linux-kernel - Re: [PATCH 1/1] vhost scsi: allocate vhost_scsi with GFP

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <5a03abaf-cbd6-242c-c965-b2da577da816@oracle.com>
Date:   Sat, 23 Jan 2021 00:01:51 -0800
From:   Dongli Zhang <dongli.zhang@...cle.com>
To:     Jason Wang <jasowang@...hat.com>,
        virtualization@...ts.linux-foundation.org, kvm@...r.kernel.org,
        netdev@...r.kernel.org
Cc:     linux-kernel@...r.kernel.org, mst@...hat.com, pbonzini@...hat.com,
        stefanha@...hat.com, joe.jin@...cle.com,
        aruna.ramakrishna@...cle.com
Subject: Re: [PATCH 1/1] vhost scsi: allocate vhost_scsi with GFP_NOWAIT to
 avoid delay



On 1/21/21 1:00 AM, Jason Wang wrote:
> 
> On 2021/1/21 13:03, Dongli Zhang wrote:
>> The size of 'struct vhost_scsi' is order-10 (~2.3MB). It may take long time
>> delay by kzalloc() to compact memory pages when there is a lack of
>> high-order pages. As a result, there is latency to create a VM (with
>> vhost-scsi) or to hotadd vhost-scsi-based storage.
>>
>> The prior commit 595cb754983d ("vhost/scsi: use vmalloc for order-10
>> allocation") prefers to fallback only when really needed, while this patch
>> changes allocation to GFP_NOWAIT in order to avoid the delay caused by
>> memory page compact.
>>
>> Cc: Aruna Ramakrishna <aruna.ramakrishna@...cle.com>
>> Cc: Joe Jin <joe.jin@...cle.com>
>> Signed-off-by: Dongli Zhang <dongli.zhang@...cle.com>
>> ---
>> Another option is to rework by reducing the size of 'struct vhost_scsi',
>> e.g., by replacing inline vhost_scsi.vqs with just memory pointers while
>> each vhost_scsi.vqs[i] should be allocated separately. Please let me
>> know if that option is better.
>>
>>   drivers/vhost/scsi.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/vhost/scsi.c b/drivers/vhost/scsi.c
>> index 4ce9f00ae10e..85eaa4e883f4 100644
>> --- a/drivers/vhost/scsi.c
>> +++ b/drivers/vhost/scsi.c
>> @@ -1814,7 +1814,7 @@ static int vhost_scsi_open(struct inode *inode, struct
>> file *f)
>>       struct vhost_virtqueue **vqs;
>>       int r = -ENOMEM, i;
>>   -    vs = kzalloc(sizeof(*vs), GFP_KERNEL | __GFP_NOWARN |
>> __GFP_RETRY_MAYFAIL);
>> +    vs = kzalloc(sizeof(*vs), GFP_NOWAIT | __GFP_NOWARN);
>>       if (!vs) {
>>           vs = vzalloc(sizeof(*vs));
>>           if (!vs)
> 
> 
> Can we use kvzalloc?
> 
Thank you very much for the suggestion.

To use 'GFP_NOWAIT' will avoid any direct compact in __alloc_pages_slowpath(),
while to use kvzalloc() will just avoid retrying direct compact for multiple times.

Although the latter will still do direct compact (without retry), I think it is
better than the former using GFP_NOWAIT.

I will send v2 with kvzalloc().

Thank you very much!

Dongli Zhang