Re: [PATCH] mm: support __GFP_REPEAT in kvmalloc_node

[Vlastimil Babka <vbabka@suse.cz>]

On 01/04/2017 07:12 PM, Michal Hocko wrote:
> While checking opencoded users I've encountered that vhost code would
> really like to use kvmalloc with __GFP_REPEAT [1] so the following patch
> adds support for __GFP_REPEAT and converts both vhost users.
> 
> So currently I am sitting on 3 patches. I will wait for more feedback -
> especially about potential split ups or cleanups few more days and then
> repost the whole series.
> 
> [1] http://lkml.kernel.org/r/20170104150800.GO25453@dhcp22.suse.cz
> ---
> From 0b92e4d2e040524b878d4e7b9ee88fbad5284b33 Mon Sep 17 00:00:00 2001
> From: Michal Hocko <mhocko@suse.com>
> Date: Wed, 4 Jan 2017 18:01:39 +0100
> Subject: [PATCH] mm: support __GFP_REPEAT in kvmalloc_node
> 
> vhost code uses __GFP_REPEAT when allocating vhost_virtqueue resp.
> vhost_vsock because it would really like to prefer kmalloc to the
> vmalloc fallback - see 23cc5a991c7a ("vhost-net: extend device
> allocation to vmalloc") for more context. Michael Tsirkin has also
> noted:
> "
> __GFP_REPEAT overhead is during allocation time.  Using vmalloc means all
> accesses are slowed down.  Allocation is not on data path, accesses are.
> "
> 
> Let's teach kvmalloc_node to handle __GFP_REPEAT properly. There are two
> things to be careful about. First we should prevent from the OOM killer
> and so have to involve __GFP_NORETRY by default and secondly override
> __GFP_REPEAT for !costly order requests as the __GFP_REPEAT is ignored
> for !costly orders.
> 
> This patch shouldn't introduce any functional change.

Which is because the converted usages are always used for costly order,
right.

> 
> Cc: "Michael S. Tsirkin" <mst@redhat.com>
> Signed-off-by: Michal Hocko <mhocko@suse.com>
> ---
>  drivers/vhost/net.c   | 9 +++------
>  drivers/vhost/vsock.c | 9 +++------
>  mm/util.c             | 9 +++++++--
>  3 files changed, 13 insertions(+), 14 deletions(-)
> 
> diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
> index 5dc34653274a..105cd04c7414 100644
> --- a/drivers/vhost/net.c
> +++ b/drivers/vhost/net.c
> @@ -797,12 +797,9 @@ static int vhost_net_open(struct inode *inode, struct file *f)
>  	struct vhost_virtqueue **vqs;
>  	int i;
>  
> -	n = kmalloc(sizeof *n, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
> -	if (!n) {
> -		n = vmalloc(sizeof *n);
> -		if (!n)
> -			return -ENOMEM;
> -	}
> +	n = kvmalloc(sizeof *n, GFP_KERNEL | __GFP_REPEAT);
> +	if (!n)
> +		return -ENOMEM;
>  	vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL);
>  	if (!vqs) {
>  		kvfree(n);
> diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c
> index bbbf588540ed..7e0159867553 100644
> --- a/drivers/vhost/vsock.c
> +++ b/drivers/vhost/vsock.c
> @@ -455,12 +455,9 @@ static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
>  	/* This struct is large and allocation could fail, fall back to vmalloc
>  	 * if there is no other way.
>  	 */
> -	vsock = kzalloc(sizeof(*vsock), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
> -	if (!vsock) {
> -		vsock = vmalloc(sizeof(*vsock));
> -		if (!vsock)
> -			return -ENOMEM;
> -	}
> +	vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_REPEAT);
> +	if (!vsock)
> +		return -ENOMEM;
>  
>  	vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
>  	if (!vqs) {
> diff --git a/mm/util.c b/mm/util.c
> index 8e4ea6cbe379..a2bfb85e60e5 100644
> --- a/mm/util.c
> +++ b/mm/util.c
> @@ -348,8 +348,13 @@ void *kvmalloc_node(size_t size, gfp_t flags, int node)
>  	 * Make sure that larger requests are not too disruptive - no OOM
>  	 * killer and no allocation failure warnings as we have a fallback
>  	 */
> -	if (size > PAGE_SIZE)
> -		kmalloc_flags |= __GFP_NORETRY | __GFP_NOWARN;
> +	if (size > PAGE_SIZE) {
> +		kmalloc_flags |= __GFP_NOWARN;
> +
> +		if (!(kmalloc_flags & __GFP_REPEAT) ||
> +				(size <= PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
> +			kmalloc_flags |= __GFP_NORETRY;

I think this would be more understandable for me if it was written in
the opposite way, i.e. "if we have costly __GFP_REPEAT allocation, don't
use __GFP_NORETRY", but nevermind, seems correct to me wrt current
handling of both flags in the page allocator. And it serves as a good
argument to have this wrapper in mm/ as we are hopefully more likely to
keep it working as intended with future changes, than all the opencoded
variants.

Acked-by: Vlastimil Babka <vbabka@suse.cz>

> +	}
>  
>  	ret = kmalloc_node(size, kmalloc_flags, node);
>  
>