Skip to contentSkip to conversation

mpfs: fix handling failed service requests

PATCH Applied Conor Dooley 3 participants 6 messages Nov 23, 2022 – Dec 26, 2022

Subsystems (4)
lore
Share & tools
lore

Conor Dooley

author maintainer
#0 lore raw

v3 cover mpfs: fix handling failed service requests

From: Conor Dooley <[email protected]>

As things stand, if a service request fails, we never know about it as the status register is not read. This is quite obviously far from ideal, as we would be reading garbage from the mailbox if a service was to fail.

This pair of patches adds handling of the service status to both sides of the mailbox implementation.

@Jassi, there should be no build, or really functional, dependency here between patches. In the failing case, the buffer is allocated by the service driver, so things will "work" just as badly as they did before if only one of the patches is present. As a result, should be safe to do the mailbox patch via your tree & the client via the soc tree.

Thanks,
Conor.

Changes since v2:
- Stop redefining the already existing #defines for status register.

Changes since v1:
- Send the patches I meant to send, with an extraneous ; removed.

Conor Dooley (2):
soc: microchip: mpfs: handle failed system service requests
mailbox: mpfs: read the system controller's status

 drivers/mailbox/mailbox-mpfs.c              | 31 +++++++++++++++++++--
 drivers/soc/microchip/mpfs-sys-controller.c |  6 ++++
 2 files changed, 34 insertions(+), 3 deletions(-)
Signature · 7 lines
2.38.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
author maintainer
#3 lore raw

PATCH v3 2/2 mailbox: mpfs: read the system controller's status

From: Conor Dooley <[email protected]>

Some services explicitly return an error code in their response, but others rely on the system controller to set a status in its status register. The meaning of the bits varies based on what service is requested, so pass it back up to the driver that requested the service in the first place. The field in the message struct already existed, but was unused until now.

If the system controller is busy, in which case we should never actually be in the interrupt handler, or if the service fails the mailbox itself should not be read. Callers should check the status before operating on the response.

There's an existing, but unused, #define for the mailbox mask - but it was incorrect. It was doing a GENMASK_ULL(32, 16) which should've just been a GENMASK(31, 16), so fix that up and start using it.

Fixes 83d7b1560810 ("mbox: add polarfire soc system controller mailbox")
---
 drivers/mailbox/mailbox-mpfs.c | 31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)
drivers/mailbox/mailbox-mpfs.c +28 -3
index cfacb3f320a6..853901acaeec 100644
@@ -2,7 +2,7 @@
 /*
  * Microchip PolarFire SoC (MPFS) system controller/mailbox controller driver
  *
- * Copyright (c) 2020 Microchip Corporation. All rights reserved.
+ * Copyright (c) 2020-2022 Microchip Corporation. All rights reserved.
  *
  * Author: Conor Dooley <[email protected]>
  *
@@ -56,7 +56,7 @@
 #define SCB_STATUS_NOTIFY_MASK BIT(SCB_STATUS_NOTIFY)
 
 #define SCB_STATUS_POS (16)
-#define SCB_STATUS_MASK GENMASK_ULL(SCB_STATUS_POS + SCB_MASK_WIDTH, SCB_STATUS_POS)
+#define SCB_STATUS_MASK GENMASK(SCB_STATUS_POS + SCB_MASK_WIDTH - 1, SCB_STATUS_POS)
 
 struct mpfs_mbox {
 	struct mbox_controller controller;
@@ -130,13 +130,38 @@ static void mpfs_mbox_rx_data(struct mbox_chan *chan)
 	struct mpfs_mbox *mbox = (struct mpfs_mbox *)chan->con_priv;
 	struct mpfs_mss_response *response = mbox->response;
 	u16 num_words = ALIGN((response->resp_size), (4)) / 4U;
-	u32 i;
+	u32 i, status;
 
 	if (!response->resp_msg) {
 		dev_err(mbox->dev, "failed to assign memory for response %d\n", -ENOMEM);
 		return;
 	}
 
+	/*
+	 * The status is stored in bits 31:16 of the SERVICES_SR register.
+	 * It is only valid when BUSY == 0.
+	 * We should *never* get an interrupt while the controller is
+	 * still in the busy state. If we do, something has gone badly
+	 * wrong & the content of the mailbox would not be valid.
+	 */
+	if (mpfs_mbox_busy(mbox)) {
+		dev_err(mbox->dev, "got an interrupt but system controller is busy\n");
+		response->resp_status = 0xDEAD;
+		return;
+	}
+
+	status = readl_relaxed(mbox->ctrl_base + SERVICES_SR_OFFSET);
+
+	/*
+	 * If the status of the individual servers is non-zero, the service has
+	 * failed. The contents of the mailbox at this point are not be valid,
+	 * so don't bother reading them. Set the status so that the driver
+	 * implementing the service can handle the result.
+	 */
+	response->resp_status = (status & SCB_STATUS_MASK) >> SCB_STATUS_POS;
+	if (response->resp_status)
+		return;
+
 	if (!mpfs_mbox_busy(mbox)) {
 		for (i = 0; i < num_words; i++) {
 			response->resp_msg[i] =
Signature · 7 lines
2.38.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
author maintainer
#1 lore raw

PATCH v3 1/2 soc: microchip: mpfs: handle failed system service requests

In reply to Conor Dooley

From: Conor Dooley <[email protected]>

If a service request fails, a non-zero, per-service error code will be set. Since the individual service drivers may wish to handle things differently, there's little point trying to do anything intelligent in the system controller driver. Let the caller know that things went wrong & leave the details of handling the error to it.

Fixes d0054a470c33 ("soc: add microchip polarfire soc system controller")
---
 drivers/soc/microchip/mpfs-sys-controller.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/soc/microchip/mpfs-sys-controller.c +6 -0
index 6e20207b5756..539fc24b397d 100644
@@ -52,6 +52,12 @@ int mpfs_blocking_transaction(struct mpfs_sys_controller *sys_controller, struct
 
 	mutex_unlock(&transaction_lock);
 
+	if (ret)
+		return ret;
+
+	if (msg->response->resp_status)
+		ret = -EIO;
+
 	return ret;
 }
 EXPORT_SYMBOL(mpfs_blocking_transaction);
Signature · 7 lines
2.38.1


_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
maintainer Reviewed-by on v3 1/2
#2 lore raw

Re: [PATCH v3 1/2] soc: microchip: mpfs: handle failed system service requests

Quoted 21 lines · Conor Dooley

From: Conor Dooley <[email protected]>

If a service request fails, a non-zero, per-service error code will be set. Since the individual service drivers may wish to handle things differently, there's little point trying to do anything intelligent in the system controller driver. Let the caller know that things went wrong & leave the details of handling the error to it.

Fixes d0054a470c33 ("soc: add microchip polarfire soc system controller")
---
 drivers/soc/microchip/mpfs-sys-controller.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/soc/microchip/mpfs-sys-controller.c +0 -0
index 6e20207b5756..539fc24b397d 100644
@@ -52,6 +52,12 @@ int mpfs_blocking_transaction(struct mpfs_sys_controller *sys_controller, struct
 
 	mutex_unlock(&transaction_lock);

drivers/soc/microchip/mpfs-sys-controller.c int mpfs_blocking_transaction(struct mpfs_sys_controller *sys_controller, struct · line 63

Quoted 10 lines
drivers/soc/microchip/mpfs-sys-controller.c +6 -0
 
+	if (ret)
+		return ret;
+
+	if (msg->response->resp_status)
+		ret = -EIO;
+
 	return ret;
 }
 EXPORT_SYMBOL(mpfs_blocking_transaction);
Reviewed-by on v3 2/2
#4 lore raw

Re: [PATCH v3 2/2] mailbox: mpfs: read the system controller's status

In reply to Conor Dooley
Quoted 76 lines · Conor Dooley

From: Conor Dooley <[email protected]>

Some services explicitly return an error code in their response, but others rely on the system controller to set a status in its status register. The meaning of the bits varies based on what service is requested, so pass it back up to the driver that requested the service in the first place. The field in the message struct already existed, but was unused until now.

If the system controller is busy, in which case we should never actually be in the interrupt handler, or if the service fails the mailbox itself should not be read. Callers should check the status before operating on the response.

There's an existing, but unused, #define for the mailbox mask - but it was incorrect. It was doing a GENMASK_ULL(32, 16) which should've just been a GENMASK(31, 16), so fix that up and start using it.

Fixes 83d7b1560810 ("mbox: add polarfire soc system controller mailbox")
---
 drivers/mailbox/mailbox-mpfs.c | 31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)
drivers/mailbox/mailbox-mpfs.c +21 -3
index cfacb3f320a6..853901acaeec 100644
@@ -2,7 +2,7 @@
 /*
  * Microchip PolarFire SoC (MPFS) system controller/mailbox controller driver
  *
- * Copyright (c) 2020 Microchip Corporation. All rights reserved.
+ * Copyright (c) 2020-2022 Microchip Corporation. All rights reserved.
  *
  * Author: Conor Dooley <[email protected]>
  *
@@ -56,7 +56,7 @@
 #define SCB_STATUS_NOTIFY_MASK BIT(SCB_STATUS_NOTIFY)
 
 #define SCB_STATUS_POS (16)
-#define SCB_STATUS_MASK GENMASK_ULL(SCB_STATUS_POS + SCB_MASK_WIDTH, SCB_STATUS_POS)
+#define SCB_STATUS_MASK GENMASK(SCB_STATUS_POS + SCB_MASK_WIDTH - 1, SCB_STATUS_POS)
 
 struct mpfs_mbox {
 	struct mbox_controller controller;
@@ -130,13 +130,38 @@ static void mpfs_mbox_rx_data(struct mbox_chan *chan)
 	struct mpfs_mbox *mbox = (struct mpfs_mbox *)chan->con_priv;
 	struct mpfs_mss_response *response = mbox->response;
 	u16 num_words = ALIGN((response->resp_size), (4)) / 4U;
-	u32 i;
+	u32 i, status;
 
 	if (!response->resp_msg) {
 		dev_err(mbox->dev, "failed to assign memory for response %d\n", -ENOMEM);
 		return;
 	}
 
+	/*
+	 * The status is stored in bits 31:16 of the SERVICES_SR register.
+	 * It is only valid when BUSY == 0.
+	 * We should *never* get an interrupt while the controller is
+	 * still in the busy state. If we do, something has gone badly
+	 * wrong & the content of the mailbox would not be valid.
+	 */
+	if (mpfs_mbox_busy(mbox)) {
+		dev_err(mbox->dev, "got an interrupt but system controller is busy\n");
+		response->resp_status = 0xDEAD;
+		return;
+	}
+
+	status = readl_relaxed(mbox->ctrl_base + SERVICES_SR_OFFSET);
+
+	/*
+	 * If the status of the individual servers is non-zero, the service has
+	 * failed. The contents of the mailbox at this point are not be valid,

drivers/mailbox/mailbox-mpfs.c static void mpfs_mbox_rx_data(struct mbox_chan *chan) · line 167

Quoted 10 lines
drivers/mailbox/mailbox-mpfs.c +7 -0
+	 * so don't bother reading them. Set the status so that the driver
+	 * implementing the service can handle the result.
+	 */
+	response->resp_status = (status & SCB_STATUS_MASK) >> SCB_STATUS_POS;
+	if (response->resp_status)
+		return;
+
 	if (!mpfs_mbox_busy(mbox)) {
 		for (i = 0; i < num_words; i++) {
 			response->resp_msg[i] =

I'm assuming this is aimed at some other tree.

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

author maintainer Applied
#5 lore raw

Re: (subset) [PATCH v3 0/2] mpfs: fix handling failed service requests

In reply to Conor Dooley

From: Conor Dooley <[email protected]>

Quoted 8 lines · Conor Dooley

From: Conor Dooley <[email protected]>

As things stand, if a service request fails, we never know about it as the status register is not read. This is quite obviously far from ideal, as we would be reading garbage from the mailbox if a service was to fail.

[...]

Applied to riscv-soc-fixes, thanks!

[1/2] soc: microchip: mpfs: handle failed system service requests

      https://git.kernel.org/conor/c/730892135b7d

Thanks,
Conor.

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv