linux/drivers/net/ethernet/intel/ice/ice_controlq.h
Michal Schmidt f86d6f9c49 ice: sleep, don't busy-wait, for ICE_CTL_Q_SQ_CMD_TIMEOUT
The driver polls for ice_sq_done() with a 100 µs period for up to 1 s
and it uses udelay to do that.

Let's use usleep_range instead. We know sleeping is allowed here,
because we're holding a mutex (cq->sq_lock). To preserve the total
max waiting time, measure the timeout in jiffies.

ICE_CTL_Q_SQ_CMD_TIMEOUT is used also in ice_release_res(), but there
the polling period is 1 ms (i.e. 10 times longer). Since the timeout was
expressed in terms of the number of loops, the total timeout in this
function is 10 s. I do not know if this is intentional. This patch keeps
it.

The patch lowers the CPU usage of the ice-gnss-<dev_name> kernel thread
on my system from ~8 % to less than 1 %.

I received a report of high CPU usage with ptp4l where the busy-waiting
in ice_sq_send_cmd dominated the profile. This patch has been tested in
that usecase too and it made a huge improvement there.

Tested-by: Brent Rowsell <browsell@redhat.com>
Signed-off-by: Michal Schmidt <mschmidt@redhat.com>
Reviewed-by: Arkadiusz Kubalewski <arkadiusz.kubalewski@intel.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Tested-by: Sunitha Mekala <sunithax.d.mekala@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
2023-04-20 16:33:14 -07:00

100 lines
2.6 KiB
C

/* SPDX-License-Identifier: GPL-2.0 */
/* Copyright (c) 2018, Intel Corporation. */
#ifndef _ICE_CONTROLQ_H_
#define _ICE_CONTROLQ_H_
#include "ice_adminq_cmd.h"
/* Maximum buffer lengths for all control queue types */
#define ICE_AQ_MAX_BUF_LEN 4096
#define ICE_MBXQ_MAX_BUF_LEN 4096
#define ICE_SBQ_MAX_BUF_LEN 512
#define ICE_CTL_Q_DESC(R, i) \
(&(((struct ice_aq_desc *)((R).desc_buf.va))[i]))
#define ICE_CTL_Q_DESC_UNUSED(R) \
((u16)((((R)->next_to_clean > (R)->next_to_use) ? 0 : (R)->count) + \
(R)->next_to_clean - (R)->next_to_use - 1))
/* Defines that help manage the driver vs FW API checks.
* Take a look at ice_aq_ver_check in ice_controlq.c for actual usage.
*/
#define EXP_FW_API_VER_BRANCH 0x00
#define EXP_FW_API_VER_MAJOR 0x01
#define EXP_FW_API_VER_MINOR 0x05
/* Different control queue types: These are mainly for SW consumption. */
enum ice_ctl_q {
ICE_CTL_Q_UNKNOWN = 0,
ICE_CTL_Q_ADMIN,
ICE_CTL_Q_MAILBOX,
ICE_CTL_Q_SB,
};
/* Control Queue timeout settings - max delay 1s */
#define ICE_CTL_Q_SQ_CMD_TIMEOUT HZ /* Wait max 1s */
#define ICE_CTL_Q_SQ_CMD_USEC 100 /* Check every 100usec */
#define ICE_CTL_Q_ADMIN_INIT_TIMEOUT 10 /* Count 10 times */
#define ICE_CTL_Q_ADMIN_INIT_MSEC 100 /* Check every 100msec */
struct ice_ctl_q_ring {
void *dma_head; /* Virtual address to DMA head */
struct ice_dma_mem desc_buf; /* descriptor ring memory */
void *cmd_buf; /* command buffer memory */
union {
struct ice_dma_mem *sq_bi;
struct ice_dma_mem *rq_bi;
} r;
u16 count; /* Number of descriptors */
/* used for interrupt processing */
u16 next_to_use;
u16 next_to_clean;
/* used for queue tracking */
u32 head;
u32 tail;
u32 len;
u32 bah;
u32 bal;
u32 len_mask;
u32 len_ena_mask;
u32 len_crit_mask;
u32 head_mask;
};
/* sq transaction details */
struct ice_sq_cd {
struct ice_aq_desc *wb_desc;
};
#define ICE_CTL_Q_DETAILS(R, i) (&(((struct ice_sq_cd *)((R).cmd_buf))[i]))
/* rq event information */
struct ice_rq_event_info {
struct ice_aq_desc desc;
u16 msg_len;
u16 buf_len;
u8 *msg_buf;
};
/* Control Queue information */
struct ice_ctl_q_info {
enum ice_ctl_q qtype;
struct ice_ctl_q_ring rq; /* receive queue */
struct ice_ctl_q_ring sq; /* send queue */
u16 num_rq_entries; /* receive queue depth */
u16 num_sq_entries; /* send queue depth */
u16 rq_buf_size; /* receive queue buffer size */
u16 sq_buf_size; /* send queue buffer size */
enum ice_aq_err sq_last_status; /* last status on send queue */
struct mutex sq_lock; /* Send queue lock */
struct mutex rq_lock; /* Receive queue lock */
};
#endif /* _ICE_CONTROLQ_H_ */