Received-SPF: pass (google.com: domain of linux-kernel+bounces-132173-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Thu, 4 Apr 2024 14:41:56 -0700
From: Krister Johansen <kjlx@templeofstupid.com>
To: Ali Saidi <alisaidi@amazon.com>
Cc: catalin.marinas@arm.com, james.morse@arm.com, kvmarm@lists.linux.dev,
	linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org,
	maz@kernel.org, me@davidreaver.com, oliver.upton@linux.dev,
	suzuki.poulose@arm.com, will@kernel.org, yuzenghui@huawei.com
Subject: Re: [PATCH] KVM: arm64: Limit stage2_apply_range() batch size to
 smallest block
Message-ID: <20240404214156.GF2178@templeofstupid.com>
References: <20240404044028.GA1976@templeofstupid.com>
 <20240404212742.11248-1-alisaidi@amazon.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240404212742.11248-1-alisaidi@amazon.com>

On Thu, Apr 04, 2024 at 09:27:42PM +0000, Ali Saidi wrote:
> > Just a quick followup that I did test Will's patches and didn't find
> > that it changed the performance of the workload that I'd been testing.
> > IOW, I wasn't able to discern a network performance difference between
> > the baseline and those changes.
> 
> That is a bit unexpected that the performance wasn't worse with the patch Will
> sent because it should have disabled the range invalidates since they these 
> invalidates will be getting rid of blocks?  Which Graviton were you testing
> this on? 

Sorry I didn't mention it earlier.  This was on a Graviton 4 with
FEAT_TLBIRANGE.  Due to the placement of the test machine and the client
the max single flow rate was 5Gbps, and with both the baseline and
Will's patches I wasn't able to discern any slowdown there, at least in
terms of impact to the adjacent networking traffic.  I saw an approx 1%
slowdown with multiple flows at 10Gbps, but didn't have a baseline for
that test case so was hesitant to offer it as conclusive.  I should be
clear that I _wasn't_ measuring the the teardown times in this test
-- just how the teardown itself impacted the networking performance.

-K