diff options
author | Martin Sperl <kernel@martin.sperl.org> | 2019-01-08 12:13:45 +0000 |
---|---|---|
committer | Mark Brown <broonie@kernel.org> | 2019-01-09 19:16:18 +0000 |
commit | 412e60373245fd1dfae8d4d44c5d1406b3d90971 (patch) | |
tree | cd716d2e49b96ff645de95863b7fa760ed7a6adc /drivers/spi | |
parent | 9400c41e77b8f1f9c711b08a6c79bdbc5681fb00 (diff) | |
download | linux-412e60373245fd1dfae8d4d44c5d1406b3d90971.tar.gz linux-412e60373245fd1dfae8d4d44c5d1406b3d90971.tar.bz2 linux-412e60373245fd1dfae8d4d44c5d1406b3d90971.zip |
spi: core: avoid waking pump thread from spi_sync instead run teardown delayed
When spi_sync is running alone with no other spi devices connected
to the bus the worker thread is woken during spi_finalize_current_message
to run the teardown code every time.
This is totally unnecessary in the case that there is no message queued.
On a multi-core system this results in one wakeup of the thread for each
spi_message processed via spi_sync where in most cases the teardown does
not happen as the hw is already in use.
This patch now delays the teardown by 1 second by using a separate
kthread_delayed_work for the teardown.
This avoids waking the kthread too often.
For spi_sync transfers in a tight loop (say 40k messages/s) this
avoids the penalty of waking the worker thread 40k times/s.
On a rasperry pi 3 with 4 cores the results in 32% of a single core
only to find out that there is nothing in the queue and it can go back
to sleep.
With this patch applied the spi-worker is woken exactly once: after
the load finishes and the spi bus is idle for 1 second.
I believe I have also seen situations where during a spi_sync loop
the worker thread (triggered by the last message finished) is slightly
faster and _wins_ the race to process the message, so we are actually
running the kthread and letting it do some work...
This is also no longer observed with this patch applied as.
Tested with a new CAN controller driver for the mcp2517fd which
uses spi_sync for interrupt handling and spi_async for scheduling
of can frames for transmission (in a different thread)
Some statistics when receiving 100000 CAN frames with the mcp25xxfd driver
on a Raspberry pi 3:
without the patch:
------------------
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0) 5
(irq/94-mcp25xxf) 0
root@raspcm3:~# vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
r b swpd free buff cache si so bi bo in cs us sy id wa st
1 0 0 821960 13592 50848 0 0 80 2 1986 105 1 2 97 0 0
0 0 0 821968 13592 50876 0 0 0 0 8046 30 0 0 100 0 0
0 0 0 821936 13592 50876 0 0 0 0 8032 24 0 0 100 0 0
0 0 0 821936 13592 50876 0 0 0 0 8035 30 0 0 100 0 0
0 0 0 821936 13592 50876 0 0 0 0 8033 22 0 0 100 0 0
2 0 0 821936 13592 50876 0 0 0 0 11598 7129 0 3 97 0 0
1 0 0 821872 13592 50876 0 0 0 0 37741 59003 0 31 69 0 0
2 0 0 821840 13592 50876 0 0 0 0 37762 59078 0 29 71 0 0
2 0 0 821776 13592 50876 0 0 0 0 37593 58792 0 28 72 0 0
1 0 0 821744 13592 50876 0 0 0 0 37642 58881 0 30 70 0 0
2 0 0 821680 13592 50876 0 0 0 0 37490 58602 0 27 73 0 0
1 0 0 821648 13592 50876 0 0 0 0 37412 58418 0 29 71 0 0
1 0 0 821584 13592 50876 0 0 0 0 37337 58288 0 27 73 0 0
1 0 0 821552 13592 50876 0 0 0 0 37584 58774 0 27 73 0 0
0 0 0 821520 13592 50876 0 0 0 0 18363 20566 0 9 91 0 0
0 0 0 821520 13592 50876 0 0 0 0 8037 32 0 0 100 0 0
0 0 0 821520 13592 50876 0 0 0 0 8031 23 0 0 100 0 0
0 0 0 821520 13592 50876 0 0 0 0 8034 26 0 0 100 0 0
0 0 0 821520 13592 50876 0 0 0 0 8033 24 0 0 100 0 0
^C
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0) 228
(irq/94-mcp25xxf) 794
root@raspcm3:~# cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3
17: 34 0 0 0 ARMCTRL-level 1 Edge 3f00b880.mailbox
27: 1 0 0 0 ARMCTRL-level 35 Edge timer
33: 1416870 0 0 0 ARMCTRL-level 41 Edge 3f980000.usb, dwc2_hsotg:usb1
34: 1 0 0 0 ARMCTRL-level 42 Edge vc4
35: 0 0 0 0 ARMCTRL-level 43 Edge 3f004000.txp
40: 1753 0 0 0 ARMCTRL-level 48 Edge DMA IRQ
42: 11 0 0 0 ARMCTRL-level 50 Edge DMA IRQ
44: 11 0 0 0 ARMCTRL-level 52 Edge DMA IRQ
45: 0 0 0 0 ARMCTRL-level 53 Edge DMA IRQ
66: 0 0 0 0 ARMCTRL-level 74 Edge vc4 crtc
69: 0 0 0 0 ARMCTRL-level 77 Edge vc4 crtc
70: 0 0 0 0 ARMCTRL-level 78 Edge vc4 crtc
77: 20 0 0 0 ARMCTRL-level 85 Edge 3f205000.i2c, 3f804000.i2c, 3f805000.i2c
78: 6346 0 0 0 ARMCTRL-level 86 Edge 3f204000.spi
80: 205 0 0 0 ARMCTRL-level 88 Edge mmc0
81: 493 0 0 0 ARMCTRL-level 89 Edge uart-pl011
89: 0 0 0 0 bcm2836-timer 0 Edge arch_timer
90: 4291 3821 2180 1649 bcm2836-timer 1 Edge arch_timer
94: 14289 0 0 0 pinctrl-bcm2835 16 Level mcp25xxfd
IPI0: 0 0 0 0 CPU wakeup interrupts
IPI1: 0 0 0 0 Timer broadcast interrupts
IPI2: 3645 242371 7919 1328 Rescheduling interrupts
IPI3: 112 543 273 194 Function call interrupts
IPI4: 0 0 0 0 CPU stop interrupts
IPI5: 1 0 0 0 IRQ work interrupts
IPI6: 0 0 0 0 completion interrupts
Err: 0
top shows 93% for the mcp25xxfd interrupt handler, 31% for spi0.
with the patch:
---------------
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0) 0
(irq/94-mcp25xxf) 0
root@raspcm3:~# vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
0 0 0 804768 13584 62628 0 0 0 0 8038 24 0 0 100 0 0
0 0 0 804768 13584 62628 0 0 0 0 8042 25 0 0 100 0 0
1 0 0 804704 13584 62628 0 0 0 0 9603 2967 0 20 80 0 0
1 0 0 804672 13584 62628 0 0 0 0 9828 3380 0 24 76 0 0
1 0 0 804608 13584 62628 0 0 0 0 9823 3375 0 23 77 0 0
1 0 0 804608 13584 62628 0 0 0 12 9829 3394 0 23 77 0 0
1 0 0 804544 13584 62628 0 0 0 0 9816 3362 0 22 78 0 0
1 0 0 804512 13584 62628 0 0 0 0 9817 3367 0 23 77 0 0
1 0 0 804448 13584 62628 0 0 0 0 9822 3370 0 22 78 0 0
1 0 0 804416 13584 62628 0 0 0 0 9815 3367 0 23 77 0 0
0 0 0 804352 13584 62628 0 0 0 84 9222 2250 0 14 86 0 0
0 0 0 804352 13592 62620 0 0 0 24 8131 209 0 0 93 7 0
0 0 0 804320 13592 62628 0 0 0 0 8041 27 0 0 100 0 0
0 0 0 804352 13592 62628 0 0 0 0 8040 26 0 0 100 0 0
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0) 0
(irq/94-mcp25xxf) 767
root@raspcm3:~# cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3
17: 29 0 0 0 ARMCTRL-level 1 Edge 3f00b880.mailbox
27: 1 0 0 0 ARMCTRL-level 35 Edge timer
33: 1024412 0 0 0 ARMCTRL-level 41 Edge 3f980000.usb, dwc2_hsotg:usb1
34: 1 0 0 0 ARMCTRL-level 42 Edge vc4
35: 0 0 0 0 ARMCTRL-level 43 Edge 3f004000.txp
40: 1773 0 0 0 ARMCTRL-level 48 Edge DMA IRQ
42: 11 0 0 0 ARMCTRL-level 50 Edge DMA IRQ
44: 11 0 0 0 ARMCTRL-level 52 Edge DMA IRQ
45: 0 0 0 0 ARMCTRL-level 53 Edge DMA IRQ
66: 0 0 0 0 ARMCTRL-level 74 Edge vc4 crtc
69: 0 0 0 0 ARMCTRL-level 77 Edge vc4 crtc
70: 0 0 0 0 ARMCTRL-level 78 Edge vc4 crtc
77: 20 0 0 0 ARMCTRL-level 85 Edge 3f205000.i2c, 3f804000.i2c, 3f805000.i2c
78: 6417 0 0 0 ARMCTRL-level 86 Edge 3f204000.spi
80: 237 0 0 0 ARMCTRL-level 88 Edge mmc0
81: 489 0 0 0 ARMCTRL-level 89 Edge uart-pl011
89: 0 0 0 0 bcm2836-timer 0 Edge arch_timer
90: 4048 3704 2383 1892 bcm2836-timer 1 Edge arch_timer
94: 14287 0 0 0 pinctrl-bcm2835 16 Level mcp25xxfd
IPI0: 0 0 0 0 CPU wakeup interrupts
IPI1: 0 0 0 0 Timer broadcast interrupts
IPI2: 2361 2948 7890 1616 Rescheduling interrupts
IPI3: 65 617 301 166 Function call interrupts
IPI4: 0 0 0 0 CPU stop interrupts
IPI5: 1 0 0 0 IRQ work interrupts
IPI6: 0 0 0 0 completion interrupts
Err: 0
top shows 91% for the mcp25xxfd interrupt handler, 0% for spi0
So we see that spi0 is no longer getting scheduled wasting CPU cycles
There are a lot less context switches and corresponding Rescheduling interrupts
All of these show that this improves efficiency of the system and reduces
CPU utilization.
Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
Diffstat (limited to 'drivers/spi')
-rw-r--r-- | drivers/spi/spi.c | 122 |
1 files changed, 89 insertions, 33 deletions
diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c index 13f447a67d67..06b9139664a3 100644 --- a/drivers/spi/spi.c +++ b/drivers/spi/spi.c @@ -1225,7 +1225,7 @@ static void __spi_pump_messages(struct spi_controller *ctlr, bool in_kthread) return; } - /* If another context is idling the device then defer */ + /* If another context is idling the device then defer to kthread */ if (ctlr->idling) { kthread_queue_work(&ctlr->kworker, &ctlr->pump_messages); spin_unlock_irqrestore(&ctlr->queue_lock, flags); @@ -1239,34 +1239,10 @@ static void __spi_pump_messages(struct spi_controller *ctlr, bool in_kthread) return; } - /* Only do teardown in the thread */ - if (!in_kthread) { - kthread_queue_work(&ctlr->kworker, - &ctlr->pump_messages); - spin_unlock_irqrestore(&ctlr->queue_lock, flags); - return; - } - - ctlr->busy = false; - ctlr->idling = true; - spin_unlock_irqrestore(&ctlr->queue_lock, flags); - - kfree(ctlr->dummy_rx); - ctlr->dummy_rx = NULL; - kfree(ctlr->dummy_tx); - ctlr->dummy_tx = NULL; - if (ctlr->unprepare_transfer_hardware && - ctlr->unprepare_transfer_hardware(ctlr)) - dev_err(&ctlr->dev, - "failed to unprepare transfer hardware\n"); - if (ctlr->auto_runtime_pm) { - pm_runtime_mark_last_busy(ctlr->dev.parent); - pm_runtime_put_autosuspend(ctlr->dev.parent); - } - trace_spi_controller_idle(ctlr); - - spin_lock_irqsave(&ctlr->queue_lock, flags); - ctlr->idling = false; + /* schedule idle teardown with a delay of 1 second */ + kthread_mod_delayed_work(&ctlr->kworker, + &ctlr->pump_idle_teardown, + HZ); spin_unlock_irqrestore(&ctlr->queue_lock, flags); return; } @@ -1359,6 +1335,77 @@ static void spi_pump_messages(struct kthread_work *work) __spi_pump_messages(ctlr, true); } +/** + * spi_pump_idle_teardown - kthread delayed work function which tears down + * the controller settings after some delay + * @work: pointer to kthread work struct contained in the controller struct + */ +static void spi_pump_idle_teardown(struct kthread_work *work) +{ + struct spi_controller *ctlr = + container_of(work, struct spi_controller, + pump_idle_teardown.work); + unsigned long flags; + + /* Lock queue */ + spin_lock_irqsave(&ctlr->queue_lock, flags); + + /* Make sure we are not already running a message */ + if (ctlr->cur_msg) + goto out; + + /* if there is anything in the list then exit */ + if (!list_empty(&ctlr->queue)) + goto out; + + /* if the controller is running then exit */ + if (ctlr->running) + goto out; + + /* if the controller is busy then exit */ + if (ctlr->busy) + goto out; + + /* if the controller is idling then exit + * this is actually a bit strange and would indicate that + * this function is scheduled twice, which should not happen + */ + if (ctlr->idling) + goto out; + + /* set up the initial states */ + ctlr->busy = false; + ctlr->idling = true; + spin_unlock_irqrestore(&ctlr->queue_lock, flags); + + /* free dummy receive buffers */ + kfree(ctlr->dummy_rx); + ctlr->dummy_rx = NULL; + kfree(ctlr->dummy_tx); + ctlr->dummy_tx = NULL; + + /* unprepare hardware */ + if (ctlr->unprepare_transfer_hardware && + ctlr->unprepare_transfer_hardware(ctlr)) + dev_err(&ctlr->dev, + "failed to unprepare transfer hardware\n"); + /* handle pm */ + if (ctlr->auto_runtime_pm) { + pm_runtime_mark_last_busy(ctlr->dev.parent); + pm_runtime_put_autosuspend(ctlr->dev.parent); + } + + /* mark controller as idle */ + trace_spi_controller_idle(ctlr); + + /* finally put us from idling into stopped */ + spin_lock_irqsave(&ctlr->queue_lock, flags); + ctlr->idling = false; + +out: + spin_unlock_irqrestore(&ctlr->queue_lock, flags); +} + static int spi_init_queue(struct spi_controller *ctlr) { struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; @@ -1374,7 +1421,8 @@ static int spi_init_queue(struct spi_controller *ctlr) return PTR_ERR(ctlr->kworker_task); } kthread_init_work(&ctlr->pump_messages, spi_pump_messages); - + kthread_init_delayed_work(&ctlr->pump_idle_teardown, + spi_pump_idle_teardown); /* * Controller config will indicate if this controller should run the * message pump with high (realtime) priority to reduce the transfer @@ -1446,7 +1494,16 @@ void spi_finalize_current_message(struct spi_controller *ctlr) spin_lock_irqsave(&ctlr->queue_lock, flags); ctlr->cur_msg = NULL; ctlr->cur_msg_prepared = false; - kthread_queue_work(&ctlr->kworker, &ctlr->pump_messages); + + /* if there is something queued, then wake the queue */ + if (!list_empty(&ctlr->queue)) + kthread_queue_work(&ctlr->kworker, &ctlr->pump_messages); + else + /* otherwise schedule delayed teardown */ + kthread_mod_delayed_work(&ctlr->kworker, + &ctlr->pump_idle_teardown, + HZ); + spin_unlock_irqrestore(&ctlr->queue_lock, flags); trace_spi_message_done(mesg); @@ -1551,7 +1608,7 @@ static int __spi_queued_transfer(struct spi_device *spi, msg->status = -EINPROGRESS; list_add_tail(&msg->queue, &ctlr->queue); - if (!ctlr->busy && need_pump) + if (need_pump) kthread_queue_work(&ctlr->kworker, &ctlr->pump_messages); spin_unlock_irqrestore(&ctlr->queue_lock, flags); @@ -3726,4 +3783,3 @@ err0: * include needing to have boardinfo data structures be much more public. */ postcore_initcall(spi_init); - |