Skip to contentSkip to conversation

xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

PATCH Superseded Stefano Stabellini 5 participants 21 messages Aug 1, 2014 – Sep 12, 2014

Subsystems (3)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

Hi all,
Xen support in Linux for ARM and ARM64 suffers from lack of support for multiple mfn to pfn mappings: whenever a frontend grants the same page multiple times to the backend, the mfn to pfn accounting in arch/arm/xen/p2m.c fails. The issue has become critical since v3.15, when xen-netfront/xen-netback switched from grant copies to grant mappings, therefore causing the issue to happen much more often.

Fixing the mfn to pfn accounting in p2m.c is difficult and expensive, therefore we are looking for alternative solutions. One idea is avoiding mfn to pfn conversions altogether. The only code path that needs them is swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

To avoid mfn to pfn conversions we rely on a second p2m mapping done by Xen (a separate patch series will be sent for Xen). In Linux we use it to perform the cache maintenance operations without mfns conversions.

The Xen side patch series that introduces XENFEAT_grant_map_identity is:
http://marc.info/?l=xen-devel&m=140690433923855

Changes in v3:
- rename XENFEAT_grant_map_11 to XENFEAT_grant_map_identity.

Changes in v2:
- introduce XENFEAT_grant_map_11;
- remeber the ptep corresponding to scratch pages so that we don't need to calculate it again every time;
- do not acutally unmap the page on xen_mm32_unmap;
- properly account preempt_enable/disable;
- do not check for mfn in xen_add_phys_to_mach_entry.

Stefano Stabellini (3):
xen/arm: introduce XENFEAT_grant_map_identity xen/arm: reimplement xen_dma_unmap_page & friends xen/arm: remove mach_to_phys rbtree

 arch/arm/include/asm/xen/page-coherent.h |   25 ++--
 arch/arm/include/asm/xen/page.h          |    9 --
 arch/arm/xen/Makefile                    |    2 +-
 arch/arm/xen/enlighten.c                 |    6 +
 arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
 arch/arm/xen/p2m.c                       |   66 +---------
 include/xen/interface/features.h         |    3 +
 7 files changed, 220 insertions(+), 93 deletions(-)
 create mode 100644 arch/arm/xen/mm32.c
#1 lore raw

PATCH v3 1/3 xen/arm: introduce XENFEAT_grant_map_identity

The flag tells us that the hypervisor maps a grant page to guest physical address == machine address of the page in addition to the normal grant mapping address. It is needed to properly issue cache maintenance operation at the completion of a DMA operation involving a foreign grant.

---
Changes in v3:
- rename XENFEAT_grant_map_11 to XENFEAT_grant_map_identity.

---
 arch/arm/xen/enlighten.c         |    6 ++++++
 include/xen/interface/features.h |    3 +++
 2 files changed, 9 insertions(+)
arch/arm/xen/enlighten.c +6 -0
index b96723e..eef324f 100644
@@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
 	xen_domain_type = XEN_HVM_DOMAIN;
 
 	xen_setup_features();
+
+	if (!xen_feature(XENFEAT_grant_map_identity)) {
+		pr_warn("Please upgrade your Xen.\n"
+				"If your platform has any non-coherent DMA devices, they won't work properly.\n");
+	}
+
 	if (xen_feature(XENFEAT_dom0))
 		xen_start_info->flags |= SIF_INITDOMAIN|SIF_PRIVILEGED;
 	else
include/xen/interface/features.h +3 -0
index 131a6cc..14334d0 100644
@@ -53,6 +53,9 @@
 /* operation as Dom0 is supported */
 #define XENFEAT_dom0                      11
 
+/* Xen also maps grant references at pfn = mfn */
+#define XENFEAT_grant_map_identity        12
+
 #define XENFEAT_NR_SUBMAPS 1
 
 #endif /* __XEN_PUBLIC_FEATURES_H__ */
Signature · 1 line
1.7.10.4
#2 lore raw

PATCH v3 1/3 xen/arm: introduce XENFEAT_grant_map_identity

Quoted 22 lines · Stefano Stabellini

The flag tells us that the hypervisor maps a grant page to guest physical address == machine address of the page in addition to the normal grant mapping address. It is needed to properly issue cache maintenance operation at the completion of a DMA operation involving a foreign grant.

---
Changes in v3:
- rename XENFEAT_grant_map_11 to XENFEAT_grant_map_identity.

---
 arch/arm/xen/enlighten.c         |    6 ++++++
 include/xen/interface/features.h |    3 +++
 2 files changed, 9 insertions(+)
arch/arm/xen/enlighten.c +0 -0
index b96723e..eef324f 100644

arch/arm/xen/enlighten.c static int __init xen_guest_init(void) · line 268

Quoted 7 lines
arch/arm/xen/enlighten.c +4 -0
@@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
 	xen_domain_type = XEN_HVM_DOMAIN;
 
 	xen_setup_features();
+
+	if (!xen_feature(XENFEAT_grant_map_identity)) {
+		pr_warn("Please upgrade your Xen.\n"
+				"If your platform has any non-coherent DMA devices, they won't work properly.\n");

I think:

"Non-coherent DMA devices require a hypervisor with
XENFEAT_grant_map_identity"

Is a more helpful message.

David

#3 lore raw

PATCH v3 1/3 xen/arm: introduce XENFEAT_grant_map_identity

In reply to Stefano Stabellini

Hi Stefano,

Quoted 22 lines · Stefano Stabellini

The flag tells us that the hypervisor maps a grant page to guest physical address == machine address of the page in addition to the normal grant mapping address. It is needed to properly issue cache maintenance operation at the completion of a DMA operation involving a foreign grant.

---
Changes in v3:
- rename XENFEAT_grant_map_11 to XENFEAT_grant_map_identity.

---
  arch/arm/xen/enlighten.c         |    6 ++++++
  include/xen/interface/features.h |    3 +++
  2 files changed, 9 insertions(+)
arch/arm/xen/enlighten.c +0 -0
index b96723e..eef324f 100644

arch/arm/xen/enlighten.c static int __init xen_guest_init(void) · line 269

Quoted 8 lines
arch/arm/xen/enlighten.c +5 -0
@@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
  	xen_domain_type = XEN_HVM_DOMAIN;
 
  	xen_setup_features();
+
+	if (!xen_feature(XENFEAT_grant_map_identity)) {
+		pr_warn("Please upgrade your Xen.\n"
+				"If your platform has any non-coherent DMA devices, they won't work properly.\n");
+	}

I know we talked about this on a previous thread but I can't find it in my mail :/

Does a Linux with this series will work on Xen 4.4? If not what's the plan?

I think we have to keep any OS working on top on Xen 4.4 as people may want to retrieve the latest kernel for drivers bug fixes...

For running a previous kernel on Xen 4.5, then it will still be buggy. It might be worse to backport thise series on LTS kernel (i.e at least 3.14).

Regards,

Signature · 1 line
Julien Grall
#4 lore raw

PATCH v3 1/3 xen/arm: introduce XENFEAT_grant_map_identity

Quoted 38 lines · Julien Grall

Hi Stefano,

Quoted 22 lines · Stefano Stabellini

The flag tells us that the hypervisor maps a grant page to guest physical address == machine address of the page in addition to the normal grant mapping address. It is needed to properly issue cache maintenance operation at the completion of a DMA operation involving a foreign grant.

---
Changes in v3:
- rename XENFEAT_grant_map_11 to XENFEAT_grant_map_identity.

---
  arch/arm/xen/enlighten.c         |    6 ++++++
  include/xen/interface/features.h |    3 +++
  2 files changed, 9 insertions(+)
arch/arm/xen/enlighten.c +0 -0
index b96723e..eef324f 100644

arch/arm/xen/enlighten.c static int __init xen_guest_init(void) · line 270

Quoted 9 lines
arch/arm/xen/enlighten.c +5 -0
@@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
  	xen_domain_type = XEN_HVM_DOMAIN;
 
  	xen_setup_features();
+
+	if (!xen_feature(XENFEAT_grant_map_identity)) {
+		pr_warn("Please upgrade your Xen.\n"
+				"If your platform has any non-coherent DMA
 devices, they won't work properly.\n");
+	}

I know we talked about this on a previous thread but I can't find it in my mail :/

Quoted 1 line

Does a Linux with this series will work on Xen 4.4? If not what's the plan?

I suggest we backport the changes to Xen 4.4.

"Still working" doesn't apply to a system without networking :-(

Quoted 2 lines

I think we have to keep any OS working on top on Xen 4.4 as people may want to retrieve the latest kernel for drivers bug fixes...

Right, that's why we need to backport the fixes. Without the backport of the Xen patches, with or without this Linux series, networking won't work once the kernel is upgraded.

Quoted 2 lines

For running a previous kernel on Xen 4.5, then it will still be buggy. It might be worse to backport thise series on LTS kernel (i.e at least 3.14).

The old thread is: http://marc.info/?l=xen-devel&m=140483414203686&w=2

#5 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

In reply to Stefano Stabellini

xen_dma_unmap_page, xen_dma_sync_single_for_cpu and xen_dma_sync_single_for_device are currently implemented by calling into the corresponding generic ARM implementation of these functions. In order to do this, firstly the dma_addr_t handle, that on Xen is a machine address, needs to be translated into a physical address. The operation is expensive and inaccurate, given that a single machine address can correspond to multiple physical addresses in one domain, because the same page can be granted multiple times by the frontend.

To avoid this problem, we introduce a Xen specific implementation of xen_dma_unmap_page, xen_dma_sync_single_for_cpu and xen_dma_sync_single_for_device, that can operate on machine addresses directly.

The new implementation relies on the fact that the hypervisor creates a second p2m mapping of any grant pages at physical address == machine address of the page for dom0. Therefore we can access memory at physical address == dma_addr_r handle and perform the cache flushing there. Some cache maintenance operations require a virtual address. Instead of using ioremap_cache, that is not safe in interrupt context, we allocate a per-cpu PAGE_KERNEL scratch page and we manually update the pte for it.

arm64 doesn't need cache maintenance operations on unmap for now.

---

Changes in v3:
- rename XENFEAT_grant_map_11 to XENFEAT_grant_map_identity.

Changes in v2:
- check for XENFEAT_grant_map_11;
- remeber the ptep corresponding to scratch pages so that we don't need to calculate it again every time;
- do not acutally unmap the page on xen_mm32_unmap;
- properly account preempt_enable/disable.

---
 arch/arm/include/asm/xen/page-coherent.h |   25 ++--
 arch/arm/xen/Makefile                    |    2 +-
 arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
 3 files changed, 210 insertions(+), 19 deletions(-)
 create mode 100644 arch/arm/xen/mm32.c
arch/arm/include/asm/xen/page-coherent.h +7 -18
index 1109017..e8275ea 100644
@@ -26,25 +26,14 @@ static inline void xen_dma_map_page(struct device *hwdev, struct page *page,
 	__generic_dma_ops(hwdev)->map_page(hwdev, page, offset, size, dir, attrs);
 }
 
-static inline void xen_dma_unmap_page(struct device *hwdev, dma_addr_t handle,
+void xen_dma_unmap_page(struct device *hwdev, dma_addr_t handle,
 		size_t size, enum dma_data_direction dir,
-		struct dma_attrs *attrs)
-{
-	if (__generic_dma_ops(hwdev)->unmap_page)
-		__generic_dma_ops(hwdev)->unmap_page(hwdev, handle, size, dir, attrs);
-}
+		struct dma_attrs *attrs);
 
-static inline void xen_dma_sync_single_for_cpu(struct device *hwdev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
-{
-	if (__generic_dma_ops(hwdev)->sync_single_for_cpu)
-		__generic_dma_ops(hwdev)->sync_single_for_cpu(hwdev, handle, size, dir);
-}
+void xen_dma_sync_single_for_cpu(struct device *hwdev,
+		dma_addr_t handle, size_t size, enum dma_data_direction dir);
+
+void xen_dma_sync_single_for_device(struct device *hwdev,
+		dma_addr_t handle, size_t size, enum dma_data_direction dir);
 
-static inline void xen_dma_sync_single_for_device(struct device *hwdev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
-{
-	if (__generic_dma_ops(hwdev)->sync_single_for_device)
-		__generic_dma_ops(hwdev)->sync_single_for_device(hwdev, handle, size, dir);
-}
 #endif /* _ASM_ARM_XEN_PAGE_COHERENT_H */
arch/arm/xen/Makefile +1 -1
index 1296952..1f85bfe 100644
@@ -1 +1 @@
-obj-y		:= enlighten.o hypercall.o grant-table.o p2m.o mm.o
+obj-y		:= enlighten.o hypercall.o grant-table.o p2m.o mm.o mm32.o
arch/arm/xen/mm32.c new file +202 -0

202 lines not on this page · raw message

Signature · 1 line
1.7.10.4
#6 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

arch/arm/xen/mm32.c line 134

Quoted 7 lines · Stefano Stabellini
arch/arm/xen/mm32.c new file +7 -0
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+		size_t size, enum dma_data_direction dir)
+{
+	/* Cannot use __dma_page_dev_to_cpu because we don't have a
+	 * struct page for handle */
+
+	if (dir == DMA_TO_DEVICE)

This should be:

    if (dir != DMA_TO_DEVICE)

Thomas, could you please confirm that with this small fix
http://pastebin.com/FPRf7pgL goes away?

Quoted 4 lines
arch/arm/xen/mm32.c new file +4 -0
+		outer_inv_range(handle, handle + size);
+
+	dma_cache_maint(handle & PAGE_MASK, handle & ~PAGE_MASK, size, dir, dmac_unmap_area);
+}
#7 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

Quoted 10 lines · Stefano Stabellini

arch/arm/xen/mm32.c line 134

Quoted 7 lines · Stefano Stabellini
arch/arm/xen/mm32.c new file +7 -0
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+		size_t size, enum dma_data_direction dir)
+{
+	/* Cannot use __dma_page_dev_to_cpu because we don't have a
+	 * struct page for handle */
+
+	if (dir == DMA_TO_DEVICE)

This should be:

Quoted 5 lines
    if (dir != DMA_TO_DEVICE)

Thomas, could you please confirm that with this small fix
http://pastebin.com/FPRf7pgL goes away?

Thomas, please try this fix with my ref-counting patch.

The old "working" version might actually cover this latent bug due to it's long delay.

Wei.

Quoted 5 lines
Quoted 5 lines
arch/arm/xen/mm32.c new file +4 -0
+		outer_inv_range(handle, handle + size);
+
+	dma_cache_maint(handle & PAGE_MASK, handle & ~PAGE_MASK, size, dir, dmac_unmap_area);
+}
#8 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

Quoted 13 lines · Wei Liu
Quoted 12 lines · Stefano Stabellini

arch/arm/xen/mm32.c line 134

Quoted 7 lines · Stefano Stabellini
arch/arm/xen/mm32.c new file +7 -0
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+           size_t size, enum dma_data_direction dir)
+{
+   /* Cannot use __dma_page_dev_to_cpu because we don't have a
+    * struct page for handle */
+
+   if (dir == DMA_TO_DEVICE)

This should be:

    if (dir != DMA_TO_DEVICE)
Quoted 8 lines
Quoted 3 lines

Thomas, could you please confirm that with this small fix
http://pastebin.com/FPRf7pgL goes away?

Thomas, please try this fix with my ref-counting patch.

The old "working" version might actually cover this latent bug due to it's long delay.

I'm not sure how to apply this. The function "__xen_dma_page_dev_to_cpu" doesn't appear in your "for-thomas" branch. If you push the change to that branch I can test it.

Quoted 7 lines

Wei.

Quoted 5 lines
Quoted 5 lines
arch/arm/xen/mm32.c new file +4 -0
+           outer_inv_range(handle, handle + size);
+
+   dma_cache_maint(handle & PAGE_MASK, handle & ~PAGE_MASK, size, dir, dmac_unmap_area);
+}
Signature · 3 lines
Dr Thomas Leonard        http://0install.net/
GPG: 9242 9807 C985 3C07 44A6  8B9A AE07 8280 59A5 3CC1
GPG: DA98 25AE CAD0 8975 7CDA  BD8E 0713 3F96 CA74 D8BA
#9 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

Quoted 18 lines · Thomas Leonard
Quoted 17 lines · Wei Liu
Quoted 15 lines · Stefano Stabellini

Review

Quoted 7 lines · Stefano Stabellini
(quoted diff) +7 -0
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+           size_t size, enum dma_data_direction dir)
+{
+   /* Cannot use __dma_page_dev_to_cpu because we don't have a
+    * struct page for handle */
+
+   if (dir == DMA_TO_DEVICE)

This should be:

    if (dir != DMA_TO_DEVICE)

Thomas, could you please confirm that with this small fix
http://pastebin.com/FPRf7pgL goes away?

Quoted 8 lines
Quoted 4 lines

Thomas, please try this fix with my ref-counting patch.

The old "working" version might actually cover this latent bug due to it's long delay.

I'm not sure how to apply this. The function "__xen_dma_page_dev_to_cpu" doesn't appear in your "for-thomas" branch. If you push the change to that branch I can test it.

That's the patch for you:

arch/arm/xen/mm32.c +1 -1
index 534f362..3b99860 100644
@@ -131,7 +131,7 @@ static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
 	/* Cannot use __dma_page_dev_to_cpu because we don't have a
 	 * struct page for handle */
 
-	if (dir == DMA_TO_DEVICE)
+	if (dir != DMA_TO_DEVICE)
 		outer_inv_range(handle, handle + size);
 
 	dma_cache_maint(handle & PAGE_MASK, handle & ~PAGE_MASK, size, dir, dmac_unmap_area);
#10 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

In reply to Thomas Leonard
Quoted 18 lines · Thomas Leonard
Quoted 17 lines · Wei Liu
Quoted 15 lines · Stefano Stabellini

arch/arm/xen/mm32.c line 134

Quoted 7 lines · Stefano Stabellini
arch/arm/xen/mm32.c new file +7 -0
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+           size_t size, enum dma_data_direction dir)
+{
+   /* Cannot use __dma_page_dev_to_cpu because we don't have a
+    * struct page for handle */
+
+   if (dir == DMA_TO_DEVICE)

This should be:

    if (dir != DMA_TO_DEVICE)

Thomas, could you please confirm that with this small fix
http://pastebin.com/FPRf7pgL goes away?

Quoted 9 lines
Quoted 4 lines

Thomas, please try this fix with my ref-counting patch.

The old "working" version might actually cover this latent bug due to it's long delay.

I'm not sure how to apply this. The function "__xen_dma_page_dev_to_cpu" doesn't appear in your "for-thomas" branch. If you push the change to that branch I can test it.

I think you can cherry-pick my three patches to your tree which contains Stefano's patches. It's probably easier because Stefano's patches are not yet in mainline while my patches should be able to apply to 3.16 mainline kernel without much effort.

I've rebased my patches on top of 3.16, in for-thomas2 branch.

Wei.

Quoted 14 lines
Quoted 7 lines

Wei.

Quoted 5 lines
Quoted 5 lines
arch/arm/xen/mm32.c new file +4 -0
+           outer_inv_range(handle, handle + size);
+
+   dma_cache_maint(handle & PAGE_MASK, handle & ~PAGE_MASK, size, dir, dmac_unmap_area);
+}
Signature · 3 lines
Dr Thomas Leonard        http://0install.net/
GPG: 9242 9807 C985 3C07 44A6  8B9A AE07 8280 59A5 3CC1
GPG: DA98 25AE CAD0 8975 7CDA  BD8E 0713 3F96 CA74 D8BA
#11 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

Quoted 29 lines · Wei Liu
Quoted 18 lines · Thomas Leonard
Quoted 17 lines · Wei Liu
Quoted 15 lines · Stefano Stabellini

arch/arm/xen/mm32.c line 134

Quoted 7 lines · Stefano Stabellini
arch/arm/xen/mm32.c new file +7 -0
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+           size_t size, enum dma_data_direction dir)
+{
+   /* Cannot use __dma_page_dev_to_cpu because we don't have a
+    * struct page for handle */
+
+   if (dir == DMA_TO_DEVICE)

This should be:

    if (dir != DMA_TO_DEVICE)

Thomas, could you please confirm that with this small fix
http://pastebin.com/FPRf7pgL goes away?

Quoted 9 lines
Quoted 4 lines

Thomas, please try this fix with my ref-counting patch.

The old "working" version might actually cover this latent bug due to it's long delay.

I'm not sure how to apply this. The function "__xen_dma_page_dev_to_cpu" doesn't appear in your "for-thomas" branch. If you push the change to that branch I can test it.

Quoted 6 lines

I think you can cherry-pick my three patches to your tree which contains Stefano's patches. It's probably easier because Stefano's patches are not yet in mainline while my patches should be able to apply to 3.16 mainline kernel without much effort.

I've rebased my patches on top of 3.16, in for-thomas2 branch.

OK, that works for me! I tried starting the quickly-exiting VM a few times and it didn't break anything, and my queue service served up the file too. Thanks!

(for reference, this is the exact commit I tested:
https://github.com/talex5/linux/commit/026c61c9649e0ef30546222a429a1b11ae7a7555)

Signature · 3 lines
Dr Thomas Leonard        http://0install.net/
GPG: 9242 9807 C985 3C07 44A6  8B9A AE07 8280 59A5 3CC1
GPG: DA98 25AE CAD0 8975 7CDA  BD8E 0713 3F96 CA74 D8BA
#12 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

On Fri, Aug 08, 2014 at 05:01:42PM +0100, Thomas Leonard wrote:
[...]

Quoted 7 lines

OK, that works for me! I tried starting the quickly-exiting VM a few times and it didn't break anything, and my queue service served up the file too. Thanks!

(for reference, this is the exact commit I tested:
https://github.com/talex5/linux/commit/026c61c9649e0ef30546222a429a1b11ae7a7555)

Thanks for testing.

I will submit my network driver patches shortly.

Wei.

#13 lore raw

PATCH v3 2/3 xen/arm: reimplement xen_dma_unmap_page & friends

In reply to Thomas Leonard
Quoted 35 lines · Thomas Leonard
Quoted 34 lines · Wei Liu
Quoted 18 lines · Thomas Leonard
Quoted 17 lines · Wei Liu
Quoted 15 lines · Stefano Stabellini

arch/arm/xen/mm32.c line 134

Quoted 7 lines · Stefano Stabellini
arch/arm/xen/mm32.c new file +7 -0
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+           size_t size, enum dma_data_direction dir)
+{
+   /* Cannot use __dma_page_dev_to_cpu because we don't have a
+    * struct page for handle */
+
+   if (dir == DMA_TO_DEVICE)

This should be:

    if (dir != DMA_TO_DEVICE)

Thomas, could you please confirm that with this small fix
http://pastebin.com/FPRf7pgL goes away?

Quoted 9 lines
Quoted 4 lines

Thomas, please try this fix with my ref-counting patch.

The old "working" version might actually cover this latent bug due to it's long delay.

I'm not sure how to apply this. The function "__xen_dma_page_dev_to_cpu" doesn't appear in your "for-thomas" branch. If you push the change to that branch I can test it.

I think you can cherry-pick my three patches to your tree which contains Stefano's patches. It's probably easier because Stefano's patches are not yet in mainline while my patches should be able to apply to 3.16 mainline kernel without much effort.

Quoted 9 lines
Quoted 1 line

I've rebased my patches on top of 3.16, in for-thomas2 branch.

OK, that works for me! I tried starting the quickly-exiting VM a few times and it didn't break anything, and my queue service served up the file too. Thanks!

(for reference, this is the exact commit I tested:
https://github.com/talex5/linux/commit/026c61c9649e0ef30546222a429a1b11ae7a7555)

Thanks for testing! I'll update my patch series with the small fix.

#14 lore raw

PATCH v3 3/3 xen/arm: remove mach_to_phys rbtree

In reply to Stefano Stabellini

Remove the rbtree used to keep track of machine to physical mappings:
the frontend can grant the same page multiple times, leading to errors inserting or removing entries from the mach_to_phys tree.

Linux only needed to know the physical address corresponding to a given machine address in swiotlb-xen. Now that swiotlb-xen can call the xen_dma_* functions passing the machine address directly, we can remove it.

---

Changes in v2:
- do not check for mfn in xen_add_phys_to_mach_entry.

---
 arch/arm/include/asm/xen/page.h |    9 ------
 arch/arm/xen/p2m.c              |   66 +--------------------------------------
 2 files changed, 1 insertion(+), 74 deletions(-)
arch/arm/include/asm/xen/page.h +0 -9
index ded062f..135c24a 100644
@@ -33,7 +33,6 @@ typedef struct xpaddr {
 #define INVALID_P2M_ENTRY      (~0UL)
 
 unsigned long __pfn_to_mfn(unsigned long pfn);
-unsigned long __mfn_to_pfn(unsigned long mfn);
 extern struct rb_root phys_to_mach;
 
 static inline unsigned long pfn_to_mfn(unsigned long pfn)
@@ -51,14 +50,6 @@ static inline unsigned long pfn_to_mfn(unsigned long pfn)
 
 static inline unsigned long mfn_to_pfn(unsigned long mfn)
 {
-	unsigned long pfn;
-
-	if (phys_to_mach.rb_node != NULL) {
-		pfn = __mfn_to_pfn(mfn);
-		if (pfn != INVALID_P2M_ENTRY)
-			return pfn;
-	}
-
 	return mfn;
 }
 
arch/arm/xen/p2m.c +1 -65

102 lines not on this page · raw message

Signature · 1 line
1.7.10.4
#15 lore raw

v3 cover xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

In reply to Stefano Stabellini
Quoted 13 lines · Stefano Stabellini

Hi all,
Xen support in Linux for ARM and ARM64 suffers from lack of support for multiple mfn to pfn mappings: whenever a frontend grants the same page multiple times to the backend, the mfn to pfn accounting in arch/arm/xen/p2m.c fails. The issue has become critical since v3.15, when xen-netfront/xen-netback switched from grant copies to grant mappings, therefore causing the issue to happen much more often.

Fixing the mfn to pfn accounting in p2m.c is difficult and expensive, therefore we are looking for alternative solutions. One idea is avoiding mfn to pfn conversions altogether. The only code path that needs them is swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

Quoted 7 lines

To avoid mfn to pfn conversions we rely on a second p2m mapping done by Xen (a separate patch series will be sent for Xen). In Linux we use it to perform the cache maintenance operations without mfns conversions.

The Xen side patch series that introduces XENFEAT_grant_map_identity is:
http://marc.info/?l=xen-devel&m=140690433923855

If the Xen side is in master, you can apply this series to devel/for-linus-3.17

David

#16 lore raw

v3 cover xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

Quoted 19 lines · David Vrabel
Quoted 18 lines · Stefano Stabellini

Hi all,
Xen support in Linux for ARM and ARM64 suffers from lack of support for multiple mfn to pfn mappings: whenever a frontend grants the same page multiple times to the backend, the mfn to pfn accounting in arch/arm/xen/p2m.c fails. The issue has become critical since v3.15, when xen-netfront/xen-netback switched from grant copies to grant mappings, therefore causing the issue to happen much more often.

Fixing the mfn to pfn accounting in p2m.c is difficult and expensive, therefore we are looking for alternative solutions. One idea is avoiding mfn to pfn conversions altogether. The only code path that needs them is swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

To avoid mfn to pfn conversions we rely on a second p2m mapping done by Xen (a separate patch series will be sent for Xen). In Linux we use it to perform the cache maintenance operations without mfns conversions.

Quoted 5 lines
Quoted 2 lines

The Xen side patch series that introduces XENFEAT_grant_map_identity is:
http://marc.info/?l=xen-devel&m=140690433923855

If the Xen side is in master, you can apply this series to devel/for-linus-3.17

Thanks. The Xen series is not in yet, I'll do as soon as it is applied.

#17 lore raw

v3 cover xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

In reply to David Vrabel
Quoted 19 lines · David Vrabel
Quoted 18 lines · Stefano Stabellini

Hi all,
Xen support in Linux for ARM and ARM64 suffers from lack of support for multiple mfn to pfn mappings: whenever a frontend grants the same page multiple times to the backend, the mfn to pfn accounting in arch/arm/xen/p2m.c fails. The issue has become critical since v3.15, when xen-netfront/xen-netback switched from grant copies to grant mappings, therefore causing the issue to happen much more often.

Fixing the mfn to pfn accounting in p2m.c is difficult and expensive, therefore we are looking for alternative solutions. One idea is avoiding mfn to pfn conversions altogether. The only code path that needs them is swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

To avoid mfn to pfn conversions we rely on a second p2m mapping done by Xen (a separate patch series will be sent for Xen). In Linux we use it to perform the cache maintenance operations without mfns conversions.

Quoted 5 lines
Quoted 2 lines

The Xen side patch series that introduces XENFEAT_grant_map_identity is:
http://marc.info/?l=xen-devel&m=140690433923855

If the Xen side is in master, you can apply this series to devel/for-linus-3.17

The Xen side patch finally went in. Unfortunately it's probably too late for 3.17.

I have applied the latest version
(http://marc.info/?l=linux-arm-kernel&m=140751761520724&w=2) to devel/for-linus-3.18.

#18 lore raw

v3 cover xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

Quoted 20 lines · Stefano Stabellini
Quoted 19 lines · David Vrabel
Quoted 18 lines · Stefano Stabellini

Hi all,
Xen support in Linux for ARM and ARM64 suffers from lack of support for multiple mfn to pfn mappings: whenever a frontend grants the same page multiple times to the backend, the mfn to pfn accounting in arch/arm/xen/p2m.c fails. The issue has become critical since v3.15, when xen-netfront/xen-netback switched from grant copies to grant mappings, therefore causing the issue to happen much more often.

Fixing the mfn to pfn accounting in p2m.c is difficult and expensive, therefore we are looking for alternative solutions. One idea is avoiding mfn to pfn conversions altogether. The only code path that needs them is swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

To avoid mfn to pfn conversions we rely on a second p2m mapping done by Xen (a separate patch series will be sent for Xen). In Linux we use it to perform the cache maintenance operations without mfns conversions.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

The Xen side patch series that introduces XENFEAT_grant_map_identity is:
http://marc.info/?l=xen-devel&m=140690433923855

If the Xen side is in master, you can apply this series to devel/for-linus-3.17

The Xen side patch finally went in. Unfortunately it's probably too late for 3.17.

Isn't this a bug fix? I think would be fine to go into 3.17.

David

author Applied
#19 lore raw

v3 cover xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

Quoted 24 lines · David Vrabel
Quoted 23 lines · Stefano Stabellini
Quoted 22 lines · David Vrabel
Quoted 20 lines · Stefano Stabellini

Hi all,
Xen support in Linux for ARM and ARM64 suffers from lack of support for multiple mfn to pfn mappings: whenever a frontend grants the same page multiple times to the backend, the mfn to pfn accounting in arch/arm/xen/p2m.c fails. The issue has become critical since v3.15, when xen-netfront/xen-netback switched from grant copies to grant mappings, therefore causing the issue to happen much more often.

Fixing the mfn to pfn accounting in p2m.c is difficult and expensive, therefore we are looking for alternative solutions. One idea is avoiding mfn to pfn conversions altogether. The only code path that needs them is swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

To avoid mfn to pfn conversions we rely on a second p2m mapping done by Xen (a separate patch series will be sent for Xen). In Linux we use it to perform the cache maintenance operations without mfns conversions.

The Xen side patch series that introduces XENFEAT_grant_map_identity is:
http://marc.info/?l=xen-devel&m=140690433923855

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

If the Xen side is in master, you can apply this series to devel/for-linus-3.17

The Xen side patch finally went in. Unfortunately it's probably too late for 3.17.

Isn't this a bug fix? I think would be fine to go into 3.17.

Yes, you are right, it is a bug fix. I was being too conservative.

I applied it to stable/for-linus-3.17-b

Do you have any more fixes queued up to send to Linus for 3.17?

#20 lore raw

v3 cover xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

Quoted 2 lines · Stefano Stabellini

Do you have any more fixes queued up to send to Linus for 3.17?

I sent a batch yesterday. Sorry for not waiting for these -- I forgot about them.

David