lkml.org 
[lkml]   [2024]   [Apr]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH v1 24/31] x86/resctrl: Move get_config_index() to a header
From
Hi Dave,

On 4/11/2024 7:25 AM, Dave Martin wrote:
> On Mon, Apr 08, 2024 at 08:25:26PM -0700, Reinette Chatre wrote:
>> Hi James,
>>
>> On 3/21/2024 9:50 AM, James Morse wrote:
>>> get_config_index() is used by the architecture specific code to map a
>>> CLOSID+type pair to an index in the configuration arrays.
>>>
>>> MPAM needs to do this too to preserve the ABI to user-space, there is
>>> no reason to do it differently.
>>>
>>> Move the helper to a header file.
>>>
>>> Signed-off-by: James Morse <james.morse@arm.com>
>>> ---
>>> arch/x86/kernel/cpu/resctrl/ctrlmondata.c | 19 +++----------------
>>> include/linux/resctrl.h | 15 +++++++++++++++
>>> 2 files changed, 18 insertions(+), 16 deletions(-)
>
> [...]
>
>>> diff --git a/include/linux/resctrl.h b/include/linux/resctrl.h
>>> index 3de5bc63ace0..73c111963433 100644
>>> --- a/include/linux/resctrl.h
>>> +++ b/include/linux/resctrl.h
>>> @@ -258,6 +258,21 @@ bool resctrl_arch_is_evt_configurable(enum resctrl_event_id evt);
>>> void resctrl_arch_mon_event_config_write(void *info);
>>> void resctrl_arch_mon_event_config_read(void *info);
>>>
>>> +/* For use by arch code to remap resctrl's smaller CDP CLOSID range */
>>> +static inline u32 resctrl_get_config_index(u32 closid,
>>> + enum resctrl_conf_type type)
>>> +{
>>> + switch (type) {
>>> + default:
>>> + case CDP_NONE:
>>> + return closid;
>>> + case CDP_CODE:
>>> + return (closid * 2) + 1;
>>> + case CDP_DATA:
>>> + return (closid * 2);
>>> + }
>>> +}
>>
>> (please check the tabs)
>
> Noted. I also see that redundant parentheses seem spuriously added
> compared with the original version of this moved code. I can make a
> note to drop them if you prefer.
>
>> This change is unexpected to me. Could you please elaborate how
>> MPAM's variant of CDP works?
>>
>> Thank you very much.
>>
>> Reinette
>
> Note: I haven't discussed this specifically with James, so the following
> is my best guess at the rationale... With that in mind:
>
> For MPAM, CDP isn't a special mode; instead, the PARTIDs for
> instructions and data are always configured independently in the CPU.
> If resctrl is not configured for CDP, we simply program the same PARTID
> value both for instructions and data on task switch.
>
> For a given resctrl control group we could pick two random unrelated
> PARTIDs, but there seems to be no advantage in doing that since resctrl
> enables cdp globally or not, and we would require more effort to
> translate resctrl closids to PARTIDs if we didn't pair the IDs up
> systematically.
>
> (See [1], [2] in James' snapshot, which illustrate how he proposes
> to do it.)
>
>
> So, we may as well stick with the same scheme already established for
> x86: nothing forces us to do that, but it looks simpler than the
> alternatives. I think that's the idea, anyway.
>
> Then, if the same scheme is used by multiple arches (and 100% of the
> arches currently known to resctrl), it probably makes sense to share the
> definition of the mapping at least as a default for arches that don't
> have their own different ways of doing it.
>
> Does this make sense?

It does, thank you very much.

>
> I can recommend adding some of this rationale to the commit message
> if it helps (and assuming I'm right!)

Sounds good, thank you.

Reinette

\
 
 \ /
  Last update: 2024-04-11 20:25    [W:0.119 / U:0.196 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site