of/irq: Fix device node refcount leak in of_check_msi_parent()
PATCH 1/1 of/irq: Fix device node refcount leak in of_check_msi_parent()
of_msi_xlate() documents @msi_np as a filter when it points at a node on entry, handing back a reference only when it points at NULL. The msi-map branch does that. of_check_msi_parent() keeps the reference from the 1:1 msi-parent match either way.
pci_msi_domain_get_msi_rid() is the only caller that passes a node in, from irq_domain_get_of_node(), and it puts nothing, so a 1:1 match there leaks the MSI controller node.
--- drivers/of/irq.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/of/irq.c +1 -1
@@ -785,7 +785,7 @@ static int of_check_msi_parent(struct device_node *dev_node, struct device_nodeif ((*msi_node && *msi_node != msi_spec.np) || msi_spec.args_count !=0 ) ret = -EINVAL ; -if (!ret) { +if (!ret && !*msi_node) {/* Return with a node reference held */ *msi_node = msi_spec.np;return 0 ;
Signature · 1 line
2.39.5
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
Pre-existing issues:
- [High] of_msi_xlate() fails to traverse the device hierarchy when msi_np is NULL, completely breaking MSI ID mapping for callers such as the fsl-mc subsystem.
- [High] msi_node reference is leaked on error path in its_v5_pci_msi_prepare()
- [High] of_msi_xlate() incorrectly handles 'msi-parent' by continuing tree traversal when it mismatches.
--
commit b8991c82a3994d7ebf04943226ad5d1d9490a8d4
Author: Fuad Tabba <[email protected]>
of/irq: Fix device node refcount leak in of_check_msi_parent()
The patch fixes a device node refcount leak in of_check_msi_parent(). It ensures that if the node was not matched in a 1:1 mapping, the reference is released correctly before returning.
drivers/of/irq.c static int of_check_msi_parent(struct device_node *dev_node, struct device_node · line 791
Quoted 13 lines
drivers/of/irq.c +1
-1@@ -785,7 +785,7 @@ static int of_check_msi_parent(struct device_node *dev_node, struct device_nodeif ((*msi_node && *msi_node != msi_spec.np) || msi_spec.args_count !=0 ) ret = -EINVAL ; -if (!ret) { +if (!ret && !*msi_node) {/* Return with a node reference held */ *msi_node = msi_spec.np;return 0 ;
[Severity: High]
This isn't a bug introduced by this patch, but of_msi_xlate() seems to incorrectly handle msi-parent mismatches.
If a device specifies an msi-parent that does not match the msi_np filter, of_check_msi_parent() returns -EINVAL.
drivers/of/irq.c:of_msi_xlate() {
/* -ENODEV: msi-map absent -> check for msi-parent */
if (msi_np && !of_check_msi_parent(parent_dev->of_node, msi_np))
break;
}The condition !of_check_msi_parent() evaluates to false, skipping the break statement and allowing traversal to continue. Will this erroneously allow traversal up the tree, potentially inheriting an incorrect msi-map from a parent bus?
[Severity: High]
This is a pre-existing issue, but does of_msi_xlate() fail to traverse the device hierarchy when msi_np is NULL?
When of_map_msi_id() (which calls of_map_id()) processes an absent map property while filter_np is NULL, it returns 0 instead of an error:
drivers/of/base.c:of_map_id() {
...
map = of_get_property(np, map_name, &map_bytes);
if (!map) {
if (filter_np)
return -ENODEV;
/* Otherwise, no map implies no translation */
arg->args[0] = id;
arg->args_count = 1;
return 0;
}
...
}In of_msi_xlate(), this return value of 0 triggers a break:
drivers/of/irq.c:of_msi_xlate() {
if (!of_map_msi_id(parent_dev->of_node, id_in, msi_np, &msi_spec)) {
...
/* msi-map present but no match -> stop walking */
break;
}
}Because of_map_id() returns 0 when the map is absent, wouldn't this early break erroneously abort the parent traversal loop on the first iteration, breaking MSI ID mapping for callers passing a NULL msi_np?
[Severity: High]
This isn't a bug introduced by this patch, but is there a fwnode reference leak on the error path in its_v5_pci_msi_prepare()?
drivers/irqchip/irq-gic-its-msi-parent.c:its_v5_pci_msi_prepare() {
rid = pci_msi_map_rid_ctlr_node(domain->parent, pdev, &msi_node);
if (!msi_node)
return -ENODEV;
ret = its_translate_frame_address(msi_node, &pa);
if (ret)
return -ENODEV;
fwnode_handle_put(msi_node);
...
}If its_translate_frame_address() fails, it returns -ENODEV without dropping the acquired msi_node reference. Can this leak the fwnode_handle if the frame address translation fails repeatedly during driver bind or unbind?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Quoted 47 lines · On Fri, 4 Sept 2026 at 08:48, <[email protected]> wrote:
Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
Pre-existing issues:
- [High]of_msi_xlate()fails to traverse the device hierarchy whenmsi_npis NULL, completely breaking MSI ID mapping for callers such as the fsl-mc subsystem.
- [High]msi_nodereference is leaked on error path inits_v5_pci_msi_prepare()
- [High]of_msi_xlate()incorrectly handles 'msi-parent' by continuing tree traversal when it mismatches.
--commit
b8991c82a3994d7ebf04943226ad5d1d9490a8d4
Author: Fuad Tabba <[email protected]>of/irq: Fix device node refcount leak in
of_check_msi_parent()The patch fixes a device node refcount leak in
of_check_msi_parent(). It ensures that if the node was not matched in a 1:1 mapping, the reference is released correctly before returning.drivers/of/irq.c static int of_check_msi_parent(struct device_node *dev_node, struct device_node · line 791
Quoted 13 lines
drivers/of/irq.c +1
-1@@ -785,7 +785,7 @@ static int of_check_msi_parent(struct device_node *dev_node, struct device_nodeif ((*msi_node && *msi_node != msi_spec.np) || msi_spec.args_count !=0 ) ret = -EINVAL ; -if (!ret) { +if (!ret && !*msi_node) {/* Return with a node reference held */ *msi_node = msi_spec.np;return 0 ;[Severity: High]
This isn't a bug introduced by this patch, butof_msi_xlate()seems to incorrectly handle msi-parent mismatches.If a device specifies an msi-parent that does not match the
msi_npfilter,of_check_msi_parent()returns-EINVAL.drivers/of/irq.c:of_msi_xlate() { /* -ENODEV: msi-map absent -> check for msi-parent */ if (msi_np && !of_check_msi_parent(parent_dev->of_node, msi_np)) break; }
Quoted 4 lines
The condition !of_check_msi_parent() evaluates to false, skipping the break statement and allowing traversal to continue. Will this erroneously allow traversal up the tree, potentially inheriting an incorrect msi-map from a parent bus?
It does continue. -EINVAL also covers #msi-cells != 0, so of_msi_xlate() cannot tell a phandle mismatch from an msi-parent that is not a 1:1 mapping, and breaking on it would stop the walk for both.
Lorenzo, Rob, should the walk stop at the first node declaring an msi-parent?
Quoted 32 lines
[Severity: High]
This is a pre-existing issue, but doesof_msi_xlate()fail to traverse the device hierarchy whenmsi_npis NULL?When
of_map_msi_id()(which callsof_map_id()) processes an absent map property whilefilter_npis NULL, it returns 0 instead of an error:drivers/of/base.c:of_map_id() { ... map = of_get_property(np, map_name, &map_bytes); if (!map) { if (filter_np) return -ENODEV; /* Otherwise, no map implies no translation */ arg->args[0] = id; arg->args_count = 1; return 0; } ... }In
of_msi_xlate(), this return value of 0 triggers a break:drivers/of/irq.c:of_msi_xlate() { if (!of_map_msi_id(parent_dev->of_node, id_in, msi_np, &msi_spec)) { ... /* msi-map present but no match -> stop walking */ break; } }
Quoted 3 lines
Because
of_map_id()returns 0 when the map is absent, wouldn't this early break erroneously abort the parent traversal loop on the first iteration, breaking MSI ID mapping for callers passing a NULLmsi_np?
It does break early there, but not for fsl-mc, the only NULL caller:
AFAICT, an fsl_mc_device never gets an of_node, so the walk carries on up to the fsl-mc platform device, which is where fsl,qoriq-mc puts msi-map.
Quoted 17 lines
[Severity: High]
This isn't a bug introduced by this patch, but is there a fwnode reference leak on the error path inits_v5_pci_msi_prepare()?drivers/irqchip/irq-gic-its-msi-parent.c:its_v5_pci_msi_prepare() { rid = pci_msi_map_rid_ctlr_node(domain->parent, pdev, &msi_node); if (!msi_node) return -ENODEV; ret = its_translate_frame_address(msi_node, &pa); if (ret) return -ENODEV; fwnode_handle_put(msi_node); ... }
Quoted 3 lines
If
its_translate_frame_address()fails, it returns-ENODEVwithout dropping the acquiredmsi_nodereference. Can this leak thefwnode_handleif the frame address translation fails repeatedly during driver bind or unbind?
Yes, every failed translation leaks one. I have a patch written, moving the put above the error check. It goes with the irqchip fixes rather than here.
/fuad
Quoted 3 lines
--
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Quoted 53 lines · Fuad Tabba
Quoted 47 lines · On Fri, 4 Sept 2026 at 08:48, <[email protected]> wrote:
Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
Pre-existing issues:
- [High]of_msi_xlate()fails to traverse the device hierarchy whenmsi_npis NULL, completely breaking MSI ID mapping for callers such as the fsl-mc subsystem.
- [High]msi_nodereference is leaked on error path inits_v5_pci_msi_prepare()
- [High]of_msi_xlate()incorrectly handles 'msi-parent' by continuing tree traversal when it mismatches.
--commit
b8991c82a3994d7ebf04943226ad5d1d9490a8d4
Author: Fuad Tabba <[email protected]>of/irq: Fix device node refcount leak in
of_check_msi_parent()The patch fixes a device node refcount leak in
of_check_msi_parent(). It ensures that if the node was not matched in a 1:1 mapping, the reference is released correctly before returning.drivers/of/irq.c static int of_check_msi_parent(struct device_node *dev_node, struct device_node · line 791
Quoted 13 lines
drivers/of/irq.c +1
-1@@ -785,7 +785,7 @@ static int of_check_msi_parent(struct device_node *dev_node, struct device_nodeif ((*msi_node && *msi_node != msi_spec.np) || msi_spec.args_count !=0 ) ret = -EINVAL ; -if (!ret) { +if (!ret && !*msi_node) {/* Return with a node reference held */ *msi_node = msi_spec.np;return 0 ;[Severity: High]
This isn't a bug introduced by this patch, butof_msi_xlate()seems to incorrectly handle msi-parent mismatches.If a device specifies an msi-parent that does not match the
msi_npfilter,of_check_msi_parent()returns-EINVAL.drivers/of/irq.c:of_msi_xlate() { /* -ENODEV: msi-map absent -> check for msi-parent */ if (msi_np && !of_check_msi_parent(parent_dev->of_node, msi_np)) break; }Quoted 4 lines
The condition !of_check_msi_parent() evaluates to false, skipping the break statement and allowing traversal to continue. Will this erroneously allow traversal up the tree, potentially inheriting an incorrect msi-map from a parent bus?
Quoted 5 lines
It does continue.
-EINVALalso covers #msi-cells != 0, soof_msi_xlate()cannot tell a phandle mismatch from an msi-parent that is not a 1:1 mapping, and breaking on it would stop the walk for both.Lorenzo, Rob, should the walk stop at the first node declaring an msi-parent?
First off, thank you for fixing this.
I think we must stop at the first msi-parent and apply the same (convoluted) logic that of_msi_xlate() executes upon of_map_msi_id()'s return to make sure ref counts are kosher (we already have an msi_spec scratch variable, we can pass that to of_check_msi_parent() - that is, always drop the msi_spec.np reference if !NULL and get one if !*msi_np to initialize it).
Quoted 59 lines
Quoted 33 lines
[Severity: High]
This is a pre-existing issue, but doesof_msi_xlate()fail to traverse the device hierarchy whenmsi_npis NULL?When
of_map_msi_id()(which callsof_map_id()) processes an absent map property whilefilter_npis NULL, it returns 0 instead of an error:drivers/of/base.c:of_map_id() { ... map = of_get_property(np, map_name, &map_bytes); if (!map) { if (filter_np) return -ENODEV; /* Otherwise, no map implies no translation */ arg->args[0] = id; arg->args_count = 1; return 0; } ... }In
of_msi_xlate(), this return value of 0 triggers a break:drivers/of/irq.c:of_msi_xlate() { if (!of_map_msi_id(parent_dev->of_node, id_in, msi_np, &msi_spec)) { ... /* msi-map present but no match -> stop walking */ break; } }Quoted 3 lines
Because
of_map_id()returns 0 when the map is absent, wouldn't this early break erroneously abort the parent traversal loop on the first iteration, breaking MSI ID mapping for callers passing a NULLmsi_np?It does break early there, but not for fsl-mc, the only NULL caller:
AFAICT, anfsl_mc_devicenever gets anof_node, so the walk carries on up to the fsl-mc platform device, which is where fsl,qoriq-mc puts msi-map.Quoted 17 lines
[Severity: High]
This isn't a bug introduced by this patch, but is there a fwnode reference leak on the error path inits_v5_pci_msi_prepare()?drivers/irqchip/irq-gic-its-msi-parent.c:its_v5_pci_msi_prepare() { rid = pci_msi_map_rid_ctlr_node(domain->parent, pdev, &msi_node); if (!msi_node) return -ENODEV; ret = its_translate_frame_address(msi_node, &pa); if (ret) return -ENODEV; fwnode_handle_put(msi_node); ... }
Quoted 7 lines
Quoted 3 lines
If
its_translate_frame_address()fails, it returns-ENODEVwithout dropping the acquiredmsi_nodereference. Can this leak thefwnode_handleif the frame address translation fails repeatedly during driver bind or unbind?Yes, every failed translation leaks one. I have a patch written, moving the put above the error check. It goes with the irqchip fixes rather than here.
Yep.
Thanks,
Lorenzo
On Fri, 4 Sept 2026 at 12:11, Lorenzo Pieralisi <[email protected]> wrote:
...
Quoted 6 lines
Quoted 5 lines
It does continue.
-EINVALalso covers #msi-cells != 0, soof_msi_xlate()cannot tell a phandle mismatch from an msi-parent that is not a 1:1 mapping, and breaking on it would stop the walk for both.Lorenzo, Rob, should the walk stop at the first node declaring an msi-parent?
Quoted 7 lines
First off, thank you for fixing this.
I think we must stop at the first msi-parent and apply the same (convoluted) logic that
of_msi_xlate()executes upon of_map_msi_id()'s return to make sure ref counts are kosher (we already have anmsi_specscratch variable, we can pass that toof_check_msi_parent()- that is, always drop the msi_spec.np reference if !NULL and get one if !*msi_np to initialize it).
Thanks for the clarification.
Cheers,
/fuad
Quoted 71 lines
Quoted 59 lines
Quoted 33 lines
[Severity: High]
This is a pre-existing issue, but doesof_msi_xlate()fail to traverse the device hierarchy whenmsi_npis NULL?When
of_map_msi_id()(which callsof_map_id()) processes an absent map property whilefilter_npis NULL, it returns 0 instead of an error:drivers/of/base.c:of_map_id() { ... map = of_get_property(np, map_name, &map_bytes); if (!map) { if (filter_np) return -ENODEV; /* Otherwise, no map implies no translation */ arg->args[0] = id; arg->args_count = 1; return 0; } ... }In
of_msi_xlate(), this return value of 0 triggers a break:drivers/of/irq.c:of_msi_xlate() { if (!of_map_msi_id(parent_dev->of_node, id_in, msi_np, &msi_spec)) { ... /* msi-map present but no match -> stop walking */ break; } }Quoted 3 lines
Because
of_map_id()returns 0 when the map is absent, wouldn't this early break erroneously abort the parent traversal loop on the first iteration, breaking MSI ID mapping for callers passing a NULLmsi_np?It does break early there, but not for fsl-mc, the only NULL caller:
AFAICT, anfsl_mc_devicenever gets anof_node, so the walk carries on up to the fsl-mc platform device, which is where fsl,qoriq-mc puts msi-map.Quoted 17 lines
[Severity: High]
This isn't a bug introduced by this patch, but is there a fwnode reference leak on the error path inits_v5_pci_msi_prepare()?drivers/irqchip/irq-gic-its-msi-parent.c:its_v5_pci_msi_prepare() { rid = pci_msi_map_rid_ctlr_node(domain->parent, pdev, &msi_node); if (!msi_node) return -ENODEV; ret = its_translate_frame_address(msi_node, &pa); if (ret) return -ENODEV; fwnode_handle_put(msi_node); ... }Quoted 7 lines
Quoted 3 lines
If
its_translate_frame_address()fails, it returns-ENODEVwithout dropping the acquiredmsi_nodereference. Can this leak thefwnode_handleif the frame address translation fails repeatedly during driver bind or unbind?Yes, every failed translation leaks one. I have a patch written, moving the put above the error check. It goes with the irqchip fixes rather than here.
Yep.
Thanks,
Lorenzo
Quoted 10 lines · Fuad Tabba
On Fri, 4 Sept 2026 at 12:11, Lorenzo Pieralisi <[email protected]> wrote:
...Quoted 8 lines
Quoted 5 lines
It does continue.
-EINVALalso covers #msi-cells != 0, soof_msi_xlate()cannot tell a phandle mismatch from an msi-parent that is not a 1:1 mapping, and breaking on it would stop the walk for both.Lorenzo, Rob, should the walk stop at the first node declaring an msi-parent?
First off, thank you for fixing this.
Quoted 7 lines
Quoted 5 lines
I think we must stop at the first msi-parent and apply the same (convoluted) logic that
of_msi_xlate()executes upon of_map_msi_id()'s return to make sure ref counts are kosher (we already have anmsi_specscratch variable, we can pass that toof_check_msi_parent()- that is, always drop the msi_spec.np reference if !NULL and get one if !*msi_np to initialize it).Thanks for the clarification.
I can send a patch myself - I understand this is a nuisance for you - I think though that since you are in the process of fixing it I don't want to get in the way either.
Let me know.
Thanks,
Lorenzo
Quoted 75 lines
Cheers,
/fuadQuoted 71 lines
Quoted 59 lines
Quoted 33 lines
[Severity: High]
This is a pre-existing issue, but doesof_msi_xlate()fail to traverse the device hierarchy whenmsi_npis NULL?When
of_map_msi_id()(which callsof_map_id()) processes an absent map property whilefilter_npis NULL, it returns 0 instead of an error:drivers/of/base.c:of_map_id() { ... map = of_get_property(np, map_name, &map_bytes); if (!map) { if (filter_np) return -ENODEV; /* Otherwise, no map implies no translation */ arg->args[0] = id; arg->args_count = 1; return 0; } ... }In
of_msi_xlate(), this return value of 0 triggers a break:drivers/of/irq.c:of_msi_xlate() { if (!of_map_msi_id(parent_dev->of_node, id_in, msi_np, &msi_spec)) { ... /* msi-map present but no match -> stop walking */ break; } }Quoted 3 lines
Because
of_map_id()returns 0 when the map is absent, wouldn't this early break erroneously abort the parent traversal loop on the first iteration, breaking MSI ID mapping for callers passing a NULLmsi_np?It does break early there, but not for fsl-mc, the only NULL caller:
AFAICT, anfsl_mc_devicenever gets anof_node, so the walk carries on up to the fsl-mc platform device, which is where fsl,qoriq-mc puts msi-map.Quoted 17 lines
[Severity: High]
This isn't a bug introduced by this patch, but is there a fwnode reference leak on the error path inits_v5_pci_msi_prepare()?drivers/irqchip/irq-gic-its-msi-parent.c:its_v5_pci_msi_prepare() { rid = pci_msi_map_rid_ctlr_node(domain->parent, pdev, &msi_node); if (!msi_node) return -ENODEV; ret = its_translate_frame_address(msi_node, &pa); if (ret) return -ENODEV; fwnode_handle_put(msi_node); ... }Quoted 7 lines
Quoted 3 lines
If
its_translate_frame_address()fails, it returns-ENODEVwithout dropping the acquiredmsi_nodereference. Can this leak thefwnode_handleif the frame address translation fails repeatedly during driver bind or unbind?Yes, every failed translation leaks one. I have a patch written, moving the put above the error check. It goes with the irqchip fixes rather than here.
Yep.
Thanks,
Lorenzo
Hi Lorenzo,
Quoted 18 lines · Lorenzo Pieralisi
Quoted 16 lines · Fuad Tabba
On Fri, 4 Sept 2026 at 12:11, Lorenzo Pieralisi <[email protected]> wrote:
...Quoted 13 lines
Quoted 5 lines
It does continue.
-EINVALalso covers #msi-cells != 0, soof_msi_xlate()cannot tell a phandle mismatch from an msi-parent that is not a 1:1 mapping, and breaking on it would stop the walk for both.Lorenzo, Rob, should the walk stop at the first node declaring an msi-parent?
First off, thank you for fixing this.
I think we must stop at the first msi-parent and apply the same (convoluted) logic that
of_msi_xlate()executes upon of_map_msi_id()'s return to make sure ref counts are kosher (we already have anmsi_specscratch variable, we can pass that toof_check_msi_parent()- that is, always drop the msi_spec.np reference if !NULL and get one if !*msi_np to initialize it).
Quoted 7 lines
Quoted 1 line
Thanks for the clarification.
I can send a patch myself - I understand this is a nuisance for you - I think though that since you are in the process of fixing it I don't want to get in the way either.
Let me know.
I practically have it ready. Will send it soon.
Thanks for the offer.
/fuad
Quoted 79 lines
Thanks,
LorenzoQuoted 75 lines
Cheers,
/fuadQuoted 71 lines
Quoted 59 lines
Quoted 33 lines
[Severity: High]
This is a pre-existing issue, but doesof_msi_xlate()fail to traverse the device hierarchy whenmsi_npis NULL?When
of_map_msi_id()(which callsof_map_id()) processes an absent map property whilefilter_npis NULL, it returns 0 instead of an error:drivers/of/base.c:of_map_id() { ... map = of_get_property(np, map_name, &map_bytes); if (!map) { if (filter_np) return -ENODEV; /* Otherwise, no map implies no translation */ arg->args[0] = id; arg->args_count = 1; return 0; } ... }In
of_msi_xlate(), this return value of 0 triggers a break:drivers/of/irq.c:of_msi_xlate() { if (!of_map_msi_id(parent_dev->of_node, id_in, msi_np, &msi_spec)) { ... /* msi-map present but no match -> stop walking */ break; } }Quoted 3 lines
Because
of_map_id()returns 0 when the map is absent, wouldn't this early break erroneously abort the parent traversal loop on the first iteration, breaking MSI ID mapping for callers passing a NULLmsi_np?It does break early there, but not for fsl-mc, the only NULL caller:
AFAICT, anfsl_mc_devicenever gets anof_node, so the walk carries on up to the fsl-mc platform device, which is where fsl,qoriq-mc puts msi-map.Quoted 17 lines
[Severity: High]
This isn't a bug introduced by this patch, but is there a fwnode reference leak on the error path inits_v5_pci_msi_prepare()?drivers/irqchip/irq-gic-its-msi-parent.c:its_v5_pci_msi_prepare() { rid = pci_msi_map_rid_ctlr_node(domain->parent, pdev, &msi_node); if (!msi_node) return -ENODEV; ret = its_translate_frame_address(msi_node, &pa); if (ret) return -ENODEV; fwnode_handle_put(msi_node); ... }Quoted 7 lines
Quoted 3 lines
If
its_translate_frame_address()fails, it returns-ENODEVwithout dropping the acquiredmsi_nodereference. Can this leak thefwnode_handleif the frame address translation fails repeatedly during driver bind or unbind?Yes, every failed translation leaks one. I have a patch written, moving the put above the error check. It goes with the irqchip fixes rather than here.
Yep.
Thanks,
Lorenzo