From: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
To: sagi@grimberg.me, hch@lst.de
Cc: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>,
linux-nvme@lists.infradead.org
Subject: [RFC PATCH 2/2] nvmet: add file-ns polling support
Date: Mon, 9 Dec 2019 22:25:57 -0800 [thread overview]
Message-ID: <20191210062557.5171-3-chaitanya.kulkarni@wdc.com> (raw)
In-Reply-To: <20191210062557.5171-1-chaitanya.kulkarni@wdc.com>
This patch adds polling support for file-ns. We only enable polling
based on underlying fs support and use_poll attr. By default we don't
poll on any operation we poll on the request with Reqd/write when we
receive -EIOCBQUEUED i.e. for direct I/O.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
---
drivers/nvme/target/core.c | 4 +++
drivers/nvme/target/io-cmd-file.c | 60 +++++++++++++++++++++++++++----
drivers/nvme/target/nvmet.h | 9 +++--
3 files changed, 63 insertions(+), 10 deletions(-)
diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
index d8f9130d1cd1..cd2f5c6f896e 100644
--- a/drivers/nvme/target/core.c
+++ b/drivers/nvme/target/core.c
@@ -514,12 +514,16 @@ inline void nvmet_req_done(struct nvmet_req *req)
{
if (req->ns->bdev)
nvmet_bdev_req_complete(req);
+ if (req->ns->file)
+ nvmet_file_req_complete(req);
}
inline void nvmet_req_poll_complete(struct nvmet_req *req)
{
if (req->ns->bdev)
nvmet_bdev_poll_complete(req);
+ if (req->ns->file)
+ nvmet_file_poll_complete(req);
}
int nvmet_ns_enable(struct nvmet_ns *ns)
diff --git a/drivers/nvme/target/io-cmd-file.c b/drivers/nvme/target/io-cmd-file.c
index caebfce06605..1f49a02fd437 100644
--- a/drivers/nvme/target/io-cmd-file.c
+++ b/drivers/nvme/target/io-cmd-file.c
@@ -16,6 +16,7 @@
void nvmet_file_ns_disable(struct nvmet_ns *ns)
{
if (ns->file) {
+ ns->poll ? nvmet_ns_stop_poll(ns) : 0;
if (ns->buffered_io)
flush_workqueue(buffered_io_wq);
mempool_destroy(ns->bvec_pool);
@@ -72,6 +73,11 @@ int nvmet_file_ns_enable(struct nvmet_ns *ns)
goto err;
}
+ ns->poll = ns->use_poll && ns->file->f_op->iopoll;
+ ret = ns->poll ? nvmet_ns_start_poll(ns) : 0;
+ if (ret)
+ goto err;
+
return ret;
err:
ns->size = 0;
@@ -114,9 +120,8 @@ static ssize_t nvmet_file_submit_bvec(struct nvmet_req *req, loff_t pos,
return call_iter(iocb, &iter);
}
-static void nvmet_file_io_done(struct kiocb *iocb, long ret, long ret2)
+void nvmet_file_req_complete(struct nvmet_req *req)
{
- struct nvmet_req *req = container_of(iocb, struct nvmet_req, f.iocb);
u16 status = NVME_SC_SUCCESS;
if (req->f.bvec != req->inline_bvec) {
@@ -126,13 +131,39 @@ static void nvmet_file_io_done(struct kiocb *iocb, long ret, long ret2)
mempool_free(req->f.bvec, req->ns->bvec_pool);
}
- if (unlikely(ret != req->transfer_len))
- status = errno_to_nvme_status(req, ret);
+ if (unlikely(req->f.iosize != req->transfer_len))
+ status = errno_to_nvme_status(req, req->f.iosize);
nvmet_req_complete(req, status);
}
+static void nvmet_file_io_done(struct kiocb *iocb, long ret, long ret2)
+{
+ struct nvmet_req *req = container_of(iocb, struct nvmet_req, f.iocb);
+
+ req->f.iosize = ret;
+ req->poll ? complete(&req->wait) : nvmet_file_req_complete(req);
+}
+
+void nvmet_file_poll_complete(struct nvmet_req *req)
+{
+ while (!completion_done(&req->wait)) {
+ int ret = req->f.iocb.ki_filp->f_op->iopoll(&req->f.iocb, true);
+
+ if (ret < 0)
+ pr_err("tid %d poll error %d", req->t->id, ret);
+ }
+ /*
+ * We are out of the lock anyway, by completing the polled request here
+ * we reduce lock contention and decrease the size of done list which
+ * reduces the size of list_lock. This allows nvmet_file_execute_rw()
+ * to make progress as and when we scheduled out.
+ */
+ nvmet_file_req_complete(req);
+}
+
static bool nvmet_file_execute_io(struct nvmet_req *req, int ki_flags)
{
+ struct kiocb *iocb = &req->f.iocb;
ssize_t nr_bvec = req->sg_cnt;
unsigned long bv_cnt = 0;
bool is_sync = false;
@@ -151,7 +182,7 @@ static bool nvmet_file_execute_io(struct nvmet_req *req, int ki_flags)
return true;
}
- memset(&req->f.iocb, 0, sizeof(struct kiocb));
+ memset(iocb, 0, sizeof(struct kiocb));
for_each_sg(req->sg, sg, req->sg_cnt, i) {
nvmet_file_init_bvec(&req->f.bvec[bv_cnt], sg);
len += req->f.bvec[bv_cnt].bv_len;
@@ -187,13 +218,20 @@ static bool nvmet_file_execute_io(struct nvmet_req *req, int ki_flags)
* A NULL ki_complete ask for synchronous execution, which we want
* for the IOCB_NOWAIT case.
*/
- if (!(ki_flags & IOCB_NOWAIT))
- req->f.iocb.ki_complete = nvmet_file_io_done;
+ if (!(ki_flags & IOCB_NOWAIT)) {
+ iocb->ki_complete = nvmet_file_io_done;
+ ki_flags |= req->ns->poll ? IOCB_HIPRI : 0;
+ }
ret = nvmet_file_submit_bvec(req, pos, bv_cnt, total_len, ki_flags);
switch (ret) {
case -EIOCBQUEUED:
+ if (req->ns->poll) {
+ req->poll = true;
+ nvmet_req_prep_poll(req);
+ nvmet_req_issue_poll(req);
+ }
return true;
case -EAGAIN:
if (WARN_ON_ONCE(!(ki_flags & IOCB_NOWAIT)))
@@ -211,6 +249,10 @@ static bool nvmet_file_execute_io(struct nvmet_req *req, int ki_flags)
}
complete:
+ /*
+ * If we are here, then I/O is synchronously completed and ret holds
+ * number of bytes transferred.
+ */
nvmet_file_io_done(&req->f.iocb, ret, 0);
return true;
}
@@ -379,15 +421,19 @@ u16 nvmet_file_parse_io_cmd(struct nvmet_req *req)
switch (cmd->common.opcode) {
case nvme_cmd_read:
case nvme_cmd_write:
+ req->poll = false;
req->execute = nvmet_file_execute_rw;
return 0;
case nvme_cmd_flush:
+ req->poll = false;
req->execute = nvmet_file_execute_flush;
return 0;
case nvme_cmd_dsm:
+ req->poll = false;
req->execute = nvmet_file_execute_dsm;
return 0;
case nvme_cmd_write_zeroes:
+ req->poll = false;
req->execute = nvmet_file_execute_write_zeroes;
return 0;
default:
diff --git a/drivers/nvme/target/nvmet.h b/drivers/nvme/target/nvmet.h
index ef2919e23e0b..e7e0e0de705e 100644
--- a/drivers/nvme/target/nvmet.h
+++ b/drivers/nvme/target/nvmet.h
@@ -317,9 +317,10 @@ struct nvmet_req {
} b;
struct {
bool mpool_alloc;
- struct kiocb iocb;
- struct bio_vec *bvec;
- struct work_struct work;
+ struct kiocb iocb;
+ struct bio_vec *bvec;
+ struct work_struct work;
+ long iosize;
} f;
};
int sg_cnt;
@@ -469,6 +470,8 @@ void nvmet_req_issue_poll(struct nvmet_req *req);
void nvmet_req_poll_complete(struct nvmet_req *req);
void nvmet_bdev_poll_complete(struct nvmet_req *req);
void nvmet_bdev_req_complete(struct nvmet_req *req);
+void nvmet_file_poll_complete(struct nvmet_req *req);
+void nvmet_file_req_complete(struct nvmet_req *req);
void nvmet_req_done(struct nvmet_req *req);
#define NVMET_QUEUE_SIZE 1024
--
2.22.1
_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme
next prev parent reply other threads:[~2019-12-10 6:26 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-12-10 6:25 [RFC PATCH 0/2] nvmet: add polling support Chaitanya Kulkarni
2019-12-10 6:25 ` [RFC PATCH 1/2] nvmet: add bdev-ns " Chaitanya Kulkarni
2020-01-20 12:52 ` Max Gurtovoy
2020-01-21 19:22 ` Chaitanya Kulkarni
2020-01-23 14:23 ` Max Gurtovoy
2020-01-30 18:19 ` Chaitanya Kulkarni
2019-12-10 6:25 ` Chaitanya Kulkarni [this message]
2019-12-12 1:01 ` [RFC PATCH 0/2] nvmet: add " Sagi Grimberg
2019-12-12 5:44 ` Chaitanya Kulkarni
2019-12-12 20:32 ` Sagi Grimberg
2020-01-20 5:13 ` Chaitanya Kulkarni
2020-01-20 4:48 ` Chaitanya Kulkarni
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20191210062557.5171-3-chaitanya.kulkarni@wdc.com \
--to=chaitanya.kulkarni@wdc.com \
--cc=hch@lst.de \
--cc=linux-nvme@lists.infradead.org \
--cc=sagi@grimberg.me \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).