Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH] hugetlb: Add nohugepages parameter to prevent hugepages
 creation
To:     "Guilherme G. Piccoli" <gpiccoli@canonical.com>, linux-mm@kvack.org
Cc:     linux-kernel@vger.kernel.org, jay.vosburgh@canonical.com,
        kernel@gpiccoli.net
References: <20191011223955.1435-1-gpiccoli@canonical.com>
From:   Mike Kravetz <mike.kravetz@oracle.com>
Message-ID: <c145a020-5ae0-e3a5-0251-199618cfaa9e@oracle.com>
Date:   Mon, 14 Oct 2019 11:25:31 -0700
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.1.0
MIME-Version: 1.0
In-Reply-To: <20191011223955.1435-1-gpiccoli@canonical.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 10/11/19 3:39 PM, Guilherme G. Piccoli wrote:
> Currently there are 2 ways for setting HugeTLB hugepages in kernel; either
> users pass parameters on kernel command-line or they can write to sysfs
> files (which is effectively the sysctl way).
> 
> Kdump kernels won't benefit from hugepages - in fact it's quite opposite,
> it may be the case hugepages on kdump kernel can lead to OOM if kernel
> gets unable to allocate demanded pages due to the fact the preallocated
> hugepages are consuming a lot of memory.
> 
> This patch proposes a new kernel parameter to prevent the creation of
> HugeTLB hugepages - we currently don't have a way to do that. We can
> even have kdump scripts removing the kernel command-line options to
> set hugepages, but it's not straightforward to prevent sysctl/sysfs
> configuration, given it happens in later boot or anytime when the
> system is running.
> 
> Signed-off-by: Guilherme G. Piccoli <gpiccoli@canonical.com>
> ---
> 
> About some decisions took in this patch:
> 
> * early_param() was used because I couldn't find a way to enforce
> parameters' ordering when using __setup(), and we need nohugepages
> processed before all other hugepages options.

I don't know much about early_param(), so I will assume this works as you
describe.  However, a quick grep shows hugepage options for ia64 also with
early_param.

> * The return when sysctl handler is prevented to progress due to
> nohugepages is -EINVAL, but could be changed; I've just followed
> present code there, but I'm OK changing that if we have suggestions.

It looks like you only have short circuited/prevented nr_hugepages via
sysfs/sysctl.  Theoretically, one could set nr_overcommit_hugepages and
still allocate hugetlb pages.  So, if you REALLY want to shut things down
you need to stop this as well.

There is already a macro hugepages_supported() that can be set by arch
specific code.  I wonder how difficult it would be to 'overwrite' the
macro if nohugepages is specified.  Perhaps just a level of naming
indirection.  This would use the existing code to prevent all hugetlb usage.

It seems like there may be some discussion about 'the right' way to
do kdump.  I can't add to that discussion, but if such an option as
nohugepages is needed, I can help.
-- 
Mike Kravetz