Re: [RFC PATCH 1/3] getcpu_cache system call: cache CPU number of running thread

[Mathieu Desnoyers]

----- On Jan 5, 2016, at 12:31 PM, Mathieu Desnoyers mathieu.desnoyers@xxxxxxxxxxxx wrote:

> ----- On Jan 5, 2016, at 7:04 AM, Will Deacon will.deacon@xxxxxxx wrote:
> 
>> Hi Mathieu,
>> 
>> On Tue, Jan 05, 2016 at 02:01:58AM -0500, Mathieu Desnoyers wrote:
>>> Expose a new system call allowing threads to register userspace memory
>>> areas where to store the CPU number on which the calling thread is
>>> running. Scheduler migration sets the TIF_NOTIFY_RESUME flag on the
>>> current thread. Upon return to user-space, a notify-resume handler
>>> updates the current CPU value within each registered user-space memory
>>> area. User-space can then read the current CPU number directly from
>>> memory.
>> 
>> What guarantees do you provide if a thread other than the one which
>> registered the cache tries to access the value? Obviously, there's a
>> potential data race here with the kernel issuing a parallel update, but
>> are you intending to have single-copy atomicity semantics (like relaxed
>> atomics in C11) or is this simply going to give you junk?
>> 
>> I ask because, in the absence of alignment checks on the cache pointer,
>> we can't guarantee single-copy atomicity on ARM when the kernel writes
>> the current CPU value.
> 
> Hi Will,
> 
> This is an excellent question. My initial thinking was that only the
> thread registering the cache would read it, but now that you ask,
> there might be use-cases where other threads would be interested in
> reading each other's current CPU number.
> 
> For instance, an application could create a linked list or hash map
> of thread control structures, which could contain the current CPU
> number of each thread. A dispatch thread could then traverse or
> lookup this structure to see on which CPU each thread is running and
> do work queue dispatch or scheduling decisions accordingly.
> 
> This use-case would imply ensuring that reading the current CPU value
> from another CPU will never result in reading a garbage value.
> 
> If we indeed intend to enable this use-case, we should:
> 
> 1) Add an alignment check on the cpu_cache pointer. Should we
>   return -EINVAL if unaligned ?
> 2) Document this alignment requirement in the man page, and the
>   atomicity guarantees it provides,

Related question: if we check that cpu_cache pointer is aligned
on 4 bytes, does put_user() then guarantee single-copy atomicity
on all architectures ?

Thanks,

Mathieu

> 
> The tiny downside of having this alignment requirement is that
> it would not be possible to put the cpu_cache into a packed
> structure. I don't think anyone would care though.
> 
> Thanks!
> 
> Mathieu
> 
> 
>> 
>> Cheers,
>> 
>> Will
> 
> --
> Mathieu Desnoyers
> EfficiOS Inc.
> http://www.efficios.com

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/