From: Selvin Xavier <selvin.xavier@broadcom.com>
To: jgg@ziepe.ca, leon@kernel.org
Cc: linux-rdma@vger.kernel.org, andrew.gospodarek@broadcom.com,
Kashyap Desai <kashyap.desai@broadcom.com>,
Selvin Xavier <selvin.xavier@broadcom.com>
Subject: [PATCH for-next 07/17] RDMA/bnxt_re: use shadow qd while posting non blocking rcfw command
Date: Thu, 8 Jun 2023 03:24:58 -0700 [thread overview]
Message-ID: <1686219908-11181-8-git-send-email-selvin.xavier@broadcom.com> (raw)
In-Reply-To: <1686219908-11181-1-git-send-email-selvin.xavier@broadcom.com>
[-- Attachment #1: Type: text/plain, Size: 4535 bytes --]
From: Kashyap Desai <kashyap.desai@broadcom.com>
Whenever there is a fast path IO and create/destroy resources from
the slow path is happening in parallel, we may notice high latency
of slow path command completion.
Introduces a shadow queue depth to prevent the outstanding requests
to the FW. Driver will not allow more than #RCFW_CMD_NON_BLOCKING_SHADOW_QD
non-blocking commands to the Firmware.
Shadow queue depth is a soft limit only for non-blocking
commands. Blocking commands will be posted to the firmware
as long as there is a free slot.
Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
drivers/infiniband/hw/bnxt_re/qplib_rcfw.c | 60 +++++++++++++++++++++++++++++-
drivers/infiniband/hw/bnxt_re/qplib_rcfw.h | 3 ++
2 files changed, 61 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
index 3c4f72a..f7d1238 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
@@ -281,8 +281,21 @@ static int __send_message(struct bnxt_qplib_rcfw *rcfw,
return 0;
}
-int bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
- struct bnxt_qplib_cmdqmsg *msg)
+/**
+ * __bnxt_qplib_rcfw_send_message - qplib interface to send
+ * and complete rcfw command.
+ * @rcfw - rcfw channel instance of rdev
+ * @msg - qplib message internal
+ *
+ * This function does not account shadow queue depth. It will send
+ * all the command unconditionally as long as send queue is not full.
+ *
+ * Returns:
+ * 0 if command completed by firmware.
+ * Non zero if the command is not completed by firmware.
+ */
+static int __bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
+ struct bnxt_qplib_cmdqmsg *msg)
{
struct creq_qp_event *evnt = (struct creq_qp_event *)msg->resp;
u16 cookie;
@@ -331,6 +344,48 @@ int bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
return rc;
}
+
+/**
+ * bnxt_qplib_rcfw_send_message - qplib interface to send
+ * and complete rcfw command.
+ * @rcfw - rcfw channel instance of rdev
+ * @msg - qplib message internal
+ *
+ * Driver interact with Firmware through rcfw channel/slow path in two ways.
+ * a. Blocking rcfw command send. In this path, driver cannot hold
+ * the context for longer period since it is holding cpu until
+ * command is not completed.
+ * b. Non-blocking rcfw command send. In this path, driver can hold the
+ * context for longer period. There may be many pending command waiting
+ * for completion because of non-blocking nature.
+ *
+ * Driver will use shadow queue depth. Current queue depth of 8K
+ * (due to size of rcfw message there can be actual ~4K rcfw outstanding)
+ * is not optimal for rcfw command processing in firmware.
+ *
+ * Restrict at max #RCFW_CMD_NON_BLOCKING_SHADOW_QD Non-Blocking rcfw commands.
+ * Allow all blocking commands until there is no queue full.
+ *
+ * Returns:
+ * 0 if command completed by firmware.
+ * Non zero if the command is not completed by firmware.
+ */
+int bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
+ struct bnxt_qplib_cmdqmsg *msg)
+{
+ int ret;
+
+ if (!msg->block) {
+ down(&rcfw->rcfw_inflight);
+ ret = __bnxt_qplib_rcfw_send_message(rcfw, msg);
+ up(&rcfw->rcfw_inflight);
+ } else {
+ ret = __bnxt_qplib_rcfw_send_message(rcfw, msg);
+ }
+
+ return ret;
+}
+
/* Completions */
static int bnxt_qplib_process_func_event(struct bnxt_qplib_rcfw *rcfw,
struct creq_func_event *func_event)
@@ -932,6 +987,7 @@ int bnxt_qplib_enable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw,
return rc;
}
+ sema_init(&rcfw->rcfw_inflight, RCFW_CMD_NON_BLOCKING_SHADOW_QD);
bnxt_qplib_start_rcfw(rcfw);
return 0;
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
index 32e5b67..862bfbf 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.h
@@ -66,6 +66,8 @@ static inline void bnxt_qplib_rcfw_cmd_prep(struct cmdq_base *req,
req->cmd_size = cmd_size;
}
+/* Shadow queue depth for non blocking command */
+#define RCFW_CMD_NON_BLOCKING_SHADOW_QD 64
#define RCFW_CMD_WAIT_TIME_MS 20000 /* 20 Seconds timeout */
/* CMDQ elements */
@@ -197,6 +199,7 @@ struct bnxt_qplib_rcfw {
u64 oos_prev;
u32 init_oos_stats;
u32 cmdq_depth;
+ struct semaphore rcfw_inflight;
};
struct bnxt_qplib_cmdqmsg {
--
2.5.5
[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4224 bytes --]
next prev parent reply other threads:[~2023-06-08 10:37 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-06-08 10:24 [PATCH for-next 00/17] RDMA/bnxt_re: Control path updates Selvin Xavier
2023-06-08 10:24 ` [PATCH for-next 01/17] RDMA/bnxt_re: wraparound mbox producer index Selvin Xavier
2023-06-08 10:24 ` [PATCH for-next 02/17] RDMA/bnxt_re: Avoid calling wake_up threads from spin_lock context Selvin Xavier
2023-06-08 10:24 ` [PATCH for-next 03/17] RDMA/bnxt_re: remove virt_func check while creating RoCE FW channel Selvin Xavier
2023-06-08 10:24 ` [PATCH for-next 04/17] RDMA/bnxt_re: set fixed command queue depth Selvin Xavier
2023-06-08 10:24 ` [PATCH for-next 05/17] RDMA/bnxt_re: Enhance the existing functions that wait for FW responses Selvin Xavier
2023-06-08 10:24 ` [PATCH for-next 06/17] RDMA/bnxt_re: Avoid the command wait if firmware is inactive Selvin Xavier
2023-06-08 10:24 ` Selvin Xavier [this message]
2023-06-08 10:24 ` [PATCH for-next 08/17] RDMA/bnxt_re: Simplify the function that sends the FW commands Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 09/17] RDMA/bnxt_re: add helper function __poll_for_resp Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 10/17] RDMA/bnxt_re: handle command completions after driver detect a timedout Selvin Xavier
2023-06-08 12:53 ` kernel test robot
2023-06-08 10:25 ` [PATCH for-next 11/17] RDMA/bnxt_re: Add firmware stall check detection Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 12/17] RDMA/bnxt_re: post destroy_ah for delayed completion of AH creation Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 13/17] RDMA/bnxt_re: consider timeout of destroy ah as success Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 14/17] RDMA/bnxt_re: cancel all control path command waiters upon error Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 15/17] RDMA/bnxt_re: use firmware provided max request timeout Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 16/17] RDMA/bnxt_re: remove redundant cmdq_bitmap Selvin Xavier
2023-06-08 10:25 ` [PATCH for-next 17/17] RDMA/bnxt_re: optimize the parameters passed to helper functions Selvin Xavier
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=1686219908-11181-8-git-send-email-selvin.xavier@broadcom.com \
--to=selvin.xavier@broadcom.com \
--cc=andrew.gospodarek@broadcom.com \
--cc=jgg@ziepe.ca \
--cc=kashyap.desai@broadcom.com \
--cc=leon@kernel.org \
--cc=linux-rdma@vger.kernel.org \
/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