The BPF_TEST_RUN code in XDP live frame mode creates a new page pool
each time it is called and uses that to allocate the frames used for the
XDP run. This works well if the syscall is used with a high repetitions
number, as it allows for efficient page recycling. However, if used with
a small number of repetitions, the overhead of creating and tearing down
the page pool is significant, and can even lead to system stalls if the
syscall is called in a tight loop.
Now that we have a persistent system page pool instance, it becomes
pretty straight forward to change the test_run code to use it. The only
wrinkle is that we can no longer rely on a custom page init callback
from page_pool itself; instead, we change the test_run code to write a
random cookie value to the beginning of the page as an indicator that
the page has been initialised and can be re-used without copying the
initial data again.
Signed-off-by: Toke Høiland-Jørgensen <toke@xxxxxxxxxx>
---
net/bpf/test_run.c | 134 ++++++++++++++++++++++-----------------------
1 file changed, 66 insertions(+), 68 deletions(-)
diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
index dfd919374017..c742869a0612 100644
--- a/net/bpf/test_run.c
+++ b/net/bpf/test_run.c
@@ -94,10 +94,14 @@ static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
}
/* We put this struct at the head of each page with a context and frame
- * initialised when the page is allocated, so we don't have to do this on each
- * repetition of the test run.
+ * initialised the first time a given page is used, saving the memcpy() of the
+ * data on subsequent repetition of the test run. The cookie value is used to
+ * mark the page data the first time we initialise it so we can skip it the next
+ * time we see that page.
*/
+
struct xdp_page_head {
+ u64 cookie;
struct xdp_buff orig_ctx;
struct xdp_buff ctx;
union {
@@ -111,10 +115,9 @@ struct xdp_test_data {
struct xdp_buff *orig_ctx;
struct xdp_rxq_info rxq;
struct net_device *dev;
- struct page_pool *pp;
struct xdp_frame **frames;
struct sk_buff **skbs;
- struct xdp_mem_info mem;
+ u64 cookie;
u32 batch_size;
u32 frame_cnt;
};
@@ -126,48 +129,9 @@ struct xdp_test_data {
#define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
#define TEST_XDP_MAX_BATCH 256
-static void xdp_test_run_init_page(struct page *page, void *arg)
-{
- struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
- struct xdp_buff *new_ctx, *orig_ctx;
- u32 headroom = XDP_PACKET_HEADROOM;
- struct xdp_test_data *xdp = arg;
- size_t frm_len, meta_len;
- struct xdp_frame *frm;
- void *data;
-
- orig_ctx = xdp->orig_ctx;
- frm_len = orig_ctx->data_end - orig_ctx->data_meta;
- meta_len = orig_ctx->data - orig_ctx->data_meta;
- headroom -= meta_len;
-
- new_ctx = &head->ctx;
- frm = head->frame;
- data = head->data;
- memcpy(data + headroom, orig_ctx->data_meta, frm_len);
-
- xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
- xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
- new_ctx->data = new_ctx->data_meta + meta_len;
-
- xdp_update_frame_from_buff(new_ctx, frm);
- frm->mem = new_ctx->rxq->mem;
-
- memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
-}
-
static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
{
- struct page_pool *pp;
int err = -ENOMEM;
- struct page_pool_params pp_params = {
- .order = 0,
- .flags = 0,
- .pool_size = xdp->batch_size,
- .nid = NUMA_NO_NODE,
- .init_callback = xdp_test_run_init_page,
- .init_arg = xdp,
- };
xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
if (!xdp->frames)
@@ -177,34 +141,21 @@ static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_c
if (!xdp->skbs)
goto err_skbs;
- pp = page_pool_create(&pp_params);
- if (IS_ERR(pp)) {
- err = PTR_ERR(pp);
- goto err_pp;
- }
-
- /* will copy 'mem.id' into pp->xdp_mem_id */
- err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
- if (err)
- goto err_mmodel;
-
- xdp->pp = pp;
-
/* We create a 'fake' RXQ referencing the original dev, but with an
* xdp_mem_info pointing to our page_pool
*/
xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
- xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
- xdp->rxq.mem.id = pp->xdp_mem_id;
+ xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL; /* mem id is set per-frame below */
xdp->dev = orig_ctx->rxq->dev;
xdp->orig_ctx = orig_ctx;
+ /* We need a random cookie for each run as pages can stick around
+ * between runs in the system page pool
+ */
+ get_random_bytes(&xdp->cookie, sizeof(xdp->cookie));
+
return 0;
-err_mmodel:
- page_pool_destroy(pp);
-err_pp:
- kvfree(xdp->skbs);
err_skbs:
kvfree(xdp->frames);
return err;
@@ -212,8 +163,6 @@ static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_c
static void xdp_test_run_teardown(struct xdp_test_data *xdp)
{
- xdp_unreg_mem_model(&xdp->mem);
- page_pool_destroy(xdp->pp);
kfree(xdp->frames);
kfree(xdp->skbs);
}
@@ -235,8 +184,12 @@ static bool ctx_was_changed(struct xdp_page_head *head)
head->orig_ctx.data_end != head->ctx.data_end;
}
-static void reset_ctx(struct xdp_page_head *head)
+static void reset_ctx(struct xdp_page_head *head, struct xdp_test_data *xdp)
{
+ /* mem id can change if we migrate CPUs between batches */
+ if (head->frame->mem.id != xdp->rxq.mem.id)
+ head->frame->mem.id = xdp->rxq.mem.id;
+
if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
return;
@@ -246,6 +199,48 @@ static void reset_ctx(struct xdp_page_head *head)
xdp_update_frame_from_buff(&head->ctx, head->frame);
}
+static struct xdp_page_head *
+xdp_test_run_init_page(struct page *page, struct xdp_test_data *xdp)
+{
+ struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
+ struct xdp_buff *new_ctx, *orig_ctx;
+ u32 headroom = XDP_PACKET_HEADROOM;
+ size_t frm_len, meta_len;
+ struct xdp_frame *frm;
+ void *data;
+
+ /* Optimise for the recycle case, which is the normal case when doing
+ * high-repetition REDIRECTS to drivers that return frames.
+ */
+ if (likely(head->cookie == xdp->cookie)) {
+ reset_ctx(head, xdp);
+ return head;
+ }
+
+ head->cookie = xdp->cookie;
+
+ orig_ctx = xdp->orig_ctx;
+ frm_len = orig_ctx->data_end - orig_ctx->data_meta;
+ meta_len = orig_ctx->data - orig_ctx->data_meta;
+ headroom -= meta_len;
+
+ new_ctx = &head->ctx;
+ frm = head->frame;
+ data = head->data;
+ memcpy(data + headroom, orig_ctx->data_meta, frm_len);
+
+ xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
+ xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
+ new_ctx->data = new_ctx->data_meta + meta_len;
+
+ xdp_update_frame_from_buff(new_ctx, frm);
+ frm->mem = new_ctx->rxq->mem;
+
+ memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
+
+ return head;
+}
+
static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
struct sk_buff **skbs,
struct net_device *dev)
@@ -287,6 +282,7 @@ static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
struct xdp_page_head *head;
struct xdp_frame *frm;
bool redirect = false;
+ struct page_pool *pp;
struct xdp_buff *ctx;
struct page *page;
@@ -295,15 +291,17 @@ static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
local_bh_disable();
xdp_set_return_frame_no_direct();
+ pp = this_cpu_read(system_page_pool);
+ xdp->rxq.mem.id = pp->xdp_mem_id;
+
for (i = 0; i < batch_sz; i++) {
- page = page_pool_dev_alloc_pages(xdp->pp);
+ page = page_pool_dev_alloc_pages(pp);
if (!page) {
err = -ENOMEM;
goto out;
}
- head = phys_to_virt(page_to_phys(page));
- reset_ctx(head);
+ head = xdp_test_run_init_page(page, xdp);
ctx = &head->ctx;
frm = head->frame;
xdp->frame_cnt++;
--
2.43.0
