[net-next,v24,06/13] memory-provider: dmabuf devmem memory provider
Commit Message
Implement a memory provider that allocates dmabuf devmem in the form of
net_iov.
The provider receives a reference to the struct netdev_dmabuf_binding
via the pool->mp_priv pointer. The driver needs to set this pointer for
the provider in the net_iov.
The provider obtains a reference on the netdev_dmabuf_binding which
guarantees the binding and the underlying mapping remains alive until
the provider is destroyed.
Usage of PP_FLAG_DMA_MAP is required for this memory provide such that
the page_pool can provide the driver with the dma-addrs of the devmem.
Support for PP_FLAG_DMA_SYNC_DEV is omitted for simplicity & p.order !=
0.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Kaiyuan Zhang <kaiyuanz@google.com>
Signed-off-by: Mina Almasry <almasrymina@google.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
---
v23:
- Sort includes (Jakub)
- Add missing linux/mm.h include found after sorting.
v21:
- Provide empty definitions of functions moved to page_pool_priv.h, so
that the build still succeeds when CONFIG_PAGE_POOL is not set.
v20:
- Moved queue pp_params field from fast path entries to slow path
entries.
- Moved page_pool_check_memory_provider() call to inside
netdev_rx_queue_restart (Pavel).
- Removed binding arg to page_pool_check_memory_provider() (Pavel).
- Removed unnecessary includes from page_pool.c
- Removed EXPORT_SYMBOL(page_pool_mem_providers) (Jakub)
- Check pool->slow.queue instead of walking binding xarray (Pavel
& Jakub).
v19:
- Add PP_FLAG_ALLOW_UNREADABLE_NETMEM flag. It serves 2 purposes, (a)
it guards drivers that don't support unreadable netmem (net_iov
backed) from accidentally getting exposed to it, and (b) drivers that
wish to create header pools can unset it for that pool to force
readable netmem.
- Add page_pool_check_memory_provider, which verifies that the driver
has created a page_pool with the expected configuration. This is used
to report to the user if the mp configuration succeeded, and also
verify that the driver is doing the right thing.
- Don't reset niov->dma_addr on allocation/free.
v17:
- Use ASSERT_RTNL (Jakub)
v16:
- Add DEBUG_NET_WARN_ON_ONCE(!rtnl_is_locked()), to catch cases if
page_pool_init without rtnl_locking when the queue is provided. In
this case, the queue configuration may be changed while we're initing
the page_pool, which could be a race.
v13:
- Return on warning (Pavel).
- Fixed pool->recycle_stats not being freed on error (Pavel).
- Applied reviewed-by from Pavel.
v11:
- Rebase to not use the ops. (Christoph)
v8:
- Use skb_frag_size instead of frag->bv_len to fix patch-by-patch build
error
v6:
- refactor new memory provider functions into net/core/devmem.c (Pavel)
v2:
- Disable devmem for p.order != 0
v1:
- static_branch check in page_is_page_pool_iov() (Willem & Paolo).
- PP_DEVMEM -> PP_IOV (David).
- Require PP_FLAG_DMA_MAP (Jakub).
---
include/net/mp_dmabuf_devmem.h | 44 +++++++++++++++
include/net/netmem.h | 1 +
include/net/page_pool/types.h | 16 +++++-
net/core/devmem.c | 66 ++++++++++++++++++++++
net/core/netdev_rx_queue.c | 7 +++
net/core/page_pool.c | 100 ++++++++++++++++++++++++---------
net/core/page_pool_priv.h | 20 +++++++
net/core/page_pool_user.c | 27 ++++++++-
8 files changed, 252 insertions(+), 29 deletions(-)
create mode 100644 include/net/mp_dmabuf_devmem.h
Comments
On Sat, 31 Aug 2024 00:43:06 +0000 Mina Almasry wrote:
> 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..
> @@ -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)
> +{
> + 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.
> #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.
> 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
> + 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?
> + 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
> + * page_pool.
> + */
> + ASSERT_RTNL();
> + pool->mp_priv = pool->slow.queue->mp_params.mp_priv;
> + }
On Tue, Sep 3, 2024 at 2:19 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Sat, 31 Aug 2024 00:43:06 +0000 Mina Almasry wrote:
> > 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.
>
Yes, those can be moved under net/core trivially. done.
> In fact the same is true for include/net/devmem.h ?
>
This turned out to be possible, but with a minor moving around of some
helpers. Basically netmem.h included devmem.h to get access to some
devmem internals for some of the net_iov helpers specific to devmem.
Moving these helpers to devmem.h enabled me to keep
include/net/netmem.h but put devmem.h under net/core. Now netmem.h
doesn't need to include devmem.h. I think this is an improvement.
> > +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)
>
FWIW I use a formatting tool (clang-format) which seems to prefer
breaking in between the args, but I'll fix this manually and wherever
else I notice.
> > 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.
>
Makes sense.
> > +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?
>
Ah, sorry I think you pointed this out earlier and I missed applying
it. Should be done in the next iteration.
--
Thanks,
Mina
new file mode 100644
@@ -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)
+{
+ 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 */
@@ -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>
@@ -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.
+ */
#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;
unsigned int flags;
/* private: used by test code only */
void (*init_callback)(netmem_ref netmem, void *arg);
@@ -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;
+
+ 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))
+ 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;
+}
@@ -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;
@@ -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
+ * page_pool.
+ */
+ ASSERT_RTNL();
+ pool->mp_priv = pool->slow.queue->mp_params.mp_priv;
+ }
+
+ if (pool->mp_priv) {
+ err = mp_dmabuf_devmem_init(pool);
+ if (err) {
+ pr_warn("%s() mem-provider init failed %d\n", __func__,
+ err);
+ goto free_ptr_ring;
+ }
+
+ static_branch_inc(&page_pool_mem_providers);
+ }
+
return 0;
+
+free_ptr_ring:
+ ptr_ring_cleanup(&pool->ring, NULL);
+#ifdef CONFIG_PAGE_POOL_STATS
+ if (!pool->system)
+ free_percpu(pool->recycle_stats);
+#endif
+ return err;
}
static void page_pool_uninit(struct page_pool *pool)
@@ -455,28 +487,6 @@ static bool page_pool_dma_map(struct page_pool *pool, netmem_ref netmem)
return false;
}
-static void page_pool_set_pp_info(struct page_pool *pool, netmem_ref netmem)
-{
- netmem_set_pp(netmem, pool);
- netmem_or_pp_magic(netmem, PP_SIGNATURE);
-
- /* Ensuring all pages have been split into one fragment initially:
- * page_pool_set_pp_info() is only called once for every page when it
- * is allocated from the page allocator and page_pool_fragment_page()
- * is dirtying the same cache line as the page->pp_magic above, so
- * the overhead is negligible.
- */
- page_pool_fragment_netmem(netmem, 1);
- if (pool->has_init_callback)
- pool->slow.init_callback(netmem, pool->slow.init_arg);
-}
-
-static void page_pool_clear_pp_info(netmem_ref netmem)
-{
- netmem_clear_pp_magic(netmem);
- netmem_set_pp(netmem, NULL);
-}
-
static struct page *__page_pool_alloc_page_order(struct page_pool *pool,
gfp_t gfp)
{
@@ -572,7 +582,10 @@ netmem_ref page_pool_alloc_netmem(struct page_pool *pool, gfp_t gfp)
return netmem;
/* Slow-path: cache empty, do real allocation */
- netmem = __page_pool_alloc_pages_slow(pool, gfp);
+ if (static_branch_unlikely(&page_pool_mem_providers) && pool->mp_priv)
+ netmem = mp_dmabuf_devmem_alloc_netmems(pool, gfp);
+ else
+ netmem = __page_pool_alloc_pages_slow(pool, gfp);
return netmem;
}
EXPORT_SYMBOL(page_pool_alloc_netmem);
@@ -608,6 +621,28 @@ s32 page_pool_inflight(const struct page_pool *pool, bool strict)
return inflight;
}
+void page_pool_set_pp_info(struct page_pool *pool, netmem_ref netmem)
+{
+ netmem_set_pp(netmem, pool);
+ netmem_or_pp_magic(netmem, PP_SIGNATURE);
+
+ /* Ensuring all pages have been split into one fragment initially:
+ * page_pool_set_pp_info() is only called once for every page when it
+ * is allocated from the page allocator and page_pool_fragment_page()
+ * is dirtying the same cache line as the page->pp_magic above, so
+ * the overhead is negligible.
+ */
+ page_pool_fragment_netmem(netmem, 1);
+ if (pool->has_init_callback)
+ pool->slow.init_callback(netmem, pool->slow.init_arg);
+}
+
+void page_pool_clear_pp_info(netmem_ref netmem)
+{
+ netmem_clear_pp_magic(netmem);
+ netmem_set_pp(netmem, NULL);
+}
+
static __always_inline void __page_pool_release_page_dma(struct page_pool *pool,
netmem_ref netmem)
{
@@ -636,8 +671,13 @@ static __always_inline void __page_pool_release_page_dma(struct page_pool *pool,
void page_pool_return_page(struct page_pool *pool, netmem_ref netmem)
{
int count;
+ bool put;
- __page_pool_release_page_dma(pool, netmem);
+ put = true;
+ if (static_branch_unlikely(&page_pool_mem_providers) && pool->mp_priv)
+ put = mp_dmabuf_devmem_release_page(pool, netmem);
+ else
+ __page_pool_release_page_dma(pool, netmem);
/* This may be the last page returned, releasing the pool, so
* it is not safe to reference pool afterwards.
@@ -645,8 +685,10 @@ void page_pool_return_page(struct page_pool *pool, netmem_ref netmem)
count = atomic_inc_return_relaxed(&pool->pages_state_release_cnt);
trace_page_pool_state_release(pool, netmem, count);
- page_pool_clear_pp_info(netmem);
- put_page(netmem_to_page(netmem));
+ if (put) {
+ page_pool_clear_pp_info(netmem);
+ put_page(netmem_to_page(netmem));
+ }
/* An optimization would be to call __free_pages(page, pool->p.order)
* knowing page is not part of page-cache (thus avoiding a
* __page_cache_release() call).
@@ -965,6 +1007,12 @@ static void __page_pool_destroy(struct page_pool *pool)
page_pool_unlist(pool);
page_pool_uninit(pool);
+
+ if (pool->mp_priv) {
+ mp_dmabuf_devmem_destroy(pool);
+ static_branch_dec(&page_pool_mem_providers);
+ }
+
kfree(pool);
}
@@ -35,4 +35,24 @@ static inline bool page_pool_set_dma_addr(struct page *page, dma_addr_t addr)
return page_pool_set_dma_addr_netmem(page_to_netmem(page), addr);
}
+#if defined(CONFIG_PAGE_POOL)
+void page_pool_set_pp_info(struct page_pool *pool, netmem_ref netmem);
+void page_pool_clear_pp_info(netmem_ref netmem);
+int page_pool_check_memory_provider(struct net_device *dev,
+ struct netdev_rx_queue *rxq);
+#else
+static inline void page_pool_set_pp_info(struct page_pool *pool,
+ netmem_ref netmem)
+{
+}
+static inline void page_pool_clear_pp_info(netmem_ref netmem)
+{
+}
+static inline int page_pool_check_memory_provider(struct net_device *dev,
+ struct netdev_rx_queue *rxq)
+{
+ return 0;
+}
+#endif
+
#endif
@@ -4,8 +4,9 @@
#include <linux/netdevice.h>
#include <linux/xarray.h>
#include <net/net_debug.h>
-#include <net/page_pool/types.h>
+#include <net/netdev_rx_queue.h>
#include <net/page_pool/helpers.h>
+#include <net/page_pool/types.h>
#include <net/sock.h>
#include "page_pool_priv.h"
@@ -344,6 +345,30 @@ void page_pool_unlist(struct page_pool *pool)
mutex_unlock(&page_pools_lock);
}
+int page_pool_check_memory_provider(struct net_device *dev,
+ struct netdev_rx_queue *rxq)
+{
+ struct net_devmem_dmabuf_binding *binding = rxq->mp_params.mp_priv;
+ struct page_pool *pool;
+ struct hlist_node *n;
+
+ if (!binding)
+ return 0;
+
+ mutex_lock(&page_pools_lock);
+ hlist_for_each_entry_safe(pool, n, &dev->page_pools, user.list) {
+ if (pool->mp_priv != binding)
+ continue;
+
+ if (pool->slow.queue == rxq) {
+ mutex_unlock(&page_pools_lock);
+ return 0;
+ }
+ }
+ mutex_unlock(&page_pools_lock);
+ return -ENODATA;
+}
+
static void page_pool_unreg_netdev_wipe(struct net_device *netdev)
{
struct page_pool *pool;