Skip to contentSkip to conversation

mmc: sdhci-acpi: AMDI0040: Allow changing HS200/HS400 driver strength

PATCH In review Victor Ding 4 participants 4 messages Oct 27, 2020 – Dec 2, 2020

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

PATCH v2 1/1 mmc: sdhci-acpi: AMDI0040: Allow changing HS200/HS400 driver strength

From: Raul E Rangel <[email protected]>

This change will allow platform designers better control over signal integrity by allowing them to tune the HS200 and HS400 driver strengths.

The driver strength was previously hard coded to A to solve boot problems with certain platforms. This driver strength does not universally apply to all platforms so we need a knob to adjust it.

All older platforms currently have the SDR104 preset hard coded to A in the firmware. This means that switching from the hard coded value in the kernel to reading the SDR104 preset is a no-op for these platforms. Newer platforms will have properly set presets. So this change will support both new and old platforms.

---

Changes in v2:
By Victor Ding <[email protected]>
- Rebased the patch by using FIELD_GET for preset value bit masks.
- (No functional changes).

The original patch was developed by Raul E Rangel.
https://patchwork.kernel.org/project/linux-mmc/patch/20200928154718.2.Ic6b6031366f090393d00a53fd69e1ada31ceb29e@changeid/

 drivers/mmc/host/sdhci-acpi.c | 39 ++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)
drivers/mmc/host/sdhci-acpi.c +36 -3
index 54205e3be9e8..225cb34cf1b9 100644
@@ -5,6 +5,7 @@
  * Copyright (c) 2012, Intel Corporation.
  */
 
+#include <linux/bitfield.h>
 #include <linux/init.h>
 #include <linux/export.h>
 #include <linux/module.h>
@@ -545,10 +546,42 @@ struct amd_sdhci_host {
 
 static int amd_select_drive_strength(struct mmc_card *card,
 				     unsigned int max_dtr, int host_drv,
-				     int card_drv, int *drv_type)
+				     int card_drv, int *host_driver_strength)
 {
-	*drv_type = MMC_SET_DRIVER_TYPE_A;
-	return MMC_SET_DRIVER_TYPE_A;
+	struct sdhci_host *host = mmc_priv(card->host);
+	u16 preset, preset_driver_strength;
+
+	/*
+	 * This method is only called by mmc_select_hs200 so we only need to
+	 * read from the HS200 (SDR104) preset register.
+	 *
+	 * Firmware that has "invalid/default" presets return a driver strength
+	 * of A. This matches the previously hard coded value.
+	 */
+	preset = sdhci_readw(host, SDHCI_PRESET_FOR_SDR104);
+	preset_driver_strength = FIELD_GET(SDHCI_PRESET_DRV_MASK, preset);
+
+	/*
+	 * We want the controller driver strength to match the card's driver
+	 * strength so they have similar rise/fall times.
+	 *
+	 * The controller driver strength set by this method is sticky for all
+	 * timings after this method is called. This unfortunately means that
+	 * while HS400 tuning is in progress we end up with mismatched driver
+	 * strengths between the controller and the card. HS400 tuning requires
+	 * switching from HS400->DDR52->HS->HS200->HS400. So the driver mismatch
+	 * happens while in DDR52 and HS modes. This has not been observed to
+	 * cause problems. Enabling presets would fix this issue.
+	 */
+	*host_driver_strength = preset_driver_strength;
+
+	/*
+	 * The resulting card driver strength is only set when switching the
+	 * card's timing to HS200 or HS400. The card will use the default driver
+	 * strength (B) for any other mode.
+	 */
+	return preset_driver_strength;
+
 }
 
 static void sdhci_acpi_amd_hs400_dll(struct sdhci_host *host, bool enable)
Signature · 1 line
2.29.0.rc2.309.g374f81d7ae-goog
maintainer Acked-by on v2 1/1
#1 lore raw
Quoted 9 lines · Victor Ding

From: Raul E Rangel <[email protected]>

This change will allow platform designers better control over signal integrity by allowing them to tune the HS200 and HS400 driver strengths.

The driver strength was previously hard coded to A to solve boot problems with certain platforms. This driver strength does not universally apply to all platforms so we need a knob to adjust it.

Quoted 8 lines

All older platforms currently have the SDR104 preset hard coded to A in the firmware. This means that switching from the hard coded value in the kernel to reading the SDR104 preset is a no-op for these platforms. Newer platforms will have properly set presets. So this change will support both new and old platforms.

Apart from unnecessary blank line after "return preset_driver_strength;"

Quoted 59 lines

---

Changes in v2:
By Victor Ding <[email protected]>
- Rebased the patch by using FIELD_GET for preset value bit masks.
- (No functional changes).

The original patch was developed by Raul E Rangel.
https://patchwork.kernel.org/project/linux-mmc/patch/20200928154718.2.Ic6b6031366f090393d00a53fd69e1ada31ceb29e@changeid/

 drivers/mmc/host/sdhci-acpi.c | 39 ++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)
drivers/mmc/host/sdhci-acpi.c +26 -3
index 54205e3be9e8..225cb34cf1b9 100644
@@ -5,6 +5,7 @@
  * Copyright (c) 2012, Intel Corporation.
  */
 
+#include <linux/bitfield.h>
 #include <linux/init.h>
 #include <linux/export.h>
 #include <linux/module.h>
@@ -545,10 +546,42 @@ struct amd_sdhci_host {
 
 static int amd_select_drive_strength(struct mmc_card *card,
 				     unsigned int max_dtr, int host_drv,
-				     int card_drv, int *drv_type)
+				     int card_drv, int *host_driver_strength)
 {
-	*drv_type = MMC_SET_DRIVER_TYPE_A;
-	return MMC_SET_DRIVER_TYPE_A;
+	struct sdhci_host *host = mmc_priv(card->host);
+	u16 preset, preset_driver_strength;
+
+	/*
+	 * This method is only called by mmc_select_hs200 so we only need to
+	 * read from the HS200 (SDR104) preset register.
+	 *
+	 * Firmware that has "invalid/default" presets return a driver strength
+	 * of A. This matches the previously hard coded value.
+	 */
+	preset = sdhci_readw(host, SDHCI_PRESET_FOR_SDR104);
+	preset_driver_strength = FIELD_GET(SDHCI_PRESET_DRV_MASK, preset);
+
+	/*
+	 * We want the controller driver strength to match the card's driver
+	 * strength so they have similar rise/fall times.
+	 *
+	 * The controller driver strength set by this method is sticky for all
+	 * timings after this method is called. This unfortunately means that
+	 * while HS400 tuning is in progress we end up with mismatched driver
+	 * strengths between the controller and the card. HS400 tuning requires
+	 * switching from HS400->DDR52->HS->HS200->HS400. So the driver mismatch
+	 * happens while in DDR52 and HS modes. This has not been observed to
+	 * cause problems. Enabling presets would fix this issue.

drivers/mmc/host/sdhci-acpi.c struct amd_sdhci_host { · line 584

Quoted 10 lines
drivers/mmc/host/sdhci-acpi.c +10 -0
+	 */
+	*host_driver_strength = preset_driver_strength;
+
+	/*
+	 * The resulting card driver strength is only set when switching the
+	 * card's timing to HS200 or HS400. The card will use the default driver
+	 * strength (B) for any other mode.
+	 */
+	return preset_driver_strength;
+

Unnecessary blank line.

Quoted 4 lines
drivers/mmc/host/sdhci-acpi.c +0 -0
 }
 
 static void sdhci_acpi_amd_hs400_dll(struct sdhci_host *host, bool enable)
#2 lore raw
In reply to Victor Ding
Quoted 10 lines · Victor Ding

From: Raul E Rangel <[email protected]>

This change will allow platform designers better control over signal integrity by allowing them to tune the HS200 and HS400 driver strengths.

The driver strength was previously hard coded to A to solve boot problems with certain platforms. This driver strength does not universally apply to all platforms so we need a knob to adjust it.

Quoted 8 lines

All older platforms currently have the SDR104 preset hard coded to A in the firmware. This means that switching from the hard coded value in the kernel to reading the SDR104 preset is a no-op for these platforms. Newer platforms will have properly set presets. So this change will support both new and old platforms.

Applied for next, thanks!

Note that I amended the patch to fix the white-space issue, as pointed out by Adrian.

Kind regards
Uffe

Quoted 75 lines

---

Changes in v2:
By Victor Ding <[email protected]>
- Rebased the patch by using FIELD_GET for preset value bit masks.
- (No functional changes).

The original patch was developed by Raul E Rangel.
https://patchwork.kernel.org/project/linux-mmc/patch/20200928154718.2.Ic6b6031366f090393d00a53fd69e1ada31ceb29e@changeid/

 drivers/mmc/host/sdhci-acpi.c | 39 ++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)
drivers/mmc/host/sdhci-acpi.c +36 -4
index 54205e3be9e8..225cb34cf1b9 100644
@@ -5,6 +5,7 @@
  * Copyright (c) 2012, Intel Corporation.
  */
 
+#include <linux/bitfield.h>
 #include <linux/init.h>
 #include <linux/export.h>
 #include <linux/module.h>
@@ -545,10 +546,42 @@ struct amd_sdhci_host {
 
 static int amd_select_drive_strength(struct mmc_card *card,
                                     unsigned int max_dtr, int host_drv,
-                                    int card_drv, int *drv_type)
+                                    int card_drv, int *host_driver_strength)
 {
-       *drv_type = MMC_SET_DRIVER_TYPE_A;
-       return MMC_SET_DRIVER_TYPE_A;
+       struct sdhci_host *host = mmc_priv(card->host);
+       u16 preset, preset_driver_strength;
+
+       /*
+        * This method is only called by mmc_select_hs200 so we only need to
+        * read from the HS200 (SDR104) preset register.
+        *
+        * Firmware that has "invalid/default" presets return a driver strength
+        * of A. This matches the previously hard coded value.
+        */
+       preset = sdhci_readw(host, SDHCI_PRESET_FOR_SDR104);
+       preset_driver_strength = FIELD_GET(SDHCI_PRESET_DRV_MASK, preset);
+
+       /*
+        * We want the controller driver strength to match the card's driver
+        * strength so they have similar rise/fall times.
+        *
+        * The controller driver strength set by this method is sticky for all
+        * timings after this method is called. This unfortunately means that
+        * while HS400 tuning is in progress we end up with mismatched driver
+        * strengths between the controller and the card. HS400 tuning requires
+        * switching from HS400->DDR52->HS->HS200->HS400. So the driver mismatch
+        * happens while in DDR52 and HS modes. This has not been observed to
+        * cause problems. Enabling presets would fix this issue.
+        */
+       *host_driver_strength = preset_driver_strength;
+
+       /*
+        * The resulting card driver strength is only set when switching the
+        * card's timing to HS200 or HS400. The card will use the default driver
+        * strength (B) for any other mode.
+        */
+       return preset_driver_strength;
+
 }
 
 static void sdhci_acpi_amd_hs400_dll(struct sdhci_host *host, bool enable)
--

2.29.0.rc2.309.g374f81d7ae-goog

#3 lore raw

Thanks Victor and Ulf!

Quoted 105 lines · Ulf Hansson
Quoted 10 lines · Victor Ding

From: Raul E Rangel <[email protected]>

This change will allow platform designers better control over signal integrity by allowing them to tune the HS200 and HS400 driver strengths.

The driver strength was previously hard coded to A to solve boot problems with certain platforms. This driver strength does not universally apply to all platforms so we need a knob to adjust it.

Quoted 8 lines

All older platforms currently have the SDR104 preset hard coded to A in the firmware. This means that switching from the hard coded value in the kernel to reading the SDR104 preset is a no-op for these platforms. Newer platforms will have properly set presets. So this change will support both new and old platforms.

Applied for next, thanks!

Note that I amended the patch to fix the white-space issue, as pointed out by Adrian.

Kind regards
Uffe

Quoted 75 lines

---

Changes in v2:
By Victor Ding <[email protected]>
- Rebased the patch by using FIELD_GET for preset value bit masks.
- (No functional changes).

The original patch was developed by Raul E Rangel.
https://patchwork.kernel.org/project/linux-mmc/patch/20200928154718.2.Ic6b6031366f090393d00a53fd69e1ada31ceb29e@changeid/

 drivers/mmc/host/sdhci-acpi.c | 39 ++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)
drivers/mmc/host/sdhci-acpi.c +36 -4
index 54205e3be9e8..225cb34cf1b9 100644
@@ -5,6 +5,7 @@
  * Copyright (c) 2012, Intel Corporation.
  */
 
+#include <linux/bitfield.h>
 #include <linux/init.h>
 #include <linux/export.h>
 #include <linux/module.h>
@@ -545,10 +546,42 @@ struct amd_sdhci_host {
 
 static int amd_select_drive_strength(struct mmc_card *card,
                                     unsigned int max_dtr, int host_drv,
-                                    int card_drv, int *drv_type)
+                                    int card_drv, int *host_driver_strength)
 {
-       *drv_type = MMC_SET_DRIVER_TYPE_A;
-       return MMC_SET_DRIVER_TYPE_A;
+       struct sdhci_host *host = mmc_priv(card->host);
+       u16 preset, preset_driver_strength;
+
+       /*
+        * This method is only called by mmc_select_hs200 so we only need to
+        * read from the HS200 (SDR104) preset register.
+        *
+        * Firmware that has "invalid/default" presets return a driver strength
+        * of A. This matches the previously hard coded value.
+        */
+       preset = sdhci_readw(host, SDHCI_PRESET_FOR_SDR104);
+       preset_driver_strength = FIELD_GET(SDHCI_PRESET_DRV_MASK, preset);
+
+       /*
+        * We want the controller driver strength to match the card's driver
+        * strength so they have similar rise/fall times.
+        *
+        * The controller driver strength set by this method is sticky for all
+        * timings after this method is called. This unfortunately means that
+        * while HS400 tuning is in progress we end up with mismatched driver
+        * strengths between the controller and the card. HS400 tuning requires
+        * switching from HS400->DDR52->HS->HS200->HS400. So the driver mismatch
+        * happens while in DDR52 and HS modes. This has not been observed to
+        * cause problems. Enabling presets would fix this issue.
+        */
+       *host_driver_strength = preset_driver_strength;
+
+       /*
+        * The resulting card driver strength is only set when switching the
+        * card's timing to HS200 or HS400. The card will use the default driver
+        * strength (B) for any other mode.
+        */
+       return preset_driver_strength;
+
 }
 
 static void sdhci_acpi_amd_hs400_dll(struct sdhci_host *host, bool enable)
--

2.29.0.rc2.309.g374f81d7ae-goog