Skip to contentSkip to conversation

iommu/dma: Add support for DMA_ATTR_SYS_CACHE

PATCH In review Isaac J. Manjarres 5 participants 9 messages Oct 26, 2019 – Oct 28, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iommu/dma: Add support for DMA_ATTR_SYS_CACHE

Currently, IOMMU_QCOM_SYS_CACHE exists to allow non-coherent I/O masters on Qualcomm SoCs to upgrade to caching their buffers in the outer-level/system cache on these platforms. However, these masters are limited to managing the mapping of these buffers themselves through the IOMMU framework, as opposed to allowing the DMA-IOMMU framework to handle it, since there is no DMA attribute that maps to the IOMMU_QCOM_SYS_CACHE protection bit.

Thus, introduce DMA_ATTR_SYS_CACHE so that clients can use the DMA-IOMMU layer to map their DMA buffers with the correct memory attributes to allow the buffers to be cached in the outer-level/system cache.

---
 drivers/iommu/dma-iommu.c   | 2 ++
 include/linux/dma-mapping.h | 8 ++++++++
 2 files changed, 10 insertions(+)
drivers/iommu/dma-iommu.c +2 -0
index f321279..c433ece 100644
@@ -369,6 +369,8 @@ static int dma_info_to_prot(enum dma_data_direction dir, bool coherent,
 
 	if (attrs & DMA_ATTR_PRIVILEGED)
 		prot |= IOMMU_PRIV;
+	if (attrs & DMA_ATTR_SYS_CACHE)
+		prot |= IOMMU_QCOM_SYS_CACHE;
 
 	switch (dir) {
 	case DMA_BIDIRECTIONAL:
include/linux/dma-mapping.h +8 -0
index 4a1c4fc..bdd4dcf 100644
@@ -71,6 +71,14 @@
 #define DMA_ATTR_PRIVILEGED		(1UL << 9)
 
 /*
+ * DMA_ATTR_SYS_CACHE: This is a hint that non-coherent masters can use to
+ * tell the DMA-IOMMU layer to map their DMA buffers with the correct memory
+ * attributes that allow these buffers to be cached in an outer-level/system
+ * cache.
+ */
+#define DMA_ATTR_SYS_CACHE		(1UL << 10)
+
+/*
  * A dma_addr_t can hold any valid DMA or bus address for the platform.
  * It can be given to a device to use as a DMA source or target.  A CPU cannot
  * reference a dma_addr_t directly because there may be translation between
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
#1 lore raw

The definition makes very little sense. Any without a user in the same series it is a complete no-go anyway.

#2 lore raw
Quoted 1 line · Christoph Hellwig

The definition makes very little sense.

Can you please clarify what part doesn’t make sense, and why? This is really just an extension of this patch that got mainlined, so that clients that use the DMA API can use IOMMU_QCOM_SYS_CACHE as well: https://patchwork.kernel.org/patch/10946099/

Quoted 1 line

Any without a user in the same series it is a complete no-go anyway.

IOMMU_QCOM_SYS_CACHE does not have any current users in the mainline, nor did it have it in the patch series in which it got merged, yet it is still present? Furthermore, there are plans to upstream support for one of our SoCs that may benefit from this, as seen here: https://www.spinics.net/lists/iommu/msg39608.html.

Thanks,
Isaac
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

#3 lore raw
Quoted 3 lines · On Sat, Oct 26, 2019 at 03:12:57AM -0700, [email protected] wrote:
Quoted 1 line · Christoph Hellwig

The definition makes very little sense.

Can you please clarify what part doesn’t make sense, and why?

It looks like complete garbage to me. That might just be because it uses tons of terms I've never heard of of and which aren't used anywhere in the DMA API. It also might be because it doesn't explain how the flag might actually be practically useful.

Quoted 10 lines

This is
really just an extension of this patch that got mainlined, so that clients that use the DMA API can use IOMMU_QCOM_SYS_CACHE as well:
https://patchwork.kernel.org/patch/10946099/

Quoted 1 line

Any without a user in the same series it is a complete no-go anyway.

IOMMU_QCOM_SYS_CACHE does not have any current users in the mainline, nor did it have it in the patch series in which it got merged, yet it is still present? Furthermore, there are plans to upstream support for one of our SoCs that may benefit from this, as seen here:
https://www.spinics.net/lists/iommu/msg39608.html.

Which means it should have never been merged. As a general policy we do not add code to the Linux kernel without actual users.

maintainer
#4 lore raw

Hi Christoph,

Quoted 9 lines · Christoph Hellwig
Quoted 3 lines · On Sat, Oct 26, 2019 at 03:12:57AM -0700, [email protected] wrote:
Quoted 1 line · Christoph Hellwig

The definition makes very little sense.

Can you please clarify what part doesn’t make sense, and why?

It looks like complete garbage to me. That might just be because it uses tons of terms I've never heard of of and which aren't used anywhere in the DMA API. It also might be because it doesn't explain how the flag might actually be practically useful.

Agreed. The way I /think/ it works is that on many SoCs there is a system/last-level cache (LLC) which effectively sits in front of memory for all masters. Even if a device isn't coherent with the CPU caches, we still want to be able to allocate into the LLC. Why this doesn't happen automatically is beyond me, but it appears that on these Qualcomm designs you actually have to set the memory attributes up in the page-table to ensure that the resulting memory transactions are non-cacheable for the CPU but cacheable for the LLC. Without any changes, the transactions are non-cacheable in both of them which assumedly has a performance cost.

But you can see that I'm piecing things together myself here. Isaac?

Quoted 5 lines
Quoted 5 lines

This is
really just an extension of this patch that got mainlined, so that clients that use the DMA API can use IOMMU_QCOM_SYS_CACHE as well:
https://patchwork.kernel.org/patch/10946099/

Quoted 1 line

Any without a user in the same series it is a complete no-go anyway.

Quoted 8 lines
Quoted 5 lines

IOMMU_QCOM_SYS_CACHE does not have any current users in the mainline, nor did it have it in the patch series in which it got merged, yet it is still present? Furthermore, there are plans to upstream support for one of our SoCs that may benefit from this, as seen here:
https://www.spinics.net/lists/iommu/msg39608.html.

Which means it should have never been merged. As a general policy we do not add code to the Linux kernel without actual users.

Yes, in this case I was hoping a user would materialise via a different tree, but it didn't happen, hence my post last week about removing this altogether:

https://lore.kernel.org/linux-iommu/[email protected]/T/#t

which I suspect prompted this patch that unfortunately fails to solve the problem.

Will
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

#5 lore raw
Quoted 11 lines · Will Deacon

Agreed. The way I /think/ it works is that on many SoCs there is a system/last-level cache (LLC) which effectively sits in front of memory for all masters. Even if a device isn't coherent with the CPU caches, we still want to be able to allocate into the LLC. Why this doesn't happen automatically is beyond me, but it appears that on these Qualcomm designs you actually have to set the memory attributes up in the page-table to ensure that the resulting memory transactions are non-cacheable for the CPU but cacheable for the LLC. Without any changes, the transactions are non-cacheable in both of them which assumedly has a performance cost.

But you can see that I'm piecing things together myself here. Isaac?

If that is the case it sounds like we'd want to drive this through DT properties, not the driver API. But again, without an actual consumer it pretty much is a moot point anyway.

maintainer
#6 lore raw
Quoted 11 lines · Christoph Hellwig
Quoted 10 lines · Will Deacon

Agreed. The way I /think/ it works is that on many SoCs there is a system/last-level cache (LLC) which effectively sits in front of memory for all masters. Even if a device isn't coherent with the CPU caches, we still want to be able to allocate into the LLC. Why this doesn't happen automatically is beyond me, but it appears that on these Qualcomm designs you actually have to set the memory attributes up in the page-table to ensure that the resulting memory transactions are non-cacheable for the CPU but cacheable for the LLC. Without any changes, the transactions are non-cacheable in both of them which assumedly has a performance cost.

Quoted 5 lines
Quoted 1 line

But you can see that I'm piecing things together myself here. Isaac?

If that is the case it sounds like we'd want to drive this through DT properties, not the driver API. But again, without an actual consumer it pretty much is a moot point anyway.

I think there's certainly a DT aspect to it so that the driver knows that the SoC is hooked up this way, but I agree that we need a consumer so that we can figure out how dynamic this needs to be. If it's just a fire-and-forget thing then it needn't escape the IOMMU layer, but I fear that it's probably more flexible than that.

If nothing shows up for 5.6, I'll send a patch to remove it (since Jordan said there was something coming soon [1])

Will

[1] http://lkml.kernel.org/r/[email protected]
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

maintainer
#7 lore raw
In reply to Will Deacon
Quoted 23 lines · Will Deacon

Hi Christoph,

Quoted 9 lines · Christoph Hellwig
Quoted 3 lines · On Sat, Oct 26, 2019 at 03:12:57AM -0700, [email protected] wrote:
Quoted 1 line · Christoph Hellwig

The definition makes very little sense.

Can you please clarify what part doesn’t make sense, and why?

It looks like complete garbage to me. That might just be because it uses tons of terms I've never heard of of and which aren't used anywhere in the DMA API. It also might be because it doesn't explain how the flag might actually be practically useful.

Agreed. The way I /think/ it works is that on many SoCs there is a system/last-level cache (LLC) which effectively sits in front of memory for all masters. Even if a device isn't coherent with the CPU caches, we still want to be able to allocate into the LLC. Why this doesn't happen automatically is beyond me, but it appears that on these Qualcomm designs you actually have to set the memory attributes up in the page-table to ensure that the resulting memory transactions are non-cacheable for the CPU but cacheable for the LLC. Without any changes, the transactions are non-cacheable in both of them which assumedly has a performance cost.

Quoted 1 line

But you can see that I'm piecing things together myself here. Isaac?

FWIW, that's pretty much how Pratik and Jordan explained it to me - the LLC sits directly in front of memory and is more or less transparent, although it might treat CPU and device accesses slightly differently (I don't remember exactly how the inner cacheablility attribute interacts). Certain devices don't get much benefit from the LLC, hence the desire for finer-grained control of their outer allocation policy to avoid more thrashing than necessary. Furthermore, for stuff in the video/GPU/display area certain jobs benefit more than others, hence the desire to go even finer-grained than a per-device control in order to maximise LLC effectiveness.

Robin.

Quoted 18 lines
Quoted 5 lines
Quoted 5 lines

This is
really just an extension of this patch that got mainlined, so that clients that use the DMA API can use IOMMU_QCOM_SYS_CACHE as well:
https://patchwork.kernel.org/patch/10946099/

Quoted 1 line

Any without a user in the same series it is a complete no-go anyway.

Quoted 8 lines
Quoted 5 lines

IOMMU_QCOM_SYS_CACHE does not have any current users in the mainline, nor did it have it in the patch series in which it got merged, yet it is still present? Furthermore, there are plans to upstream support for one of our SoCs that may benefit from this, as seen here:
https://www.spinics.net/lists/iommu/msg39608.html.

Which means it should have never been merged. As a general policy we do not add code to the Linux kernel without actual users.

Yes, in this case I was hoping a user would materialise via a different tree, but it didn't happen, hence my post last week about removing this altogether:

Quoted 7 lines

https://lore.kernel.org/linux-iommu/[email protected]/T/#t

which I suspect prompted this patch that unfortunately fails to solve the problem.

Will

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

#8 lore raw
Quoted 26 lines · Robin Murphy
Quoted 23 lines · Will Deacon

Hi Christoph,

Quoted 9 lines · Christoph Hellwig
Quoted 3 lines · On Sat, Oct 26, 2019 at 03:12:57AM -0700, [email protected] wrote:
Quoted 1 line · Christoph Hellwig

The definition makes very little sense.

Can you please clarify what part doesn’t make sense, and why?

It looks like complete garbage to me. That might just be because it uses tons of terms I've never heard of of and which aren't used anywhere in the DMA API. It also might be because it doesn't explain how the flag might actually be practically useful.

Agreed. The way I /think/ it works is that on many SoCs there is a system/last-level cache (LLC) which effectively sits in front of memory for all masters. Even if a device isn't coherent with the CPU caches, we still want to be able to allocate into the LLC. Why this doesn't happen automatically is beyond me, but it appears that on these Qualcomm designs you actually have to set the memory attributes up in the page-table to ensure that the resulting memory transactions are non-cacheable for the CPU but cacheable for the LLC. Without any changes, the transactions are non-cacheable in both of them which assumedly has a performance cost.

Quoted 1 line

But you can see that I'm piecing things together myself here. Isaac?

Quoted 10 lines

FWIW, that's pretty much how Pratik and Jordan explained it to me - the LLC sits directly in front of memory and is more or less transparent, although it might treat CPU and device accesses slightly differently (I don't remember exactly how the inner cacheablility attribute interacts). Certain devices don't get much benefit from the LLC, hence the desire for finer-grained control of their outer allocation policy to avoid more thrashing than necessary. Furthermore, for stuff in the video/GPU/display area certain jobs benefit more than others, hence the desire to go even finer-grained than a per-device control in order to maximise LLC effectiveness.

Robin's description is correct. And we did have a patch for an in-kernel user but it got lost in the wash. I'm hoping Sharat can get a respin in time for 5.5.

https://lore.kernel.org/linux-arm-msm/[email protected]/

Jordan

Signature · 6 lines
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu