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/[email protected] > --- > From 0b92e4d2e040524b878d4e7b9ee88fbad5284b33 Mon Sep 17 00:00:00 2001 > From: Michal Hocko <[email protected]> > 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" <[email protected]> > Signed-off-by: Michal Hocko <[email protected]> > --- > 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 <[email protected]> > + } > > ret = kmalloc_node(size, kmalloc_flags, node); > >
