sdmmc: I/O phase adjustments

1. Fix incorrect meaning of SDMMC.clock bits, synchronize the names
   with the TRM.
2. Choose input and output phases to satisfy typical timing
   requirements.
3. Move use_hold_reg setting into the host driver, since it is related
   to timing.

Closes #8521
Related to #8257

components/driver/sdmmc/sdmmc_host.c

@@ -116,33 +116,29 @@ esp_err_t sdmmc_host_reset(void)
 static void sdmmc_host_set_clk_div(int div)
 {
     // Set frequency to 160MHz / div
-    // div = p + 1
-    // duty cycle = (h + 1)/(p + 1) (should be = 1/2)
+    // div = l + 1
+    // duty cycle = (h + 1)/(l + 1) (should be = 1/2)
     assert (div > 1 && div <= 16);
-    int p = div - 1;
+    int l = div - 1;
     int h = div / 2 - 1;
 
-    SDMMC.clock.div_factor_p = p;
     SDMMC.clock.div_factor_h = h;
-    SDMMC.clock.div_factor_m = p;
+    SDMMC.clock.div_factor_l = l;
+    SDMMC.clock.div_factor_n = l;
 
     // Make sure 160 MHz source clock is used
 #if SOC_SDMMC_SUPPORT_XTAL_CLOCK
     SDMMC.clock.clk_sel = 1;
 #endif
-#if SOC_SDMMC_USE_GPIO_MATRIX
-    // 90 degree phase on input and output clocks
-    const int inout_clock_phase = 1;
-#else
-    // 180 degree phase on input and output clocks
-    const int inout_clock_phase = 4;
-#endif
-    // Set phases for in/out clocks
-    SDMMC.clock.phase_dout = inout_clock_phase;
-    SDMMC.clock.phase_din = inout_clock_phase;
+
     SDMMC.clock.phase_core = 0;
-    // Wait for the clock to propagate
-    esp_rom_delay_us(10);
+    /* 90 deg. delay for cclk_out to satisfy large hold time for SDR12 (up to 25MHz) and SDR25 (up to 50MHz) modes.
+     * Whether this delayed clock will be used depends on use_hold_reg bit in CMD structure,
+     * determined when sending out the command.
+     */
+    SDMMC.clock.phase_dout = 1;
+    SDMMC.clock.phase_din = 0;
+    esp_rom_delay_us(1);
 }
 
 static void sdmmc_host_input_clk_disable(void)
@@ -300,7 +296,7 @@ esp_err_t sdmmc_host_get_real_freq(int slot, int* real_freq_khz)
         return ESP_ERR_INVALID_ARG;
     }
 
-    int host_div = SDMMC.clock.div_factor_p + 1;
+    int host_div = SDMMC.clock.div_factor_l + 1;
     int card_div = slot == 0 ? SDMMC.clkdiv.div0 : SDMMC.clkdiv.div1;
     *real_freq_khz = sdmmc_host_calc_freq(host_div, card_div);
 
@@ -317,6 +313,9 @@ esp_err_t sdmmc_host_start_command(int slot, sdmmc_hw_cmd_t cmd, uint32_t arg) {
     if (cmd.data_expected && cmd.rw && (SDMMC.wrtprt.cards & BIT(slot)) != 0) {
         return ESP_ERR_INVALID_STATE;
     }
+    /* Outputs should be synchronized to cclk_out */
+    cmd.use_hold_reg = 1;
+
     int64_t t0 = esp_timer_get_time();
     while (SDMMC.cmd.start_command == 1) {
         if (esp_timer_get_time() - t0 > SDMMC_HOST_START_CMD_TIMEOUT_US) {

components/driver/sdmmc/sdmmc_transaction.c

@@ -305,7 +305,6 @@ static sdmmc_hw_cmd_t make_hw_cmd(sdmmc_command_t* cmd)
     if (cmd->flags & SCF_RSP_CRC) {
         res.check_response_crc = 1;
     }
-    res.use_hold_reg = 1;
     if (cmd->data) {
         res.data_expected = 1;
         if ((cmd->flags & SCF_CMD_READ) == 0) {

components/soc/esp32/include/soc/sdmmc_struct.h

@@ -379,9 +379,9 @@ typedef volatile struct sdmmc_dev_s {
             uint32_t phase_dout: 3;         ///< phase of data output clock (0x0: 0, 0x1: 90, 0x4: 180, 0x6: 270)
             uint32_t phase_din: 3;          ///< phase of data input clock
             uint32_t phase_core: 3;         ///< phase of the clock to SDMMC peripheral
-            uint32_t div_factor_p: 4;       ///< controls clock period; it will be (div_factor_p + 1) / 160MHz
             uint32_t div_factor_h: 4;       ///< controls length of high pulse; it will be (div_factor_h + 1) / 160MHz
-            uint32_t div_factor_m: 4;       ///< should be equal to div_factor_p
+            uint32_t div_factor_l: 4;       ///< controls clock period; it will be (div_factor_l + 1) / 160MHz
+            uint32_t div_factor_n: 4;       ///< should be equal to div_factor_l
             uint32_t reserved21: 11;
         };
         uint32_t val;

components/soc/esp32s3/include/soc/sdmmc_struct.h

@@ -376,9 +376,9 @@ typedef volatile struct sdmmc_dev_s {
             uint32_t phase_dout: 3;         ///< phase of data output clock (0x0: 0, 0x1: 90, 0x4: 180, 0x6: 270)
             uint32_t phase_din: 3;          ///< phase of data input clock
             uint32_t phase_core: 3;         ///< phase of the clock to SDMMC peripheral
-            uint32_t div_factor_p: 4;       ///< controls clock period; it will be (div_factor_p + 1) / 160MHz
             uint32_t div_factor_h: 4;       ///< controls length of high pulse; it will be (div_factor_h + 1) / 160MHz
-            uint32_t div_factor_m: 4;       ///< should be equal to div_factor_p
+            uint32_t div_factor_l: 4;       ///< controls clock period; it will be (div_factor_l + 1) / 160MHz
+            uint32_t div_factor_n: 4;       ///< should be equal to div_factor_l
             uint32_t reserved1 : 2;
             uint32_t clk_sel : 1;           ///< clock source select (0: XTAL, 1: 160 MHz from PLL)
             uint32_t reserved24: 8;