Skip to contentSkip to conversation

nvme: handle persistent internal error AER from NVMe controller

PATCH Applied Michael Kelley 3 participants 6 messages Jun 8, 2022 – Jun 13, 2022 NVMe

lore
Share & tools
lore
#0 lore raw

PATCH v4 1/1 nvme: handle persistent internal error AER from NVMe controller

In the NVM Express Revision 1.4 spec, Figure 145 describes possible values for an AER with event type "Error" (value 000b). For a Persistent Internal Error (value 03h), the host should perform a controller reset.

Add support for this error using code that already exists for doing a controller reset. As part of this support, introduce two utility functions for parsing the AER type and subtype.

This new support was tested in a lab environment where we can generate the persistent internal error on demand, and observe both the Linux side and NVMe controller side to see that the controller reset has been done.

---
Changes since v3:
* Removed call to nvme_should_reset() and dropped the original patch 1/2 that moved nvme_should_reset() from pci.c to core.c [Christoph Hellwig]

Changes since v2:
* Instead of reading CSTS, use a constant value as input to nvme_should_reset() [Keith Busch]
* Introduce helper functions for parsing the AER result fields [Keith Busch]

 drivers/nvme/host/core.c | 31 +++++++++++++++++++++++++++++--
 include/linux/nvme.h     |  4 ++++
 2 files changed, 33 insertions(+), 2 deletions(-)
drivers/nvme/host/core.c +29 -2
index 72f7c95..bb8c91e 100644
@@ -4502,9 +4502,19 @@ static void nvme_fw_act_work(struct work_struct *work)
 	nvme_get_fw_slot_info(ctrl);
 }
 
+static u32 nvme_aer_type(u32 result)
+{
+	return result & 0x7;
+}
+
+static u32 nvme_aer_subtype(u32 result)
+{
+	return (result & 0xff00) >> 8;
+}
+
 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
 {
-	u32 aer_notice_type = (result & 0xff00) >> 8;
+	u32 aer_notice_type = nvme_aer_subtype(result);
 
 	trace_nvme_async_event(ctrl, aer_notice_type);
 
@@ -4537,11 +4547,19 @@ static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
 	}
 }
 
+static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
+{
+	trace_nvme_async_event(ctrl, NVME_AER_ERROR);
+	dev_warn(ctrl->device, "resetting controller due to AER\n");
+	nvme_reset_ctrl(ctrl);
+}
+
 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
 		volatile union nvme_result *res)
 {
 	u32 result = le32_to_cpu(res->u32);
-	u32 aer_type = result & 0x07;
+	u32 aer_type = nvme_aer_type(result);
+	u32 aer_subtype = nvme_aer_subtype(result);
 
 	if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
 		return;
@@ -4551,6 +4569,15 @@ void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
 		nvme_handle_aen_notice(ctrl, result);
 		break;
 	case NVME_AER_ERROR:
+		/*
+		 * For a persistent internal error, don't run async_event_work
+		 * to submit a new AER. The controller reset will do it.
+		 */
+		if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
+			nvme_handle_aer_persistent_error(ctrl);
+			return;
+		}
+		fallthrough;
 	case NVME_AER_SMART:
 	case NVME_AER_CSS:
 	case NVME_AER_VS:
include/linux/nvme.h +4 -0
index 29ec3e3..8ced243 100644
@@ -712,6 +712,10 @@ enum {
 };
 
 enum {
+	NVME_AER_ERROR_PERSIST_INT_ERR	= 0x03,
+};
+
+enum {
 	NVME_AER_NOTICE_NS_CHANGED	= 0x00,
 	NVME_AER_NOTICE_FW_ACT_STARTING = 0x01,
 	NVME_AER_NOTICE_ANA		= 0x03,
Signature · 1 line
1.8.3.1
Reviewed-by on v4 1/1
#1 lore raw
Quoted 9 lines · Michael Kelley

In the NVM Express Revision 1.4 spec, Figure 145 describes possible values for an AER with event type "Error" (value 000b). For a Persistent Internal Error (value 03h), the host should perform a controller reset.

Add support for this error using code that already exists for doing a controller reset. As part of this support, introduce two utility functions for parsing the AER type and subtype.

Quoted 7 lines

This new support was tested in a lab environment where we can generate the persistent internal error on demand, and observe both the Linux side and NVMe controller side to see that the controller reset has been done.

---

Looks good. Thanks a lot for testing this, perhaps consider writing the testcase for it in the blktests under nvme category, that way it will get tested by everyone else.

-ck

#2 lore raw
Quoted 10 lines · Chaitanya Kulkarni
Quoted 9 lines · Michael Kelley

In the NVM Express Revision 1.4 spec, Figure 145 describes possible values for an AER with event type "Error" (value 000b). For a Persistent Internal Error (value 03h), the host should perform a controller reset.

Add support for this error using code that already exists for doing a controller reset. As part of this support, introduce two utility functions for parsing the AER type and subtype.

Quoted 6 lines
Quoted 6 lines

This new support was tested in a lab environment where we can generate the persistent internal error on demand, and observe both the Linux side and NVMe controller side to see that the controller reset has been done.

Can you please clarify that which transports you have tested such as RDMA, TCP, and PCIe ?

-ck

#3 lore raw

From: Chaitanya Kulkarni <[email protected]>

Quoted 19 lines · Chaitanya Kulkarni
Quoted 10 lines · Michael Kelley
Quoted 9 lines

In the NVM Express Revision 1.4 spec, Figure 145 describes possible values for an AER with event type "Error" (value 000b). For a Persistent Internal Error (value 03h), the host should perform a controller reset.

Add support for this error using code that already exists for doing a controller reset. As part of this support, introduce two utility functions for parsing the AER type and subtype.

Quoted 6 lines
Quoted 6 lines

This new support was tested in a lab environment where we can generate the persistent internal error on demand, and observe both the Linux side and NVMe controller side to see that the controller reset has been done.

Quoted 3 lines

Can you please clarify that which transports you have tested such as RDMA, TCP, and PCIe ?

I've tested PCIe only -- that's all I have access to. I can tweak the commit message to be more specific.

Michael

#4 lore raw
Quoted 20 lines · Michael Kelley (LINUX)

From: Chaitanya Kulkarni <[email protected]>

Quoted 19 lines · Chaitanya Kulkarni
Quoted 16 lines · Michael Kelley
Quoted 15 lines

In the NVM Express Revision 1.4 spec, Figure 145 describes possible values for an AER with event type "Error" (value 000b). For a Persistent Internal Error (value 03h), the host should perform a controller reset.

Add support for this error using code that already exists for doing a controller reset. As part of this support, introduce two utility functions for parsing the AER type and subtype.

This new support was tested in a lab environment where we can generate the persistent internal error on demand, and observe both the Linux side and NVMe controller side to see that the controller reset has been done.

Quoted 8 lines
Quoted 3 lines

Can you please clarify that which transports you have tested such as RDMA, TCP, and PCIe ?

I've tested PCIe only -- that's all I have access to. I can tweak the commit message to be more specific.

Michael

It's okay we have it documented now, thanks again.

-ck