iommu: Add IOMMU_DEBUG_PAGEALLOC sanitizer
Subsystems (5)
v2 cover iommu: Add IOMMU_DEBUG_PAGEALLOC sanitizer
Overview
--------
This patch series introduces a new debugging feature, IOMMU_DEBUG_PAGEALLOC, designed to catch DMA use-after-free bugs and IOMMU mapping leaks from buggy drivers.
The kernel has powerful sanitizers like KASAN and DEBUG_PAGEALLOC for catching CPU-side memory corruption. However, there is limited runtime sanitization for DMA mappings managed by the IOMMU. A buggy driver can free a page while it is still mapped for DMA, leading to memory corruption or use-after-free vulnerabilities when that page is reallocated and used for a different purpose.
Inspired by DEBUG_PAGEALLOC, this sanitizer tracks IOMMU mappings on a per-page basis, as it’s not possible to unmap the pages, because it requires to lock and walk all domains on every kernel free, instead we rely on page_ext to add an IOMMU-specific mapping reference count for each page.
And on each page allocated/freed from the kernel we simply check the count and WARN if it is not zero.
Concurrency
-----------
By design this check is racy where one caller can map pages just after the check, which can lead to false negatives.
In my opinion this is acceptable for sanitizers (for ex KCSAN have that property).
Otherwise we have to implement locks in iommu_map/unmap for all domains which is not favourable even for a debug feature.
The sanitizer only guarantees that the refcount itself doesn’t get corrupted using atomics. And there are no false positives.
CPU vs IOMMU Page Size
----------------------
IOMMUs can use different page sizes and which can be non-homogeneous; not even all of them have the same page size.
To solve this, the refcount is always incremented and decremented in units of the smallest page size supported by the IOMMU domain. This ensures the accounting remains consistent regardless of the size of the map or unmap operation, otherwise double counting can happen.
Testing & Performance
---------------------
This was tested on Morello with Arm64 + SMMUv3
Also I booted RockPi-4b with Rockchip IOMMU.
Did some tests on Qemu including different SMMUv3/CPU page size (arm64).
I also ran dma_map_benchmark on Morello:
echo dma_map_benchmark > /sys/bus/pci/devices/0000\:06\:00.0/driver_override echo 0000:06:00.0 > /sys/bus/pci/devices/0000\:06\:00.0/driver/unbind echo 0000:06:00.0 > /sys/bus/pci/drivers/dma_map_benchmark/bind ./dma_map_benchmark -t $threads -g $nr_pages
CONFIG refers to "CONFIG_IOMMU_DEBUG_PAGEALLOC" cmdline refers to "iommu.debug_pagealloc"
Numbers are (map latency)/(unmap latency), lower is better.
CONFIG=n CONFIG=y CONFIG=y cmdline=0 cmdline=1 4K - 1 thread 0.1/0.6 0.1/0.6 0.1/0.7
4K - 4 threads 0.1/1.0 0.1/1.0 0.1/1.1
1M - 1 thread 0.8/21.2 0.8/21.2 5.6/42.4
1M - 4 threads 1.1/45.9 1.1/46.0 6.0/45.4
Main changes v2:
v1: https://lore.kernel.org/linux-iommu/[email protected]/
- Address Jörg comments about #ifdefs and static keys
- Reword the KCONFIG help
- Drop RFC
- Collect t-b from Qinxin
- Minor cleanups
Mostafa Saleh (4):
drivers/iommu: Add page_ext for IOMMU_DEBUG_PAGEALLOC drivers/iommu: Add calls for IOMMU_DEBUG_PAGEALLOC drivers/iommu-debug-pagealloc: Track IOMMU pages drivers/iommu-debug-pagealloc: Check mapped/unmapped kernel memory
.../admin-guide/kernel-parameters.txt | 6 + drivers/iommu/Kconfig | 15 ++ drivers/iommu/Makefile | 1 + drivers/iommu/iommu-debug-pagealloc.c | 148 ++++++++++++++++++ drivers/iommu/iommu.c | 14 +- include/linux/iommu-debug-pagealloc.h | 83 ++++++++++ include/linux/mm.h | 5 + mm/page_ext.c | 4 + 8 files changed, 274 insertions(+), 2 deletions(-) create mode 100644 drivers/iommu/iommu-debug-pagealloc.c create mode 100644 include/linux/iommu-debug-pagealloc.h
base-commit: dc77806cf3b4788d328fddf245e86c5b529f31a2
Signature · 1 line
2.51.2.1026.g39e6a42477-goog
PATCH v2 1/4 drivers/iommu: Add page_ext for IOMMU_DEBUG_PAGEALLOC
Add a new config IOMMU_DEBUG_PAGEALLOC, which registers new data to page_ext.
This config will be used by the IOMMU API to track pages mapped in the IOMMU to catch drivers trying to free kernel memory that they still map in their domains, causing all types of memory corruption. This behaviour is disabled by default and can be enabled using kernel cmdline iommu.debug_pagealloc.
--- .../admin-guide/kernel-parameters.txt | 6 ++++ drivers/iommu/Kconfig | 15 +++++++++ drivers/iommu/Makefile | 1 + drivers/iommu/iommu-debug-pagealloc.c | 32 +++++++++++++++++++ include/linux/iommu-debug-pagealloc.h | 17 ++++++++++ mm/page_ext.c | 4 +++ 6 files changed, 75 insertions(+) create mode 100644 drivers/iommu/iommu-debug-pagealloc.c create mode 100644 include/linux/iommu-debug-pagealloc.h
Documentation/admin-guide/kernel-parameters.txt +6 -0
@@ -2557,6 +2557,12 @@ 1 - Bypass the IOMMU for DMA. unset - Use value of CONFIG_IOMMU_DEFAULT_PASSTHROUGH. + iommu.debug_pagealloc= + [KNL,EARLY] When CONFIG_IOMMU_DEBUG_PAGEALLOC is set, this + parameter enables the feature at boot time. By default, it + is disabled and the system will work mostly the same as a + kernel built without CONFIG_IOMMU_DEBUG_PAGEALLOC. + io7= [HW] IO7 for Marvel-based Alpha systems See comment before marvel_specify_io7 in arch/alpha/kernel/core_marvel.c.
drivers/iommu/Kconfig +15 -0
@@ -383,4 +383,19 @@ config SPRD_IOMMU Say Y here if you want to use the multimedia devices listed above. +config IOMMU_DEBUG_PAGEALLOC + bool "Debug page memory allocations against IOMMU" + depends on DEBUG_PAGEALLOC && IOMMU_API && PAGE_EXTENSION + help + This config checks that a page is freed(unmapped) or mapped by the + kernel is not mapped in any IOMMU domain. It can help with debugging + use-after-free or out-of-bound maps from drivers doing DMA through + the IOMMU API. + This santaizer can have false-negative cases where some problems + won't be detected. + Expect overhead when enabling this and enabling the kernel command + line iommu.debug_pagealloc. + + If unsure, say N here. + endif # IOMMU_SUPPORT
drivers/iommu/Makefile +1 -0
@@ -34,3 +34,4 @@ obj-$(CONFIG_IOMMU_SVA) += iommu-sva.o obj-$(CONFIG_IOMMU_IOPF) += io-pgfault.o obj-$(CONFIG_SPRD_IOMMU) += sprd-iommu.o obj-$(CONFIG_APPLE_DART) += apple-dart.o +obj-$(CONFIG_IOMMU_DEBUG_PAGEALLOC) += iommu-debug-pagealloc.o
drivers/iommu/iommu-debug-pagealloc.c new file +32 -0
@@ -0,0 +1,32 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (C) 2025 - Google Inc +* Author: Mostafa Saleh <[email protected]> +* IOMMU API debug page alloc sanitizer +*/ +#include <linux/atomic.h> +#include <linux/iommu-debug-pagealloc.h> +#include <linux/kernel.h> +#include <linux/page_ext.h> + +static bool needed; + +struct iommu_debug_metadate { +atomic_t ref; +}; + +static __init bool need_iommu_debug (void ) +{ +return needed; +} + +struct page_ext_operations page_iommu_debug_ops = { + .size =sizeof (struct iommu_debug_metadate ), + .need = need_iommu_debug, +}; + +static int __init iommu_debug_pagealloc (char *str) +{ +return kstrtobool (str, &needed); +} +early_param ("iommu.debug_pagealloc" , iommu_debug_pagealloc);
include/linux/iommu-debug-pagealloc.h new file +17 -0
@@ -0,0 +1,17 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (C) 2025 - Google Inc +* Author: Mostafa Saleh <[email protected]> +* IOMMU API debug page alloc sanitizer +*/ + +#ifndef __LINUX_IOMMU_DEBUG_PAGEALLOC_H +#define __LINUX_IOMMU_DEBUG_PAGEALLOC_H + +#ifdef CONFIG_IOMMU_DEBUG_PAGEALLOC + +extern struct page_ext_operations page_iommu_debug_ops; + +#endif /* CONFIG_IOMMU_DEBUG_PAGEALLOC */ + +#endif /* __LINUX_IOMMU_DEBUG_PAGEALLOC_H */
mm/page_ext.c +4 -0
@@ -11,6 +11,7 @@#include <linux/page_table_check.h> #include <linux/rcupdate.h> #include <linux/pgalloc_tag.h> +#include <linux/iommu-debug-pagealloc.h> /* * struct page extension
@@ -89,6 +90,9 @@ static struct page_ext_operations *page_ext_ops[] __initdata = {#ifdef CONFIG_PAGE_TABLE_CHECK &page_table_check_ops,#endif +#ifdef CONFIG_IOMMU_DEBUG_PAGEALLOC + &page_iommu_debug_ops, +#endif };unsigned long page_ext_size;
Signature · 1 line
2.51.2.1026.g39e6a42477-goog
Randy Dunlap
reviewerRe: [PATCH v2 1/4] drivers/iommu: Add page_ext for IOMMU_DEBUG_PAGEALLOC
Quoted 9 lines
drivers/iommu/Kconfig +0
-0@@ -383,4 +383,19 @@ config SPRD_IOMMU Say Y here if you want to use the multimedia devices listed above.
drivers/iommu/Kconfig config SPRD_IOMMU · line 394
Quoted 10 lines
drivers/iommu/Kconfig +9
-0+config IOMMU_DEBUG_PAGEALLOC + bool "Debug page memory allocations against IOMMU" + depends on DEBUG_PAGEALLOC && IOMMU_API && PAGE_EXTENSION + help + This config checks that a page is freed(unmapped) or mapped by the + kernel is not mapped in any IOMMU domain. It can help with debugging + use-after-free or out-of-bound maps from drivers doing DMA through + the IOMMU API. + This santaizer can have false-negative cases where some problems
sanitizer
Quoted 7 lines
drivers/iommu/Kconfig +6
-0+ won't be detected. + Expect overhead when enabling this and enabling the kernel command + line iommu.debug_pagealloc. + + If unsure, say N here. + endif # IOMMU_SUPPORT
Signature · 1 line
~Randy
Re: [PATCH v2 1/4] drivers/iommu: Add page_ext for IOMMU_DEBUG_PAGEALLOC
Quoted 10 lines · Randy Dunlap
Quoted 10 lines
drivers/iommu/Kconfig +0
-0@@ -383,4 +383,19 @@ config SPRD_IOMMU Say Y here if you want to use the multimedia devices listed above.
Quoted 11 lines
drivers/iommu/Kconfig config SPRD_IOMMU · line 394
Quoted 9 lines
drivers/iommu/Kconfig +9
-0+config IOMMU_DEBUG_PAGEALLOC + bool "Debug page memory allocations against IOMMU" + depends on DEBUG_PAGEALLOC && IOMMU_API && PAGE_EXTENSION + help + This config checks that a page is freed(unmapped) or mapped by the + kernel is not mapped in any IOMMU domain. It can help with debugging + use-after-free or out-of-bound maps from drivers doing DMA through + the IOMMU API. + This santaizer can have false-negative cases where some problemssanitizer
I will fix it.
Thanks,
Mostafa
Quoted 13 lines
Quoted 8 lines
drivers/iommu/Kconfig +6
-0+ won't be detected. + Expect overhead when enabling this and enabling the kernel command + line iommu.debug_pagealloc. + + If unsure, say N here. + endif # IOMMU_SUPPORTSignature · 1 line
~Randy
Will Deacon
maintainerRe: [PATCH v2 1/4] drivers/iommu: Add page_ext for IOMMU_DEBUG_PAGEALLOC
In reply to Mostafa SalehHi Mostafa,
Quoted 10 lines · Mostafa Saleh
Add a new config
IOMMU_DEBUG_PAGEALLOC, which registers new data topage_ext.
This config will be used by the IOMMU API to track pages mapped in the IOMMU to catch drivers trying to free kernel memory that they still map in their domains, causing all types of memory corruption. This behaviour is disabled by default and can be enabled using kernel cmdline iommu.debug_pagealloc.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>
Quoted 10 lines
--- .../admin-guide/kernel-parameters.txt | 6 ++++ drivers/iommu/Kconfig | 15 +++++++++ drivers/iommu/Makefile | 1 + drivers/iommu/iommu-debug-pagealloc.c | 32 +++++++++++++++++++ include/linux/iommu-debug-pagealloc.h | 17 ++++++++++ mm/page_ext.c | 4 +++ 6 files changed, 75 insertions(+) create mode 100644 drivers/iommu/iommu-debug-pagealloc.c create mode 100644 include/linux/iommu-debug-pagealloc.h
This looks like a pretty handy feature to me, but I have some nits below.
Documentation/admin-guide/kernel-parameters.txt line 2564
Quoted 13 lines
Documentation/admin-guide/kernel-parameters.txt +5
-0@@ -2557,6 +2557,12 @@ 1 - Bypass the IOMMU for DMA. unset - Use value of CONFIG_IOMMU_DEFAULT_PASSTHROUGH. + iommu.debug_pagealloc= + [KNL,EARLY] When CONFIG_IOMMU_DEBUG_PAGEALLOC is set, this + parameter enables the feature at boot time. By default, it + is disabled and the system will work mostly the same as a + kernel built without CONFIG_IOMMU_DEBUG_PAGEALLOC.
Can you be more specific about "mostly the same"?
Quoted 9 lines
Documentation/admin-guide/kernel-parameters.txt +1
-0+ io7= [HW] IO7 for Marvel-based Alpha systems See comment before marvel_specify_io7 in arch/alpha/kernel/core_marvel.c.drivers/iommu/Kconfig +0
-0
drivers/iommu/Kconfig config SPRD_IOMMU · line 387
Quoted 5 lines
drivers/iommu/Kconfig +2
-0@@ -383,4 +383,19 @@ config SPRD_IOMMU Say Y here if you want to use the multimedia devices listed above. +config IOMMU_DEBUG_PAGEALLOC + bool "Debug page memory allocations against IOMMU"
Perhaps "IOMMU mappings" would make this a little clearer?
drivers/iommu/Kconfig config SPRD_IOMMU
Quoted 4 lines
drivers/iommu/Kconfig +4
-0+ depends on DEBUG_PAGEALLOC && IOMMU_API && PAGE_EXTENSION + help + This config checks that a page is freed(unmapped) or mapped by the + kernel is not mapped in any IOMMU domain.
I can't really parse this sentence :/
drivers/iommu/Kconfig config SPRD_IOMMU · line 395
Quoted 5 lines
It can help with debugging
drivers/iommu/Kconfig +4
-0+ use-after-free or out-of-bound maps from drivers doing DMA through + the IOMMU API. + This santaizer can have false-negative cases where some problems + won't be detected.
Maybe just say "The sanitizer is best-effort and can fail to detect problems in the case that ...".
drivers/iommu/Kconfig config SPRD_IOMMU · line 397
Quoted 2 lines
drivers/iommu/Kconfig +2
-0+ Expect overhead when enabling this and enabling the kernel command + line iommu.debug_pagealloc.
I'd reword this to say something like "Due to the overhead of the sanitiser, iommu.debug_pagealloc must also be passed on the kernel command-line to enable this feature".
Quoted 25 lines
drivers/iommu/Kconfig +3
-0+ + If unsure, say N here. + endif # IOMMU_SUPPORTdrivers/iommu/Makefile +1
-0@@ -34,3 +34,4 @@ obj-$(CONFIG_IOMMU_SVA) += iommu-sva.o obj-$(CONFIG_IOMMU_IOPF) += io-pgfault.o obj-$(CONFIG_SPRD_IOMMU) += sprd-iommu.o obj-$(CONFIG_APPLE_DART) += apple-dart.o +obj-$(CONFIG_IOMMU_DEBUG_PAGEALLOC) += iommu-debug-pagealloc.odrivers/iommu/iommu-debug-pagealloc.c new file +6
-0@@ -0,0 +1,32 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (C) 2025 - Google Inc +* Author: Mostafa Saleh <[email protected]> +* IOMMU API debug page alloc sanitizer +*/
drivers/iommu/iommu-debug-pagealloc.c line 16
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c new file +10
-0+#include <linux/atomic.h> +#include <linux/iommu-debug-pagealloc.h> +#include <linux/kernel.h> +#include <linux/page_ext.h> + +static bool needed; + +struct iommu_debug_metadate { +atomic_t ref; +};
s/metadate/metadata/
Will
Re: [PATCH v2 1/4] drivers/iommu: Add page_ext for IOMMU_DEBUG_PAGEALLOC
Quoted 24 lines · Will Deacon
Hi Mostafa,
Quoted 10 lines · Mostafa Saleh
Add a new config
IOMMU_DEBUG_PAGEALLOC, which registers new data topage_ext.
This config will be used by the IOMMU API to track pages mapped in the IOMMU to catch drivers trying to free kernel memory that they still map in their domains, causing all types of memory corruption. This behaviour is disabled by default and can be enabled using kernel cmdline iommu.debug_pagealloc.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>Quoted 10 lines
--- .../admin-guide/kernel-parameters.txt | 6 ++++ drivers/iommu/Kconfig | 15 +++++++++ drivers/iommu/Makefile | 1 + drivers/iommu/iommu-debug-pagealloc.c | 32 +++++++++++++++++++ include/linux/iommu-debug-pagealloc.h | 17 ++++++++++ mm/page_ext.c | 4 +++ 6 files changed, 75 insertions(+) create mode 100644 drivers/iommu/iommu-debug-pagealloc.c create mode 100644 include/linux/iommu-debug-pagealloc.h
Quoted 1 line
This looks like a pretty handy feature to me, but I have some nits below.
Thanks for taking the time to review the patches!
Quoted 9 lines
Quoted 9 lines
Documentation/admin-guide/kernel-parameters.txt +0
-0@@ -2557,6 +2557,12 @@ 1 - Bypass the IOMMU for DMA. unset - Use value of CONFIG_IOMMU_DEFAULT_PASSTHROUGH.
Quoted 7 lines
Documentation/admin-guide/kernel-parameters.txt line 2564
Quoted 5 lines
Documentation/admin-guide/kernel-parameters.txt +5
-0+ iommu.debug_pagealloc= + [KNL,EARLY] When CONFIG_IOMMU_DEBUG_PAGEALLOC is set, this + parameter enables the feature at boot time. By default, it + is disabled and the system will work mostly the same as a + kernel built without CONFIG_IOMMU_DEBUG_PAGEALLOC.Can you be more specific about "mostly the same"?
The only difference is that the static key to gate the calls, I was not sure if saying “exactly the same” is correct, but I think it’s better avoid “mostly” as it might be confusing and as the data in the cover letter shows no overhead, I will re-write the whole help anyway.
Quoted 11 lines
Quoted 11 lines
Documentation/admin-guide/kernel-parameters.txt +1
-0+ io7= [HW] IO7 for Marvel-based Alpha systems See comment before marvel_specify_io7 in arch/alpha/kernel/core_marvel.c.drivers/iommu/Kconfig +0
-0@@ -383,4 +383,19 @@ config SPRD_IOMMU
Quoted 6 lines
drivers/iommu/Kconfig config SPRD_IOMMU · line 387
Quoted 4 lines
drivers/iommu/Kconfig +2
-0Say Y here if you want to use the multimedia devices listed above. +config IOMMU_DEBUG_PAGEALLOC + bool "Debug page memory allocations against IOMMU"Perhaps "IOMMU mappings" would make this a little clearer?
Will do.
Quoted 7 lines
drivers/iommu/Kconfig config SPRD_IOMMU
Quoted 5 lines
drivers/iommu/Kconfig +4
-0+ depends on DEBUG_PAGEALLOC && IOMMU_API && PAGE_EXTENSION + help + This config checks that a page is freed(unmapped) or mapped by the + kernel is not mapped in any IOMMU domain.I can't really parse this sentence :/
I will re-write it.
Quoted 9 lines
drivers/iommu/Kconfig config SPRD_IOMMU · line 395
Quoted 6 lines
It can help with debugging
drivers/iommu/Kconfig +4
-0+ use-after-free or out-of-bound maps from drivers doing DMA through + the IOMMU API. + This santaizer can have false-negative cases where some problems + won't be detected.Maybe just say "The sanitizer is best-effort and can fail to detect problems in the case that ...".
Makes sense, will do.
Quoted 7 lines
drivers/iommu/Kconfig config SPRD_IOMMU · line 397
Quoted 3 lines
drivers/iommu/Kconfig +2
-0+ Expect overhead when enabling this and enabling the kernel command + line iommu.debug_pagealloc.I'd reword this to say something like "Due to the overhead of the sanitiser, iommu.debug_pagealloc must also be passed on the kernel command-line to enable this feature".
Will do.
Quoted 30 lines
Quoted 30 lines
drivers/iommu/Kconfig +3
-0+ + If unsure, say N here. + endif # IOMMU_SUPPORTdrivers/iommu/Makefile +1
-0@@ -34,3 +34,4 @@ obj-$(CONFIG_IOMMU_SVA) += iommu-sva.o obj-$(CONFIG_IOMMU_IOPF) += io-pgfault.o obj-$(CONFIG_SPRD_IOMMU) += sprd-iommu.o obj-$(CONFIG_APPLE_DART) += apple-dart.o +obj-$(CONFIG_IOMMU_DEBUG_PAGEALLOC) += iommu-debug-pagealloc.odrivers/iommu/iommu-debug-pagealloc.c new file +10
-0@@ -0,0 +1,32 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (C) 2025 - Google Inc +* Author: Mostafa Saleh <[email protected]> +* IOMMU API debug page alloc sanitizer +*/ +#include <linux/atomic.h> +#include <linux/iommu-debug-pagealloc.h> +#include <linux/kernel.h> +#include <linux/page_ext.h>
Quoted 8 lines
drivers/iommu/iommu-debug-pagealloc.c line 16
Quoted 6 lines
drivers/iommu/iommu-debug-pagealloc.c new file +6
-0+ +static bool needed; + +struct iommu_debug_metadate { +atomic_t ref; +};s/metadate/metadata/
Ah, that's embarrassing, I will fix it.
Thanks,
Mostafa
Quoted 2 lines
Will
Re: [PATCH v2 1/4] drivers/iommu: Add page_ext for IOMMU_DEBUG_PAGEALLOC
Quoted 47 lines · Mostafa Saleh
Quoted 24 lines · Will Deacon
Hi Mostafa,
Quoted 10 lines · Mostafa Saleh
Add a new config
IOMMU_DEBUG_PAGEALLOC, which registers new data topage_ext.
This config will be used by the IOMMU API to track pages mapped in the IOMMU to catch drivers trying to free kernel memory that they still map in their domains, causing all types of memory corruption. This behaviour is disabled by default and can be enabled using kernel cmdline iommu.debug_pagealloc.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>Quoted 10 lines
--- .../admin-guide/kernel-parameters.txt | 6 ++++ drivers/iommu/Kconfig | 15 +++++++++ drivers/iommu/Makefile | 1 + drivers/iommu/iommu-debug-pagealloc.c | 32 +++++++++++++++++++ include/linux/iommu-debug-pagealloc.h | 17 ++++++++++ mm/page_ext.c | 4 +++ 6 files changed, 75 insertions(+) create mode 100644 drivers/iommu/iommu-debug-pagealloc.c create mode 100644 include/linux/iommu-debug-pagealloc.hQuoted 1 line
This looks like a pretty handy feature to me, but I have some nits below.
Thanks for taking the time to review the patches!
Quoted 9 lines
Quoted 9 lines
Documentation/admin-guide/kernel-parameters.txt +0
-0@@ -2557,6 +2557,12 @@ 1 - Bypass the IOMMU for DMA. unset - Use value of CONFIG_IOMMU_DEFAULT_PASSTHROUGH.Quoted 7 lines
Documentation/admin-guide/kernel-parameters.txt line 2564
Quoted 5 lines
Documentation/admin-guide/kernel-parameters.txt +5
-0+ iommu.debug_pagealloc= + [KNL,EARLY] When CONFIG_IOMMU_DEBUG_PAGEALLOC is set, this + parameter enables the feature at boot time. By default, it + is disabled and the system will work mostly the same as a + kernel built without CONFIG_IOMMU_DEBUG_PAGEALLOC.Can you be more specific about "mostly the same"?
Quoted 5 lines
The only difference is that the static key to gate the calls, I was not sure if saying “exactly the same” is correct, but I think it’s better avoid “mostly” as it might be confusing and as the data in the cover letter shows no overhead, I will re-write the whole help anyway.
Ah, I think I copied the base of this from the documenation of "debug_pagealloc="
which actually slightly changes system behviour by not using THP on some archs, which is not relevant in our case, I will drop it then.
Thanks,
Mostafa
Quoted 98 lines
Quoted 11 lines
Quoted 11 lines
Documentation/admin-guide/kernel-parameters.txt +1
-0+ io7= [HW] IO7 for Marvel-based Alpha systems See comment before marvel_specify_io7 in arch/alpha/kernel/core_marvel.c.drivers/iommu/Kconfig +0
-0@@ -383,4 +383,19 @@ config SPRD_IOMMUQuoted 6 lines
drivers/iommu/Kconfig config SPRD_IOMMU · line 387
Quoted 4 lines
drivers/iommu/Kconfig +2
-0Say Y here if you want to use the multimedia devices listed above. +config IOMMU_DEBUG_PAGEALLOC + bool "Debug page memory allocations against IOMMU"Perhaps "IOMMU mappings" would make this a little clearer?
Will do.
Quoted 7 lines
drivers/iommu/Kconfig config SPRD_IOMMU
Quoted 5 lines
drivers/iommu/Kconfig +4
-0+ depends on DEBUG_PAGEALLOC && IOMMU_API && PAGE_EXTENSION + help + This config checks that a page is freed(unmapped) or mapped by the + kernel is not mapped in any IOMMU domain.I can't really parse this sentence :/
I will re-write it.
Quoted 9 lines
drivers/iommu/Kconfig config SPRD_IOMMU · line 395
Quoted 6 lines
It can help with debugging
drivers/iommu/Kconfig +4
-0+ use-after-free or out-of-bound maps from drivers doing DMA through + the IOMMU API. + This santaizer can have false-negative cases where some problems + won't be detected.Maybe just say "The sanitizer is best-effort and can fail to detect problems in the case that ...".
Makes sense, will do.
Quoted 7 lines
drivers/iommu/Kconfig config SPRD_IOMMU · line 397
Quoted 3 lines
drivers/iommu/Kconfig +2
-0+ Expect overhead when enabling this and enabling the kernel command + line iommu.debug_pagealloc.I'd reword this to say something like "Due to the overhead of the sanitiser, iommu.debug_pagealloc must also be passed on the kernel command-line to enable this feature".
Will do.
Quoted 30 lines
Quoted 30 lines
drivers/iommu/Kconfig +3
-0+ + If unsure, say N here. + endif # IOMMU_SUPPORTdrivers/iommu/Makefile +1
-0@@ -34,3 +34,4 @@ obj-$(CONFIG_IOMMU_SVA) += iommu-sva.o obj-$(CONFIG_IOMMU_IOPF) += io-pgfault.o obj-$(CONFIG_SPRD_IOMMU) += sprd-iommu.o obj-$(CONFIG_APPLE_DART) += apple-dart.o +obj-$(CONFIG_IOMMU_DEBUG_PAGEALLOC) += iommu-debug-pagealloc.odrivers/iommu/iommu-debug-pagealloc.c new file +10
-0@@ -0,0 +1,32 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (C) 2025 - Google Inc +* Author: Mostafa Saleh <[email protected]> +* IOMMU API debug page alloc sanitizer +*/ +#include <linux/atomic.h> +#include <linux/iommu-debug-pagealloc.h> +#include <linux/kernel.h> +#include <linux/page_ext.h> Quoted 8 lines
drivers/iommu/iommu-debug-pagealloc.c line 16
Quoted 6 lines
drivers/iommu/iommu-debug-pagealloc.c new file +6
-0+ +static bool needed; + +struct iommu_debug_metadate { +atomic_t ref; +};s/metadate/metadata/
Ah, that's embarrassing, I will fix it.
Thanks,
MostafaQuoted 2 lines
Will
PATCH v2 2/4 drivers/iommu: Add calls for IOMMU_DEBUG_PAGEALLOC
In reply to Mostafa SalehAdd calls for the new iommu debug config IOMMU_DEBUG_PAGEALLOC:
- iommu_debug_init: Enable the debug mode if configured by the user.
- iommu_debug_map: Track iommu pages mapped, using physical address.
- iommu_debug_unmap: Track iommu pages unmapped, using IO virtual address.
- iommu_debug_remap: Track iommu pages, already mapped using IOVA.
We have to do the unmap/remap as once pages are unmapped we lose the information of the physical address.
This is racy, but the API is racy by construction as it uses refcounts and doesn't attempt to lock/synchronize with the IOMMU API as that will be costly, meaning that possibility of false negative exists.
--- drivers/iommu/iommu-debug-pagealloc.c | 23 ++++++++++++ drivers/iommu/iommu.c | 14 ++++++- include/linux/iommu-debug-pagealloc.h | 54 +++++++++++++++++++++++++++ 3 files changed, 89 insertions(+), 2 deletions(-)
drivers/iommu/iommu-debug-pagealloc.c +23 -0
@@ -5,11 +5,13 @@* IOMMU API debug page alloc sanitizer */ #include <linux/atomic.h> +#include <linux/iommu.h> #include <linux/iommu-debug-pagealloc.h> #include <linux/kernel.h> #include <linux/page_ext.h> static bool needed; +DEFINE_STATIC_KEY_FALSE (iommu_debug_initialized);struct iommu_debug_metadate {atomic_t ref;
@@ -25,6 +27,27 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, }; +void __iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size) +{ +} + +void __iommu_debug_unmap (struct iommu_domain *domain,unsigned long iova,size_t size) +{ +} + +void __iommu_debug_remap (struct iommu_domain *domain,unsigned long iova,size_t size) +{ +} + +void iommu_debug_init (void ) +{ +if (!needed) +return ; + +pr_info ("iommu: Debugging page allocations, expect overhead or disable iommu.debug_pagealloc" ); +static_branch_enable (&iommu_debug_initialized); +} +static int __init iommu_debug_pagealloc (char *str) {return kstrtobool (str, &needed);
drivers/iommu/iommu.c +12 -2
@@ -18,6 +18,7 @@#include <linux/errno.h> #include <linux/host1x_context_bus.h> #include <linux/iommu.h> +#include <linux/iommu-debug-pagealloc.h> #include <linux/iommufd.h> #include <linux/idr.h> #include <linux/err.h>
@@ -231,6 +232,8 @@ static int __init iommu_subsys_init(void)if (!nb)return -ENOMEM ; +iommu_debug_init (); +for (int i =0 ; i <ARRAY_SIZE (iommu_buses); i++) { nb[i].notifier_call = iommu_bus_notifier;bus_register_notifier (iommu_buses[i], &nb[i]);
@@ -2544,10 +2547,12 @@ int iommu_map_nosync(struct iommu_domain *domain, unsigned long iova, }/* unroll mapping in case something went wrong */ -if (ret) +if (ret) {iommu_unmap (domain, orig_iova, orig_size - size); -else + }else {trace_map (orig_iova, orig_paddr, orig_size); +iommu_debug_map (domain, orig_paddr, orig_size); + }return ret; }
@@ -2609,6 +2614,8 @@ static size_t __iommu_unmap(struct iommu_domain *domain,pr_debug ("unmap this: iova 0x%lx size 0x%zx\n" , iova, size); +iommu_debug_unmap (domain, iova, size); +/* * Keep iterating until we either unmap 'size' bytes (or more) * or we hit an area that isn't mapped.
@@ -2628,6 +2635,9 @@ static size_t __iommu_unmap(struct iommu_domain *domain, unmapped += unmapped_page; } +if (unmapped < size) +iommu_debug_remap (domain, iova, size - unmapped); +trace_unmap (orig_iova, size, unmapped);return unmapped; }
include/linux/iommu-debug-pagealloc.h +54 -0
64 lines not on this page · raw message
Signature · 1 line
2.51.2.1026.g39e6a42477-goog
Will Deacon
maintainerRe: [PATCH v2 2/4] drivers/iommu: Add calls for IOMMU_DEBUG_PAGEALLOC
Quoted 44 lines · Mostafa Saleh
Add calls for the new iommu debug config
IOMMU_DEBUG_PAGEALLOC:
-iommu_debug_init: Enable the debug mode if configured by the user.
-iommu_debug_map: Track iommu pages mapped, using physical address.
-iommu_debug_unmap: Track iommu pages unmapped, using IO virtual address.
-iommu_debug_remap: Track iommu pages, already mapped using IOVA.We have to do the unmap/remap as once pages are unmapped we lose the information of the physical address.
This is racy, but the API is racy by construction as it uses refcounts and doesn't attempt to lock/synchronize with the IOMMU API as that will be costly, meaning that possibility of false negative exists.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 23 ++++++++++++ drivers/iommu/iommu.c | 14 ++++++- include/linux/iommu-debug-pagealloc.h | 54 +++++++++++++++++++++++++++ 3 files changed, 89 insertions(+), 2 deletions(-)drivers/iommu/iommu-debug-pagealloc.c +3
-0@@ -5,11 +5,13 @@* IOMMU API debug page alloc sanitizer */ #include <linux/atomic.h> +#include <linux/iommu.h> #include <linux/iommu-debug-pagealloc.h> #include <linux/kernel.h> #include <linux/page_ext.h> static bool needed; +DEFINE_STATIC_KEY_FALSE (iommu_debug_initialized);struct iommu_debug_metadate {atomic_t ref;@@ -25,6 +27,27 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, }; +void __iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size)
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 40
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +10
-0+{ +} + +void __iommu_debug_unmap (struct iommu_domain *domain,unsigned long iova,size_t size) +{ +} + +void __iommu_debug_remap (struct iommu_domain *domain,unsigned long iova,size_t size) +{ +}
Since the IOMMU API doesn't really have a "remap" operation, I wonder whether it would be clearer to have unmap_begin() and unmap_end() functions instead? You'd probably want to call them as a pair, so the check for unmapped < size would move into unmap_end().
Will
Re: [PATCH v2 2/4] drivers/iommu: Add calls for IOMMU_DEBUG_PAGEALLOC
Quoted 44 lines · Will Deacon
Quoted 43 lines · Mostafa Saleh
Add calls for the new iommu debug config
IOMMU_DEBUG_PAGEALLOC:
-iommu_debug_init: Enable the debug mode if configured by the user.
-iommu_debug_map: Track iommu pages mapped, using physical address.
-iommu_debug_unmap: Track iommu pages unmapped, using IO virtual address.
-iommu_debug_remap: Track iommu pages, already mapped using IOVA.We have to do the unmap/remap as once pages are unmapped we lose the information of the physical address.
This is racy, but the API is racy by construction as it uses refcounts and doesn't attempt to lock/synchronize with the IOMMU API as that will be costly, meaning that possibility of false negative exists.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 23 ++++++++++++ drivers/iommu/iommu.c | 14 ++++++- include/linux/iommu-debug-pagealloc.h | 54 +++++++++++++++++++++++++++ 3 files changed, 89 insertions(+), 2 deletions(-)drivers/iommu/iommu-debug-pagealloc.c +2
-0@@ -5,11 +5,13 @@* IOMMU API debug page alloc sanitizer */ #include <linux/atomic.h> +#include <linux/iommu.h> #include <linux/iommu-debug-pagealloc.h> #include <linux/kernel.h> #include <linux/page_ext.h> static bool needed; +DEFINE_STATIC_KEY_FALSE (iommu_debug_initialized);struct iommu_debug_metadate {atomic_t ref;@@ -25,6 +27,27 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, };
Quoted 16 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 40
Quoted 11 lines
drivers/iommu/iommu-debug-pagealloc.c +11
-0+void __iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size) +{ +} + +void __iommu_debug_unmap (struct iommu_domain *domain,unsigned long iova,size_t size) +{ +} + +void __iommu_debug_remap (struct iommu_domain *domain,unsigned long iova,size_t size) +{ +}Since the IOMMU API doesn't really have a "remap" operation, I wonder whether it would be clearer to have
unmap_begin()andunmap_end()functions instead? You'd probably want to call them as a pair, so the check for unmapped < size would move intounmap_end().
I guess that would be cleaner, but we have to pass both size and unmapped to unmap_end, I will give it a try.
Thanks,
Mostafa
Quoted 2 lines
Will
PATCH v2 3/4 drivers/iommu-debug-pagealloc: Track IOMMU pages
In reply to Mostafa SalehUsing the new calls, use an atomic refcount to track how many times a page is mapped in any of the IOMMUs.
For unmap we need to use iova_to_phys() to get the physical address of the pages.
We use the smallest supported page size as the granularity of tracking per domain.
This is important as it possible to map pages and unmap them with larger sizes (as in map_sg()) cases.
--- drivers/iommu/iommu-debug-pagealloc.c | 74 +++++++++++++++++++++++++++ 1 file changed, 74 insertions(+)
drivers/iommu/iommu-debug-pagealloc.c +74 -0
@@ -27,16 +27,90 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, }; +static struct page_ext *get_iommu_page_ext (phys_addr_t phys) +{ +struct page *page =phys_to_page (phys); +struct page_ext *page_ext =page_ext_get (page); + +return page_ext; +} + +static struct iommu_debug_metadate *get_iommu_data (struct page_ext *page_ext) +{ +return page_ext_data (page_ext, &page_iommu_debug_ops); +} + +static void iommu_debug_inc_page (phys_addr_t phys) +{ +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_inc_return (&d->ref) <=0 ); +page_ext_put (page_ext); +} + +static void iommu_debug_dec_page (phys_addr_t phys) +{ +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_dec_return (&d->ref) <0 ); +page_ext_put (page_ext); +} + +/* +* IOMMU page size might not match the CPU page size, in that case, we use +* the smallest IOMMU page size to refcount the pages in the vmemmap. +* That is important as both map and unmap has to use the same page size +* to update the refcount to avoid double counting the same page. +* And as we can't know from iommu_unmap() what was the original page size +* used for map, we just use the minimum supported one for both. +*/ +static size_t iommu_debug_page_size (struct iommu_domain *domain) +{ +return 1UL <<__ffs (domain->pgsize_bitmap); +} +void __iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain); + +for (off =0 ; off < size ; off += page_size) { +if (!pfn_valid (__phys_to_pfn (phys + off))) +continue ; +iommu_debug_inc_page (phys + off); + } }void __iommu_debug_unmap (struct iommu_domain *domain,unsigned long iova,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain); + +for (off =0 ; off < size ; off += page_size) { +phys_addr_t phys =iommu_iova_to_phys (domain, iova + off); + +if (!phys || !pfn_valid (__phys_to_pfn (phys + off))) +continue ; + +iommu_debug_dec_page (phys); + } }void __iommu_debug_remap (struct iommu_domain *domain,unsigned long iova,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain); + +for (off =0 ; off < size ; off += page_size) { +phys_addr_t phys =iommu_iova_to_phys (domain, iova + off); + +if (!phys || !pfn_valid (__phys_to_pfn (phys + off))) +continue ; + +iommu_debug_inc_page (phys); + } }void iommu_debug_init (void )
Signature · 1 line
2.51.2.1026.g39e6a42477-goog
Will Deacon
maintainerRe: [PATCH v2 3/4] drivers/iommu-debug-pagealloc: Track IOMMU pages
Quoted 34 lines · Mostafa Saleh
Using the new calls, use an atomic refcount to track how many times a page is mapped in any of the IOMMUs.
For unmap we need to use
iova_to_phys()to get the physical address of the pages.We use the smallest supported page size as the granularity of tracking per domain.
This is important as it possible to map pages and unmap them with larger sizes (as inmap_sg()) cases.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 74 +++++++++++++++++++++++++++ 1 file changed, 74 insertions(+)drivers/iommu/iommu-debug-pagealloc.c +9
-0@@ -27,16 +27,90 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, }; +static struct page_ext *get_iommu_page_ext (phys_addr_t phys) +{ +struct page *page =phys_to_page (phys); +struct page_ext *page_ext =page_ext_get (page); + +return page_ext; +} + +static struct iommu_debug_metadate *get_iommu_data (struct page_ext *page_ext)
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 48
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +10
-0+{ +return page_ext_data (page_ext, &page_iommu_debug_ops); +} + +static void iommu_debug_inc_page (phys_addr_t phys) +{ +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_inc_return (&d->ref) <=0 );
Is it worth dumping some information about the page in addition to the WARN_ON()? That way, you might be able to benefit from other debug options (e.g. PAGE_OWNER) if they are enabled.
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 57
Quoted 9 lines
drivers/iommu/iommu-debug-pagealloc.c +9
-0+page_ext_put (page_ext); +} + +static void iommu_debug_dec_page (phys_addr_t phys) +{ +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_dec_return (&d->ref) <0 );
nit: I can't see why you need memory ordering guarantees for the refcount, so you could use the relaxed variants for the inc/dec operations.
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +10
-0+page_ext_put (page_ext); +} + +/* +* IOMMU page size might not match the CPU page size, in that case, we use +* the smallest IOMMU page size to refcount the pages in the vmemmap. +* That is important as both map and unmap has to use the same page size +* to update the refcount to avoid double counting the same page. +* And as we can't know from iommu_unmap() what was the original page size +* used for map, we just use the minimum supported one for both.
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 77
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +8
-0+*/ +static size_t iommu_debug_page_size (struct iommu_domain *domain) +{ +return 1UL <<__ffs (domain->pgsize_bitmap); +} +void __iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain);
Since this is a debug feature, is it worth checking other properties of the arguments too? For example, that phys is non-zero and that phys + size doesn't overflow?
Quoted 7 lines
drivers/iommu/iommu-debug-pagealloc.c +5
-0+for (off =0 ; off < size ; off += page_size) { +if (!pfn_valid (__phys_to_pfn (phys + off))) +continue ; +iommu_debug_inc_page (phys + off); + } }
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 95
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +8
-0void __iommu_debug_unmap (struct iommu_domain *domain,unsigned long iova,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain); + +for (off =0 ; off < size ; off += page_size) { +phys_addr_t phys =iommu_iova_to_phys (domain, iova + off); + +if (!phys || !pfn_valid (__phys_to_pfn (phys + off))) +continue ;
Hmm, it looks weird to add 'off' to both 'iova' _and_ the resulting physical address. Is that correct?
Quoted 7 lines
drivers/iommu/iommu-debug-pagealloc.c +3
-0+iommu_debug_dec_page (phys); + } }void __iommu_debug_remap (struct iommu_domain *domain,unsigned long iova,size_t size) { +size_t off;
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 113
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +10
-0+size_t page_size =iommu_debug_page_size (domain); + +for (off =0 ; off < size ; off += page_size) { +phys_addr_t phys =iommu_iova_to_phys (domain, iova + off); + +if (!phys || !pfn_valid (__phys_to_pfn (phys + off))) +continue ; + +iommu_debug_inc_page (phys); + }
You can make the bulk of this code common with the unmap function.
Will
Re: [PATCH v2 3/4] drivers/iommu-debug-pagealloc: Track IOMMU pages
Quoted 41 lines · Will Deacon
Quoted 40 lines · Mostafa Saleh
Using the new calls, use an atomic refcount to track how many times a page is mapped in any of the IOMMUs.
For unmap we need to use
iova_to_phys()to get the physical address of the pages.We use the smallest supported page size as the granularity of tracking per domain.
This is important as it possible to map pages and unmap them with larger sizes (as inmap_sg()) cases.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 74 +++++++++++++++++++++++++++ 1 file changed, 74 insertions(+)drivers/iommu/iommu-debug-pagealloc.c +15
-0@@ -27,16 +27,90 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, }; +static struct page_ext *get_iommu_page_ext (phys_addr_t phys) +{ +struct page *page =phys_to_page (phys); +struct page_ext *page_ext =page_ext_get (page); + +return page_ext; +} + +static struct iommu_debug_metadate *get_iommu_data (struct page_ext *page_ext) +{ +return page_ext_data (page_ext, &page_iommu_debug_ops); +} + +static void iommu_debug_inc_page (phys_addr_t phys) +{
Quoted 8 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 48
Quoted 4 lines
drivers/iommu/iommu-debug-pagealloc.c +4
-0+struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_inc_return (&d->ref) <=0 );Is it worth dumping some information about the page in addition to the
WARN_ON()? That way, you might be able to benefit from other debug options (e.g.PAGE_OWNER) if they are enabled.
These WARN_ON are for overflows, which should never happen. I initially thought about using the refcount_t, but it didn’t seem suitable as refcount_add() expects that the refcount is already “1” indicating that an object was already created which doesn’t fit in the semantics of what this is. Similar for refcount_dec().
In the next patch there is a WARN_ON for the refcount check to capture the mis-behaving context, I will add a debug print with the leaked physical address in that case as this is the important one.
Quoted 13 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 57
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +9
-0+page_ext_put (page_ext); +} + +static void iommu_debug_dec_page (phys_addr_t phys) +{ +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_dec_return (&d->ref) <0 );nit: I can't see why you need memory ordering guarantees for the refcount, so you could use the relaxed variants for the inc/dec operations.
Will do.
Quoted 17 lines
Quoted 17 lines
drivers/iommu/iommu-debug-pagealloc.c +16
-0+page_ext_put (page_ext); +} + +/* +* IOMMU page size might not match the CPU page size, in that case, we use +* the smallest IOMMU page size to refcount the pages in the vmemmap. +* That is important as both map and unmap has to use the same page size +* to update the refcount to avoid double counting the same page. +* And as we can't know from iommu_unmap() what was the original page size +* used for map, we just use the minimum supported one for both. +*/ +static size_t iommu_debug_page_size (struct iommu_domain *domain) +{ +return 1UL <<__ffs (domain->pgsize_bitmap); +} +
Quoted 9 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 77
Quoted 4 lines
drivers/iommu/iommu-debug-pagealloc.c +2
-0void __iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain);Since this is a debug feature, is it worth checking other properties of the arguments too? For example, that phys is non-zero and that phys + size doesn't overflow?
Makes sense, I will add some more checks.
Quoted 7 lines
Quoted 7 lines
drivers/iommu/iommu-debug-pagealloc.c +5
-0+for (off =0 ; off < size ; off += page_size) { +if (!pfn_valid (__phys_to_pfn (phys + off))) +continue ; +iommu_debug_inc_page (phys + off); + } }
Quoted 14 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 95
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +8
-0void __iommu_debug_unmap (struct iommu_domain *domain,unsigned long iova,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain); + +for (off =0 ; off < size ; off += page_size) { +phys_addr_t phys =iommu_iova_to_phys (domain, iova + off); + +if (!phys || !pfn_valid (__phys_to_pfn (phys + off))) +continue ;Hmm, it looks weird to add 'off' to both 'iova' _and_ the resulting physical address. Is that correct?
Yes, that's a bug, I will fix it.
Quoted 4 lines
Quoted 4 lines
drivers/iommu/iommu-debug-pagealloc.c +2
-0+iommu_debug_dec_page (phys); + } }
Quoted 15 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 113
Quoted 13 lines
drivers/iommu/iommu-debug-pagealloc.c +11
-0void __iommu_debug_remap (struct iommu_domain *domain,unsigned long iova,size_t size) { +size_t off; +size_t page_size =iommu_debug_page_size (domain); + +for (off =0 ; off < size ; off += page_size) { +phys_addr_t phys =iommu_iova_to_phys (domain, iova + off); + +if (!phys || !pfn_valid (__phys_to_pfn (phys + off))) +continue ; + +iommu_debug_inc_page (phys); + }You can make the bulk of this code common with the unmap function.
Will do.
Thank,
Mostafa
Quoted 2 lines
Will
Will Deacon
maintainerRe: [PATCH v2 3/4] drivers/iommu-debug-pagealloc: Track IOMMU pages
Quoted 57 lines · Mostafa Saleh
Quoted 41 lines · Will Deacon
Quoted 40 lines · Mostafa Saleh
Using the new calls, use an atomic refcount to track how many times a page is mapped in any of the IOMMUs.
For unmap we need to use
iova_to_phys()to get the physical address of the pages.We use the smallest supported page size as the granularity of tracking per domain.
This is important as it possible to map pages and unmap them with larger sizes (as inmap_sg()) cases.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 74 +++++++++++++++++++++++++++ 1 file changed, 74 insertions(+)drivers/iommu/iommu-debug-pagealloc.c +15
-0@@ -27,16 +27,90 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, }; +static struct page_ext *get_iommu_page_ext (phys_addr_t phys) +{ +struct page *page =phys_to_page (phys); +struct page_ext *page_ext =page_ext_get (page); + +return page_ext; +} + +static struct iommu_debug_metadate *get_iommu_data (struct page_ext *page_ext) +{ +return page_ext_data (page_ext, &page_iommu_debug_ops); +} + +static void iommu_debug_inc_page (phys_addr_t phys) +{Quoted 8 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 48
Quoted 4 lines
drivers/iommu/iommu-debug-pagealloc.c +4
-0+struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_inc_return (&d->ref) <=0 );Is it worth dumping some information about the page in addition to the
WARN_ON()? That way, you might be able to benefit from other debug options (e.g.PAGE_OWNER) if they are enabled.These
WARN_ONare for overflows, which should never happen. I initially thought about using therefcount_t, but it didn’t seem suitable asrefcount_add()expects that the refcount is already “1” indicating that an object was already created which doesn’t fit in the semantics of what this is. Similar forrefcount_dec().
Quoted 3 lines
In the next patch there is a
WARN_ONfor the refcount check to capture the mis-behaving context, I will add a debug print with the leaked physical address in that case as this is the important one.
I was thinking specifically about calling dump_page_owner().
Will
Re: [PATCH v2 3/4] drivers/iommu-debug-pagealloc: Track IOMMU pages
Quoted 58 lines · Will Deacon
Quoted 57 lines · Mostafa Saleh
Quoted 41 lines · Will Deacon
Quoted 40 lines · Mostafa Saleh
Using the new calls, use an atomic refcount to track how many times a page is mapped in any of the IOMMUs.
For unmap we need to use
iova_to_phys()to get the physical address of the pages.We use the smallest supported page size as the granularity of tracking per domain.
This is important as it possible to map pages and unmap them with larger sizes (as inmap_sg()) cases.Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 74 +++++++++++++++++++++++++++ 1 file changed, 74 insertions(+)drivers/iommu/iommu-debug-pagealloc.c +15
-0@@ -27,16 +27,90 @@ struct page_ext_operations page_iommu_debug_ops = { .need = need_iommu_debug, }; +static struct page_ext *get_iommu_page_ext (phys_addr_t phys) +{ +struct page *page =phys_to_page (phys); +struct page_ext *page_ext =page_ext_get (page); + +return page_ext; +} + +static struct iommu_debug_metadate *get_iommu_data (struct page_ext *page_ext) +{ +return page_ext_data (page_ext, &page_iommu_debug_ops); +} + +static void iommu_debug_inc_page (phys_addr_t phys) +{Quoted 8 lines
drivers/iommu/iommu-debug-pagealloc.c struct page_ext_operations page_iommu_debug_ops = { · line 48
Quoted 4 lines
drivers/iommu/iommu-debug-pagealloc.c +4
-0+struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + +WARN_ON (atomic_inc_return (&d->ref) <=0 );Is it worth dumping some information about the page in addition to the
WARN_ON()? That way, you might be able to benefit from other debug options (e.g.PAGE_OWNER) if they are enabled.These
WARN_ONare for overflows, which should never happen. I initially thought about using therefcount_t, but it didn’t seem suitable asrefcount_add()expects that the refcount is already “1” indicating that an object was already created which doesn’t fit in the semantics of what this is. Similar forrefcount_dec().
Quoted 5 lines
Quoted 3 lines
In the next patch there is a
WARN_ONfor the refcount check to capture the mis-behaving context, I will add a debug print with the leaked physical address in that case as this is the important one.I was thinking specifically about calling
dump_page_owner().
I see, that makes sense.
Thanks,
Mostafa
Quoted 2 lines
Will
PATCH v2 4/4 drivers/iommu-debug-pagealloc: Check mapped/unmapped kernel memory
In reply to Mostafa SalehNow, as the page_ext holds count of IOMMU mappings, we can use it to assert that any page allocated/freed is indeed not in the IOMMU.
The sanitizer doesn’t protect against mapping/unmapping during this period. However, that’s less harmful as the page is not used by the kernel.
--- drivers/iommu/iommu-debug-pagealloc.c | 19 +++++++++++++++++++ include/linux/iommu-debug-pagealloc.h | 12 ++++++++++++ include/linux/mm.h | 5 +++++ 3 files changed, 36 insertions(+)
drivers/iommu/iommu-debug-pagealloc.c +19 -0
@@ -71,6 +71,25 @@ static size_t iommu_debug_page_size(struct iommu_domain *domain)return 1UL <<__ffs (domain->pgsize_bitmap); } +static unsigned int iommu_debug_page_count (unsigned long phys) +{ +unsigned int ref; +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext); + + ref =atomic_read (&d->ref); +page_ext_put (page_ext); +return ref; +} + +void __iommu_debug_check_unmapped (const struct page *page,int numpages) +{ +while (numpages--) { +WARN_ON (iommu_debug_page_count (page_to_phys (page))); + page++; + } +} +void __iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size) {size_t off;
include/linux/iommu-debug-pagealloc.h +12 -0
@@ -22,6 +22,7 @@ void __iommu_debug_unmap(struct iommu_domain *domain, unsigned long iova, size_t size);void __iommu_debug_remap (struct iommu_domain *domain,unsigned long iova,size_t size); +void __iommu_debug_check_unmapped (const struct page *page,int numpages);static inline void iommu_debug_map (struct iommu_domain *domain,phys_addr_t phys,size_t size)
@@ -44,6 +45,12 @@ static inline void iommu_debug_remap(struct iommu_domain *domain, __iommu_debug_remap(domain, iova, size); } +static inline void iommu_debug_check_unmapped (const struct page *page,int numpages) +{ +if (static_branch_unlikely (&iommu_debug_initialized)) +__iommu_debug_check_unmapped (page, numpages); +} +void iommu_debug_init (void );#else
@@ -66,6 +73,11 @@ static inline void iommu_debug_init(void) { } +static inline void iommu_debug_check_unmapped (const struct page *page, +int numpages) +{ +} +#endif /* CONFIG_IOMMU_DEBUG_PAGEALLOC */ #endif /* __LINUX_IOMMU_DEBUG_PAGEALLOC_H */
include/linux/mm.h +5 -0
@@ -36,6 +36,7 @@#include <linux/rcuwait.h> #include <linux/bitmap.h> #include <linux/bitops.h> +#include <linux/iommu-debug-pagealloc.h> struct mempolicy ;struct anon_vma ;
@@ -3811,12 +3812,16 @@ extern void __kernel_map_pages(struct page *page, int numpages, int enable);#ifdef CONFIG_DEBUG_PAGEALLOC static inline void debug_pagealloc_map_pages (struct page *page,int numpages) { +iommu_debug_check_unmapped (page, numpages); +if (debug_pagealloc_enabled_static ())__kernel_map_pages (page, numpages,1 ); }static inline void debug_pagealloc_unmap_pages (struct page *page,int numpages) { +iommu_debug_check_unmapped (page, numpages); +if (debug_pagealloc_enabled_static ())__kernel_map_pages (page, numpages,0 ); }
Signature · 1 line
2.51.2.1026.g39e6a42477-goog
Will Deacon
maintainerRe: [PATCH v2 4/4] drivers/iommu-debug-pagealloc: Check mapped/unmapped kernel memory
Quoted 20 lines · Mostafa Saleh
Now, as the
page_extholds count of IOMMU mappings, we can use it to assert that any page allocated/freed is indeed not in the IOMMU.The sanitizer doesn’t protect against mapping/unmapping during this period. However, that’s less harmful as the page is not used by the kernel.
Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 19 +++++++++++++++++++ include/linux/iommu-debug-pagealloc.h | 12 ++++++++++++ include/linux/mm.h | 5 +++++ 3 files changed, 36 insertions(+)drivers/iommu/iommu-debug-pagealloc.c +0
-0
drivers/iommu/iommu-debug-pagealloc.c static size_t iommu_debug_page_size(struct iommu_domain *domain) · line 74
Quoted 4 lines
drivers/iommu/iommu-debug-pagealloc.c +1
-0@@ -71,6 +71,25 @@ static size_t iommu_debug_page_size(struct iommu_domain *domain)return 1UL <<__ffs (domain->pgsize_bitmap); } +static unsigned int iommu_debug_page_count (unsigned long phys)
'phys_addr_t phys' ?
But having said that, wouldn't you be better off taking the 'struct page *' here rather than converting it to a physical address only for get_iommu_page_ext() to convert it straight back again?
Quoted 4 lines
drivers/iommu/iommu-debug-pagealloc.c +4
-0+{ +unsigned int ref; +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext);
drivers/iommu/iommu-debug-pagealloc.c static size_t iommu_debug_page_size(struct iommu_domain *domain) · line 88
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +10
-0+ + ref =atomic_read (&d->ref); +page_ext_put (page_ext); +return ref; +} + +void __iommu_debug_check_unmapped (const struct page *page,int numpages) +{ +while (numpages--) { +WARN_ON (iommu_debug_page_count (page_to_phys (page)));
Since you only care about the count being non-zero, perhaps tweak iommu_debug_page_count() to be something like:
bool iommu_debug_page_referenced(struct page *);
Will
Re: [PATCH v2 4/4] drivers/iommu-debug-pagealloc: Check mapped/unmapped kernel memory
Quoted 24 lines · Will Deacon
Quoted 23 lines · Mostafa Saleh
Now, as the
page_extholds count of IOMMU mappings, we can use it to assert that any page allocated/freed is indeed not in the IOMMU.The sanitizer doesn’t protect against mapping/unmapping during this period. However, that’s less harmful as the page is not used by the kernel.
Signed-off-by Mostafa Saleh <[email protected]>Tested-by Qinxin Xia <[email protected]>--- drivers/iommu/iommu-debug-pagealloc.c | 19 +++++++++++++++++++ include/linux/iommu-debug-pagealloc.h | 12 ++++++++++++ include/linux/mm.h | 5 +++++ 3 files changed, 36 insertions(+)drivers/iommu/iommu-debug-pagealloc.c +0
-0@@ -71,6 +71,25 @@ static size_t iommu_debug_page_size(struct iommu_domain *domain)return 1UL <<__ffs (domain->pgsize_bitmap); }
Quoted 7 lines
drivers/iommu/iommu-debug-pagealloc.c static size_t iommu_debug_page_size(struct iommu_domain *domain) · line 74
Quoted 1 line
drivers/iommu/iommu-debug-pagealloc.c +1
-0+static unsigned int iommu_debug_page_count (unsigned long phys)'
phys_addr_tphys' ?But having said that, wouldn't you be better off taking the 'struct page *' here rather than converting it to a physical address only for
get_iommu_page_ext()to convert it straight back again?
Will do, we will need the physical address anyway for the error message.
Quoted 19 lines
Quoted 5 lines
drivers/iommu/iommu-debug-pagealloc.c +4
-0+{ +unsigned int ref; +struct page_ext *page_ext =get_iommu_page_ext (phys); +struct iommu_debug_metadate *d =get_iommu_data (page_ext);drivers/iommu/iommu-debug-pagealloc.c static size_t iommu_debug_page_size(struct iommu_domain *domain) · line 88
Quoted 10 lines
drivers/iommu/iommu-debug-pagealloc.c +10
-0+ + ref =atomic_read (&d->ref); +page_ext_put (page_ext); +return ref; +} + +void __iommu_debug_check_unmapped (const struct page *page,int numpages) +{ +while (numpages--) { +WARN_ON (iommu_debug_page_count (page_to_phys (page)));Since you only care about the count being non-zero, perhaps tweak
iommu_debug_page_count()to be something like:
Quoted 2 lines
bool iommu_debug_page_referenced(struct page *);
Will do.
Thanks,
Mostafa
Quoted 1 line
Will