[PATCH v2 0/5] Fix MMC tuning algorithm

The following patch series includes a MMC tuning algorithm fix according to the following published paper [0].
This seris also includes fixes for OTAP/ITAP delay values in j721e_4bit_sdhci_set_ios_post and for HS400 mode.
For DDR52 mode, also set ENDLL=1 and call am654_sdhci_setup_dll() instead of am654_sdhci_setup_delay_chain() according to device datasheet[1].
[0] https://www.ti.com/lit/an/spract9/spract9.pdf [1] https://www.ti.com/lit/ds/symlink/am62p.pdf
Link to v1: https://lore.kernel.org/u-boot/20240415212747.2678974-1-jm@ti.com/
Changes since v1: - Use ENABLE macro instead of 0x1 - Fix assignment to val variable in j721e_4bit_sdhci_set_ios_post() - Remove extra param from mmc_send_tunning() call - Add Reviewed by
Judith Mendez (4): mmc: am654_sdhci: Add tuning algorithm for delay chain mmc: am654_sdhci: Add itap_del_ena[] to store itapdlyena bit mmc: am654_sdhci: Set ENDLL=1 for DDR52 mode mmc: am654_sdhci: Fix ITAPDLY for HS400 timing
Nitin Yadav (1): mmc: am654_sdhci: Fix OTAP/ITAP delay values
drivers/mmc/am654_sdhci.c | 172 +++++++++++++++++++++++++++++++------- 1 file changed, 140 insertions(+), 32 deletions(-)
base-commit: 3434b88d2c2fdad3cc947f9e9b03dabfdd3feb5c

Currently the sdhci_am654 driver only supports one tuning algorithm which should be used only when DLL is enabled. The ITAPDLY is selected from the largest passing window and the buffer is viewed as a circular buffer.
The new tuning algorithm should be used when the delay chain is enabled; the ITAPDLY is selected from the largest passing window and the buffer is not viewed as a circular buffer.
This implementation is based off of the following paper: [1].
Also add support for multiple failing windows.
[1] https://www.ti.com/lit/an/spract9/spract9.pdf
Fixes: a759abf569d4 ("mmc: am654_sdhci: Add support for software tuning") Signed-off-by: Judith Mendez jm@ti.com --- Changes since v1: - Remove extra param from mmc_send_tunning() call --- drivers/mmc/am654_sdhci.c | 107 +++++++++++++++++++++++++++++++------- 1 file changed, 89 insertions(+), 18 deletions(-)
diff --git a/drivers/mmc/am654_sdhci.c b/drivers/mmc/am654_sdhci.c index 2139fea04d5..7e41dd91f8e 100644 --- a/drivers/mmc/am654_sdhci.c +++ b/drivers/mmc/am654_sdhci.c @@ -97,6 +97,7 @@ struct am654_sdhci_plat { u32 strb_sel; u32 clkbuf_sel; u32 flags; + bool dll_enable; #define DLL_PRESENT BIT(0) #define IOMUX_PRESENT BIT(1) #define FREQSEL_2_BIT BIT(2) @@ -110,6 +111,12 @@ struct timing_data { u32 capability; };
+struct window { + u8 start; + u8 end; + u8 length; +}; + static const struct timing_data td[] = { [MMC_LEGACY] = {"ti,otap-del-sel-legacy", "ti,itap-del-sel-legacy", @@ -280,8 +287,11 @@ static int am654_sdhci_set_ios_post(struct sdhci_host *host) ret = am654_sdhci_setup_dll(plat, speed); if (ret) return ret; + + plat->dll_enable = true; } else { am654_sdhci_setup_delay_chain(plat, mode); + plat->dll_enable = false; }
regmap_update_bits(plat->base, PHY_CTRL5, CLKBUFSEL_MASK, @@ -375,38 +385,99 @@ static void am654_sdhci_write_b(struct sdhci_host *host, u8 val, int reg) writeb(val, host->ioaddr + reg); } #ifdef MMC_SUPPORTS_TUNING -#define ITAP_MAX 32 +#define ITAPDLY_LENGTH 32 +#define ITAPDLY_LAST_INDEX (ITAPDLY_LENGTH - 1) + +static u32 am654_sdhci_calculate_itap(struct udevice *dev, struct window + *fail_window, u8 num_fails, bool circular_buffer) +{ + u8 itap = 0, start_fail = 0, end_fail = 0, pass_length = 0; + u8 first_fail_start = 0, last_fail_end = 0; + struct window pass_window = {0, 0, 0}; + int prev_fail_end = -1; + u8 i; + + if (!num_fails) + return ITAPDLY_LAST_INDEX >> 1; + + if (fail_window->length == ITAPDLY_LENGTH) { + dev_err(dev, "No passing ITAPDLY, return 0\n"); + return 0; + } + + first_fail_start = fail_window->start; + last_fail_end = fail_window[num_fails - 1].end; + + for (i = 0; i < num_fails; i++) { + start_fail = fail_window[i].start; + end_fail = fail_window[i].end; + pass_length = start_fail - (prev_fail_end + 1); + + if (pass_length > pass_window.length) { + pass_window.start = prev_fail_end + 1; + pass_window.length = pass_length; + } + prev_fail_end = end_fail; + } + + if (!circular_buffer) + pass_length = ITAPDLY_LAST_INDEX - last_fail_end; + else + pass_length = ITAPDLY_LAST_INDEX - last_fail_end + first_fail_start; + + if (pass_length > pass_window.length) { + pass_window.start = last_fail_end + 1; + pass_window.length = pass_length; + } + + if (!circular_buffer) + itap = pass_window.start + (pass_window.length >> 1); + else + itap = (pass_window.start + (pass_window.length >> 1)) % ITAPDLY_LENGTH; + + return (itap > ITAPDLY_LAST_INDEX) ? ITAPDLY_LAST_INDEX >> 1 : itap; +} + static int am654_sdhci_execute_tuning(struct mmc *mmc, u8 opcode) { struct udevice *dev = mmc->dev; struct am654_sdhci_plat *plat = dev_get_plat(dev); - int cur_val, prev_val = 1, fail_len = 0, pass_window = 0, pass_len; - u32 itap; + struct window fail_window[ITAPDLY_LENGTH]; + u8 curr_pass, itap; + u8 fail_index = 0; + u8 prev_pass = 1; + + memset(fail_window, 0, sizeof(fail_window));
/* Enable ITAPDLY */ regmap_update_bits(plat->base, PHY_CTRL4, ITAPDLYENA_MASK, 1 << ITAPDLYENA_SHIFT);
- for (itap = 0; itap < ITAP_MAX; itap++) { + for (itap = 0; itap < ITAPDLY_LENGTH; itap++) { am654_sdhci_write_itapdly(plat, itap);
- cur_val = !mmc_send_tuning(mmc, opcode); - if (cur_val && !prev_val) - pass_window = itap; + curr_pass = !mmc_send_tuning(mmc, opcode);
- if (!cur_val) - fail_len++; + if (!curr_pass && prev_pass) + fail_window[fail_index].start = itap;
- prev_val = cur_val; + if (!curr_pass) { + fail_window[fail_index].end = itap; + fail_window[fail_index].length++; + } + + if (curr_pass && !prev_pass) + fail_index++; + + prev_pass = curr_pass; } - /* - * Having determined the length of the failing window and start of - * the passing window calculate the length of the passing window and - * set the final value halfway through it considering the range as a - * circular buffer - */ - pass_len = ITAP_MAX - fail_len; - itap = (pass_window + (pass_len >> 1)) % ITAP_MAX; + + if (fail_window[fail_index].length != 0) + fail_index++; + + itap = am654_sdhci_calculate_itap(dev, fail_window, fail_index, + plat->dll_enable); + am654_sdhci_write_itapdly(plat, itap);
return 0;

From: Nitin Yadav n-yadav@ti.com
U-Boot is failing to boot class U1 UHS SD cards due to incorrect OTAP and ITAP delay select values. Update OTAP and ITAP delay select values from DT.
Fixes: c7d106b4eb3 ("mmc: am654_sdhci: Update output tap delay writes") Signed-off-by: Nitin Yadav n-yadav@ti.com Signed-off-by: Judith Mendez jm@ti.com --- Changes since v1: - Fix assignment to val variable in j721e_4bit_sdhci_set_ios_post() --- drivers/mmc/am654_sdhci.c | 23 +++++++++++++++++++---- 1 file changed, 19 insertions(+), 4 deletions(-)
diff --git a/drivers/mmc/am654_sdhci.c b/drivers/mmc/am654_sdhci.c index 7e41dd91f8e..becb3550899 100644 --- a/drivers/mmc/am654_sdhci.c +++ b/drivers/mmc/am654_sdhci.c @@ -513,12 +513,27 @@ static int j721e_4bit_sdhci_set_ios_post(struct sdhci_host *host) { struct udevice *dev = host->mmc->dev; struct am654_sdhci_plat *plat = dev_get_plat(dev); - u32 otap_del_sel, mask, val; + int mode = host->mmc->selected_mode; + u32 otap_del_sel; + u32 itap_del_sel; + u32 mask, val; + + otap_del_sel = plat->otap_del_sel[mode];
- otap_del_sel = plat->otap_del_sel[host->mmc->selected_mode]; mask = OTAPDLYENA_MASK | OTAPDLYSEL_MASK; - val = (1 << OTAPDLYENA_SHIFT) | (otap_del_sel << OTAPDLYSEL_SHIFT); + val = (1 << OTAPDLYENA_SHIFT) | + (otap_del_sel << OTAPDLYSEL_SHIFT); + + itap_del_sel = plat->itap_del_sel[mode]; + + mask |= ITAPDLYENA_MASK | ITAPDLYSEL_MASK; + val |= (1 << ITAPDLYENA_SHIFT) | + (itap_del_sel << ITAPDLYSEL_SHIFT); + + regmap_update_bits(plat->base, PHY_CTRL4, ITAPCHGWIN_MASK, + 1 << ITAPCHGWIN_SHIFT); regmap_update_bits(plat->base, PHY_CTRL4, mask, val); + regmap_update_bits(plat->base, PHY_CTRL4, ITAPCHGWIN_MASK, 0);
regmap_update_bits(plat->base, PHY_CTRL5, CLKBUFSEL_MASK, plat->clkbuf_sel); @@ -572,7 +587,7 @@ static int sdhci_am654_get_otap_delay(struct udevice *dev, * Remove the corresponding capability if an otap-del-sel * value is not found */ - for (i = MMC_HS; i <= MMC_HS_400; i++) { + for (i = MMC_LEGACY; i <= MMC_HS_400; i++) { ret = dev_read_u32(dev, td[i].otap_binding, &plat->otap_del_sel[i]); if (ret) {

Set itap_del_ena if ITAPDLY is found in DT or if the tuning algorithm was executed and found the optimal ITAPDLY. Add the functionality to save ITAPDLYENA that can be referenced later by storing the bit in array itap_del_ena[].
Signed-off-by: Judith Mendez jm@ti.com --- Changes since v1: - Use ENABLE macro instead of 0x1 --- drivers/mmc/am654_sdhci.c | 32 ++++++++++++++++++++++---------- 1 file changed, 22 insertions(+), 10 deletions(-)
diff --git a/drivers/mmc/am654_sdhci.c b/drivers/mmc/am654_sdhci.c index becb3550899..62007ebd0f4 100644 --- a/drivers/mmc/am654_sdhci.c +++ b/drivers/mmc/am654_sdhci.c @@ -85,6 +85,8 @@ #define AM654_SDHCI_MIN_FREQ 400000 #define CLOCK_TOO_SLOW_HZ 50000000
+#define ENABLE 0x1 + struct am654_sdhci_plat { struct mmc_config cfg; struct mmc mmc; @@ -92,6 +94,7 @@ struct am654_sdhci_plat { bool non_removable; u32 otap_del_sel[MMC_MODES_END]; u32 itap_del_sel[MMC_MODES_END]; + u32 itap_del_ena[MMC_MODES_END]; u32 trm_icp; u32 drv_strength; u32 strb_sel; @@ -223,8 +226,10 @@ static int am654_sdhci_setup_dll(struct am654_sdhci_plat *plat, }
static void am654_sdhci_write_itapdly(struct am654_sdhci_plat *plat, - u32 itapdly) + u32 itapdly, u32 enable) { + regmap_update_bits(plat->base, PHY_CTRL4, ITAPDLYENA_MASK, + enable << ITAPDLYENA_SHIFT); /* Set ITAPCHGWIN before writing to ITAPDLY */ regmap_update_bits(plat->base, PHY_CTRL4, ITAPCHGWIN_MASK, 1 << ITAPCHGWIN_SHIFT); @@ -242,7 +247,8 @@ static void am654_sdhci_setup_delay_chain(struct am654_sdhci_plat *plat, mask = SELDLYTXCLK_MASK | SELDLYRXCLK_MASK; regmap_update_bits(plat->base, PHY_CTRL5, mask, val);
- am654_sdhci_write_itapdly(plat, plat->itap_del_sel[mode]); + am654_sdhci_write_itapdly(plat, plat->itap_del_sel[mode], + plat->itap_del_ena[mode]); }
static int am654_sdhci_set_ios_post(struct sdhci_host *host) @@ -443,6 +449,7 @@ static int am654_sdhci_execute_tuning(struct mmc *mmc, u8 opcode) struct udevice *dev = mmc->dev; struct am654_sdhci_plat *plat = dev_get_plat(dev); struct window fail_window[ITAPDLY_LENGTH]; + int mode = mmc->selected_mode; u8 curr_pass, itap; u8 fail_index = 0; u8 prev_pass = 1; @@ -450,11 +457,10 @@ static int am654_sdhci_execute_tuning(struct mmc *mmc, u8 opcode) memset(fail_window, 0, sizeof(fail_window));
/* Enable ITAPDLY */ - regmap_update_bits(plat->base, PHY_CTRL4, ITAPDLYENA_MASK, - 1 << ITAPDLYENA_SHIFT); + plat->itap_del_ena[mode] = ENABLE;
for (itap = 0; itap < ITAPDLY_LENGTH; itap++) { - am654_sdhci_write_itapdly(plat, itap); + am654_sdhci_write_itapdly(plat, itap, plat->itap_del_ena[mode]);
curr_pass = !mmc_send_tuning(mmc, opcode);
@@ -478,7 +484,7 @@ static int am654_sdhci_execute_tuning(struct mmc *mmc, u8 opcode) itap = am654_sdhci_calculate_itap(dev, fail_window, fail_index, plat->dll_enable);
- am654_sdhci_write_itapdly(plat, itap); + am654_sdhci_write_itapdly(plat, itap, plat->itap_del_ena[mode]);
return 0; } @@ -515,6 +521,7 @@ static int j721e_4bit_sdhci_set_ios_post(struct sdhci_host *host) struct am654_sdhci_plat *plat = dev_get_plat(dev); int mode = host->mmc->selected_mode; u32 otap_del_sel; + u32 itap_del_ena; u32 itap_del_sel; u32 mask, val;
@@ -524,10 +531,11 @@ static int j721e_4bit_sdhci_set_ios_post(struct sdhci_host *host) val = (1 << OTAPDLYENA_SHIFT) | (otap_del_sel << OTAPDLYSEL_SHIFT);
+ itap_del_ena = plat->itap_del_ena[mode]; itap_del_sel = plat->itap_del_sel[mode];
mask |= ITAPDLYENA_MASK | ITAPDLYSEL_MASK; - val |= (1 << ITAPDLYENA_SHIFT) | + val |= (itap_del_ena << ITAPDLYENA_SHIFT) | (itap_del_sel << ITAPDLYSEL_SHIFT);
regmap_update_bits(plat->base, PHY_CTRL4, ITAPCHGWIN_MASK, @@ -599,9 +607,13 @@ static int sdhci_am654_get_otap_delay(struct udevice *dev, cfg->host_caps &= ~td[i].capability; }
- if (td[i].itap_binding) - dev_read_u32(dev, td[i].itap_binding, - &plat->itap_del_sel[i]); + if (td[i].itap_binding) { + ret = dev_read_u32(dev, td[i].itap_binding, + &plat->itap_del_sel[i]); + + if (!ret) + plat->itap_del_ena[i] = ENABLE; + } }
return 0;

According to the device datasheet [0], ENDLL=1 for DDR52 mode, so call am654_sdhci_setup_dll() and write itapdly after since we do not carry out tuning.
[0] https://www.ti.com/lit/ds/symlink/am62p.pdf Fixes: c964447ea3d6 ("mmc: am654_sdhci: Add support for input tap delay") Signed-off-by: Judith Mendez jm@ti.com Reviewed-by: Jaehoon Chung jh80.chung@samsung.com --- Changes since v1: - Add Reviewed-by --- drivers/mmc/am654_sdhci.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/mmc/am654_sdhci.c b/drivers/mmc/am654_sdhci.c index 62007ebd0f4..e1047812fa8 100644 --- a/drivers/mmc/am654_sdhci.c +++ b/drivers/mmc/am654_sdhci.c @@ -289,12 +289,14 @@ static int am654_sdhci_set_ios_post(struct sdhci_host *host)
regmap_update_bits(plat->base, PHY_CTRL4, mask, val);
- if (mode > UHS_SDR25 && speed >= CLOCK_TOO_SLOW_HZ) { + if ((mode > UHS_SDR25 || mode == MMC_DDR_52) && speed >= CLOCK_TOO_SLOW_HZ) { ret = am654_sdhci_setup_dll(plat, speed); if (ret) return ret;
plat->dll_enable = true; + am654_sdhci_write_itapdly(plat, plat->itap_del_sel[mode], + plat->itap_del_ena[mode]); } else { am654_sdhci_setup_delay_chain(plat, mode); plat->dll_enable = false;

At HS400 mode the ITAPDLY value is that from High Speed mode which is incorrect and may cause boot failures.
The ITAPDLY for HS400 speed mode should be the same as ITAPDLY as HS200 timing after tuning is executed. Add the functionality to save ITAPDLY from HS200 tuning and save as HS400 ITAPDLY.
Fixes: c964447ea3d6 ("mmc: am654_sdhci: Add support for input tap delay") Signed-off-by: Judith Mendez jm@ti.com --- Changes since v1: - Use ENABLE macro instead of 0x1 --- drivers/mmc/am654_sdhci.c | 8 ++++++++ 1 file changed, 8 insertions(+)
diff --git a/drivers/mmc/am654_sdhci.c b/drivers/mmc/am654_sdhci.c index e1047812fa8..fadab7d40bb 100644 --- a/drivers/mmc/am654_sdhci.c +++ b/drivers/mmc/am654_sdhci.c @@ -295,6 +295,11 @@ static int am654_sdhci_set_ios_post(struct sdhci_host *host) return ret;
plat->dll_enable = true; + if (mode == MMC_HS_400) { + plat->itap_del_ena[mode] = ENABLE; + plat->itap_del_sel[mode] = plat->itap_del_sel[mode - 1]; + } + am654_sdhci_write_itapdly(plat, plat->itap_del_sel[mode], plat->itap_del_ena[mode]); } else { @@ -486,6 +491,9 @@ static int am654_sdhci_execute_tuning(struct mmc *mmc, u8 opcode) itap = am654_sdhci_calculate_itap(dev, fail_window, fail_index, plat->dll_enable);
+ /* Save ITAPDLY */ + plat->itap_del_sel[mode] = itap; + am654_sdhci_write_itapdly(plat, itap, plat->itap_del_ena[mode]);
return 0;

On Thu, 18 Apr 2024 14:00:55 -0500, Judith Mendez wrote:
The following patch series includes a MMC tuning algorithm fix according to the following published paper [0].
This seris also includes fixes for OTAP/ITAP delay values in j721e_4bit_sdhci_set_ios_post and for HS400 mode.
For DDR52 mode, also set ENDLL=1 and call am654_sdhci_setup_dll() instead of am654_sdhci_setup_delay_chain() according to device datasheet[1].
[...]
Applied to u-boot/master, thanks!
participants (2)
-
Judith Mendez
-
Tom Rini