[PATCH net-next v4] xsk: support use vaddr as ring

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
contiguous physical memory that will cause the start failure.

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 to these machine with long running time.

Here, that we actually waste 15 pages. 4-Order memory is 32 pages, but
we only use 17 pages.

This patch replaces __get_free_pages() by vmalloc() to allocate memory
to solve these problems.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
---
 net/xdp/xsk.c       |  9 ++-------
 net/xdp/xsk_queue.c | 11 +++++------
 net/xdp/xsk_queue.h |  1 +
 3 files changed, 8 insertions(+), 13 deletions(-)

diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 9f0561b67c12..0a047a09a10f 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c

@@ -1295,8 +1295,6 @@ static int xsk_mmap(struct file *file, struct socket *sock,
 	unsigned long size = vma->vm_end - vma->vm_start;
 	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;

@@ -1319,13 +1317,10 @@ 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 (size > q->ring_vmalloc_size)
 		return -EINVAL;
 
-	pfn = virt_to_phys(q->ring) >> PAGE_SHIFT;
-	return remap_pfn_range(vma, vma->vm_start, pfn,
-			       size, vma->vm_page_prot);
+	return remap_vmalloc_range(vma, q->ring, 0);
 }
 
 static int xsk_notifier(struct notifier_block *this,

diff --git a/net/xdp/xsk_queue.c b/net/xdp/xsk_queue.c
index 6cf9586e5027..f8905400ee07 100644
--- a/net/xdp/xsk_queue.c
+++ b/net/xdp/xsk_queue.c

@@ -6,6 +6,7 @@
 #include <linux/log2.h>
 #include <linux/slab.h>
 #include <linux/overflow.h>
+#include <linux/vmalloc.h>
 #include <net/xdp_sock_drv.h>
 
 #include "xsk_queue.h"

@@ -23,7 +24,6 @@ static size_t xskq_get_ring_size(struct xsk_queue *q, bool umem_queue)
 struct xsk_queue *xskq_create(u32 nentries, bool umem_queue)
 {
 	struct xsk_queue *q;
-	gfp_t gfp_flags;
 	size_t size;
 
 	q = kzalloc(sizeof(*q), GFP_KERNEL);

@@ -33,17 +33,16 @@ struct xsk_queue *xskq_create(u32 nentries, bool umem_queue)
 	q->nentries = nentries;
 	q->ring_mask = nentries - 1;
 
-	gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN |
-		    __GFP_COMP  | __GFP_NORETRY;
 	size = xskq_get_ring_size(q, umem_queue);
+	size = PAGE_ALIGN(size);
 
-	q->ring = (struct xdp_ring *)__get_free_pages(gfp_flags,
-						      get_order(size));
+	q->ring = vmalloc_user(size);
 	if (!q->ring) {
 		kfree(q);
 		return NULL;
 	}
 
+	q->ring_vmalloc_size = size;
 	return q;
 }
 

@@ -52,6 +51,6 @@ void xskq_destroy(struct xsk_queue *q)
 	if (!q)
 		return;
 
-	page_frag_free(q->ring);
+	vfree(q->ring);
 	kfree(q);
 }

diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h
index c6fb6b763658..bfb2a7e50c26 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_vmalloc_size;
 };
 
 /* The structure of the shared state of the rings are a simple

-- 
2.32.0.3.g01195cf9f

From: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Date: Thu, 16 Feb 2023 16:30:47 +0800

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
contiguous physical memory that will cause the start failure.

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 to these machine with long running time.

Here, that we actually waste 15 pages. 4-Order memory is 32 pages, but
we only use 17 pages.

This patch replaces __get_free_pages() by vmalloc() to allocate memory
to solve these problems.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
---

[...]

The name looks a bit long to me, but that might be just personal
preference. The code itself now looks good to me.

Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>

 };
 
 /* The structure of the shared state of the rings are a simple

Next time pls make sure you added all of the reviewers to the Cc list
when sending a new revision. I noticed you posted v4 only by monitoring
the ML.

Thanks,
Olek

On Thu, 16 Feb 2023 14:04:47 +0100, Alexander Lobakin [off-list ref] wrote:

From: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Date: Thu, 16 Feb 2023 16:30:47 +0800

quoted

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
contiguous physical memory that will cause the start failure.

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 to these machine with long running time.

Here, that we actually waste 15 pages. 4-Order memory is 32 pages, but
we only use 17 pages.

This patch replaces __get_free_pages() by vmalloc() to allocate memory
to solve these problems.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
---

[...]

quoted

diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h
index c6fb6b763658..bfb2a7e50c26 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_vmalloc_size;

The name looks a bit long to me, but that might be just personal
preference. The code itself now looks good to me.

Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>

quoted

 };

 /* The structure of the shared state of the rings are a simple

Next time pls make sure you added all of the reviewers to the Cc list
when sending a new revision. I noticed you posted v4 only by monitoring
the ML.

Oh, sorry. I always thought you were in the list. I did not notice this
situation.

I will pay attention next time. Thank you for your reply.

Thanks.

Thanks,
Olek

Hello:

This patch was applied to netdev/net-next.git (master)
by David S. Miller [off-list ref]:

On Thu, 16 Feb 2023 16:30:47 +0800 you 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
contiguous physical memory that will cause the start failure.

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 to these machine with long running time.

[...]

Here is the summary with links:
  - [net-next,v4] xsk: support use vaddr as ring
    https://git.kernel.org/netdev/net-next/c/9f78bf330a66

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html