2020-11-06 07:58:53

by 王擎

[permalink] [raw]
Subject: [PATCH] net/ethernet: update ret when ptp_clock is ERROR

We always have to update the value of ret, otherwise the
error value may be the previous one.

Signed-off-by: Wang Qing <[email protected]>
---
drivers/net/ethernet/ti/am65-cpts.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/ti/am65-cpts.c b/drivers/net/ethernet/ti/am65-cpts.c
index 75056c1..b77ff61
--- a/drivers/net/ethernet/ti/am65-cpts.c
+++ b/drivers/net/ethernet/ti/am65-cpts.c
@@ -1001,8 +1001,7 @@ struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
if (IS_ERR_OR_NULL(cpts->ptp_clock)) {
dev_err(dev, "Failed to register ptp clk %ld\n",
PTR_ERR(cpts->ptp_clock));
- if (!cpts->ptp_clock)
- ret = -ENODEV;
+ ret = cpts->ptp_clock ? cpts->ptp_clock : (-ENODEV);
goto refclk_disable;
}
cpts->phc_index = ptp_clock_index(cpts->ptp_clock);
--
2.7.4


2020-11-06 11:38:21

by Grygorii Strashko

[permalink] [raw]
Subject: Re: [PATCH] net/ethernet: update ret when ptp_clock is ERROR



On 06/11/2020 09:56, Wang Qing wrote:
> We always have to update the value of ret, otherwise the
> error value may be the previous one.
>
> Signed-off-by: Wang Qing <[email protected]>
> ---
> drivers/net/ethernet/ti/am65-cpts.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/ti/am65-cpts.c b/drivers/net/ethernet/ti/am65-cpts.c
> index 75056c1..b77ff61
> --- a/drivers/net/ethernet/ti/am65-cpts.c
> +++ b/drivers/net/ethernet/ti/am65-cpts.c
> @@ -1001,8 +1001,7 @@ struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,

there is
cpts->ptp_clock = ptp_clock_register(&cpts->ptp_info, cpts->dev);


> if (IS_ERR_OR_NULL(cpts->ptp_clock)) {

And ptp_clock_register() can return NULL only if PTP support is disabled.
In which case, we should not even get here.

So, I'd propose to s/IS_ERR_OR_NULL/IS_ERR above,
and just assign ret = PTR_ERR(cpts->ptp_clock) here.

> dev_err(dev, "Failed to register ptp clk %ld\n",
> PTR_ERR(cpts->ptp_clock));
> - if (!cpts->ptp_clock)
> - ret = -ENODEV;
> + ret = cpts->ptp_clock ? cpts->ptp_clock : (-ENODEV);
> goto refclk_disable;
> }
> cpts->phc_index = ptp_clock_index(cpts->ptp_clock);
>

--
Best regards,
grygorii

2020-11-06 12:16:24

by Arnd Bergmann

[permalink] [raw]
Subject: Re: [PATCH] net/ethernet: update ret when ptp_clock is ERROR

On Fri, Nov 6, 2020 at 12:35 PM Grygorii Strashko
<[email protected]> wrote:
> On 06/11/2020 09:56, Wang Qing wrote:

> > +++ b/drivers/net/ethernet/ti/am65-cpts.c
> > @@ -1001,8 +1001,7 @@ struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
>
> there is
> cpts->ptp_clock = ptp_clock_register(&cpts->ptp_info, cpts->dev);
>
>
> > if (IS_ERR_OR_NULL(cpts->ptp_clock)) {
>
> And ptp_clock_register() can return NULL only if PTP support is disabled.
> In which case, we should not even get here.
>
> So, I'd propose to s/IS_ERR_OR_NULL/IS_ERR above,
> and just assign ret = PTR_ERR(cpts->ptp_clock) here.

Right, using IS_ERR_OR_NULL() is almost ever a mistake, either
from misunderstanding the interface, or from a badly designed
interface that needs to be changed.

Arnd

2020-11-06 13:00:20

by Kurt Kanzenbach

[permalink] [raw]
Subject: Re: [PATCH] net/ethernet: update ret when ptp_clock is ERROR

On Fri Nov 06 2020, Arnd Bergmann wrote:
> On Fri, Nov 6, 2020 at 12:35 PM Grygorii Strashko
> <[email protected]> wrote:
>> On 06/11/2020 09:56, Wang Qing wrote:
>
>> > +++ b/drivers/net/ethernet/ti/am65-cpts.c
>> > @@ -1001,8 +1001,7 @@ struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
>>
>> there is
>> cpts->ptp_clock = ptp_clock_register(&cpts->ptp_info, cpts->dev);
>>
>>
>> > if (IS_ERR_OR_NULL(cpts->ptp_clock)) {
>>
>> And ptp_clock_register() can return NULL only if PTP support is disabled.
>> In which case, we should not even get here.
>>
>> So, I'd propose to s/IS_ERR_OR_NULL/IS_ERR above,
>> and just assign ret = PTR_ERR(cpts->ptp_clock) here.
>
> Right, using IS_ERR_OR_NULL() is almost ever a mistake, either
> from misunderstanding the interface, or from a badly designed
> interface that needs to be changed.

The NULL case should be handled differently and it is documented:

/**
* ptp_clock_register() - register a PTP hardware clock driver
[...]
* Returns a valid pointer on success or PTR_ERR on failure. If PHC
* support is missing at the configuration level, this function
* returns NULL, and drivers are expected to gracefully handle that
* case separately.
*/

Thanks,
Kurt


Attachments:
signature.asc (847.00 B)

2020-11-06 14:50:46

by Grygorii Strashko

[permalink] [raw]
Subject: Re: [PATCH] net/ethernet: update ret when ptp_clock is ERROR



On 06/11/2020 14:58, Kurt Kanzenbach wrote:
> On Fri Nov 06 2020, Arnd Bergmann wrote:
>> On Fri, Nov 6, 2020 at 12:35 PM Grygorii Strashko
>> <[email protected]> wrote:
>>> On 06/11/2020 09:56, Wang Qing wrote:
>>
>>>> +++ b/drivers/net/ethernet/ti/am65-cpts.c
>>>> @@ -1001,8 +1001,7 @@ struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
>>>
>>> there is
>>> cpts->ptp_clock = ptp_clock_register(&cpts->ptp_info, cpts->dev);
>>>
>>>
>>>> if (IS_ERR_OR_NULL(cpts->ptp_clock)) {
>>>
>>> And ptp_clock_register() can return NULL only if PTP support is disabled.
>>> In which case, we should not even get here.
>>>
>>> So, I'd propose to s/IS_ERR_OR_NULL/IS_ERR above,
>>> and just assign ret = PTR_ERR(cpts->ptp_clock) here.
>>
>> Right, using IS_ERR_OR_NULL() is almost ever a mistake, either
>> from misunderstanding the interface, or from a badly designed
>> interface that needs to be changed.
>
> The NULL case should be handled differently and it is documented:
>
> /**
> * ptp_clock_register() - register a PTP hardware clock driver
> [...]
> * Returns a valid pointer on success or PTR_ERR on failure. If PHC
> * support is missing at the configuration level, this function
> * returns NULL, and drivers are expected to gracefully handle that
> * case separately.
> */

I think, it's not the first time such question triggered, I've found [1]

I've managed to find 6 drivers which uses IS_ERR_OR_NULL check with ptp_clock_register() and, kinda,
assume to be able to work with !CONFIG_PTP_1588_CLOCK. List below with some comments:

hv
hv_util.c
hv_timesync_init, line 697: hv_ptp_clock = ptp_clock_register(&ptp_hyperv_info, NULL);

net/ethernet
sja1105 (depends on PTP_1588_CLOCK, use IS_ERR())
sja1105_ptp.c
sja1105_ptp_clock_register, line 867: ptp_data->clock = ptp_clock_register(&ptp_data->caps, ds->dev);

net/ethernet
chelsio (can be fixed by adding !IS_ENABLED(CONFIG_PTP_1588_CLOCK))
cxgb4
cxgb4_ptp.c
cxgb4_ptp_init, line 431: adapter->ptp_clock = ptp_clock_register(&adapter->ptp_clock_info,

net/ethernet
octeontx2 (can be fixed by adding !IS_ENABLED(CONFIG_PTP_1588_CLOCK))
nic
otx2_ptp.c
otx2_ptp_init, line 170: ptp_ptr->ptp_clock = ptp_clock_register(&ptp_ptr->ptp_info, pfvf->dev);

net/ethernet
renesas (no checks - will crash if init failed or !PTP), uses imply PTP_1588_CLOCK
ravb_ptp.c
ravb_ptp_init, line 345: priv->ptp.clock = ptp_clock_register(&priv->ptp.info, &pdev->dev);

net/phy
mscc (no checks - will crash if init failed or !PTP, depends on NETWORK_PHY_TIMESTAMPING)
mscc_ptp.c
__vsc8584_init_ptp, line 1495: vsc8531->ptp->ptp_clock = ptp_clock_register(&vsc8531->ptp->caps,

In general, if above updated, the return value for ptp_clock_register() can be changed to ERR_PTR(-EOPNOTSUPP)
for the !CONFIG_PTP_1588_CLOCK and so question resolved.

For hv/sja1105/cxgb4/octeontx2 below diff should solve the case, but I'm not sure about
renesas/ravb_ptp and net/phy/mscc.

For renesas/ravb_ptp - seems strict "depends on PTP_1588_CLOCK" can be the choice
For net/phy/mscc - seems code dependencies need to be changed from CONFIG_NETWORK_PHY_TIMESTAMPING to
CONFIG_PTP_1588_CLOCK.

[1] https://lore.kernel.org/lkml/[email protected]/
--
Best regards,
grygorii

-------------
diff --git a/drivers/hv/hv_util.c b/drivers/hv/hv_util.c
index 05566ecdbe4b..0fa8f6cb2394 100644
--- a/drivers/hv/hv_util.c
+++ b/drivers/hv/hv_util.c
@@ -681,6 +681,11 @@ static struct ptp_clock *hv_ptp_clock;

static int hv_timesync_init(struct hv_util_service *srv)
{
+ int ret = 0;
+
+ if (!IS_ENABLED(CONFIG_PTP_1588_CLOCK))
+ return -EOPNOTSUPP;
+
/* TimeSync requires Hyper-V clocksource. */
if (!hv_read_reference_counter)
return -ENODEV;
@@ -695,13 +700,13 @@ static int hv_timesync_init(struct hv_util_service *srv)
* as it still handles the ICTIMESYNCFLAG_SYNC case.
*/
hv_ptp_clock = ptp_clock_register(&ptp_hyperv_info, NULL);
- if (IS_ERR_OR_NULL(hv_ptp_clock)) {
- pr_err("cannot register PTP clock: %ld\n",
- PTR_ERR(hv_ptp_clock));
+ if (IS_ERR(hv_ptp_clock)) {
+ ret = PTR_ERR(hv_ptp_clock);
+ pr_err("cannot register PTP clock: %ld\n", ret);
hv_ptp_clock = NULL;
}

- return 0;
+ return ret;
}

static void hv_timesync_cancel_work(void)
diff --git a/drivers/net/dsa/sja1105/sja1105_ptp.c b/drivers/net/dsa/sja1105/sja1105_ptp.c
index 1b90570b257b..1e41d491c854 100644
--- a/drivers/net/dsa/sja1105/sja1105_ptp.c
+++ b/drivers/net/dsa/sja1105/sja1105_ptp.c
@@ -865,7 +865,7 @@ int sja1105_ptp_clock_register(struct dsa_switch *ds)
spin_lock_init(&tagger_data->meta_lock);

ptp_data->clock = ptp_clock_register(&ptp_data->caps, ds->dev);
- if (IS_ERR_OR_NULL(ptp_data->clock))
+ if (IS_ERR(ptp_data->clock))
return PTR_ERR(ptp_data->clock);

ptp_data->cmd.corrclk4ts = true;
diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ptp.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ptp.c
index 70dbee89118e..e43a3e73762b 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ptp.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ptp.c
@@ -420,6 +420,10 @@ static const struct ptp_clock_info cxgb4_ptp_clock_info = {
void cxgb4_ptp_init(struct adapter *adapter)
{
struct timespec64 now;
+
+ if (!IS_ENABLED(CONFIG_PTP_1588_CLOCK))
+ return;
+
/* no need to create a clock device if we already have one */
if (!IS_ERR_OR_NULL(adapter->ptp_clock))
return;
@@ -430,7 +434,7 @@ void cxgb4_ptp_init(struct adapter *adapter)

adapter->ptp_clock = ptp_clock_register(&adapter->ptp_clock_info,
&adapter->pdev->dev);
- if (IS_ERR_OR_NULL(adapter->ptp_clock)) {
+ if (IS_ERR(adapter->ptp_clock)) {
adapter->ptp_clock = NULL;
dev_err(adapter->pdev_dev,
"PTP %s Clock registration has failed\n", __func__);
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ptp.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ptp.c
index 7bcf5246350f..22b736c2e9d6 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ptp.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ptp.c
@@ -119,6 +119,9 @@ int otx2_ptp_init(struct otx2_nic *pfvf)
struct ptp_req *req;
int err;

+ if (!IS_ENABLED(CONFIG_PTP_1588_CLOCK))
+ return 0;
+
mutex_lock(&pfvf->mbox.lock);
/* check if PTP block is available */
req = otx2_mbox_alloc_msg_ptp_op(&pfvf->mbox);
@@ -168,9 +171,8 @@ int otx2_ptp_init(struct otx2_nic *pfvf)
};

ptp_ptr->ptp_clock = ptp_clock_register(&ptp_ptr->ptp_info, pfvf->dev);
- if (IS_ERR_OR_NULL(ptp_ptr->ptp_clock)) {
- err = ptp_ptr->ptp_clock ?
- PTR_ERR(ptp_ptr->ptp_clock) : -ENODEV;
+ if (IS_ERR(ptp_ptr->ptp_clock)) {
+ err = PTR_ERR(ptp_ptr->ptp_clock);
kfree(ptp_ptr);
goto error;
}

2020-11-07 15:09:03

by Richard Cochran

[permalink] [raw]
Subject: Re: [PATCH] net/ethernet: update ret when ptp_clock is ERROR

On Fri, Nov 06, 2020 at 01:34:04PM +0200, Grygorii Strashko wrote:
> And ptp_clock_register() can return NULL only if PTP support is disabled.

Not true in general ...

> In which case, we should not even get here.

only because the Kconfig uses "depends on" instead of "implies"
PTP_1588_CLOCK.

> So, I'd propose to s/IS_ERR_OR_NULL/IS_ERR above,
> and just assign ret = PTR_ERR(cpts->ptp_clock) here.

No, please no -- don't make another bad example for people to
copy/paste.

Thanks,
Richard

2020-11-07 15:10:45

by Richard Cochran

[permalink] [raw]
Subject: Re: [PATCH] net/ethernet: update ret when ptp_clock is ERROR

On Fri, Nov 06, 2020 at 03:56:45PM +0800, Wang Qing wrote:
> We always have to update the value of ret, otherwise the
> error value may be the previous one.
>
> Signed-off-by: Wang Qing <[email protected]>

Acked-by: Richard Cochran <[email protected]>


> ---
> drivers/net/ethernet/ti/am65-cpts.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/ti/am65-cpts.c b/drivers/net/ethernet/ti/am65-cpts.c
> index 75056c1..b77ff61
> --- a/drivers/net/ethernet/ti/am65-cpts.c
> +++ b/drivers/net/ethernet/ti/am65-cpts.c
> @@ -1001,8 +1001,7 @@ struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
> if (IS_ERR_OR_NULL(cpts->ptp_clock)) {
> dev_err(dev, "Failed to register ptp clk %ld\n",
> PTR_ERR(cpts->ptp_clock));
> - if (!cpts->ptp_clock)
> - ret = -ENODEV;
> + ret = cpts->ptp_clock ? cpts->ptp_clock : (-ENODEV);
> goto refclk_disable;
> }
> cpts->phc_index = ptp_clock_index(cpts->ptp_clock);
> --
> 2.7.4
>

2020-11-11 13:07:21

by Grygorii Strashko

[permalink] [raw]
Subject: Re: [PATCH] net/ethernet: update ret when ptp_clock is ERROR



On 07/11/2020 17:08, Richard Cochran wrote:
> On Fri, Nov 06, 2020 at 03:56:45PM +0800, Wang Qing wrote:
>> We always have to update the value of ret, otherwise the
>> error value may be the previous one.
>>
>> Signed-off-by: Wang Qing <[email protected]>
>
> Acked-by: Richard Cochran <[email protected]>
>

Following Richard's comments:

Reviewed-by: Grygorii Strashko <[email protected]>

>
>> ---
>> drivers/net/ethernet/ti/am65-cpts.c | 3 +--
>> 1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/ti/am65-cpts.c b/drivers/net/ethernet/ti/am65-cpts.c
>> index 75056c1..b77ff61
>> --- a/drivers/net/ethernet/ti/am65-cpts.c
>> +++ b/drivers/net/ethernet/ti/am65-cpts.c
>> @@ -1001,8 +1001,7 @@ struct am65_cpts *am65_cpts_create(struct device *dev, void __iomem *regs,
>> if (IS_ERR_OR_NULL(cpts->ptp_clock)) {
>> dev_err(dev, "Failed to register ptp clk %ld\n",
>> PTR_ERR(cpts->ptp_clock));
>> - if (!cpts->ptp_clock)
>> - ret = -ENODEV;
>> + ret = cpts->ptp_clock ? cpts->ptp_clock : (-ENODEV);
>> goto refclk_disable;
>> }
>> cpts->phc_index = ptp_clock_index(cpts->ptp_clock);
>> --
>> 2.7.4
>>

--
Best regards,
grygorii