This add power management suspend/resume support for the fsl-edma
driver.
eDMA acted as a basic function used by others. What it needs to do is
the two steps below to support power management.
In fsl_edma_suspend_late:
Check whether the DMA chan is idle and if it is not idle, stop PM
operation.
In fsl_edma_resume_early:
Enable the eDMA and wait for being used.
Signed-off-by: Yuan Yao <[email protected]>
---
drivers/dma/fsl-edma.c | 83 ++++++++++++++++++++++++++++++++++++++++++++++++--
1 file changed, 80 insertions(+), 3 deletions(-)
diff --git a/drivers/dma/fsl-edma.c b/drivers/dma/fsl-edma.c
index 915eec3..cf8b06c 100644
--- a/drivers/dma/fsl-edma.c
+++ b/drivers/dma/fsl-edma.c
@@ -116,6 +116,10 @@
BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
BIT(DMA_SLAVE_BUSWIDTH_8_BYTES)
+enum fsl_edma_pm_state {
+ RUNNING = 0,
+ SUSPENDED,
+};
struct fsl_edma_hw_tcd {
__le32 saddr;
@@ -147,6 +151,9 @@ struct fsl_edma_slave_config {
struct fsl_edma_chan {
struct virt_dma_chan vchan;
enum dma_status status;
+ enum fsl_edma_pm_state pm_state;
+ bool idle;
+ u32 slave_id;
struct fsl_edma_engine *edma;
struct fsl_edma_desc *edesc;
struct fsl_edma_slave_config fsc;
@@ -298,6 +305,7 @@ static int fsl_edma_terminate_all(struct dma_chan *chan)
spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
fsl_edma_disable_request(fsl_chan);
fsl_chan->edesc = NULL;
+ fsl_chan->idle = true;
vchan_get_all_descriptors(&fsl_chan->vchan, &head);
spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
vchan_dma_desc_free_list(&fsl_chan->vchan, &head);
@@ -313,6 +321,7 @@ static int fsl_edma_pause(struct dma_chan *chan)
if (fsl_chan->edesc) {
fsl_edma_disable_request(fsl_chan);
fsl_chan->status = DMA_PAUSED;
+ fsl_chan->idle = true;
}
spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
return 0;
@@ -327,6 +336,7 @@ static int fsl_edma_resume(struct dma_chan *chan)
if (fsl_chan->edesc) {
fsl_edma_enable_request(fsl_chan);
fsl_chan->status = DMA_IN_PROGRESS;
+ fsl_chan->idle = false;
}
spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
return 0;
@@ -648,6 +658,7 @@ static void fsl_edma_xfer_desc(struct fsl_edma_chan *fsl_chan)
fsl_edma_set_tcd_regs(fsl_chan, fsl_chan->edesc->tcd[0].vtcd);
fsl_edma_enable_request(fsl_chan);
fsl_chan->status = DMA_IN_PROGRESS;
+ fsl_chan->idle = false;
}
static irqreturn_t fsl_edma_tx_handler(int irq, void *dev_id)
@@ -676,6 +687,7 @@ static irqreturn_t fsl_edma_tx_handler(int irq, void *dev_id)
vchan_cookie_complete(&fsl_chan->edesc->vdesc);
fsl_chan->edesc = NULL;
fsl_chan->status = DMA_COMPLETE;
+ fsl_chan->idle = true;
} else {
vchan_cyclic_callback(&fsl_chan->edesc->vdesc);
}
@@ -704,6 +716,7 @@ static irqreturn_t fsl_edma_err_handler(int irq, void *dev_id)
edma_writeb(fsl_edma, EDMA_CERR_CERR(ch),
fsl_edma->membase + EDMA_CERR);
fsl_edma->chans[ch].status = DMA_ERROR;
+ fsl_edma->chans[ch].idle = true;
}
}
return IRQ_HANDLED;
@@ -724,6 +737,12 @@ static void fsl_edma_issue_pending(struct dma_chan *chan)
spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
+ if (unlikely(fsl_chan->pm_state != RUNNING)) {
+ spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
+ /* cannot submit due to suspend */
+ return;
+ }
+
if (vchan_issue_pending(&fsl_chan->vchan) && !fsl_chan->edesc)
fsl_edma_xfer_desc(fsl_chan);
@@ -735,6 +754,7 @@ static struct dma_chan *fsl_edma_xlate(struct of_phandle_args *dma_spec,
{
struct fsl_edma_engine *fsl_edma = ofdma->of_dma_data;
struct dma_chan *chan, *_chan;
+ struct fsl_edma_chan *fsl_chan;
unsigned long chans_per_mux = fsl_edma->n_chans / DMAMUX_NR;
if (dma_spec->args_count != 2)
@@ -748,8 +768,10 @@ static struct dma_chan *fsl_edma_xlate(struct of_phandle_args *dma_spec,
chan = dma_get_slave_channel(chan);
if (chan) {
chan->device->privatecnt++;
- fsl_edma_chan_mux(to_fsl_edma_chan(chan),
- dma_spec->args[1], true);
+ fsl_chan = to_fsl_edma_chan(chan);
+ fsl_chan->slave_id = dma_spec->args[1];
+ fsl_edma_chan_mux(fsl_chan, fsl_chan->slave_id,
+ true);
mutex_unlock(&fsl_edma->fsl_edma_mutex);
return chan;
}
@@ -888,7 +910,9 @@ static int fsl_edma_probe(struct platform_device *pdev)
struct fsl_edma_chan *fsl_chan = &fsl_edma->chans[i];
fsl_chan->edma = fsl_edma;
-
+ fsl_chan->pm_state = RUNNING;
+ fsl_chan->slave_id = 0;
+ fsl_chan->idle = true;
fsl_chan->vchan.desc_free = fsl_edma_free_desc;
vchan_init(&fsl_chan->vchan, &fsl_edma->dma_dev);
@@ -959,6 +983,58 @@ static int fsl_edma_remove(struct platform_device *pdev)
return 0;
}
+static int fsl_edma_suspend_late(struct device *dev)
+{
+ struct fsl_edma_engine *fsl_edma = dev_get_drvdata(dev);
+ struct fsl_edma_chan *fsl_chan;
+ unsigned long flags;
+ int i;
+
+ for (i = 0; i < fsl_edma->n_chans; i++) {
+ fsl_chan = &fsl_edma->chans[i];
+ spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
+ /* Make sure chan is idle or can not into suspend. */
+ if (unlikely(!fsl_chan->idle))
+ goto out;
+ fsl_chan->pm_state = SUSPENDED;
+ spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
+ }
+ return 0;
+
+out:
+ for (; i >= 0; i--) {
+ fsl_chan = &fsl_edma->chans[i];
+ fsl_chan->pm_state = RUNNING;
+ spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
+ }
+ return -EBUSY;
+}
+
+static int fsl_edma_resume_early(struct device *dev)
+{
+ struct fsl_edma_engine *fsl_edma = dev_get_drvdata(dev);
+ struct fsl_edma_chan *fsl_chan;
+ int i;
+
+ for (i = 0; i < fsl_edma->n_chans; i++) {
+ fsl_chan = &fsl_edma->chans[i];
+ fsl_chan->pm_state = RUNNING;
+ edma_writew(fsl_edma, 0x0, fsl_edma->membase + EDMA_TCD_CSR(i));
+ if (fsl_chan->slave_id != 0)
+ fsl_edma_chan_mux(fsl_chan, fsl_chan->slave_id, true);
+ }
+
+ edma_writel(fsl_edma, EDMA_CR_ERGA | EDMA_CR_ERCA,
+ fsl_edma->membase + EDMA_CR);
+
+ return 0;
+}
+
+static const struct dev_pm_ops fsl_edma_pm_ops = {
+ .suspend_late = fsl_edma_suspend_late,
+ .resume_early = fsl_edma_resume_early,
+};
+
static const struct of_device_id fsl_edma_dt_ids[] = {
{ .compatible = "fsl,vf610-edma", },
{ /* sentinel */ }
@@ -969,6 +1045,7 @@ static struct platform_driver fsl_edma_driver = {
.driver = {
.name = "fsl-edma",
.of_match_table = fsl_edma_dt_ids,
+ .pm = &fsl_edma_pm_ops,
},
.probe = fsl_edma_probe,
.remove = fsl_edma_remove,
--
2.1.0.27.g96db324
Hi Vinod,
After our discuss, I have already send the v2.
Could you please take some times help to review this patch again?
Thanks.
Best Regards,
Yuan Yao
> -----Original Message-----
> From: linux-arm-kernel [mailto:[email protected]]
> On Behalf Of Yuan Yao
> Sent: Tuesday, July 21, 2015 4:57 PM
> To: [email protected]; [email protected]; [email protected]
> Cc: [email protected]; [email protected]; linux-
> [email protected]; [email protected]
> Subject: [PATCH v2] dmaengine: fsl-edma: add PM suspend/resume support
>
> This add power management suspend/resume support for the fsl-edma driver.
>
> eDMA acted as a basic function used by others. What it needs to do is the two
> steps below to support power management.
>
> In fsl_edma_suspend_late:
> Check whether the DMA chan is idle and if it is not idle, stop PM operation.
>
> In fsl_edma_resume_early:
> Enable the eDMA and wait for being used.
>
> Signed-off-by: Yuan Yao <[email protected]>
> ---
> drivers/dma/fsl-edma.c | 83
> ++++++++++++++++++++++++++++++++++++++++++++++++--
> 1 file changed, 80 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/dma/fsl-edma.c b/drivers/dma/fsl-edma.c index
> 915eec3..cf8b06c 100644
> --- a/drivers/dma/fsl-edma.c
> +++ b/drivers/dma/fsl-edma.c
> @@ -116,6 +116,10 @@
> BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
> BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
> BIT(DMA_SLAVE_BUSWIDTH_8_BYTES)
> +enum fsl_edma_pm_state {
> + RUNNING = 0,
> + SUSPENDED,
> +};
>
> struct fsl_edma_hw_tcd {
> __le32 saddr;
> @@ -147,6 +151,9 @@ struct fsl_edma_slave_config { struct fsl_edma_chan {
> struct virt_dma_chan vchan;
> enum dma_status status;
> + enum fsl_edma_pm_state pm_state;
> + bool idle;
> + u32 slave_id;
> struct fsl_edma_engine *edma;
> struct fsl_edma_desc *edesc;
> struct fsl_edma_slave_config fsc;
> @@ -298,6 +305,7 @@ static int fsl_edma_terminate_all(struct dma_chan
> *chan)
> spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
> fsl_edma_disable_request(fsl_chan);
> fsl_chan->edesc = NULL;
> + fsl_chan->idle = true;
> vchan_get_all_descriptors(&fsl_chan->vchan, &head);
> spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> vchan_dma_desc_free_list(&fsl_chan->vchan, &head); @@ -313,6
> +321,7 @@ static int fsl_edma_pause(struct dma_chan *chan)
> if (fsl_chan->edesc) {
> fsl_edma_disable_request(fsl_chan);
> fsl_chan->status = DMA_PAUSED;
> + fsl_chan->idle = true;
> }
> spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> return 0;
> @@ -327,6 +336,7 @@ static int fsl_edma_resume(struct dma_chan *chan)
> if (fsl_chan->edesc) {
> fsl_edma_enable_request(fsl_chan);
> fsl_chan->status = DMA_IN_PROGRESS;
> + fsl_chan->idle = false;
> }
> spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> return 0;
> @@ -648,6 +658,7 @@ static void fsl_edma_xfer_desc(struct fsl_edma_chan
> *fsl_chan)
> fsl_edma_set_tcd_regs(fsl_chan, fsl_chan->edesc->tcd[0].vtcd);
> fsl_edma_enable_request(fsl_chan);
> fsl_chan->status = DMA_IN_PROGRESS;
> + fsl_chan->idle = false;
> }
>
> static irqreturn_t fsl_edma_tx_handler(int irq, void *dev_id) @@ -676,6
> +687,7 @@ static irqreturn_t fsl_edma_tx_handler(int irq, void *dev_id)
> vchan_cookie_complete(&fsl_chan->edesc-
> >vdesc);
> fsl_chan->edesc = NULL;
> fsl_chan->status = DMA_COMPLETE;
> + fsl_chan->idle = true;
> } else {
> vchan_cyclic_callback(&fsl_chan->edesc-
> >vdesc);
> }
> @@ -704,6 +716,7 @@ static irqreturn_t fsl_edma_err_handler(int irq, void
> *dev_id)
> edma_writeb(fsl_edma, EDMA_CERR_CERR(ch),
> fsl_edma->membase + EDMA_CERR);
> fsl_edma->chans[ch].status = DMA_ERROR;
> + fsl_edma->chans[ch].idle = true;
> }
> }
> return IRQ_HANDLED;
> @@ -724,6 +737,12 @@ static void fsl_edma_issue_pending(struct dma_chan
> *chan)
>
> spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
>
> + if (unlikely(fsl_chan->pm_state != RUNNING)) {
> + spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> + /* cannot submit due to suspend */
> + return;
> + }
> +
> if (vchan_issue_pending(&fsl_chan->vchan) && !fsl_chan->edesc)
> fsl_edma_xfer_desc(fsl_chan);
>
> @@ -735,6 +754,7 @@ static struct dma_chan *fsl_edma_xlate(struct
> of_phandle_args *dma_spec, {
> struct fsl_edma_engine *fsl_edma = ofdma->of_dma_data;
> struct dma_chan *chan, *_chan;
> + struct fsl_edma_chan *fsl_chan;
> unsigned long chans_per_mux = fsl_edma->n_chans / DMAMUX_NR;
>
> if (dma_spec->args_count != 2)
> @@ -748,8 +768,10 @@ static struct dma_chan *fsl_edma_xlate(struct
> of_phandle_args *dma_spec,
> chan = dma_get_slave_channel(chan);
> if (chan) {
> chan->device->privatecnt++;
> - fsl_edma_chan_mux(to_fsl_edma_chan(chan),
> - dma_spec->args[1], true);
> + fsl_chan = to_fsl_edma_chan(chan);
> + fsl_chan->slave_id = dma_spec->args[1];
> + fsl_edma_chan_mux(fsl_chan, fsl_chan-
> >slave_id,
> + true);
> mutex_unlock(&fsl_edma->fsl_edma_mutex);
> return chan;
> }
> @@ -888,7 +910,9 @@ static int fsl_edma_probe(struct platform_device *pdev)
> struct fsl_edma_chan *fsl_chan = &fsl_edma->chans[i];
>
> fsl_chan->edma = fsl_edma;
> -
> + fsl_chan->pm_state = RUNNING;
> + fsl_chan->slave_id = 0;
> + fsl_chan->idle = true;
> fsl_chan->vchan.desc_free = fsl_edma_free_desc;
> vchan_init(&fsl_chan->vchan, &fsl_edma->dma_dev);
>
> @@ -959,6 +983,58 @@ static int fsl_edma_remove(struct platform_device
> *pdev)
> return 0;
> }
>
> +static int fsl_edma_suspend_late(struct device *dev) {
> + struct fsl_edma_engine *fsl_edma = dev_get_drvdata(dev);
> + struct fsl_edma_chan *fsl_chan;
> + unsigned long flags;
> + int i;
> +
> + for (i = 0; i < fsl_edma->n_chans; i++) {
> + fsl_chan = &fsl_edma->chans[i];
> + spin_lock_irqsave(&fsl_chan->vchan.lock, flags);
> + /* Make sure chan is idle or can not into suspend. */
> + if (unlikely(!fsl_chan->idle))
> + goto out;
> + fsl_chan->pm_state = SUSPENDED;
> + spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> + }
> + return 0;
> +
> +out:
> + for (; i >= 0; i--) {
> + fsl_chan = &fsl_edma->chans[i];
> + fsl_chan->pm_state = RUNNING;
> + spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
> + }
> + return -EBUSY;
> +}
> +
> +static int fsl_edma_resume_early(struct device *dev) {
> + struct fsl_edma_engine *fsl_edma = dev_get_drvdata(dev);
> + struct fsl_edma_chan *fsl_chan;
> + int i;
> +
> + for (i = 0; i < fsl_edma->n_chans; i++) {
> + fsl_chan = &fsl_edma->chans[i];
> + fsl_chan->pm_state = RUNNING;
> + edma_writew(fsl_edma, 0x0, fsl_edma->membase +
> EDMA_TCD_CSR(i));
> + if (fsl_chan->slave_id != 0)
> + fsl_edma_chan_mux(fsl_chan, fsl_chan->slave_id,
> true);
> + }
> +
> + edma_writel(fsl_edma, EDMA_CR_ERGA | EDMA_CR_ERCA,
> + fsl_edma->membase + EDMA_CR);
> +
> + return 0;
> +}
> +
> +static const struct dev_pm_ops fsl_edma_pm_ops = {
> + .suspend_late = fsl_edma_suspend_late,
> + .resume_early = fsl_edma_resume_early,
> +};
> +
> static const struct of_device_id fsl_edma_dt_ids[] = {
> { .compatible = "fsl,vf610-edma", },
> { /* sentinel */ }
> @@ -969,6 +1045,7 @@ static struct platform_driver fsl_edma_driver = {
> .driver = {
> .name = "fsl-edma",
> .of_match_table = fsl_edma_dt_ids,
> + .pm = &fsl_edma_pm_ops,
> },
> .probe = fsl_edma_probe,
> .remove = fsl_edma_remove,
> --
> 2.1.0.27.g96db324
>
>
> _______________________________________________
> linux-arm-kernel mailing list
> [email protected]
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
On Tue, Jul 21, 2015 at 3:56 AM, Yuan Yao <[email protected]> wrote:
> This add power management suspend/resume support for the fsl-edma
> driver.
>
> eDMA acted as a basic function used by others. What it needs to do is
> the two steps below to support power management.
>
> In fsl_edma_suspend_late:
> Check whether the DMA chan is idle and if it is not idle, stop PM
> operation.
You should try to quiesce the device on suspend instead of depending
on itself to be happen in idle and failing if it is not.
Regards,
Leo
Hi Leo,
Thanks for your review.
About those two methods for DMA suspend that you have mentioned. We have a lot of the discussions in other DMA driver like DMA for Freescale PowerPC.
Finally, we think the device which used the DMA transmission service should cancel the transmission service in its suspend.
So DMA in suspend should be idle.
Once the DMA in late_suspend is not be idle, we think some driver haven't canceled the DMA transmission. So maybe something is error when other driver in suspend.
In the case, we should return failed to stop PM. DMA should not make a choice for other drivers(which used DMA) to force stop DMA transmission.
Thanks.
Best Regards,
Yuan Yao
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of Li Yang
> Sent: Friday, August 14, 2015 4:58 AM
> To: Yuan Yao-B46683
> Cc: Vinod Koul; [email protected]; Arnd Bergmann; Dan Williams;
> [email protected]; lkml; [email protected]; linux-
> [email protected]
> Subject: Re: [PATCH v2] dmaengine: fsl-edma: add PM suspend/resume
> support
>
> On Tue, Jul 21, 2015 at 3:56 AM, Yuan Yao <[email protected]> wrote:
> > This add power management suspend/resume support for the fsl-edma
> > driver.
> >
> > eDMA acted as a basic function used by others. What it needs to do is
> > the two steps below to support power management.
> >
> > In fsl_edma_suspend_late:
> > Check whether the DMA chan is idle and if it is not idle, stop PM
> > operation.
>
> You should try to quiesce the device on suspend instead of depending on itself
> to be happen in idle and failing if it is not.
>
> Regards,
> Leo
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
On Fri, Aug 14, 2015 at 1:24 AM, Yao Yuan <[email protected]> wrote:
> Hi Leo,
>
> Thanks for your review.
> About those two methods for DMA suspend that you have mentioned. We have a lot of the discussions in other DMA driver like DMA for Freescale PowerPC.
>
> Finally, we think the device which used the DMA transmission service should cancel the transmission service in its suspend.
> So DMA in suspend should be idle.
If that's the case you should clearly state this in the commit message
and in code, although I don't know if it is safe to make such
assumption. There could be user of the DMA that doesn't track the
completion of transfers.
>
> Once the DMA in late_suspend is not be idle, we think some driver haven't canceled the DMA transmission. So maybe something is error when other driver in suspend.
>
> In the case, we should return failed to stop PM. DMA should not make a choice for other drivers(which used DMA) to force stop DMA transmission.
The suspend entrance should be terminated by wakeup events and only
critical issues. I don't think we should just terminate the suspend
entrance just because having on-going I/O without even try to stop it.
>
> Thanks.
>
> Best Regards,
> Yuan Yao
>
>> -----Original Message-----
>> From: [email protected] [mailto:[email protected]] On Behalf Of Li Yang
>> Sent: Friday, August 14, 2015 4:58 AM
>> To: Yuan Yao-B46683
>> Cc: Vinod Koul; [email protected]; Arnd Bergmann; Dan Williams;
>> [email protected]; lkml; [email protected]; linux-
>> [email protected]
>> Subject: Re: [PATCH v2] dmaengine: fsl-edma: add PM suspend/resume
>> support
>>
>> On Tue, Jul 21, 2015 at 3:56 AM, Yuan Yao <[email protected]> wrote:
>> > This add power management suspend/resume support for the fsl-edma
>> > driver.
>> >
>> > eDMA acted as a basic function used by others. What it needs to do is
>> > the two steps below to support power management.
>> >
>> > In fsl_edma_suspend_late:
>> > Check whether the DMA chan is idle and if it is not idle, stop PM
>> > operation.
>>
>> You should try to quiesce the device on suspend instead of depending on itself
>> to be happen in idle and failing if it is not.
>>
>> Regards,
>> Leo
--
- Leo
On Sat, Aug 15, 2015 at 7:48 AM, pku.leo < [email protected] > wrote:
> On Fri, Aug 14, 2015 at 1:24 AM, Yao Yuan <[email protected]> wrote:
> > Hi Leo,
> >
> > Thanks for your review.
> > About those two methods for DMA suspend that you have mentioned. We
> have a lot of the discussions in other DMA driver like DMA for Freescale
> PowerPC.
> >
> > Finally, we think the device which used the DMA transmission service should
> cancel the transmission service in its suspend.
> > So DMA in suspend should be idle.
>
> If that's the case you should clearly state this in the commit message and in
> code, although I don't know if it is safe to make such assumption. There could
> be user of the DMA that doesn't track the completion of transfers.
I think it should be safe. In my opinion, even some client(the user of the DMA) forget to cancel its DMA transmission,
It will just lead to PM failed but no other system and data risk.
Although we should first fix the behavior of the client.
Once you are no need the DMA transmission, why not stop it?
Is it right?
> >
> > Once the DMA in late_suspend is not be idle, we think some driver haven't
> canceled the DMA transmission. So maybe something is error when other
> driver in suspend.
> >
> > In the case, we should return failed to stop PM. DMA should not make a
> choice for other drivers(which used DMA) to force stop DMA transmission.
>
> The suspend entrance should be terminated by wakeup events and only critical
> issues. I don't think we should just terminate the suspend entrance just
> because having on-going I/O without even try to stop it.
The graceful behavior would be to for client to PAUSE or terminate and then suspend
followed by DMA suspend.
We need to rely on client doing the right thing here.
The DMA should not make a decision instead of client.
If the DMA is not idle in DMA suspend, it should be the client's issue.
We don't know what the client really want to do, so just return the non-success value.
I'm not sure my description is clear. So we may refer the discussion about the DMA PM support before.
Such as "DMA: Freescale: add suspend resume functions for DMA driver"
Like:https://lkml.org/lkml/2014/5/21/1
> >
> > Thanks.
> >
> > Best Regards,
> > Yuan Yao
> >
> >> -----Original Message-----
> >> From: [email protected] [mailto:[email protected]] On Behalf Of Li
> >> Yang
> >> Sent: Friday, August 14, 2015 4:58 AM
> >> To: Yuan Yao-B46683
> >> Cc: Vinod Koul; [email protected]; Arnd Bergmann; Dan Williams;
> >> [email protected]; lkml;
> >> [email protected]; linux- [email protected]
> >> Subject: Re: [PATCH v2] dmaengine: fsl-edma: add PM suspend/resume
> >> support
> >>
> >> On Tue, Jul 21, 2015 at 3:56 AM, Yuan Yao <[email protected]> wrote:
> >> > This add power management suspend/resume support for the fsl-edma
> >> > driver.
> >> >
> >> > eDMA acted as a basic function used by others. What it needs to do
> >> > is the two steps below to support power management.
> >> >
> >> > In fsl_edma_suspend_late:
> >> > Check whether the DMA chan is idle and if it is not idle, stop PM
> >> > operation.
> >>
> >> You should try to quiesce the device on suspend instead of depending
> >> on itself to be happen in idle and failing if it is not.
> >>
> >> Regards,
> >> Leo
>
>
>
> --
> - Leo
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
Hi Yao.
On 17/08/15 13:59, Yao Yuan wrote:
> On Sat, Aug 15, 2015 at 7:48 AM, pku.leo < [email protected] > wrote:
>> On Fri, Aug 14, 2015 at 1:24 AM, Yao Yuan <[email protected]> wrote:
>>> Hi Leo,
>>>
>>> Thanks for your review.
>>> About those two methods for DMA suspend that you have mentioned. We
>> have a lot of the discussions in other DMA driver like DMA for Freescale
>> PowerPC.
>>> Finally, we think the device which used the DMA transmission service should
>> cancel the transmission service in its suspend.
>>> So DMA in suspend should be idle.
>> If that's the case you should clearly state this in the commit message and in
>> code, although I don't know if it is safe to make such assumption. There could
>> be user of the DMA that doesn't track the completion of transfers.
> I think it should be safe. In my opinion, even some client(the user of the DMA) forget to cancel its DMA transmission,
> It will just lead to PM failed but no other system and data risk.
> Although we should first fix the behavior of the client.
> Once you are no need the DMA transmission, why not stop it?
>
> Is it right?
Think of it from the end user perspective. Would you like your laptop (or whatever) to refuse to suspend because of this condition? The user may well expect that closing the lid on their laptop will reliably lead to it suspending to ram. Returning a failure here could result in a loss of data if the condition is not detected and the machine subsequently runs out of power.
I do agree that whatever is submitting DMA should be stopped first; ideally this driver would always be idle because whatever producers of work exist would already have been quiesced and output flushed.
Regards,
Nigel
Hi Nigel,
On Mon, Aug 17, 2015 at 2:49 PM, Nigel Cunningham < [email protected] > wrote:
> On 17/08/15 13:59, Yao Yuan wrote:
> > On Sat, Aug 15, 2015 at 7:48 AM, pku.leo < [email protected] > wrote:
> >> On Fri, Aug 14, 2015 at 1:24 AM, Yao Yuan <[email protected]> wrote:
> >>> Hi Leo,
> >>>
> >>> Thanks for your review.
> >>> About those two methods for DMA suspend that you have mentioned.
> We
> >> have a lot of the discussions in other DMA driver like DMA for
> >> Freescale PowerPC.
> >>> Finally, we think the device which used the DMA transmission service
> >>> should
> >> cancel the transmission service in its suspend.
> >>> So DMA in suspend should be idle.
> >> If that's the case you should clearly state this in the commit
> >> message and in code, although I don't know if it is safe to make such
> >> assumption. There could be user of the DMA that doesn't track the
> completion of transfers.
> > I think it should be safe. In my opinion, even some client(the user of
> > the DMA) forget to cancel its DMA transmission, It will just lead to PM failed
> but no other system and data risk.
> > Although we should first fix the behavior of the client.
> > Once you are no need the DMA transmission, why not stop it?
> >
> > Is it right?
> Think of it from the end user perspective. Would you like your laptop (or
> whatever) to refuse to suspend because of this condition? The user may well
> expect that closing the lid on their laptop will reliably lead to it suspending to
> ram. Returning a failure here could result in a loss of data if the condition is not
> detected and the machine subsequently runs out of power.
>
Yes, the user may well expect that closing the lid on their laptop will reliably lead to it suspending to ram.
So the client(the user of the DMA) must to PAUSE or terminate the DMA transmission.
We need to rely on client doing the right thing here.
The DMA should not make a decision instead of client.
If the DMA is not idle in DMA suspend, it should be the client's issue.
We don't know what the client really want to do, so just return the non-success value.
> I do agree that whatever is submitting DMA should be stopped first; ideally this
> driver would always be idle because whatever producers of work exist would
> already have been quiesced and output flushed.
>
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
On Mon, Aug 17, 2015 at 2:22 AM, Yao Yuan <[email protected]> wrote:
> Hi Nigel,
>
> On Mon, Aug 17, 2015 at 2:49 PM, Nigel Cunningham < [email protected] > wrote:
>> On 17/08/15 13:59, Yao Yuan wrote:
>> > On Sat, Aug 15, 2015 at 7:48 AM, pku.leo < [email protected] > wrote:
>> >> On Fri, Aug 14, 2015 at 1:24 AM, Yao Yuan <[email protected]> wrote:
>> >>> Hi Leo,
>> >>>
>> >>> Thanks for your review.
>> >>> About those two methods for DMA suspend that you have mentioned.
>> We
>> >> have a lot of the discussions in other DMA driver like DMA for
>> >> Freescale PowerPC.
>> >>> Finally, we think the device which used the DMA transmission service
>> >>> should
>> >> cancel the transmission service in its suspend.
>> >>> So DMA in suspend should be idle.
>> >> If that's the case you should clearly state this in the commit
>> >> message and in code, although I don't know if it is safe to make such
>> >> assumption. There could be user of the DMA that doesn't track the
>> completion of transfers.
>> > I think it should be safe. In my opinion, even some client(the user of
>> > the DMA) forget to cancel its DMA transmission, It will just lead to PM failed
>> but no other system and data risk.
>> > Although we should first fix the behavior of the client.
>> > Once you are no need the DMA transmission, why not stop it?
>> >
>> > Is it right?
>> Think of it from the end user perspective. Would you like your laptop (or
>> whatever) to refuse to suspend because of this condition? The user may well
>> expect that closing the lid on their laptop will reliably lead to it suspending to
>> ram. Returning a failure here could result in a loss of data if the condition is not
>> detected and the machine subsequently runs out of power.
>>
>
> Yes, the user may well expect that closing the lid on their laptop will reliably lead to it suspending to ram.
> So the client(the user of the DMA) must to PAUSE or terminate the DMA transmission.
>
> We need to rely on client doing the right thing here.
> The DMA should not make a decision instead of client.
> If the DMA is not idle in DMA suspend, it should be the client's issue.
> We don't know what the client really want to do, so just return the non-success value.
The problem here is that neither the client nor the DMA controller
driver should easily decide to stop the suspend entrance and rollback.
I don't think the non-idle situation is serious enough to cause a
rollback. You should do whatever can be done with the DMA
controller(such as stop the controller and leave whatever to be done
to the wake up) and continue with the suspend.
Regards,
Leo
On Mon, Aug 17, 2015 at 02:10:46PM -0500, Li Yang wrote:
> >> Think of it from the end user perspective. Would you like your laptop (or
> >> whatever) to refuse to suspend because of this condition? The user may well
> >> expect that closing the lid on their laptop will reliably lead to it suspending to
> >> ram. Returning a failure here could result in a loss of data if the condition is not
> >> detected and the machine subsequently runs out of power.
> >>
> >
> > Yes, the user may well expect that closing the lid on their laptop will reliably lead to it suspending to ram.
> > So the client(the user of the DMA) must to PAUSE or terminate the DMA transmission.
> >
> > We need to rely on client doing the right thing here.
> > The DMA should not make a decision instead of client.
> > If the DMA is not idle in DMA suspend, it should be the client's issue.
> > We don't know what the client really want to do, so just return the non-success value.
>
> The problem here is that neither the client nor the DMA controller
> driver should easily decide to stop the suspend entrance and rollback.
> I don't think the non-idle situation is serious enough to cause a
> rollback. You should do whatever can be done with the DMA
> controller(such as stop the controller and leave whatever to be done
> to the wake up) and continue with the suspend.
Ideally yes client should suspend first and dmaengine driver then being idle
when suspend is invoked. But we know we are in idle world!
So, driver should ensure it suspends the active channels and then goes to
suspend and restores that on resume
--
~Vinod