Skip to contentSkip to conversation

soc: qcom: cmd-db: Map shared memory as WC, not WB

PATCH Applied Maulik Shah 4 participants 8 messages Jul 18, 2024 – Jul 29, 2024

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 soc: qcom: cmd-db: Map shared memory as WC, not WB

From: Volodymyr Babchuk <[email protected]>

Linux does not write into cmd-db region. This region of memory is write protected by XPU. XPU may sometime falsely detect clean cache eviction as "write" into the write protected region leading to secure interrupt which causes an endless loop somewhere in Trust Zone.

The only reason it is working right now is because Qualcomm Hypervisor maps the same region as Non-Cacheable memory in Stage 2 translation tables. The issue manifests if we want to use another hypervisor (like Xen or KVM), which does not know anything about those specific mappings.

Changing the mapping of cmd-db memory from MEMREMAP_WB to MEMREMAP_WT/WC removes dependency on correct mappings in Stage 2 tables. This patch fixes the issue by updating the mapping to MEMREMAP_WC.

I tested this on SA8155P with Xen.

Fixes 312416d9171a ("drivers: qcom: add command DB driver")

---
Changes in v2:
- Use MEMREMAP_WC instead of MEMREMAP_WT
- Update commit message from comments in v1
- Add Fixes tag and Cc to stable
- Link to v1: https://lore.kernel.org/lkml/[email protected]

---
 drivers/soc/qcom/cmd-db.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/soc/qcom/cmd-db.c +1 -1
index d84572662017..ae66c2623d25 100644
@@ -349,7 +349,7 @@ static int cmd_db_dev_probe(struct platform_device *pdev)
 		return -EINVAL;
 	}
 
-	cmd_db_header = memremap(rmem->base, rmem->size, MEMREMAP_WB);
+	cmd_db_header = memremap(rmem->base, rmem->size, MEMREMAP_WC);
 	if (!cmd_db_header) {
 		ret = -ENOMEM;
 		cmd_db_header = NULL;

---
base-commit: 797012914d2d031430268fe512af0ccd7d8e46ef
change-id: 20240718-cmd_db_uncached-e896da5c5296

Best regards,

Signature · 1 line
Maulik Shah <[email protected]>
Tested-by on v2 1/1
#1 lore raw
Quoted 39 lines · Maulik Shah

From: Volodymyr Babchuk <[email protected]>

Linux does not write into cmd-db region. This region of memory is write protected by XPU. XPU may sometime falsely detect clean cache eviction as "write" into the write protected region leading to secure interrupt which causes an endless loop somewhere in Trust Zone.

The only reason it is working right now is because Qualcomm Hypervisor maps the same region as Non-Cacheable memory in Stage 2 translation tables. The issue manifests if we want to use another hypervisor (like Xen or KVM), which does not know anything about those specific mappings.

Changing the mapping of cmd-db memory from MEMREMAP_WB to MEMREMAP_WT/WC removes dependency on correct mappings in Stage 2 tables. This patch fixes the issue by updating the mapping to MEMREMAP_WC.

I tested this on SA8155P with Xen.

Fixes 312416d9171a ("drivers: qcom: add command DB driver")

---
Changes in v2:
- Use MEMREMAP_WC instead of MEMREMAP_WT
- Update commit message from comments in v1
- Add Fixes tag and Cc to stable
- Link to v1: https://lore.kernel.org/lkml/[email protected]

---
 drivers/soc/qcom/cmd-db.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/soc/qcom/cmd-db.c +0 -0
index d84572662017..ae66c2623d25 100644

drivers/soc/qcom/cmd-db.c static int cmd_db_dev_probe(struct platform_device *pdev) · line 355

Quoted 8 lines
drivers/soc/qcom/cmd-db.c +1 -1
@@ -349,7 +349,7 @@ static int cmd_db_dev_probe(struct platform_device *pdev)
 		return -EINVAL;
 	}
 
-	cmd_db_header = memremap(rmem->base, rmem->size, MEMREMAP_WB);
+	cmd_db_header = memremap(rmem->base, rmem->size, MEMREMAP_WC);
 	if (!cmd_db_header) {
 		ret = -ENOMEM;
 		cmd_db_header = NULL;

Thanks Maulik for sharing the patch. It works as expected. Feel free to use

Thanks,
Pavan

#2 lore raw

Hi Pavan,

Quoted 3 lines

Thanks Maulik for sharing the patch. It works as expected. Feel free to use

Can I ask how you're testing this?

Kind regards,

Quoted 5 lines

Thanks,
Pavan

Signature · 1 line
// Caleb (they/them)
#3 lore raw
Quoted 13 lines · Caleb Connolly

Hi Pavan,

Quoted 3 lines

Thanks Maulik for sharing the patch. It works as expected. Feel free to use

Can I ask how you're testing this?

Kind regards,

Quoted 3 lines

The QCM6490 RB3 board boots from upstream kernel. As part of releases here [1] we plan to support booting Linux in EL2. Currently, I have an internal board/build with firmware allowing this already. I have boot tested Maulik's patch (and as well v1 patch). The targets gets reset early in the boot up w/o this patch and I could boot w/ this patch.

Thanks,
Pavan

[1] https://github.com/quic-yocto/qcom-manifest

#4 lore raw
Quoted 15 lines · Pavan Kondeti
Quoted 13 lines · Caleb Connolly

Hi Pavan,

Quoted 3 lines

Thanks Maulik for sharing the patch. It works as expected. Feel free to use

Can I ask how you're testing this?

Kind regards,

Quoted 3 lines
Quoted 5 lines

The QCM6490 RB3 board boots from upstream kernel. As part of releases here [1] we plan to support booting Linux in EL2. Currently, I have an internal board/build with firmware allowing this already. I have boot tested Maulik's patch (and as well v1 patch). The targets gets reset early in the boot up w/o this patch and I could boot w/ this patch.

Ahh awesome! Thanks for the info :)

Quoted 5 lines

Thanks,
Pavan

[1] https://github.com/quic-yocto/qcom-manifest

Signature · 1 line
// Caleb (they/them)
Reviewed-by on v2 1/1
#5 lore raw
In reply to Maulik Shah
Quoted 16 lines · Maulik Shah

From: Volodymyr Babchuk <[email protected]>

Linux does not write into cmd-db region. This region of memory is write protected by XPU. XPU may sometime falsely detect clean cache eviction as "write" into the write protected region leading to secure interrupt which causes an endless loop somewhere in Trust Zone.

The only reason it is working right now is because Qualcomm Hypervisor maps the same region as Non-Cacheable memory in Stage 2 translation tables. The issue manifests if we want to use another hypervisor (like Xen or KVM), which does not know anything about those specific mappings.

Changing the mapping of cmd-db memory from MEMREMAP_WB to MEMREMAP_WT/WC removes dependency on correct mappings in Stage 2 tables. This patch fixes the issue by updating the mapping to MEMREMAP_WC.

Quoted 7 lines

I tested this on SA8155P with Xen.

Fixes 312416d9171a ("drivers: qcom: add command DB driver")
Quoted 29 lines

---
Changes in v2:
- Use MEMREMAP_WC instead of MEMREMAP_WT
- Update commit message from comments in v1
- Add Fixes tag and Cc to stable
- Link to v1: https://lore.kernel.org/lkml/[email protected]

---
  drivers/soc/qcom/cmd-db.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/soc/qcom/cmd-db.c +1 -2
index d84572662017..ae66c2623d25 100644
@@ -349,7 +349,7 @@ static int cmd_db_dev_probe(struct platform_device *pdev)
  		return -EINVAL;
  	}
 
-	cmd_db_header = memremap(rmem->base, rmem->size, MEMREMAP_WB);
+	cmd_db_header = memremap(rmem->base, rmem->size, MEMREMAP_WC);
  	if (!cmd_db_header) {
  		ret = -ENOMEM;
  		cmd_db_header = NULL;
 
---

base-commit: 797012914d2d031430268fe512af0ccd7d8e46ef
change-id: 20240718-cmd_db_uncached-e896da5c5296

Best regards,

Signature · 1 line
// Caleb (they/them)
#6 lore raw

Hi Bjorn,

Quoted 21 lines · Caleb Connolly
Quoted 18 lines · Maulik Shah

From: Volodymyr Babchuk <[email protected]>

Linux does not write into cmd-db region. This region of memory is write protected by XPU. XPU may sometime falsely detect clean cache eviction as "write" into the write protected region leading to secure interrupt which causes an endless loop somewhere in Trust Zone.

The only reason it is working right now is because Qualcomm Hypervisor maps the same region as Non-Cacheable memory in Stage 2 translation tables. The issue manifests if we want to use another hypervisor (like Xen or KVM), which does not know anything about those specific mappings.

Changing the mapping of cmd-db memory from MEMREMAP_WB to MEMREMAP_WT/WC removes dependency on correct mappings in Stage 2 tables. This patch fixes the issue by updating the mapping to MEMREMAP_WC.

I tested this on SA8155P with Xen.

Quoted 7 lines
Quoted 5 lines
Fixes 312416d9171a ("drivers: qcom: add command DB driver")

Is it possible to include it in v6.11-rc?

Thanks,
Pavan

maintainer Applied
#7 lore raw
In reply to Maulik Shah
Quoted 11 lines · Maulik Shah

Linux does not write into cmd-db region. This region of memory is write protected by XPU. XPU may sometime falsely detect clean cache eviction as "write" into the write protected region leading to secure interrupt which causes an endless loop somewhere in Trust Zone.

The only reason it is working right now is because Qualcomm Hypervisor maps the same region as Non-Cacheable memory in Stage 2 translation tables. The issue manifests if we want to use another hypervisor (like Xen or KVM), which does not know anything about those specific mappings.

[...]

Applied, thanks!

[1/1] soc: qcom: cmd-db: Map shared memory as WC, not WB

      commit: f9bb896eab221618927ae6a2f1d566567999839d

Best regards,

Signature · 1 line
Bjorn Andersson <[email protected]>