Hi,
On 03-12-18 15:46, Miquel Raynal wrote:
Hi Hans,
Hans de Goede <hdegoede@xxxxxxxxxx> wrote on Sun, 2 Dec 2018 14:19:49
+0100:
Hi,
On 30-11-18 16:38, Miquel Raynal wrote:
Current implementation of the libahci does not take into account the
new PHY framework. Correct the situation by adding a call to
phy_set_mode() before phy_power_on() and by adding calls to
ahci_platform_enable/disable_phys() at suspend/resume_host() time.
Suggested-by: Grzegorz Jaszczyk <jaz@xxxxxxxxxxxx>
Signed-off-by: Miquel Raynal <miquel.raynal@xxxxxxxxxxx>
Much better. Some remarks inline.
---
drivers/ata/ahci.h | 2 ++
drivers/ata/libahci_platform.c | 19 ++++++++++++++++++-
2 files changed, 20 insertions(+), 1 deletion(-)
diff --git a/drivers/ata/ahci.h b/drivers/ata/ahci.h
index ef356e70e6de..982638b37fee 100644
--- a/drivers/ata/ahci.h
+++ b/drivers/ata/ahci.h
@@ -254,6 +254,8 @@ enum {
AHCI_HFLAG_IS_MOBILE = (1 << 25), /* mobile chipset, use
SATA_MOBILE_LPM_POLICY
as default lpm_policy */
+ AHCI_HFLAG_MANAGE_PHYS = (1 << 26), /* let the core manage the
+ PHYs when relevant */
Using a hflag for this is a good idea, but the MANAGE_PHYS names is a bit
generic, even if this is not set then ahci_platform_enable_resources() will
still enable the phy. So I would name this AHCI_HFLAG_SUSPEND_PHYS to make
clear this only influences suspend/resume behavior.
I understand and will modify the HFLAG name as you suggest.
Also see below.
/* ap->flags bits */
> diff --git a/drivers/ata/libahci_platform.c b/drivers/ata/libahci_platform.c
index 4b900fc659f7..f5a64eb1fea8 100644
--- a/drivers/ata/libahci_platform.c
+++ b/drivers/ata/libahci_platform.c
@@ -56,6 +56,12 @@ static int ahci_platform_enable_phys(struct ahci_host_priv *hpriv)
if (rc)
goto disable_phys;
> + rc = phy_set_mode(hpriv->phys[i], PHY_MODE_SATA);
+ if (rc) {
+ phy_exit(hpriv->phys[i]);
+ goto disable_phys;
+ }
+
rc = phy_power_on(hpriv->phys[i]);
if (rc) {
phy_exit(hpriv->phys[i]);
@@ -378,7 +384,8 @@ static int ahci_platform_get_regulator(struct ahci_host_priv *hpriv, u32 port,
* 3) 0 - AHCI_MAX_CLKS clocks, as specified in the devs devicetree node,
* or for non devicetree enabled platforms a single clock
* 4) resets, if flags has AHCI_PLATFORM_GET_RESETS (optional)
- * 5) phys (optional)
+ * 5) phys (optional), PHY handling during suspend/resume will be skipped if the
+ * flag AHCI_HFLAG_MANAGE_PHYS is missing.
*
* RETURNS:
* The allocated ahci_host_priv on success, otherwise an ERR_PTR value
@@ -458,6 +465,9 @@ struct ahci_host_priv *ahci_platform_get_resources(struct platform_device *pdev,
}
}
> + if (flags & AHCI_HFLAG_MANAGE_PHYS)
+ hpriv->flags |= AHCI_HFLAG_MANAGE_PHYS;
+
hpriv->nports = child_nodes = of_get_child_count(dev->of_node);
> /*
Interpreting a flag passed to ahci_platform_get_resources using a
AHCI_HFLAG_* mask is asking for trouble in the future.
Since you are using a hflag for this, you can just drop this chunk / change
and in the caller of ahci_platform_get_resources() do:
hpriv = ahci_platform_get_resources(pdev, ...);
if (IS_ERR(hpriv))
...
hpriv->flags |= AHCI_HFLAG_MANAGE_PHYS;
Settting hflags like this after calling ahci_platform_get_resources()
is already done in several place.
Indeed, this is close to my first draft but after re-reading your
previous review I understood you wanted this flag to be passed to
ahci_platform_get_resources().
Yes that was my initial thought, but sine you are using a hflag for
it that is no longer a good idea.
I do agree that the current form is a bit redundant so I will let users
add the flag manually during the probe after the
ahci_platform_get_resources() call, as suggested.
Sounds good, thanks.
Regards,
Hans
