* This file may also be available under a different license from Cavium.
* Contact Cavium, Inc. for more information
**********************************************************************/
-#include <linux/version.h>
-#include <linux/types.h>
-#include <linux/list.h>
-#include <linux/interrupt.h>
#include <linux/pci.h>
-#include <linux/kthread.h>
#include <linux/netdevice.h>
#include <linux/vmalloc.h>
-#include "octeon_config.h"
#include "liquidio_common.h"
#include "octeon_droq.h"
#include "octeon_iq.h"
#include "response_manager.h"
#include "octeon_device.h"
-#include "octeon_nic.h"
#include "octeon_main.h"
#include "octeon_network.h"
-#include "cn66xx_regs.h"
#include "cn66xx_device.h"
-#include "cn68xx_regs.h"
-#include "cn68xx_device.h"
-#include "liquidio_image.h"
#define INCR_INSTRQUEUE_PKT_COUNT(octeon_dev_ptr, iq_no, field, count) \
(octeon_dev_ptr->instr_queue[iq_no]->stats.field += count)
};
static void check_db_timeout(struct work_struct *work);
-static void __check_db_timeout(struct octeon_device *oct, unsigned long iq_no);
+static void __check_db_timeout(struct octeon_device *oct, u64 iq_no);
static void (*reqtype_free_fn[MAX_OCTEON_DEVICES][REQTYPE_LAST + 1]) (void *);
/* Initialize the spinlock for this instruction queue */
spin_lock_init(&iq->lock);
+ spin_lock_init(&iq->post_lock);
- oct->io_qmask.iq |= (1 << iq_no);
+ spin_lock_init(&iq->iq_flush_running_lock);
+
+ oct->io_qmask.iq |= (1ULL << iq_no);
/* Set the 32B/64B mode for each input queue */
oct->io_qmask.iq64B |= ((conf->instr_type == 64) << iq_no);
instr_cnt = 0;
/*for (i = 0; i < oct->num_iqs; i++) {*/
- for (i = 0; i < MAX_OCTEON_INSTR_QUEUES; i++) {
- if (!(oct->io_qmask.iq & (1UL << i)))
+ for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) {
+ if (!(oct->io_qmask.iq & (1ULL << i)))
continue;
pending =
atomic_read(&oct->
memcpy(iqptr, cmd, cmdsize);
}
-static inline int
-__post_command(struct octeon_device *octeon_dev __attribute__((unused)),
- struct octeon_instr_queue *iq,
- u32 force_db __attribute__((unused)), u8 *cmd)
-{
- u32 index = -1;
-
- /* This ensures that the read index does not wrap around to the same
- * position if queue gets full before Octeon could fetch any instr.
- */
- if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 1))
- return -1;
-
- __copy_cmd_into_iq(iq, cmd);
-
- /* "index" is returned, host_write_index is modified. */
- index = iq->host_write_index;
- INCR_INDEX_BY1(iq->host_write_index, iq->max_count);
- iq->fill_cnt++;
-
- /* Flush the command into memory. We need to be sure the data is in
- * memory before indicating that the instruction is pending.
- */
- wmb();
-
- atomic_inc(&iq->instr_pending);
-
- return index;
-}
-
static inline struct iq_post_status
-__post_command2(struct octeon_device *octeon_dev __attribute__((unused)),
- struct octeon_instr_queue *iq,
- u32 force_db __attribute__((unused)), u8 *cmd)
+__post_command2(struct octeon_instr_queue *iq, u8 *cmd)
{
struct iq_post_status st;
iq->request_list[idx].reqtype = reqtype;
}
+/* Can only run in process context */
int
lio_process_iq_request_list(struct octeon_device *oct,
- struct octeon_instr_queue *iq)
+ struct octeon_instr_queue *iq, u32 napi_budget)
{
int reqtype;
void *buf;
u32 old = iq->flush_index;
u32 inst_count = 0;
- unsigned pkts_compl = 0, bytes_compl = 0;
+ unsigned int pkts_compl = 0, bytes_compl = 0;
struct octeon_soft_command *sc;
struct octeon_instr_irh *irh;
+ unsigned long flags;
while (old != iq->octeon_read_index) {
reqtype = iq->request_list[old].reqtype;
* command response list because we expect
* a response from Octeon.
*/
- spin_lock_bh(&oct->response_list
- [OCTEON_ORDERED_SC_LIST].lock);
+ spin_lock_irqsave
+ (&oct->response_list
+ [OCTEON_ORDERED_SC_LIST].lock,
+ flags);
atomic_inc(&oct->response_list
[OCTEON_ORDERED_SC_LIST].
pending_req_count);
list_add_tail(&sc->node, &oct->response_list
[OCTEON_ORDERED_SC_LIST].head);
- spin_unlock_bh(&oct->response_list
- [OCTEON_ORDERED_SC_LIST].lock);
+ spin_unlock_irqrestore
+ (&oct->response_list
+ [OCTEON_ORDERED_SC_LIST].lock,
+ flags);
} else {
if (sc->callback) {
+ /* This callback must not sleep */
sc->callback(oct, OCTEON_REQUEST_DONE,
sc->callback_arg);
}
skip_this:
inst_count++;
INCR_INDEX_BY1(old, iq->max_count);
+
+ if ((napi_budget) && (inst_count >= napi_budget))
+ break;
}
if (bytes_compl)
octeon_report_tx_completion_to_bql(iq->app_ctx, pkts_compl,
return inst_count;
}
-static inline void
-update_iq_indices(struct octeon_device *oct, struct octeon_instr_queue *iq)
+/* Can only be called from process context */
+int
+octeon_flush_iq(struct octeon_device *oct, struct octeon_instr_queue *iq,
+ u32 pending_thresh, u32 napi_budget)
{
u32 inst_processed = 0;
+ u32 tot_inst_processed = 0;
+ int tx_done = 1;
- /* Calculate how many commands Octeon has read and move the read index
- * accordingly.
- */
- iq->octeon_read_index = oct->fn_list.update_iq_read_idx(oct, iq);
+ if (!spin_trylock(&iq->iq_flush_running_lock))
+ return tx_done;
- /* Move the NORESPONSE requests to the per-device completion list. */
- if (iq->flush_index != iq->octeon_read_index)
- inst_processed = lio_process_iq_request_list(oct, iq);
+ spin_lock_bh(&iq->lock);
- if (inst_processed) {
- atomic_sub(inst_processed, &iq->instr_pending);
- iq->stats.instr_processed += inst_processed;
- }
-}
+ iq->octeon_read_index = oct->fn_list.update_iq_read_idx(iq);
-static void
-octeon_flush_iq(struct octeon_device *oct, struct octeon_instr_queue *iq,
- u32 pending_thresh)
-{
if (atomic_read(&iq->instr_pending) >= (s32)pending_thresh) {
- spin_lock_bh(&iq->lock);
- update_iq_indices(oct, iq);
- spin_unlock_bh(&iq->lock);
+ do {
+ /* Process any outstanding IQ packets. */
+ if (iq->flush_index == iq->octeon_read_index)
+ break;
+
+ if (napi_budget)
+ inst_processed = lio_process_iq_request_list
+ (oct, iq,
+ napi_budget - tot_inst_processed);
+ else
+ inst_processed =
+ lio_process_iq_request_list(oct, iq, 0);
+
+ if (inst_processed) {
+ atomic_sub(inst_processed, &iq->instr_pending);
+ iq->stats.instr_processed += inst_processed;
+ }
+
+ tot_inst_processed += inst_processed;
+ inst_processed = 0;
+
+ } while (tot_inst_processed < napi_budget);
+
+ if (napi_budget && (tot_inst_processed >= napi_budget))
+ tx_done = 0;
}
+
+ iq->last_db_time = jiffies;
+
+ spin_unlock_bh(&iq->lock);
+
+ spin_unlock(&iq->iq_flush_running_lock);
+
+ return tx_done;
}
-static void __check_db_timeout(struct octeon_device *oct, unsigned long iq_no)
+/* Process instruction queue after timeout.
+ * This routine gets called from a workqueue or when removing the module.
+ */
+static void __check_db_timeout(struct octeon_device *oct, u64 iq_no)
{
struct octeon_instr_queue *iq;
u64 next_time;
if (!iq)
return;
+ /* return immediately, if no work pending */
+ if (!atomic_read(&iq->instr_pending))
+ return;
/* If jiffies - last_db_time < db_timeout do nothing */
next_time = iq->last_db_time + iq->db_timeout;
if (!time_after(jiffies, (unsigned long)next_time))
return;
iq->last_db_time = jiffies;
- /* Get the lock and prevent tasklets. This routine gets called from
- * the poll thread. Instructions can now be posted in tasklet context
- */
- spin_lock_bh(&iq->lock);
- if (iq->fill_cnt != 0)
- ring_doorbell(oct, iq);
-
- spin_unlock_bh(&iq->lock);
-
/* Flush the instruction queue */
- if (iq->do_auto_flush)
- octeon_flush_iq(oct, iq, 1);
+ octeon_flush_iq(oct, iq, 1, 0);
}
/* Called by the Poll thread at regular intervals to check the instruction
{
struct cavium_wk *wk = (struct cavium_wk *)work;
struct octeon_device *oct = (struct octeon_device *)wk->ctxptr;
- unsigned long iq_no = wk->ctxul;
+ u64 iq_no = wk->ctxul;
struct cavium_wq *db_wq = &oct->check_db_wq[iq_no];
+ u32 delay = 10;
__check_db_timeout(oct, iq_no);
- queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(1));
+ queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(delay));
}
int
struct iq_post_status st;
struct octeon_instr_queue *iq = oct->instr_queue[iq_no];
- spin_lock_bh(&iq->lock);
+ /* Get the lock and prevent other tasks and tx interrupt handler from
+ * running.
+ */
+ spin_lock_bh(&iq->post_lock);
- st = __post_command2(oct, iq, force_db, cmd);
+ st = __post_command2(iq, cmd);
if (st.status != IQ_SEND_FAILED) {
octeon_report_sent_bytes_to_bql(buf, reqtype);
INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, bytes_sent, datasize);
INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_posted, 1);
- if (iq->fill_cnt >= iq->fill_threshold || force_db)
+ if (force_db)
ring_doorbell(oct, iq);
} else {
INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_dropped, 1);
}
- spin_unlock_bh(&iq->lock);
+ spin_unlock_bh(&iq->post_lock);
- if (iq->do_auto_flush)
- octeon_flush_iq(oct, iq, 2);
+ /* This is only done here to expedite packets being flushed
+ * for cases where there are no IQ completion interrupts.
+ */
+ /*if (iq->do_auto_flush)*/
+ /* octeon_flush_iq(oct, iq, 2, 0);*/
return st.status;
}
struct octeon_instr_irh *irh;
struct octeon_instr_rdp *rdp;
- BUG_ON(opcode > 15);
- BUG_ON(subcode > 127);
+ WARN_ON(opcode > 15);
+ WARN_ON(subcode > 127);
oct_cfg = octeon_get_conf(oct);
{
struct octeon_instr_ih2 *ih2;
struct octeon_instr_irh *irh;
- struct octeon_instr_rdp *rdp;
u32 len;
ih2 = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2;
}
irh = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh;
if (irh->rflag) {
- BUG_ON(!sc->dmarptr);
- BUG_ON(!sc->status_word);
+ WARN_ON(!sc->dmarptr);
+ WARN_ON(!sc->status_word);
*sc->status_word = COMPLETION_WORD_INIT;
- rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd2.rdp;
-
sc->cmd.cmd2.rptr = sc->dmarptr;
}
len = (u32)ih2->dlengsz;
struct list_head *tmp, *tmp2;
struct octeon_soft_command *sc;
- spin_lock(&oct->sc_buf_pool.lock);
+ spin_lock_bh(&oct->sc_buf_pool.lock);
list_for_each_safe(tmp, tmp2, &oct->sc_buf_pool.head) {
list_del(tmp);
INIT_LIST_HEAD(&oct->sc_buf_pool.head);
- spin_unlock(&oct->sc_buf_pool.lock);
+ spin_unlock_bh(&oct->sc_buf_pool.lock);
return 0;
}
struct octeon_soft_command *sc = NULL;
struct list_head *tmp;
- BUG_ON((offset + datasize + rdatasize + ctxsize) >
+ WARN_ON((offset + datasize + rdatasize + ctxsize) >
SOFT_COMMAND_BUFFER_SIZE);
- spin_lock(&oct->sc_buf_pool.lock);
+ spin_lock_bh(&oct->sc_buf_pool.lock);
if (list_empty(&oct->sc_buf_pool.head)) {
- spin_unlock(&oct->sc_buf_pool.lock);
+ spin_unlock_bh(&oct->sc_buf_pool.lock);
return NULL;
}
atomic_inc(&oct->sc_buf_pool.alloc_buf_count);
- spin_unlock(&oct->sc_buf_pool.lock);
+ spin_unlock_bh(&oct->sc_buf_pool.lock);
sc = (struct octeon_soft_command *)tmp;
offset = (offset + datasize + 127) & 0xffffff80;
if (rdatasize) {
- BUG_ON(rdatasize < 16);
+ WARN_ON(rdatasize < 16);
sc->virtrptr = (u8 *)sc + offset;
sc->dmarptr = dma_addr + offset;
sc->rdatasize = rdatasize;
void octeon_free_soft_command(struct octeon_device *oct,
struct octeon_soft_command *sc)
{
- spin_lock(&oct->sc_buf_pool.lock);
+ spin_lock_bh(&oct->sc_buf_pool.lock);
list_add_tail(&sc->node, &oct->sc_buf_pool.head);
atomic_dec(&oct->sc_buf_pool.alloc_buf_count);
- spin_unlock(&oct->sc_buf_pool.lock);
+ spin_unlock_bh(&oct->sc_buf_pool.lock);
}