diff mbox series

[v1] dmaengine: tegra: Use relaxed versions of readl/writel

Message ID 20190424231708.21219-1-digetx@gmail.com (mailing list archive)
State Changes Requested
Headers show
Series [v1] dmaengine: tegra: Use relaxed versions of readl/writel | expand

Commit Message

Dmitry Osipenko April 24, 2019, 11:17 p.m. UTC
The readl/writel functions are inserting memory barrier in order to
ensure that memory stores are completed. On Tegra20 and Tegra30 this
results in L2 cache syncing which isn't a cheapest operation. The
tegra20-apb-dma driver doesn't need to synchronize generic memory
accesses, hence use the relaxed versions of the functions.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

Comments

Jon Hunter April 26, 2019, 9:52 a.m. UTC | #1
On 25/04/2019 00:17, Dmitry Osipenko wrote:
> The readl/writel functions are inserting memory barrier in order to
> ensure that memory stores are completed. On Tegra20 and Tegra30 this
> results in L2 cache syncing which isn't a cheapest operation. The
> tegra20-apb-dma driver doesn't need to synchronize generic memory
> accesses, hence use the relaxed versions of the functions.

Do you mean device-io accesses here as this is not generic memory?

Although there may not be any issues with this change, I think I need a
bit more convincing that we should do this given that we have had it
this way for sometime and I would not like to see us introduce any
regressions as this point without being 100% certain we would not.
Ideally, if I had some good extensive tests I could run to hammer the
DMA for all configurations with different combinations of channels
running simultaneously then we could test this, but right now I don't :-(

Have you ...
1. Tested both cyclic and scatter-gather transfers?
2. Stress tested simultaneous transfers with various different
   configurations?
3. Quantified the actual performance benefit of this change so we can
   understand how much of a performance boost this offers?

Cheers
Jon
Dmitry Osipenko April 26, 2019, 10:45 a.m. UTC | #2
26.04.2019 12:52, Jon Hunter пишет:
> 
> On 25/04/2019 00:17, Dmitry Osipenko wrote:
>> The readl/writel functions are inserting memory barrier in order to
>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
>> results in L2 cache syncing which isn't a cheapest operation. The
>> tegra20-apb-dma driver doesn't need to synchronize generic memory
>> accesses, hence use the relaxed versions of the functions.
> 
> Do you mean device-io accesses here as this is not generic memory?

Yes. The IOMEM accesses within are always ordered and uncached, while
generic memory accesses are out-of-order and cached.

> Although there may not be any issues with this change, I think I need a
> bit more convincing that we should do this given that we have had it
> this way for sometime and I would not like to see us introduce any
> regressions as this point without being 100% certain we would not.
> Ideally, if I had some good extensive tests I could run to hammer the
> DMA for all configurations with different combinations of channels
> running simultaneously then we could test this, but right now I don't :-(
> 
> Have you ...
> 1. Tested both cyclic and scatter-gather transfers?
> 2. Stress tested simultaneous transfers with various different
>    configurations?
> 3. Quantified the actual performance benefit of this change so we can
>    understand how much of a performance boost this offers?

Actually I found a case where this change causes a problem, I'm seeing
I2C transfer timeout for touchscreen and it breaks the touch input.
Indeed, I haven't tested this patch very well.

And the fix is this:

@@ -1592,6 +1592,8 @@ static int tegra_dma_runtime_suspend(struct device
*dev)
 						  TEGRA_APBDMA_CHAN_WCOUNT);
 	}

+	dsb();
+
 	clk_disable_unprepare(tdma->dma_clk);

 	return 0;


Apparently the problem is that CLK/DMA (PPSB/APB) accesses are
incoherent and CPU disables clock before writes are reaching DMA controller.

I'd say that cyclic and scatter-gather transfers are now tested. I also
made some more testing of simultaneous transfers.

Quantifying performance probably won't be easy to make as the DMA
read/writes are not on any kind of code's hot-path.

Jon, are you still insisting about to drop this patch or you will be
fine with the v2 that will have the dsb() in place?
Jon Hunter April 26, 2019, 11:13 a.m. UTC | #3
On 26/04/2019 11:45, Dmitry Osipenko wrote:
> 26.04.2019 12:52, Jon Hunter пишет:
>>
>> On 25/04/2019 00:17, Dmitry Osipenko wrote:
>>> The readl/writel functions are inserting memory barrier in order to
>>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
>>> results in L2 cache syncing which isn't a cheapest operation. The
>>> tegra20-apb-dma driver doesn't need to synchronize generic memory
>>> accesses, hence use the relaxed versions of the functions.
>>
>> Do you mean device-io accesses here as this is not generic memory?
> 
> Yes. The IOMEM accesses within are always ordered and uncached, while
> generic memory accesses are out-of-order and cached.
> 
>> Although there may not be any issues with this change, I think I need a
>> bit more convincing that we should do this given that we have had it
>> this way for sometime and I would not like to see us introduce any
>> regressions as this point without being 100% certain we would not.
>> Ideally, if I had some good extensive tests I could run to hammer the
>> DMA for all configurations with different combinations of channels
>> running simultaneously then we could test this, but right now I don't :-(
>>
>> Have you ...
>> 1. Tested both cyclic and scatter-gather transfers?
>> 2. Stress tested simultaneous transfers with various different
>>    configurations?
>> 3. Quantified the actual performance benefit of this change so we can
>>    understand how much of a performance boost this offers?
> 
> Actually I found a case where this change causes a problem, I'm seeing
> I2C transfer timeout for touchscreen and it breaks the touch input.
> Indeed, I haven't tested this patch very well.
> 
> And the fix is this:
> 
> @@ -1592,6 +1592,8 @@ static int tegra_dma_runtime_suspend(struct device
> *dev)
>  						  TEGRA_APBDMA_CHAN_WCOUNT);
>  	}
> 
> +	dsb();
> +
>  	clk_disable_unprepare(tdma->dma_clk);
> 
>  	return 0;
> 
> 
> Apparently the problem is that CLK/DMA (PPSB/APB) accesses are
> incoherent and CPU disables clock before writes are reaching DMA controller.
> 
> I'd say that cyclic and scatter-gather transfers are now tested. I also
> made some more testing of simultaneous transfers.
> 
> Quantifying performance probably won't be easy to make as the DMA
> read/writes are not on any kind of code's hot-path.

So why make the change?
> Jon, are you still insisting about to drop this patch or you will be
> fine with the v2 that will have the dsb() in place?

If we can't quantify the performance gain, then it is difficult to
justify the change. I would also be concerned if that is the only place
we need an explicit dsb.

Cheers
Jon
Vinod Koul April 26, 2019, 12:01 p.m. UTC | #4
On 25-04-19, 02:17, Dmitry Osipenko wrote:
> The readl/writel functions are inserting memory barrier in order to
> ensure that memory stores are completed. On Tegra20 and Tegra30 this
> results in L2 cache syncing which isn't a cheapest operation. The
> tegra20-apb-dma driver doesn't need to synchronize generic memory
> accesses, hence use the relaxed versions of the functions.

Subsystem name is **dmaengine** not dma! Please fix that
Dmitry Osipenko April 26, 2019, 12:18 p.m. UTC | #5
26.04.2019 14:13, Jon Hunter пишет:
> 
> On 26/04/2019 11:45, Dmitry Osipenko wrote:
>> 26.04.2019 12:52, Jon Hunter пишет:
>>>
>>> On 25/04/2019 00:17, Dmitry Osipenko wrote:
>>>> The readl/writel functions are inserting memory barrier in order to
>>>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
>>>> results in L2 cache syncing which isn't a cheapest operation. The
>>>> tegra20-apb-dma driver doesn't need to synchronize generic memory
>>>> accesses, hence use the relaxed versions of the functions.
>>>
>>> Do you mean device-io accesses here as this is not generic memory?
>>
>> Yes. The IOMEM accesses within are always ordered and uncached, while
>> generic memory accesses are out-of-order and cached.
>>
>>> Although there may not be any issues with this change, I think I need a
>>> bit more convincing that we should do this given that we have had it
>>> this way for sometime and I would not like to see us introduce any
>>> regressions as this point without being 100% certain we would not.
>>> Ideally, if I had some good extensive tests I could run to hammer the
>>> DMA for all configurations with different combinations of channels
>>> running simultaneously then we could test this, but right now I don't :-(
>>>
>>> Have you ...
>>> 1. Tested both cyclic and scatter-gather transfers?
>>> 2. Stress tested simultaneous transfers with various different
>>>    configurations?
>>> 3. Quantified the actual performance benefit of this change so we can
>>>    understand how much of a performance boost this offers?
>>
>> Actually I found a case where this change causes a problem, I'm seeing
>> I2C transfer timeout for touchscreen and it breaks the touch input.
>> Indeed, I haven't tested this patch very well.
>>
>> And the fix is this:
>>
>> @@ -1592,6 +1592,8 @@ static int tegra_dma_runtime_suspend(struct device
>> *dev)
>>  						  TEGRA_APBDMA_CHAN_WCOUNT);
>>  	}
>>
>> +	dsb();
>> +
>>  	clk_disable_unprepare(tdma->dma_clk);
>>
>>  	return 0;
>>
>>
>> Apparently the problem is that CLK/DMA (PPSB/APB) accesses are
>> incoherent and CPU disables clock before writes are reaching DMA controller.
>>
>> I'd say that cyclic and scatter-gather transfers are now tested. I also
>> made some more testing of simultaneous transfers.
>>
>> Quantifying performance probably won't be easy to make as the DMA
>> read/writes are not on any kind of code's hot-path.
> 
> So why make the change?

For consistency.

>> Jon, are you still insisting about to drop this patch or you will be
>> fine with the v2 that will have the dsb() in place?
> 
> If we can't quantify the performance gain, then it is difficult to
> justify the change. I would also be concerned if that is the only place
> we need an explicit dsb.

Maybe it won't hurt to add dsb to the ISR as well. But okay, let's drop
this patch for now.
Dmitry Osipenko April 26, 2019, 12:19 p.m. UTC | #6
26.04.2019 15:01, Vinod Koul пишет:
> On 25-04-19, 02:17, Dmitry Osipenko wrote:
>> The readl/writel functions are inserting memory barrier in order to
>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
>> results in L2 cache syncing which isn't a cheapest operation. The
>> tegra20-apb-dma driver doesn't need to synchronize generic memory
>> accesses, hence use the relaxed versions of the functions.
> 
> Subsystem name is **dmaengine** not dma! Please fix that
> 

Looks like you answered to a wrong email.
Dmitry Osipenko April 26, 2019, 12:42 p.m. UTC | #7
26.04.2019 15:18, Dmitry Osipenko пишет:
> 26.04.2019 14:13, Jon Hunter пишет:
>>
>> On 26/04/2019 11:45, Dmitry Osipenko wrote:
>>> 26.04.2019 12:52, Jon Hunter пишет:
>>>>
>>>> On 25/04/2019 00:17, Dmitry Osipenko wrote:
>>>>> The readl/writel functions are inserting memory barrier in order to
>>>>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
>>>>> results in L2 cache syncing which isn't a cheapest operation. The
>>>>> tegra20-apb-dma driver doesn't need to synchronize generic memory
>>>>> accesses, hence use the relaxed versions of the functions.
>>>>
>>>> Do you mean device-io accesses here as this is not generic memory?
>>>
>>> Yes. The IOMEM accesses within are always ordered and uncached, while
>>> generic memory accesses are out-of-order and cached.
>>>
>>>> Although there may not be any issues with this change, I think I need a
>>>> bit more convincing that we should do this given that we have had it
>>>> this way for sometime and I would not like to see us introduce any
>>>> regressions as this point without being 100% certain we would not.
>>>> Ideally, if I had some good extensive tests I could run to hammer the
>>>> DMA for all configurations with different combinations of channels
>>>> running simultaneously then we could test this, but right now I don't :-(
>>>>
>>>> Have you ...
>>>> 1. Tested both cyclic and scatter-gather transfers?
>>>> 2. Stress tested simultaneous transfers with various different
>>>>    configurations?
>>>> 3. Quantified the actual performance benefit of this change so we can
>>>>    understand how much of a performance boost this offers?
>>>
>>> Actually I found a case where this change causes a problem, I'm seeing
>>> I2C transfer timeout for touchscreen and it breaks the touch input.
>>> Indeed, I haven't tested this patch very well.
>>>
>>> And the fix is this:
>>>
>>> @@ -1592,6 +1592,8 @@ static int tegra_dma_runtime_suspend(struct device
>>> *dev)
>>>  						  TEGRA_APBDMA_CHAN_WCOUNT);
>>>  	}
>>>
>>> +	dsb();
>>> +
>>>  	clk_disable_unprepare(tdma->dma_clk);
>>>
>>>  	return 0;
>>>
>>>
>>> Apparently the problem is that CLK/DMA (PPSB/APB) accesses are
>>> incoherent and CPU disables clock before writes are reaching DMA controller.
>>>
>>> I'd say that cyclic and scatter-gather transfers are now tested. I also
>>> made some more testing of simultaneous transfers.
>>>
>>> Quantifying performance probably won't be easy to make as the DMA
>>> read/writes are not on any kind of code's hot-path.
>>
>> So why make the change?
> 
> For consistency.
> 
>>> Jon, are you still insisting about to drop this patch or you will be
>>> fine with the v2 that will have the dsb() in place?
>>
>> If we can't quantify the performance gain, then it is difficult to
>> justify the change. I would also be concerned if that is the only place
>> we need an explicit dsb.
> 
> Maybe it won't hurt to add dsb to the ISR as well. But okay, let's drop
> this patch for now.
> 

Jon, it occurred to me that there still should be a problem with the
writel() ordering in the driver because writel() ensures that memory
stores are completed *before* the write occurs and hence translates into
iowmb() + writel_relaxed() [0]. Thus the last write will always happen
asynchronously in regards to clk accesses.

[0]
https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/arch/arm/include/asm/io.h#n311
Vinod Koul April 26, 2019, 12:47 p.m. UTC | #8
On 26-04-19, 15:19, Dmitry Osipenko wrote:
> 26.04.2019 15:01, Vinod Koul пишет:
> > On 25-04-19, 02:17, Dmitry Osipenko wrote:
> >> The readl/writel functions are inserting memory barrier in order to
> >> ensure that memory stores are completed. On Tegra20 and Tegra30 this
> >> results in L2 cache syncing which isn't a cheapest operation. The
> >> tegra20-apb-dma driver doesn't need to synchronize generic memory
> >> accesses, hence use the relaxed versions of the functions.
> > 
> > Subsystem name is **dmaengine** not dma! Please fix that
> > 
> 
> Looks like you answered to a wrong email.

Yes looks like I did, apologies for that :(
Dmitry Osipenko April 26, 2019, 1:03 p.m. UTC | #9
26.04.2019 15:42, Dmitry Osipenko пишет:
> 26.04.2019 15:18, Dmitry Osipenko пишет:
>> 26.04.2019 14:13, Jon Hunter пишет:
>>>
>>> On 26/04/2019 11:45, Dmitry Osipenko wrote:
>>>> 26.04.2019 12:52, Jon Hunter пишет:
>>>>>
>>>>> On 25/04/2019 00:17, Dmitry Osipenko wrote:
>>>>>> The readl/writel functions are inserting memory barrier in order to
>>>>>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
>>>>>> results in L2 cache syncing which isn't a cheapest operation. The
>>>>>> tegra20-apb-dma driver doesn't need to synchronize generic memory
>>>>>> accesses, hence use the relaxed versions of the functions.
>>>>>
>>>>> Do you mean device-io accesses here as this is not generic memory?
>>>>
>>>> Yes. The IOMEM accesses within are always ordered and uncached, while
>>>> generic memory accesses are out-of-order and cached.
>>>>
>>>>> Although there may not be any issues with this change, I think I need a
>>>>> bit more convincing that we should do this given that we have had it
>>>>> this way for sometime and I would not like to see us introduce any
>>>>> regressions as this point without being 100% certain we would not.
>>>>> Ideally, if I had some good extensive tests I could run to hammer the
>>>>> DMA for all configurations with different combinations of channels
>>>>> running simultaneously then we could test this, but right now I don't :-(
>>>>>
>>>>> Have you ...
>>>>> 1. Tested both cyclic and scatter-gather transfers?
>>>>> 2. Stress tested simultaneous transfers with various different
>>>>>    configurations?
>>>>> 3. Quantified the actual performance benefit of this change so we can
>>>>>    understand how much of a performance boost this offers?
>>>>
>>>> Actually I found a case where this change causes a problem, I'm seeing
>>>> I2C transfer timeout for touchscreen and it breaks the touch input.
>>>> Indeed, I haven't tested this patch very well.
>>>>
>>>> And the fix is this:
>>>>
>>>> @@ -1592,6 +1592,8 @@ static int tegra_dma_runtime_suspend(struct device
>>>> *dev)
>>>>  						  TEGRA_APBDMA_CHAN_WCOUNT);
>>>>  	}
>>>>
>>>> +	dsb();
>>>> +
>>>>  	clk_disable_unprepare(tdma->dma_clk);
>>>>
>>>>  	return 0;
>>>>
>>>>
>>>> Apparently the problem is that CLK/DMA (PPSB/APB) accesses are
>>>> incoherent and CPU disables clock before writes are reaching DMA controller.
>>>>
>>>> I'd say that cyclic and scatter-gather transfers are now tested. I also
>>>> made some more testing of simultaneous transfers.
>>>>
>>>> Quantifying performance probably won't be easy to make as the DMA
>>>> read/writes are not on any kind of code's hot-path.
>>>
>>> So why make the change?
>>
>> For consistency.
>>
>>>> Jon, are you still insisting about to drop this patch or you will be
>>>> fine with the v2 that will have the dsb() in place?
>>>
>>> If we can't quantify the performance gain, then it is difficult to
>>> justify the change. I would also be concerned if that is the only place
>>> we need an explicit dsb.
>>
>> Maybe it won't hurt to add dsb to the ISR as well. But okay, let's drop
>> this patch for now.
>>
> 
> Jon, it occurred to me that there still should be a problem with the
> writel() ordering in the driver because writel() ensures that memory
> stores are completed *before* the write occurs and hence translates into
> iowmb() + writel_relaxed() [0]. Thus the last write will always happen
> asynchronously in regards to clk accesses.
> 
> [0]
> https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/arch/arm/include/asm/io.h#n311
> 

Also please note that iowmb() translates into wmb() if
CONFIG_ARM_DMA_MEM_BUFFERABLE=y and sometime ago I was profiling host1x
driver job submission performance and have seen cases where wmb() could
take up to 1ms on T20 due to L2 syncing if there are outstanding memory
writes in the cache (or even more, I don't remember exactly already how
bad it was..).

Altogether, I think the usage of readl/writel in pretty much all of
Tegra drivers is plainly wrong and explicit dsb() shall be used in
places where hardware synchronization is really needed.
Thierry Reding April 26, 2019, 3:11 p.m. UTC | #10
On Fri, Apr 26, 2019 at 04:03:08PM +0300, Dmitry Osipenko wrote:
> 26.04.2019 15:42, Dmitry Osipenko пишет:
> > 26.04.2019 15:18, Dmitry Osipenko пишет:
> >> 26.04.2019 14:13, Jon Hunter пишет:
> >>>
> >>> On 26/04/2019 11:45, Dmitry Osipenko wrote:
> >>>> 26.04.2019 12:52, Jon Hunter пишет:
> >>>>>
> >>>>> On 25/04/2019 00:17, Dmitry Osipenko wrote:
> >>>>>> The readl/writel functions are inserting memory barrier in order to
> >>>>>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
> >>>>>> results in L2 cache syncing which isn't a cheapest operation. The
> >>>>>> tegra20-apb-dma driver doesn't need to synchronize generic memory
> >>>>>> accesses, hence use the relaxed versions of the functions.
> >>>>>
> >>>>> Do you mean device-io accesses here as this is not generic memory?
> >>>>
> >>>> Yes. The IOMEM accesses within are always ordered and uncached, while
> >>>> generic memory accesses are out-of-order and cached.
> >>>>
> >>>>> Although there may not be any issues with this change, I think I need a
> >>>>> bit more convincing that we should do this given that we have had it
> >>>>> this way for sometime and I would not like to see us introduce any
> >>>>> regressions as this point without being 100% certain we would not.
> >>>>> Ideally, if I had some good extensive tests I could run to hammer the
> >>>>> DMA for all configurations with different combinations of channels
> >>>>> running simultaneously then we could test this, but right now I don't :-(
> >>>>>
> >>>>> Have you ...
> >>>>> 1. Tested both cyclic and scatter-gather transfers?
> >>>>> 2. Stress tested simultaneous transfers with various different
> >>>>>    configurations?
> >>>>> 3. Quantified the actual performance benefit of this change so we can
> >>>>>    understand how much of a performance boost this offers?
> >>>>
> >>>> Actually I found a case where this change causes a problem, I'm seeing
> >>>> I2C transfer timeout for touchscreen and it breaks the touch input.
> >>>> Indeed, I haven't tested this patch very well.
> >>>>
> >>>> And the fix is this:
> >>>>
> >>>> @@ -1592,6 +1592,8 @@ static int tegra_dma_runtime_suspend(struct device
> >>>> *dev)
> >>>>  						  TEGRA_APBDMA_CHAN_WCOUNT);
> >>>>  	}
> >>>>
> >>>> +	dsb();
> >>>> +
> >>>>  	clk_disable_unprepare(tdma->dma_clk);
> >>>>
> >>>>  	return 0;
> >>>>
> >>>>
> >>>> Apparently the problem is that CLK/DMA (PPSB/APB) accesses are
> >>>> incoherent and CPU disables clock before writes are reaching DMA controller.
> >>>>
> >>>> I'd say that cyclic and scatter-gather transfers are now tested. I also
> >>>> made some more testing of simultaneous transfers.
> >>>>
> >>>> Quantifying performance probably won't be easy to make as the DMA
> >>>> read/writes are not on any kind of code's hot-path.
> >>>
> >>> So why make the change?
> >>
> >> For consistency.
> >>
> >>>> Jon, are you still insisting about to drop this patch or you will be
> >>>> fine with the v2 that will have the dsb() in place?
> >>>
> >>> If we can't quantify the performance gain, then it is difficult to
> >>> justify the change. I would also be concerned if that is the only place
> >>> we need an explicit dsb.
> >>
> >> Maybe it won't hurt to add dsb to the ISR as well. But okay, let's drop
> >> this patch for now.
> >>
> > 
> > Jon, it occurred to me that there still should be a problem with the
> > writel() ordering in the driver because writel() ensures that memory
> > stores are completed *before* the write occurs and hence translates into
> > iowmb() + writel_relaxed() [0]. Thus the last write will always happen
> > asynchronously in regards to clk accesses.
> > 
> > [0]
> > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/arch/arm/include/asm/io.h#n311
> > 
> 
> Also please note that iowmb() translates into wmb() if
> CONFIG_ARM_DMA_MEM_BUFFERABLE=y and sometime ago I was profiling host1x
> driver job submission performance and have seen cases where wmb() could
> take up to 1ms on T20 due to L2 syncing if there are outstanding memory
> writes in the cache (or even more, I don't remember exactly already how
> bad it was..).

This looks to be primarily caused by the fact that we have the L2X0
cache on Tegra20. So there's not really anything that can be done there
without potentially compromising correctness of the code.

> Altogether, I think the usage of readl/writel in pretty much all of
> Tegra drivers is plainly wrong and explicit dsb() shall be used in
> places where hardware synchronization is really needed.

I don't think that's an accurate observation. readl()/writel() are more
likely to be correct than the relaxed versions. You already saw yourself
that using the relaxed versions can easily introduce regressions.

Granted, readl()/writel() might add more memory barriers than strictly
necessary, and therefore they might in many cases be suboptimal. But, we
can't just go and engage in a wholesale conversion of all drivers. If we
do this, we need to very carefully audit every conversion to make sure
no regressions are introduced. This is especially complicated because
these would be subtle regressions and may be difficult to catch or
reproduce.

Also, we should avoid using primitives such as dsb in driver code to
avoid making the code too architecture specific.

Thierry
Dmitry Osipenko April 30, 2019, 12:25 p.m. UTC | #11
26.04.2019 18:11, Thierry Reding пишет:
> On Fri, Apr 26, 2019 at 04:03:08PM +0300, Dmitry Osipenko wrote:
>> 26.04.2019 15:42, Dmitry Osipenko пишет:
>>> 26.04.2019 15:18, Dmitry Osipenko пишет:
>>>> 26.04.2019 14:13, Jon Hunter пишет:
>>>>>
>>>>> On 26/04/2019 11:45, Dmitry Osipenko wrote:
>>>>>> 26.04.2019 12:52, Jon Hunter пишет:
>>>>>>>
>>>>>>> On 25/04/2019 00:17, Dmitry Osipenko wrote:
>>>>>>>> The readl/writel functions are inserting memory barrier in order to
>>>>>>>> ensure that memory stores are completed. On Tegra20 and Tegra30 this
>>>>>>>> results in L2 cache syncing which isn't a cheapest operation. The
>>>>>>>> tegra20-apb-dma driver doesn't need to synchronize generic memory
>>>>>>>> accesses, hence use the relaxed versions of the functions.
>>>>>>>
>>>>>>> Do you mean device-io accesses here as this is not generic memory?
>>>>>>
>>>>>> Yes. The IOMEM accesses within are always ordered and uncached, while
>>>>>> generic memory accesses are out-of-order and cached.
>>>>>>
>>>>>>> Although there may not be any issues with this change, I think I need a
>>>>>>> bit more convincing that we should do this given that we have had it
>>>>>>> this way for sometime and I would not like to see us introduce any
>>>>>>> regressions as this point without being 100% certain we would not.
>>>>>>> Ideally, if I had some good extensive tests I could run to hammer the
>>>>>>> DMA for all configurations with different combinations of channels
>>>>>>> running simultaneously then we could test this, but right now I don't :-(
>>>>>>>
>>>>>>> Have you ...
>>>>>>> 1. Tested both cyclic and scatter-gather transfers?
>>>>>>> 2. Stress tested simultaneous transfers with various different
>>>>>>>    configurations?
>>>>>>> 3. Quantified the actual performance benefit of this change so we can
>>>>>>>    understand how much of a performance boost this offers?
>>>>>>
>>>>>> Actually I found a case where this change causes a problem, I'm seeing
>>>>>> I2C transfer timeout for touchscreen and it breaks the touch input.
>>>>>> Indeed, I haven't tested this patch very well.
>>>>>>
>>>>>> And the fix is this:
>>>>>>
>>>>>> @@ -1592,6 +1592,8 @@ static int tegra_dma_runtime_suspend(struct device
>>>>>> *dev)
>>>>>>  						  TEGRA_APBDMA_CHAN_WCOUNT);
>>>>>>  	}
>>>>>>
>>>>>> +	dsb();
>>>>>> +
>>>>>>  	clk_disable_unprepare(tdma->dma_clk);
>>>>>>
>>>>>>  	return 0;
>>>>>>
>>>>>>
>>>>>> Apparently the problem is that CLK/DMA (PPSB/APB) accesses are
>>>>>> incoherent and CPU disables clock before writes are reaching DMA controller.
>>>>>>
>>>>>> I'd say that cyclic and scatter-gather transfers are now tested. I also
>>>>>> made some more testing of simultaneous transfers.
>>>>>>
>>>>>> Quantifying performance probably won't be easy to make as the DMA
>>>>>> read/writes are not on any kind of code's hot-path.
>>>>>
>>>>> So why make the change?
>>>>
>>>> For consistency.
>>>>
>>>>>> Jon, are you still insisting about to drop this patch or you will be
>>>>>> fine with the v2 that will have the dsb() in place?
>>>>>
>>>>> If we can't quantify the performance gain, then it is difficult to
>>>>> justify the change. I would also be concerned if that is the only place
>>>>> we need an explicit dsb.
>>>>
>>>> Maybe it won't hurt to add dsb to the ISR as well. But okay, let's drop
>>>> this patch for now.
>>>>
>>>
>>> Jon, it occurred to me that there still should be a problem with the
>>> writel() ordering in the driver because writel() ensures that memory
>>> stores are completed *before* the write occurs and hence translates into
>>> iowmb() + writel_relaxed() [0]. Thus the last write will always happen
>>> asynchronously in regards to clk accesses.
>>>
>>> [0]
>>> https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/arch/arm/include/asm/io.h#n311
>>>
>>
>> Also please note that iowmb() translates into wmb() if
>> CONFIG_ARM_DMA_MEM_BUFFERABLE=y and sometime ago I was profiling host1x
>> driver job submission performance and have seen cases where wmb() could
>> take up to 1ms on T20 due to L2 syncing if there are outstanding memory
>> writes in the cache (or even more, I don't remember exactly already how
>> bad it was..).
> 
> This looks to be primarily caused by the fact that we have the L2X0
> cache on Tegra20. So there's not really anything that can be done there
> without potentially compromising correctness of the code.
> 
>> Altogether, I think the usage of readl/writel in pretty much all of
>> Tegra drivers is plainly wrong and explicit dsb() shall be used in
>> places where hardware synchronization is really needed.
> 
> I don't think that's an accurate observation. readl()/writel() are more
> likely to be correct than the relaxed versions. You already saw yourself
> that using the relaxed versions can easily introduce regressions.
> 
> Granted, readl()/writel() might add more memory barriers than strictly
> necessary, and therefore they might in many cases be suboptimal. But, we
> can't just go and engage in a wholesale conversion of all drivers. If we
> do this, we need to very carefully audit every conversion to make sure
> no regressions are introduced. This is especially complicated because
> these would be subtle regressions and may be difficult to catch or
> reproduce.
> 
> Also, we should avoid using primitives such as dsb in driver code to
> avoid making the code too architecture specific.

I was testing this a bit more for a couple of days and my current
conclusion that there is likely some problem that is getting masked by
writel/readl because I tried to manually insert the syncing that
writel/readl does for the relaxed versions (and more) and that slight
shuffling of the code makes the problem to occur intermittently. My
observations show that it's only the I2C-DMA that has the trouble, other
DMA clients are working fine. Maybe there is some timing problem or
missing ready-state polling somewhere, for now I don't know what's the
actual problem is.
diff mbox series

Patch

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index cf462b1abc0b..e646e1c7b299 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -241,23 +241,23 @@  struct tegra_dma {
 
 static inline void tdma_write(struct tegra_dma *tdma, u32 reg, u32 val)
 {
-	writel(val, tdma->base_addr + reg);
+	writel_relaxed(val, tdma->base_addr + reg);
 }
 
 static inline u32 tdma_read(struct tegra_dma *tdma, u32 reg)
 {
-	return readl(tdma->base_addr + reg);
+	return readl_relaxed(tdma->base_addr + reg);
 }
 
 static inline void tdc_write(struct tegra_dma_channel *tdc,
 		u32 reg, u32 val)
 {
-	writel(val, tdc->chan_addr + reg);
+	writel_relaxed(val, tdc->chan_addr + reg);
 }
 
 static inline u32 tdc_read(struct tegra_dma_channel *tdc, u32 reg)
 {
-	return readl(tdc->chan_addr + reg);
+	return readl_relaxed(tdc->chan_addr + reg);
 }
 
 static inline struct tegra_dma_channel *to_tegra_dma_chan(struct dma_chan *dc)