[RFC v2 6/6] pata_marvell: use async probe

[RFC v2 6/6] pata_marvell: use async probe

[Luis R. Rodriguez]

From: "Luis R. Rodriguez" <mcgrof@suse.com>

Alexander reported that on his Sony VAIO VPCZ23A4R laptop
experiences long delays on boot when connected to its dock
station on pre 3.9 kernels but anything after 3.9 will cause
the device to not be detected at all ending with:

[   38.065673] pata_marvell 0000:1a:00.0: no available native port
[   38.065769] pata_acpi 0000:1a:00.0: no available native port

This laptop has a Marvell 88SE6121 SATA II Controller [11ab:6121]
and a BluRay writer attached. The reason for the delays are
caused by SRST errors and the link being slow to respond.
The pata_marvell driver is a simple libata wrapper so the
real required changes need to be made on libata however not
many folks are around and available anymore with intimate
knowledge and experience with these devices. Alexander notes
that it may be that *any* ATA BMDMA controller that fails to
respond to an identify command until a reset or other device
poking might suffer from similar fate, this needs to be
investigated further. Using async probe the issue caused
by systemd killing the driver after taking over 30 seconds
on probe.

[0] https://bugzilla.kernel.org/show_bug.cgi?id=59581

Cc: Tejun Heo <tj@kernel.org>
Cc: linux-ide@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: One Thousand Gnomes <gnomes@lxorguk.ukuu.org.uk>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Benjamin Poirier <bpoirier@suse.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: patrakov@gmail.com
Reported-by: "Alexander E. Patrakov" <patrakov@gmail.com>
Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
---
 drivers/ata/pata_marvell.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/ata/pata_marvell.c b/drivers/ata/pata_marvell.c
index ae9feb1..6a543b9 100644
--- a/drivers/ata/pata_marvell.c
+++ b/drivers/ata/pata_marvell.c
@@ -175,6 +175,7 @@ static struct pci_driver marvell_pci_driver = {
 	.suspend		= ata_pci_device_suspend,
 	.resume			= ata_pci_device_resume,
 #endif
+	.driver.async_probe = true,
 };
 
 module_pci_driver(marvell_pci_driver);
-- 
2.0.3

Re: [RFC v2 6/6] pata_marvell: use async probe

[Alexander E. Patrakov]

05.09.2014 12:37, Luis R. Rodriguez пишет:
> From: "Luis R. Rodriguez" <mcgrof@suse.com>
>
> Alexander reported that on his Sony VAIO VPCZ23A4R laptop
> experiences long delays on boot when connected to its dock
> station on pre 3.9 kernels but anything after 3.9 will cause
> the device to not be detected at all ending with:
>
> [   38.065673] pata_marvell 0000:1a:00.0: no available native port
> [   38.065769] pata_acpi 0000:1a:00.0: no available native port

I object to this commit message, it is based on outdated information and 
is due to a different bug that was fixed in 3.10 as a last-minute fix. 
Modern kernels just experience long delays during boot.

> This laptop has a Marvell 88SE6121 SATA II Controller [11ab:6121]
> and a BluRay writer attached. The reason for the delays are
> caused by SRST errors and the link being slow to respond.
> The pata_marvell driver is a simple libata wrapper so the
> real required changes need to be made on libata however not
> many folks are around and available anymore with intimate
> knowledge and experience with these devices. Alexander notes
> that it may be that *any* ATA BMDMA controller that fails to
> respond to an identify command until a reset or other device
> poking might suffer from similar fate, this needs to be
> investigated further. Using async probe the issue caused
> by systemd killing the driver after taking over 30 seconds
> on probe.
>
> [0] https://bugzilla.kernel.org/show_bug.cgi?id=59581
>
> Cc: Tejun Heo <tj@kernel.org>
> Cc: linux-ide@vger.kernel.org
> Cc: linux-kernel@vger.kernel.org
> Cc: One Thousand Gnomes <gnomes@lxorguk.ukuu.org.uk>
> Cc: Oleg Nesterov <oleg@redhat.com>
> Cc: Benjamin Poirier <bpoirier@suse.de>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: patrakov@gmail.com
> Reported-by: "Alexander E. Patrakov" <patrakov@gmail.com>
> Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
> ---
>   drivers/ata/pata_marvell.c | 1 +
>   1 file changed, 1 insertion(+)
>
> diff --git a/drivers/ata/pata_marvell.c b/drivers/ata/pata_marvell.c
> index ae9feb1..6a543b9 100644
> --- a/drivers/ata/pata_marvell.c
> +++ b/drivers/ata/pata_marvell.c
> @@ -175,6 +175,7 @@ static struct pci_driver marvell_pci_driver = {
>   	.suspend		= ata_pci_device_suspend,
>   	.resume			= ata_pci_device_resume,
>   #endif
> +	.driver.async_probe = true,
>   };
>
>   module_pci_driver(marvell_pci_driver);
>


-- 
Alexander E. Patrakov

Re: [RFC v2 6/6] pata_marvell: use async probe

[Tejun Heo]

On Thu, Sep 04, 2014 at 11:37:27PM -0700, Luis R. Rodriguez wrote:
> diff --git a/drivers/ata/pata_marvell.c b/drivers/ata/pata_marvell.c
> index ae9feb1..6a543b9 100644
> --- a/drivers/ata/pata_marvell.c
> +++ b/drivers/ata/pata_marvell.c
> @@ -175,6 +175,7 @@ static struct pci_driver marvell_pci_driver = {
>  	.suspend		= ata_pci_device_suspend,
>  	.resume			= ata_pci_device_resume,
>  #endif
> +	.driver.async_probe = true,

You can't do this.  There's nothing special about pata_marvell.  Sure
there was a bug report which made long probe durations more common on
this driver on certain configurations but those long durations can
happen on *any* libata driver and singling out pata_marvell for async
probing is adding a different probing behavior basically arbitrarily.
I really can't see how this marking random drivers with async probing
would work, so one driver does synchronous probing while the
equivalent next one doesn't?  That's crazy.

Thanks.

-- 
tejun