Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <ca71fc1f-280e-19ff-ce0f-8befa29f45cd@gmail.com>
Date:   Wed, 28 Jun 2023 12:25:16 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:102.0)
 Gecko/20100101 Thunderbird/102.12.0
Subject: Re: [PATCH 1/4] perf/x86/intel: Get shared reg constraints first for
 vLBR
Content-Language: en-US
To:     Xiong Zhang <xiong.y.zhang@intel.com>,
        Sean Christopherson <seanjc@google.com>
Cc:     pbonzini@redhat.com, peterz@infradead.org,
        kan.liang@linux.intel.com, zhenyuw@linux.intel.com,
        zhiyuan.lv@intel.com, kvm list <kvm@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
References: <20230616113353.45202-1-xiong.y.zhang@intel.com>
 <20230616113353.45202-2-xiong.y.zhang@intel.com>
From:   Like Xu <like.xu.linux@gmail.com>
In-Reply-To: <20230616113353.45202-2-xiong.y.zhang@intel.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 16/6/2023 7:33 pm, Xiong Zhang wrote:
> When host has per cpu pinned LBR event and guest use LBR also, host
> couldn't get correct LBR data, as the physical LBR is preempted by
> guest.
> 
> The rule for multi events sharing LBR is defined in
> __intel_shared_reg_get_constraints(), but guest vLBR event skips this
> function, so even if host has per cpu pinned LBR event, guest vLBR event
> could get constraints successfully and make vlbr_exclude_host returns true,
> finally host couldn't enable LBR in intel_pmu_lbr_enable_all().

Although it goes against the "per cpu pinned LBR event" priority expectation,
the order is intentionally specified. For two reasons:

- vlbr uses the fake event mechanism in its implementation, a presence similar to
   BTS event, thus the question here is whether we can get the per cpu pinned BTS
   event to work as expected;

- this change should not be applied first before KVM has done a good job
   of making guest lbr and other lbr events coexist correctly;

In treating vlbr event as an ordinary perf_event behind a guest counter that
is expected to comply equally with the scheduling rules of host perf, the first
thing we need to address is how a guest counter should continue to function
during the time when the backend event is preempted by a higher priority one.

> 
> This commit move intel_vlbr_constraints() behind
> intel_shared_regs_constraints(), guest vLBR event will use LBR also and it
> should get LBR resource through intel_shared_regs_constraints().
> 
> Fixes: 097e4311cda9 ("perf/x86: Add constraint to create guest LBR event without hw counter")
> Signed-off-by: Xiong Zhang <xiong.y.zhang@intel.com>
> ---
>   arch/x86/events/intel/core.c | 6 +++---
>   1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
> index 6fd3cd97a6ac..2e27a69e9725 100644
> --- a/arch/x86/events/intel/core.c
> +++ b/arch/x86/events/intel/core.c
> @@ -3347,15 +3347,15 @@ __intel_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
>   {
>   	struct event_constraint *c;
>   
> -	c = intel_vlbr_constraints(event);
> +	c = intel_bts_constraints(event);
>   	if (c)
>   		return c;
>   
> -	c = intel_bts_constraints(event);
> +	c = intel_shared_regs_constraints(cpuc, event);
>   	if (c)
>   		return c;
>   
> -	c = intel_shared_regs_constraints(cpuc, event);
> +	c = intel_vlbr_constraints(event);
>   	if (c)
>   		return c;
>