Skip to contentSkip to conversation

tpm, tpm_crb: fix unaligned read of the command buffer address

PATCH In review Jarkko Sakkinen 3 participants 5 messages Sep 15, 2015 – Oct 9, 2015 TPM device driver

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 tpm, tpm_crb: fix unaligned read of the command buffer address

The command buffer address must be read with exactly two 32-bit reads. Otherwise, on some HW platforms, it seems that HW will abort the read operation, which causes CPU to fill the read bytes with 1's. Therefore, we cannot rely on memcpy_fromio() but must call ioread32() two times instead.

Also, this matches the PC Client Platform TPM Profile specification, which defines command buffer address with two 32-bit fields.

---
 drivers/char/tpm/tpm_crb.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/char/tpm/tpm_crb.c +4 -3
index b4564b6..c09b370 100644
@@ -68,7 +68,8 @@ struct crb_control_area {
 	u32 int_enable;
 	u32 int_sts;
 	u32 cmd_size;
-	u64 cmd_pa;
+	u32 cmd_pa_low;
+	u32 cmd_pa_high;
 	u32 rsp_size;
 	u64 rsp_pa;
 } __packed;
@@ -263,8 +264,8 @@ static int crb_acpi_add(struct acpi_device *device)
 		return -ENOMEM;
 	}
 
-	memcpy_fromio(&pa, &priv->cca->cmd_pa, 8);
-	pa = le64_to_cpu(pa);
+	pa = ((u64) le32_to_cpu(ioread32(&priv->cca->cmd_pa_high)) << 32) +
+		(u64) le32_to_cpu(ioread32(&priv->cca->cmd_pa_low));
 	priv->cmd = devm_ioremap_nocache(dev, pa,
 					 ioread32(&priv->cca->cmd_size));
 	if (!priv->cmd) {
Signature · 1 line
2.5.0
Reviewed-by on v3 1/1
#1 lore raw
Quoted 17 lines · Jarkko Sakkinen

The command buffer address must be read with exactly two 32-bit reads. Otherwise, on some HW platforms, it seems that HW will abort the read operation, which causes CPU to fill the read bytes with 1's. Therefore, we cannot rely on memcpy_fromio() but must call ioread32() two times instead.

Also, this matches the PC Client Platform TPM Profile specification, which defines command buffer address with two 32-bit fields.

 drivers/char/tpm/tpm_crb.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/char/tpm/tpm_crb.c +1 -0
index b4564b6..c09b370 100644
+++ b/drivers/char/tpm/tpm_crb.c

drivers/char/tpm/tpm_crb.c struct crb_control_area { · line 74

Quoted 8 lines
drivers/char/tpm/tpm_crb.c +2 -1
@@ -68,7 +68,8 @@ struct crb_control_area {
 	u32 int_enable;
 	u32 int_sts;
 	u32 cmd_size;
-	u64 cmd_pa;
+	u32 cmd_pa_low;
+	u32 cmd_pa_high;
 	u32 rsp_size;
 	u64 rsp_pa;

What about rsp_pa?

drivers/char/tpm/tpm_crb.c struct crb_control_area { · line 75

Quoted 1 line
drivers/char/tpm/tpm_crb.c +0 -0
 } __packed;

As I mentioned, drop the __packed.

Otherwise:

Jason

#2 lore raw
Quoted 14 lines · Jason Gunthorpe
Quoted 13 lines · Jarkko Sakkinen

The command buffer address must be read with exactly two 32-bit reads. Otherwise, on some HW platforms, it seems that HW will abort the read operation, which causes CPU to fill the read bytes with 1's. Therefore, we cannot rely on memcpy_fromio() but must call ioread32() two times instead.

Also, this matches the PC Client Platform TPM Profile specification, which defines command buffer address with two 32-bit fields.

 drivers/char/tpm/tpm_crb.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
Quoted 14 lines

drivers/char/tpm/tpm_crb.c struct crb_control_area { · line 74

Quoted 12 lines
drivers/char/tpm/tpm_crb.c +3 -1
index b4564b6..c09b370 100644
+++ b/drivers/char/tpm/tpm_crb.c
@@ -68,7 +68,8 @@ struct crb_control_area {
 	u32 int_enable;
 	u32 int_sts;
 	u32 cmd_size;
-	u64 cmd_pa;
+	u32 cmd_pa_low;
+	u32 cmd_pa_high;
 	u32 rsp_size;
 	u64 rsp_pa;

What about rsp_pa?

It's naturally aligned and defined as 64-bit field in:

http://www.trustedcomputinggroup.org/developers/pc_client/specifications

Also in a platform (not available in public yet) it gets read correctly.

Quoted 3 lines

drivers/char/tpm/tpm_crb.c struct crb_control_area { · line 75

Quoted 1 line
drivers/char/tpm/tpm_crb.c +0 -0
 } __packed;

As I mentioned, drop the __packed.

I didn't do that because it is not mandatory for the bug fix.

I would rather submit it as a separate commit later on.

Quoted 5 lines

Otherwise:

Jason

/Jarkko

#3 lore raw
Quoted 30 lines · Jarkko Sakkinen
Quoted 14 lines · Jason Gunthorpe
Quoted 13 lines · Jarkko Sakkinen

The command buffer address must be read with exactly two 32-bit reads. Otherwise, on some HW platforms, it seems that HW will abort the read operation, which causes CPU to fill the read bytes with 1's. Therefore, we cannot rely on memcpy_fromio() but must call ioread32() two times instead.

Also, this matches the PC Client Platform TPM Profile specification, which defines command buffer address with two 32-bit fields.

 drivers/char/tpm/tpm_crb.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
Quoted 14 lines

drivers/char/tpm/tpm_crb.c struct crb_control_area { · line 74

Quoted 12 lines
drivers/char/tpm/tpm_crb.c +3 -1
index b4564b6..c09b370 100644
+++ b/drivers/char/tpm/tpm_crb.c
@@ -68,7 +68,8 @@ struct crb_control_area {
 	u32 int_enable;
 	u32 int_sts;
 	u32 cmd_size;
-	u64 cmd_pa;
+	u32 cmd_pa_low;
+	u32 cmd_pa_high;
 	u32 rsp_size;
 	u64 rsp_pa;

What about rsp_pa?

Quoted 5 lines

It's naturally aligned and defined as 64-bit field in:

http://www.trustedcomputinggroup.org/developers/pc_client/specifications

Also in a platform (not available in public yet) it gets read correctly.

memcpy_64.S uses rep movsq and rep movsb for the remainder. That's why the read failed in the case of cmd address but it does not fail with rsp address.

See http://lxr.free-electrons.com/source/arch/x86/lib/memcpy_64.S

Quoted 8 lines
Quoted 3 lines

drivers/char/tpm/tpm_crb.c struct crb_control_area { · line 75

Quoted 1 line
drivers/char/tpm/tpm_crb.c +0 -0
 } __packed;

As I mentioned, drop the __packed.

I didn't do that because it is not mandatory for the bug fix.

I would rather submit it as a separate commit later on.

Quoted 7 lines
Quoted 5 lines

Otherwise:

Jason

/Jarkko

/Jarkko

#4 lore raw
In reply to Jarkko Sakkinen

Hi

This would desperately neeed Tested-by's (with Haswell PTT).

/Jarkko

Quoted 42 lines · Jarkko Sakkinen

The command buffer address must be read with exactly two 32-bit reads. Otherwise, on some HW platforms, it seems that HW will abort the read operation, which causes CPU to fill the read bytes with 1's. Therefore, we cannot rely on memcpy_fromio() but must call ioread32() two times instead.

Also, this matches the PC Client Platform TPM Profile specification, which defines command buffer address with two 32-bit fields.

---
 drivers/char/tpm/tpm_crb.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/char/tpm/tpm_crb.c +4 -3
index b4564b6..c09b370 100644
@@ -68,7 +68,8 @@ struct crb_control_area {
 	u32 int_enable;
 	u32 int_sts;
 	u32 cmd_size;
-	u64 cmd_pa;
+	u32 cmd_pa_low;
+	u32 cmd_pa_high;
 	u32 rsp_size;
 	u64 rsp_pa;
 } __packed;
@@ -263,8 +264,8 @@ static int crb_acpi_add(struct acpi_device *device)
 		return -ENOMEM;
 	}
 
-	memcpy_fromio(&pa, &priv->cca->cmd_pa, 8);
-	pa = le64_to_cpu(pa);
+	pa = ((u64) le32_to_cpu(ioread32(&priv->cca->cmd_pa_high)) << 32) +
+		(u64) le32_to_cpu(ioread32(&priv->cca->cmd_pa_low));
 	priv->cmd = devm_ioremap_nocache(dev, pa,
 					 ioread32(&priv->cca->cmd_size));
 	if (!priv->cmd) {
Signature · 1 line
2.5.0