEST, + .subsequence_start_pointer = subseq_start, + .subsequence_end_pointer = subseq_end, + .start_test_delay = 2, + }; + mchbar_write64(REUT_ch_SEQ_CFG(channel), reut_seq_cfg.raw); + program_loop_count(ctrl, channel, lc_exp); + mchbar_write32(REUT_ch_SEQ_CTL(channel), (union reut_seq_ctl_reg) { + .clear_errors = 1, + }.raw); + + /* + * Program subsequences + */ + uint32_t subseq_a = 0; + + /* Number of cachelines and scale */ + subseq_a |= (num_clcr & 0x00ff) << 0; + subseq_a |= (subseq_wait & 0x3fff) << 8; + + /* Reset current base address to start */ + subseq_a |= BIT(27); + + uint32_t subseq_b = 0; + + /* Number of cachelines and scale */ + subseq_b |= (num_cl2cr & 0x00ff) << 0; + subseq_b |= (subseq_wait & 0x3fff) << 8; + + /* Reset current base address to start */ + subseq_b |= BIT(27); + + program_subseq(ctrl, channel, cmd_pat, subseq_a, subseq_b); + + /* Program sequence address */ + program_seq_addr(channel, reut_addr, false); + + /* Program WDB */ + const bool is_linear = pat->inc_rate < 32; + mchbar_write32(REUT_ch_WDB_CL_CTRL(channel), (union reut_pat_wdb_cl_ctrl_reg) { + .start_ptr = pat->start_ptr, + .end_ptr = pat->stop_ptr, + .inc_rate = is_linear ? pat->inc_rate : log2_ceil(pat->inc_rate), + .inc_scale = is_linear, + }.raw); + + /* Enable LMN in LMN or CADB modes, used to create lots of supply noise */ + const bool use_lmn = pat->dq_pattern == LMN_VA || pat->dq_pattern == CADB; + union reut_pat_wdb_cl_mux_cfg_reg pat_wdb_cl_mux_cfg = { + .mux_0_control = use_lmn ? REUT_MUX_LMN : REUT_MUX_LFSR, + .mux_1_control = REUT_MUX_LFSR, + .mux_2_control = REUT_MUX_LFSR, + .ecc_data_source_sel = 1, + }; + + /* Program LFSR save/restore, too complex unless everything is power of 2 */ + if (cmd_pat == PAT_ODT_TA || cmd_pat == PAT_WR_RD_TA) { + pat_wdb_cl_mux_cfg.reload_lfsr_seed_rate = log2_ceil(num_cl) + 1; + pat_wdb_cl_mux_cfg.save_lfsr_seed_rate = 1; + } + mchbar_write32(REUT_ch_PAT_WDB_CL_MUX_CFG(channel), pat_wdb_cl_mux_cfg.raw); + + /* Inversion mask is not used */ + mchbar_write32(REUT_ch_PAT_WDB_INV(channel), 0); + + /* Program error checking */ + const union reut_err_ctl_reg reut_err_ctl = { + .selective_err_enable_cacheline = 0xff, + .selective_err_enable_chunk = 0xff, + .stop_on_error_control = soe, + .stop_on_nth_error = 1, + }; + mchbar_write32(REUT_ch_ERR_CONTROL(channel), reut_err_ctl.raw); + mchbar_write64(REUT_ch_ERR_DATA_MASK(channel), 0); + mchbar_write8(REUT_ch_ERR_ECC_MASK(channel), 0); + } + + /* Always do a ZQ short before the beginning of a test */ + reut_issue_zq(ctrl, chanmask, ZQ_SHORT); +} + +void setup_io_test_cadb( + struct sysinfo *ctrl, + const uint8_t chanmask, + const uint8_t lc, + const enum test_stop soe) +{ + const struct reut_box reut_addr = { + .rank = { + .start = 0, + .stop = 0, + .inc_rate = 32, + .inc_val = 1, + }, + .bank = { + .start = 0, + .stop = 7, + .inc_rate = 3, + .inc_val = 1, + }, + .row = { + .start = 0, + .stop = 2047, + .inc_rate = 3, + .inc_val = 73, + }, + .col = { + .start = 0, + .stop = 1023, + .inc_rate = 0, + .inc_val = 53, + }, + }; + const struct wdb_pat pattern = { + .start_ptr = 0, + .stop_ptr = 9, + .inc_rate = 4, + .dq_pattern = CADB, + }; + setup_io_test( + ctrl, + chanmask, + PAT_WR_RD, + 128, + lc, + &reut_addr, + soe, + &pattern, + 1, + 0); + + ctrl->dq_pat_lc = MAX(lc - 2 - 3, 0) + 1; + ctrl->dq_pat = CADB; +} + +void setup_io_test_basic_va( + struct sysinfo *ctrl, + const uint8_t chanmask, + const uint8_t lc, + const enum test_stop soe) +{ + const uint32_t spread = 8; + const struct reut_box reut_addr = { + .rank = { + .start = 0, + .stop = 0, + .inc_rate = 32, + .inc_val = 1, + }, + .col = { + .start = 0, + .stop = 1023, + .inc_rate = 0, + .inc_val = 1, + }, + }; + const struct wdb_pat pattern = { + .start_ptr = 0, + .stop_ptr = spread - 1, + .inc_rate = 4, + .dq_pattern = BASIC_VA, + }; + setup_io_test( + ctrl, + chanmask, + PAT_WR_RD, + 128, + lc, + &reut_addr, + soe, + &pattern, + 0, + 0); + + ctrl->dq_pat_lc = MAX(lc - 8, 0) + 1; + ctrl->dq_pat = BASIC_VA; +} + +void setup_io_test_mpr( + struct sysinfo *ctrl, + const uint8_t chanmask, + const uint8_t lc, + const enum test_stop soe) +{ + const struct reut_box reut_addr_ddr = { + .rank = { + .start = 0, + .stop = 0, + .inc_rate = 32, + .inc_val = 1, + }, + .col = { + .start = 0, + .stop = 1023, + .inc_rate = 0, + .inc_val = 1, + }, + }; + const struct reut_box reut_addr_lpddr = { + .bank = { + .start = 4, + .stop = 4, + .inc_rate = 0, + .inc_val = 0, + }, + }; + const struct wdb_pat pattern = { + .start_ptr = 0, + .stop_ptr = 9, + .inc_rate = 4, + .dq_pattern = BASIC_VA, + }; + setup_io_test( + ctrl, + chanmask, + PAT_RD, + 128, + lc, + ctrl->lpddr ? &reut_addr_lpddr : &reut_addr_ddr, + soe, + &pattern, + 0, + 0); + + ctrl->dq_pat_lc = 1; + ctrl->dq_pat = BASIC_VA; +} + +uint8_t select_reut_ranks(struct sysinfo *ctrl, const uint8_t channel, uint8_t rankmask) +{ + rankmask &= ctrl->rankmap[channel]; + + uint8_t rank_count = 0; + uint32_t rank_log_to_phys = 0; + for (uint8_t rank = 0; rank < NUM_SLOTRANKS; rank++) { + if (!rank_in_mask(rank, rankmask)) + continue; + + rank_log_to_phys |= rank << (4 * rank_count); + rank_count++; + } + mchbar_write32(REUT_ch_RANK_LOG_TO_PHYS(channel), rank_log_to_phys); + + union reut_seq_cfg_reg reut_seq_cfg = { + .raw = mchbar_read64(REUT_ch_SEQ_CFG(channel)), + }; + if (!rank_count) { + reut_seq_cfg.global_control = 0; + mchbar_write64(REUT_ch_SEQ_CFG(channel), reut_seq_cfg.raw); + return 0; + } + union reut_seq_base_addr_reg reut_seq_addr_stop = { + .raw = mchbar_read64(REUT_ch_SEQ_ADDR_WRAP(channel)), + }; + reut_seq_addr_stop.rank_addr = rank_count - 1; + mchbar_write64(REUT_ch_SEQ_ADDR_WRAP(channel), reut_seq_addr_stop.raw); + + reut_seq_cfg.global_control = 1; + mchbar_write64(REUT_ch_SEQ_CFG(channel), reut_seq_cfg.raw); + return BIT(channel); +} + +void run_mpr_io_test(const bool clear_errors) +{ + io_reset(); + mchbar_write32(REUT_GLOBAL_CTL, (union reut_seq_ctl_reg) { + .start_test = 1, + .clear_errors = clear_errors, + }.raw); + tick_delay(2); + io_reset(); + tick_delay(2); + mchbar_write32(REUT_GLOBAL_CTL, (union reut_seq_ctl_reg) { + .stop_test = 1, + }.raw); +} + +static uint8_t get_num_tests(const uint8_t dq_pat) +{ + switch (dq_pat) { + case SEGMENT_WDB: return 4; + case CADB: return 7; + case TURN_AROUND_WR: return 8; + case TURN_AROUND_ODT: return 4; + case RD_RD_TA: return 2; + case RD_RD_TA_ALL: return 8; + default: return 1; + } +} + +uint8_t run_io_test( + struct sysinfo *const ctrl, + const uint8_t chanmask, + const uint8_t dq_pat, + const bool clear_errors) +{ + /* SEGMENT_WDB only runs 4 tests */ + const uint8_t segment_wdb_lc[4] = { 0, 0, 4, 2 }; + const union reut_pat_wdb_cl_ctrl_reg pat_wdb_cl[4] = { + [0] = { + .start_ptr = 0, + .end_ptr = 9, + .inc_rate = 25, + .inc_scale = SCALE_LINEAR, + }, + [1] = { + .start_ptr = 0, + .end_ptr = 9, + .inc_rate = 25, + .inc_scale = SCALE_LINEAR, + }, + [2] = { + .start_ptr = 10, + .end_ptr = 63, + .inc_rate = 19, + .inc_scale = SCALE_LINEAR, + }, + [3] = { + .start_ptr = 10, + .end_ptr = 63, + .inc_rate = 10, + .inc_scale = SCALE_LINEAR, + }, + }; + const bool is_turnaround = dq_pat == RD_RD_TA || dq_pat == RD_RD_TA_ALL; + const uint8_t num_tests = get_num_tests(dq_pat); + union tc_bank_rank_a_reg tc_bank_rank_a[NUM_CHANNELS] = { 0 }; + if (is_turnaround) { + for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) { + if (!(chanmask & BIT(channel))) + continue; + + tc_bank_rank_a[channel].raw = ctrl->tc_bankrank_a[channel].raw; + } + } + for (uint8_t t = 0; t < num_tests; t++) { + for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) { + if (!(chanmask & BIT(channel))) + continue; + + if (dq_pat == SEGMENT_WDB) { + mchbar_write32(REUT_ch_WDB_CL_CTRL(channel), pat_wdb_cl[t].raw); + /* + * Skip programming LFSR save/restore. Too complex + * unless power of 2. Program desired loopcount. + */ + const uint8_t pat_lc = ctrl->dq_pat_lc + segment_wdb_lc[t]; + program_loop_count(ctrl, channel, pat_lc); + } else if (dq_pat == CADB) { + setup_cadb(ctrl, channel, num_tests, t); + } else if (dq_pat == TURN_AROUND_WR || dq_pat == TURN_AROUND_ODT) { + union reut_seq_cfg_reg reut_seq_cfg = { + .raw = mchbar_read64(REUT_ch_SEQ_CFG(channel)), + }; + reut_seq_cfg.subsequence_start_pointer = t; + reut_seq_cfg.subsequence_end_pointer = t; + mchbar_write64(REUT_ch_SEQ_CFG(channel), reut_seq_cfg.raw); + union reut_seq_addr_inc_ctl_reg addr_inc_ctl = { + .raw = mchbar_read64(REUT_ch_SEQ_ADDR_INC_CTL(channel)), + }; + uint8_t ta_inc_rate = 1; + if (dq_pat == TURN_AROUND_WR && (t == 0 || t == 7)) + ta_inc_rate = 0; + else if (dq_pat == TURN_AROUND_ODT && (t == 0 || t == 2)) + ta_inc_rate = 0; + + /* Program increment rate as linear value */ + addr_inc_ctl.rank_addr_update = BIT(7) | ta_inc_rate; + addr_inc_ctl.col_addr_update = BIT(7) | ta_inc_rate; + mchbar_write64(REUT_ch_SEQ_ADDR_INC_CTL(channel), + addr_inc_ctl.raw); + } else if (dq_pat == RD_RD_TA) { + tc_bank_rank_a[channel].tRDRD_sr = (t == 0) ? 4 : 5; + mchbar_write32(TC_BANK_RANK_A_ch(channel), + tc_bank_rank_a[channel].raw); + } else if (dq_pat == RD_RD_TA_ALL) { + /* + * Program tRDRD for SR and DR. Run 8 tests, covering + * tRDRD_sr = 4, 5, 6, 7 and tRDRD_dr = min, +1, +2, +3 + */ + const uint32_t tRDRD_dr = ctrl->tc_bankrank_a[channel].tRDRD_dr; + tc_bank_rank_a[channel].tRDRD_sr = (t % 4) + 4; + tc_bank_rank_a[channel].tRDRD_dr = (t % 4) + tRDRD_dr; + mchbar_write32(TC_BANK_RANK_A_ch(channel), + tc_bank_rank_a[channel].raw); + + /* Program linear rank increment rate */ + union reut_seq_addr_inc_ctl_reg addr_inc_ctl = { + .raw = mchbar_read64(REUT_ch_SEQ_ADDR_INC_CTL(channel)), + }; + addr_inc_ctl.rank_addr_update = BIT(7) | (t / 4) ? 0 : 31; + mchbar_write64(REUT_ch_SEQ_ADDR_INC_CTL(channel), + addr_inc_ctl.raw); + } + } + bool test_soe = false; + for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) { + if (!(chanmask & BIT(channel))) + continue; + + const union reut_err_ctl_reg reut_err_ctl = { + .raw = mchbar_read32(REUT_ch_ERR_CONTROL(channel)), + }; + const uint8_t soe = reut_err_ctl.stop_on_error_control; + if (soe != NSOE) { + test_soe = true; + break; + } + } + io_reset(); + mchbar_write32(REUT_GLOBAL_CTL, (union reut_seq_ctl_reg) { + .start_test = 1, + .clear_errors = clear_errors && t == 0, + }.raw); + struct mono_time prev, curr; + timer_monotonic_get(&prev); + union reut_global_err_reg global_err; + do { + global_err.raw = mchbar_read32(REUT_GLOBAL_ERR); + /** TODO: Clean up this mess **/ + timer_monotonic_get(&curr); + if (mono_time_diff_microseconds(&prev, &curr) > 1000 * 1000) { + mchbar_write32(REUT_GLOBAL_CTL, (union reut_seq_ctl_reg) { + .stop_test = 1, + }.raw); + printk(BIOS_ERR, "REUT timed out, ch_done: %x\n", + global_err.ch_test_done); + break; + } + } while ((global_err.ch_test_done & chanmask) != chanmask); + if (test_soe && global_err.ch_error & chanmask) + break; + } + if (is_turnaround) { + for (uint8_t channel = 0; channel < NUM_CHANNELS; channel++) { + if (!(chanmask & BIT(channel))) + continue; + + mchbar_write32(TC_BANK_RANK_A_ch(channel), + ctrl->tc_bankrank_a[channel].raw); + } + } + return ((union reut_global_err_reg)mchbar_read32(REUT_GLOBAL_ERR)).ch_error; +} diff --git a/src/northbridge/intel/haswell/registers/mchbar.h b/src/northbridge/intel/haswell/registers/mchbar.h index f8408e51a0..817a9f8bf8 100644 --- a/src/northbridge/intel/haswell/registers/mchbar.h +++ b/src/northbridge/intel/haswell/registers/mchbar.h @@ -94,20 +94,35 @@ #define TC_BANK_RANK_D_ch(ch) _MCMAIN_C(0x4014, ch) #define SC_ROUNDT_LAT_ch(ch) _MCMAIN_C(0x4024, ch) +#define REUT_ch_PAT_WDB_CL_MUX_CFG(ch) _MCMAIN_C(0x4040, ch) + #define REUT_ch_PAT_WDB_CL_MUX_WR_x(ch, x) _MCMAIN_C_X(0x4048, ch, x) /* x in 0 .. 2 */ #define REUT_ch_PAT_WDB_CL_MUX_RD_x(ch, x) _MCMAIN_C_X(0x4054, ch, x) /* x in 0 .. 2 */ #define REUT_ch_PAT_WDB_CL_MUX_LMN(ch) _MCMAIN_C(0x4078, ch) +#define REUT_ch_PAT_WDB_INV(ch) _MCMAIN_C(0x4084, ch) + +#define REUT_ch_ERR_CONTROL(ch) _MCMAIN_C(0x4098, ch) +#define REUT_ch_ERR_ECC_MASK(ch) _MCMAIN_C(0x409c, ch) + #define SC_WR_ADD_DELAY_ch(ch) _MCMAIN_C(0x40d0, ch) +#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_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) +#define REUT_ch_PAT_CADB_MUX_x(ch, x) _MCMAIN_C_X(0x41a4, ch, x) /* x in 0 .. 2 */ +#define REUT_ch_PAT_CADB_CL_MUX_LMN(ch) _MCMAIN_C(0x41b0, ch) #define REUT_ch_PAT_CADB_WRITE_PTR(ch) _MCMAIN_C(0x41bc, ch) #define REUT_ch_PAT_CADB_PROG(ch) _MCMAIN_C(0x41c0, ch) +#define REUT_ch_WDB_CL_CTRL(ch) _MCMAIN_C(0x4200, ch) + #define TC_ZQCAL_ch(ch) _MCMAIN_C(0x4290, ch) #define TC_RFP_ch(ch) _MCMAIN_C(0x4294, ch) #define TC_RFTP_ch(ch) _MCMAIN_C(0x4298, ch) @@ -119,12 +134,27 @@ #define QCLK_ch_LDAT_SDAT(ch) _MCMAIN_C(0x42d4, ch) #define QCLK_ch_LDAT_DATA_IN_x(ch, x) _MCMAIN_C_X(0x42dc, ch, x) /* x in 0 .. 1 */ +#define REUT_GLOBAL_CTL 0x4800 #define REUT_GLOBAL_ERR 0x4804 +#define REUT_ch_SUBSEQ_x_CTL(ch, x) (0x4808 + 40 * (ch) + 4 * (x)) + #define REUT_ch_SEQ_CFG(ch) (0x48a8 + 8 * (ch)) #define REUT_ch_SEQ_CTL(ch) (0x48b8 + 4 * (ch)) +#define REUT_ch_SEQ_ADDR_START(ch) (0x48d8 + 8 * (ch)) + +#define REUT_ch_SEQ_ADDR_WRAP(ch) (0x48e8 + 8 * (ch)) + +#define REUT_ch_SEQ_MISC_CTL(ch) (0x4908 + 4 * (ch)) + +#define REUT_ch_SEQ_ADDR_INC_CTL(ch) (0x4910 + 8 * (ch)) + +#define REUT_ch_RANK_LOG_TO_PHYS(ch) (0x4930 + 4 * (ch)) /* 4 bits per rank */ + +#define HSW_REUT_ch_SEQ_LOOP_COUNT(ch) (0x4980 + 4 * (ch)) /* *** only on C0 *** */ + /* MCMAIN broadcast */ #define MCSCHEDS_CBIT 0x4c20 -- 2.39.5