[PATCH] linux-next remove wmb() from ide-dma-sff.c and scc_pata.c

[PATCH] linux-next remove wmb() from ide-dma-sff.c and scc_pata.c

[Grant Grundler]

[-- Attachment #1: Type: text/plain, Size: 1178 bytes --]

Followup to "[PATCH 03/10] ide: destroy DMA mappings after ending DMA"
email on March 14th:
    http://lkml.org/lkml/2009/3/14/17

No maintainer is listed for "Toshiba CELL Reference Set IDE" (BLK_DEV_CELLEB)
or tx4939ide.c in MAINTAINERS. I've CC'd "Ishizaki Kou" @Toshiba (Maintainer for
"Spidernet Network Driver for CELL") and linuxppc-dev list in the hope
someone else
would know or would be able to ACK this patch.

This patch:
o replaces "mask" variable in ide_dma_end() with #define.
o removes use of wmb() in ide-dma-sff.c and scc_pata.c.
o is not tested - I don't have (or want) the HW.

I did NOT remove wmb() use in tx4939ide.c. tx4939ide.c __raw_writeb()
for MMIO transactions. __raw_writeb() does NOT guarantee memory
transaction ordering.

tx4939ide also uses mmiowb(). AFAIK, mmiowb() only has an effect on
SGI IA64 NUMA machines. I'm not going to guess how this driver might work.

Gmail is broken for sending patches (word wrap). My apologies.
I've attached the patch: diff-next-remove_wmb_from_ide-01.txt

Patch is against linux-next tree:
     git://git.kernel.org/pub/scm/linux/kernel/git/sfr/linux-next

Signed-off-by: Grant Grundler <grundler@google.com>

[-- Attachment #2: diff-next-remove_wmb_from_ide-01.txt --]
[-- Type: text/plain, Size: 1702 bytes --]

diff --git a/drivers/ide/ide-dma-sff.c b/drivers/ide/ide-dma-sff.c
index 16fc46e..e4cdf78 100644
--- a/drivers/ide/ide-dma-sff.c
+++ b/drivers/ide/ide-dma-sff.c
@@ -277,8 +277,6 @@ void ide_dma_start(ide_drive_t *drive)
 		dma_cmd = inb(hwif->dma_base + ATA_DMA_CMD);
 		outb(dma_cmd | ATA_DMA_START, hwif->dma_base + ATA_DMA_CMD);
 	}
-
-	wmb();
 }
 EXPORT_SYMBOL_GPL(ide_dma_start);
 
@@ -286,7 +284,7 @@ EXPORT_SYMBOL_GPL(ide_dma_start);
 int ide_dma_end(ide_drive_t *drive)
 {
 	ide_hwif_t *hwif = drive->hwif;
-	u8 dma_stat = 0, dma_cmd = 0, mask;
+	u8 dma_stat = 0, dma_cmd = 0;
 
 	/* stop DMA */
 	if (hwif->host_flags & IDE_HFLAG_MMIO) {
@@ -304,11 +302,10 @@ int ide_dma_end(ide_drive_t *drive)
 	/* clear INTR & ERROR bits */
 	ide_dma_sff_write_status(hwif, dma_stat | ATA_DMA_ERR | ATA_DMA_INTR);
 
-	wmb();
+#define CHECK_DMA_MASK (ATA_DMA_ACTIVE | ATA_DMA_ERR | ATA_DMA_INTR)
 
 	/* verify good DMA status */
-	mask = ATA_DMA_ACTIVE | ATA_DMA_ERR | ATA_DMA_INTR;
-	if ((dma_stat & mask) != ATA_DMA_INTR)
+	if ((dma_stat & CHECK_DMA_MASK) != ATA_DMA_INTR)
 		return 0x10 | dma_stat;
 	return 0;
 }
diff --git a/drivers/ide/scc_pata.c b/drivers/ide/scc_pata.c
index 97f8e0e..dcbb299 100644
--- a/drivers/ide/scc_pata.c
+++ b/drivers/ide/scc_pata.c
@@ -337,7 +337,6 @@ static void scc_dma_start(ide_drive_t *drive)
 
 	/* start DMA */
 	scc_ide_outb(dma_cmd | 1, hwif->dma_base);
-	wmb();
 }
 
 static int __scc_dma_end(ide_drive_t *drive)
@@ -354,7 +353,6 @@ static int __scc_dma_end(ide_drive_t *drive)
 	/* clear the INTR & ERROR bits */
 	scc_ide_outb(dma_stat | 6, hwif->dma_base + 4);
 	/* verify good DMA status */
-	wmb();
 	return (dma_stat & 7) != 4 ? (0x10 | dma_stat) : 0;
 }
 

Re: [PATCH] linux-next remove wmb() from ide-dma-sff.c and scc_pata.c

[Geert Uytterhoeven]

On Mon, 30 Mar 2009, Grant Grundler wrote:
> Followup to "[PATCH 03/10] ide: destroy DMA mappings after ending DMA"
> email on March 14th:
>     http://lkml.org/lkml/2009/3/14/17
> 
> No maintainer is listed for "Toshiba CELL Reference Set IDE" (BLK_DEV_CELLEB)
> or tx4939ide.c in MAINTAINERS. I've CC'd "Ishizaki Kou" @Toshiba (Maintainer for
> "Spidernet Network Driver for CELL") and linuxppc-dev list in the hope
> someone else
> would know or would be able to ACK this patch.

tx49xx is MIPS, for Nemoto-san.

> This patch:
> o replaces "mask" variable in ide_dma_end() with #define.
> o removes use of wmb() in ide-dma-sff.c and scc_pata.c.
> o is not tested - I don't have (or want) the HW.
> 
> I did NOT remove wmb() use in tx4939ide.c. tx4939ide.c __raw_writeb()
> for MMIO transactions. __raw_writeb() does NOT guarantee memory
> transaction ordering.
> 
> tx4939ide also uses mmiowb(). AFAIK, mmiowb() only has an effect on
> SGI IA64 NUMA machines. I'm not going to guess how this driver might work.
> 
> Gmail is broken for sending patches (word wrap). My apologies.
> I've attached the patch: diff-next-remove_wmb_from_ide-01.txt
> 
> Patch is against linux-next tree:
>      git://git.kernel.org/pub/scm/linux/kernel/git/sfr/linux-next
> 
> Signed-off-by: Grant Grundler <grundler@google.com>

[ attachment converted to inline ]

> diff --git a/drivers/ide/ide-dma-sff.c b/drivers/ide/ide-dma-sff.c
> index 16fc46e..e4cdf78 100644
> --- a/drivers/ide/ide-dma-sff.c
> +++ b/drivers/ide/ide-dma-sff.c
> @@ -277,8 +277,6 @@ void ide_dma_start(ide_drive_t *drive)
>  		dma_cmd = inb(hwif->dma_base + ATA_DMA_CMD);
>  		outb(dma_cmd | ATA_DMA_START, hwif->dma_base + ATA_DMA_CMD);
>  	}
> -
> -	wmb();
>  }
>  EXPORT_SYMBOL_GPL(ide_dma_start);
>  
> @@ -286,7 +284,7 @@ EXPORT_SYMBOL_GPL(ide_dma_start);
>  int ide_dma_end(ide_drive_t *drive)
>  {
>  	ide_hwif_t *hwif = drive->hwif;
> -	u8 dma_stat = 0, dma_cmd = 0, mask;
> +	u8 dma_stat = 0, dma_cmd = 0;
>  
>  	/* stop DMA */
>  	if (hwif->host_flags & IDE_HFLAG_MMIO) {
> @@ -304,11 +302,10 @@ int ide_dma_end(ide_drive_t *drive)
>  	/* clear INTR & ERROR bits */
>  	ide_dma_sff_write_status(hwif, dma_stat | ATA_DMA_ERR | ATA_DMA_INTR);
>  
> -	wmb();
> +#define CHECK_DMA_MASK (ATA_DMA_ACTIVE | ATA_DMA_ERR | ATA_DMA_INTR)
>  
>  	/* verify good DMA status */
> -	mask = ATA_DMA_ACTIVE | ATA_DMA_ERR | ATA_DMA_INTR;
> -	if ((dma_stat & mask) != ATA_DMA_INTR)
> +	if ((dma_stat & CHECK_DMA_MASK) != ATA_DMA_INTR)
>  		return 0x10 | dma_stat;
>  	return 0;
>  }
> diff --git a/drivers/ide/scc_pata.c b/drivers/ide/scc_pata.c
> index 97f8e0e..dcbb299 100644
> --- a/drivers/ide/scc_pata.c
> +++ b/drivers/ide/scc_pata.c
> @@ -337,7 +337,6 @@ static void scc_dma_start(ide_drive_t *drive)
>  
>  	/* start DMA */
>  	scc_ide_outb(dma_cmd | 1, hwif->dma_base);
> -	wmb();
>  }
>  
>  static int __scc_dma_end(ide_drive_t *drive)
> @@ -354,7 +353,6 @@ static int __scc_dma_end(ide_drive_t *drive)
>  	/* clear the INTR & ERROR bits */
>  	scc_ide_outb(dma_stat | 6, hwif->dma_base + 4);
>  	/* verify good DMA status */
> -	wmb();
>  	return (dma_stat & 7) != 4 ? (0x10 | dma_stat) : 0;
>  }

With kind regards,

Geert Uytterhoeven
Software Architect

Sony Techsoft Centre Europe
The Corporate Village · Da Vincilaan 7-D1 · B-1935 Zaventem · Belgium

Phone:    +32 (0)2 700 8453
Fax:      +32 (0)2 700 8622
E-mail:   Geert.Uytterhoeven@sonycom.com
Internet: http://www.sony-europe.com/

A division of Sony Europe (Belgium) N.V.
VAT BE 0413.825.160 · RPR Brussels
Fortis · BIC GEBABEBB · IBAN BE41293037680010

Re: [PATCH] linux-next remove wmb() from ide-dma-sff.c and scc_pata.c

[KOBAYASHI Yoshitake]

2009/03/31 16:51, Geert Uytterhoeven wrote:
> On Mon, 30 Mar 2009, Grant Grundler wrote:
>> Followup to "[PATCH 03/10] ide: destroy DMA mappings after ending DMA"
>> email on March 14th:
>>     http://lkml.org/lkml/2009/3/14/17
>>
>> No maintainer is listed for "Toshiba CELL Reference Set IDE" (BLK_DEV_CELLEB)
>> or tx4939ide.c in MAINTAINERS. I've CC'd "Ishizaki Kou" @Toshiba (Maintainer for
>> "Spidernet Network Driver for CELL") and linuxppc-dev list in the hope
>> someone else
>> would know or would be able to ACK this patch.
> 
> tx49xx is MIPS, for Nemoto-san.

The patch looks good for Toshiba Cell Reference Set.
I think the patch will be acked by IDE maintainer.

Thank you for informing me of the contribution.

Regards,

-- Yoshi

Re: [PATCH] linux-next remove wmb() from ide-dma-sff.c and scc_pata.c

[Bartlomiej Zolnierkiewicz]

On Tuesday 31 March 2009, KOBAYASHI Yoshitake wrote:
> 2009/03/31 16:51, Geert Uytterhoeven wrote:
> > On Mon, 30 Mar 2009, Grant Grundler wrote:
> >> Followup to "[PATCH 03/10] ide: destroy DMA mappings after ending DMA"
> >> email on March 14th:
> >>     http://lkml.org/lkml/2009/3/14/17
> >>
> >> No maintainer is listed for "Toshiba CELL Reference Set IDE" (BLK_DEV_CELLEB)
> >> or tx4939ide.c in MAINTAINERS. I've CC'd "Ishizaki Kou" @Toshiba (Maintainer for
> >> "Spidernet Network Driver for CELL") and linuxppc-dev list in the hope
> >> someone else
> >> would know or would be able to ACK this patch.
> > 
> > tx49xx is MIPS, for Nemoto-san.
> 
> The patch looks good for Toshiba Cell Reference Set.

Great, I applied the patch.

Thanks,
Bart

Re: [PATCH] linux-next remove wmb() from ide-dma-sff.c and scc_pata.c

[Atsushi Nemoto]

On Tue, 31 Mar 2009 09:51:53 +0200 (CEST), Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com> wrote:
> > Followup to "[PATCH 03/10] ide: destroy DMA mappings after ending DMA"
> > email on March 14th:
> >     http://lkml.org/lkml/2009/3/14/17
> > 
> > No maintainer is listed for "Toshiba CELL Reference Set IDE" (BLK_DEV_CELLEB)
> > or tx4939ide.c in MAINTAINERS. I've CC'd "Ishizaki Kou" @Toshiba (Maintainer for
> > "Spidernet Network Driver for CELL") and linuxppc-dev list in the hope
> > someone else
> > would know or would be able to ACK this patch.
> 
> tx49xx is MIPS, for Nemoto-san.
> 
> > This patch:
> > o replaces "mask" variable in ide_dma_end() with #define.
> > o removes use of wmb() in ide-dma-sff.c and scc_pata.c.
> > o is not tested - I don't have (or want) the HW.
> > 
> > I did NOT remove wmb() use in tx4939ide.c. tx4939ide.c __raw_writeb()
> > for MMIO transactions. __raw_writeb() does NOT guarantee memory
> > transaction ordering.

The wmb() in tx4939ide.c was just copied from ide_dma_end().  On this
MIPS core memory operations are strictly ordered so that the wmb() can
be removed.

And on MIPS __raw_writeb() and writeb() do same thing except for
endian conversion.

I will send a patch just for tx4939ide.c.  Thank you for suggestion.

> > tx4939ide also uses mmiowb(). AFAIK, mmiowb() only has an effect on
> > SGI IA64 NUMA machines. I'm not going to guess how this driver might work.

On MIPS mmiowb() can be (ab)used to flush write buffer.  Please do not
drop this mmiowb().

---
Atsushi Nemoto