* [PATCH liburing 1/3] examples: remove zcrx size limiting
2025-04-20 11:24 [PATCH liburing 0/3] zcrx refill queue allocation modes Pavel Begunkov
@ 2025-04-20 11:24 ` Pavel Begunkov
2025-04-21 0:22 ` David Wei
2025-04-20 11:24 ` [PATCH liburing 2/3] examples/zcrx: constants for request types Pavel Begunkov
2025-04-20 11:24 ` [PATCH liburing 3/3] examples/zcrx: add refill queue allocation modes Pavel Begunkov
2 siblings, 1 reply; 10+ messages in thread
From: Pavel Begunkov @ 2025-04-20 11:24 UTC (permalink / raw)
To: io-uring; +Cc: asml.silence, David Wei
It's not handled too well and is not great at show casing the feature,
remove it for now, we may add it back later in a different form.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
examples/zcrx.c | 34 ++++------------------------------
1 file changed, 4 insertions(+), 30 deletions(-)
diff --git a/examples/zcrx.c b/examples/zcrx.c
index 32d9e4ae..afc05642 100644
--- a/examples/zcrx.c
+++ b/examples/zcrx.c
@@ -46,8 +46,6 @@ static long page_size;
static int cfg_port = 8000;
static const char *cfg_ifname;
static int cfg_queue_id = -1;
-static bool cfg_oneshot;
-static int cfg_oneshot_recvs;
static bool cfg_verify_data = false;
static struct sockaddr_in6 cfg_addr;
@@ -150,16 +148,6 @@ static void add_recvzc(struct io_uring *ring, int sockfd)
sqe->user_data = 2;
}
-static void add_recvzc_oneshot(struct io_uring *ring, int sockfd, size_t len)
-{
- struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
-
- io_uring_prep_rw(IORING_OP_RECV_ZC, sqe, sockfd, NULL, len, 0);
- sqe->ioprio |= IORING_RECV_MULTISHOT;
- sqe->zcrx_ifq_idx = zcrx_id;
- sqe->user_data = 2;
-}
-
static void process_accept(struct io_uring *ring, struct io_uring_cqe *cqe)
{
if (cqe->res < 0)
@@ -168,10 +156,7 @@ static void process_accept(struct io_uring *ring, struct io_uring_cqe *cqe)
t_error(1, 0, "Unexpected second connection");
connfd = cqe->res;
- if (cfg_oneshot)
- add_recvzc_oneshot(ring, connfd, page_size);
- else
- add_recvzc(ring, connfd);
+ add_recvzc(ring, connfd);
}
static void verify_data(char *data, size_t size, unsigned long seq)
@@ -200,19 +185,13 @@ static void process_recvzc(struct io_uring *ring, struct io_uring_cqe *cqe)
if (cqe->res < 0)
t_error(1, 0, "recvzc(): %d", cqe->res);
- if (cqe->res == 0 && cqe->flags == 0 && cfg_oneshot_recvs == 0) {
+ if (cqe->res == 0 && cqe->flags == 0) {
stop = true;
return;
}
- if (cfg_oneshot) {
- if (cqe->res == 0 && cqe->flags == 0 && cfg_oneshot_recvs) {
- add_recvzc_oneshot(ring, connfd, page_size);
- cfg_oneshot_recvs--;
- }
- } else if (!(cqe->flags & IORING_CQE_F_MORE)) {
+ if (!(cqe->flags & IORING_CQE_F_MORE))
add_recvzc(ring, connfd);
- }
rcqe = (struct io_uring_zcrx_cqe *)(cqe + 1);
mask = (1ULL << IORING_ZCRX_AREA_SHIFT) - 1;
@@ -299,7 +278,7 @@ static void parse_opts(int argc, char **argv)
if (argc <= 1)
usage(argv[0]);
- while ((c = getopt(argc, argv, "vp:i:q:o:")) != -1) {
+ while ((c = getopt(argc, argv, "vp:i:q:")) != -1) {
switch (c) {
case 'p':
cfg_port = strtoul(optarg, NULL, 0);
@@ -307,11 +286,6 @@ static void parse_opts(int argc, char **argv)
case 'i':
cfg_ifname = optarg;
break;
- case 'o': {
- cfg_oneshot = true;
- cfg_oneshot_recvs = strtoul(optarg, NULL, 0);
- break;
- }
case 'q':
cfg_queue_id = strtoul(optarg, NULL, 0);
break;
--
2.48.1
^ permalink raw reply related [flat|nested] 10+ messages in thread
* Re: [PATCH liburing 1/3] examples: remove zcrx size limiting
2025-04-20 11:24 ` [PATCH liburing 1/3] examples: remove zcrx size limiting Pavel Begunkov
@ 2025-04-21 0:22 ` David Wei
2025-04-21 6:40 ` Pavel Begunkov
0 siblings, 1 reply; 10+ messages in thread
From: David Wei @ 2025-04-21 0:22 UTC (permalink / raw)
To: Pavel Begunkov, io-uring
On 2025-04-20 04:24, Pavel Begunkov wrote:
> It's not handled too well and is not great at show casing the feature,
> remove it for now, we may add it back later in a different form.
>
> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
> ---
> examples/zcrx.c | 34 ++++------------------------------
> 1 file changed, 4 insertions(+), 30 deletions(-)
>
I'm relying on this selftest. Can the deletion wait until the
replacement is ready?
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH liburing 1/3] examples: remove zcrx size limiting
2025-04-21 0:22 ` David Wei
@ 2025-04-21 6:40 ` Pavel Begunkov
2025-04-21 17:38 ` David Wei
0 siblings, 1 reply; 10+ messages in thread
From: Pavel Begunkov @ 2025-04-21 6:40 UTC (permalink / raw)
To: David Wei, io-uring
On 4/21/25 01:22, David Wei wrote:
> On 2025-04-20 04:24, Pavel Begunkov wrote:
>> It's not handled too well and is not great at show casing the feature,
>> remove it for now, we may add it back later in a different form.
>>
>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>> ---
>> examples/zcrx.c | 34 ++++------------------------------
>> 1 file changed, 4 insertions(+), 30 deletions(-)
>>
>
> I'm relying on this selftest. Can the deletion wait until the
> replacement is ready?
I assumed you're using the selftest, and not this liburing/examples/*
Anyway, let's see what I can do with it
--
Pavel Begunkov
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH liburing 1/3] examples: remove zcrx size limiting
2025-04-21 6:40 ` Pavel Begunkov
@ 2025-04-21 17:38 ` David Wei
0 siblings, 0 replies; 10+ messages in thread
From: David Wei @ 2025-04-21 17:38 UTC (permalink / raw)
To: Pavel Begunkov, io-uring
On 2025-04-20 23:40, Pavel Begunkov wrote:
> On 4/21/25 01:22, David Wei wrote:
>> On 2025-04-20 04:24, Pavel Begunkov wrote:
>>> It's not handled too well and is not great at show casing the feature,
>>> remove it for now, we may add it back later in a different form.
>>>
>>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>>> ---
>>> examples/zcrx.c | 34 ++++------------------------------
>>> 1 file changed, 4 insertions(+), 30 deletions(-)
>>>
>>
>> I'm relying on this selftest. Can the deletion wait until the
>> replacement is ready?
>
> I assumed you're using the selftest, and not this liburing/examples/*
>
> Anyway, let's see what I can do with it
>
Oh my mistake :facepalm: I got this mixed up with the selfteset.
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH liburing 2/3] examples/zcrx: constants for request types
2025-04-20 11:24 [PATCH liburing 0/3] zcrx refill queue allocation modes Pavel Begunkov
2025-04-20 11:24 ` [PATCH liburing 1/3] examples: remove zcrx size limiting Pavel Begunkov
@ 2025-04-20 11:24 ` Pavel Begunkov
2025-04-21 0:25 ` David Wei
2025-04-20 11:24 ` [PATCH liburing 3/3] examples/zcrx: add refill queue allocation modes Pavel Begunkov
2 siblings, 1 reply; 10+ messages in thread
From: Pavel Begunkov @ 2025-04-20 11:24 UTC (permalink / raw)
To: io-uring; +Cc: asml.silence, David Wei
Instead of hard coding user_data, name request types we need and use
them.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
examples/zcrx.c | 22 +++++++++++++++++-----
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/examples/zcrx.c b/examples/zcrx.c
index afc05642..8989c9a4 100644
--- a/examples/zcrx.c
+++ b/examples/zcrx.c
@@ -43,6 +43,14 @@ static long page_size;
#define AREA_SIZE (8192 * page_size)
#define SEND_SIZE (512 * 4096)
+#define REQ_TYPE_SHIFT 3
+#define REQ_TYPE_MASK ((1UL << REQ_TYPE_SHIFT) - 1)
+
+enum request_type {
+ REQ_TYPE_ACCEPT = 1,
+ REQ_TYPE_RX = 2,
+};
+
static int cfg_port = 8000;
static const char *cfg_ifname;
static int cfg_queue_id = -1;
@@ -135,7 +143,7 @@ static void add_accept(struct io_uring *ring, int sockfd)
struct io_uring_sqe *sqe = io_uring_get_sqe(ring);
io_uring_prep_accept(sqe, sockfd, NULL, NULL, 0);
- sqe->user_data = 1;
+ sqe->user_data = REQ_TYPE_ACCEPT;
}
static void add_recvzc(struct io_uring *ring, int sockfd)
@@ -145,7 +153,7 @@ static void add_recvzc(struct io_uring *ring, int sockfd)
io_uring_prep_rw(IORING_OP_RECV_ZC, sqe, sockfd, NULL, 0, 0);
sqe->ioprio |= IORING_RECV_MULTISHOT;
sqe->zcrx_ifq_idx = zcrx_id;
- sqe->user_data = 2;
+ sqe->user_data = REQ_TYPE_RX;
}
static void process_accept(struct io_uring *ring, struct io_uring_cqe *cqe)
@@ -215,12 +223,16 @@ static void server_loop(struct io_uring *ring)
io_uring_submit_and_wait(ring, 1);
io_uring_for_each_cqe(ring, head, cqe) {
- if (cqe->user_data == 1)
+ switch (cqe->user_data & REQ_TYPE_MASK) {
+ case REQ_TYPE_ACCEPT:
process_accept(ring, cqe);
- else if (cqe->user_data == 2)
+ break;
+ case REQ_TYPE_RX:
process_recvzc(ring, cqe);
- else
+ break;
+ default:
t_error(1, 0, "unknown cqe");
+ }
count++;
}
io_uring_cq_advance(ring, count);
--
2.48.1
^ permalink raw reply related [flat|nested] 10+ messages in thread
* [PATCH liburing 3/3] examples/zcrx: add refill queue allocation modes
2025-04-20 11:24 [PATCH liburing 0/3] zcrx refill queue allocation modes Pavel Begunkov
2025-04-20 11:24 ` [PATCH liburing 1/3] examples: remove zcrx size limiting Pavel Begunkov
2025-04-20 11:24 ` [PATCH liburing 2/3] examples/zcrx: constants for request types Pavel Begunkov
@ 2025-04-20 11:24 ` Pavel Begunkov
2025-04-21 0:28 ` David Wei
2 siblings, 1 reply; 10+ messages in thread
From: Pavel Begunkov @ 2025-04-20 11:24 UTC (permalink / raw)
To: io-uring; +Cc: asml.silence, David Wei
Refill queue creating is backed by the region api, which can either be
user or kernel allocated. Add an option to switch between the modes.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
examples/zcrx.c | 46 ++++++++++++++++++++++++++++++++++++----------
1 file changed, 36 insertions(+), 10 deletions(-)
diff --git a/examples/zcrx.c b/examples/zcrx.c
index 8989c9a4..eafe1969 100644
--- a/examples/zcrx.c
+++ b/examples/zcrx.c
@@ -39,6 +39,13 @@
#include "liburing.h"
#include "helpers.h"
+enum {
+ RQ_ALLOC_USER,
+ RQ_ALLOC_KERNEL,
+
+ __RQ_ALLOC_MAX,
+};
+
static long page_size;
#define AREA_SIZE (8192 * page_size)
#define SEND_SIZE (512 * 4096)
@@ -55,6 +62,7 @@ static int cfg_port = 8000;
static const char *cfg_ifname;
static int cfg_queue_id = -1;
static bool cfg_verify_data = false;
+static unsigned cfg_rq_alloc_mode = RQ_ALLOC_USER;
static struct sockaddr_in6 cfg_addr;
static void *area_ptr;
@@ -79,6 +87,7 @@ static void setup_zcrx(struct io_uring *ring)
{
unsigned int ifindex;
unsigned int rq_entries = 4096;
+ unsigned rq_flags = 0;
int ret;
ifindex = if_nametoindex(cfg_ifname);
@@ -95,19 +104,22 @@ static void setup_zcrx(struct io_uring *ring)
t_error(1, 0, "mmap(): zero copy area");
ring_size = get_refill_ring_size(rq_entries);
- ring_ptr = mmap(NULL,
- ring_size,
- PROT_READ | PROT_WRITE,
- MAP_ANONYMOUS | MAP_PRIVATE,
- 0,
- 0);
- if (ring_ptr == MAP_FAILED)
- t_error(1, 0, "mmap(): refill ring");
+
+ ring_ptr = NULL;
+ if (cfg_rq_alloc_mode == RQ_ALLOC_USER) {
+ ring_ptr = mmap(NULL, ring_size,
+ PROT_READ | PROT_WRITE,
+ MAP_ANONYMOUS | MAP_PRIVATE,
+ 0, 0);
+ if (ring_ptr == MAP_FAILED)
+ t_error(1, 0, "mmap(): refill ring");
+ rq_flags |= IORING_MEM_REGION_TYPE_USER;
+ }
struct io_uring_region_desc region_reg = {
.size = ring_size,
.user_addr = (__u64)(unsigned long)ring_ptr,
- .flags = IORING_MEM_REGION_TYPE_USER,
+ .flags = rq_flags,
};
struct io_uring_zcrx_area_reg area_reg = {
@@ -128,6 +140,15 @@ static void setup_zcrx(struct io_uring *ring)
if (ret)
t_error(1, 0, "io_uring_register_ifq(): %d", ret);
+ if (cfg_rq_alloc_mode == RQ_ALLOC_USER) {
+ ring_ptr = mmap(NULL, ring_size,
+ PROT_READ | PROT_WRITE,
+ MAP_SHARED | MAP_POPULATE,
+ ring->ring_fd, region_reg.mmap_offset);
+ if (ring_ptr == MAP_FAILED)
+ t_error(1, 0, "mmap(): refill ring");
+ }
+
rq_ring.khead = (unsigned int *)((char *)ring_ptr + reg.offsets.head);
rq_ring.ktail = (unsigned int *)((char *)ring_ptr + reg.offsets.tail);
rq_ring.rqes = (struct io_uring_zcrx_rqe *)((char *)ring_ptr + reg.offsets.rqes);
@@ -290,7 +311,7 @@ static void parse_opts(int argc, char **argv)
if (argc <= 1)
usage(argv[0]);
- while ((c = getopt(argc, argv, "vp:i:q:")) != -1) {
+ while ((c = getopt(argc, argv, "vp:i:q:a:")) != -1) {
switch (c) {
case 'p':
cfg_port = strtoul(optarg, NULL, 0);
@@ -304,6 +325,11 @@ static void parse_opts(int argc, char **argv)
case 'v':
cfg_verify_data = true;
break;
+ case 'a':
+ cfg_rq_alloc_mode = strtoul(optarg, NULL, 0);
+ if (cfg_rq_alloc_mode >= __RQ_ALLOC_MAX)
+ t_error(1, 0, "invalid RQ allocation mode");
+ break;
}
}
--
2.48.1
^ permalink raw reply related [flat|nested] 10+ messages in thread
* Re: [PATCH liburing 3/3] examples/zcrx: add refill queue allocation modes
2025-04-20 11:24 ` [PATCH liburing 3/3] examples/zcrx: add refill queue allocation modes Pavel Begunkov
@ 2025-04-21 0:28 ` David Wei
2025-04-21 6:38 ` Pavel Begunkov
0 siblings, 1 reply; 10+ messages in thread
From: David Wei @ 2025-04-21 0:28 UTC (permalink / raw)
To: Pavel Begunkov, io-uring
On 2025-04-20 04:24, Pavel Begunkov wrote:
> Refill queue creating is backed by the region api, which can either be
> user or kernel allocated. Add an option to switch between the modes.
>
> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
> ---
> examples/zcrx.c | 46 ++++++++++++++++++++++++++++++++++++----------
> 1 file changed, 36 insertions(+), 10 deletions(-)
>
> diff --git a/examples/zcrx.c b/examples/zcrx.c
> index 8989c9a4..eafe1969 100644
> --- a/examples/zcrx.c
> +++ b/examples/zcrx.c
> @@ -39,6 +39,13 @@
[...]
> @@ -128,6 +140,15 @@ static void setup_zcrx(struct io_uring *ring)
> if (ret)
> t_error(1, 0, "io_uring_register_ifq(): %d", ret);
>
> + if (cfg_rq_alloc_mode == RQ_ALLOC_USER) {
cfg_rq_alloc_mode == RQ_ALLOC_KERNEL?
> + ring_ptr = mmap(NULL, ring_size,
> + PROT_READ | PROT_WRITE,
> + MAP_SHARED | MAP_POPULATE,
> + ring->ring_fd, region_reg.mmap_offset);
> + if (ring_ptr == MAP_FAILED)
> + t_error(1, 0, "mmap(): refill ring");
> + }
> +
> rq_ring.khead = (unsigned int *)((char *)ring_ptr + reg.offsets.head);
> rq_ring.ktail = (unsigned int *)((char *)ring_ptr + reg.offsets.tail);
> rq_ring.rqes = (struct io_uring_zcrx_rqe *)((char *)ring_ptr + reg.offsets.rqes);
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH liburing 3/3] examples/zcrx: add refill queue allocation modes
2025-04-21 0:28 ` David Wei
@ 2025-04-21 6:38 ` Pavel Begunkov
0 siblings, 0 replies; 10+ messages in thread
From: Pavel Begunkov @ 2025-04-21 6:38 UTC (permalink / raw)
To: David Wei, io-uring
On 4/21/25 01:28, David Wei wrote:
> On 2025-04-20 04:24, Pavel Begunkov wrote:
>> Refill queue creating is backed by the region api, which can either be
>> user or kernel allocated. Add an option to switch between the modes.
>>
>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>> ---
>> examples/zcrx.c | 46 ++++++++++++++++++++++++++++++++++++----------
>> 1 file changed, 36 insertions(+), 10 deletions(-)
>>
>> diff --git a/examples/zcrx.c b/examples/zcrx.c
>> index 8989c9a4..eafe1969 100644
>> --- a/examples/zcrx.c
>> +++ b/examples/zcrx.c
>> @@ -39,6 +39,13 @@
> [...]
>> @@ -128,6 +140,15 @@ static void setup_zcrx(struct io_uring *ring)
>> if (ret)
>> t_error(1, 0, "io_uring_register_ifq(): %d", ret);
>>
>> + if (cfg_rq_alloc_mode == RQ_ALLOC_USER) {
>
> cfg_rq_alloc_mode == RQ_ALLOC_KERNEL?
Good catch! I found it during testing but apparently forgot to
commit.
>
>> + ring_ptr = mmap(NULL, ring_size,
>> + PROT_READ | PROT_WRITE,
>> + MAP_SHARED | MAP_POPULATE,
>> + ring->ring_fd, region_reg.mmap_offset);
>> + if (ring_ptr == MAP_FAILED)
>> + t_error(1, 0, "mmap(): refill ring");
>> + }
>> +
>> rq_ring.khead = (unsigned int *)((char *)ring_ptr + reg.offsets.head);
>> rq_ring.ktail = (unsigned int *)((char *)ring_ptr + reg.offsets.tail);
>> rq_ring.rqes = (struct io_uring_zcrx_rqe *)((char *)ring_ptr + reg.offsets.rqes);
>
--
Pavel Begunkov
^ permalink raw reply [flat|nested] 10+ messages in thread