[PATCH 0/3] mmc: avoid vicious circle when retuning

[PATCH 0/3] mmc: avoid vicious circle when retuning

[Wolfram Sang]

See patch 1 for a description of the problem. This series implements the
alternative approach suggested by Adrian (thanks!). It also adds some
documentation and a minor cleanup which I came up with while working on
the fix. Patch 1 can go to stable as is, the rest built on top of that.

This series fixes the performance issue which we saw when injecting CRC
errors on Renesas R-Car Gen3 hardware.

Looking forward to comments!


Wolfram Sang (3):
  mmc: core: clear flags before allowing to retune
  mmc: host: add kdoc for mmc_retune_{en|dis}able
  mmc: host: factor out clearing the retune state

 drivers/mmc/core/core.c |  6 ++++--
 drivers/mmc/core/host.c | 13 +++++++++++--
 drivers/mmc/core/host.h |  6 ++++++
 3 files changed, 21 insertions(+), 4 deletions(-)

-- 
2.30.2

[PATCH 1/3] mmc: core: clear flags before allowing to retune

[Wolfram Sang]

It might be that something goes wrong during tuning so the MMC core will
immediately trigger a retune. In our case it was:

 - we sent a tuning block
 - there was an error so we need to send an abort cmd to the eMMC
 - the abort cmd had a CRC error
 - retune was set by the MMC core

This lead to a vicious circle causing a performance regression of 75%.
So, clear retuning flags before we enable retuning to start with a known
cleared state.

Reported-by Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Suggested-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
---
 drivers/mmc/core/core.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
index 4e52eb14198a..f397cf051b8d 100644
--- a/drivers/mmc/core/core.c
+++ b/drivers/mmc/core/core.c
@@ -937,11 +937,14 @@ int mmc_execute_tuning(struct mmc_card *card)
 
 	err = host->ops->execute_tuning(host, opcode);
 
-	if (err)
+	if (err) {
 		pr_err("%s: tuning execution failed: %d\n",
 			mmc_hostname(host), err);
-	else
+	} else {
+		host->retune_now = 0;
+		host->need_retune = 0;
 		mmc_retune_enable(host);
+	}
 
 	return err;
 }
-- 
2.30.2

[PATCH 2/3] mmc: host: add kdoc for mmc_retune_{en|dis}able

[Wolfram Sang]

I wanted to use it in a wrong way, so document the intended way.

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
---
 drivers/mmc/core/host.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
index eda4a1892c33..0f084c9b2684 100644
--- a/drivers/mmc/core/host.c
+++ b/drivers/mmc/core/host.c
@@ -95,6 +95,10 @@ void mmc_unregister_host_class(void)
 	class_unregister(&mmc_host_class);
 }
 
+/**
+ * mmc_retune_enable() - enter a transfer mode that requires retuning
+ * @host: host which should retune now
+ */
 void mmc_retune_enable(struct mmc_host *host)
 {
 	host->can_retune = 1;
@@ -126,6 +130,12 @@ void mmc_retune_unpause(struct mmc_host *host)
 }
 EXPORT_SYMBOL(mmc_retune_unpause);
 
+/**
+ * mmc_retune_disable() - exit a transfer mode that requires retuning
+ * @host: host which should not retune anymore
+ *
+ * It is not meant for temporarily preventing retuning!
+ */
 void mmc_retune_disable(struct mmc_host *host)
 {
 	mmc_retune_unpause(host);
-- 
2.30.2

[PATCH 3/3] mmc: host: factor out clearing the retune state

[Wolfram Sang]

We have this in two places, so let's have a dedicated function. It is
also more readable.

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
---
 drivers/mmc/core/core.c | 3 +--
 drivers/mmc/core/host.c | 3 +--
 drivers/mmc/core/host.h | 6 ++++++
 3 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
index f397cf051b8d..469bc58c2113 100644
--- a/drivers/mmc/core/core.c
+++ b/drivers/mmc/core/core.c
@@ -941,8 +941,7 @@ int mmc_execute_tuning(struct mmc_card *card)
 		pr_err("%s: tuning execution failed: %d\n",
 			mmc_hostname(host), err);
 	} else {
-		host->retune_now = 0;
-		host->need_retune = 0;
+		mmc_retune_clear(host);
 		mmc_retune_enable(host);
 	}
 
diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
index 0f084c9b2684..52d37587cf45 100644
--- a/drivers/mmc/core/host.c
+++ b/drivers/mmc/core/host.c
@@ -141,8 +141,7 @@ void mmc_retune_disable(struct mmc_host *host)
 	mmc_retune_unpause(host);
 	host->can_retune = 0;
 	del_timer_sync(&host->retune_timer);
-	host->retune_now = 0;
-	host->need_retune = 0;
+	mmc_retune_clear(host);
 }
 
 void mmc_retune_timer_stop(struct mmc_host *host)
diff --git a/drivers/mmc/core/host.h b/drivers/mmc/core/host.h
index ba407617ed23..48c4952512a5 100644
--- a/drivers/mmc/core/host.h
+++ b/drivers/mmc/core/host.h
@@ -21,6 +21,12 @@ int mmc_retune(struct mmc_host *host);
 void mmc_retune_pause(struct mmc_host *host);
 void mmc_retune_unpause(struct mmc_host *host);
 
+static inline void mmc_retune_clear(struct mmc_host *host)
+{
+	host->retune_now = 0;
+	host->need_retune = 0;
+}
+
 static inline void mmc_retune_hold_now(struct mmc_host *host)
 {
 	host->retune_now = 0;
-- 
2.30.2

Re: [PATCH 0/3] mmc: avoid vicious circle when retuning

[Adrian Hunter]

On 24/06/21 6:16 pm, Wolfram Sang wrote:
> See patch 1 for a description of the problem. This series implements the
> alternative approach suggested by Adrian (thanks!). It also adds some
> documentation and a minor cleanup which I came up with while working on
> the fix. Patch 1 can go to stable as is, the rest built on top of that.
> 
> This series fixes the performance issue which we saw when injecting CRC
> errors on Renesas R-Car Gen3 hardware.
> 
> Looking forward to comments!

Looks good to me.  For all 3 patches:

Acked-by: Adrian Hunter <adrian.hunter@intel.com>

> 
> 
> Wolfram Sang (3):
>   mmc: core: clear flags before allowing to retune
>   mmc: host: add kdoc for mmc_retune_{en|dis}able
>   mmc: host: factor out clearing the retune state
> 
>  drivers/mmc/core/core.c |  6 ++++--
>  drivers/mmc/core/host.c | 13 +++++++++++--
>  drivers/mmc/core/host.h |  6 ++++++
>  3 files changed, 21 insertions(+), 4 deletions(-)
> 

RE: [PATCH 0/3] mmc: avoid vicious circle when retuning

[Yoshihiro Shimoda]

Hi Wolfram-san,

Thank you for the patches!

> From: Wolfram Sang, Sent: Friday, June 25, 2021 12:16 AM
> 
> See patch 1 for a description of the problem. This series implements the
> alternative approach suggested by Adrian (thanks!). It also adds some
> documentation and a minor cleanup which I came up with while working on
> the fix. Patch 1 can go to stable as is, the rest built on top of that.

Perhaps, we can add a Fixes tag into the patch 1 like below?

Fixes: bd11e8bd03ca ("mmc: core: Flag re-tuning is needed on CRC errors")

> This series fixes the performance issue which we saw when injecting CRC
> errors on Renesas R-Car Gen3 hardware.
> 
> Looking forward to comments!

The patches look good to me. So,

Reviewed-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

And, I tested on my environment (r8a77951-salvator-xs) with local error
injection, and the performance issue disappeared. So,

Tested-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>

Best regards,
Yoshihiro Shimoda

Re: [PATCH 0/3] mmc: avoid vicious circle when retuning

[Ulf Hansson]

On Thu, 24 Jun 2021 at 17:16, Wolfram Sang
<wsa+renesas@sang-engineering.com> wrote:
>
> See patch 1 for a description of the problem. This series implements the
> alternative approach suggested by Adrian (thanks!). It also adds some
> documentation and a minor cleanup which I came up with while working on
> the fix. Patch 1 can go to stable as is, the rest built on top of that.
>
> This series fixes the performance issue which we saw when injecting CRC
> errors on Renesas R-Car Gen3 hardware.
>
> Looking forward to comments!
>
>
> Wolfram Sang (3):
>   mmc: core: clear flags before allowing to retune
>   mmc: host: add kdoc for mmc_retune_{en|dis}able
>   mmc: host: factor out clearing the retune state
>
>  drivers/mmc/core/core.c |  6 ++++--
>  drivers/mmc/core/host.c | 13 +++++++++++--
>  drivers/mmc/core/host.h |  6 ++++++
>  3 files changed, 21 insertions(+), 4 deletions(-)
>
> --
> 2.30.2
>

Patch1 applied for fixes and by adding a fixes+stable tag. Patch2 and
patch3, queued up for v5.15 (temporary on the devel branch).

Thanks and kind regards
Uffe