FSI scom driver overhaul

Otherwise, multiple clients can open the driver and attempt to access the PIB at the same time, thus clobbering each other in the process.

Signed-off-by Benjamin Herrenschmidt <[email protected]>

---
  drivers/fsi/fsi-scom.c | 25 ++++++++++++++++++-------
  1 file changed, 18 insertions(+), 7 deletions(-)

drivers/fsi/fsi-scom.c +18 -7

index c8eb5e5b94a7..3cba0eb645e1 100644
@@ -37,6 +37,7 @@ struct scom_device {
  	struct list_head link;
  	struct fsi_device *fsi_dev;
  	struct miscdevice mdev;
+	struct mutex lock;
  	char	name[32];
  	int idx;
  };

@@ -53,21 +54,26 @@ static int put_scom(struct scom_device *scom_dev, uint64_t value,
  	int rc;
  	uint32_t data;
 
+	mutex_lock(&scom_dev->lock);
+
  	data = cpu_to_be32((value >> 32) & 0xffffffff);
  	rc = fsi_device_write(scom_dev->fsi_dev, SCOM_DATA0_REG, &data,
  				sizeof(uint32_t));
  	if (rc)
-		return rc;
+		goto bail;
 
  	data = cpu_to_be32(value & 0xffffffff);
  	rc = fsi_device_write(scom_dev->fsi_dev, SCOM_DATA1_REG, &data,
  				sizeof(uint32_t));
  	if (rc)
-		return rc;
+		goto bail;
 
  	data = cpu_to_be32(SCOM_WRITE_CMD | addr);
-	return fsi_device_write(scom_dev->fsi_dev, SCOM_CMD_REG, &data,
+	rc = fsi_device_write(scom_dev->fsi_dev, SCOM_CMD_REG, &data,
  				sizeof(uint32_t));
+ bail:
+	mutex_unlock(&scom_dev->lock);
+	return rc;
  }
 
  static int get_scom(struct scom_device *scom_dev, uint64_t *value,

@@ -76,27 +82,31 @@ static int get_scom(struct scom_device *scom_dev, uint64_t *value,
  	uint32_t result, data;
  	int rc;
 
+
+	mutex_lock(&scom_dev->lock);
  	*value = 0ULL;
  	data = cpu_to_be32(addr);
  	rc = fsi_device_write(scom_dev->fsi_dev, SCOM_CMD_REG, &data,
  				sizeof(uint32_t));
  	if (rc)
-		return rc;
+		goto bail;
 
  	rc = fsi_device_read(scom_dev->fsi_dev, SCOM_DATA0_REG, &result,
  				sizeof(uint32_t));
  	if (rc)
-		return rc;
+		goto bail;
 
  	*value |= (uint64_t)cpu_to_be32(result) << 32;
  	rc = fsi_device_read(scom_dev->fsi_dev, SCOM_DATA1_REG, &result,
  				sizeof(uint32_t));
  	if (rc)
-		return rc;
+		goto bail;
 
  	*value |= cpu_to_be32(result);
 
-	return 0;
+ bail:
+	mutex_unlock(&scom_dev->lock);
+	return rc;
  }
 
  static ssize_t scom_read(struct file *filep, char __user *buf, size_t len,

@@ -183,6 +193,7 @@ static int scom_probe(struct device *dev)
  	if (!scom)
  		return -ENOMEM;
 
+	mutex_init(&scom->lock);
  	scom->idx = ida_simple_get(&scom_ida, 1, INT_MAX, GFP_KERNEL);
  	snprintf(scom->name, sizeof(scom->name), "scom%d", scom->idx);
  	scom->fsi_dev = fsi_dev;

No functional changes

Signed-off-by Benjamin Herrenschmidt <[email protected]>

---
  drivers/fsi/fsi-scom.c | 8 ++++----
  1 file changed, 4 insertions(+), 4 deletions(-)

drivers/fsi/fsi-scom.c +4 -4

index 3cba0eb645e1..8a608db0aa07 100644
@@ -49,7 +49,7 @@ static struct list_head scom_devices;
  static DEFINE_IDA(scom_ida);
 
  static int put_scom(struct scom_device *scom_dev, uint64_t value,
-			uint32_t addr)
+		    uint32_t addr)
  {
  	int rc;
  	uint32_t data;

@@ -77,7 +77,7 @@ static int put_scom(struct scom_device *scom_dev, uint64_t value,
  }
 
  static int get_scom(struct scom_device *scom_dev, uint64_t *value,
-			uint32_t addr)
+		    uint32_t addr)
  {
  	uint32_t result, data;
  	int rc;

@@ -110,7 +110,7 @@ static int get_scom(struct scom_device *scom_dev, uint64_t *value,
  }
 
  static ssize_t scom_read(struct file *filep, char __user *buf, size_t len,
-			loff_t *offset)
+			 loff_t *offset)
  {
  	int rc;
  	struct miscdevice *mdev =

@@ -136,7 +136,7 @@ static ssize_t scom_read(struct file *filep, char __user *buf, size_t len,
  }
 
  static ssize_t scom_write(struct file *filep, const char __user *buf,
-			size_t len, loff_t *offset)
+			  size_t len, loff_t *offset)
  {
  	int rc;
  	struct miscdevice *mdev = filep->private_data;

Use the proper annotated type __be32 and fixup the accessor used for get_scom()

Signed-off-by Benjamin Herrenschmidt <[email protected]>

---
  drivers/fsi/fsi-scom.c | 9 ++++-----
  1 file changed, 4 insertions(+), 5 deletions(-)

drivers/fsi/fsi-scom.c +4 -5

index 8a608db0aa07..6ddfb6021420 100644
@@ -51,8 +51,8 @@ static DEFINE_IDA(scom_ida);
  static int put_scom(struct scom_device *scom_dev, uint64_t value,
  		    uint32_t addr)
  {
+	__be32 data;
  	int rc;
-	uint32_t data;
 
  	mutex_lock(&scom_dev->lock);
 

@@ -79,7 +79,7 @@ static int put_scom(struct scom_device *scom_dev, uint64_t value,
  static int get_scom(struct scom_device *scom_dev, uint64_t *value,
  		    uint32_t addr)
  {
-	uint32_t result, data;
+	__be32 result, data;
  	int rc;
 
 

@@ -96,14 +96,13 @@ static int get_scom(struct scom_device *scom_dev, uint64_t *value,
  	if (rc)
  		goto bail;
 
-	*value |= (uint64_t)cpu_to_be32(result) << 32;
+	*value |= (uint64_t)be32_to_cpu(result) << 32;
  	rc = fsi_device_read(scom_dev->fsi_dev, SCOM_DATA1_REG, &result,
  				sizeof(uint32_t));
  	if (rc)
  		goto bail;
 
-	*value |= cpu_to_be32(result);
-
+	*value |= be32_to_cpu(result);
   bail:
  	mutex_unlock(&scom_dev->lock);
  	return rc;

Add a few more register and bit definitions, also define and use SCOM_READ_CMD (which is 0 but it makes the code clearer)

Signed-off-by Benjamin Herrenschmidt <[email protected]>

---
  drivers/fsi/fsi-scom.c | 19 ++++++++++++++++++-
  1 file changed, 18 insertions(+), 1 deletion(-)

drivers/fsi/fsi-scom.c +18 -1

index 6ddfb6021420..e98573ecdae1 100644
@@ -30,8 +30,25 @@
  #define SCOM_DATA0_REG		0x00
  #define SCOM_DATA1_REG		0x04
  #define SCOM_CMD_REG		0x08
+#define SCOM_FSI2PIB_RESET_REG	0x18
+#define SCOM_STATUS_REG		0x1C /* Read */
+#define SCOM_PIB_RESET_REG	0x1C /* Write */
 
+/* Command register */
  #define SCOM_WRITE_CMD		0x80000000
+#define SCOM_READ_CMD		0x00000000
+
+/* Status register bits */
+#define SCOM_STATUS_ERR_SUMMARY		0x80000000
+#define SCOM_STATUS_PROTECTION		0x01000000
+#define SCOM_STATUS_PIB_ABORT		0x00100000
+#define SCOM_STATUS_PIB_RESP_MASK	0x00007000
+#define SCOM_STATUS_PIB_RESP_SHIFT	12
+
+#define SCOM_STATUS_ANY_ERR		(SCOM_STATUS_ERR_SUMMARY | \
+					 SCOM_STATUS_PROTECTION | \
+					 SCOM_STATUS_PIB_ABORT | \
+					 SCOM_STATUS_PIB_RESP_MASK)
 
  struct scom_device {
  	struct list_head link;

@@ -85,7 +102,7 @@ static int get_scom(struct scom_device *scom_dev, uint64_t *value,
 
  	mutex_lock(&scom_dev->lock);
  	*value = 0ULL;
-	data = cpu_to_be32(addr);
+	data = cpu_to_be32(SCOM_READ_CMD | addr);
  	rc = fsi_device_write(scom_dev->fsi_dev, SCOM_CMD_REG, &data,
  				sizeof(uint32_t));
  	if (rc)

This was too hard to split ... this adds a number of features to the SCOM user interface:

- Support for indirect SCOMs

- read()/write() interface now handle errors and retries

- New ioctl() "raw" interface for use by debuggers

Signed-off-by Benjamin Herrenschmidt <[email protected]>

---
  drivers/fsi/fsi-scom.c   | 424 ++++++++++++++++++++++++++++++++++++---
  include/uapi/linux/fsi.h |  56 ++++++
  2 files changed, 450 insertions(+), 30 deletions(-)

create mode 100644 include/uapi/linux/fsi.h

drivers/fsi/fsi-scom.c +1 -0

index e98573ecdae1..39c74351f1bf 100644
@@ -24,6 +24,8 @@
  #include <linux/list.h>
  #include <linux/idr.h>
 
+

drivers/fsi/fsi-scom.c +43 -4

+	if (!rc && (flags & (SCOM_RESET_PIB | SCOM_RESET_INTF)))
+		rc = fsi_device_write(scom->fsi_dev, SCOM_FSI2PIB_RESET_REG, &dummy,
+				      sizeof(uint32_t));
+	return rc;
+}
+
+static int scom_check(struct scom_device *scom, void __user *argp)
+{
+	/* Still need to find out how to get "protected" */
+	return put_user(SCOM_CHECK_SUPPORTED, (__u32 __user *)argp);
+}
+
+static long scom_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	struct miscdevice *mdev = file->private_data;
+	struct scom_device *scom = to_scom_dev(mdev);
+	void __user *argp = (void __user *)arg;
+	int rc = -ENOTTY;
+
+	mutex_lock(&scom->lock);
+	switch(cmd) {
+	case FSI_SCOM_CHECK:
+		rc = scom_check(scom, argp);
+		break;
+	case FSI_SCOM_READ:
+		rc = scom_raw_read(scom, argp);
+		break;
+	case FSI_SCOM_WRITE:
+		rc = scom_raw_write(scom, argp);
+		break;
+	case FSI_SCOM_RESET:
+		rc = scom_reset(scom, argp);
+		break;
+	}
+	mutex_unlock(&scom->lock);
+	return rc;
+}
+
  static const struct file_operations scom_fops = {
-	.owner	= THIS_MODULE,
-	.llseek	= scom_llseek,
-	.read	= scom_read,
-	.write	= scom_write,
+	.owner		= THIS_MODULE,
+	.llseek		= scom_llseek,
+	.read		= scom_read,
+	.write		= scom_write,
+	.unlocked_ioctl	= scom_ioctl,
  };
 
  static int scom_probe(struct device *dev)

include/uapi/linux/fsi.h new file +56 -0

new file mode 100644
index 000000000000..6008d93f2e48
@@ -0,0 +1,56 @@
+#ifndef _UAPI_LINUX_FSI_H
+#define _UAPI_LINUX_FSI_H
+
+#include <linux/ioctl.h>
+
+/*
+ * /dev/scom "raw" ioctl interface
+ *
+ * The driver supports a high level "read/write" interface which
+ * handles retries and converts the status to Linux error codes,
+ * however low level tools an debugger need to access the "raw"
+ * HW status information and interpret it themselves, so this
+ * ioctl interface is also provided for their use case.
+ */
+
+/* Structure for SCOM read/write */
+struct scom_access {
+	__u64	addr;		/* SCOM address, supports indirect */
+	__u64	data;		/* SCOM data (in for write, out for read) */
+	__u64	mask;		/* Data mask for writes */
+	__u32	intf_errors;	/* Interface error flags */
+#define SCOM_INTF_ERR_PARITY		0x00000001 /* Parity error */
+#define SCOM_INTF_ERR_PROTECTION	0x00000002 /* Blocked by secure boot */
+#define SCOM_INTF_ERR_ABORT		0x00000004 /* PIB reset during access */
+#define SCOM_INTF_ERR_UNKNOWN		0x80000000 /* Unknown error */
+	/*
+	 * Note: Any other bit set in intf_errors need to be considered as an
+	 * error. Future implementations may define new error conditions. The
+	 * pib_status below is only valid if intf_errors is 0.
+	 */
+	__u8	pib_status;	/* 3-bit PIB status */
+#define SCOM_PIB_SUCCESS	0	/* Access successful */
+#define SCOM_PIB_BLOCKED	1	/* PIB blocked, pls retry */
+#define SCOM_PIB_OFFLINE	2	/* Chiplet offline */
+#define SCOM_PIB_PARTIAL	3	/* Partial good */
+#define SCOM_PIB_BAD_ADDR	4	/* Invalid address */
+#define SCOM_PIB_CLK_ERR	5	/* Clock error */
+#define SCOM_PIB_PARITY_ERR	6	/* Parity error on the PIB bus */
+#define SCOM_PIB_TIMEOUT	7	/* Bus timeout */
+	__u8	pad;
+};
+
+/* Flags for SCOM check */
+#define SCOM_CHECK_SUPPORTED	0x00000001	/* Interface supported */
+#define SCOM_CHECK_PROTECTED	0x00000002	/* Interface blocked by secure boot */
+
+/* Flags for SCOM reset */
+#define SCOM_RESET_INTF		0x00000001	/* Reset interface */
+#define SCOM_RESET_PIB		0x00000002	/* Reset PIB */
+
+#define FSI_SCOM_CHECK	_IOR('s', 0x00, __u32)
+#define FSI_SCOM_READ	_IOWR('s', 0x01, struct scom_access)
+#define FSI_SCOM_WRITE	_IOWR('s', 0x02, struct scom_access)
+#define FSI_SCOM_RESET	_IOW('s', 0x03, __u32)
+
+#endif /* _UAPI_LINUX_FSI_H */

Quoted 30 lines · Benjamin Herrenschmidt

This was too hard to split ... this adds a number of features to the SCOM user interface:

- Support for indirect SCOMs

- read()/write() interface now handle errors and retries

- New ioctl() "raw" interface for use by debuggers

Signed-off-by Benjamin Herrenschmidt <[email protected]>

---
  drivers/fsi/fsi-scom.c   | 424 ++++++++++++++++++++++++++++++++++++---
  include/uapi/linux/fsi.h |  56 ++++++
  2 files changed, 450 insertions(+), 30 deletions(-)

create mode 100644 include/uapi/linux/fsi.h

drivers/fsi/fsi-scom.c +7 -0

index e98573ecdae1..39c74351f1bf 100644
@@ -24,6 +24,8 @@
  #include <linux/list.h>
  #include <linux/idr.h>
+
+static int scom_reset(struct scom_device *scom, void __user *argp)
+{
+	uint32_t flags, dummy = -1;
+	int rc = 0;
+
+	if (get_user(flags, (__u32 __user *)argp))

drivers/fsi/fsi-scom.c static loff_t scom_llseek(struct file *file, loff_t offset, int whence) · line 520

Quoted 4 lines

drivers/fsi/fsi-scom.c +4 -0

+		return -EFAULT;
+	if (flags & SCOM_RESET_PIB)
+		rc = fsi_device_write(scom->fsi_dev, SCOM_PIB_RESET_REG, &dummy,
+				      sizeof(uint32_t));

I realize this is a user requested flag but I believe the BMC is never supposed to issue this type of reset, due to the possibility of breaking stuff on the host side. Not sure if it should even be available?

Otherwise, looks good!
Thanks,
Eddie

Quoted 113 lines

drivers/fsi/fsi-scom.c +43 -4

+	if (!rc && (flags & (SCOM_RESET_PIB | SCOM_RESET_INTF)))
+		rc = fsi_device_write(scom->fsi_dev, SCOM_FSI2PIB_RESET_REG, &dummy,
+				      sizeof(uint32_t));
+	return rc;
+}
+
+static int scom_check(struct scom_device *scom, void __user *argp)
+{
+	/* Still need to find out how to get "protected" */
+	return put_user(SCOM_CHECK_SUPPORTED, (__u32 __user *)argp);
+}
+
+static long scom_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	struct miscdevice *mdev = file->private_data;
+	struct scom_device *scom = to_scom_dev(mdev);
+	void __user *argp = (void __user *)arg;
+	int rc = -ENOTTY;
+
+	mutex_lock(&scom->lock);
+	switch(cmd) {
+	case FSI_SCOM_CHECK:
+		rc = scom_check(scom, argp);
+		break;
+	case FSI_SCOM_READ:
+		rc = scom_raw_read(scom, argp);
+		break;
+	case FSI_SCOM_WRITE:
+		rc = scom_raw_write(scom, argp);
+		break;
+	case FSI_SCOM_RESET:
+		rc = scom_reset(scom, argp);
+		break;
+	}
+	mutex_unlock(&scom->lock);
+	return rc;
+}
+
  static const struct file_operations scom_fops = {
-	.owner	= THIS_MODULE,
-	.llseek	= scom_llseek,
-	.read	= scom_read,
-	.write	= scom_write,
+	.owner		= THIS_MODULE,
+	.llseek		= scom_llseek,
+	.read		= scom_read,
+	.write		= scom_write,
+	.unlocked_ioctl	= scom_ioctl,
  };
 
  static int scom_probe(struct device *dev)

include/uapi/linux/fsi.h new file +56 -0

new file mode 100644
index 000000000000..6008d93f2e48
@@ -0,0 +1,56 @@
+#ifndef _UAPI_LINUX_FSI_H
+#define _UAPI_LINUX_FSI_H
+
+#include <linux/ioctl.h>
+
+/*
+ * /dev/scom "raw" ioctl interface
+ *
+ * The driver supports a high level "read/write" interface which
+ * handles retries and converts the status to Linux error codes,
+ * however low level tools an debugger need to access the "raw"
+ * HW status information and interpret it themselves, so this
+ * ioctl interface is also provided for their use case.
+ */
+
+/* Structure for SCOM read/write */
+struct scom_access {
+	__u64	addr;		/* SCOM address, supports indirect */
+	__u64	data;		/* SCOM data (in for write, out for read) */
+	__u64	mask;		/* Data mask for writes */
+	__u32	intf_errors;	/* Interface error flags */
+#define SCOM_INTF_ERR_PARITY		0x00000001 /* Parity error */
+#define SCOM_INTF_ERR_PROTECTION	0x00000002 /* Blocked by secure boot */
+#define SCOM_INTF_ERR_ABORT		0x00000004 /* PIB reset during access */
+#define SCOM_INTF_ERR_UNKNOWN		0x80000000 /* Unknown error */
+	/*
+	 * Note: Any other bit set in intf_errors need to be considered as an
+	 * error. Future implementations may define new error conditions. The
+	 * pib_status below is only valid if intf_errors is 0.
+	 */
+	__u8	pib_status;	/* 3-bit PIB status */
+#define SCOM_PIB_SUCCESS	0	/* Access successful */
+#define SCOM_PIB_BLOCKED	1	/* PIB blocked, pls retry */
+#define SCOM_PIB_OFFLINE	2	/* Chiplet offline */
+#define SCOM_PIB_PARTIAL	3	/* Partial good */
+#define SCOM_PIB_BAD_ADDR	4	/* Invalid address */
+#define SCOM_PIB_CLK_ERR	5	/* Clock error */
+#define SCOM_PIB_PARITY_ERR	6	/* Parity error on the PIB bus */
+#define SCOM_PIB_TIMEOUT	7	/* Bus timeout */
+	__u8	pad;
+};
+
+/* Flags for SCOM check */
+#define SCOM_CHECK_SUPPORTED	0x00000001	/* Interface supported */
+#define SCOM_CHECK_PROTECTED	0x00000002	/* Interface blocked by secure boot */
+
+/* Flags for SCOM reset */
+#define SCOM_RESET_INTF		0x00000001	/* Reset interface */
+#define SCOM_RESET_PIB		0x00000002	/* Reset PIB */
+
+#define FSI_SCOM_CHECK	_IOR('s', 0x00, __u32)
+#define FSI_SCOM_READ	_IOWR('s', 0x01, struct scom_access)
+#define FSI_SCOM_WRITE	_IOWR('s', 0x02, struct scom_access)
+#define FSI_SCOM_RESET	_IOW('s', 0x03, __u32)
+
+#endif /* _UAPI_LINUX_FSI_H */

Quoted 30 lines · Eddie James

Quoted 29 lines · Benjamin Herrenschmidt

This was too hard to split ... this adds a number of features to the SCOM user interface:

- Support for indirect SCOMs

- read()/write() interface now handle errors and retries

- New ioctl() "raw" interface for use by debuggers

Signed-off-by Benjamin Herrenschmidt <[email protected]>

---
   drivers/fsi/fsi-scom.c   | 424 ++++++++++++++++++++++++++++++++++++---
   include/uapi/linux/fsi.h |  56 ++++++
   2 files changed, 450 insertions(+), 30 deletions(-)

create mode 100644 include/uapi/linux/fsi.h

drivers/fsi/fsi-scom.c +6 -0

index e98573ecdae1..39c74351f1bf 100644
@@ -24,6 +24,8 @@
   #include <linux/list.h>
   #include <linux/idr.h>
+
+static int scom_reset(struct scom_device *scom, void __user *argp)
+{
+	uint32_t flags, dummy = -1;
+	int rc = 0;
+

Quoted 8 lines

drivers/fsi/fsi-scom.c static loff_t scom_llseek(struct file *file, loff_t offset, int whence) · line 520

Quoted 5 lines

drivers/fsi/fsi-scom.c +5 -0

+	if (get_user(flags, (__u32 __user *)argp))
+		return -EFAULT;
+	if (flags & SCOM_RESET_PIB)
+		rc = fsi_device_write(scom->fsi_dev, SCOM_PIB_RESET_REG, &dummy,
+				      sizeof(uint32_t));

I realize this is a user requested flag but I believe the BMC is never supposed to issue this type of reset, due to the possibility of breaking stuff on the host side. Not sure if it should even be available?

It's for use by cronus or similar low level system debuggers.

Cheers,
Ben.

Quoted 118 lines

Otherwise, looks good!
Thanks,
Eddie

Quoted 113 lines

drivers/fsi/fsi-scom.c +43 -4

+	if (!rc && (flags & (SCOM_RESET_PIB | SCOM_RESET_INTF)))
+		rc = fsi_device_write(scom->fsi_dev, SCOM_FSI2PIB_RESET_REG, &dummy,
+				      sizeof(uint32_t));
+	return rc;
+}
+
+static int scom_check(struct scom_device *scom, void __user *argp)
+{
+	/* Still need to find out how to get "protected" */
+	return put_user(SCOM_CHECK_SUPPORTED, (__u32 __user *)argp);
+}
+
+static long scom_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	struct miscdevice *mdev = file->private_data;
+	struct scom_device *scom = to_scom_dev(mdev);
+	void __user *argp = (void __user *)arg;
+	int rc = -ENOTTY;
+
+	mutex_lock(&scom->lock);
+	switch(cmd) {
+	case FSI_SCOM_CHECK:
+		rc = scom_check(scom, argp);
+		break;
+	case FSI_SCOM_READ:
+		rc = scom_raw_read(scom, argp);
+		break;
+	case FSI_SCOM_WRITE:
+		rc = scom_raw_write(scom, argp);
+		break;
+	case FSI_SCOM_RESET:
+		rc = scom_reset(scom, argp);
+		break;
+	}
+	mutex_unlock(&scom->lock);
+	return rc;
+}
+
   static const struct file_operations scom_fops = {
-	.owner	= THIS_MODULE,
-	.llseek	= scom_llseek,
-	.read	= scom_read,
-	.write	= scom_write,
+	.owner		= THIS_MODULE,
+	.llseek		= scom_llseek,
+	.read		= scom_read,
+	.write		= scom_write,
+	.unlocked_ioctl	= scom_ioctl,
   };
 
   static int scom_probe(struct device *dev)

include/uapi/linux/fsi.h new file +56 -0

new file mode 100644
index 000000000000..6008d93f2e48
@@ -0,0 +1,56 @@
+#ifndef _UAPI_LINUX_FSI_H
+#define _UAPI_LINUX_FSI_H
+
+#include <linux/ioctl.h>
+
+/*
+ * /dev/scom "raw" ioctl interface
+ *
+ * The driver supports a high level "read/write" interface which
+ * handles retries and converts the status to Linux error codes,
+ * however low level tools an debugger need to access the "raw"
+ * HW status information and interpret it themselves, so this
+ * ioctl interface is also provided for their use case.
+ */
+
+/* Structure for SCOM read/write */
+struct scom_access {
+	__u64	addr;		/* SCOM address, supports indirect */
+	__u64	data;		/* SCOM data (in for write, out for read) */
+	__u64	mask;		/* Data mask for writes */
+	__u32	intf_errors;	/* Interface error flags */
+#define SCOM_INTF_ERR_PARITY		0x00000001 /* Parity error */
+#define SCOM_INTF_ERR_PROTECTION	0x00000002 /* Blocked by secure boot */
+#define SCOM_INTF_ERR_ABORT		0x00000004 /* PIB reset during access */
+#define SCOM_INTF_ERR_UNKNOWN		0x80000000 /* Unknown error */
+	/*
+	 * Note: Any other bit set in intf_errors need to be considered as an
+	 * error. Future implementations may define new error conditions. The
+	 * pib_status below is only valid if intf_errors is 0.
+	 */
+	__u8	pib_status;	/* 3-bit PIB status */
+#define SCOM_PIB_SUCCESS	0	/* Access successful */
+#define SCOM_PIB_BLOCKED	1	/* PIB blocked, pls retry */
+#define SCOM_PIB_OFFLINE	2	/* Chiplet offline */
+#define SCOM_PIB_PARTIAL	3	/* Partial good */
+#define SCOM_PIB_BAD_ADDR	4	/* Invalid address */
+#define SCOM_PIB_CLK_ERR	5	/* Clock error */
+#define SCOM_PIB_PARITY_ERR	6	/* Parity error on the PIB bus */
+#define SCOM_PIB_TIMEOUT	7	/* Bus timeout */
+	__u8	pad;
+};
+
+/* Flags for SCOM check */
+#define SCOM_CHECK_SUPPORTED	0x00000001	/* Interface supported */
+#define SCOM_CHECK_PROTECTED	0x00000002	/* Interface blocked by secure boot */
+
+/* Flags for SCOM reset */
+#define SCOM_RESET_INTF		0x00000001	/* Reset interface */
+#define SCOM_RESET_PIB		0x00000002	/* Reset PIB */
+
+#define FSI_SCOM_CHECK	_IOR('s', 0x00, __u32)
+#define FSI_SCOM_READ	_IOWR('s', 0x01, struct scom_access)
+#define FSI_SCOM_WRITE	_IOWR('s', 0x02, struct scom_access)
+#define FSI_SCOM_RESET	_IOW('s', 0x03, __u32)
+
+#endif /* _UAPI_LINUX_FSI_H */

This was too hard to split ... this adds a number of features to the SCOM user interface:

- Support for indirect SCOMs

- read()/write() interface now handle errors and retries

- New ioctl() "raw" interface for use by debuggers

Signed-off-by Benjamin Herrenschmidt <[email protected]>

include/uapi/linux/fsi.h new file +52 -1

+
+/*
+ * /dev/scom "raw" ioctl interface
+ *
+ * The driver supports a high level "read/write" interface which
+ * handles retries and converts the status to Linux error codes,
+ * however low level tools an debugger need to access the "raw"
+ * HW status information and interpret it themselves, so this
+ * ioctl interface is also provided for their use case.
+ */
+
+/* Structure for SCOM read/write */
+struct scom_access {
+       __u64   addr;           /* SCOM address, supports indirect */
+       __u64   data;           /* SCOM data (in for write, out for read) */
+       __u64   mask;           /* Data mask for writes */
+       __u32   intf_errors;    /* Interface error flags */
+#define SCOM_INTF_ERR_PARITY           0x00000001 /* Parity error */
+#define SCOM_INTF_ERR_PROTECTION       0x00000002 /* Blocked by secure boot */
+#define SCOM_INTF_ERR_ABORT            0x00000004 /* PIB reset during access */
+#define SCOM_INTF_ERR_UNKNOWN          0x80000000 /* Unknown error */
+       /*
+        * Note: Any other bit set in intf_errors need to be considered as an
+        * error. Future implementations may define new error conditions. The
+        * pib_status below is only valid if intf_errors is 0.
+        */
+       __u8    pib_status;     /* 3-bit PIB status */
+#define SCOM_PIB_SUCCESS       0       /* Access successful */
+#define SCOM_PIB_BLOCKED       1       /* PIB blocked, pls retry */
+#define SCOM_PIB_OFFLINE       2       /* Chiplet offline */
+#define SCOM_PIB_PARTIAL       3       /* Partial good */
+#define SCOM_PIB_BAD_ADDR      4       /* Invalid address */
+#define SCOM_PIB_CLK_ERR       5       /* Clock error */
+#define SCOM_PIB_PARITY_ERR    6       /* Parity error on the PIB bus */
+#define SCOM_PIB_TIMEOUT       7       /* Bus timeout */
+       __u8    pad;
+};
+
+/* Flags for SCOM check */
+#define SCOM_CHECK_SUPPORTED   0x00000001      /* Interface supported */
+#define SCOM_CHECK_PROTECTED   0x00000002      /* Interface blocked by secure boot */
+
+/* Flags for SCOM reset */
+#define SCOM_RESET_INTF                0x00000001      /* Reset interface */
+#define SCOM_RESET_PIB         0x00000002      /* Reset PIB */
+
+#define FSI_SCOM_CHECK _IOR('s', 0x00, __u32)
+#define FSI_SCOM_READ  _IOWR('s', 0x01, struct scom_access)
+#define FSI_SCOM_WRITE _IOWR('s', 0x02, struct scom_access)
+#define FSI_SCOM_RESET _IOW('s', 0x03, __u32)
+
+#endif /* _UAPI_LINUX_FSI_H */
--

2.17.0

Quoted 10 lines · Benjamin Herrenschmidt

This was too hard to split ... this adds a number of features to the SCOM user interface:

- Support for indirect SCOMs

- read()/write() interface now handle errors and retries

- New ioctl() "raw" interface for use by debuggers

Signed-off-by Benjamin Herrenschmidt <[email protected]>

Looks okay to me. I will put it in the openbmc tree with Eddie's ack for more testing.

I got a warning from the 0day infra, I made comments below.

We should get Alistair review the ioctl interface to ensure we have everything that pdbg might need.

include/uapi/linux/fsi.h line 4

Quoted 7 lines

+++ b/include/uapi/linux/fsi.h

include/uapi/linux/fsi.h new file +4 -0

@@ -0,0 +1,56 @@
+#ifndef _UAPI_LINUX_FSI_H
+#define _UAPI_LINUX_FSI_H
+
+#include <linux/ioctl.h>

This needs to include <linux/types.h> for the __u64 etc types.

We should also put a licence in the header. Probably:

/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */

Cheers,

Joel

Quoted 55 lines

include/uapi/linux/fsi.h new file +52 -1

+
+/*
+ * /dev/scom "raw" ioctl interface
+ *
+ * The driver supports a high level "read/write" interface which
+ * handles retries and converts the status to Linux error codes,
+ * however low level tools an debugger need to access the "raw"
+ * HW status information and interpret it themselves, so this
+ * ioctl interface is also provided for their use case.
+ */
+
+/* Structure for SCOM read/write */
+struct scom_access {
+       __u64   addr;           /* SCOM address, supports indirect */
+       __u64   data;           /* SCOM data (in for write, out for read) */
+       __u64   mask;           /* Data mask for writes */
+       __u32   intf_errors;    /* Interface error flags */
+#define SCOM_INTF_ERR_PARITY           0x00000001 /* Parity error */
+#define SCOM_INTF_ERR_PROTECTION       0x00000002 /* Blocked by secure boot */
+#define SCOM_INTF_ERR_ABORT            0x00000004 /* PIB reset during access */
+#define SCOM_INTF_ERR_UNKNOWN          0x80000000 /* Unknown error */
+       /*
+        * Note: Any other bit set in intf_errors need to be considered as an
+        * error. Future implementations may define new error conditions. The
+        * pib_status below is only valid if intf_errors is 0.
+        */
+       __u8    pib_status;     /* 3-bit PIB status */
+#define SCOM_PIB_SUCCESS       0       /* Access successful */
+#define SCOM_PIB_BLOCKED       1       /* PIB blocked, pls retry */
+#define SCOM_PIB_OFFLINE       2       /* Chiplet offline */
+#define SCOM_PIB_PARTIAL       3       /* Partial good */
+#define SCOM_PIB_BAD_ADDR      4       /* Invalid address */
+#define SCOM_PIB_CLK_ERR       5       /* Clock error */
+#define SCOM_PIB_PARITY_ERR    6       /* Parity error on the PIB bus */
+#define SCOM_PIB_TIMEOUT       7       /* Bus timeout */
+       __u8    pad;
+};
+
+/* Flags for SCOM check */
+#define SCOM_CHECK_SUPPORTED   0x00000001      /* Interface supported */
+#define SCOM_CHECK_PROTECTED   0x00000002      /* Interface blocked by secure boot */
+
+/* Flags for SCOM reset */
+#define SCOM_RESET_INTF                0x00000001      /* Reset interface */
+#define SCOM_RESET_PIB         0x00000002      /* Reset PIB */
+
+#define FSI_SCOM_CHECK _IOR('s', 0x00, __u32)
+#define FSI_SCOM_READ  _IOWR('s', 0x01, struct scom_access)
+#define FSI_SCOM_WRITE _IOWR('s', 0x02, struct scom_access)
+#define FSI_SCOM_RESET _IOW('s', 0x03, __u32)
+
+#endif /* _UAPI_LINUX_FSI_H */
--

2.17.0

We have everything that cronus needs and more than pdbg needs afaik :-)

That said, cronus does a bunch of other stupid things that I'm still trying to figure out how to fix.

We might need to create a /dev/cfam rather than going through that magic sysfs "raw" file, and I wouldn't mind using a single IDA so that all the devices below a given FSI slace (cfam itself, sbefifo, occ, ...) have the same "number".

Quoted 2 lines · Benjamin Herrenschmidt

We have everything that cronus needs and more than pdbg needs afaik :-)

drivers/fsi/fsi-scom.c +15 -0

+	for (retries = 0; retries < SCOM_MAX_RETRIES; retries++) {
+		rc = raw_get_scom(scom, value, addr, &status);
+		if (rc) {
+			/* Try resetting the bridge if FSI fails */
+			if (rc != -ENODEV && retries == 0) {
+				fsi_device_write(scom->fsi_dev, SCOM_FSI2PIB_RESET_REG,
+						 &dummy, sizeof(uint32_t));
+				rc = -EBUSY;
+			} else
+				return rc;
+		} else
+			rc = handle_fsi2pib_status(scom, status);
+		if (rc && rc != -EBUSY)
+			break;
+		if (rc == 0) {

Quoted 8 lines

That said, cronus does a bunch of other stupid things that I'm still trying to figure out how to fix.

We might need to create a /dev/cfam rather than going through that magic sysfs "raw" file, and I wouldn't mind using a single IDA so that all the devices below a given FSI slace (cfam itself, sbefifo, occ, ...) have the same "number".

Also while we're at reworking how all this is exposed to our broken userspace, I wouldn't mind putting all these dev entries under a directory, if I can figure out how to do that (I haven't really looked yet).

/dev/fsi/{cfamN,sbefifoN,occN, ...} and possibly similar by-id and by- path that other subsystems use, so we have something more deterministic than the "random number" crap we do now.

We can always keep hacks to do symlinks in our kernel tree until we have converted all our userspace users.

We currently control the only userspace users of this, so now is a good time to cleanup how we expose things. This won't always be the case.

Cheers,
Ben.

Quoted 3 lines · AEST Benjamin Herrenschmidt

Quoted 2 lines · Benjamin Herrenschmidt

We have everything that cronus needs and more than pdbg needs afaik :-)

Yep, has what we need and more (such as put scom under mask and indirect scom). Only other useful thing would be repeated getsom/putscom operations (eg. read the same scom address n times) as they would help with ADU access which can do autoincrement or scanscom (although we should just use the scan engine directly for that so not a big issue).

Quoted 20 lines

drivers/fsi/fsi-scom.c +20 -0

+	for (retries = 0; retries < SCOM_MAX_RETRIES; retries++) {
+		rc = raw_get_scom(scom, value, addr, &status);
+		if (rc) {
+			/* Try resetting the bridge if FSI fails */
+			if (rc != -ENODEV && retries == 0) {
+				fsi_device_write(scom->fsi_dev, SCOM_FSI2PIB_RESET_REG,
+						 &dummy, sizeof(uint32_t));
+				rc = -EBUSY;
+			} else
+				return rc;
+		} else
+			rc = handle_fsi2pib_status(scom, status);
+		if (rc && rc != -EBUSY)
+			break;
+		if (rc == 0) {
+			rc = handle_pib_status(scom,
+					       (status & SCOM_STATUS_PIB_RESP_MASK)
+					       >> SCOM_STATUS_PIB_RESP_SHIFT);
+			if (rc && rc != -EBUSY)
+				break;

drivers/fsi/fsi-scom.c static struct list_head scom_devices; · line 336

Quoted 5 lines

drivers/fsi/fsi-scom.c +5 -0

+		}
+		if (rc == 0)
+			break;
+		msleep(1);
+	}

The rc handling above took me a little while to grok but I didn't come up with a cleaner alternative and I think it's correct.

- Alistair

Quoted 28 lines

Quoted 8 lines

That said, cronus does a bunch of other stupid things that I'm still trying to figure out how to fix.

We might need to create a /dev/cfam rather than going through that magic sysfs "raw" file, and I wouldn't mind using a single IDA so that all the devices below a given FSI slace (cfam itself, sbefifo, occ, ...) have the same "number".

Also while we're at reworking how all this is exposed to our broken userspace, I wouldn't mind putting all these dev entries under a directory, if I can figure out how to do that (I haven't really looked yet).

/dev/fsi/{cfamN,sbefifoN,occN, ...} and possibly similar by-id and by- path that other subsystems use, so we have something more deterministic than the "random number" crap we do now.

We can always keep hacks to do symlinks in our kernel tree until we have converted all our userspace users.

We currently control the only userspace users of this, so now is a good time to cleanup how we expose things. This won't always be the case.

Cheers,
Ben.

Quoted 31 lines · Alistair Popple

Quoted 3 lines · AEST Benjamin Herrenschmidt

Quoted 2 lines · Benjamin Herrenschmidt

We have everything that cronus needs and more than pdbg needs afaik :-)

Yep, has what we need and more (such as put scom under mask and indirect scom). Only other useful thing would be repeated getsom/putscom operations (eg. read the same scom address n times) as they would help with ADU access which can do autoincrement or scanscom (although we should just use the scan engine directly for that so not a big issue).

Quoted 20 lines

drivers/fsi/fsi-scom.c +20 -0

+	for (retries = 0; retries < SCOM_MAX_RETRIES; retries++) {
+		rc = raw_get_scom(scom, value, addr, &status);
+		if (rc) {
+			/* Try resetting the bridge if FSI fails */
+			if (rc != -ENODEV && retries == 0) {
+				fsi_device_write(scom->fsi_dev, SCOM_FSI2PIB_RESET_REG,
+						 &dummy, sizeof(uint32_t));
+				rc = -EBUSY;
+			} else
+				return rc;
+		} else
+			rc = handle_fsi2pib_status(scom, status);
+		if (rc && rc != -EBUSY)
+			break;
+		if (rc == 0) {
+			rc = handle_pib_status(scom,
+					       (status & SCOM_STATUS_PIB_RESP_MASK)
+					       >> SCOM_STATUS_PIB_RESP_SHIFT);
+			if (rc && rc != -EBUSY)
+				break;

Quoted 8 lines

drivers/fsi/fsi-scom.c static struct list_head scom_devices; · line 336

Quoted 5 lines

drivers/fsi/fsi-scom.c +5 -0

+		}
+		if (rc == 0)
+			break;
+		msleep(1);
+	}

The rc handling above took me a little while to grok but I didn't come up with a cleaner alternative and I think it's correct.

Ack-by or Reviewed-by tag pls ? :-)

Cheers,
Ben.

Quoted 32 lines

- Alistair

Quoted 28 lines

Quoted 8 lines

That said, cronus does a bunch of other stupid things that I'm still trying to figure out how to fix.

We might need to create a /dev/cfam rather than going through that magic sysfs "raw" file, and I wouldn't mind using a single IDA so that all the devices below a given FSI slace (cfam itself, sbefifo, occ, ...) have the same "number".

Also while we're at reworking how all this is exposed to our broken userspace, I wouldn't mind putting all these dev entries under a directory, if I can figure out how to do that (I haven't really looked yet).

/dev/fsi/{cfamN,sbefifoN,occN, ...} and possibly similar by-id and by- path that other subsystems use, so we have something more deterministic than the "random number" crap we do now.

We can always keep hacks to do symlinks in our kernel tree until we have converted all our userspace users.

We currently control the only userspace users of this, so now is a good time to cleanup how we expose things. This won't always be the case.

Cheers,
Ben.