Skip to contentSkip to conversation

misc: pci_endpoint_test: Handle BAR sizes larger than INT_MAX

PATCH Superseded Niklas Cassel 2 participants 5 messages Jan 23, 2025 – Jan 24, 2025

Subsystems (3)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 misc: pci_endpoint_test: Handle BAR sizes larger than INT_MAX

Running 'pcitest -b 0' fails with "TEST FAILED" when the BAR0 size is e.g. 8 GB.

The return value of the pci_resource_len() macro can be larger than that of a signed integer type. Thus, when using 'pcitest' with an 8 GB BAR, the bar_size of the integer type will overflow.

Change bar_size from integer to resource_size_t to prevent integer overflow for large BAR sizes with 32-bit compilers.

---
Hans submitted a patch for this that was reverted because apparently some gcc-7 arm32 compiler doesn't like div_u64(). In order to avoid debugging gcc-7 arm32 compiler issues, simply replace the division with addition, which arguably makes the code simpler as well.

 drivers/misc/pci_endpoint_test.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
drivers/misc/pci_endpoint_test.c +10 -8
index d5ac71a49386..8e48a15100f1 100644
@@ -272,9 +272,9 @@ static const u32 bar_test_pattern[] = {
 };
 
 static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
-					enum pci_barno barno, int offset,
-					void *write_buf, void *read_buf,
-					int size)
+					enum pci_barno barno,
+					resource_size_t offset, void *write_buf,
+					void *read_buf, int size)
 {
 	memset(write_buf, bar_test_pattern[barno], size);
 	memcpy_toio(test->bar[barno] + offset, write_buf, size);
@@ -287,10 +287,11 @@ static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
 static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 				  enum pci_barno barno)
 {
-	int j, bar_size, buf_size, iters;
+	resource_size_t bar_size, offset = 0;
 	void *write_buf __free(kfree) = NULL;
 	void *read_buf __free(kfree) = NULL;
 	struct pci_dev *pdev = test->pdev;
+	int buf_size;
 
 	if (!test->bar[barno])
 		return -ENOMEM;
@@ -314,11 +315,12 @@ static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 	if (!read_buf)
 		return -ENOMEM;
 
-	iters = bar_size / buf_size;
-	for (j = 0; j < iters; j++)
-		if (pci_endpoint_test_bar_memcmp(test, barno, buf_size * j,
-						 write_buf, read_buf, buf_size))
+	while (offset < bar_size) {
+		if (pci_endpoint_test_bar_memcmp(test, barno, offset, write_buf,
+						 read_buf, buf_size))
 			return -EIO;
+		offset += buf_size;
+	}
 
 	return 0;
 }
Signature · 1 line
2.48.1
#1 lore raw
Quoted 56 lines · Niklas Cassel

Running 'pcitest -b 0' fails with "TEST FAILED" when the BAR0 size is e.g. 8 GB.

The return value of the pci_resource_len() macro can be larger than that of a signed integer type. Thus, when using 'pcitest' with an 8 GB BAR, the bar_size of the integer type will overflow.

Change bar_size from integer to resource_size_t to prevent integer overflow for large BAR sizes with 32-bit compilers.

---
Hans submitted a patch for this that was reverted because apparently some gcc-7 arm32 compiler doesn't like div_u64(). In order to avoid debugging gcc-7 arm32 compiler issues, simply replace the division with addition, which arguably makes the code simpler as well.

 drivers/misc/pci_endpoint_test.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
drivers/misc/pci_endpoint_test.c +5 -4
index d5ac71a49386..8e48a15100f1 100644
@@ -272,9 +272,9 @@ static const u32 bar_test_pattern[] = {
 };
 
 static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
-					enum pci_barno barno, int offset,
-					void *write_buf, void *read_buf,
-					int size)
+					enum pci_barno barno,
+					resource_size_t offset, void *write_buf,
+					void *read_buf, int size)
 {
 	memset(write_buf, bar_test_pattern[barno], size);
 	memcpy_toio(test->bar[barno] + offset, write_buf, size);
@@ -287,10 +287,11 @@ static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
 static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 				  enum pci_barno barno)
 {
-	int j, bar_size, buf_size, iters;
+	resource_size_t bar_size, offset = 0;
 	void *write_buf __free(kfree) = NULL;
 	void *read_buf __free(kfree) = NULL;
 	struct pci_dev *pdev = test->pdev;
+	int buf_size;
 
 	if (!test->bar[barno])
 		return -ENOMEM;
@@ -314,11 +315,12 @@ static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 	if (!read_buf)
 		return -ENOMEM;
 

drivers/misc/pci_endpoint_test.c static int pci_endpoint_test_bar(struct pci_endpoint_test *test, · line 323

Quoted 10 lines
drivers/misc/pci_endpoint_test.c +5 -4
-	iters = bar_size / buf_size;
-	for (j = 0; j < iters; j++)
-		if (pci_endpoint_test_bar_memcmp(test, barno, buf_size * j,
-						 write_buf, read_buf, buf_size))
+	while (offset < bar_size) {
+		if (pci_endpoint_test_bar_memcmp(test, barno, offset, write_buf,
+						 read_buf, buf_size))
 			return -EIO;
+		offset += buf_size;
+	}

Actually, you change code logic although functionality is the same. I feel like you should mention at commit message or use origial code by just change variable type.

#ifdef CONFIG_PHYS_ADDR_T_64BIT
typedef u64 phys_addr_t;
#else
typedef u32 phys_addr_t;
#endif

typedef phys_addr_t resource_size_t;

resource_size_t may 32bit at some configuration. But I don't know what happen when 8G bar pci device attached to such system.

Frank

Quoted 6 lines
drivers/misc/pci_endpoint_test.c +0 -1
 	return 0;
 }
--

2.48.1

#2 lore raw
Quoted 72 lines · Frank Li
Quoted 56 lines · Niklas Cassel

Running 'pcitest -b 0' fails with "TEST FAILED" when the BAR0 size is e.g. 8 GB.

The return value of the pci_resource_len() macro can be larger than that of a signed integer type. Thus, when using 'pcitest' with an 8 GB BAR, the bar_size of the integer type will overflow.

Change bar_size from integer to resource_size_t to prevent integer overflow for large BAR sizes with 32-bit compilers.

---
Hans submitted a patch for this that was reverted because apparently some gcc-7 arm32 compiler doesn't like div_u64(). In order to avoid debugging gcc-7 arm32 compiler issues, simply replace the division with addition, which arguably makes the code simpler as well.

 drivers/misc/pci_endpoint_test.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
drivers/misc/pci_endpoint_test.c +5 -4
index d5ac71a49386..8e48a15100f1 100644
@@ -272,9 +272,9 @@ static const u32 bar_test_pattern[] = {
 };
 
 static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
-					enum pci_barno barno, int offset,
-					void *write_buf, void *read_buf,
-					int size)
+					enum pci_barno barno,
+					resource_size_t offset, void *write_buf,
+					void *read_buf, int size)
 {
 	memset(write_buf, bar_test_pattern[barno], size);
 	memcpy_toio(test->bar[barno] + offset, write_buf, size);
@@ -287,10 +287,11 @@ static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
 static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 				  enum pci_barno barno)
 {
-	int j, bar_size, buf_size, iters;
+	resource_size_t bar_size, offset = 0;
 	void *write_buf __free(kfree) = NULL;
 	void *read_buf __free(kfree) = NULL;
 	struct pci_dev *pdev = test->pdev;
+	int buf_size;
 
 	if (!test->bar[barno])
 		return -ENOMEM;
@@ -314,11 +315,12 @@ static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 	if (!read_buf)
 		return -ENOMEM;
 

drivers/misc/pci_endpoint_test.c static int pci_endpoint_test_bar(struct pci_endpoint_test *test, · line 323

Quoted 10 lines
drivers/misc/pci_endpoint_test.c +5 -4
-	iters = bar_size / buf_size;
-	for (j = 0; j < iters; j++)
-		if (pci_endpoint_test_bar_memcmp(test, barno, buf_size * j,
-						 write_buf, read_buf, buf_size))
+	while (offset < bar_size) {
+		if (pci_endpoint_test_bar_memcmp(test, barno, offset, write_buf,
+						 read_buf, buf_size))
 			return -EIO;
+		offset += buf_size;
+	}

Actually, you change code logic although functionality is the same. I feel like you should mention at commit message or use origial code by just change variable type.

Quoted 5 lines

#ifdef CONFIG_PHYS_ADDR_T_64BIT
typedef u64 phys_addr_t;
#else
typedef u32 phys_addr_t;
#endif

Hello Frank,

I personally think that is a horrible idea :)

We do not want to introduce ifdefs in the middle of the code, unless in exceptional circumstances, like architecture specific optimized code.

Kind regards,
Niklas

#3 lore raw
Quoted 79 lines · Niklas Cassel
Quoted 72 lines · Frank Li
Quoted 56 lines · Niklas Cassel

Running 'pcitest -b 0' fails with "TEST FAILED" when the BAR0 size is e.g. 8 GB.

The return value of the pci_resource_len() macro can be larger than that of a signed integer type. Thus, when using 'pcitest' with an 8 GB BAR, the bar_size of the integer type will overflow.

Change bar_size from integer to resource_size_t to prevent integer overflow for large BAR sizes with 32-bit compilers.

---
Hans submitted a patch for this that was reverted because apparently some gcc-7 arm32 compiler doesn't like div_u64(). In order to avoid debugging gcc-7 arm32 compiler issues, simply replace the division with addition, which arguably makes the code simpler as well.

 drivers/misc/pci_endpoint_test.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
drivers/misc/pci_endpoint_test.c +5 -4
index d5ac71a49386..8e48a15100f1 100644
@@ -272,9 +272,9 @@ static const u32 bar_test_pattern[] = {
 };
 
 static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
-					enum pci_barno barno, int offset,
-					void *write_buf, void *read_buf,
-					int size)
+					enum pci_barno barno,
+					resource_size_t offset, void *write_buf,
+					void *read_buf, int size)
 {
 	memset(write_buf, bar_test_pattern[barno], size);
 	memcpy_toio(test->bar[barno] + offset, write_buf, size);
@@ -287,10 +287,11 @@ static int pci_endpoint_test_bar_memcmp(struct pci_endpoint_test *test,
 static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 				  enum pci_barno barno)
 {
-	int j, bar_size, buf_size, iters;
+	resource_size_t bar_size, offset = 0;
 	void *write_buf __free(kfree) = NULL;
 	void *read_buf __free(kfree) = NULL;
 	struct pci_dev *pdev = test->pdev;
+	int buf_size;
 
 	if (!test->bar[barno])
 		return -ENOMEM;
@@ -314,11 +315,12 @@ static int pci_endpoint_test_bar(struct pci_endpoint_test *test,
 	if (!read_buf)
 		return -ENOMEM;
 

drivers/misc/pci_endpoint_test.c static int pci_endpoint_test_bar(struct pci_endpoint_test *test, · line 323

Quoted 10 lines
drivers/misc/pci_endpoint_test.c +5 -4
-	iters = bar_size / buf_size;
-	for (j = 0; j < iters; j++)
-		if (pci_endpoint_test_bar_memcmp(test, barno, buf_size * j,
-						 write_buf, read_buf, buf_size))
+	while (offset < bar_size) {
+		if (pci_endpoint_test_bar_memcmp(test, barno, offset, write_buf,
+						 read_buf, buf_size))
 			return -EIO;
+		offset += buf_size;
+	}

Actually, you change code logic although functionality is the same. I feel like you should mention at commit message or use origial code by just change variable type.

Quoted 5 lines

#ifdef CONFIG_PHYS_ADDR_T_64BIT
typedef u64 phys_addr_t;
#else
typedef u32 phys_addr_t;
#endif

Quoted 6 lines

Hello Frank,

I personally think that is a horrible idea :)

We do not want to introduce ifdefs in the middle of the code, unless in exceptional circumstances, like architecture specific optimized code.

You miss understand what my means. I copy it from type.h to indicate resource_size_t is not 64bit at all platforms.

Frank

Quoted 4 lines

Kind regards,
Niklas

#4 lore raw
Quoted 14 lines · Frank Li
Quoted 13 lines · Niklas Cassel
Quoted 10 lines

Actually, you change code logic although functionality is the same. I feel like you should mention at commit message or use origial code by just change variable type.

#ifdef CONFIG_PHYS_ADDR_T_64BIT
typedef u64 phys_addr_t;
#else
typedef u32 phys_addr_t;
#endif

Hello Frank,

Quoted 7 lines
Quoted 4 lines

I personally think that is a horrible idea :)

We do not want to introduce ifdefs in the middle of the code, unless in exceptional circumstances, like architecture specific optimized code.

You miss understand what my means. I copy it from type.h to indicate resource_size_t is not 64bit at all platforms.

I know that resource_size_t is typedefed to phys_addr_t, which can be 32-bit or 64-bit. (I compile tested this patch on 32-bit both with and without PAE.)

resource_size_t is the type returned by pci_resource_len(). That is why the patch in subject changes the type to use resource_size_t. IMO, it does not make sense to use any other type (e.g. u64), since the value returned by pci_resource_len() will still be limited to what can be represented by resource_size_t.

A BARs larger than 4GB, on systems with 32-bit resource_size_t, will get disabled by PCI core:
https://github.com/torvalds/linux/blob/v6.13/drivers/pci/probe.c#L265-L270

So all good.

As for your question why I don't keep the division, please read the comment section in this patch (where the changelog usually is), or read the thread:
https://lore.kernel.org/linux-pci/[email protected]/T/#t

I guess I could have added:
"
In order to handle 64-bit resource_type_t on 32-bit platforms, we would have needed to use a function like div_u64() or similar. Instead, change the code to use addition instead of division. This avoids the need for div_u64() or similar, while also simplifying the code.
"

Let me send a V2 with that senctence added to address your review comment.

Kind regards,
Niklas