Skip to contentSkip to conversation

firmware/clk: arm_scpi hardening (leak, OPP bounds, cpufreq)

PATCH Superseded Xixin Liu 2 participants 19 messages Jul 27 – Sep 7

Subsystems (2)
lore
Share & tools
lore

Xixin Liu

author
#0 lore raw

cover firmware/clk: arm_scpi hardening (leak, OPP bounds, cpufreq)

Hi,

This series hardens the OF SCPI firmware and clock paths against a few real defects found while reviewing linux-next:

1) device_node leak in scpi_dev_domain_id() after of_parse_phandle_with_args()
2) DVFS OPP count from SCP trusted beyond MAX_DVFS_OPPS (OOB read / bad OPP table size)
3) DVFS index used as clock rate without an upper bound check
4) scpi-cpufreq registered once only, and cleared on register failure so remove() does not unregister an ERR_PTR

Please review.

Thanks,
Xixin Liu

---

Xixin Liu (4):
firmware: arm_scpi: fix device_node leak in scpi_dev_domain_id
firmware: arm_scpi: reject DVFS OPP count above MAX_DVFS_OPPS
clk: scpi: bound-check DVFS index in scpi_dvfs_recalc_rate
clk: scpi: register scpi-cpufreq once and clear on failure

 drivers/clk/clk-scpi.c      | 8 ++++++--
 drivers/firmware/arm_scpi.c | 9 ++++++---
 2 files changed, 12 insertions(+), 5 deletions(-)
Signature · 1 line
2.53.0
author
#1 lore raw

PATCH 1/4 firmware: arm_scpi: fix device_node leak in scpi_dev_domain_id

of_parse_phandle_with_args() takes a reference on clkspec.np that must be released with of_node_put(). scpi_dev_domain_id() returned clkspec.args[0] without dropping that reference, so every domain lookup leaked a device node. Paths such as scpi_dvfs_info() / cpufreq init call this per CPU, so the leak accumulates over time.

Save the domain id, of_node_put(clkspec.np), then return the saved value.

---
 drivers/firmware/arm_scpi.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/firmware/arm_scpi.c +4 -1
index 87c323de17b9..6056754ef48c 100644
@@ -660,12 +660,15 @@ static struct scpi_dvfs_info *scpi_dvfs_get_info(u8 domain)
 static int scpi_dev_domain_id(struct device *dev)
 {
 	struct of_phandle_args clkspec;
+	int domain;
 
 	if (of_parse_phandle_with_args(dev->of_node, "clocks", "#clock-cells",
 				       0, &clkspec))
 		return -EINVAL;
 
-	return clkspec.args[0];
+	domain = clkspec.args[0];
+	of_node_put(clkspec.np);
+	return domain;
 }
 
 static struct scpi_dvfs_info *scpi_dvfs_info(struct device *dev)
Signature · 1 line
2.53.0
author
#2 lore raw

PATCH 2/4 firmware: arm_scpi: reject DVFS OPP count above MAX_DVFS_OPPS

In reply to Xixin Liu

scpi_dvfs_get_info() already rejected a zero opp_count, but still trusted any larger value from the SCP firmware. The shared-memory reply only holds MAX_DVFS_OPPS entries in buf.opps[]; a bigger count over-reads that array and then sizes the allocated OPP table incorrectly (garbage OPPs / OOB).

Reject zero and out-of-range counts in one check and return -EINVAL.

---
 drivers/firmware/arm_scpi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/firmware/arm_scpi.c +2 -2
index 6056754ef48c..5db58bdfc947 100644
@@ -630,8 +630,8 @@ static struct scpi_dvfs_info *scpi_dvfs_get_info(u8 domain)
 	if (ret)
 		return ERR_PTR(ret);
 
-	if (!buf.opp_count)
-		return ERR_PTR(-ENOENT);
+	if (!buf.opp_count || buf.opp_count > MAX_DVFS_OPPS)
+		return ERR_PTR(-EINVAL);
 
 	info = kmalloc(sizeof(*info), GFP_KERNEL);
 	if (!info)
Signature · 1 line
2.53.0
maintainer
#3 lore raw

Re: [PATCH v1 2/4] firmware: arm_scpi: reject DVFS OPP count above MAX_DVFS_OPPS

Quoted 7 lines · Xixin Liu

scpi_dvfs_get_info() already rejected a zero opp_count, but still trusted any larger value from the SCP firmware. The shared-memory reply only holds MAX_DVFS_OPPS entries in buf.opps[]; a bigger count over-reads that array and then sizes the allocated OPP table incorrectly (garbage OPPs / OOB).

Reject zero and out-of-range counts in one check and return -EINVAL.

This fails to apply, you need to rebase to the latest.

Signature · 2 lines
Regards,
Sudeep
maintainer
#4 lore raw

Re: [PATCH v1 2/4] firmware: arm_scpi: reject DVFS OPP count above MAX_DVFS_OPPS

In reply to Xixin Liu
Quoted 7 lines · Xixin Liu

scpi_dvfs_get_info() already rejected a zero opp_count, but still trusted any larger value from the SCP firmware. The shared-memory reply only holds MAX_DVFS_OPPS entries in buf.opps[]; a bigger count over-reads that array and then sizes the allocated OPP table incorrectly (garbage OPPs / OOB).

Reject zero and out-of-range counts in one check and return -EINVAL.

This issue of out-of-bounds read is present since the original SCPI DVFS implementation. So please include the below fixes tag:

Fixes 8cb7cf56c9fe ("firmware: add support for ARM System Control and Power Interface(SCPI) protocol")
Signature · 2 lines
Regards,
Sudeep
author
#5 lore raw

PATCH 3/4 clk: scpi: bound-check DVFS index in scpi_dvfs_recalc_rate

In reply to Xixin Liu

dvfs_get_idx() may return an out-of-range index if the SCP firmware is buggy or returns a stale value. Only negative indexes were rejected, so a large index walked past info->opps and could treat garbage as a clock rate (KASAN OOB / wrong frequency to consumers).

Treat indexes >= opp count as invalid and return 0, same as idx < 0.

---
 drivers/clk/clk-scpi.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/clk/clk-scpi.c +1 -1
index 19d530d52e64..4e22ba12b157 100644
@@ -85,7 +85,7 @@ static unsigned long scpi_dvfs_recalc_rate(struct clk_hw *hw,
 	int idx = clk->scpi_ops->dvfs_get_idx(clk->id);
 	const struct scpi_opp *opp;
 
-	if (idx < 0)
+	if (idx < 0 || idx >= clk->info->count)
 		return 0;
 
 	opp = clk->info->opps + idx;
Signature · 1 line
2.53.0
maintainer
#6 lore raw

Re: [PATCH v1 3/4] clk: scpi: bound-check DVFS index in scpi_dvfs_recalc_rate

Quoted 7 lines · Xixin Liu

dvfs_get_idx() may return an out-of-range index if the SCP firmware is buggy or returns a stale value. Only negative indexes were rejected, so a large index walked past info->opps and could treat garbage as a clock rate (KASAN OOB / wrong frequency to consumers).

Treat indexes >= opp count as invalid and return 0, same as idx < 0.

Again this issue of out-of-bounds read is in the original SCPI clock driver. So include the below fixes tag:

Fixes cd52c2a4b5c4 ("clk: add support for clocks provided by SCP(System Control Processor)")
Signature · 2 lines
Regards,
Sudeep
author
#7 lore raw

PATCH 4/4 clk: scpi: register scpi-cpufreq once and clear on failure

In reply to Xixin Liu

scpi_clk_probe() walks clock children and, for each DVFS provider, calls platform_device_register_simple("scpi-cpufreq"). Two related bugs:

1) Multiple DVFS children each spawned another virtual cpufreq device. If cpufreq_dev is already set, continue and skip a second register.

2) On IS_ERR, the pointer still held ERR_PTR (e.g. -ENOMEM). remove() treats any non-NULL cpufreq_dev as live and would call platform_device_unregister() on the error pointer (oops). Clear cpufreq_dev to NULL after the failure warn so remove() skips it.

---
 drivers/clk/clk-scpi.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/clk/clk-scpi.c +5 -1
index 4e22ba12b157..ce6baea8f82e 100644
@@ -283,10 +283,14 @@ static int scpi_clocks_probe(struct platform_device *pdev)
 		if (match->data != &scpi_dvfs_ops)
 			continue;
 		/* Add the virtual cpufreq device if it's DVFS clock provider */
+		if (cpufreq_dev)
+			continue;
 		cpufreq_dev = platform_device_register_simple("scpi-cpufreq",
 							      -1, NULL, 0);
-		if (IS_ERR(cpufreq_dev))
+		if (IS_ERR(cpufreq_dev)) {
 			pr_warn("unable to register cpufreq device");
+			cpufreq_dev = NULL;
+		}
 	}
 	return 0;
 }
Signature · 1 line
2.53.0
maintainer
#8 lore raw

Re: [PATCH v1 4/4] clk: scpi: register scpi-cpufreq once and clear on failure

Quoted 3 lines · Xixin Liu

scpi_clk_probe() walks clock children and, for each DVFS provider, calls platform_device_register_simple("scpi-cpufreq"). Two related bugs:

The function named above appears to be scpi_clocks_probe(), not scpi_clk_probe(). Fix that.

Quoted 3 lines

1) Multiple DVFS children each spawned another virtual cpufreq device. If cpufreq_dev is already set, continue and skip a second register.

The registration uses id -1, so every device must be named "scpi-cpufreq". A second device_add() cannot register that name and must return -EEXIST; platform_device_register_full() then drops the second pdev.

So the above information could be misleading.

Can you state it as each additional DVFS child attempts a duplicate registration and overwrites the first valid pointer with ERR_PTR(-EEXIST), rather than saying that each child spawns another device?

Quoted 5 lines

2) On IS_ERR, the pointer still held ERR_PTR (e.g. -ENOMEM). remove() treats any non-NULL cpufreq_dev as live and would call platform_device_unregister() on the error pointer (oops). Clear cpufreq_dev to NULL after the failure warn so remove() skips it.

This also fixes bugs introduced by the original virtual device support and by moving registration inside the child loop. So

Fixes 9490f01e2471 ("clk: scpi: add support for cpufreq virtual device")
Fixes 67bcc2c5f1da ("clk: scpi: don't add cpufreq device if the scpi dvfs node is disabled")
Signature · 2 lines
Regards,
Sudeep
author
#9 lore raw

v2 cover firmware/clk: arm_scpi hardening (leak, OPP bounds, cpufreq)

In reply to Xixin Liu

Hi Sudeep,

Thanks for reviewing v1.

This v2 hardens the OF SCPI firmware and clock paths against a few real defects found while reviewing linux-next:

1) device_node leak in scpi_dev_domain_id() after of_parse_phandle_with_args()
2) DVFS OPP count from SCP trusted beyond MAX_DVFS_OPPS (OOB read / bad OPP table size)
3) DVFS index used as clock rate without an upper bound check
4) scpi-cpufreq registered once only, and cleared on register failure so remove() does not unregister an ERR_PTR
5) use PLATFORM_DEVID_NONE instead of bare -1 for the scpi-cpufreq platform device id (readability; Fixes the original registration)

Changes since v1:
- 2/5: add Fixes: 8cb7cf56c9fe ("firmware: add support for ARM System Control and Power Interface(SCPI) protocol")
- 3/5: add Fixes: cd52c2a4b5c4 ("clk: add support for clocks provided by SCP(System Control Processor)")
- 4/5: correct commit message (scpi_clocks_probe; multi-DVFS child overwrites the first good cpufreq_dev with ERR_PTR(-EEXIST)) and add Fixes: 9490f01e2471 / 67bcc2c5f1da
All per maintainer feedback; code unchanged.
- add 5/5: clk: scpi: use PLATFORM_DEVID_NONE for scpi-cpufreq (readability; Fixes: 9490f01e2471)
- Rebased onto linux-next next-20260727 (0d33d21e47d9); 2/5 context updated for kmalloc_obj() (logic unchanged).

Please review.

Thanks,
Xixin Liu

---

Xixin Liu (5):
firmware: arm_scpi: fix device_node leak in scpi_dev_domain_id
firmware: arm_scpi: reject DVFS OPP count above MAX_DVFS_OPPS
clk: scpi: bound-check DVFS index in scpi_dvfs_recalc_rate
clk: scpi: register scpi-cpufreq once and clear on failure
clk: scpi: use PLATFORM_DEVID_NONE for scpi-cpufreq

 drivers/clk/clk-scpi.c      | 10 ++++++++--
 drivers/firmware/arm_scpi.c |  9 ++++++---
 2 files changed, 13 insertions(+), 6 deletions(-)
Signature · 1 line
2.53.0
author
#10 lore raw

PATCH v2 1/5 firmware: arm_scpi: fix device_node leak in scpi_dev_domain_id

of_parse_phandle_with_args() takes a reference on clkspec.np that must be released with of_node_put(). scpi_dev_domain_id() returned clkspec.args[0] without dropping that reference, so every domain lookup leaked a device node. Paths such as scpi_dvfs_info() / cpufreq init call this per CPU, so the leak accumulates over time.

Save the domain id, of_node_put(clkspec.np), then return the saved value.

---
 drivers/firmware/arm_scpi.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/firmware/arm_scpi.c +4 -1
index 2acad5fa5a28..43d6d9bbc7a9 100644
@@ -661,12 +661,15 @@ static struct scpi_dvfs_info *scpi_dvfs_get_info(u8 domain)
 static int scpi_dev_domain_id(struct device *dev)
 {
 	struct of_phandle_args clkspec;
+	int domain;
 
 	if (of_parse_phandle_with_args(dev->of_node, "clocks", "#clock-cells",
 				       0, &clkspec))
 		return -EINVAL;
 
-	return clkspec.args[0];
+	domain = clkspec.args[0];
+	of_node_put(clkspec.np);
+	return domain;
 }
 
 static struct scpi_dvfs_info *scpi_dvfs_info(struct device *dev)
Signature · 1 line
2.53.0
author
#11 lore raw

PATCH v2 2/5 firmware: arm_scpi: reject DVFS OPP count above MAX_DVFS_OPPS

In reply to Xixin Liu

scpi_dvfs_get_info() already rejected a zero opp_count, but still trusted any larger value from the SCP firmware. The shared-memory reply only holds MAX_DVFS_OPPS entries in buf.opps[]; a bigger count over-reads that array and then sizes the allocated OPP table incorrectly (garbage OPPs / OOB). The missing upper bound dates back to the original SCPI DVFS support.

Reject zero and out-of-range counts in one check and return -EINVAL.

Fixes 8cb7cf56c9fe ("firmware: add support for ARM System Control and Power Interface(SCPI) protocol")
---
 drivers/firmware/arm_scpi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/firmware/arm_scpi.c +2 -2
index 43d6d9bbc7a9..68a730d22037 100644
@@ -631,8 +631,8 @@ static struct scpi_dvfs_info *scpi_dvfs_get_info(u8 domain)
 	if (ret)
 		return ERR_PTR(ret);
 
-	if (!buf.opp_count)
-		return ERR_PTR(-ENOENT);
+	if (!buf.opp_count || buf.opp_count > MAX_DVFS_OPPS)
+		return ERR_PTR(-EINVAL);
 
 	info = kmalloc_obj(*info);
 	if (!info)
Signature · 1 line
2.53.0
author
#12 lore raw

PATCH v2 3/5 clk: scpi: bound-check DVFS index in scpi_dvfs_recalc_rate

In reply to Xixin Liu

dvfs_get_idx() may return an out-of-range index if the SCP firmware is buggy or returns a stale value. Only negative indexes were rejected, so a large index walked past info->opps and could treat garbage as a clock rate (KASAN OOB / wrong frequency to consumers). The missing upper bound dates back to the original SCPI clock driver.

Treat indexes >= opp count as invalid and return 0, same as idx < 0.

Fixes cd52c2a4b5c4 ("clk: add support for clocks provided by SCP(System Control Processor)")
---
 drivers/clk/clk-scpi.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/clk/clk-scpi.c +1 -1
index 24cee7c9fda6..3c33a9e3c690 100644
@@ -85,7 +85,7 @@ static unsigned long scpi_dvfs_recalc_rate(struct clk_hw *hw,
 	int idx = clk->scpi_ops->dvfs_get_idx(clk->id);
 	const struct scpi_opp *opp;
 
-	if (idx < 0)
+	if (idx < 0 || idx >= clk->info->count)
 		return 0;
 
 	opp = clk->info->opps + idx;
Signature · 1 line
2.53.0
author
#13 lore raw

PATCH v2 4/5 clk: scpi: register scpi-cpufreq once and clear on failure

In reply to Xixin Liu

scpi_clocks_probe() walks clock children and, for each DVFS provider, calls platform_device_register_simple("scpi-cpufreq", -1, ...). Two related bugs:

1) Registration uses id = -1 and the fixed name "scpi-cpufreq". A second registration of the same name returns -EEXIST and overwrites the first good cpufreq_dev with ERR_PTR(-EEXIST). When cpufreq_dev is already set, continue so only the first successful registration remains.

2) On IS_ERR, cpufreq_dev still held the ERR_PTR; remove() would then call platform_device_unregister() on it (oops). Clear cpufreq_dev to NULL after the failure warn so remove() skips it.

Fixes 9490f01e2471 ("clk: scpi: add support for cpufreq virtual device")
Fixes 67bcc2c5f1da ("clk: scpi: don't add cpufreq device if the scpi dvfs node is disabled")
---
 drivers/clk/clk-scpi.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/clk/clk-scpi.c +5 -1
index 3c33a9e3c690..e1891351c221 100644
@@ -284,10 +284,14 @@ static int scpi_clocks_probe(struct platform_device *pdev)
 		if (match->data != &scpi_dvfs_ops)
 			continue;
 		/* Add the virtual cpufreq device if it's DVFS clock provider */
+		if (cpufreq_dev)
+			continue;
 		cpufreq_dev = platform_device_register_simple("scpi-cpufreq",
 							      -1, NULL, 0);
-		if (IS_ERR(cpufreq_dev))
+		if (IS_ERR(cpufreq_dev)) {
 			pr_warn("unable to register cpufreq device");
+			cpufreq_dev = NULL;
+		}
 	}
 	return 0;
 }
Signature · 1 line
2.53.0
author
#14 lore raw

PATCH v2 5/5 clk: scpi: use PLATFORM_DEVID_NONE for scpi-cpufreq

In reply to Xixin Liu

Replace the magic -1 passed to platform_device_register_simple() with PLATFORM_DEVID_NONE. The constant is defined as (-1) in platform_device.h; this is a readability cleanup only and does not change behavior.

Fixes 9490f01e2471 ("clk: scpi: add support for cpufreq virtual device")
---
 drivers/clk/clk-scpi.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/clk/clk-scpi.c +1 -1
index e1891351c221..900488dedf95 100644
@@ -287,7 +287,7 @@ static int scpi_clocks_probe(struct platform_device *pdev)
 		if (cpufreq_dev)
 			continue;
 		cpufreq_dev = platform_device_register_simple("scpi-cpufreq",
-							      -1, NULL, 0);
+							      PLATFORM_DEVID_NONE, NULL, 0);
 		if (IS_ERR(cpufreq_dev)) {
 			pr_warn("unable to register cpufreq device");
 			cpufreq_dev = NULL;
Signature · 1 line
2.53.0
maintainer
#15 lore raw

Re: [PATCH v2 5/5] clk: scpi: use PLATFORM_DEVID_NONE for scpi-cpufreq

Quoted 6 lines · Xixin Liu

Replace the magic -1 passed to platform_device_register_simple() with PLATFORM_DEVID_NONE. The constant is defined as (-1) in platform_device.h; this is a readability cleanup only and does not change behavior.

Fixes 9490f01e2471 ("clk: scpi: add support for cpufreq virtual device")

This is not a fix, just a clean. I will drop this when applying.

Signature · 2 lines
Regards,
Sudeep
author
#16 lore raw

Re: [PATCH v2 5/5] clk: scpi: use PLATFORM_DEVID_NONE for scpi-cpufreq

Hi Sudeep,

Just a gentle ping on this v2 series from 2026-07-28. Dropping the 5/5 cleanup on apply is fine with me. Could you please apply the rest when you have a chance?

Thanks,
Xixin Liu

maintainer Applied
#17 lore raw
In reply to Xixin Liu
Quoted 15 lines · Xixin Liu

Thanks for reviewing v1.

This v2 hardens the OF SCPI firmware and clock paths against a few real defects found while reviewing linux-next:

1) device_node leak in scpi_dev_domain_id() after of_parse_phandle_with_args()
2) DVFS OPP count from SCP trusted beyond MAX_DVFS_OPPS (OOB read / bad OPP table size)
3) DVFS index used as clock rate without an upper bound check
4) scpi-cpufreq registered once only, and cleared on register failure so remove() does not unregister an ERR_PTR
5) use PLATFORM_DEVID_NONE instead of bare -1 for the scpi-cpufreq platform device id (readability; Fixes the original registration)

Quoted 1 line

[...]

Applied to sudeep.holla/linux (for-next/scmi/fixes), thanks!

[1/5] firmware: arm_scpi: fix device_node leak in scpi_dev_domain_id

      https://git.kernel.org/sudeep.holla/c/65320b642025
[2/5] firmware: arm_scpi: reject DVFS OPP count above MAX_DVFS_OPPS
      https://git.kernel.org/sudeep.holla/c/32471d84a487
[3/5] clk: scpi: bound-check DVFS index in scpi_dvfs_recalc_rate
      https://git.kernel.org/sudeep.holla/c/70f4b78d560e
[4/5] clk: scpi: register scpi-cpufreq once and clear on failure
      https://git.kernel.org/sudeep.holla/c/ab06cf8152da
[5/5] clk: scpi: use PLATFORM_DEVID_NONE for scpi-cpufreq
      https://git.kernel.org/sudeep.holla/c/86d923a882f4
Signature · 2 lines
Regards,
Sudeep
author
#18 lore raw

Hi Sudeep,

Thanks for applying the series.

Thanks,
Xixin Liu