[PATCH 3.17-rc3] serial: asc: Adopt readl_/writel_relaxed()

[PATCH 3.17-rc3] serial: asc: Adopt readl_/writel_relaxed()

[Daniel Thompson]

The architectures where this peripheral exists (ARM and SH) have expensive
implementations of writel(), reliant on spin locks and explicit L2 cache
management. These architectures provide a cheaper writel_relaxed() which
is much better suited to peripherals that do not perform DMA. The
situation with readl()/readl_relaxed()is similar although less acute.

This driver does not use DMA and will be more power efficient and more
robust (due to absense of spin locks during console I/O) if it uses the
relaxed variants.

This change means the driver is no longer portable and therefore no
longer suitable for compile testing.

Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Cc: Srinivas Kandagatla <srinivas.kandagatla@gmail.com>
Cc: Patrice Chotard <patrice.chotard@st.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: kernel@stlinux.com
Cc: linux-serial@vger.kernel.org
Acked-by: Maxime Coquelin <maxime.coquelin@st.com>
Acked-by: Peter Griffin <peter.griffin@linaro.org>
---
 drivers/tty/serial/Kconfig  | 2 +-
 drivers/tty/serial/st-asc.c | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/tty/serial/Kconfig b/drivers/tty/serial/Kconfig
index 26cec64d..e9b1735 100644
--- a/drivers/tty/serial/Kconfig
+++ b/drivers/tty/serial/Kconfig
@@ -1527,7 +1527,7 @@ config SERIAL_FSL_LPUART_CONSOLE
 config SERIAL_ST_ASC
 	tristate "ST ASC serial port support"
 	select SERIAL_CORE
-	depends on ARM || COMPILE_TEST
+	depends on ARM
 	help
 	  This driver is for the on-chip Asychronous Serial Controller on
 	  STMicroelectronics STi SoCs.
diff --git a/drivers/tty/serial/st-asc.c b/drivers/tty/serial/st-asc.c
index 8b2d735..adadbc1 100644
--- a/drivers/tty/serial/st-asc.c
+++ b/drivers/tty/serial/st-asc.c
@@ -151,12 +151,12 @@ static inline struct asc_port *to_asc_port(struct uart_port *port)

 static inline u32 asc_in(struct uart_port *port, u32 offset)
 {
-	return readl(port->membase + offset);
+	return readl_relaxed(port->membase + offset);
 }

 static inline void asc_out(struct uart_port *port, u32 offset, u32 value)
 {
-	writel(value, port->membase + offset);
+	writel_relaxed(value, port->membase + offset);
 }

 /*
--
1.9.3

Re: [PATCH 3.17-rc3] serial: asc: Adopt readl_/writel_relaxed()

[Greg Kroah-Hartman]

On Wed, Sep 03, 2014 at 01:00:51PM +0100, Daniel Thompson wrote:
> The architectures where this peripheral exists (ARM and SH) have expensive
> implementations of writel(), reliant on spin locks and explicit L2 cache
> management. These architectures provide a cheaper writel_relaxed() which
> is much better suited to peripherals that do not perform DMA. The
> situation with readl()/readl_relaxed()is similar although less acute.
> 
> This driver does not use DMA and will be more power efficient and more
> robust (due to absense of spin locks during console I/O) if it uses the
> relaxed variants.
> 
> This change means the driver is no longer portable and therefore no
> longer suitable for compile testing.
> 
> Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
> Cc: Srinivas Kandagatla <srinivas.kandagatla@gmail.com>
> Cc: Patrice Chotard <patrice.chotard@st.com>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: Jiri Slaby <jslaby@suse.cz>
> Cc: kernel@stlinux.com
> Cc: linux-serial@vger.kernel.org
> Acked-by: Maxime Coquelin <maxime.coquelin@st.com>
> Acked-by: Peter Griffin <peter.griffin@linaro.org>
> ---
>  drivers/tty/serial/Kconfig  | 2 +-
>  drivers/tty/serial/st-asc.c | 4 ++--
>  2 files changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/tty/serial/Kconfig b/drivers/tty/serial/Kconfig
> index 26cec64d..e9b1735 100644
> --- a/drivers/tty/serial/Kconfig
> +++ b/drivers/tty/serial/Kconfig
> @@ -1527,7 +1527,7 @@ config SERIAL_FSL_LPUART_CONSOLE
>  config SERIAL_ST_ASC
>  	tristate "ST ASC serial port support"
>  	select SERIAL_CORE
> -	depends on ARM || COMPILE_TEST
> +	depends on ARM

I really don't like stuff that does this, sorry.  I want to test build
as many drivers as I can.  COMPILE_TEST does not mean that the driver is
"portable", only that it builds properly on all platforms.

>  	help
>  	  This driver is for the on-chip Asychronous Serial Controller on
>  	  STMicroelectronics STi SoCs.
> diff --git a/drivers/tty/serial/st-asc.c b/drivers/tty/serial/st-asc.c
> index 8b2d735..adadbc1 100644
> --- a/drivers/tty/serial/st-asc.c
> +++ b/drivers/tty/serial/st-asc.c
> @@ -151,12 +151,12 @@ static inline struct asc_port *to_asc_port(struct uart_port *port)
> 
>  static inline u32 asc_in(struct uart_port *port, u32 offset)
>  {
> -	return readl(port->membase + offset);
> +	return readl_relaxed(port->membase + offset);

What plaforms do not provide readl_relaxed()?

thanks,

greg k-h

Re: [PATCH 3.17-rc3] serial: asc: Adopt readl_/writel_relaxed()

[Daniel Thompson]

On 09/09/14 00:27, Greg Kroah-Hartman wrote:
> On Wed, Sep 03, 2014 at 01:00:51PM +0100, Daniel Thompson wrote:
>> The architectures where this peripheral exists (ARM and SH) have expensive
>> implementations of writel(), reliant on spin locks and explicit L2 cache
>> management. These architectures provide a cheaper writel_relaxed() which
>> is much better suited to peripherals that do not perform DMA. The
>> situation with readl()/readl_relaxed()is similar although less acute.
>>
>> This driver does not use DMA and will be more power efficient and more
>> robust (due to absense of spin locks during console I/O) if it uses the
>> relaxed variants.
>>
>> This change means the driver is no longer portable and therefore no
>> longer suitable for compile testing.
>>
>> Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
>> Cc: Srinivas Kandagatla <srinivas.kandagatla@gmail.com>
>> Cc: Patrice Chotard <patrice.chotard@st.com>
>> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
>> Cc: Jiri Slaby <jslaby@suse.cz>
>> Cc: kernel@stlinux.com
>> Cc: linux-serial@vger.kernel.org
>> Acked-by: Maxime Coquelin <maxime.coquelin@st.com>
>> Acked-by: Peter Griffin <peter.griffin@linaro.org>
>> ---
>>  drivers/tty/serial/Kconfig  | 2 +-
>>  drivers/tty/serial/st-asc.c | 4 ++--
>>  2 files changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/tty/serial/Kconfig b/drivers/tty/serial/Kconfig
>> index 26cec64d..e9b1735 100644
>> --- a/drivers/tty/serial/Kconfig
>> +++ b/drivers/tty/serial/Kconfig
>> @@ -1527,7 +1527,7 @@ config SERIAL_FSL_LPUART_CONSOLE
>>  config SERIAL_ST_ASC
>>  	tristate "ST ASC serial port support"
>>  	select SERIAL_CORE
>> -	depends on ARM || COMPILE_TEST
>> +	depends on ARM
> 
> I really don't like stuff that does this, sorry.  I want to test build
> as many drivers as I can.  COMPILE_TEST does not mean that the driver is
> "portable", only that it builds properly on all platforms.

I originally made this change compilable (and portable) but was asked to
change it during review:
http://thread.gmane.org/gmane.linux.ports.arm.kernel/331027/focus=333911

It sounds like I gave in too quickly. I'll re-post the original version
shortly.


>>  	help
>>  	  This driver is for the on-chip Asychronous Serial Controller on
>>  	  STMicroelectronics STi SoCs.
>> diff --git a/drivers/tty/serial/st-asc.c b/drivers/tty/serial/st-asc.c
>> index 8b2d735..adadbc1 100644
>> --- a/drivers/tty/serial/st-asc.c
>> +++ b/drivers/tty/serial/st-asc.c
>> @@ -151,12 +151,12 @@ static inline struct asc_port *to_asc_port(struct uart_port *port)
>>
>>  static inline u32 asc_in(struct uart_port *port, u32 offset)
>>  {
>> -	return readl(port->membase + offset);
>> +	return readl_relaxed(port->membase + offset);
> 
> What plaforms do not provide readl_relaxed()?

I'd never thought to ask that. However I think the answer is "only those
that use asm-generic/io.h" meaning: blackfin, m68k, metag, openrisc,
score and sparc. I'll look into a patch to fix that...

The reason I never thought much about readl_relaxed() is that the
compilability concerns centre around writel_relaxed() instead. This is
much less widely implemented. It appears mostly on architectures where
writel() is both expensive and (sometimes) overkill. It is currently
found only on: alpha, arm, arm64, avr32, hexagon, microblaze, mips and sh.

My original code to conceal the difference between the two looked like
this (and the change is to a single accessor function, not littered
though the code):
--- cut here ---
 static inline void asc_out(struct uart_port *port, u32 offset,
 			    u32 value)
{
+#ifdef writel_relaxed
+	writel_relaxed(value, port->membase + offset);
+	barrier();
+#else
 	writel(value, port->membase + offset);
+#endif
 }
--- cut here ---

Note that barrier() is not needed if our only goal is for the driver to
pass the COMPILE_TEST but was included because different architectures
have different rules about inclusion of barrier() within the _relaxed()
macros making explicit barriers useful if this code were consumed by a
copy 'n paste operation...

[PATCH 3.17-rc3 v2] serial: asc: Adopt readl_/writel_relaxed()

[Daniel Thompson]

The architectures supported by this driver, arm and sh, have expensive
implementations of writel(), reliant on spin locks and explicit L2 cache
management. These architectures provide a cheaper writel_relaxed() which
is much better suited to peripherals that do not perform DMA. The
situation with readl()/readl_relaxed()is similar although less acute.

This driver does not use DMA and will be more power efficient and more
robust (due to absence of spin locks during console I/O) if it uses the
relaxed variants.

The driver supports COMPILE_TEST and therefore falls back to writel()
when writel_relaxed() does not exist.

Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Cc: Srinivas Kandagatla <srinivas.kandagatla@gmail.com>
Cc: Maxime Coquelin <maxime.coquelin@st.com>
Cc: Patrice Chotard <patrice.chotard@st.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: kernel@stlinux.com
Cc: linux-serial@vger.kernel.org
---

Notes:
    Changes since v1:
    
    * Added fallback to writel() to permit COMPILE_TESTing (review
      of Greg Kroah-Hartmam).

 drivers/tty/serial/st-asc.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/tty/serial/st-asc.c b/drivers/tty/serial/st-asc.c
index 8b2d735..a3fc167 100644
--- a/drivers/tty/serial/st-asc.c
+++ b/drivers/tty/serial/st-asc.c
@@ -151,12 +151,16 @@ static inline struct asc_port *to_asc_port(struct uart_port *port)

 static inline u32 asc_in(struct uart_port *port, u32 offset)
 {
-	return readl(port->membase + offset);
+	return readl_relaxed(port->membase + offset);
 }

 static inline void asc_out(struct uart_port *port, u32 offset, u32 value)
 {
+#ifdef writel_relaxed
+	writel_relaxed(value, port->membase + offset);
+#else
 	writel(value, port->membase + offset);
+#endif
 }

 /*
--
1.9.3

Re: [PATCH 3.17-rc3 v2] serial: asc: Adopt readl_/writel_relaxed()

[Daniel Thompson]

On 09/09/14 12:08, Srinivas Kandagatla wrote:
> Hi Dan,
> 
> After Greg's comment, Am ok with the patch and sorry for the noise.
> Good that you removed the barier() too.

Yes.

I got interested in Greg's question about which architectures don't have
readl_relaxed() and, in the process, discovered I was wrong about that
being needed anyway.


> You can have my Ack.
> 
> --srini
> 
> On Tue, Sep 9, 2014 at 11:03 AM, Daniel Thompson
> <daniel.thompson@linaro.org <mailto:daniel.thompson@linaro.org>> wrote:
> 
>     The architectures supported by this driver, arm and sh, have expensive
>     implementations of writel(), reliant on spin locks and explicit L2 cache
>     management. These architectures provide a cheaper writel_relaxed() which
>     is much better suited to peripherals that do not perform DMA. The
>     situation with readl()/readl_relaxed()is similar although less acute.
> 
>     This driver does not use DMA and will be more power efficient and more
>     robust (due to absence of spin locks during console I/O) if it uses the
>     relaxed variants.
> 
>     The driver supports COMPILE_TEST and therefore falls back to writel()
>     when writel_relaxed() does not exist.
> 
>     Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org
>     <mailto:daniel.thompson@linaro.org>>
>     Cc: Srinivas Kandagatla <srinivas.kandagatla@gmail.com
>     <mailto:srinivas.kandagatla@gmail.com>>
>     Cc: Maxime Coquelin <maxime.coquelin@st.com
>     <mailto:maxime.coquelin@st.com>>
>     Cc: Patrice Chotard <patrice.chotard@st.com
>     <mailto:patrice.chotard@st.com>>
>     Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org
>     <mailto:gregkh@linuxfoundation.org>>
>     Cc: Jiri Slaby <jslaby@suse.cz <mailto:jslaby@suse.cz>>
>     Cc: kernel@stlinux.com <mailto:kernel@stlinux.com>
>     Cc: linux-serial@vger.kernel.org <mailto:linux-serial@vger.kernel.org>
>     ---
> 
>     Notes:
>         Changes since v1:
> 
>         * Added fallback to writel() to permit COMPILE_TESTing (review
>           of Greg Kroah-Hartmam).
> 
>      drivers/tty/serial/st-asc.c | 6 +++++-
>      1 file changed, 5 insertions(+), 1 deletion(-)
> 
>     diff --git a/drivers/tty/serial/st-asc.c b/drivers/tty/serial/st-asc.c
>     index 8b2d735..a3fc167 100644
>     --- a/drivers/tty/serial/st-asc.c
>     +++ b/drivers/tty/serial/st-asc.c
>     @@ -151,12 +151,16 @@ static inline struct asc_port
>     *to_asc_port(struct uart_port *port)
> 
>      static inline u32 asc_in(struct uart_port *port, u32 offset)
>      {
>     -       return readl(port->membase + offset);
>     +       return readl_relaxed(port->membase + offset);
>      }
> 
>      static inline void asc_out(struct uart_port *port, u32 offset, u32
>     value)
>      {
>     +#ifdef writel_relaxed
>     +       writel_relaxed(value, port->membase + offset);
>     +#else
>             writel(value, port->membase + offset);
>     +#endif
>      }
> 
>      /*
>     --
>     1.9.3
> 
>