On Fri, 10 Feb 2023 10:52:20 +0100, Magnus Karlsson <magnus.karlsson@xxxxxxxxx> wrote: > On Fri, 10 Feb 2023 at 03:14, Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote: > > > > When we try to start AF_XDP on some machines with long running time, due > > to the machine's memory fragmentation problem, there is no sufficient > > continuous physical memory that will cause the start failure. > > > > After AF_XDP fails to apply for continuous physical memory, this patch > > tries to use vmalloc() to allocate memory to solve this problem. > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> > > Reported-by: kernel test robot <lkp@xxxxxxxxx> > > Link: https://lore.kernel.org/oe-kbuild-all/202302091850.0HBmsDAq-lkp@xxxxxxxxx > > --- > > net/xdp/xsk.c | 8 +++++--- > > net/xdp/xsk_queue.c | 21 +++++++++++++++------ > > net/xdp/xsk_queue.h | 1 + > > 3 files changed, 21 insertions(+), 9 deletions(-) > > > > diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c > > index 9f0561b67c12..33db57548ee3 100644 > > --- a/net/xdp/xsk.c > > +++ b/net/xdp/xsk.c > > @@ -1296,7 +1296,6 @@ static int xsk_mmap(struct file *file, struct socket *sock, > > struct xdp_sock *xs = xdp_sk(sock->sk); > > struct xsk_queue *q = NULL; > > unsigned long pfn; > > - struct page *qpg; > > > > if (READ_ONCE(xs->state) != XSK_READY) > > return -EBUSY; > > @@ -1319,10 +1318,13 @@ static int xsk_mmap(struct file *file, struct socket *sock, > > > > /* Matches the smp_wmb() in xsk_init_queue */ > > smp_rmb(); > > - qpg = virt_to_head_page(q->ring); > > - if (size > page_size(qpg)) > > + > > + if (PAGE_ALIGN(q->ring_size) < size) > > return -EINVAL; > > > > + if (is_vmalloc_addr(q->ring)) > > + return remap_vmalloc_range(vma, q->ring, 0); > > + > > pfn = virt_to_phys(q->ring) >> PAGE_SHIFT; > > return remap_pfn_range(vma, vma->vm_start, pfn, > > size, vma->vm_page_prot); > > diff --git a/net/xdp/xsk_queue.c b/net/xdp/xsk_queue.c > > index 6cf9586e5027..7b03102d1672 100644 > > --- a/net/xdp/xsk_queue.c > > +++ b/net/xdp/xsk_queue.c > > @@ -7,6 +7,7 @@ > > #include <linux/slab.h> > > #include <linux/overflow.h> > > #include <net/xdp_sock_drv.h> > > +#include <linux/vmalloc.h> > > > > #include "xsk_queue.h" > > > > @@ -37,14 +38,18 @@ struct xsk_queue *xskq_create(u32 nentries, bool umem_queue) > > __GFP_COMP | __GFP_NORETRY; > > size = xskq_get_ring_size(q, umem_queue); > > > > + q->ring_size = size; > > q->ring = (struct xdp_ring *)__get_free_pages(gfp_flags, > > get_order(size)); > > - if (!q->ring) { > > - kfree(q); > > - return NULL; > > - } > > + if (q->ring) > > + return q; > > + > > + q->ring = (struct xdp_ring *)vmalloc_user(size); > > + if (q->ring) > > + return q; > > Thanks for bringing this to attention. Interesting to see how hard it > gets after a while to find consecutive memory since this is not a > large area. If the size of the queue is 8 * 1024, then the size of the desc[] is 8 * 1024 * 8 = 16 * PAGE, but we also add struct xdp_ring size, so it is 16page+. This is necessary to apply for a 4-order memory. If there are a lot of queues, it is difficult. Here, that we actually waste 15 pages. 4-Order memory is 32 pages, but we only use 17 pages. > > I am wondering if it would be better to remove the __get_free_pages() > and just go for vmalloc_user. There is no particular reason here for > allocating consecutive physical pages for the ring. Does anyone see > any problem with removing this? If not, please just remove > __get_free_pages(), test it, and post a v2. I agree. Thanks. > > > - return q; > > + kfree(q); > > + return NULL; > > } > > > > void xskq_destroy(struct xsk_queue *q) > > @@ -52,6 +57,10 @@ void xskq_destroy(struct xsk_queue *q) > > if (!q) > > return; > > > > - page_frag_free(q->ring); > > + if (is_vmalloc_addr(q->ring)) > > + vfree(q->ring); > > + else > > + page_frag_free(q->ring); > > + > > kfree(q); > > } > > diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h > > index c6fb6b763658..35922b8b92a8 100644 > > --- a/net/xdp/xsk_queue.h > > +++ b/net/xdp/xsk_queue.h > > @@ -45,6 +45,7 @@ struct xsk_queue { > > struct xdp_ring *ring; > > u64 invalid_descs; > > u64 queue_empty_descs; > > + size_t ring_size; > > }; > > > > /* The structure of the shared state of the rings are a simple > > -- > > 2.32.0.3.g01195cf9f > >
