[PATCH bpf] cacheinfo: move get_cpu_cacheinfo_id() back out

[Reinette Chatre]

Hi Everybody,

On 11/25/2021 7:59 AM, Alexei Starovoitov wrote:
> On Wed, Nov 24, 2021 at 1:14 AM Song Liu <song at kernel.org> wrote:
>>
>> On Tue, Nov 23, 2021 at 8:49 AM James Morse <james.morse at arm.com> wrote:
>>>
>>> Hello,
>>>
>>> On 23/11/2021 17:45, Song Liu wrote:
>>>> On Sat, Nov 20, 2021 at 6:55 AM Jakub Kicinski <kuba at kernel.org> wrote:
>>>>>
>>>>> This commit more or less reverts commit 709c4362725a ("cacheinfo:
>>>>> Move resctrl's get_cache_id() to the cacheinfo header file").
>>>>>
>>>>> There are no users of the static inline helper outside of resctrl/core.c
>>>>> and cpu.h is a pretty heavy include, it pulls in device.h etc. This
>>>>> trips up architectures like riscv which want to access cacheinfo
>>>>> in low level headers like elf.h.
>>>>>
>>>>> Link: https://lore.kernel.org/all/20211120035253.72074-1-kuba@kernel.org/
>>>>> Signed-off-by: Jakub Kicinski <kuba at kernel.org>
>>>>> ---
>>>
>>>>> x86 resctrl folks, does this look okay?
>>>>>
>>>>> I'd like to do some bpf header cleanups in -next which this is blocking.
>>>>> How would you like to handle that? This change looks entirely harmless,
>>>>> can I get an ack and take this via bpf/netdev to Linus ASAP so it
>>>>> propagates to all trees?
>>>>
>>>> Does this patch target the bpf tree, or the bpf-next tree? If we want to unblock
>>>> bpf header cleanup in -next, we can simply include it in a set for bpf-next.
>>>
>>>
>>> Some background: this is part of the mpam tree that wires up resctrl for arm64. This patch
>>> floated to the top and got merged with some cleanup as it was independent of the wider
>>> resctrl changes.
>>>
>>> If the cpu.h include is the problem, I can't see what that is needed for. It almost
>>> certainly came in with a lockdep annotation that got replaced by a comment instead.
>>
>> Thanks for the information.
>>
>> I can ack the patch for the patch itself.
>>
>> Acked-by: Song Liu <songliubraving at fb.com>
>>
>> But I am not sure whether we should ship it via bpf tree. It seems to
>> me that the
>> only reason we ship it via bpf tree is to get it to upstream ASAP?
>>
>> Alexei/Daniel/Andrii, what do you think about this?
> 
> I don't completely understand why it cannot go via -next along
> with other patches, but if Jakub needs it asap here is my
> Acked-by: Alexei Starovoitov <ast at kernel.org>
> and probably the fastest is for Jakub to take it via net tree directly.

These responses seems to ignore the information provided by James.

How about just the hunk below. It is not needed by the other parts 
removed and addresses the issue described in the changelog.

--- a/include/linux/cacheinfo.h
+++ b/include/linux/cacheinfo.h
@@ -3,7 +3,6 @@
   #define _LINUX_CACHEINFO_H

   #include <linux/bitops.h>
-#include <linux/cpu.h>
   #include <linux/cpumask.h>
   #include <linux/smp.h>

Reinette