Skip to contentSkip to conversation

mmc: sdhci-pci: Add AMD HS200 mode tuning function

PATCH In review Wan Zongshun 5 participants 10 messages Dec 22, 2015 – Jan 15, 2016

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/3 mmc: sdhci-pci: Add AMD HS200 mode tuning function

From: Wan Zongshun <[email protected]>

This patch is to add software tuning functions for AMD hs200 mode.

---
 drivers/mmc/host/sdhci-pci-core.c | 146 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 146 insertions(+)
drivers/mmc/host/sdhci-pci-core.c +146 -0
index 08f4a9f..01c5723 100644
@@ -729,6 +729,152 @@ enum amd_chipset_gen {
 	AMD_CHIPSET_UNKNOWN,
 };
 
+struct tuning_descriptor {
+	unsigned char tune_around;
+	bool this_tune_ok;
+	bool last_tune_ok;
+	bool valid_front_end;
+	unsigned char valid_front;
+	unsigned char valid_window_max;
+	unsigned char tune_low_max;
+	unsigned char tune_low;
+	unsigned char valid_window;
+	unsigned char tune_result;
+};
+
+#define AMD_EMMC_TUNE_REG 0xb8
+static struct tuning_descriptor tdescriptor;
+
+static int tuning_reset(struct sdhci_host *host)
+{
+	unsigned int val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	val = sdhci_readw(host, SDHCI_HOST_CONTROL2);
+	val |= SDHCI_CTRL_PRESET_VAL_ENABLE | SDHCI_CTRL_EXEC_TUNING;
+	sdhci_writew(host, val, SDHCI_HOST_CONTROL2);
+
+	val = sdhci_readw(host, SDHCI_HOST_CONTROL2);
+	val &= ~SDHCI_CTRL_EXEC_TUNING;
+	sdhci_writew(host, val, SDHCI_HOST_CONTROL2);
+
+	spin_unlock_irqrestore(&host->lock, flags);
+
+	return 0;
+}
+
+static int config_tuning_phase(struct sdhci_host *host, unsigned char phase)
+{
+	struct sdhci_pci_slot *slot = sdhci_priv(host);
+	struct pci_dev *pdev = slot->chip->pdev;
+	unsigned int val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	pci_read_config_dword(pdev, AMD_EMMC_TUNE_REG, &val);
+	val &= ~0xf;
+	val |= (0x10800 | phase);
+	pci_write_config_dword(pdev, AMD_EMMC_TUNE_REG, val);
+
+	spin_unlock_irqrestore(&host->lock, flags);
+
+	return 0;
+}
+
+static int find_good_phase(struct sdhci_host *host)
+{
+	struct tuning_descriptor *td = &tdescriptor;
+	struct sdhci_pci_slot *slot = sdhci_priv(host);
+	struct pci_dev *pdev = slot->chip->pdev;
+	unsigned int val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	if (td->this_tune_ok == false)
+		td->valid_front_end = 1;
+
+	if (td->valid_front_end)
+		td->valid_front = td->valid_front;
+	else if (td->this_tune_ok)
+		td->valid_front = td->valid_front + 1;
+
+	if ((!td->this_tune_ok && td->last_tune_ok) ||
+					(td->tune_around == 11)) {
+		if (td->valid_window > td->valid_window_max) {
+			td->valid_window_max = td->valid_window;
+			td->tune_low_max = td->tune_low;
+		}
+	}
+
+	if (td->this_tune_ok) {
+		if (!td->last_tune_ok)
+			td->tune_low = td->tune_around;
+		td->valid_window = td->valid_window + 1;
+	} else {
+		if (td->last_tune_ok)
+			td->valid_window = 0x0;
+	}
+
+	td->last_tune_ok = td->this_tune_ok;
+
+	if (td->tune_around == 11) {
+		if ((td->valid_front + td->valid_window) >
+						td->valid_window_max) {
+			if (td->valid_front > td->valid_window)
+				td->tune_result =
+				((td->valid_front - td->valid_window) >> 1);
+			else
+				td->tune_result = td->tune_low +
+				((td->valid_window + td->valid_front) >> 1);
+		} else {
+			td->tune_result =
+				td->tune_low_max + (td->valid_window_max >> 1);
+		}
+
+		if (td->tune_result > 0x0b)
+			td->tune_result = 0x0b;
+
+		pci_read_config_dword(pdev, AMD_EMMC_TUNE_REG, &val);
+		val &= ~0xf;
+		val |= (0x10800 | td->tune_result);
+		pci_write_config_dword(pdev, AMD_EMMC_TUNE_REG, val);
+	}
+
+	spin_unlock_irqrestore(&host->lock, flags);
+
+	return 0;
+}
+
+static int amd_execute_tuning(struct sdhci_host *host, u32 opcode)
+{
+	struct tuning_descriptor *td = &tdescriptor;
+
+	tuning_reset(host);
+
+	for (td->tune_around = 0; td->tune_around < 12; td->tune_around++) {
+
+		config_tuning_phase(host, td->tune_around);
+
+		if (mmc_send_tuning(host->mmc, opcode, NULL)) {
+			td->this_tune_ok = false;
+			host->mmc->need_retune = 0;
+			mdelay(4);
+		} else {
+			td->this_tune_ok = true;
+		}
+
+		find_good_phase(host);
+	}
+
+	host->mmc->retune_period = 0;
+
+	return 0;
+}
+
 static int amd_probe(struct sdhci_pci_chip *chip)
 {
 	struct pci_dev	*smbus_dev;
Signature · 1 line
1.9.1
#1 lore raw

PATCH 2/3 mmc: sdhci-pci: Add platform tuning callback for amd hs200 mode

From: Wan Zongshun <[email protected]>

AMD hs200 mode tuning mode is not compatible with standard tuning process, so we need .platform_execute_tuning callback support in sdhci-pci-core.c, this patch is to do:

1. Add platform_execute_tuning callback in sdhci_pci_slot.
2. Implement sdhci_pci_ops.platform_execute_tuning function.

---
Hi Ulf,

Though modifying sdhci_pci_ops to be not const that is easy to implement my requirement, I am not sure it is right to do this.

So I just follow sdhci_pci_select_drive_strength style to add new callback:
sdhci_pci_platform_execute_tuning.

But I also met trouble in sdhci_execute_tuning of sdhci.c, I have to suppose only sdhci_pci_platform_execute_tuning is returning -EPERM(current code, my assumption is right), so that those vendor that has no slot->platform_execute_tuning could be skipped and go next standard tuning process.

If you have better idea for my requirement, please correct me.

Thanks!
Wan Zongshun.

---
 drivers/mmc/host/sdhci-pci-core.c | 23 +++++++++++++++++++++++
 drivers/mmc/host/sdhci-pci.h      |  1 +
 drivers/mmc/host/sdhci.c          |  3 ++-
 3 files changed, 26 insertions(+), 1 deletion(-)
drivers/mmc/host/sdhci-pci-core.c +23 -0
index 01c5723..e7b2bbe 100644
@@ -905,8 +905,19 @@ static int amd_probe(struct sdhci_pci_chip *chip)
 	return 0;
 }
 
+static int amd_probe_slot(struct sdhci_pci_slot *slot)
+{
+	struct sdhci_host *host = slot->host;
+
+	if (host->quirks2 & SDHCI_QUIRK2_CLEAR_TRANSFERMODE_REG_BEFORE_CMD)
+		slot->platform_execute_tuning = amd_execute_tuning;
+
+	return 0;
+}
+
 static const struct sdhci_pci_fixes sdhci_amd = {
 	.probe		= amd_probe,
+	.probe_slot	= amd_probe_slot,
 };
 
 static const struct pci_device_id pci_ids[] = {
@@ -1508,6 +1519,17 @@ static int sdhci_pci_select_drive_strength(struct sdhci_host *host,
 					   card_drv, drv_type);
 }
 
+static int sdhci_pci_platform_execute_tuning(struct sdhci_host *host,
+					     u32 opcode)
+{
+	struct sdhci_pci_slot *slot = sdhci_priv(host);
+
+	if (!slot->platform_execute_tuning)
+		return -EPERM;
+
+	return slot->platform_execute_tuning(host, opcode);
+}
+
 static const struct sdhci_ops sdhci_pci_ops = {
 	.set_clock	= sdhci_set_clock,
 	.enable_dma	= sdhci_pci_enable_dma,
@@ -1516,6 +1538,7 @@ static const struct sdhci_ops sdhci_pci_ops = {
 	.set_uhs_signaling = sdhci_set_uhs_signaling,
 	.hw_reset		= sdhci_pci_hw_reset,
 	.select_drive_strength	= sdhci_pci_select_drive_strength,
+	.platform_execute_tuning = sdhci_pci_platform_execute_tuning,
 };
 
 /*****************************************************************************\
drivers/mmc/host/sdhci-pci.h +1 -0
index d1a0b4d..48d98f1 100644
@@ -83,6 +83,7 @@ struct sdhci_pci_slot {
 				     struct mmc_card *card,
 				     unsigned int max_dtr, int host_drv,
 				     int card_drv, int *drv_type);
+	int (*platform_execute_tuning)(struct sdhci_host *host, u32 opcode);
 };
 
 struct sdhci_pci_chip {
drivers/mmc/host/sdhci.c +2 -1
index 2753b722d..2a5a6ce 100644
@@ -1937,7 +1937,8 @@ static int sdhci_execute_tuning(struct mmc_host *mmc, u32 opcode)
 		spin_unlock_irqrestore(&host->lock, flags);
 		err = host->ops->platform_execute_tuning(host, opcode);
 		sdhci_runtime_pm_put(host);
-		return err;
+		if (err != EPERM)
+			return err;
 	}
 
 	ctrl = sdhci_readw(host, SDHCI_HOST_CONTROL2);
Signature · 1 line
1.9.1
#2 lore raw

Re: [PATCH 2/3] mmc: sdhci-pci: Add platform tuning callback for amd hs200 mode

Quoted 61 lines · Wan Zongshun

From: Wan Zongshun <[email protected]>

AMD hs200 mode tuning mode is not compatible with standard tuning process, so we need .platform_execute_tuning callback support in sdhci-pci-core.c, this patch is to do:

1. Add platform_execute_tuning callback in sdhci_pci_slot.
2. Implement sdhci_pci_ops.platform_execute_tuning function.

---
Hi Ulf,

Though modifying sdhci_pci_ops to be not const that is easy to implement my requirement, I am not sure it is right to do this.

So I just follow sdhci_pci_select_drive_strength style to add new callback:
sdhci_pci_platform_execute_tuning.

But I also met trouble in sdhci_execute_tuning of sdhci.c, I have to suppose only sdhci_pci_platform_execute_tuning is returning -EPERM(current code, my assumption is right), so that those vendor that has no slot->platform_execute_tuning could be skipped and go next standard tuning process.

If you have better idea for my requirement, please correct me.

Thanks!
Wan Zongshun.

---
 drivers/mmc/host/sdhci-pci-core.c | 23 +++++++++++++++++++++++
 drivers/mmc/host/sdhci-pci.h      |  1 +
 drivers/mmc/host/sdhci.c          |  3 ++-
 3 files changed, 26 insertions(+), 1 deletion(-)
drivers/mmc/host/sdhci-pci-core.c +11 -0
index 01c5723..e7b2bbe 100644
@@ -905,8 +905,19 @@ static int amd_probe(struct sdhci_pci_chip *chip)
        return 0;
 }
 
+static int amd_probe_slot(struct sdhci_pci_slot *slot)
+{
+       struct sdhci_host *host = slot->host;
+
+       if (host->quirks2 & SDHCI_QUIRK2_CLEAR_TRANSFERMODE_REG_BEFORE_CMD)
+               slot->platform_execute_tuning = amd_execute_tuning;
+
+       return 0;
+}
+
 static const struct sdhci_pci_fixes sdhci_amd = {
        .probe          = amd_probe,
+       .probe_slot     = amd_probe_slot,
 };
 
 static const struct pci_device_id pci_ids[] = {

drivers/mmc/host/sdhci-pci-core.c static int sdhci_pci_select_drive_strength(struct sdhci_host *host, · line 1528

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +7 -0
@@ -1508,6 +1519,17 @@ static int sdhci_pci_select_drive_strength(struct sdhci_host *host,
                                           card_drv, drv_type);
 }
 
+static int sdhci_pci_platform_execute_tuning(struct sdhci_host *host,
+                                            u32 opcode)
+{
+       struct sdhci_pci_slot *slot = sdhci_priv(host);
+
+       if (!slot->platform_execute_tuning)
+               return -EPERM;

Here you return an error code.

Quoted 32 lines
drivers/mmc/host/sdhci-pci-core.c +5 -0
+
+       return slot->platform_execute_tuning(host, opcode);
+}
+
 static const struct sdhci_ops sdhci_pci_ops = {
        .set_clock      = sdhci_set_clock,
        .enable_dma     = sdhci_pci_enable_dma,
@@ -1516,6 +1538,7 @@ static const struct sdhci_ops sdhci_pci_ops = {
        .set_uhs_signaling = sdhci_set_uhs_signaling,
        .hw_reset               = sdhci_pci_hw_reset,
        .select_drive_strength  = sdhci_pci_select_drive_strength,
+       .platform_execute_tuning = sdhci_pci_platform_execute_tuning,
 };
 
 /*****************************************************************************\
drivers/mmc/host/sdhci-pci.h +1 -0
index d1a0b4d..48d98f1 100644
@@ -83,6 +83,7 @@ struct sdhci_pci_slot {
                                     struct mmc_card *card,
                                     unsigned int max_dtr, int host_drv,
                                     int card_drv, int *drv_type);
+       int (*platform_execute_tuning)(struct sdhci_host *host, u32 opcode);
 };
 
 struct sdhci_pci_chip {
drivers/mmc/host/sdhci.c +0 -0
index 2753b722d..2a5a6ce 100644

drivers/mmc/host/sdhci.c static int sdhci_execute_tuning(struct mmc_host *mmc, u32 opcode) · line 1940

Quoted 5 lines
drivers/mmc/host/sdhci.c +1 -1
@@ -1937,7 +1937,8 @@ static int sdhci_execute_tuning(struct mmc_host *mmc, u32 opcode)
                spin_unlock_irqrestore(&host->lock, flags);
                err = host->ops->platform_execute_tuning(host, opcode);
                sdhci_runtime_pm_put(host);
-               return err;
+               if (err != EPERM)

…and here you ignore it. Perhaps better to introduce various positive codes

#define AMD_TUNNING_DONE 0
#define AMD_TUNNING_NA 1

Quoted 12 lines
drivers/mmc/host/sdhci.c +1 -1
+                       return err;
        }
 
        ctrl = sdhci_readw(host, SDHCI_HOST_CONTROL2);
--

1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#3 lore raw

Re: [PATCH 2/3] mmc: sdhci-pci: Add platform tuning callback for amd hs200 mode

Quoted 61 lines
Quoted 8 lines
drivers/mmc/host/sdhci-pci-core.c +2 -0
+
 static const struct sdhci_pci_fixes sdhci_amd = {
        .probe          = amd_probe,
+       .probe_slot     = amd_probe_slot,
 };
 
 static const struct pci_device_id pci_ids[] = {

drivers/mmc/host/sdhci-pci-core.c static int sdhci_pci_select_drive_strength(struct sdhci_host *host, · line 1528

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +7 -0
@@ -1508,6 +1519,17 @@ static int sdhci_pci_select_drive_strength(struct sdhci_host *host,
                                           card_drv, drv_type);
 }
 
+static int sdhci_pci_platform_execute_tuning(struct sdhci_host *host,
+                                            u32 opcode)
+{
+       struct sdhci_pci_slot *slot = sdhci_priv(host);
+
+       if (!slot->platform_execute_tuning)
+               return -EPERM;

Here you return an error code.

Quoted 32 lines
drivers/mmc/host/sdhci-pci-core.c +5 -0
+
+       return slot->platform_execute_tuning(host, opcode);
+}
+
 static const struct sdhci_ops sdhci_pci_ops = {
        .set_clock      = sdhci_set_clock,
        .enable_dma     = sdhci_pci_enable_dma,
@@ -1516,6 +1538,7 @@ static const struct sdhci_ops sdhci_pci_ops = {
        .set_uhs_signaling = sdhci_set_uhs_signaling,
        .hw_reset               = sdhci_pci_hw_reset,
        .select_drive_strength  = sdhci_pci_select_drive_strength,
+       .platform_execute_tuning = sdhci_pci_platform_execute_tuning,
 };
 
 /*****************************************************************************\
drivers/mmc/host/sdhci-pci.h +1 -0
index d1a0b4d..48d98f1 100644
@@ -83,6 +83,7 @@ struct sdhci_pci_slot {
                                     struct mmc_card *card,
                                     unsigned int max_dtr, int host_drv,
                                     int card_drv, int *drv_type);
+       int (*platform_execute_tuning)(struct sdhci_host *host, u32 opcode);
 };
 
 struct sdhci_pci_chip {
drivers/mmc/host/sdhci.c +0 -0
index 2753b722d..2a5a6ce 100644

drivers/mmc/host/sdhci.c static int sdhci_execute_tuning(struct mmc_host *mmc, u32 opcode) · line 1940

Quoted 5 lines
drivers/mmc/host/sdhci.c +1 -1
@@ -1937,7 +1937,8 @@ static int sdhci_execute_tuning(struct mmc_host *mmc, u32 opcode)
                spin_unlock_irqrestore(&host->lock, flags);
                err = host->ops->platform_execute_tuning(host, opcode);
                sdhci_runtime_pm_put(host);
-               return err;
+               if (err != EPERM)

…and here you ignore it. Perhaps better to introduce various positive codes

Quoted 2 lines

#define AMD_TUNNING_DONE 0
#define AMD_TUNNING_NA 1

Andy, It looks a good idea, thanks.

Quoted 20 lines
Quoted 9 lines
drivers/mmc/host/sdhci.c +1 -1
+                       return err;
        }
 
        ctrl = sdhci_readw(host, SDHCI_HOST_CONTROL2);
--

1.9.1

Quoted 5 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

--
With Best Regards,
Andy Shevchenko

Signature · 3 lines
---
Vincent Wan(Zongshun)
www.mcuos.com
#4 lore raw

Re: [PATCH 2/3] mmc: sdhci-pci: Add platform tuning callback for amd hs200 mode

In reply to Wan Zongshun
Quoted 10 lines · Wan Zongshun

From: Wan Zongshun <[email protected]>

AMD hs200 mode tuning mode is not compatible with standard tuning process, so we need .platform_execute_tuning callback support in sdhci-pci-core.c, this patch is to do:

1. Add platform_execute_tuning callback in sdhci_pci_slot.
2. Implement sdhci_pci_ops.platform_execute_tuning function.

Quoted 8 lines

---
Hi Ulf,

Though modifying sdhci_pci_ops to be not const that is easy to implement my requirement, I am not sure it is right to do this.

So I just follow sdhci_pci_select_drive_strength style to add new callback:
sdhci_pci_platform_execute_tuning.

No thanks, please don't add new sdhci callbacks (or quirks).

Quoted 8 lines

But I also met trouble in sdhci_execute_tuning of sdhci.c, I have to suppose only sdhci_pci_platform_execute_tuning is returning -EPERM(current code, my assumption is right), so that those vendor that has no slot->platform_execute_tuning could be skipped and go next standard tuning process.

If you have better idea for my requirement, please correct me.

sdhci needs to become a set of library functions.

Typically the mmc_host_ops ->execute_tuning() callback for sdhci, should be assigned to a default function, unless the sdhci variant has assigned it to something else.

Yes, I realize that it requires core changes to sdhci to allow this. Although it's necessary do this conversion as I won't accept any more changes for sdhci that doesn't move the code into this direction.

Kind regards
Uffe

Quoted 89 lines

Thanks!
Wan Zongshun.

---
 drivers/mmc/host/sdhci-pci-core.c | 23 +++++++++++++++++++++++
 drivers/mmc/host/sdhci-pci.h      |  1 +
 drivers/mmc/host/sdhci.c          |  3 ++-
 3 files changed, 26 insertions(+), 1 deletion(-)
drivers/mmc/host/sdhci-pci-core.c +23 -0
index 01c5723..e7b2bbe 100644
@@ -905,8 +905,19 @@ static int amd_probe(struct sdhci_pci_chip *chip)
        return 0;
 }
 
+static int amd_probe_slot(struct sdhci_pci_slot *slot)
+{
+       struct sdhci_host *host = slot->host;
+
+       if (host->quirks2 & SDHCI_QUIRK2_CLEAR_TRANSFERMODE_REG_BEFORE_CMD)
+               slot->platform_execute_tuning = amd_execute_tuning;
+
+       return 0;
+}
+
 static const struct sdhci_pci_fixes sdhci_amd = {
        .probe          = amd_probe,
+       .probe_slot     = amd_probe_slot,
 };
 
 static const struct pci_device_id pci_ids[] = {
@@ -1508,6 +1519,17 @@ static int sdhci_pci_select_drive_strength(struct sdhci_host *host,
                                           card_drv, drv_type);
 }
 
+static int sdhci_pci_platform_execute_tuning(struct sdhci_host *host,
+                                            u32 opcode)
+{
+       struct sdhci_pci_slot *slot = sdhci_priv(host);
+
+       if (!slot->platform_execute_tuning)
+               return -EPERM;
+
+       return slot->platform_execute_tuning(host, opcode);
+}
+
 static const struct sdhci_ops sdhci_pci_ops = {
        .set_clock      = sdhci_set_clock,
        .enable_dma     = sdhci_pci_enable_dma,
@@ -1516,6 +1538,7 @@ static const struct sdhci_ops sdhci_pci_ops = {
        .set_uhs_signaling = sdhci_set_uhs_signaling,
        .hw_reset               = sdhci_pci_hw_reset,
        .select_drive_strength  = sdhci_pci_select_drive_strength,
+       .platform_execute_tuning = sdhci_pci_platform_execute_tuning,
 };
 
 /*****************************************************************************\
drivers/mmc/host/sdhci-pci.h +1 -0
index d1a0b4d..48d98f1 100644
@@ -83,6 +83,7 @@ struct sdhci_pci_slot {
                                     struct mmc_card *card,
                                     unsigned int max_dtr, int host_drv,
                                     int card_drv, int *drv_type);
+       int (*platform_execute_tuning)(struct sdhci_host *host, u32 opcode);
 };
 
 struct sdhci_pci_chip {
drivers/mmc/host/sdhci.c +2 -2
index 2753b722d..2a5a6ce 100644
@@ -1937,7 +1937,8 @@ static int sdhci_execute_tuning(struct mmc_host *mmc, u32 opcode)
                spin_unlock_irqrestore(&host->lock, flags);
                err = host->ops->platform_execute_tuning(host, opcode);
                sdhci_runtime_pm_put(host);
-               return err;
+               if (err != EPERM)
+                       return err;
        }
 
        ctrl = sdhci_readw(host, SDHCI_HOST_CONTROL2);
--

1.9.1

#5 lore raw

Re: [PATCH 2/3] mmc: sdhci-pci: Add platform tuning callback for amd hs200 mode

Quoted 11 lines
Quoted 8 lines

But I also met trouble in sdhci_execute_tuning of sdhci.c, I have to suppose only sdhci_pci_platform_execute_tuning is returning -EPERM(current code, my assumption is right), so that those vendor that has no slot->platform_execute_tuning could be skipped and go next standard tuning process.

If you have better idea for my requirement, please correct me.

sdhci needs to become a set of library functions.

Quoted 8 lines

Typically the mmc_host_ops ->execute_tuning() callback for sdhci, should be assigned to a default function, unless the sdhci variant has assigned it to something else.

Yes, I realize that it requires core changes to sdhci to allow this. Although it's necessary do this conversion as I won't accept any more changes for sdhci that doesn't move the code into this direction.

Ulf,

Then Can you point me what's my next step for submitting tuning workaround for AMD emmc4.5 driver?

What your mean is you will change sdhci-pci-core.c to a core and library function? And then I can implement a AMD specific emmc-pci driver call to those libs?

Quoted 21 lines

Kind regards
Uffe

Quoted 18 lines

Thanks!
Wan Zongshun.

---
  drivers/mmc/host/sdhci-pci-core.c | 23 +++++++++++++++++++++++
  drivers/mmc/host/sdhci-pci.h      |  1 +
  drivers/mmc/host/sdhci.c          |  3 ++-
  3 files changed, 26 insertions(+), 1 deletion(-)
drivers/mmc/host/sdhci-pci-core.c +0 -0
index 01c5723..e7b2bbe 100644
@@ -905,8 +905,19 @@ static int amd_probe(struct sdhci_pci_chip *chip)
         return 0;
  }
 
#6 lore raw

Re: [PATCH 2/3] mmc: sdhci-pci: Add platform tuning callback for amd hs200 mode

Quoted 22 lines · Wan Zongshun
Quoted 13 lines
Quoted 9 lines

But I also met trouble in sdhci_execute_tuning of sdhci.c, I have to suppose
only sdhci_pci_platform_execute_tuning is returning -EPERM(current code, my assumption is right), so that those vendor that has no slot->platform_execute_tuning could be skipped and go next standard tuning process.

If you have better idea for my requirement, please correct me.

sdhci needs to become a set of library functions.

Quoted 8 lines

Typically the mmc_host_ops ->execute_tuning() callback for sdhci, should be assigned to a default function, unless the sdhci variant has assigned it to something else.

Yes, I realize that it requires core changes to sdhci to allow this. Although it's necessary do this conversion as I won't accept any more changes for sdhci that doesn't move the code into this direction.

Quoted 7 lines

Ulf,

Then Can you point me what's my next step for submitting tuning workaround for AMD emmc4.5 driver?

What your mean is you will change sdhci-pci-core.c to a core and library function?

Not me personally as I don't have the bandwidth to do it. Anybody that cares about sdhci are encouraged to give it a try!

Quoted 1 line

And then I can implement a AMD specific emmc-pci driver call to those libs?

The lib should provide common functionality needed among sdhci variants.

If there specific needs for any sdhci variant, that variant should implement that part separately without affecting other variants. This isn't the case when adding an sdhci callback/quirk to the sdhci core, as that would affect more or less *all* sdhci variants.

[...]

Kind regards
Uffe

#7 lore raw

PATCH 3/3 mmc: sdhci-pci: enable AMD emmc hs200 mode

In reply to Wan Zongshun

From: Wan Zongshun <[email protected]>

This patch is to remove the SDHCI_QUIRK2_BROKEN_HS200 quirk and enable the emmc hs200 mode for AMD current platforms.

---
 drivers/mmc/host/sdhci-pci-core.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/mmc/host/sdhci-pci-core.c +1 -3
index e7b2bbe..17697fd 100644
@@ -897,10 +897,8 @@ static int amd_probe(struct sdhci_pci_chip *chip)
 		}
 	}
 
-	if ((gen == AMD_CHIPSET_BEFORE_ML) || (gen == AMD_CHIPSET_CZ)) {
+	if ((gen == AMD_CHIPSET_BEFORE_ML) || (gen == AMD_CHIPSET_CZ))
 		chip->quirks2 |= SDHCI_QUIRK2_CLEAR_TRANSFERMODE_REG_BEFORE_CMD;
-		chip->quirks2 |= SDHCI_QUIRK2_BROKEN_HS200;
-	}
 
 	return 0;
 }
Signature · 1 line
1.9.1
#8 lore raw
In reply to Wan Zongshun
Quoted 23 lines · Wan Zongshun

From: Wan Zongshun <[email protected]>

This patch is to add software tuning functions for AMD hs200 mode.

---
 drivers/mmc/host/sdhci-pci-core.c | 146 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 146 insertions(+)
drivers/mmc/host/sdhci-pci-core.c +5 -0
index 08f4a9f..01c5723 100644
@@ -729,6 +729,152 @@ enum amd_chipset_gen {
        AMD_CHIPSET_UNKNOWN,
 };
 
+struct tuning_descriptor {
+       unsigned char tune_around;
+       bool this_tune_ok;
+       bool last_tune_ok;
+       bool valid_front_end;

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 746

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+       unsigned char valid_front;
+       unsigned char valid_window_max;
+       unsigned char tune_low_max;
+       unsigned char tune_low;
+       unsigned char valid_window;
+       unsigned char tune_result;
+};
+
+#define AMD_EMMC_TUNE_REG 0xb8
+static struct tuning_descriptor tdescriptor;

Global variable?!

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 748

Quoted 2 lines
drivers/mmc/host/sdhci-pci-core.c +2 -0
+
+static int tuning_reset(struct sdhci_host *host)

Better prefixes?

Quoted 21 lines
drivers/mmc/host/sdhci-pci-core.c +21 -0
+{
+       unsigned int val;
+       unsigned long flags;
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       val = sdhci_readw(host, SDHCI_HOST_CONTROL2);
+       val |= SDHCI_CTRL_PRESET_VAL_ENABLE | SDHCI_CTRL_EXEC_TUNING;
+       sdhci_writew(host, val, SDHCI_HOST_CONTROL2);
+
+       val = sdhci_readw(host, SDHCI_HOST_CONTROL2);
+       val &= ~SDHCI_CTRL_EXEC_TUNING;
+       sdhci_writew(host, val, SDHCI_HOST_CONTROL2);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+
+       return 0;
+}
+
+static int config_tuning_phase(struct sdhci_host *host, unsigned char phase)
+{

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 779

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+       struct sdhci_pci_slot *slot = sdhci_priv(host);
+       struct pci_dev *pdev = slot->chip->pdev;
+       unsigned int val;
+       unsigned long flags;
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       pci_read_config_dword(pdev, AMD_EMMC_TUNE_REG, &val);
+       val &= ~0xf;
+       val |= (0x10800 | phase);

Magic.

Quoted 17 lines
drivers/mmc/host/sdhci-pci-core.c +17 -0
+       pci_write_config_dword(pdev, AMD_EMMC_TUNE_REG, val);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+
+       return 0;
+}
+
+static int find_good_phase(struct sdhci_host *host)
+{
+       struct tuning_descriptor *td = &tdescriptor;
+       struct sdhci_pci_slot *slot = sdhci_priv(host);
+       struct pci_dev *pdev = slot->chip->pdev;
+       unsigned int val;
+       unsigned long flags;
+
+       spin_lock_irqsave(&host->lock, flags);
+

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 806

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+       if (td->this_tune_ok == false)
+               td->valid_front_end = 1;
+
+       if (td->valid_front_end)
+               td->valid_front = td->valid_front;
+       else if (td->this_tune_ok)
+               td->valid_front = td->valid_front + 1;
+
+       if ((!td->this_tune_ok && td->last_tune_ok) ||
+                                       (td->tune_around == 11)) {

Magic.

Quoted 27 lines
drivers/mmc/host/sdhci-pci-core.c +27 -0
+               if (td->valid_window > td->valid_window_max) {
+                       td->valid_window_max = td->valid_window;
+                       td->tune_low_max = td->tune_low;
+               }
+       }
+
+       if (td->this_tune_ok) {
+               if (!td->last_tune_ok)
+                       td->tune_low = td->tune_around;
+               td->valid_window = td->valid_window + 1;
+       } else {
+               if (td->last_tune_ok)
+                       td->valid_window = 0x0;
+       }
+
+       td->last_tune_ok = td->this_tune_ok;
+
+       if (td->tune_around == 11) {
+               if ((td->valid_front + td->valid_window) >
+                                               td->valid_window_max) {
+                       if (td->valid_front > td->valid_window)
+                               td->tune_result =
+                               ((td->valid_front - td->valid_window) >> 1);
+                       else
+                               td->tune_result = td->tune_low +
+                               ((td->valid_window + td->valid_front) >> 1);
+               } else {

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 843

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+                       td->tune_result =
+                               td->tune_low_max + (td->valid_window_max >> 1);
+               }
+
+               if (td->tune_result > 0x0b)
+                       td->tune_result = 0x0b;
+
+               pci_read_config_dword(pdev, AMD_EMMC_TUNE_REG, &val);
+               val &= ~0xf;
+               val |= (0x10800 | td->tune_result);

Magic.

Quoted 5 lines
drivers/mmc/host/sdhci-pci-core.c +5 -0
+               pci_write_config_dword(pdev, AMD_EMMC_TUNE_REG, val);
+       }
+
+       spin_unlock_irqrestore(&host->lock, flags);
+

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 858

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+       return 0;
+}
+
+static int amd_execute_tuning(struct sdhci_host *host, u32 opcode)
+{
+       struct tuning_descriptor *td = &tdescriptor;
+
+       tuning_reset(host);
+
+       for (td->tune_around = 0; td->tune_around < 12; td->tune_around++) {

Magic.

Why loop is done with non-local variable?

Quoted 12 lines
drivers/mmc/host/sdhci-pci-core.c +12 -0
+
+               config_tuning_phase(host, td->tune_around);
+
+               if (mmc_send_tuning(host->mmc, opcode, NULL)) {
+                       td->this_tune_ok = false;
+                       host->mmc->need_retune = 0;
+                       mdelay(4);
+               } else {
+                       td->this_tune_ok = true;
+               }
+
+               find_good_phase(host);

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 880

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +7 -0
+       }
+
+       host->mmc->retune_period = 0;
+
+       return 0;
+}
+
 static int amd_probe(struct sdhci_pci_chip *chip)
 {
        struct pci_dev  *smbus_dev;

No users for such code. I don't think it makes sense to push it separately.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#9 lore raw

2015-12-22 17:52 GMT+08:00 Andy Shevchenko <[email protected]>:

Quoted 28 lines · Wan Zongshun
Quoted 27 lines

From: Wan Zongshun <[email protected]>

This patch is to add software tuning functions for AMD hs200 mode.

---
 drivers/mmc/host/sdhci-pci-core.c | 146 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 146 insertions(+)
drivers/mmc/host/sdhci-pci-core.c +9 -0
index 08f4a9f..01c5723 100644
@@ -729,6 +729,152 @@ enum amd_chipset_gen {
        AMD_CHIPSET_UNKNOWN,
 };
 
+struct tuning_descriptor {
+       unsigned char tune_around;
+       bool this_tune_ok;
+       bool last_tune_ok;
+       bool valid_front_end;
+       unsigned char valid_front;
+       unsigned char valid_window_max;
+       unsigned char tune_low_max;
+       unsigned char tune_low;
Quoted 8 lines

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 746

Quoted 6 lines
drivers/mmc/host/sdhci-pci-core.c +6 -0
+       unsigned char valid_window;
+       unsigned char tune_result;
+};
+
+#define AMD_EMMC_TUNE_REG 0xb8
+static struct tuning_descriptor tdescriptor;

Global variable?!

Okay, will change it to local.

Quoted 5 lines

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 748

Quoted 3 lines
drivers/mmc/host/sdhci-pci-core.c +2 -0
+
+static int tuning_reset(struct sdhci_host *host)

Better prefixes?

Do you mean I should not name this function to tuning reset?

Quoted 26 lines
Quoted 26 lines
drivers/mmc/host/sdhci-pci-core.c +25 -0
+{
+       unsigned int val;
+       unsigned long flags;
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       val = sdhci_readw(host, SDHCI_HOST_CONTROL2);
+       val |= SDHCI_CTRL_PRESET_VAL_ENABLE | SDHCI_CTRL_EXEC_TUNING;
+       sdhci_writew(host, val, SDHCI_HOST_CONTROL2);
+
+       val = sdhci_readw(host, SDHCI_HOST_CONTROL2);
+       val &= ~SDHCI_CTRL_EXEC_TUNING;
+       sdhci_writew(host, val, SDHCI_HOST_CONTROL2);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+
+       return 0;
+}
+
+static int config_tuning_phase(struct sdhci_host *host, unsigned char phase)
+{
+       struct sdhci_pci_slot *slot = sdhci_priv(host);
+       struct pci_dev *pdev = slot->chip->pdev;
+       unsigned int val;
+       unsigned long flags;
Quoted 8 lines

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 779

Quoted 6 lines
drivers/mmc/host/sdhci-pci-core.c +6 -0
+
+       spin_lock_irqsave(&host->lock, flags);
+
+       pci_read_config_dword(pdev, AMD_EMMC_TUNE_REG, &val);
+       val &= ~0xf;
+       val |= (0x10800 | phase);

Magic.

Okay, I will make 0x10800 to be see more clearly, will define each bit Macro for it.

Quoted 89 lines
Quoted 18 lines
drivers/mmc/host/sdhci-pci-core.c +17 -0
+       pci_write_config_dword(pdev, AMD_EMMC_TUNE_REG, val);
+
+       spin_unlock_irqrestore(&host->lock, flags);
+
+       return 0;
+}
+
+static int find_good_phase(struct sdhci_host *host)
+{
+       struct tuning_descriptor *td = &tdescriptor;
+       struct sdhci_pci_slot *slot = sdhci_priv(host);
+       struct pci_dev *pdev = slot->chip->pdev;
+       unsigned int val;
+       unsigned long flags;
+
+       spin_lock_irqsave(&host->lock, flags);
+

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 806

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+       if (td->this_tune_ok == false)
+               td->valid_front_end = 1;
+
+       if (td->valid_front_end)
+               td->valid_front = td->valid_front;
+       else if (td->this_tune_ok)
+               td->valid_front = td->valid_front + 1;
+
+       if ((!td->this_tune_ok && td->last_tune_ok) ||
+                                       (td->tune_around == 11)) {

Magic.

Quoted 27 lines
drivers/mmc/host/sdhci-pci-core.c +27 -0
+               if (td->valid_window > td->valid_window_max) {
+                       td->valid_window_max = td->valid_window;
+                       td->tune_low_max = td->tune_low;
+               }
+       }
+
+       if (td->this_tune_ok) {
+               if (!td->last_tune_ok)
+                       td->tune_low = td->tune_around;
+               td->valid_window = td->valid_window + 1;
+       } else {
+               if (td->last_tune_ok)
+                       td->valid_window = 0x0;
+       }
+
+       td->last_tune_ok = td->this_tune_ok;
+
+       if (td->tune_around == 11) {
+               if ((td->valid_front + td->valid_window) >
+                                               td->valid_window_max) {
+                       if (td->valid_front > td->valid_window)
+                               td->tune_result =
+                               ((td->valid_front - td->valid_window) >> 1);
+                       else
+                               td->tune_result = td->tune_low +
+                               ((td->valid_window + td->valid_front) >> 1);
+               } else {

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 843

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+                       td->tune_result =
+                               td->tune_low_max + (td->valid_window_max >> 1);
+               }
+
+               if (td->tune_result > 0x0b)
+                       td->tune_result = 0x0b;
+
+               pci_read_config_dword(pdev, AMD_EMMC_TUNE_REG, &val);
+               val &= ~0xf;
+               val |= (0x10800 | td->tune_result);

Magic.

Quoted 5 lines
drivers/mmc/host/sdhci-pci-core.c +5 -0
+               pci_write_config_dword(pdev, AMD_EMMC_TUNE_REG, val);
+       }
+
+       spin_unlock_irqrestore(&host->lock, flags);
+

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 858

Quoted 10 lines
drivers/mmc/host/sdhci-pci-core.c +10 -0
+       return 0;
+}
+
+static int amd_execute_tuning(struct sdhci_host *host, u32 opcode)
+{
+       struct tuning_descriptor *td = &tdescriptor;
+
+       tuning_reset(host);
+
+       for (td->tune_around = 0; td->tune_around < 12; td->tune_around++) {

Magic.

Quoted 1 line

Why loop is done with non-local variable?

It will be changed at next version.
thanks!

Quoted 17 lines
Quoted 17 lines
drivers/mmc/host/sdhci-pci-core.c +16 -0
+
+               config_tuning_phase(host, td->tune_around);
+
+               if (mmc_send_tuning(host->mmc, opcode, NULL)) {
+                       td->this_tune_ok = false;
+                       host->mmc->need_retune = 0;
+                       mdelay(4);
+               } else {
+                       td->this_tune_ok = true;
+               }
+
+               find_good_phase(host);
+       }
+
+       host->mmc->retune_period = 0;
+
Quoted 8 lines

drivers/mmc/host/sdhci-pci-core.c enum amd_chipset_gen { · line 880

Quoted 6 lines
drivers/mmc/host/sdhci-pci-core.c +3 -0
+       return 0;
+}
+
 static int amd_probe(struct sdhci_pci_chip *chip)
 {
        struct pci_dev  *smbus_dev;

No users for such code. I don't think it makes sense to push it separately.

Since I am not sure the patch 2/3 is ok to every body, so it just is my try. If we can make decision for patch2/3, I can integrate them into one patch per your suggestion.

Thanks Andy.

Quoted 4 lines

--
With Best Regards,
Andy Shevchenko

Signature · 3 lines
---
Vincent Wan(Zongshun)
www.mcuos.com