Skip to contentSkip to conversation

mmc: omap_hsmmc: Restore the init_stream sequence

PATCH Applied Mathieu Othacehe 2 participants 5 messages Apr 10, 2025 – Apr 23, 2025 MMC, SD and SDIO

lore
Share & tools
lore
#1 lore raw

PATCH 1/2 mmc: Add a new callback function to perform the 74 clocks cycle sequence

In reply to Mathieu Othacehe

From: Jean-Jacques Hiblot <[email protected]>

Add a new callback function *send_init_stream* which start a sequence of at least 74 clock cycles.
The mmc core uses *mmc_send_init_stream* in order to invoke the callback function. This will be used during power cycle where the specification requires such a sequence after power up.

---
 drivers/mmc/mmc-uclass.c | 13 +++++++++++++
 drivers/mmc/mmc.c        |  6 ++++++
 include/mmc.h            |  9 +++++++++
 3 files changed, 28 insertions(+)
drivers/mmc/mmc-uclass.c +13 -0
index c8db4f811c2..ca5bf1d11e6 100644
@@ -83,6 +83,19 @@ int mmc_wait_dat0(struct mmc *mmc, int state, int timeout_us)
 	return dm_mmc_wait_dat0(mmc->dev, state, timeout_us);
 }
 
+void dm_mmc_send_init_stream(struct udevice *dev)
+{
+	struct dm_mmc_ops *ops = mmc_get_ops(dev);
+
+	if (ops->send_init_stream)
+		ops->send_init_stream(dev);
+}
+
+void mmc_send_init_stream(struct mmc *mmc)
+{
+	dm_mmc_send_init_stream(mmc->dev);
+}
+
 static int dm_mmc_get_wp(struct udevice *dev)
 {
 	struct dm_mmc_ops *ops = mmc_get_ops(dev);
drivers/mmc/mmc.c +6 -0
index 31a72366206..086826c668c 100644
@@ -1663,6 +1663,10 @@ static int mmc_execute_tuning(struct mmc *mmc, uint opcode)
 }
 #endif
 
+static void mmc_send_init_stream(struct mmc *mmc)
+{
+}
+
 static int mmc_set_ios(struct mmc *mmc)
 {
 	int ret = 0;
@@ -2929,6 +2933,8 @@ int mmc_get_op_cond(struct mmc *mmc, bool quiet)
 retry:
 	mmc_set_initial_state(mmc);
 
+	mmc_send_init_stream(mmc);
+
 	/* Reset the Card */
 	err = mmc_go_idle(mmc);
 
include/mmc.h +9 -0
index 81bccb4cf12..c2d275cdc3e 100644
@@ -487,6 +487,14 @@ struct dm_mmc_ops {
 	 */
 	int (*set_ios)(struct udevice *dev);
 
+	/**
+	 * send_init_stream() - send the initialization stream: 74 clock cycles
+	 * This is used after power up before sending the first command
+	 *
+	 * @dev:	Device to update
+	 */
+	void (*send_init_stream)(struct udevice *dev);
+
 	/**
 	 * get_cd() - See whether a card is present
 	 *
@@ -566,6 +574,7 @@ struct dm_mmc_ops {
 
 /* Transition functions for compatibility */
 int mmc_set_ios(struct mmc *mmc);
+void mmc_send_init_stream(struct mmc *mmc);
 int mmc_getcd(struct mmc *mmc);
 int mmc_getwp(struct mmc *mmc);
 int mmc_execute_tuning(struct mmc *mmc, uint opcode);
Signature · 1 line
2.47.1
#2 lore raw

PATCH 2/2 mmc: omap_hsmmc: implement send_init_stream callback

In reply to Mathieu Othacehe

This callback is used to send the 74 clock cycles after power up.

---
 drivers/mmc/omap_hsmmc.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
drivers/mmc/omap_hsmmc.c +11 -2
index e66ab25d02a..92bc72b267c 100644
@@ -780,6 +780,14 @@ tuning_error:
 	return ret;
 }
 #endif
+
+static void omap_hsmmc_send_init_stream(struct udevice *dev)
+{
+	struct omap_hsmmc_data *priv = dev_get_priv(dev);
+	struct hsmmc *mmc_base = priv->base_addr;
+
+	mmc_init_stream(mmc_base);
+}
 #endif
 
 static void mmc_enable_irq(struct mmc *mmc, struct mmc_cmd *cmd)
@@ -1515,9 +1523,10 @@ static const struct dm_mmc_ops omap_hsmmc_ops = {
 	.get_wp		= omap_hsmmc_getwp,
 #endif
 #if CONFIG_IS_ENABLED(MMC_SUPPORTS_TUNING)
-	.execute_tuning = omap_hsmmc_execute_tuning,
+	.execute_tuning    = omap_hsmmc_execute_tuning,
 #endif
-	.wait_dat0	= omap_hsmmc_wait_dat0,
+	.send_init_stream  = omap_hsmmc_send_init_stream,
+	.wait_dat0	   = omap_hsmmc_wait_dat0,
 };
 #else
 static const struct mmc_ops omap_hsmmc_ops = {
Signature · 1 line
2.47.1
#0 lore raw

cover mmc: omap_hsmmc: Restore the init_stream sequence

Hello,

Back in 2019, the init_stream sequence was disabled with db52e19ced because:

This is not required. The MMC core sends CMD0 right after the initialization and it serves the same purpose.

That is wrong. It does not serve the same purpose at all. The init_stream function role is to keep the CMD line high for 74 clock cycles which is required by the SD specification[1]:

A device shall be ready to accept the first command within 1ms from detecting VDD min. Device may use up to 74 clocks for preparation before receiving the first command.

It turns out that one of the devices I am speaking to is requiring those 74 clocks sequence before the send of the CMD0 as described in the specification.

It was then broken since 2019.

I guess that most of the other devices out there are able to cope with those 74 clocks sequence missing and can respond directly to CMD0, which is why that went unnoticed with the omap_hsmmc driver users.

I am proposing to restore that sequence, which is also used on the Linux side.

Thanks,

Mathieu

[1]: https://academy.cba.mit.edu/classes/networking_communications/SD/SD.pdf

Jean-Jacques Hiblot (1):
mmc: Add a new callback function to perform the 74 clocks cycle
sequence

Mathieu Othacehe (1):
mmc: omap_hsmmc: implement send_init_stream callback

 drivers/mmc/mmc-uclass.c | 13 +++++++++++++
 drivers/mmc/mmc.c        |  6 ++++++
 drivers/mmc/omap_hsmmc.c | 13 +++++++++++--
 include/mmc.h            |  9 +++++++++
 4 files changed, 39 insertions(+), 2 deletions(-)
Signature · 1 line
2.47.1
#3 lore raw

Hi Mathieu,

Quoted 14 lines · Mathieu Othacehe

Hello,

Back in 2019, the init_stream sequence was disabled with db52e19ced because:

This is not required. The MMC core sends CMD0 right after the initialization and it serves the same purpose.

That is wrong. It does not serve the same purpose at all. The init_stream function role is to keep the CMD line high for 74 clock cycles which is required by the SD specification[1]:

A device shall be ready to accept the first command within 1ms from detecting VDD min. Device may use up to 74 clocks for preparation before receiving the first command.

Quoted 8 lines

It turns out that one of the devices I am speaking to is requiring those 74 clocks sequence before the send of the CMD0 as described in the specification.

It was then broken since 2019.

I guess that most of the other devices out there are able to cope with those 74 clocks sequence missing and can respond directly to CMD0, which is why that went unnoticed with the omap_hsmmc driver users.

This patchset looks good to me.

But before apply the patchset. I would like see whether Jean has any comments.

Thanks,
Peng

Quoted 25 lines

I am proposing to restore that sequence, which is also used on the Linux side.

Thanks,

Mathieu

[1]: https://academy.cba.mit.edu/classes/networking_communications/SD/SD.pdf

Jean-Jacques Hiblot (1):
mmc: Add a new callback function to perform the 74 clocks cycle
sequence

Mathieu Othacehe (1):
mmc: omap_hsmmc: implement send_init_stream callback

drivers/mmc/mmc-uclass.c | 13 +++++++++++++
drivers/mmc/mmc.c        |  6 ++++++
drivers/mmc/omap_hsmmc.c | 13 +++++++++++--
include/mmc.h            |  9 +++++++++
4 files changed, 39 insertions(+), 2 deletions(-)
Signature · 1 line
2.47.1
#4 lore raw
In reply to Mathieu Othacehe

From: Peng Fan <[email protected]>

Quoted 10 lines · Mathieu Othacehe

Back in 2019, the init_stream sequence was disabled with db52e19ced because:

This is not required. The MMC core sends CMD0 right after the initialization and it serves the same purpose.

That is wrong. It does not serve the same purpose at all. The init_stream function role is to keep the CMD line high for 74 clock cycles which is required by the SD specification[1]:

[...]

Applied, thanks!

[1/2] mmc: Add a new callback function to perform the 74 clocks cycle sequence

      commit: d770e325e9d2fd3ba94481fc6ee41c25e799b325
[2/2] mmc: omap_hsmmc: implement send_init_stream callback
      commit: 991c9530fc0d52229e6196f6be2dd6e1ef17857c

Best regards,

Signature · 1 line
Peng Fan <[email protected]>