Messages in this thread | | | Date | Tue, 17 Jan 2023 15:18:46 +0530 | Subject | Re: [PATCH net-next v2] net: ethernet: ti: am65-cpsw/cpts: Fix CPTS release action | From | Siddharth Vadapalli <> |
| |
On 17/01/23 14:57, Roger Quadros wrote: > Siddharth, > > On 17/01/2023 07:00, Siddharth Vadapalli wrote: >> 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@ti.com> >>>>>> Reviewed-by: Roger Quadros <rogerq@kernel.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@intel.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(). >> > > Yes, I agree. > >>> >>>> 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. > > Correct. > >> >> >> 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. > > Do you really need a separate am65_cpsw_cpts_cleanup() or can just add > the NULL check in am65_cpts_release()?
Adding a NULL check in am65_cpts_release() works too. I will implement this in the v3 patch, if there are no objections to this approach. Leon, please let me know if this is acceptable.
Regards, Siddharth.
| |