On Sat, 31 Aug 2024 00:43:06 +0000 Mina Almasry wrote:
quoted hunk ↗ jump to hunk
diff --git a/include/net/mp_dmabuf_devmem.h b/include/net/mp_dmabuf_devmem.h
new file mode 100644
index 000000000000..6d1cf2a77f6b
--- /dev/null
+++ b/include/net/mp_dmabuf_devmem.h
this header can live under net/core/ like netmem_priv.h right?
devmem internals should be of no interest outside of core networking.
In fact the same is true for include/net/devmem.h ?
Sorry for pushing back on all the header exports, we have had bad
experiences with people treating anything under include/ as public
API for any subsystem to use..
quoted hunk ↗ jump to hunk
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Dmabuf device memory provider.
+ *
+ * Authors: Mina Almasry <almasrymina@google.com>
+ *
+ */
+#ifndef _NET_MP_DMABUF_DEVMEM_H
+#define _NET_MP_DMABUF_DEVMEM_H
+
+#include <net/netmem.h>
+
+#if defined(CONFIG_NET_DEVMEM)
+int mp_dmabuf_devmem_init(struct page_pool *pool);
+
+netmem_ref mp_dmabuf_devmem_alloc_netmems(struct page_pool *pool, gfp_t gfp);
+
+void mp_dmabuf_devmem_destroy(struct page_pool *pool);
+
+bool mp_dmabuf_devmem_release_page(struct page_pool *pool, netmem_ref netmem);
+#else
+static inline int mp_dmabuf_devmem_init(struct page_pool *pool)
+{
+ return -EOPNOTSUPP;
+}
+
+static inline netmem_ref mp_dmabuf_devmem_alloc_netmems(struct page_pool *pool,
+ gfp_t gfp)
Please break the lines after the return type if the line gets long:
static inline netmem_ref
mp_dmabuf_devmem_alloc_netmems(struct page_pool *pool, gfp_t gfp)
Please fix where you can (at least where it cases going over 80 chars)
quoted hunk ↗ jump to hunk
+{
+ return 0;
+}
+
+static inline void mp_dmabuf_devmem_destroy(struct page_pool *pool)
+{
+}
+
+static inline bool mp_dmabuf_devmem_release_page(struct page_pool *pool,
+ netmem_ref netmem)
+{
+ return false;
+}
+#endif
+
+#endif /* _NET_MP_DMABUF_DEVMEM_H */diff --git a/include/net/netmem.h b/include/net/netmem.h
index ac6c7945117b..61400d4b0d66 100644
--- a/include/net/netmem.h
+++ b/include/net/netmem.h
@@ -8,6 +8,7 @@
#ifndef _NET_NETMEM_H
#define _NET_NETMEM_H
+#include <linux/mm.h>
#include <net/devmem.h>
#include <net/net_debug.h>
diff --git a/include/net/page_pool/types.h b/include/net/page_pool/types.h
index 4afd6dd56351..1b4698710f25 100644
--- a/include/net/page_pool/types.h
+++ b/include/net/page_pool/types.h
@@ -20,8 +20,17 @@
* device driver responsibility
*/
#define PP_FLAG_SYSTEM_POOL BIT(2) /* Global system page_pool */
+#define PP_FLAG_ALLOW_UNREADABLE_NETMEM BIT(3) /* Allow unreadable (net_iov
+ * backed) netmem in this
+ * page_pool. Drivers setting
+ * this must be able to support
+ * unreadable netmem, where
+ * netmem_address() would return
+ * NULL. This flag should not be
+ * set for header page_pools.
+ */
Maybe move the comment before the define:
/* Allow unreadable (net_iov backed) netmem in this page_pool. Drivers
setting
* this must be able to support unreadable netmem, where netmem_address() would
* return NULL. This flag should not be set for header page_pools.
*/
#define PP_FLAG_ALLOW_UNREADABLE_NETMEM BIT(3)
? up to you.
quoted hunk ↗ jump to hunk
#define PP_FLAG_ALL (PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV | \
- PP_FLAG_SYSTEM_POOL)
+ PP_FLAG_SYSTEM_POOL | PP_FLAG_ALLOW_UNREADABLE_NETMEM)
/*
* Fast allocation side cache array/stack
@@ -57,7 +66,9 @@ struct pp_alloc_cache {
* @offset: DMA sync address offset for PP_FLAG_DMA_SYNC_DEV
* @slow: params with slowpath access only (initialization and Netlink)
* @netdev: netdev this pool will serve (leave as NULL if none or multiple)
- * @flags: PP_FLAG_DMA_MAP, PP_FLAG_DMA_SYNC_DEV, PP_FLAG_SYSTEM_POOL
+ * @queue: struct netdev_rx_queue this page_pool is being created for.
+ * @flags: PP_FLAG_DMA_MAP, PP_FLAG_DMA_SYNC_DEV, PP_FLAG_SYSTEM_POOL,
+ * PP_FLAG_ALLOW_UNREADABLE_NETMEM.
*/
struct page_pool_params {
struct_group_tagged(page_pool_params_fast, fast,@@ -72,6 +83,7 @@ struct page_pool_params {
);
struct_group_tagged(page_pool_params_slow, slow,
struct net_device *netdev;
+ struct netdev_rx_queue *queue;
Why set a pointer? It should work but drivers don't usually deal with
netdev_rx_queue struct directly. struct xdp_rxq_info takes an integer
queue id, and it serves a somewhat similar function.
Keep in mind that there will be more drivers than core code, so
convenience for them matters more.
quoted hunk ↗ jump to hunk
unsigned int flags;
/* private: used by test code only */
void (*init_callback)(netmem_ref netmem, void *arg);
diff --git a/net/core/devmem.c b/net/core/devmem.c
index 727e5ee39f30..c8c112360caa 100644
--- a/net/core/devmem.c
+++ b/net/core/devmem.c
@@ -13,6 +13,7 @@
#include <linux/netdevice.h>
#include <linux/types.h>
#include <net/devmem.h>
+#include <net/mp_dmabuf_devmem.h>
#include <net/netdev_queues.h>
#include <net/netdev_rx_queue.h>
#include <net/page_pool/helpers.h>
@@ -320,3 +321,68 @@ void dev_dmabuf_uninstall(struct net_device *dev)
}
}
}
+
+/*** "Dmabuf devmem memory provider" ***/
+
+int mp_dmabuf_devmem_init(struct page_pool *pool)
+{
+ struct net_devmem_dmabuf_binding *binding = pool->mp_priv;
+
+ if (!binding)
+ return -EINVAL;
+
+ if (!pool->dma_map)
+ return -EOPNOTSUPP;
+
+ if (pool->dma_sync)
+ return -EOPNOTSUPP;
+
+ if (pool->p.order != 0)
+ return -E2BIG;
+
+ net_devmem_dmabuf_binding_get(binding);
+ return 0;
+}
+
+netmem_ref mp_dmabuf_devmem_alloc_netmems(struct page_pool *pool, gfp_t gfp)
+{
+ struct net_devmem_dmabuf_binding *binding = pool->mp_priv;
+ netmem_ref netmem;
+ struct net_iov *niov;
nit: reverse xmas tree
quoted hunk ↗ jump to hunk
+ niov = net_devmem_alloc_dmabuf(binding);
+ if (!niov)
+ return 0;
+
+ netmem = net_iov_to_netmem(niov);
+
+ page_pool_set_pp_info(pool, netmem);
+
+ pool->pages_state_hold_cnt++;
+ trace_page_pool_state_hold(pool, netmem, pool->pages_state_hold_cnt);
+ return netmem;
+}
+
+void mp_dmabuf_devmem_destroy(struct page_pool *pool)
+{
+ struct net_devmem_dmabuf_binding *binding = pool->mp_priv;
+
+ net_devmem_dmabuf_binding_put(binding);
+}
+
+bool mp_dmabuf_devmem_release_page(struct page_pool *pool, netmem_ref netmem)
+{
+ if (WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
+ return false;
+
+ if (WARN_ON_ONCE(atomic_long_read(netmem_get_pp_ref_count_ref(netmem)) !=
+ 1))
something needs factoring out here, to make this line shorter, please..
either netmem -> net_iov conversion or at least reading of the ref
count?
quoted hunk ↗ jump to hunk
+ return false;
+
+ page_pool_clear_pp_info(netmem);
+
+ net_devmem_free_dmabuf(netmem_to_net_iov(netmem));
+
+ /* We don't want the page pool put_page()ing our net_iovs. */
+ return false;
+}
diff --git a/net/core/netdev_rx_queue.c b/net/core/netdev_rx_queue.c
index da11720a5983..e217a5838c87 100644
--- a/net/core/netdev_rx_queue.c
+++ b/net/core/netdev_rx_queue.c
@@ -4,8 +4,11 @@
#include <net/netdev_queues.h>
#include <net/netdev_rx_queue.h>
+#include "page_pool_priv.h"
+
int netdev_rx_queue_restart(struct net_device *dev, unsigned int rxq_idx)
{
+ struct netdev_rx_queue *rxq = __netif_get_rx_queue(dev, rxq_idx);
void *new_mem, *old_mem;
int err;
@@ -31,6 +34,10 @@ int netdev_rx_queue_restart(struct net_device *dev, unsigned int rxq_idx)
if (err)
goto err_free_old_mem;
+ err = page_pool_check_memory_provider(dev, rxq);
+ if (err)
+ goto err_free_new_queue_mem;
+
err = dev->queue_mgmt_ops->ndo_queue_stop(dev, old_mem, rxq_idx);
if (err)
goto err_free_new_queue_mem;
diff --git a/net/core/page_pool.c b/net/core/page_pool.c
index 52659db2d765..6e24950f2be4 100644
--- a/net/core/page_pool.c
+++ b/net/core/page_pool.c
@@ -11,6 +11,8 @@
#include <linux/slab.h>
#include <linux/device.h>
+#include <net/mp_dmabuf_devmem.h>
+#include <net/netdev_rx_queue.h>
#include <net/page_pool/helpers.h>
#include <net/xdp.h>
@@ -190,6 +192,7 @@ static int page_pool_init(struct page_pool *pool,
int cpuid)
{
unsigned int ring_qsize = 1024; /* Default */
+ int err;
page_pool_struct_check();
@@ -271,7 +274,36 @@ static int page_pool_init(struct page_pool *pool,
if (pool->dma_map)
get_device(pool->p.dev);
+ if (pool->slow.queue &&
+ pool->slow.flags & PP_FLAG_ALLOW_UNREADABLE_NETMEM) {
+ /* We rely on rtnl_lock()ing to make sure netdev_rx_queue
+ * configuration doesn't change while we're initializing the
nit: 'the' to next line
quoted hunk ↗ jump to hunk
+ * page_pool.
+ */
+ ASSERT_RTNL();
+ pool->mp_priv = pool->slow.queue->mp_params.mp_priv;
+ }