linuxsh-dev

Why flush cache in insw/outsw ?

[<tak...@op...>]

Currently, shmin can't compiled because of flush_dcache_all() in io_generic.c.
SH3 doesn't have flush_"d"cache_all().
Could you please tell me why the flush operations has come here?
Adding dummy flush_dcache_all() in cache_flush.h is easy, though.
/yoshii

cg-diff -pr 9c57548f17806ffd8e4dc4f7973ce78bbfbc2079 arch/sh/kernel/io_generic.c
...
> @@ -96,6 +96,7 @@ void generic_insw(unsigned long port, vo
>         while (count--)
>                 *buf++ = *port_addr;
>  
> +       flush_dcache_all();
>         dummy_read();
>  }
>  
> @@ -170,6 +171,7 @@ void generic_outsw(unsigned long port, c
>         while (count--)
>                 *port_addr = *buf++;
>  
> +       flush_dcache_all();
>         dummy_read();
>  }

Re: Why flush cache in insw/outsw ?

[Paul M. <le...@li...>]

On Mon, Feb 19, 2007 at 11:38:15PM +0900, tak...@op... wrote:
> Currently, shmin can't compiled because of flush_dcache_all() in io_generic.c.
> SH3 doesn't have flush_"d"cache_all().

Ah, that's right, I forgot to stub it for SH-3.

> Could you please tell me why the flush operations has come here?
> Adding dummy flush_dcache_all() in cache_flush.h is easy, though.

We needed these for some of the IDE cases, it was something that was hit
in the PG_dcache_dirty conversion. We used to do this in the CONFIG_IDE
case by overloading the PIO I/O ops with an explicit flush, but that sort
of infrastructure no longer exists with libata. As the IDE code is really
the only user of the insw/outsw routines, this was the easy place to add
it.

You may want to add a sensible definition for SH-3 as well, but I'm not
sure how heavily you're hitting the libata drivers.

Re: Why flush cache in insw/outsw ?

[<tak...@op...>]

Thank you, Paul.

> We needed these for some of the IDE cases, it was something that was hit
> in the PG_dcache_dirty conversion. <snip>
Oh, that's exactly what I guess. Happy.

> You may want to add a sensible definition for SH-3 as well, but I'm not
> sure how heavily you're hitting the libata drivers.
mmm, Because I don't have SH3 board with libata enabled, I'd like to leave it blank now.
How about following patch? I put them into io_generic.c near insw/outsw, but cache_flush.h,
 since these seems a kind of workaround.
# SH3/4 compatible cache flush API may be wanted in somewhere else, someday, though.

Regards,
/yoshii
# sorry for trivial patches always...

Signed-off-by: Takashi YOSHII <tak...@op...>

diff --git a/arch/sh/kernel/io_generic.c b/arch/sh/kernel/io_generic.c
index 66626c0..7893718 100644
--- a/arch/sh/kernel/io_generic.c
+++ b/arch/sh/kernel/io_generic.c
@@ -25,6 +25,13 @@ #else
 #define dummy_read()
 #endif
 
+#if defined(CONFIG_CPU_SH4)
+/* We need these for some of the IDE cases. */
+#define flush_cache()	flush_dcache_all()
+#else
+#define flush_cache()	/* XXX: flushing might be needed */
+#endif
+
 unsigned long generic_io_base;
 
 static inline void delay(void)
@@ -96,7 +103,7 @@ void generic_insw(unsigned long port, vo
 	while (count--)
 		*buf++ = *port_addr;
 
-	flush_dcache_all();
+	flush_cache();
 	dummy_read();
 }
 
@@ -171,7 +178,7 @@ void generic_outsw(unsigned long port, c
 	while (count--)
 		*port_addr = *buf++;
 
-	flush_dcache_all();
+	flush_cache();
 	dummy_read();
 }
 

Re: Why flush cache in insw/outsw ?

[Paul M. <le...@li...>]

On Tue, Feb 20, 2007 at 07:35:46AM +0900, tak...@op... wrote:
> > You may want to add a sensible definition for SH-3 as well, but I'm not
> > sure how heavily you're hitting the libata drivers.
> mmm, Because I don't have SH3 board with libata enabled, I'd like to leave it blank now.

You aren't booting off of CF or PCMCIA with pata_platform?

> How about following patch? I put them into io_generic.c near
> insw/outsw, but cache_flush.h, since these seems a kind of workaround.
> # SH3/4 compatible cache flush API may be wanted in somewhere else,
> someday, though.
> 
Yes, we can do that. It would be interesting to know if the SH-3 caches
will need this too, at the very least, SH7705 will.

Iwamatsu-san, are you able to add shmin in to your build scripts? It
would be good to get at least one SH-3 (and an SH-2) in to the nightly
builds.

Re: Why flush cache in insw/outsw ?

[Nobuhiro I. <he...@t-...>]

Hi , Paul .

On Tue, 20 Feb 2007 09:50:58 +0900
Paul Mundt <le...@li...> wrote:

> 
> Iwamatsu-san, are you able to add shmin in to your build scripts? It
> would be good to get at least one SH-3 (and an SH-2) in to the nightly
> builds.

OK , I will add shmin(SH3-machine) to  nightly build .
I think that after the matter of SH2 prepares the compiler of SH2.

regards,
 Nobuhiro 
-- 
Nobuhiro Iwamatsu
	E-Mail : he...@t-...
	GPG ID : 3170EBE9