mmc: sdhci-of-dwcmshc: add reset control support in suspend/resume
PATCH 1/1 mmc: sdhci-of-dwcmshc: add reset control support in suspend/resume
The driver currently handles reset control only during initializations, but does not manage the controller reset signal during system suspend/resume. This can leave the hardware in an undefined state after resume. Additionally, the reset control is stored in vendor structs making it inaccessible for power management operations.
Move the reset control to dwcmshc_priv structure and add proper assertion during suspend and deassertion during resume. This ensures the controller is properly reset before entering low power state and correctly reinitialized upon wake.
---
Hello maintainers and reviewers,
This patch adds reset control support for the MMC controller during system suspend/resume.
Currently, the reset control is only handled during initialization and is stored in vendor-specific private structures. This makes it inaccessible to the common power management code and leaves the controller in an undefined state after resume.
The changes move the reset control to the generic dwcmshc_priv structure, update all platform-specific users (Rockchip, EIC7700), and add proper reset assertion/deassertion in dwcmshc_suspend/resume.
Kindly, please review.
Many thanks.
--
Regards,
Artem
drivers/mmc/host/sdhci-of-dwcmshc.c | 43 +++++++++++++++++------------ 1 file changed, 25 insertions(+), 18 deletions(-)
drivers/mmc/host/sdhci-of-dwcmshc.c +25 -18
@@ -258,13 +258,11 @@ enum dwcmshc_rk_type { };struct rk35xx_priv { -struct reset_control *reset;enum dwcmshc_rk_type devtype;u8 txclk_tapnum; };struct eic7700_priv { -struct reset_control *reset;unsigned int drive_impedance; };
@@ -272,6 +270,7 @@ struct eic7700_priv {struct dwcmshc_priv {struct clk *bus_clk; +struct reset_control *reset;int vendor_specific_area1;/* P_VENDOR_SPECIFIC_AREA1 reg */ int vendor_specific_area2;/* P_VENDOR_SPECIFIC_AREA2 reg */
@@ -829,16 +828,15 @@ static void rk35xx_sdhci_reset(struct sdhci_host *host, u8 mask) {struct sdhci_pltfm_host *pltfm_host =sdhci_priv (host);struct dwcmshc_priv *dwc_priv =sdhci_pltfm_priv (pltfm_host); -struct rk35xx_priv *priv = dwc_priv->priv;u32 extra =sdhci_readl (host,DECMSHC_EMMC_MISC_CON );if ((host->mmc->caps2 &MMC_CAP2_CQE ) && (mask &SDHCI_RESET_ALL ))cqhci_deactivate (host->mmc); -if (mask &SDHCI_RESET_ALL && priv->reset) { -reset_control_assert (priv->reset); +if (mask &SDHCI_RESET_ALL && dwc_priv->reset) { +reset_control_assert (dwc_priv->reset);udelay (1 ); -reset_control_deassert (priv->reset); +reset_control_deassert (dwc_priv->reset); }sdhci_reset (host, mask);
@@ -863,9 +861,9 @@ static int dwcmshc_rk35xx_init(struct device *dev, struct sdhci_host *host,else priv->devtype =DWCMSHC_RK3568 ; - priv->reset =devm_reset_control_array_get_optional_exclusive (mmc_dev (host->mmc)); -if (IS_ERR (priv->reset)) { - err =PTR_ERR (priv->reset); + dwc_priv->reset =devm_reset_control_array_get_optional_exclusive (mmc_dev (host->mmc)); +if (IS_ERR (dwc_priv->reset)) { + err =PTR_ERR (dwc_priv->reset);dev_err (mmc_dev (host->mmc),"failed to get reset control %d\n" , err);return err; }
@@ -1331,24 +1329,24 @@ static void sdhci_eic7700_reset(struct sdhci_host *host, u8 mask)sdhci_eic7700_config_phy (host); } -static int sdhci_eic7700_reset_init (struct device *dev,struct eic7700_priv *priv) +static int sdhci_eic7700_reset_init (struct device *dev,struct dwcmshc_priv *dwc_priv) {int ret; - priv->reset =devm_reset_control_array_get_optional_exclusive (dev); -if (IS_ERR (priv->reset)) { - ret =PTR_ERR (priv->reset); + dwc_priv->reset =devm_reset_control_array_get_optional_exclusive (dev); +if (IS_ERR (dwc_priv->reset)) { + ret =PTR_ERR (dwc_priv->reset);dev_err (dev,"failed to get reset control %d\n" , ret);return ret; } - ret =reset_control_assert (priv->reset); + ret =reset_control_assert (dwc_priv->reset);if (ret) {dev_err (dev,"Failed to assert reset signals: %d\n" , ret);return ret; }usleep_range (2000 ,2100 ); - ret =reset_control_deassert (priv->reset); + ret =reset_control_deassert (dwc_priv->reset);if (ret) {dev_err (dev,"Failed to deassert reset signals: %d\n" , ret);return ret;
@@ -1607,7 +1605,7 @@ static int eic7700_init(struct device *dev, struct sdhci_host *host, struct dwcm dwc_priv->priv = priv; - ret =sdhci_eic7700_reset_init (dev, dwc_priv->priv); + ret =sdhci_eic7700_reset_init (dev, dwc_priv);if (ret) {dev_err (dev,"failed to reset\n" );return ret;
@@ -2122,6 +2120,8 @@ static int dwcmshc_suspend(struct device *dev)if (ret)return ret; +reset_control_assert (priv->reset); +clk_disable_unprepare (pltfm_host->clk);if (!IS_ERR (priv->bus_clk))clk_disable_unprepare (priv->bus_clk);
@@ -2152,18 +2152,25 @@ static int dwcmshc_resume(struct device *dev)if (ret) goto disable_bus_clk; - ret =sdhci_resume_host (host); + ret =reset_control_deassert (priv->reset);if (ret) goto disable_other_clks; + + ret =sdhci_resume_host (host); +if (ret) + goto assert_reset; +if (host->mmc->caps2 &MMC_CAP2_CQE ) { ret =cqhci_resume (host->mmc);if (ret) - goto disable_other_clks; + goto assert_reset; }return 0 ; +assert_reset: +reset_control_assert (priv->reset); disable_other_clks:clk_bulk_disable_unprepare (priv->num_other_clks, priv->other_clks); disable_bus_clk:
Signature · 1 line
2.43.0
Quoted 10 lines · Artem Shimko
The driver currently handles reset control only during initializations, but does not manage the controller reset signal during system suspend/resume. This can leave the hardware in an undefined state after resume. Additionally, the reset control is stored in vendor structs making it inaccessible for power management operations.
Move the reset control to
dwcmshc_privstructure and add proper assertion during suspend and deassertion during resume. This ensures the controller is properly reset before entering low power state and correctly reinitialized upon wake.
...
Quoted 3 lines
---
Hello maintainers and reviewers,
No need to repeat the commit message in the comments. It does not add any value.
Quoted 11 lines
This patch adds reset control support for the MMC controller during system suspend/resume.
Currently, the reset control is only handled during initialization and is stored in vendor-specific private structures. This makes it inaccessible to the common power management code and leaves the controller in an undefined state after resume.
The changes move the reset control to the generic
dwcmshc_privstructure, update all platform-specific users (Rockchip, EIC7700), and add proper reset assertion/deassertion in dwcmshc_suspend/resume.
This is serious behaviour change (might not be, one needs to understand what it does in comparison to no reset (and in accordance with datasheet).
Do you have any HW to test?
...
drivers/mmc/host/sdhci-of-dwcmshc.c static int dwcmshc_suspend(struct device *dev) · line 2127
Quoted 3 lines
drivers/mmc/host/sdhci-of-dwcmshc.c +0
-0clk_disable_unprepare (pltfm_host->clk);if (!IS_ERR (priv->bus_clk))clk_disable_unprepare (priv->bus_clk);
The driver seems can be refactored a lot (with no functional changes) by:
- replacing *sleep() with fsleep() API
- dropping unneeded checks like above, clk_disable_unprepare() is error pointer-aware IIRC (or it can be made so it is either NULL or valid one)
- using 'return dev_err_probe()'
This is just a side note in case you are interested.
Quoted 2 lines
drivers/mmc/host/sdhci-of-dwcmshc.c +0
-0if (ret) goto disable_bus_clk;
...
drivers/mmc/host/sdhci-of-dwcmshc.c static int dwcmshc_resume(struct device *dev) · line 2159
Quoted 6 lines
drivers/mmc/host/sdhci-of-dwcmshc.c +2
-1- ret =sdhci_resume_host (host); + ret =reset_control_deassert (priv->reset);if (ret) goto disable_other_clks; +
No need to add an extra blank line.
Quoted 3 lines
drivers/mmc/host/sdhci-of-dwcmshc.c +3
-0+ ret =sdhci_resume_host (host); +if (ret) + goto assert_reset;
Signature · 2 lines
With Best Regards, Andy Shevchenko
Hi Andy,
Thank you for your review!
Quoted 1 line · Andy Shevchenko
No need to repeat the commit message in the comments. It does not add any value.
Got it =)
Quoted 4 lines
This is serious behaviour change (might not be, one needs to understand what it does in comparison to no reset (and in accordance with datasheet).
Do you have any HW to test?
I made these changes so that our MMC controller (dwc) can handle the reset signal not only during the initialization function, but also during suspend and resume.
Since the reset control was previously stored in vendor-specific private structures (rk35xx_priv and eic7700_priv), it was inaccessible from the common PM code.
To make suspend/resume work for our platform, I had to move the reset pointer to the generic dwcmshc_priv structure, which required updating the Rockchip and EIC7700 variants accordingly.
Quoted 7 lines
The driver seems can be refactored a lot (with no functional changes) by:
- replacing*sleep()withfsleep()API
- dropping unneeded checks like above,clk_disable_unprepare()is error pointer-aware IIRC (or it can be made so it is either NULL or valid one)
- using 'returndev_err_probe()'This is just a side note in case you are interested.
Yes, sure, thank you!
Quoted 1 line
No need to add an extra blank line.
okay =)
Instead of moving the reset control directly into the common path, perhaps a cleaner solution would be to introduce a set of platform PM ops (e.g., suspend/resume callbacks in the variant data). These ops could be checked for NULL in the common dwcmshc_suspend/resume functions. For platforms that don't implement them, the behavior remains exactly as it was before (no reset assertion),
ensuring zero risk of regression for Rockchip and EIC7700. Our platform would then simply implement these ops to handle the custom reset sequence.
What do you think about it?
--
Regards,
Artem
Adrian Hunter
maintainerQuoted 52 lines · Artem Shimko
Hi Andy,
Thank you for your review!
Quoted 1 line · Andy Shevchenko
No need to repeat the commit message in the comments. It does not add any value.
Got it =)
Quoted 4 lines
This is serious behaviour change (might not be, one needs to understand what it does in comparison to no reset (and in accordance with datasheet).
Do you have any HW to test?
I made these changes so that our MMC controller (dwc) can handle the reset signal not only during the initialization function, but also during suspend and resume.
Since the reset control was previously stored in vendor-specific private structures (rk35xx_privandeic7700_priv), it was inaccessible from the common PM code.
To make suspend/resume work for our platform, I had to move the reset pointer to the genericdwcmshc_privstructure, which required updating the Rockchip and EIC7700 variants accordingly.Quoted 7 lines
The driver seems can be refactored a lot (with no functional changes) by:
- replacing*sleep()withfsleep()API
- dropping unneeded checks like above,clk_disable_unprepare()is error pointer-aware IIRC (or it can be made so it is either NULL or valid one)
- using 'returndev_err_probe()'This is just a side note in case you are interested.
Yes, sure, thank you!
Quoted 1 line
No need to add an extra blank line.
okay =)
Instead of moving the reset control directly into the common path, perhaps a cleaner solution would be to introduce a set of platform PM ops (e.g., suspend/resume callbacks in the variant data). These ops could be checked for NULL in the common dwcmshc_suspend/resume functions. For platforms that don't implement them, the behavior remains exactly as it was before (no reset assertion),
ensuring zero risk of regression for Rockchip and EIC7700. Our platform would then simply implement these ops to handle the custom reset sequence.
Quoted 1 line
What do you think about it?
Only dwcmshc_rk35xx_init() and sdhci_eic7700_reset_init() assign dwc_priv->reset, so I am confused about what devices you intend this patch for.
You definitely need to limit the change to devices that you know for certain it will work.
Quoted 3 lines
--
Regards,
Artem
Hi Adrian,
Quoted 3 lines · Adrian Hunter
Only
dwcmshc_rk35xx_init()andsdhci_eic7700_reset_init()assigndwc_priv->reset, so I am confused about what devices you intend this patch for.
Yeah, the change as proposed would affect other devices, and I should not assume it's safe for them without testing. Yes, I was so fast to send it.
My intention was to make the reset control accessible for PM operations, but only for the specific SoC I'm working with. However, since the custom MMC driver code is not part of this upstream submission, the patch currently lacks the necessary condition to limit the new behavior.
Could you please check my plan to prepare v2 to:
1. Keep the reset control moved to dwcmshc_priv (so it's accessible where needed).
2. Add a dedicated flag (e.g., bool needs_reset_on_pm or similar) to dwcmshc_priv so that only devices that explicitly opt in will trigger the reset during suspend/resume.
That ensures the existing behavior for Rockchip RK35xx and EIC7700 remains unchanged.
This way, the infrastructure is in place for devices that need it, without imposing it on platforms where it hasn't been validated. Thank you again for the review!
Regards,
Artem
Adrian Hunter
maintainerQuoted 16 lines · Artem Shimko
Hi Adrian,
Quoted 3 lines · Adrian Hunter
Only
dwcmshc_rk35xx_init()andsdhci_eic7700_reset_init()assigndwc_priv->reset, so I am confused about what devices you intend this patch for.Yeah, the change as proposed would affect other devices, and I should not assume it's safe for them without testing. Yes, I was so fast to send it.
My intention was to make the reset control accessible for PM operations, but only for the specific SoC I'm working with. However, since the custom MMC driver code is not part of this upstream submission, the patch currently lacks the necessary condition to limit the new behavior.
Quoted 6 lines
Could you please check my plan to prepare v2 to:
1. Keep the reset control moved todwcmshc_priv(so it's accessible where needed).
2. Add a dedicated flag (e.g., boolneeds_reset_on_pmor similar) todwcmshc_privso that only devices that explicitly opt in will trigger the reset during suspend/resume.
Sounds fine
Quoted 9 lines
That ensures the existing behavior for Rockchip RK35xx and EIC7700 remains unchanged.
This way, the infrastructure is in place for devices that need it, without imposing it on platforms where it hasn't been validated. Thank you again for the review!Regards,
Artem
Hi Adrian,
Quoted 1 line · Adrian Hunter
Sounds fine
Great! Let me send v2.
Thank you.
Regards,
Artem