Use spin_lock_bh to instead of spin_lock_irqsave for improving performance.
Cc: Dan Williams <[email protected]>
Cc: Vinod Koul <[email protected]>
Cc: Li Yang <[email protected]>
Signed-off-by: Qiang Liu <[email protected]>
---
drivers/dma/fsldma.c | 29 ++++++++++++-----------------
1 files changed, 12 insertions(+), 17 deletions(-)
diff --git a/drivers/dma/fsldma.c b/drivers/dma/fsldma.c
index 4ee1b8f..e975719 100644
--- a/drivers/dma/fsldma.c
+++ b/drivers/dma/fsldma.c
@@ -616,10 +616,9 @@ static dma_cookie_t fsl_dma_tx_submit(struct dma_async_tx_descriptor *tx)
struct fsldma_chan *chan = to_fsl_chan(tx->chan);
struct fsl_desc_sw *desc = tx_to_fsl_desc(tx);
struct fsl_desc_sw *child;
- unsigned long flags;
dma_cookie_t cookie;
- spin_lock_irqsave(&chan->desc_lock, flags);
+ spin_lock_bh(&chan->desc_lock);
/*
* assign cookies to all of the software descriptors
@@ -632,7 +631,7 @@ static dma_cookie_t fsl_dma_tx_submit(struct dma_async_tx_descriptor *tx)
/* put this transaction onto the tail of the pending queue */
append_ld_queue(chan, desc);
- spin_unlock_irqrestore(&chan->desc_lock, flags);
+ spin_unlock_bh(&chan->desc_lock);
return cookie;
}
@@ -741,15 +740,14 @@ static void fsldma_free_desc_list_reverse(struct fsldma_chan *chan,
static void fsl_dma_free_chan_resources(struct dma_chan *dchan)
{
struct fsldma_chan *chan = to_fsl_chan(dchan);
- unsigned long flags;
chan_dbg(chan, "free all channel resources\n");
- spin_lock_irqsave(&chan->desc_lock, flags);
+ spin_lock_bh(&chan->desc_lock);
fsldma_cleanup_descriptor(chan);
fsldma_free_desc_list(chan, &chan->ld_pending);
fsldma_free_desc_list(chan, &chan->ld_running);
fsldma_free_desc_list(chan, &chan->ld_completed);
- spin_unlock_irqrestore(&chan->desc_lock, flags);
+ spin_unlock_bh(&chan->desc_lock);
dma_pool_destroy(chan->desc_pool);
chan->desc_pool = NULL;
@@ -968,7 +966,6 @@ static int fsl_dma_device_control(struct dma_chan *dchan,
{
struct dma_slave_config *config;
struct fsldma_chan *chan;
- unsigned long flags;
int size;
if (!dchan)
@@ -978,7 +975,7 @@ static int fsl_dma_device_control(struct dma_chan *dchan,
switch (cmd) {
case DMA_TERMINATE_ALL:
- spin_lock_irqsave(&chan->desc_lock, flags);
+ spin_lock_bh(&chan->desc_lock);
/* Halt the DMA engine */
dma_halt(chan);
@@ -988,7 +985,7 @@ static int fsl_dma_device_control(struct dma_chan *dchan,
fsldma_free_desc_list(chan, &chan->ld_running);
chan->idle = true;
- spin_unlock_irqrestore(&chan->desc_lock, flags);
+ spin_unlock_bh(&chan->desc_lock);
return 0;
case DMA_SLAVE_CONFIG:
@@ -1030,11 +1027,10 @@ static int fsl_dma_device_control(struct dma_chan *dchan,
static void fsl_dma_memcpy_issue_pending(struct dma_chan *dchan)
{
struct fsldma_chan *chan = to_fsl_chan(dchan);
- unsigned long flags;
- spin_lock_irqsave(&chan->desc_lock, flags);
+ spin_lock_bh(&chan->desc_lock);
fsl_chan_xfer_ld_queue(chan);
- spin_unlock_irqrestore(&chan->desc_lock, flags);
+ spin_unlock_bh(&chan->desc_lock);
}
/**
@@ -1047,7 +1043,6 @@ static enum dma_status fsl_tx_status(struct dma_chan *dchan,
{
struct fsldma_chan *chan = to_fsl_chan(dchan);
enum dma_status ret;
- unsigned long flags;
ret = dma_cookie_status(dchan, cookie, txstate);
if (ret == DMA_SUCCESS) {
@@ -1055,9 +1050,9 @@ static enum dma_status fsl_tx_status(struct dma_chan *dchan,
return ret;
}
- spin_lock_irqsave(&chan->desc_lock, flags);
+ spin_lock_bh(&chan->desc_lock);
fsldma_cleanup_descriptor(chan);
- spin_unlock_irqrestore(&chan->desc_lock, flags);
+ spin_unlock_bh(&chan->desc_lock);
return dma_cookie_status(dchan, cookie, txstate);
}
@@ -1140,7 +1135,7 @@ static void dma_do_tasklet(unsigned long data)
chan_dbg(chan, "tasklet entry\n");
- spin_lock_irqsave(&chan->desc_lock, flags);
+ spin_lock_bh(&chan->desc_lock);
/* Run all cleanup for this descriptor */
fsldma_cleanup_descriptor(chan);
@@ -1149,7 +1144,7 @@ static void dma_do_tasklet(unsigned long data)
chan->idle = true;
fsl_chan_xfer_ld_queue(chan);
- spin_unlock_irqrestore(&chan->desc_lock, flags);
+ spin_unlock_bh(&chan->desc_lock);
chan_dbg(chan, "tasklet exit\n");
}
--
1.7.5.1
Qiang Liu wrote:
> Use spin_lock_bh to instead of spin_lock_irqsave for improving performance.
You forgot to include the evidence that performance has improved, as well
as an explanation why it's okay to use spin_lock_bh, and why it's faster.
I told you to respin the patch with that information in the patch
description.
--
Timur Tabi
Linux kernel developer at Freescale
> -----Original Message-----
> From: Tabi Timur-B04825
> Sent: Monday, July 16, 2012 10:25 PM
> To: Liu Qiang-B32616
> Cc: [email protected]; [email protected]; Vinod
> Koul; [email protected]; Dan Williams; Li Yang-R58472;
> [email protected]
> Subject: Re: [linuxppc-release] [PATCH v3 4/4] fsl-dma: use spin_lock_bh
> to instead of spin_lock_irqsave
>
> Qiang Liu wrote:
> > Use spin_lock_bh to instead of spin_lock_irqsave for improving
> performance.
>
> You forgot to include the evidence that performance has improved, as well
> as an explanation why it's okay to use spin_lock_bh, and why it's faster.
> I told you to respin the patch with that information in the patch
> description.
I attached the test result in v3 0/4, performance is improved by 2%.
For my understanding, there is not any place to access descriptor lists
in fsl-dma interrupt service handler except its tasklet, spin_lock_bh()
is born for this. Interrupts will be turned off and context will be save in
irqsave, there is needless to use irqsave in our case. You can refer to the
implement of mv_xor.c or ioap-adma.c.
If you think my explanation is ok, I can add it in the patch.
Thanks.
>
> --
> Timur Tabi
> Linux kernel developer at Freescale
On Tue, Jul 17, 2012 at 12:23 PM, Liu Qiang-B32616 <[email protected]> wrote:
>> -----Original Message-----
>> From: Tabi Timur-B04825
>> Sent: Monday, July 16, 2012 10:25 PM
>> To: Liu Qiang-B32616
>> Cc: [email protected]; [email protected]; Vinod
>> Koul; [email protected]; Dan Williams; Li Yang-R58472;
>> [email protected]
>> Subject: Re: [linuxppc-release] [PATCH v3 4/4] fsl-dma: use spin_lock_bh
>> to instead of spin_lock_irqsave
>>
>> Qiang Liu wrote:
>> > Use spin_lock_bh to instead of spin_lock_irqsave for improving
>> performance.
>>
>> You forgot to include the evidence that performance has improved, as well
>> as an explanation why it's okay to use spin_lock_bh, and why it's faster.
>> I told you to respin the patch with that information in the patch
>> description.
> I attached the test result in v3 0/4, performance is improved by 2%.
> For my understanding, there is not any place to access descriptor lists
> in fsl-dma interrupt service handler except its tasklet, spin_lock_bh()
> is born for this. Interrupts will be turned off and context will be save in
> irqsave, there is needless to use irqsave in our case. You can refer to the
> implement of mv_xor.c or ioap-adma.c.
> If you think my explanation is ok, I can add it in the patch.
Disabling bottom half is not faster than disabling the interrupt. But
by using it we reduce the time when interrupt is disabled. We can get
better real time performance in term of interrupt handling latency.
Leo
Liu Qiang-B32616 wrote:
> I attached the test result in v3 0/4, performance is improved by 2%.
> For my understanding, there is not any place to access descriptor lists
> in fsl-dma interrupt service handler except its tasklet, spin_lock_bh()
> is born for this. Interrupts will be turned off and context will be save in
> irqsave, there is needless to use irqsave in our case. You can refer to the
> implement of mv_xor.c or ioap-adma.c.
> If you think my explanation is ok, I can add it in the patch.
Yes, the explanation is ok. Please make it part of the patch description,
so that it is saved in the git history.
--
Timur Tabi
Linux kernel developer at Freescale