Re: [PATCH net-next v2 1/3] devlink: introduce framework for selftests

[Jiri Pirko]

Wed, Jul 13, 2022 at 12:16:03PM CEST, vikas.gupta@xxxxxxxxxxxx wrote:
>Hi Jiri,
>
>On Wed, Jul 13, 2022 at 12:58 PM Jiri Pirko <jiri@xxxxxxxxxx> wrote:
>>
>> Wed, Jul 13, 2022 at 08:40:50AM CEST, vikas.gupta@xxxxxxxxxxxx wrote:
>> >Hi Jiri,
>> >
>> >On Tue, Jul 12, 2022 at 11:38 PM Jiri Pirko <jiri@xxxxxxxxxxx> wrote:
>> >>
>> >> Tue, Jul 12, 2022 at 06:41:49PM CEST, vikas.gupta@xxxxxxxxxxxx wrote:
>> >> >Hi Jiri,
>> >> >
>> >> >On Tue, Jul 12, 2022 at 11:58 AM Jiri Pirko <jiri@xxxxxxxxxx> wrote:
>> >> >>
>> >> >> Tue, Jul 12, 2022 at 08:16:11AM CEST, vikas.gupta@xxxxxxxxxxxx wrote:
>> >> >> >Hi Jiri,
>> >> >> >
>> >> >> >On Mon, Jul 11, 2022 at 6:10 PM Jiri Pirko <jiri@xxxxxxxxxx> wrote:
>> >> >> >
>> >> >> >> Thu, Jul 07, 2022 at 08:29:48PM CEST, vikas.gupta@xxxxxxxxxxxx wrote:
>> >>
>> >> [...]
>> >>
>> >>
>> >> >> >> >  * enum devlink_trap_action - Packet trap action.
>> >> >> >> >  * @DEVLINK_TRAP_ACTION_DROP: Packet is dropped by the device and a copy
>> >> >> >> is not
>> >> >> >> >@@ -576,6 +598,10 @@ enum devlink_attr {
>> >> >> >> >       DEVLINK_ATTR_LINECARD_TYPE,             /* string */
>> >> >> >> >       DEVLINK_ATTR_LINECARD_SUPPORTED_TYPES,  /* nested */
>> >> >> >> >
>> >> >> >> >+      DEVLINK_ATTR_SELFTESTS_MASK,            /* u32 */
>> >> >> >>
>> >> >> >> I don't see why this is u32 bitset. Just have one attr per test
>> >> >> >> (NLA_FLAG) in a nested attr instead.
>> >> >> >>
>> >> >> >
>> >> >> >As per your suggestion, for an example it should be like as below
>> >> >> >
>> >> >> >        DEVLINK_ATTR_SELFTESTS,                 /* nested */
>> >> >> >
>> >> >> >        DEVLINK_ATTR_SELFTESTS_SOMETEST1            /* flag */
>> >> >> >
>> >> >> >        DEVLINK_ATTR_SELFTESTS_SOMETEST2           /* flag */
>> >> >>
>> >> >> Yeah, but have the flags in separate enum, no need to pullute the
>> >> >> devlink_attr enum by them.
>> >> >>
>> >> >>
>> >> >> >
>> >> >> >....    <SOME MORE TESTS>
>> >> >> >
>> >> >> >.....
>> >> >> >
>> >> >> >        DEVLINK_ATTR_SLEFTESTS_RESULT_VAL,      /* u8 */
>> >> >> >
>> >> >> >
>> >> >> >
>> >> >> > If we have this way then we need to have a mapping (probably a function)
>> >> >> >for drivers to tell them what tests need to be executed based on the flags
>> >> >> >that are set.
>> >> >> > Does this look OK?
>> >> >> >  The rationale behind choosing a mask is that we could directly pass the
>> >> >> >mask-value to the drivers.
>> >> >>
>> >> >> If you have separate enum, you can use the attrs as bits internally in
>> >> >> kernel. Add a helper that would help the driver to work with it.
>> >> >> Pass a struct containing u32 (or u8) not to drivers. Once there are more
>> >> >> tests than that, this structure can be easily extended and the helpers
>> >> >> changed. This would make this scalable. No need for UAPI change or even
>> >> >> internel driver api change.
>> >> >
>> >> >As per your suggestion, selftest attributes can be declared in separate
>> >> >enum as below
>> >> >
>> >> >enum {
>> >> >
>> >> >        DEVLINK_SELFTEST_SOMETEST,         /* flag */
>> >> >
>> >> >        DEVLINK_SELFTEST_SOMETEST1,
>> >> >
>> >> >        DEVLINK_SELFTEST_SOMETEST2,
>> >> >
>> >> >....
>> >> >
>> >> >......
>> >> >
>> >> >        __DEVLINK_SELFTEST_MAX,
>> >> >
>> >> >        DEVLINK_SELFTEST_MAX = __DEVLINK_SELFTEST_MAX - 1
>> >> >
>> >> >};
>> >> >Below  examples could be the flow of parameters/data from user to
>> >> >kernel and vice-versa
>> >> >
>> >> >
>> >> >Kernel to user for show command . Users can know what all tests are
>> >> >supported by the driver. A return from kernel to user.
>> >> >______
>> >> >|NEST |
>> >> >|_____ |TEST1|TEST4|TEST7|...
>> >> >
>> >> >
>> >> >User to kernel to execute test: If user wants to execute test4, test8, test1...
>> >> >______
>> >> >|NEST |
>> >> >|_____ |TEST4|TEST8|TEST1|...
>> >> >
>> >> >
>> >> >Result Kernel to user execute test RES(u8)
>> >> >______
>> >> >|NEST |
>> >> >|_____ |RES4|RES8|RES1|...
>> >>
>> >> Hmm, I think it is not good idea to rely on the order, a netlink library
>> >> can perhaps reorder it? Not sure here.
>> >>
>> >> >
>> >> >Results are populated in the same order as the user passed the TESTs
>> >> >flags. Does the above result format from kernel to user look OK ?
>> >> >Else we need to have below way to form a result format, a nest should
>> >> >be made for <test_flag,
>> >> >result> but since test flags are in different enum other than
>> >> >devlink_attr and RES being part of devlink_attr, I believe it's not
>> >> >good practice to make the below structure.
>> >>
>> >> Not a structure, no. Have it as another nest (could be the same attr as
>> >> the parent nest:
>> >>
>> >> ______
>> >> |NEST |
>> >> |_____ |NEST|       |NEST|       |NEST|
>> >>         TEST4,RES4   TEST8,RES8   TEST1, RES1
>> >>
>> >> also, it is flexible to add another attr if needed (like maybe result
>> >> message string containing error message? IDK).
>> >
>> >For above nesting we can have the attributes defined as below
>> >
>> >Attribute in  devlink_attr
>> >enum devlink_attr {
>> >  ....
>> >  ....
>> >        DEVLINK_SELFTESTS_INFO, /* nested */
>> >  ...
>> >...
>> >}
>> >
>> >enum devlink_selftests {
>> >        DEVLINK_SELFTESTS_SOMETEST0,   /* flag */
>> >        DEVLINK_SELFTESTS_SOMETEST1,
>> >        DEVLINK_SELFTESTS_SOMETEST2,
>> >        ...
>> >        ...
>> >}
>> >
>> >enum devlink_selftest_result {
>>
>> for attrs, have "attr" in the name of the enum and "ATTR" in name of the
>> value.
>>
>> >        DEVLINK_SELFTESTS_RESULT,       /* nested */
>> >        DEVLINK_SELFTESTS_TESTNUM,      /* u32  indicating the test
>>
>> You can have 1 enum, containing both these and the test flags from
>> above.
> I think it's better to keep enum devlink_selftests_attr (containing
>flags) and devlink_selftest_result_attr separately as it will have an
>advantage.
> For example, for show commands the kernel can iterate through and
>check with the driver if it supports a particular test.
>
>    for (i = 0; i < DEVLINK_SELFTEST_ATTR_MAX, i++) {
>                   if (devlink->ops->selftest_info(devlink, i,
>extack)) {  // supports selftest or not
>                         nla_put_flag(msg, i);
>                }
>        }
>      Also flags in devlink_selftests_attr can be used as bitwise, if required.
>      Let me know what you think.

Okay.


>
>Thanks,
>Vikas
>
>>
>>
>> >number in devlink_selftests enum */
>> >        DEVLINK_SELFTESTS_RESULT_VAL,   /* u8  skip, pass, fail.. */
>>
>> Put enum name in the comment, instead of list possible values.
>>
>>
>> >        ...some future attrr...
>> >
>> >}
>> >enums in devlink_selftest_result can be put in devlink_attr though.
>>
>> You can have them separate, I think it is about the time we try to put
>> new attrs what does not have potencial to be re-used to a separate enum.
>>
>>
>> >
>> >Does this look OK?
>> >
>> >Thanks,
>> >Vikas
>> >
>> >>
>> >>
>> >>
>> >> >______
>> >> >|NEST |
>> >> >|_____ | TEST4, RES4|TEST8,RES8|TEST1,RES1|...
>> >> >
>> >> >Let me know if my understanding is correct.
>> >>
>> >> [...]
>>
>>