[Intel-wired-lan] [PATCH v2 2/4] net: xsk: add a simple buffer reuse queue
Björn Töpel
bjorn.topel at gmail.com
Fri Sep 7 08:18:46 UTC 2018
From: Jakub Kicinski <jakub.kicinski at netronome.com>
XSK UMEM is strongly single producer single consumer so reuse of
frames is challenging. Add a simple "stash" of FILL packets to
reuse for drivers to optionally make use of. This is useful
when driver has to free (ndo_stop) or resize a ring with an active
AF_XDP ZC socket.
v2: Fixed build issues for !CONFIG_XDP_SOCKETS.
Signed-off-by: Jakub Kicinski <jakub.kicinski at netronome.com>
---
include/net/xdp_sock.h | 69 ++++++++++++++++++++++++++++++++++++++++++
net/xdp/xdp_umem.c | 2 ++
net/xdp/xsk_queue.c | 55 +++++++++++++++++++++++++++++++++
net/xdp/xsk_queue.h | 3 ++
4 files changed, 129 insertions(+)
diff --git a/include/net/xdp_sock.h b/include/net/xdp_sock.h
index 932ca0dad6f3..70a115bea4f4 100644
--- a/include/net/xdp_sock.h
+++ b/include/net/xdp_sock.h
@@ -21,6 +21,12 @@ struct xdp_umem_page {
dma_addr_t dma;
};
+struct xdp_umem_fq_reuse {
+ u32 nentries;
+ u32 length;
+ u64 handles[];
+};
+
struct xdp_umem {
struct xsk_queue *fq;
struct xsk_queue *cq;
@@ -37,6 +43,7 @@ struct xdp_umem {
struct page **pgs;
u32 npgs;
struct net_device *dev;
+ struct xdp_umem_fq_reuse *fq_reuse;
u16 queue_id;
bool zc;
spinlock_t xsk_list_lock;
@@ -75,6 +82,10 @@ void xsk_umem_discard_addr(struct xdp_umem *umem);
void xsk_umem_complete_tx(struct xdp_umem *umem, u32 nb_entries);
bool xsk_umem_consume_tx(struct xdp_umem *umem, dma_addr_t *dma, u32 *len);
void xsk_umem_consume_tx_done(struct xdp_umem *umem);
+struct xdp_umem_fq_reuse *xsk_reuseq_prepare(u32 nentries);
+struct xdp_umem_fq_reuse *xsk_reuseq_swap(struct xdp_umem *umem,
+ struct xdp_umem_fq_reuse *newq);
+void xsk_reuseq_free(struct xdp_umem_fq_reuse *rq);
static inline char *xdp_umem_get_data(struct xdp_umem *umem, u64 addr)
{
@@ -85,6 +96,35 @@ static inline dma_addr_t xdp_umem_get_dma(struct xdp_umem *umem, u64 addr)
{
return umem->pages[addr >> PAGE_SHIFT].dma + (addr & (PAGE_SIZE - 1));
}
+
+/* Reuse-queue aware version of FILL queue helpers */
+static inline u64 *xsk_umem_peek_addr_rq(struct xdp_umem *umem, u64 *addr)
+{
+ struct xdp_umem_fq_reuse *rq = umem->fq_reuse;
+
+ if (!rq->length)
+ return xsk_umem_peek_addr(umem, addr);
+
+ *addr = rq->handles[rq->length - 1];
+ return addr;
+}
+
+static inline void xsk_umem_discard_addr_rq(struct xdp_umem *umem)
+{
+ struct xdp_umem_fq_reuse *rq = umem->fq_reuse;
+
+ if (!rq->length)
+ xsk_umem_discard_addr(umem);
+ else
+ rq->length--;
+}
+
+static inline void xsk_umem_fq_reuse(struct xdp_umem *umem, u64 addr)
+{
+ struct xdp_umem_fq_reuse *rq = umem->fq_reuse;
+
+ rq->handles[rq->length++] = addr;
+}
#else
static inline int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
{
@@ -128,6 +168,21 @@ static inline void xsk_umem_consume_tx_done(struct xdp_umem *umem)
{
}
+static inline struct xdp_umem_fq_reuse *xsk_reuseq_prepare(u32 nentries)
+{
+ return NULL;
+}
+
+static inline struct xdp_umem_fq_reuse *xsk_reuseq_swap(
+ struct xdp_umem *umem,
+ struct xdp_umem_fq_reuse *newq)
+{
+ return NULL;
+}
+static inline void xsk_reuseq_free(struct xdp_umem_fq_reuse *rq)
+{
+}
+
static inline char *xdp_umem_get_data(struct xdp_umem *umem, u64 addr)
{
return NULL;
@@ -137,6 +192,20 @@ static inline dma_addr_t xdp_umem_get_dma(struct xdp_umem *umem, u64 addr)
{
return 0;
}
+
+static inline u64 *xsk_umem_peek_addr_rq(struct xdp_umem *umem, u64 *addr)
+{
+ return NULL;
+}
+
+static inline void xsk_umem_discard_addr_rq(struct xdp_umem *umem)
+{
+}
+
+static inline void xsk_umem_fq_reuse(struct xdp_umem *umem, u64 addr)
+{
+}
+
#endif /* CONFIG_XDP_SOCKETS */
#endif /* _LINUX_XDP_SOCK_H */
diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
index b3b632c5aeae..555427b3e0fe 100644
--- a/net/xdp/xdp_umem.c
+++ b/net/xdp/xdp_umem.c
@@ -165,6 +165,8 @@ static void xdp_umem_release(struct xdp_umem *umem)
umem->cq = NULL;
}
+ xsk_reuseq_destroy(umem);
+
xdp_umem_unpin_pages(umem);
task = get_pid_task(umem->pid, PIDTYPE_PID);
diff --git a/net/xdp/xsk_queue.c b/net/xdp/xsk_queue.c
index 2dc1384d9f27..b66504592d9b 100644
--- a/net/xdp/xsk_queue.c
+++ b/net/xdp/xsk_queue.c
@@ -3,7 +3,9 @@
* Copyright(c) 2018 Intel Corporation.
*/
+#include <linux/log2.h>
#include <linux/slab.h>
+#include <linux/overflow.h>
#include "xsk_queue.h"
@@ -62,3 +64,56 @@ void xskq_destroy(struct xsk_queue *q)
page_frag_free(q->ring);
kfree(q);
}
+
+struct xdp_umem_fq_reuse *xsk_reuseq_prepare(u32 nentries)
+{
+ struct xdp_umem_fq_reuse *newq;
+
+ /* Check for overflow */
+ if (nentries > (u32)roundup_pow_of_two(nentries))
+ return NULL;
+ nentries = roundup_pow_of_two(nentries);
+
+ newq = kvmalloc(struct_size(newq, handles, nentries), GFP_KERNEL);
+ if (!newq)
+ return NULL;
+ memset(newq, 0, offsetof(typeof(*newq), handles));
+
+ newq->nentries = nentries;
+ return newq;
+}
+EXPORT_SYMBOL_GPL(xsk_reuseq_prepare);
+
+struct xdp_umem_fq_reuse *xsk_reuseq_swap(struct xdp_umem *umem,
+ struct xdp_umem_fq_reuse *newq)
+{
+ struct xdp_umem_fq_reuse *oldq = umem->fq_reuse;
+
+ if (!oldq) {
+ umem->fq_reuse = newq;
+ return NULL;
+ }
+
+ if (newq->nentries < oldq->length)
+ return newq;
+
+ memcpy(newq->handles, oldq->handles,
+ array_size(oldq->length, sizeof(u64)));
+ newq->length = oldq->length;
+
+ umem->fq_reuse = newq;
+ return oldq;
+}
+EXPORT_SYMBOL_GPL(xsk_reuseq_swap);
+
+void xsk_reuseq_free(struct xdp_umem_fq_reuse *rq)
+{
+ kvfree(rq);
+}
+EXPORT_SYMBOL_GPL(xsk_reuseq_free);
+
+void xsk_reuseq_destroy(struct xdp_umem *umem)
+{
+ xsk_reuseq_free(umem->fq_reuse);
+ umem->fq_reuse = NULL;
+}
diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h
index 82252cccb4e0..bcb5cbb40419 100644
--- a/net/xdp/xsk_queue.h
+++ b/net/xdp/xsk_queue.h
@@ -258,4 +258,7 @@ void xskq_set_umem(struct xsk_queue *q, u64 size, u64 chunk_mask);
struct xsk_queue *xskq_create(u32 nentries, bool umem_queue);
void xskq_destroy(struct xsk_queue *q_ops);
+/* Executed by the core when the entire UMEM gets freed */
+void xsk_reuseq_destroy(struct xdp_umem *umem);
+
#endif /* _LINUX_XSK_QUEUE_H */
--
2.17.1
More information about the Intel-wired-lan
mailing list