Date:   Mon, 6 Aug 2018 12:50:50 -0700
From:   Reinette Chatre <reinette.chatre@...el.com>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     Dave Hansen <dave.hansen@...el.com>, tglx@...utronix.de,
        mingo@...hat.com, fenghua.yu@...el.com, tony.luck@...el.com,
        vikas.shivappa@...ux.intel.com, gavin.hindman@...el.com,
        jithu.joseph@...el.com, hpa@...or.com, x86@...nel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 0/2] x86/intel_rdt and perf/x86: Fix lack of coordination
 with perf

Hi Peter,

On 8/3/2018 11:37 AM, Reinette Chatre wrote:
> On 8/3/2018 8:25 AM, Peter Zijlstra wrote:
>> On Fri, Aug 03, 2018 at 08:18:09AM -0700, Reinette Chatre wrote:
>>> You state that you understand what we are trying to do and I hope that I
>>> convinced you that we are not able to accomplish the same by following
>>> your guidance.
>>
>> No, I said I understood your pmc reserve patch and its implications.
>>
>> I have no clue what you're trying to do with resctl, nor why you think
>> this is not feasible with perf. And if it really is not feasible, you'll
>> have to live without it.

In my previous email I provided the details of the Cache Pseudo-Locking
feature implemented on top of resctrl. Please let me know if you would
like any more details about that. I can send you more materials.

In my previous message I also provided the thoughts on why I believe
same is not feasible with perf as commented below ...

> Looking at if we were to build on top of the kernel perf event API
> (perf_event_create_kernel_counter(), perf_event_enable(),
> perf_event_disable(), ...). Just looking at perf_event_enable() -
> ideally this would be as lean as possible to only enable the event and
> not result in itself contributing the the measurement. First, the
> enabling of the event is not as lean to fulfill this requirement since
> it executes more code after the event was actually enabled. Also, the
> code relies on a mutex so we cannot use it with interrupts disabled.

I proceeded to modify the implemented debugfs measurements to build on
top of the perf APIs mentioned above. As anticipated the events could
not be enabled in interrupt context. I get a clear message in this regard:

BUG: sleeping function called from invalid context at
kernel/locking/mutex.c:748

I thus continued to use the API with interrupts enabled did the following:

Two new event attributes:
static struct perf_event_attr l2_miss_attr = {
        .type           = PERF_TYPE_RAW,
        .config         = (0x10ULL << 8) | 0xd1,
        .size           = sizeof(struct perf_event_attr),
        .pinned         = 1,
        .disabled       = 1,
        .exclude_user   = 1
};

static struct perf_event_attr l2_hit_attr = {
        .type           = PERF_TYPE_RAW,
        .config         = (0x2ULL << 8) | 0xd1,
        .size           = sizeof(struct perf_event_attr),
        .pinned         = 1,
        .disabled       = 1,
        .exclude_user   = 1
};

Create the two new events using these attributes:
l2_miss_event = perf_event_create_kernel_counter(&l2_miss_attr, cpu,
NULL, NULL, NULL);
l2_hit_event = perf_event_create_kernel_counter(&l2_hit_attr, cpu, NULL,
NULL, NULL);

Take measurements:
perf_event_enable(l2_miss_event);
perf_event_enable(l2_hit_event);
local_irq_disable();
/* Disable hardware prefetchers */
/* Loop through pseudo-locked memory */
/* Enable hardware prefetchers */
local_irq_enable();
perf_event_disable(l2_hit_event);
perf_event_disable(l2_miss_event);

Read results:
l2_hits = perf_event_read_value(l2_hit_event, &enabled, &running);
l2_miss = perf_event_read_value(l2_miss_event, &enabled, &running);
/* Make results available in tracepoints */


With the above implementation and a 256KB pseudo-locked memory region I
obtain the following results:
pseudo_lock_mea-755   [002] ....   396.946953: pseudo_lock_l2: hits=4140
miss=5
pseudo_lock_mea-762   [002] ....   397.998864: pseudo_lock_l2: hits=4138
miss=8
pseudo_lock_mea-765   [002] ....   399.041868: pseudo_lock_l2: hits=4142
miss=5
pseudo_lock_mea-768   [002] ....   400.086871: pseudo_lock_l2: hits=4141
miss=7
pseudo_lock_mea-771   [002] ....   401.132921: pseudo_lock_l2: hits=4138
miss=10
pseudo_lock_mea-774   [002] ....   402.216700: pseudo_lock_l2: hits=4238
miss=46
pseudo_lock_mea-777   [002] ....   403.312148: pseudo_lock_l2: hits=4142
miss=5
pseudo_lock_mea-780   [002] ....   404.381674: pseudo_lock_l2: hits=4139
miss=8
pseudo_lock_mea-783   [002] ....   405.422820: pseudo_lock_l2: hits=4472
miss=79
pseudo_lock_mea-786   [002] ....   406.495065: pseudo_lock_l2: hits=4140
miss=8
pseudo_lock_mea-793   [002] ....   407.561383: pseudo_lock_l2: hits=4143
miss=4

The above results are not accurate since it does not reflect the success
of the pseudo-locked region. Expected results are as we can currently
obtain (copying results from previous email):
pseudo_lock_mea-26090 [002] .... 61838.488027: pseudo_lock_l2: hits=4096
miss=0
pseudo_lock_mea-26097 [002] .... 61843.689381: pseudo_lock_l2: hits=4096
miss=0
pseudo_lock_mea-26100 [002] .... 61848.751411: pseudo_lock_l2: hits=4096
miss=0
pseudo_lock_mea-26108 [002] .... 61853.820361: pseudo_lock_l2: hits=4096
miss=0
pseudo_lock_mea-26111 [002] .... 61858.880364: pseudo_lock_l2: hits=4096
miss=0
pseudo_lock_mea-26118 [002] .... 61863.937343: pseudo_lock_l2: hits=4096
miss=0
pseudo_lock_mea-26121 [002] .... 61869.008341: pseudo_lock_l2: hits=4096
miss=0

Could you please guide me on how you would prefer us to use perf in
order to obtain the same accurate results we can now?

Thank you very much

Reinette

Powered by blists - more mailing lists