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: <de39c555-e092-8e57-43a4-7a2c56d1e66c@redhat.com>
Date:   Thu, 13 Jul 2023 17:43:22 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.12.0
Subject: Re: [PATCH 3/3] dax/kmem: Always enroll hotplugged memory for
 memmap_on_memory
Content-Language: en-US
To:     "Verma, Vishal L" <vishal.l.verma@intel.com>,
        "akpm@linux-foundation.org" <akpm@linux-foundation.org>,
        "rafael@kernel.org" <rafael@kernel.org>,
        "osalvador@suse.de" <osalvador@suse.de>,
        "aneesh.kumar@linux.ibm.com" <aneesh.kumar@linux.ibm.com>,
        "Williams, Dan J" <dan.j.williams@intel.com>,
        "Jiang, Dave" <dave.jiang@intel.com>,
        "lenb@kernel.org" <lenb@kernel.org>
Cc:     "Huang, Ying" <ying.huang@intel.com>,
        "linux-mm@kvack.org" <linux-mm@kvack.org>,
        "linux-cxl@vger.kernel.org" <linux-cxl@vger.kernel.org>,
        "nvdimm@lists.linux.dev" <nvdimm@lists.linux.dev>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "linux-acpi@vger.kernel.org" <linux-acpi@vger.kernel.org>,
        "dave.hansen@linux.intel.com" <dave.hansen@linux.intel.com>
References: <20230613-vv-kmem_memmap-v1-0-f6de9c6af2c6@intel.com>
 <20230613-vv-kmem_memmap-v1-3-f6de9c6af2c6@intel.com>
 <aadbedeb-424d-a146-392d-d56680263691@redhat.com>
 <87edleplkn.fsf@linux.ibm.com>
 <1df12885-9ae4-6aef-1a31-91ecd5a18d24@redhat.com>
 <5a8e9b1b6c8d6d9e5405ca35abb9be3ed09761c3.camel@intel.com>
 <ee0c84ff-6d97-3b7c-88a8-dd00797c2999@redhat.com>
 <6cb5624ebf3039b18f5180262fc6383b402d26ea.camel@intel.com>
 <80c8654e-21fb-b187-3475-9a1410a53a4e@redhat.com>
 <b62572356db07ae0c3305ed03916b0ff40b14426.camel@intel.com>
From:   David Hildenbrand <david@redhat.com>
Organization: Red Hat
In-Reply-To: <b62572356db07ae0c3305ed03916b0ff40b14426.camel@intel.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 13.07.23 17:40, Verma, Vishal L wrote:
> On Thu, 2023-07-13 at 17:23 +0200, David Hildenbrand wrote:
>> On 13.07.23 17:15, Verma, Vishal L wrote:
>>> On Thu, 2023-07-13 at 09:23 +0200, David Hildenbrand wrote:
>>>> On 13.07.23 08:45, Verma, Vishal L wrote:
>>>>>
>>>>> I'm taking a shot at implementing the splitting internally in
>>>>> memory_hotplug.c. The caller (kmem) side does become trivial with this
>>>>> approach, but there's a slight complication if I don't have the module
>>>>> param override (patch 1 of this series).
>>>>>
>>>>> The kmem diff now looks like:
>>>>>
>>>>>       diff --git a/drivers/dax/kmem.c b/drivers/dax/kmem.c
>>>>>       index 898ca9505754..8be932f63f90 100644
>>>>>       --- a/drivers/dax/kmem.c
>>>>>       +++ b/drivers/dax/kmem.c
>>>>>       @@ -105,6 +105,8 @@ static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
>>>>>               data->mgid = rc;
>>>>>        
>>>>>               for (i = 0; i < dev_dax->nr_range; i++) {
>>>>>       +               mhp_t mhp_flags = MHP_NID_IS_MGID | MHP_MEMMAP_ON_MEMORY |
>>>>>       +                                 MHP_SPLIT_MEMBLOCKS;
>>>>>                       struct resource *res;
>>>>>                       struct range range;
>>>>>        
>>>>>       @@ -141,7 +143,7 @@ static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
>>>>>                        * this as RAM automatically.
>>>>>                        */
>>>>>                       rc = add_memory_driver_managed(data->mgid, range.start,
>>>>>       -                               range_len(&range), kmem_name, MHP_NID_IS_MGID);
>>>>>       +                               range_len(&range), kmem_name, mhp_flags);
>>>>>        
>>>>>                       if (rc) {
>>>>>                               dev_warn(dev, "mapping%d: %#llx-%#llx memory add failed\n",
>>>>>       
>>>>>
>>>>
>>>> Why do we need the MHP_SPLIT_MEMBLOCKS?
>>>
>>> I thought we still wanted either an opt-in or opt-out for the kmem
>>> driver to be able to do memmap_on_memory, in case there were
>>> performance implications or the lack of 1GiB PUDs. I haven't
>>> implemented that yet, but I was thinking along the lines of a sysfs
>>> knob exposed by kmem, that controls setting of this new
>>> MHP_SPLIT_MEMBLOCKS flag.
>>
>> Why is MHP_MEMMAP_ON_MEMORY not sufficient for that?
>>
>>
> Ah I see what you mean now - knob just controls MHP_MEMMAP_ON_MEMORY,
> and memory_hotplug is free to split to memblocks if it needs to to
> satisfy that.

And if you don't want memmap holes in a larger area you're adding (for 
example to runtime-allocate 1 GiB pages), simply check the size your 
adding, and if it's, say, less than 1 G, don't set the flag.

But that's probably a corner case use case not worth considering for now.

> 
> That sounds reasonable. Let me give this a try and see if I run into
> anything else. Thanks David!

Sure!

-- 
Cheers,

David / dhildenb