drivers/iommu/iova.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
When iova_domain_init_rcaches() fails to allocate an iova_magazine
during the initialization of per-cpu rcaches, it jumps to out_err and
calls free_iova_rcaches() for cleanup.
In free_iova_rcaches(), the code iterates through all possible CPUs to
free both cpu_rcache->loaded and cpu_rcache->prev. However, if the
original allocation failed mid-way through the CPU loop, the pointers
for the remaining CPUs remain NULL.
Since kmem_cache_free() does not explicitly handle NULL pointers like
kfree() does, passing these NULL pointers leads to a kernel paging
request fault.
The crash can be observed when fault injection triggers a failure in
iova_magazine_alloc():
pc : _compound_head include/linux/page-flags.h:251 [inline]
pc : virt_to_folio include/linux/mm.h:1395 [inline]
pc : virt_to_slab mm/slab.h:215 [inline]
pc : kmem_cache_free+0x90/0x5c4 mm/slub.c:4769
...
Call trace:
trace_kmem_cache_free include/trace/events/kmem.h:114 [inline]
kmem_cache_free+0x90/0x5c4 mm/slub.c:4768
iova_magazine_free drivers/iommu/iova.c:623 [inline]
free_iova_rcaches+0x114/0x344 drivers/iommu/iova.c:894
iova_domain_init_rcaches drivers/iommu/iova.c:762 [inline]
iova_domain_init_rcaches+0x3d0/0x528 drivers/iommu/iova.c:717
Add a NULL check in iova_magazine_free() to safely handle partially
initialized rcaches in error paths.
Fixes: 84e6f56be9c6 ("iommu/iova: use named kmem_cache for iova magazines")
Signed-off-by: Lynn Liu <liulynn@google.com>
---
v1: https://lore.kernel.org/linux-iommu/20260214080919.390846-1-liulynn@google.com/T/#u
Changelog in v1 -> v2:
- Update author name to full name.
- Add Fixes tag.
- Include the Call trace from the kernel crash.
drivers/iommu/iova.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c
index 18f839721813..e026be5e068b 100644
--- a/drivers/iommu/iova.c
+++ b/drivers/iommu/iova.c
@@ -611,7 +611,8 @@ static struct iova_magazine *iova_magazine_alloc(gfp_t flags)
static void iova_magazine_free(struct iova_magazine *mag)
{
- kmem_cache_free(iova_magazine_cache, mag);
+ if (mag)
+ kmem_cache_free(iova_magazine_cache, mag);
}
static void
--
2.53.0.473.g4a7958ca14-goog
On 2026-03-04 8:06 am, Lynn Liu wrote:
> When iova_domain_init_rcaches() fails to allocate an iova_magazine
> during the initialization of per-cpu rcaches, it jumps to out_err and
> calls free_iova_rcaches() for cleanup.
>
> In free_iova_rcaches(), the code iterates through all possible CPUs to
> free both cpu_rcache->loaded and cpu_rcache->prev. However, if the
> original allocation failed mid-way through the CPU loop, the pointers
> for the remaining CPUs remain NULL.
>
> Since kmem_cache_free() does not explicitly handle NULL pointers like
> kfree() does, passing these NULL pointers leads to a kernel paging
> request fault.
>
> The crash can be observed when fault injection triggers a failure in
> iova_magazine_alloc():
>
> pc : _compound_head include/linux/page-flags.h:251 [inline]
> pc : virt_to_folio include/linux/mm.h:1395 [inline]
> pc : virt_to_slab mm/slab.h:215 [inline]
> pc : kmem_cache_free+0x90/0x5c4 mm/slub.c:4769
> ...
> Call trace:
> trace_kmem_cache_free include/trace/events/kmem.h:114 [inline]
> kmem_cache_free+0x90/0x5c4 mm/slub.c:4768
> iova_magazine_free drivers/iommu/iova.c:623 [inline]
> free_iova_rcaches+0x114/0x344 drivers/iommu/iova.c:894
> iova_domain_init_rcaches drivers/iommu/iova.c:762 [inline]
> iova_domain_init_rcaches+0x3d0/0x528 drivers/iommu/iova.c:717
>
> Add a NULL check in iova_magazine_free() to safely handle partially
> initialized rcaches in error paths.
Wouldn't it be even better to just break the for_each_possible_cpu()
loop in free_iova_rcaches() on the first NULL, much like the outer
rcache->cpu_rcaches condition? Since the cleanup is already trying to be
symmetric to the allocation, I'd rather fix that than add a blanket
check that's redundant for every other caller.
Thanks,
Robin.
> Fixes: 84e6f56be9c6 ("iommu/iova: use named kmem_cache for iova magazines")
> Signed-off-by: Lynn Liu <liulynn@google.com>
> ---
> v1: https://lore.kernel.org/linux-iommu/20260214080919.390846-1-liulynn@google.com/T/#u
> Changelog in v1 -> v2:
> - Update author name to full name.
> - Add Fixes tag.
> - Include the Call trace from the kernel crash.
>
> drivers/iommu/iova.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c
> index 18f839721813..e026be5e068b 100644
> --- a/drivers/iommu/iova.c
> +++ b/drivers/iommu/iova.c
> @@ -611,7 +611,8 @@ static struct iova_magazine *iova_magazine_alloc(gfp_t flags)
>
> static void iova_magazine_free(struct iova_magazine *mag)
> {
> - kmem_cache_free(iova_magazine_cache, mag);
> + if (mag)
> + kmem_cache_free(iova_magazine_cache, mag);
> }
>
> static void
In iova_domain_init_rcaches(), when an iova_magazine allocation fails,
the code jumps to the error path and calls free_iova_rcaches().
In free_iova_rcaches(), the code iterates through all possible CPUs to
free both cpu_rcache->loaded and cpu_rcache->prev. However, if the
original allocation failed mid-way through the CPU loop, the pointers
for the remaining CPUs remain NULL. free_iova_rcaches() then blindly
attempts to free these NULL pointers, which causes a kernel paging
request fault in kmem_cache_free().
The crash can be observed when fault injection triggers a failure in
iova_magazine_alloc():
pc : _compound_head include/linux/page-flags.h:251 [inline]
pc : virt_to_folio include/linux/mm.h:1395 [inline]
pc : virt_to_slab mm/slab.h:215 [inline]
pc : kmem_cache_free+0x90/0x5c4 mm/slub.c:4769
...
Call trace:
trace_kmem_cache_free include/trace/events/kmem.h:114 [inline]
kmem_cache_free+0x90/0x5c4 mm/slub.c:4768
iova_magazine_free drivers/iommu/iova.c:623 [inline]
free_iova_rcaches+0x114/0x344 drivers/iommu/iova.c:894
iova_domain_init_rcaches drivers/iommu/iova.c:762 [inline]
iova_domain_init_rcaches+0x3d0/0x528 drivers/iommu/iova.c:717
These changes update iova_domain_init_rcaches() to check for allocation
failure immediately after each individual magazine allocation.
Correspondingly, update free_iova_rcaches() to break the cleanup loop
when it encounters a NULL pointer, ensuring only successfully allocated
resources are processed.
Fixes: 84e6f56be9c6 ("iommu/iova: use named kmem_cache for iova magazines")
Signed-off-by: Lynn Liu <liulynn@google.com>
---
v3:
- Split magazine allocation checks in iova_domain_init_rcaches().
- Add early breaks in free_iova_rcaches() to safely handle partial allocations as suggested.
Link to v2: https://lore.kernel.org/linux-iommu/116b79a2-fadb-4796-8cb7-9c2dfde54999@arm.com/T/#t
v2:
- Update author name to full name.
- Add Fixes tag.
- Include the Call trace from the kernel crash.
Link to v1: https://lore.kernel.org/linux-iommu/20260214080919.390846-1-liulynn@google.com/T/#u
drivers/iommu/iova.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c
index 18f839721813..cf6cd9f23b97 100644
--- a/drivers/iommu/iova.c
+++ b/drivers/iommu/iova.c
@@ -740,8 +740,12 @@ int iova_domain_init_rcaches(struct iova_domain *iovad)
spin_lock_init(&cpu_rcache->lock);
cpu_rcache->loaded = iova_magazine_alloc(GFP_KERNEL);
+ if (!cpu_rcache->loaded) {
+ ret = -ENOMEM;
+ goto out_err;
+ }
cpu_rcache->prev = iova_magazine_alloc(GFP_KERNEL);
- if (!cpu_rcache->loaded || !cpu_rcache->prev) {
+ if (!cpu_rcache->prev) {
ret = -ENOMEM;
goto out_err;
}
@@ -886,7 +890,11 @@ static void free_iova_rcaches(struct iova_domain *iovad)
break;
for_each_possible_cpu(cpu) {
cpu_rcache = per_cpu_ptr(rcache->cpu_rcaches, cpu);
+ if (!cpu_rcache->loaded)
+ break;
iova_magazine_free(cpu_rcache->loaded);
+ if (!cpu_rcache->prev)
+ break;
iova_magazine_free(cpu_rcache->prev);
}
free_percpu(rcache->cpu_rcaches);
--
2.53.0.473.g4a7958ca14-goog
© 2016 - 2026 Red Hat, Inc.