Message-ID: <aebaa171-bf4e-c143-a186-a37cd34b724e@ti.com>
Date:   Tue, 17 Jan 2023 10:30:26 +0530
From:   Siddharth Vadapalli <s-vadapalli@...com>
To:     Roger Quadros <rogerq@...nel.org>,
        Leon Romanovsky <leon@...nel.org>
CC:     <davem@...emloft.net>, <edumazet@...gle.com>, <kuba@...nel.org>,
        <linux@...linux.org.uk>, <pabeni@...hat.com>,
        <netdev@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
        <linux-arm-kernel@...ts.infradead.org>, <vigneshr@...com>,
        <srk@...com>, <s-vadapalli@...com>
Subject: Re: [PATCH net-next v2] net: ethernet: ti: am65-cpsw/cpts: Fix CPTS
 release action

Roger, Leon,

On 16/01/23 21:31, Roger Quadros wrote:
> Hi Siddharth,
> 
> On 16/01/2023 09:43, Siddharth Vadapalli wrote:
>>
>>
>> On 16/01/23 13:00, Leon Romanovsky wrote:
>>> On Mon, Jan 16, 2023 at 10:15:17AM +0530, Siddharth Vadapalli wrote:
>>>> The am65_cpts_release() function is registered as a devm_action in the
>>>> am65_cpts_create() function in am65-cpts driver. When the am65-cpsw driver
>>>> invokes am65_cpts_create(), am65_cpts_release() is added in the set of devm
>>>> actions associated with the am65-cpsw driver's device.
>>>>
>>>> In the event of probe failure or probe deferral, the platform_drv_probe()
>>>> function invokes dev_pm_domain_detach() which powers off the CPSW and the
>>>> CPSW's CPTS hardware, both of which share the same power domain. Since the
>>>> am65_cpts_disable() function invoked by the am65_cpts_release() function
>>>> attempts to reset the CPTS hardware by writing to its registers, the CPTS
>>>> hardware is assumed to be powered on at this point. However, the hardware
>>>> is powered off before the devm actions are executed.
>>>>
>>>> Fix this by getting rid of the devm action for am65_cpts_release() and
>>>> invoking it directly on the cleanup and exit paths.
>>>>
>>>> Fixes: f6bd59526ca5 ("net: ethernet: ti: introduce am654 common platform time sync driver")
>>>> Signed-off-by: Siddharth Vadapalli <s-vadapalli@...com>
>>>> Reviewed-by: Roger Quadros <rogerq@...nel.org>
>>>> ---
>>>> Changes from v1:
>>>> 1. Fix the build issue when "CONFIG_TI_K3_AM65_CPTS" is not set. This
>>>>    error was reported by kernel test robot <lkp@...el.com> at:
>>>>    https://lore.kernel.org/r/202301142105.lt733Lt3-lkp@intel.com/
>>>> 2. Collect Reviewed-by tag from Roger Quadros.
>>>>
>>>> v1:
>>>> https://lore.kernel.org/r/20230113104816.132815-1-s-vadapalli@ti.com/
>>>>
>>>>  drivers/net/ethernet/ti/am65-cpsw-nuss.c |  8 ++++++++
>>>>  drivers/net/ethernet/ti/am65-cpts.c      | 15 +++++----------
>>>>  drivers/net/ethernet/ti/am65-cpts.h      |  5 +++++
>>>>  3 files changed, 18 insertions(+), 10 deletions(-)
>>>>
>>>> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>>>> index 5cac98284184..00f25d8a026b 100644
>>>> --- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>>>> +++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>>>> @@ -1913,6 +1913,12 @@ static int am65_cpsw_am654_get_efuse_macid(struct device_node *of_node,
>>>>  	return 0;
>>>>  }
>>>>  
>>>> +static void am65_cpsw_cpts_cleanup(struct am65_cpsw_common *common)
>>>> +{
>>>> +	if (IS_ENABLED(CONFIG_TI_K3_AM65_CPTS) && common->cpts)
>>>
>>> Why do you have IS_ENABLED(CONFIG_TI_K3_AM65_CPTS), if
>>> am65_cpts_release() defined as empty when CONFIG_TI_K3_AM65_CPTS not set?
>>>
>>> How is it possible to have common->cpts == NULL?
>>
>> Thank you for reviewing the patch. I realize now that checking
>> CONFIG_TI_K3_AM65_CPTS is unnecessary.
>>
>> common->cpts remains NULL in the following cases:

I realized that the cases I mentioned are not explained clearly. Therefore, I
will mention the cases again, along with the section of code they correspond to,
in order to make it clear.

Case-1: am65_cpsw_init_cpts() returns 0 since CONFIG_TI_K3_AM65_CPTS is not
enabled. This corresponds to the following section within am65_cpsw_init_cpts():

if (!IS_ENABLED(CONFIG_TI_K3_AM65_CPTS))
	return 0;

In this case, common->cpts remains NULL, but it is not a problem even if the
am65_cpsw_nuss_probe() fails later, since the am65_cpts_release() function is
NOP. Thus, this case is not an issue.

Case-2: am65_cpsw_init_cpts() returns -ENOENT since the cpts node is not present
in the device tree. This corresponds to the following section within
am65_cpsw_init_cpts():

node = of_get_child_by_name(dev->of_node, "cpts");
if (!node) {
	dev_err(dev, "%s cpts not found\n", __func__);
	return -ENOENT;
}

In this case as well, common->cpts remains NULL, but it is not a problem because
the probe fails and the execution jumps to "err_of_clear", which doesn't invoke
am65_cpsw_cpts_cleanup(). Therefore, common->cpts being NULL is not a problem.

Case-3 and Case-4 are described later in this mail.

>> 1. am65_cpsw_init_cpts() returns 0 since CONFIG_TI_K3_AM65_CPTS is not enabled.
>> 2. am65_cpsw_init_cpts() returns -ENOENT since the cpts node is not defined.
>> 3. The call to am65_cpts_create() fails within the am65_cpsw_init_cpts()
>> function with a return value of 0 when cpts is disabled.
> 
> In this case common->cpts is not NULL and is set to error pointer.
> Probe will continue normally.
> Is it OK to call any of the cpts APIs with invalid handle?
> Also am65_cpts_release() will be called with invalid handle.

Yes Roger, thank you for pointing it out. When I wrote "cpts is disabled", I had
meant that the following section is executed within the am65_cpsw_init_cpts()
function:

Case-3:

cpts = am65_cpts_create(dev, reg_base, node);
if (IS_ERR(cpts)) {
	int ret = PTR_ERR(cpts);

	of_node_put(node);
	if (ret == -EOPNOTSUPP) {
		dev_info(dev, "cpts disabled\n");
		return 0;
	}

......
}

Leon,

In the above code, when the section corresponding to:
dev_info(dev, "cpts disabled\n");

is executed, CONFIG_TI_K3_AM65_CPTS is enabled. Therefore, the
am65_cpts_release() is not NOP. If the probe fails after the call to
am65_cpsw_init_cpts(), then the am65_cpsw_cpts_cleanup() function will be called
in the cleanup path of probe, which needs to check for common->cpts not being
NULL. This is because common->cpts is NULL after returning 0 from the
am65_cpsw_init_cpts() function at the
dev_info(dev, "cpts disabled\n");

section. Thus, I believe that in this case, am65_cpts_release() shouldn't be
invoked from the am65_cpsw_cpts_cleanup() function, since it would have already
been invoked from am65_cpts_create()'s cleanup path. This can be ensured by
checking whether common->cpts is NULL or not, before invoking
am65_cpts_release() within am65_cpsw_cpts_cleanup().

> 
>> 4. The call to am65_cpts_create() within the am65_cpsw_init_cpts() function
>> fails with an error.
> 
> In this case common->cpts is not NULL and will invoke am65_cpts_release() with
> invalid handle.

Case-4: The call to am65_cpts_create() within the am65_cpsw_init_cpts() function
fails with an error. This corresponds to the following section within
am65_cpsw_init_cpts():

cpts = am65_cpts_create(dev, reg_base, node);
if (IS_ERR(cpts)) {
......
	dev_err(dev, "cpts create err %d\n", ret);
	return ret;
}
	

Roger,

If the call to am65_cpts_create() fails with an error other than -EOPNOTSUPP,
which corresponds to Case-4, the call to am65_cpts_release() would have been
invoked within the am65_cpts_create()'s cleanup path itself if necessary. Also,
when the error is not -EOPNOTSUPP, the am65_cpsw_init_cpts() function returns an
error, due to which the execution jumps to "err_of_clear" in
am65_cpsw_nuss_probe(). Therefore, am65_cpsw_cpts_cleanup() is not invoked in
this case, due to which common->cpts being NULL is not a problem.


Roger, Leon, please review my comments and let me know. I think that Case-3
demands checking whether common->cpts is NULL or not, within the
am65_cpsw_cpts_cleanup() function.

Regards,
Siddharth.

Powered by blists - more mailing lists