Skip to contentSkip to conversation

misc: pci_endpoint_test: doorbell fixes

PATCH Superseded [email protected] 3 participants 6 messages Apr 4 – Apr 10

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover misc: pci_endpoint_test: doorbell fixes

From: Carlos Bilbao <[email protected]>

pci_endpoint_test_doorbell() reads the BAR number from the endpoint's test register space and uses it directly as an index into test->bar[] without bounds checking. Since the value is a raw u32 written by the endpoint firmware, any value is possible; values >= PCI_STD_NUM_BARS result in an out-of-bounds array access.

Patch 1 adds the missing bounds check.

Patch 2 removes a dead read of the same register that precedes the writel sequence; the DB_BAR register carries no read side effect and the assigned value is unconditionally overwritten before use.

Carlos Bilbao (2):
misc: pci_endpoint_test: validate BAR index in doorbell test
misc: pci_endpoint_test: remove dead BAR read before doorbell trigger

---
 drivers/misc/pci_endpoint_test.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
#1 lore raw

PATCH 1/2 misc: pci_endpoint_test: validate BAR index in doorbell test

From: Carlos Bilbao <[email protected]>

pci_endpoint_test_doorbell() reads the BAR number directly from an endpoint test register and uses it as an index into test->bar[] without bounds checking. Since the value is a raw u32 from device MMIO, any value is possible and if greater than or equal to PCI_STD_NUM_BARS the access goes out of bounds.

Add a bounds check before dereferencing test->bar[bar].

Fixes eefb83790a0d ("misc: pci_endpoint_test: Add doorbell test case")
---
 drivers/misc/pci_endpoint_test.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/misc/pci_endpoint_test.c +5 -0
index 74ab5b5b9011..276bed3f1c18 100644
@@ -1089,6 +1089,11 @@ static int pci_endpoint_test_doorbell(struct pci_endpoint_test *test)
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_STATUS, 0);
 
 	bar = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_BAR);
+	if (bar >= PCI_STD_NUM_BARS) {
+		dev_err(dev, "BAR %u reported by endpoint out of range [0, %u]\n",
+			bar, PCI_STD_NUM_BARS - 1);
+		return -EINVAL;
+	}
 
 	writel(data, test->bar[bar] + addr);
 
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH 1/2] misc: pci_endpoint_test: validate BAR index in doorbell test

Quoted 21 lines · On Fri, Apr 03, 2026 at 06:20:01PM -0700, [email protected] wrote:

From: Carlos Bilbao <[email protected]>

pci_endpoint_test_doorbell() reads the BAR number directly from an endpoint test register and uses it as an index into test->bar[] without bounds checking. Since the value is a raw u32 from device MMIO, any value is possible and if greater than or equal to PCI_STD_NUM_BARS the access goes out of bounds.

Add a bounds check before dereferencing test->bar[bar].

Fixes eefb83790a0d ("misc: pci_endpoint_test: Add doorbell test case")
---
 drivers/misc/pci_endpoint_test.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/misc/pci_endpoint_test.c +0 -0
index 74ab5b5b9011..276bed3f1c18 100644

drivers/misc/pci_endpoint_test.c static int pci_endpoint_test_doorbell(struct pci_endpoint_test *test) · line 1096

Quoted 8 lines
drivers/misc/pci_endpoint_test.c +5 -0
@@ -1089,6 +1089,11 @@ static int pci_endpoint_test_doorbell(struct pci_endpoint_test *test)
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_STATUS, 0);
 
 	bar = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_BAR);
+	if (bar >= PCI_STD_NUM_BARS) {
+		dev_err(dev, "BAR %u reported by endpoint out of range [0, %u]\n",
+			bar, PCI_STD_NUM_BARS - 1);
+		return -EINVAL;
+	}

Hi,

On the EP-side (pci-epf-test.c), doorbell_bar is set to NO_BAR or a valid BAR number in the range 0..(PCI_STD_NUM_BARS-1), so if we add such a safeguard here, we might as well handle NO_BAR explicitly.

That said, I'm not sure this check is really needed. At this point COMMAND_ENABLE_DOORBELL should have succeeded, so bar is expected to always be within the valid range. Whether such defensive checks are desirable probably depends on the maintainers' preference. In any case, I don't think a Fixes tag is appropriate here.

Also, even if the intention is also to guard against an "all 1's" value scenario on errors, then I think it might be better to handle that in a more centralized way, since the same concern would apply to other similar reads as well.

Best regards,
Koichiro

Quoted 7 lines
drivers/misc/pci_endpoint_test.c +0 -0
 	writel(data, test->bar[bar] + addr);
 
Signature · 1 line
2.43.0
#3 lore raw

Re: [PATCH 1/2] misc: pci_endpoint_test: validate BAR index in doorbell test

Hello,

Quoted 42 lines · Koichiro Den
Quoted 21 lines · On Fri, Apr 03, 2026 at 06:20:01PM -0700, [email protected] wrote:

From: Carlos Bilbao <[email protected]>

pci_endpoint_test_doorbell() reads the BAR number directly from an endpoint test register and uses it as an index into test->bar[] without bounds checking. Since the value is a raw u32 from device MMIO, any value is possible and if greater than or equal to PCI_STD_NUM_BARS the access goes out of bounds.

Add a bounds check before dereferencing test->bar[bar].

Fixes eefb83790a0d ("misc: pci_endpoint_test: Add doorbell test case")
---
  drivers/misc/pci_endpoint_test.c | 5 +++++
  1 file changed, 5 insertions(+)
drivers/misc/pci_endpoint_test.c +0 -0
index 74ab5b5b9011..276bed3f1c18 100644

drivers/misc/pci_endpoint_test.c static int pci_endpoint_test_doorbell(struct pci_endpoint_test *test) · line 1096

Quoted 8 lines
drivers/misc/pci_endpoint_test.c +5 -0
@@ -1089,6 +1089,11 @@ static int pci_endpoint_test_doorbell(struct pci_endpoint_test *test)
  	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_STATUS, 0);
 
  	bar = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_BAR);
+	if (bar >= PCI_STD_NUM_BARS) {
+		dev_err(dev, "BAR %u reported by endpoint out of range [0, %u]\n",
+			bar, PCI_STD_NUM_BARS - 1);
+		return -EINVAL;
+	}

Hi,

On the EP-side (pci-epf-test.c), doorbell_bar is set to NO_BAR or a valid BAR number in the range 0..(PCI_STD_NUM_BARS-1), so if we add such a safeguard here, we might as well handle NO_BAR explicitly.

That said, I'm not sure this check is really needed. At this point COMMAND_ENABLE_DOORBELL should have succeeded, so bar is expected to always be within the valid range. Whether such defensive checks are desirable probably depends on the maintainers' preference. In any case, I don't think a Fixes tag is appropriate here.

Quoted 3 lines

Also, even if the intention is also to guard against an "all 1's" value scenario on errors, then I think it might be better to handle that in a more centralized way, since the same concern would apply to other similar reads as well.

Thanks for the feedback. I'll send a v2 with changes here to handle bar < BAR_0 as well as dropping the Fixes tag.

Quoted 11 lines

Best regards,
Koichiro

Quoted 7 lines
drivers/misc/pci_endpoint_test.c +0 -0
  	writel(data, test->bar[bar] + addr);
 
Signature · 1 line
2.43.0

Thanks,

Carlos

#4 lore raw

PATCH 2/2 misc: pci_endpoint_test: remove dead BAR read before doorbell trigger

In reply to [email protected]

From: Carlos Bilbao <[email protected]>

The assignment before the writel sequence is dead code (bar is unconditionally overwritten by the re-read immediately after) so remove the call entirely.

Note that the DB_BAR register is a plain value written by the endpoint firmware; reading it carries no side effect.

---
 drivers/misc/pci_endpoint_test.c | 1 -
 1 file changed, 1 deletion(-)
drivers/misc/pci_endpoint_test.c +0 -1
index 276bed3f1c18..07c5f19d5c5a 100644
@@ -1081,7 +1081,6 @@ static int pci_endpoint_test_doorbell(struct pci_endpoint_test *test)
 
 	data = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_DATA);
 	addr = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_OFFSET);
-	bar = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_BAR);
 
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_TYPE, irq_type);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_NUMBER, 1);
Signature · 1 line
2.43.0
Reviewed-by on v1 2/2
#5 lore raw

Re: [PATCH 2/2] misc: pci_endpoint_test: remove dead BAR read before doorbell trigger

Quoted 10 lines · On Fri, Apr 03, 2026 at 06:20:02PM -0700, [email protected] wrote:

From: Carlos Bilbao <[email protected]>

The assignment before the writel sequence is dead code (bar is unconditionally overwritten by the re-read immediately after) so remove the call entirely.

Note that the DB_BAR register is a plain value written by the endpoint firmware; reading it carries no side effect.

Quoted 20 lines
---
 drivers/misc/pci_endpoint_test.c | 1 -
 1 file changed, 1 deletion(-)
drivers/misc/pci_endpoint_test.c +0 -1
index 276bed3f1c18..07c5f19d5c5a 100644
@@ -1081,7 +1081,6 @@ static int pci_endpoint_test_doorbell(struct pci_endpoint_test *test)
 
 	data = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_DATA);
 	addr = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_OFFSET);
-	bar = pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_DB_BAR);
 
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_TYPE, irq_type);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_NUMBER, 1);
Signature · 1 line
2.43.0