Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

[Sergei Shtylyov]

Hello.

On 15-06-2012 12:09, Huacai Chen wrote:

> Signed-off-by: Huacai Chen<chenhc@lemote.com>
> Signed-off-by: Hongliang Tao<taohl@lemote.com>
> Signed-off-by: Hua Yan<yanh@lemote.com>

    You  should have CCed the 'linux-ide' mailing list.

> diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c
> index ebaf67e..3e3cfd8 100644
> --- a/drivers/ata/ahci.c
> +++ b/drivers/ata/ahci.c
> @@ -183,7 +183,12 @@ static const struct ata_port_info ahci_port_info[] = {
>   	},
>   	[board_ahci_sb700] =	/* for SB700 and SB800 */
>   	{
> +#ifndef CONFIG_CPU_LOONGSON3
>   		AHCI_HFLAGS	(AHCI_HFLAG_IGN_SERR_INTERNAL),
> +#else
> +		AHCI_HFLAGS	(AHCI_HFLAG_IGN_SERR_INTERNAL |
> +						AHCI_HFLAG_32BIT_ONLY),
> +#endif

    No, this #ifdef'ery won't do. You should add a new board type.

MBR, Sergei

Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

[huacai chen]

On Fri, Jun 15, 2012 at 6:04 PM, Sergei Shtylyov <sshtylyov@mvista.com> wrote:
> Hello.
>
>
> On 15-06-2012 12:09, Huacai Chen wrote:
>
>> Signed-off-by: Huacai Chen<chenhc@lemote.com>
>> Signed-off-by: Hongliang Tao<taohl@lemote.com>
>> Signed-off-by: Hua Yan<yanh@lemote.com>
>
>
>   You  should have CCed the 'linux-ide' mailing list.
>
>
>> diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c
>> index ebaf67e..3e3cfd8 100644
>> --- a/drivers/ata/ahci.c
>> +++ b/drivers/ata/ahci.c
>> @@ -183,7 +183,12 @@ static const struct ata_port_info ahci_port_info[] =
>> {
>>        },
>>        [board_ahci_sb700] =    /* for SB700 and SB800 */
>>        {
>> +#ifndef CONFIG_CPU_LOONGSON3
>>                AHCI_HFLAGS     (AHCI_HFLAG_IGN_SERR_INTERNAL),
>> +#else
>> +               AHCI_HFLAGS     (AHCI_HFLAG_IGN_SERR_INTERNAL |
>> +                                               AHCI_HFLAG_32BIT_ONLY),
>> +#endif
>
>
>   No, this #ifdef'ery won't do. You should add a new board type.
All Loongson-3 based machines use AMD SB700 chipsets, add a new board
type is better than #ifdef?

>
> MBR, Sergei

Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

[Borislav Petkov]

On Fri, Jun 15, 2012 at 08:42:47PM +0800, huacai chen wrote:
> On Fri, Jun 15, 2012 at 6:04 PM, Sergei Shtylyov <sshtylyov@mvista.com> wrote:
> > Hello.
> >
> >
> > On 15-06-2012 12:09, Huacai Chen wrote:
> >
> >> Signed-off-by: Huacai Chen<chenhc@lemote.com>
> >> Signed-off-by: Hongliang Tao<taohl@lemote.com>
> >> Signed-off-by: Hua Yan<yanh@lemote.com>
> >
> >
> >   You  should have CCed the 'linux-ide' mailing list.
> >
> >
> >> diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c
> >> index ebaf67e..3e3cfd8 100644
> >> --- a/drivers/ata/ahci.c
> >> +++ b/drivers/ata/ahci.c
> >> @@ -183,7 +183,12 @@ static const struct ata_port_info ahci_port_info[] =
> >> {
> >>        },
> >>        [board_ahci_sb700] =    /* for SB700 and SB800 */
> >>        {
> >> +#ifndef CONFIG_CPU_LOONGSON3
> >>                AHCI_HFLAGS     (AHCI_HFLAG_IGN_SERR_INTERNAL),
> >> +#else
> >> +               AHCI_HFLAGS     (AHCI_HFLAG_IGN_SERR_INTERNAL |
> >> +                                               AHCI_HFLAG_32BIT_ONLY),
> >> +#endif
> >
> >
> >   No, this #ifdef'ery won't do. You should add a new board type.
> All Loongson-3 based machines use AMD SB700 chipsets, add a new board
> type is better than #ifdef?

SB700/800 chipsets don't need to set a 32-bit only DMA flag; why do you
need it when you use the same chipset?

-- 
Regards/Gruss,
    Boris.

Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

[huacai chen]

Do you means it is a better idea to modify "enum board_ids" and add a
new board id such as board_ahci_sb700_loongson, and then add a new
entry in ahci_port_info[]?
If so, I think there is a problem: the pci id of our AHCI controller
is 1002:4390, if I add board_ahci_sb700_loongson, then I should also
add
{ PCI_VDEVICE(ATI, 0x4390), board_ahci_sb700_loongson },
in ahci_pci_tbl[], but ahci_pci_tbl[] already has a line
{ PCI_VDEVICE(ATI, 0x4390), board_ahci_sb700 },
Then which entry will match the device?

On Sun, Jun 17, 2012 at 8:05 PM, Borislav Petkov <bp@alien8.de> wrote:
> On Fri, Jun 15, 2012 at 08:42:47PM +0800, huacai chen wrote:
>> On Fri, Jun 15, 2012 at 6:04 PM, Sergei Shtylyov <sshtylyov@mvista.com> wrote:
>> > Hello.
>> >
>> >
>> > On 15-06-2012 12:09, Huacai Chen wrote:
>> >
>> >> Signed-off-by: Huacai Chen<chenhc@lemote.com>
>> >> Signed-off-by: Hongliang Tao<taohl@lemote.com>
>> >> Signed-off-by: Hua Yan<yanh@lemote.com>
>> >
>> >
>> >   You  should have CCed the 'linux-ide' mailing list.
>> >
>> >
>> >> diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c
>> >> index ebaf67e..3e3cfd8 100644
>> >> --- a/drivers/ata/ahci.c
>> >> +++ b/drivers/ata/ahci.c
>> >> @@ -183,7 +183,12 @@ static const struct ata_port_info ahci_port_info[] =
>> >> {
>> >>        },
>> >>        [board_ahci_sb700] =    /* for SB700 and SB800 */
>> >>        {
>> >> +#ifndef CONFIG_CPU_LOONGSON3
>> >>                AHCI_HFLAGS     (AHCI_HFLAG_IGN_SERR_INTERNAL),
>> >> +#else
>> >> +               AHCI_HFLAGS     (AHCI_HFLAG_IGN_SERR_INTERNAL |
>> >> +                                               AHCI_HFLAG_32BIT_ONLY),
>> >> +#endif
>> >
>> >
>> >   No, this #ifdef'ery won't do. You should add a new board type.
>> All Loongson-3 based machines use AMD SB700 chipsets, add a new board
>> type is better than #ifdef?
>
> SB700/800 chipsets don't need to set a 32-bit only DMA flag; why do you
> need it when you use the same chipset?
>
> --
> Regards/Gruss,
>    Boris.

Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

[Borislav Petkov]

On Mon, Jun 18, 2012 at 05:04:14PM +0800, huacai chen wrote:
> Do you means it is a better idea to modify "enum board_ids" and add a
> new board id such as board_ahci_sb700_loongson, and then add a new
> entry in ahci_port_info[]?
> If so, I think there is a problem: the pci id of our AHCI controller
> is 1002:4390, if I add board_ahci_sb700_loongson, then I should also
> add
> { PCI_VDEVICE(ATI, 0x4390), board_ahci_sb700_loongson },
> in ahci_pci_tbl[], but ahci_pci_tbl[] already has a line
> { PCI_VDEVICE(ATI, 0x4390), board_ahci_sb700 },
> Then which entry will match the device?

Before you do anything, my question is:

SB700/800 chipsets don't need to set a 32-bit only DMA flag; why do you
need it when you use the same chipset?

So why do you need to do 32-bit DMA only when the chipset supports
64-bit DMA just fine?

Thanks.

-- 
Regards/Gruss,
    Boris.

Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

[Huacai Chen]

Ohh, this is because Loongson-3 has a hardware bug, when the HT
controller receive a 64-bit DMA address, the high bits is lost. So set
to 32-bit DMA is just a workaround.

I'm sorry that I didn't describe this clearly.

On Mon, Jun 18, 2012 at 6:10 PM, Borislav Petkov <bp@alien8.de> wrote:
> On Mon, Jun 18, 2012 at 05:04:14PM +0800, huacai chen wrote:
>> Do you means it is a better idea to modify "enum board_ids" and add a
>> new board id such as board_ahci_sb700_loongson, and then add a new
>> entry in ahci_port_info[]?
>> If so, I think there is a problem: the pci id of our AHCI controller
>> is 1002:4390, if I add board_ahci_sb700_loongson, then I should also
>> add
>> { PCI_VDEVICE(ATI, 0x4390), board_ahci_sb700_loongson },
>> in ahci_pci_tbl[], but ahci_pci_tbl[] already has a line
>> { PCI_VDEVICE(ATI, 0x4390), board_ahci_sb700 },
>> Then which entry will match the device?
>
> Before you do anything, my question is:
>
> SB700/800 chipsets don't need to set a 32-bit only DMA flag; why do you
> need it when you use the same chipset?
>
> So why do you need to do 32-bit DMA only when the chipset supports
> 64-bit DMA just fine?
>
> Thanks.
>
> --
> Regards/Gruss,
>    Boris.

Re: [PATCH 09/14] ata: Use 32bit DMA in AHCI for Loongson 3.

[Borislav Petkov]

On Mon, Jun 18, 2012 at 06:41:58PM +0800, Huacai Chen wrote:
> Ohh, this is because Loongson-3 has a hardware bug, when the HT
> controller receive a 64-bit DMA address, the high bits is lost. So set
> to 32-bit DMA is just a workaround.

Ok, this should definitely go into the commit message of your patch then
since it explains why you need that flag set.

Thanks.

-- 
Regards/Gruss,
Boris.