linux/Documentation/networking/page_pool.rst
Jakub Kicinski 8c48eea3ad page_pool: allow caching from safely localized NAPI
Recent patches to mlx5 mentioned a regression when moving from
driver local page pool to only using the generic page pool code.
Page pool has two recycling paths (1) direct one, which runs in
safe NAPI context (basically consumer context, so producing
can be lockless); and (2) via a ptr_ring, which takes a spin
lock because the freeing can happen from any CPU; producer
and consumer may run concurrently.

Since the page pool code was added, Eric introduced a revised version
of deferred skb freeing. TCP skbs are now usually returned to the CPU
which allocated them, and freed in softirq context. This places the
freeing (producing of pages back to the pool) enticingly close to
the allocation (consumer).

If we can prove that we're freeing in the same softirq context in which
the consumer NAPI will run - lockless use of the cache is perfectly fine,
no need for the lock.

Let drivers link the page pool to a NAPI instance. If the NAPI instance
is scheduled on the same CPU on which we're freeing - place the pages
in the direct cache.

With that and patched bnxt (XDP enabled to engage the page pool, sigh,
bnxt really needs page pool work :() I see a 2.6% perf boost with
a TCP stream test (app on a different physical core than softirq).

The CPU use of relevant functions decreases as expected:

  page_pool_refill_alloc_cache   1.17% -> 0%
  _raw_spin_lock                 2.41% -> 0.98%

Only consider lockless path to be safe when NAPI is scheduled
- in practice this should cover majority if not all of steady state
workloads. It's usually the NAPI kicking in that causes the skb flush.

The main case we'll miss out on is when application runs on the same
CPU as NAPI. In that case we don't use the deferred skb free path.

Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Tested-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-04-14 18:56:12 -07:00

225 lines
8.1 KiB
ReStructuredText

.. SPDX-License-Identifier: GPL-2.0
=============
Page Pool API
=============
The page_pool allocator is optimized for the XDP mode that uses one frame
per-page, but it can fallback on the regular page allocator APIs.
Basic use involves replacing alloc_pages() calls with the
page_pool_alloc_pages() call. Drivers should use page_pool_dev_alloc_pages()
replacing dev_alloc_pages().
API keeps track of in-flight pages, in order to let API user know
when it is safe to free a page_pool object. Thus, API users
must run page_pool_release_page() when a page is leaving the page_pool or
call page_pool_put_page() where appropriate in order to maintain correct
accounting.
API user must call page_pool_put_page() once on a page, as it
will either recycle the page, or in case of refcnt > 1, it will
release the DMA mapping and in-flight state accounting.
Architecture overview
=====================
.. code-block:: none
+------------------+
| Driver |
+------------------+
^
|
|
|
v
+--------------------------------------------+
| request memory |
+--------------------------------------------+
^ ^
| |
| Pool empty | Pool has entries
| |
v v
+-----------------------+ +------------------------+
| alloc (and map) pages | | get page from cache |
+-----------------------+ +------------------------+
^ ^
| |
| cache available | No entries, refill
| | from ptr-ring
| |
v v
+-----------------+ +------------------+
| Fast cache | | ptr-ring cache |
+-----------------+ +------------------+
API interface
=============
The number of pools created **must** match the number of hardware queues
unless hardware restrictions make that impossible. This would otherwise beat the
purpose of page pool, which is allocate pages fast from cache without locking.
This lockless guarantee naturally comes from running under a NAPI softirq.
The protection doesn't strictly have to be NAPI, any guarantee that allocating
a page will cause no race conditions is enough.
* page_pool_create(): Create a pool.
* flags: PP_FLAG_DMA_MAP, PP_FLAG_DMA_SYNC_DEV
* order: 2^order pages on allocation
* pool_size: size of the ptr_ring
* nid: preferred NUMA node for allocation
* dev: struct device. Used on DMA operations
* dma_dir: DMA direction
* max_len: max DMA sync memory size
* offset: DMA address offset
* page_pool_put_page(): The outcome of this depends on the page refcnt. If the
driver bumps the refcnt > 1 this will unmap the page. If the page refcnt is 1
the allocator owns the page and will try to recycle it in one of the pool
caches. If PP_FLAG_DMA_SYNC_DEV is set, the page will be synced for_device
using dma_sync_single_range_for_device().
* page_pool_put_full_page(): Similar to page_pool_put_page(), but will DMA sync
for the entire memory area configured in area pool->max_len.
* page_pool_recycle_direct(): Similar to page_pool_put_full_page() but caller
must guarantee safe context (e.g NAPI), since it will recycle the page
directly into the pool fast cache.
* page_pool_release_page(): Unmap the page (if mapped) and account for it on
in-flight counters.
* page_pool_dev_alloc_pages(): Get a page from the page allocator or page_pool
caches.
* page_pool_get_dma_addr(): Retrieve the stored DMA address.
* page_pool_get_dma_dir(): Retrieve the stored DMA direction.
* page_pool_put_page_bulk(): Tries to refill a number of pages into the
ptr_ring cache holding ptr_ring producer lock. If the ptr_ring is full,
page_pool_put_page_bulk() will release leftover pages to the page allocator.
page_pool_put_page_bulk() is suitable to be run inside the driver NAPI tx
completion loop for the XDP_REDIRECT use case.
Please note the caller must not use data area after running
page_pool_put_page_bulk(), as this function overwrites it.
* page_pool_get_stats(): Retrieve statistics about the page_pool. This API
is only available if the kernel has been configured with
``CONFIG_PAGE_POOL_STATS=y``. A pointer to a caller allocated ``struct
page_pool_stats`` structure is passed to this API which is filled in. The
caller can then report those stats to the user (perhaps via ethtool,
debugfs, etc.). See below for an example usage of this API.
Stats API and structures
------------------------
If the kernel is configured with ``CONFIG_PAGE_POOL_STATS=y``, the API
``page_pool_get_stats()`` and structures described below are available. It
takes a pointer to a ``struct page_pool`` and a pointer to a ``struct
page_pool_stats`` allocated by the caller.
The API will fill in the provided ``struct page_pool_stats`` with
statistics about the page_pool.
The stats structure has the following fields::
struct page_pool_stats {
struct page_pool_alloc_stats alloc_stats;
struct page_pool_recycle_stats recycle_stats;
};
The ``struct page_pool_alloc_stats`` has the following fields:
* ``fast``: successful fast path allocations
* ``slow``: slow path order-0 allocations
* ``slow_high_order``: slow path high order allocations
* ``empty``: ptr ring is empty, so a slow path allocation was forced.
* ``refill``: an allocation which triggered a refill of the cache
* ``waive``: pages obtained from the ptr ring that cannot be added to
the cache due to a NUMA mismatch.
The ``struct page_pool_recycle_stats`` has the following fields:
* ``cached``: recycling placed page in the page pool cache
* ``cache_full``: page pool cache was full
* ``ring``: page placed into the ptr ring
* ``ring_full``: page released from page pool because the ptr ring was full
* ``released_refcnt``: page released (and not recycled) because refcnt > 1
Coding examples
===============
Registration
------------
.. code-block:: c
/* Page pool registration */
struct page_pool_params pp_params = { 0 };
struct xdp_rxq_info xdp_rxq;
int err;
pp_params.order = 0;
/* internal DMA mapping in page_pool */
pp_params.flags = PP_FLAG_DMA_MAP;
pp_params.pool_size = DESC_NUM;
pp_params.nid = NUMA_NO_NODE;
pp_params.dev = priv->dev;
pp_params.napi = napi; /* only if locking is tied to NAPI */
pp_params.dma_dir = xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
page_pool = page_pool_create(&pp_params);
err = xdp_rxq_info_reg(&xdp_rxq, ndev, 0);
if (err)
goto err_out;
err = xdp_rxq_info_reg_mem_model(&xdp_rxq, MEM_TYPE_PAGE_POOL, page_pool);
if (err)
goto err_out;
NAPI poller
-----------
.. code-block:: c
/* NAPI Rx poller */
enum dma_data_direction dma_dir;
dma_dir = page_pool_get_dma_dir(dring->page_pool);
while (done < budget) {
if (some error)
page_pool_recycle_direct(page_pool, page);
if (packet_is_xdp) {
if XDP_DROP:
page_pool_recycle_direct(page_pool, page);
} else (packet_is_skb) {
page_pool_release_page(page_pool, page);
new_page = page_pool_dev_alloc_pages(page_pool);
}
}
Stats
-----
.. code-block:: c
#ifdef CONFIG_PAGE_POOL_STATS
/* retrieve stats */
struct page_pool_stats stats = { 0 };
if (page_pool_get_stats(page_pool, &stats)) {
/* perhaps the driver reports statistics with ethool */
ethtool_print_allocation_stats(&stats.alloc_stats);
ethtool_print_recycle_stats(&stats.recycle_stats);
}
#endif
Driver unload
-------------
.. code-block:: c
/* Driver unload */
page_pool_put_full_page(page_pool, page, false);
xdp_rxq_info_unreg(&xdp_rxq);