Skip to contentSkip to conversation

Introduce i3c device userspace interface

RFC Superseded Vitor Soares 4 participants 26 messages Jan 29, 2020 – Feb 17, 2020 I3C

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#4 lore raw

RFC v2 2/4 i3c: master: export i3c_bus_type symbol

In reply to Vitor Soares

Export i3c_bus_type symbol so i3cdev can register a notifier chain for i3c bus.

---
 drivers/i3c/master.c | 1 +
 1 file changed, 1 insertion(+)
drivers/i3c/master.c +1 -0
index 8a0ba34..21c4372 100644
@@ -321,6 +321,7 @@ struct bus_type i3c_bus_type = {
 	.probe = i3c_device_probe,
 	.remove = i3c_device_remove,
 };
+EXPORT_SYMBOL_GPL(i3c_bus_type);
 
 static enum i3c_addr_slot_status
 i3c_bus_get_addr_slot_status(struct i3c_bus *bus, u16 addr)
Signature · 1 line
2.7.4
#6 lore raw

RFC v2 4/4 i3c: add i3cdev module to expose i3c dev in /dev

In reply to Vitor Soares

This patch adds user mode support to I3C SDR transfers.

The module is based on i2c-dev.c with the follow features:
- expose on /dev the i3c devices dynamically based on if they have a device driver bound.
- Dynamically allocate the char device Major number.

---
 drivers/i3c/Kconfig             |  15 ++
 drivers/i3c/Makefile            |   1 +
 drivers/i3c/i3cdev.c            | 429 ++++++++++++++++++++++++++++++++++++++++
 include/uapi/linux/i3c/i3cdev.h |  38 ++++
 4 files changed, 483 insertions(+)
 create mode 100644 drivers/i3c/i3cdev.c
 create mode 100644 include/uapi/linux/i3c/i3cdev.h
drivers/i3c/Kconfig +15 -0
index 30a4415..0164276 100644
@@ -20,5 +20,20 @@ menuconfig I3C
 	  will be called i3c.
 
 if I3C
+
+config I3CDEV
+	tristate "I3C device interface"
+	depends on I3C
+	help
+	  Say Y here to use i3c-* device files, usually found in the /dev
+	  directory on your system.  They make it possible to have user-space
+	  programs use the I3C devices.
+
+	  This support is also available as a module.  If so, the module
+	  will be called i3cdev.
+
+	  Note that this application programming interface is EXPERIMENTAL
+	  and hence SUBJECT TO CHANGE WITHOUT NOTICE while it stabilizes.
+
 source "drivers/i3c/master/Kconfig"
 endif # I3C
drivers/i3c/Makefile +1 -0
index 11982ef..606d422 100644
@@ -1,4 +1,5 @@
 # SPDX-License-Identifier: GPL-2.0
 i3c-y				:= device.o master.o
 obj-$(CONFIG_I3C)		+= i3c.o
+obj-$(CONFIG_I3CDEV)		+= i3cdev.o
 obj-$(CONFIG_I3C)		+= master/
drivers/i3c/i3cdev.c new file +429 -0

429 lines not on this page · raw message

include/uapi/linux/i3c/i3cdev.h new file +38 -0
new file mode 100644
index 0000000..0897313
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/*
+ * Copyright (c) 2019 Synopsys, Inc. and/or its affiliates.
+ *
+ * Author: Vitor Soares <[email protected]>
+ */
+
+#ifndef _UAPI_I3C_DEV_H_
+#define _UAPI_I3C_DEV_H_
+
+#include <linux/types.h>
+#include <linux/ioctl.h>
+
+/* IOCTL commands */
+#define I3C_DEV_IOC_MAGIC	0x07
+
+/**
+ * struct i3c_ioc_priv_xfer - I3C SDR ioctl private transfer
+ * @data: Holds pointer to userspace buffer with transmit data.
+ * @len: Length of data buffer buffers, in bytes.
+ * @rnw: encodes the transfer direction. true for a read, false for a write
+ */
+struct i3c_ioc_priv_xfer {
+	__u64 data;
+	__u16 len;
+	__u8 rnw;
+	__u8 pad[5];
+};
+
+
+#define I3C_PRIV_XFER_SIZE(N)	\
+	((((sizeof(struct i3c_ioc_priv_xfer)) * (N)) < (1 << _IOC_SIZEBITS)) \
+	? ((sizeof(struct i3c_ioc_priv_xfer)) * (N)) : 0)
+
+#define I3C_IOC_PRIV_XFER(N)	\
+	_IOC(_IOC_READ|_IOC_WRITE, I3C_DEV_IOC_MAGIC, 30, I3C_PRIV_XFER_SIZE(N))
+
+#endif
Signature · 1 line
2.7.4
#5 lore raw

RFC v2 3/4 i3c: master: add i3c_for_each_dev helper

In reply to Vitor Soares

Introduce i3c_for_each_dev(), an i3c device iterator for use by i3cdev.

---
 drivers/i3c/internals.h |  1 +
 drivers/i3c/master.c    | 12 ++++++++++++
 2 files changed, 13 insertions(+)
drivers/i3c/internals.h +1 -0
index bc062e8..a6deedf 100644
@@ -24,4 +24,5 @@ int i3c_dev_enable_ibi_locked(struct i3c_dev_desc *dev);
 int i3c_dev_request_ibi_locked(struct i3c_dev_desc *dev,
 			       const struct i3c_ibi_setup *req);
 void i3c_dev_free_ibi_locked(struct i3c_dev_desc *dev);
+int i3c_for_each_dev(void *data, int (*fn)(struct device *, void *));
 #endif /* I3C_INTERNAL_H */
drivers/i3c/master.c +12 -0
index 21c4372..8e22da2 100644
@@ -2640,6 +2640,18 @@ void i3c_dev_free_ibi_locked(struct i3c_dev_desc *dev)
 	dev->ibi = NULL;
 }
 
+int i3c_for_each_dev(void *data, int (*fn)(struct device *, void *))
+{
+	int res;
+
+	mutex_lock(&i3c_core_lock);
+	res = bus_for_each_dev(&i3c_bus_type, NULL, data, fn);
+	mutex_unlock(&i3c_core_lock);
+
+	return res;
+}
+EXPORT_SYMBOL_GPL(i3c_for_each_dev);
+
 static int __init i3c_init(void)
 {
 	return bus_register(&i3c_bus_type);
Signature · 1 line
2.7.4
#0 lore raw

v2 cover Introduce i3c device userspace interface

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

- For now, the module exposes i3c devices without device driver on dev/i3c-<bus>-<pid>, but we can change the path to dev/bus/i3c/<bus>-<pid> or dev/i3c/<bus>-<pid>.

- As in the i2c subsystem, here it is exposed the i3c_priv_xfer to userspace. I tried to use a dedicated structure as in spidev but I don't see any obvious advantage.

- Since the i3c API only exposes i3c_priv_xfer to devices, for now, the module just makes use of one ioctl(). This can change in the future with the introduction hdr commands or by the need of exposing some CCC commands to the device API (private contract between master-slave). Regarding the i3c device info, some information is already available through sysfs. We can add more device attributes to expose more information or add a dedicated ioctl() request for that purpose or both.

- Similar to i2c, I have also created a tool that you can find in [2] for testing purposes. If you have some time available I would appreciate your feedback about it as well.

[1] https://lkml.org/lkml/2018/11/15/853
[2] https://github.com/vitor-soares-snps/i3c-tools.git

Changes in v2:
Use IDR api for minor numbering
Modify ioctl struct
Fix SPDX license

Vitor Soares (4):
i3c: master: export i3c_masterdev_type
i3c: master: export i3c_bus_type symbol
i3c: master: add i3c_for_each_dev helper
i3c: add i3cdev module to expose i3c dev in /dev

 drivers/i3c/Kconfig             |  15 ++
 drivers/i3c/Makefile            |   1 +
 drivers/i3c/i3cdev.c            | 429 ++++++++++++++++++++++++++++++++++++++++
 drivers/i3c/internals.h         |   2 +
 drivers/i3c/master.c            |  16 +-
 include/uapi/linux/i3c/i3cdev.h |  38 ++++
 6 files changed, 500 insertions(+), 1 deletion(-)
 create mode 100644 drivers/i3c/i3cdev.c
 create mode 100644 include/uapi/linux/i3c/i3cdev.h
Signature · 1 line
2.7.4
#1 lore raw

RFC v2 1/4 i3c: master: export i3c_masterdev_type

Exporte i3c_masterdev_type so i3cdev module can verify if an i3c device is a master.

---
 drivers/i3c/internals.h | 1 +
 drivers/i3c/master.c    | 3 ++-
 2 files changed, 3 insertions(+), 1 deletion(-)
drivers/i3c/internals.h +1 -0
index 86b7b44..bc062e8 100644
@@ -11,6 +11,7 @@
 #include <linux/i3c/master.h>
 
 extern struct bus_type i3c_bus_type;
+extern const struct device_type i3c_masterdev_type;
 
 void i3c_bus_normaluse_lock(struct i3c_bus *bus);
 void i3c_bus_normaluse_unlock(struct i3c_bus *bus);
drivers/i3c/master.c +2 -1
index 7f8f896..8a0ba34 100644
@@ -523,9 +523,10 @@ static void i3c_masterdev_release(struct device *dev)
 	of_node_put(dev->of_node);
 }
 
-static const struct device_type i3c_masterdev_type = {
+const struct device_type i3c_masterdev_type = {
 	.groups	= i3c_masterdev_groups,
 };
+EXPORT_SYMBOL_GPL(i3c_masterdev_type);
 
 static int i3c_bus_set_mode(struct i3c_bus *i3cbus, enum i3c_bus_mode mode,
 			    unsigned long max_i2c_scl_rate)
Signature · 1 line
2.7.4
#7 lore raw

Re: [RFC v2 4/4] i3c: add i3cdev module to expose i3c dev in /dev

In reply to Vitor Soares
Quoted 6 lines · Vitor Soares
drivers/i3c/i3cdev.c new file +5 -0
+
+struct i3cdev_data {
+       struct list_head list;
+       struct i3c_device *i3c;
+       struct cdev cdev;

drivers/i3c/i3cdev.c line 38

Quoted 10 lines
drivers/i3c/i3cdev.c new file +10 -0
+       struct device *dev;
+       int id;
+};
+
+static DEFINE_IDA(i3cdev_ida);
+static dev_t i3cdev_number;
+#define I3C_MINORS 16 /* 16 I3C devices supported for now */
+
+static LIST_HEAD(i3cdev_list);
+static DEFINE_SPINLOCK(i3cdev_list_lock);

Please try to avoid arbitrarily limiting the number of devices you support.

Searching through the list feels a little clumsy. If the i3c user interface is supposed to become a standard feature of the subsystem, it would seem appropriate to put a pointer into the device to simplify the lookup, or just embed the cdev inside of i3c_device.

Quoted 7 lines
drivers/i3c/i3cdev.c new file +7 -0
+static int
+i3cdev_do_priv_xfer(struct i3c_device *dev, struct i3c_ioc_priv_xfer *xfers,
+                   unsigned int nxfers)
+{
+       struct i3c_priv_xfer *k_xfers;
+       u8 **data_ptrs;
+       int i, ret = 0;

drivers/i3c/i3cdev.c line 160

Quoted 10 lines
drivers/i3c/i3cdev.c new file +10 -0
+
+       k_xfers = kcalloc(nxfers, sizeof(*k_xfers), GFP_KERNEL);
+       if (!k_xfers)
+               return -ENOMEM;
+
+       data_ptrs = kcalloc(nxfers, sizeof(*data_ptrs), GFP_KERNEL);
+       if (!data_ptrs) {
+               ret = -ENOMEM;
+               goto err_free_k_xfer;
+       }

Maybe use a combined allocation to simplify the error handling?

drivers/i3c/i3cdev.c line 193

Quoted 8 lines
drivers/i3c/i3cdev.c new file +7 -0
+       for (i = 0; i < nxfers; i++) {
+               data_ptrs[i] = memdup_user((const u8 __user *)
+                                          (uintptr_t)xfers[i].data,
+                                          xfers[i].len);
 
+               if (xfers[i].rnw) {
+                       if (copy_to_user((void __user *)(uintptr_t)xfers[i].data,
+                                        data_ptrs[i], xfers[i].len))

Use u64_to_user_ptr() here.

Quoted 20 lines
drivers/i3c/i3cdev.c new file +20 -0
+
+static struct i3c_ioc_priv_xfer *
+i3cdev_get_ioc_priv_xfer(unsigned int cmd, struct i3c_ioc_priv_xfer *u_xfers,
+                        unsigned int *nxfers)
+{
+       u32 tmp = _IOC_SIZE(cmd);
+
+       if ((tmp % sizeof(struct i3c_ioc_priv_xfer)) != 0)
+               return ERR_PTR(-EINVAL);
+
+       *nxfers = tmp / sizeof(struct i3c_ioc_priv_xfer);
+       if (*nxfers == 0)
+               return NULL;
+
+       return memdup_user(u_xfers, tmp);
+}
+
+static int
+i3cdev_ioc_priv_xfer(struct i3c_device *i3c, unsigned int cmd,
+                    struct i3c_ioc_priv_xfer *u_xfers)

drivers/i3c/i3cdev.c line 235

Quoted 10 lines
drivers/i3c/i3cdev.c new file +10 -0
+{
+       struct i3c_ioc_priv_xfer *k_xfers;
+       unsigned int nxfers;
+       int ret;
+
+       k_xfers = i3cdev_get_ioc_priv_xfer(cmd, u_xfers, &nxfers);
+       if (IS_ERR_OR_NULL(k_xfers))
+               return PTR_ERR(k_xfers);
+
+       ret = i3cdev_do_priv_xfer(i3c, k_xfers, nxfers);

The IS_ERR_OR_NULL() usage looks suspicious. It's generally better to avoid interfaces that require this. What does it mean to return NULL from i3cdev_get_ioc_priv_xfer() and turn that into success? Could you handle this condition in the caller instead, or turn it into an error?

drivers/i3c/i3cdev.c line 403

Quoted 7 lines
drivers/i3c/i3cdev.c new file +7 -0
+       /* Keep track of busses which have devices to add or remove later */
+       res = bus_register_notifier(&i3c_bus_type, &i3c_notifier);
+       if (res)
+               goto out_unreg_class;
+
+       /* Bind to already existing device without driver right away */
+       i3c_for_each_dev(NULL, i3cdev_attach);

The combination of the notifier and searching through the devices seems to be racy. What happens when a device appears just before or during the i3c_for_each_dev() traversal?

What happens when a driver attaches to a device that is currently transferring data on the user interface?

Is there any guarantee that the notifiers for attach and detach are serialized?

Quoted 10 lines
include/uapi/linux/i3c/i3cdev.h new file +10 -0
+/**
+ * struct i3c_ioc_priv_xfer - I3C SDR ioctl private transfer
+ * @data: Holds pointer to userspace buffer with transmit data.
+ * @len: Length of data buffer buffers, in bytes.
+ * @rnw: encodes the transfer direction. true for a read, false for a write
+ */
+struct i3c_ioc_priv_xfer {
+       __u64 data;
+       __u16 len;
+       __u8 rnw;

include/uapi/linux/i3c/i3cdev.h line 36

Quoted 10 lines
include/uapi/linux/i3c/i3cdev.h new file +10 -0
+       __u8 pad[5];
+};
+
+
+#define I3C_PRIV_XFER_SIZE(N)  \
+       ((((sizeof(struct i3c_ioc_priv_xfer)) * (N)) < (1 << _IOC_SIZEBITS)) \
+       ? ((sizeof(struct i3c_ioc_priv_xfer)) * (N)) : 0)
+
+#define I3C_IOC_PRIV_XFER(N)   \
+       _IOC(_IOC_READ|_IOC_WRITE, I3C_DEV_IOC_MAGIC, 30, I3C_PRIV_XFER_SIZE(N))

This looks like a reasonable ioctl definition, avoiding the usual problems with compat mode etc.

Arnd

#8 lore raw

RE: [RFC v2 4/4] i3c: add i3cdev module to expose i3c dev in /dev

Hi Arnd,

From: Arnd Bergmann <[email protected]>
Date: Wed, Jan 29, 2020 at 14:30:56

Quoted 11 lines · Vitor Soares
Quoted 10 lines
drivers/i3c/i3cdev.c new file +9 -0
+
+struct i3cdev_data {
+       struct list_head list;
+       struct i3c_device *i3c;
+       struct cdev cdev;
+       struct device *dev;
+       int id;
+};
+
Quoted 8 lines

drivers/i3c/i3cdev.c line 38

Quoted 6 lines
drivers/i3c/i3cdev.c new file +6 -0
+static DEFINE_IDA(i3cdev_ida);
+static dev_t i3cdev_number;
+#define I3C_MINORS 16 /* 16 I3C devices supported for now */
+
+static LIST_HEAD(i3cdev_list);
+static DEFINE_SPINLOCK(i3cdev_list_lock);

Please try to avoid arbitrarily limiting the number of devices you support.

Should I use all minors range instead?

Quoted 4 lines

Searching through the list feels a little clumsy. If the i3c user interface is supposed to become a standard feature of the subsystem, it would seem appropriate to put a pointer into the device to simplify the lookup,

Do you mean i3c->dev ?

Quoted 2 lines

or
just embed the cdev inside of i3c_device.

I would prefer to have a pointer in i3c_device for i3cdev_data, but I see others using it in drvdata.

Quoted 12 lines
Quoted 12 lines
drivers/i3c/i3cdev.c new file +11 -0
+static int
+i3cdev_do_priv_xfer(struct i3c_device *dev, struct i3c_ioc_priv_xfer *xfers,
+                   unsigned int nxfers)
+{
+       struct i3c_priv_xfer *k_xfers;
+       u8 **data_ptrs;
+       int i, ret = 0;
+
+       k_xfers = kcalloc(nxfers, sizeof(*k_xfers), GFP_KERNEL);
+       if (!k_xfers)
+               return -ENOMEM;
Quoted 8 lines

drivers/i3c/i3cdev.c line 160

Quoted 6 lines
drivers/i3c/i3cdev.c new file +6 -0
+
+       data_ptrs = kcalloc(nxfers, sizeof(*data_ptrs), GFP_KERNEL);
+       if (!data_ptrs) {
+               ret = -ENOMEM;
+               goto err_free_k_xfer;
+       }

Maybe use a combined allocation to simplify the error handling?

Could you please provide an example?

Quoted 11 lines

drivers/i3c/i3cdev.c line 193

Quoted 9 lines
drivers/i3c/i3cdev.c new file +7 -0
+       for (i = 0; i < nxfers; i++) {
+               data_ptrs[i] = memdup_user((const u8 __user *)
+                                          (uintptr_t)xfers[i].data,
+                                          xfers[i].len);
 
+               if (xfers[i].rnw) {
+                       if (copy_to_user((void __user *)(uintptr_t)xfers[i].data,
+                                        data_ptrs[i], xfers[i].len))

Use u64_to_user_ptr() here.

You are right, it wasn't available went I did it.

Quoted 29 lines
Quoted 29 lines
drivers/i3c/i3cdev.c new file +28 -0
+
+static struct i3c_ioc_priv_xfer *
+i3cdev_get_ioc_priv_xfer(unsigned int cmd, struct i3c_ioc_priv_xfer *u_xfers,
+                        unsigned int *nxfers)
+{
+       u32 tmp = _IOC_SIZE(cmd);
+
+       if ((tmp % sizeof(struct i3c_ioc_priv_xfer)) != 0)
+               return ERR_PTR(-EINVAL);
+
+       *nxfers = tmp / sizeof(struct i3c_ioc_priv_xfer);
+       if (*nxfers == 0)
+               return NULL;
+
+       return memdup_user(u_xfers, tmp);
+}
+
+static int
+i3cdev_ioc_priv_xfer(struct i3c_device *i3c, unsigned int cmd,
+                    struct i3c_ioc_priv_xfer *u_xfers)
+{
+       struct i3c_ioc_priv_xfer *k_xfers;
+       unsigned int nxfers;
+       int ret;
+
+       k_xfers = i3cdev_get_ioc_priv_xfer(cmd, u_xfers, &nxfers);
+       if (IS_ERR_OR_NULL(k_xfers))
+               return PTR_ERR(k_xfers);
Quoted 8 lines

drivers/i3c/i3cdev.c line 235

Quoted 2 lines
drivers/i3c/i3cdev.c new file +2 -0
+
+       ret = i3cdev_do_priv_xfer(i3c, k_xfers, nxfers);

The IS_ERR_OR_NULL() usage looks suspicious. It's generally better to avoid interfaces that require this. What does it mean to return NULL from i3cdev_get_ioc_priv_xfer() and turn that into success? Could you handle this condition in the caller instead, or turn it into an error?

In both cases something is not correct. I will turn both conditions to return ERR_PTR(-EINVAL) and them just check if (IS_ERR(k_xfer)).

Quoted 9 lines

drivers/i3c/i3cdev.c line 403

Quoted 8 lines
drivers/i3c/i3cdev.c new file +7 -0
+       /* Keep track of busses which have devices to add or remove later */
+       res = bus_register_notifier(&i3c_bus_type, &i3c_notifier);
+       if (res)
+               goto out_unreg_class;
+
+       /* Bind to already existing device without driver right away */
+       i3c_for_each_dev(NULL, i3cdev_attach);
Quoted 3 lines

The combination of the notifier and searching through the devices seems to be racy. What happens when a device appears just before or during the i3c_for_each_dev() traversal?

The i3c core is locked during this phase.

Quoted 4 lines

What happens when a driver attaches to a device that is currently transferring data on the user interface?

It may lost references for inode and file. I need to guarantee there no tranfer going on during the detach.
Do you have any suggestion?

Quoted 3 lines

Is there any guarantee that the notifiers for attach and detach are serialized?

Sorry I didn't get this part.

Quoted 15 lines
Quoted 15 lines
include/uapi/linux/i3c/i3cdev.h new file +15 -0
+/**
+ * struct i3c_ioc_priv_xfer - I3C SDR ioctl private transfer
+ * @data: Holds pointer to userspace buffer with transmit data.
+ * @len: Length of data buffer buffers, in bytes.
+ * @rnw: encodes the transfer direction. true for a read, false for a write
+ */
+struct i3c_ioc_priv_xfer {
+       __u64 data;
+       __u16 len;
+       __u8 rnw;
+       __u8 pad[5];
+};
+
+
+#define I3C_PRIV_XFER_SIZE(N)  \
Quoted 8 lines

include/uapi/linux/i3c/i3cdev.h line 36

Quoted 5 lines
include/uapi/linux/i3c/i3cdev.h new file +5 -0
+       ((((sizeof(struct i3c_ioc_priv_xfer)) * (N)) < (1 << _IOC_SIZEBITS)) \
+       ? ((sizeof(struct i3c_ioc_priv_xfer)) * (N)) : 0)
+
+#define I3C_IOC_PRIV_XFER(N)   \
+       _IOC(_IOC_READ|_IOC_WRITE, I3C_DEV_IOC_MAGIC, 30, I3C_PRIV_XFER_SIZE(N))

This looks like a reasonable ioctl definition, avoiding the usual problems with compat mode etc.

Do you think I should add more reserved fields for future?

Quoted 7 lines

Arnd

_______________________________________________
linux-i3c mailing list
[email protected]
https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.infradead.org_mailman_listinfo_linux-2Di3c&d=DwICAg&c=DPL6_X_6JkXFx7AXWqB0tg&r=qVuU64u9x77Y0Kd0PhDK_lpxFgg6PK9PateHwjb_DY0&m=pv8xU_wOpDLOkwdQiuBDso73EKvNPX2jXLtBHDVWRFo&s=S-Tesk8Hi3Ok6y9d_ysocHXGt2dmnn-WcM0BxurcDdQ&e=

Thanks for your comments 😊

Best regards,
Vitor Soares

#9 lore raw

Re: [RFC v2 4/4] i3c: add i3cdev module to expose i3c dev in /dev

Quoted 26 lines · Vitor Soares

Hi Arnd,

From: Arnd Bergmann <[email protected]>
Date: Wed, Jan 29, 2020 at 14:30:56

Quoted 11 lines · Vitor Soares
Quoted 10 lines
drivers/i3c/i3cdev.c new file +9 -0
+
+struct i3cdev_data {
+       struct list_head list;
+       struct i3c_device *i3c;
+       struct cdev cdev;
+       struct device *dev;
+       int id;
+};
+
Quoted 8 lines

drivers/i3c/i3cdev.c line 38

Quoted 6 lines
drivers/i3c/i3cdev.c new file +6 -0
+static DEFINE_IDA(i3cdev_ida);
+static dev_t i3cdev_number;
+#define I3C_MINORS 16 /* 16 I3C devices supported for now */
+
+static LIST_HEAD(i3cdev_list);
+static DEFINE_SPINLOCK(i3cdev_list_lock);

Please try to avoid arbitrarily limiting the number of devices you support.

Quoted 1 line

Should I use all minors range instead?

Yes, I'm fairly sure that if you use a dynamic major number, there is no downside in using all of them.

Quoted 5 lines
Quoted 3 lines

Searching through the list feels a little clumsy. If the i3c user interface is supposed to become a standard feature of the subsystem, it would seem appropriate to put a pointer into the device to simplify the lookup,

Do you mean i3c->dev ?

I was thinking you could add another member in i3c_device, next to ->dev.

Quoted 5 lines
Quoted 2 lines

or
just embed the cdev inside of i3c_device.

I would prefer to have a pointer in i3c_device for i3cdev_data, but I see others using it in drvdata.

Ok, I think drvdata should work, but you should check that this is correct when the device goes back between being bound to a device driver and used through the chardev.

Quoted 21 lines
Quoted 12 lines
Quoted 12 lines
drivers/i3c/i3cdev.c new file +11 -0
+static int
+i3cdev_do_priv_xfer(struct i3c_device *dev, struct i3c_ioc_priv_xfer *xfers,
+                   unsigned int nxfers)
+{
+       struct i3c_priv_xfer *k_xfers;
+       u8 **data_ptrs;
+       int i, ret = 0;
+
+       k_xfers = kcalloc(nxfers, sizeof(*k_xfers), GFP_KERNEL);
+       if (!k_xfers)
+               return -ENOMEM;
Quoted 8 lines

drivers/i3c/i3cdev.c line 160

Quoted 6 lines
drivers/i3c/i3cdev.c new file +6 -0
+
+       data_ptrs = kcalloc(nxfers, sizeof(*data_ptrs), GFP_KERNEL);
+       if (!data_ptrs) {
+               ret = -ENOMEM;
+               goto err_free_k_xfer;
+       }

Maybe use a combined allocation to simplify the error handling?

Quoted 1 line

Could you please provide an example?

Something like

k_xfers = kcalloc(nxfers, sizeof(*k_xfers) +

sizeof(*data_ptrs), GFP_KERNEL);
       data_ptrs = (void *)k_xfers + (nxfers, sizeof(*k_xfers));

This would need a comment to explain the pointer math, but the resulting object code is slightly simpler.

Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
drivers/i3c/i3cdev.c new file +7 -0
+       /* Keep track of busses which have devices to add or remove later */
+       res = bus_register_notifier(&i3c_bus_type, &i3c_notifier);
+       if (res)
+               goto out_unreg_class;
+
+       /* Bind to already existing device without driver right away */
+       i3c_for_each_dev(NULL, i3cdev_attach);
Quoted 5 lines
Quoted 3 lines

The combination of the notifier and searching through the devices seems to be racy. What happens when a device appears just before or during the i3c_for_each_dev() traversal?

The i3c core is locked during this phase.

Ok.

Quoted 7 lines
Quoted 3 lines

What happens when a driver attaches to a device that is currently transferring data on the user interface?

It may lost references for inode and file. I need to guarantee there no tranfer going on during the detach.
Do you have any suggestion?

If the notifier is blocking, you could hold another mutex during the transfer I think.

Quoted 5 lines
Quoted 3 lines

Is there any guarantee that the notifiers for attach and detach are serialized?

Sorry I didn't get this part.

I think you answered this above: if the i3c code is locked while calling the notifier, this cannot happen.

Quoted 24 lines
Quoted 15 lines
Quoted 15 lines
include/uapi/linux/i3c/i3cdev.h new file +15 -0
+/**
+ * struct i3c_ioc_priv_xfer - I3C SDR ioctl private transfer
+ * @data: Holds pointer to userspace buffer with transmit data.
+ * @len: Length of data buffer buffers, in bytes.
+ * @rnw: encodes the transfer direction. true for a read, false for a write
+ */
+struct i3c_ioc_priv_xfer {
+       __u64 data;
+       __u16 len;
+       __u8 rnw;
+       __u8 pad[5];
+};
+
+
+#define I3C_PRIV_XFER_SIZE(N)  \
Quoted 8 lines

include/uapi/linux/i3c/i3cdev.h line 36

Quoted 5 lines
include/uapi/linux/i3c/i3cdev.h new file +5 -0
+       ((((sizeof(struct i3c_ioc_priv_xfer)) * (N)) < (1 << _IOC_SIZEBITS)) \
+       ? ((sizeof(struct i3c_ioc_priv_xfer)) * (N)) : 0)
+
+#define I3C_IOC_PRIV_XFER(N)   \
+       _IOC(_IOC_READ|_IOC_WRITE, I3C_DEV_IOC_MAGIC, 30, I3C_PRIV_XFER_SIZE(N))

This looks like a reasonable ioctl definition, avoiding the usual problems with compat mode etc.

Quoted 1 line

Do you think I should add more reserved fields for future?

No, what I meant is that I like it the way it is.

Arnd

#10 lore raw

RE: [RFC v2 4/4] i3c: add i3cdev module to expose i3c dev in /dev

From: Arnd Bergmann <[email protected]>
Date: Wed, Jan 29, 2020 at 19:39:41

Quoted 43 lines · Vitor Soares
Quoted 26 lines

Hi Arnd,

From: Arnd Bergmann <[email protected]>
Date: Wed, Jan 29, 2020 at 14:30:56

Quoted 11 lines · Vitor Soares
Quoted 10 lines
drivers/i3c/i3cdev.c new file +9 -0
+
+struct i3cdev_data {
+       struct list_head list;
+       struct i3c_device *i3c;
+       struct cdev cdev;
+       struct device *dev;
+       int id;
+};
+
Quoted 8 lines

drivers/i3c/i3cdev.c line 38

Quoted 6 lines
drivers/i3c/i3cdev.c new file +6 -0
+static DEFINE_IDA(i3cdev_ida);
+static dev_t i3cdev_number;
+#define I3C_MINORS 16 /* 16 I3C devices supported for now */
+
+static LIST_HEAD(i3cdev_list);
+static DEFINE_SPINLOCK(i3cdev_list_lock);

Please try to avoid arbitrarily limiting the number of devices you support.

Quoted 1 line

Should I use all minors range instead?

Yes, I'm fairly sure that if you use a dynamic major number, there is no downside in using all of them.

Quoted 5 lines
Quoted 3 lines

Searching through the list feels a little clumsy. If the i3c user interface is supposed to become a standard feature of the subsystem, it would seem appropriate to put a pointer into the device to simplify the lookup,

Do you mean i3c->dev ?

I was thinking you could add another member in i3c_device, next to ->dev.

Quoted 3 lines
Quoted 2 lines

or
just embed the cdev inside of i3c_device.

Quoted 6 lines
Quoted 2 lines

I would prefer to have a pointer in i3c_device for i3cdev_data, but I see others using it in drvdata.

Ok, I think drvdata should work, but you should check that this is correct when the device goes back between being bound to a device driver and used through the chardev.

I changed the detach to be done in BUS_NOTIFY_BIND_DRIVER.

Quoted 24 lines
Quoted 22 lines
Quoted 13 lines
Quoted 13 lines
drivers/i3c/i3cdev.c new file +11 -0
+static int
+i3cdev_do_priv_xfer(struct i3c_device *dev, struct i3c_ioc_priv_xfer *xfers,
+                   unsigned int nxfers)
+{
+       struct i3c_priv_xfer *k_xfers;
+       u8 **data_ptrs;
+       int i, ret = 0;
+
+       k_xfers = kcalloc(nxfers, sizeof(*k_xfers), GFP_KERNEL);
+       if (!k_xfers)
+               return -ENOMEM;
Quoted 8 lines

drivers/i3c/i3cdev.c line 160

Quoted 6 lines
drivers/i3c/i3cdev.c new file +6 -0
+
+       data_ptrs = kcalloc(nxfers, sizeof(*data_ptrs), GFP_KERNEL);
+       if (!data_ptrs) {
+               ret = -ENOMEM;
+               goto err_free_k_xfer;
+       }

Maybe use a combined allocation to simplify the error handling?

Quoted 1 line

Could you please provide an example?

Quoted 8 lines

Something like

k_xfers = kcalloc(nxfers, sizeof(*k_xfers) +

sizeof(*data_ptrs), GFP_KERNEL);
       data_ptrs = (void *)k_xfers + (nxfers, sizeof(*k_xfers));

This would need a comment to explain the pointer math, but the resulting object code is slightly simpler.

As we have nferxs, there is no problem to allocate k_xfers more than needed, right?

Quoted 21 lines
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
drivers/i3c/i3cdev.c new file +7 -0
+       /* Keep track of busses which have devices to add or remove later */
+       res = bus_register_notifier(&i3c_bus_type, &i3c_notifier);
+       if (res)
+               goto out_unreg_class;
+
+       /* Bind to already existing device without driver right away */
+       i3c_for_each_dev(NULL, i3cdev_attach);
Quoted 5 lines
Quoted 3 lines

The combination of the notifier and searching through the devices seems to be racy. What happens when a device appears just before or during the i3c_for_each_dev() traversal?

The i3c core is locked during this phase.

Ok.

Quoted 4 lines
Quoted 3 lines

What happens when a driver attaches to a device that is currently transferring data on the user interface?

Quoted 6 lines
Quoted 3 lines

It may lost references for inode and file. I need to guarantee there no tranfer going on during the detach.
Do you have any suggestion?

If the notifier is blocking, you could hold another mutex during the transfer I think.

A mutex during the transfer will solve the detach issue, I doing some tests but even with the change to BUS_NOTIFY_BIND_DRIVER I'm not sure if it can race with driver probe function.

Quoted 10 lines
Quoted 6 lines
Quoted 4 lines

Is there any guarantee that the notifiers for attach and detach are serialized?

Sorry I didn't get this part.

I think you answered this above: if the i3c code is locked while calling the notifier, this cannot happen.

The i3c code is only locked during the i3c_for_each_dev().

Quoted 24 lines
Quoted 24 lines
Quoted 23 lines
Quoted 10 lines
include/uapi/linux/i3c/i3cdev.h new file +10 -0
+/**
+ * struct i3c_ioc_priv_xfer - I3C SDR ioctl private transfer
+ * @data: Holds pointer to userspace buffer with transmit data.
+ * @len: Length of data buffer buffers, in bytes.
+ * @rnw: encodes the transfer direction. true for a read, false for a write
+ */
+struct i3c_ioc_priv_xfer {
+       __u64 data;
+       __u16 len;
+       __u8 rnw;

include/uapi/linux/i3c/i3cdev.h line 36

Quoted 10 lines
include/uapi/linux/i3c/i3cdev.h new file +10 -0
+       __u8 pad[5];
+};
+
+
+#define I3C_PRIV_XFER_SIZE(N)  \
+       ((((sizeof(struct i3c_ioc_priv_xfer)) * (N)) < (1 << _IOC_SIZEBITS)) \
+       ? ((sizeof(struct i3c_ioc_priv_xfer)) * (N)) : 0)
+
+#define I3C_IOC_PRIV_XFER(N)   \
+       _IOC(_IOC_READ|_IOC_WRITE, I3C_DEV_IOC_MAGIC, 30, I3C_PRIV_XFER_SIZE(N))

This looks like a reasonable ioctl definition, avoiding the usual problems with compat mode etc.

Quoted 5 lines
Quoted 1 line

Do you think I should add more reserved fields for future?

No, what I meant is that I like it the way it is.

Arnd

Best regards,
Vitor Soares

#12 lore raw
In reply to Vitor Soares

Hello Vitor,

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 4 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

Quoted 8 lines

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Regards,

Boris

#2 lore raw

Re: [RFC v2 1/4] i3c: master: export i3c_masterdev_type

In reply to Vitor Soares
Quoted 26 lines · On Wed, 29 Jan 2020 13:17:32 +0100 Vitor Soares <[email protected]>…

Exporte i3c_masterdev_type so i3cdev module can verify if an i3c device is a master.

---
 drivers/i3c/internals.h | 1 +
 drivers/i3c/master.c    | 3 ++-
 2 files changed, 3 insertions(+), 1 deletion(-)
drivers/i3c/internals.h +1 -0
index 86b7b44..bc062e8 100644
@@ -11,6 +11,7 @@
 #include <linux/i3c/master.h>
 
 extern struct bus_type i3c_bus_type;
+extern const struct device_type i3c_masterdev_type;
 
 void i3c_bus_normaluse_lock(struct i3c_bus *bus);
 void i3c_bus_normaluse_unlock(struct i3c_bus *bus);
drivers/i3c/master.c +0 -0
index 7f8f896..8a0ba34 100644

drivers/i3c/master.c static void i3c_masterdev_release(struct device *dev) · line 529

Quoted 8 lines
drivers/i3c/master.c +2 -1
@@ -523,9 +523,10 @@ static void i3c_masterdev_release(struct device *dev)
 	of_node_put(dev->of_node);
 }
 
-static const struct device_type i3c_masterdev_type = {
+const struct device_type i3c_masterdev_type = {
 	.groups	= i3c_masterdev_groups,
 };
+EXPORT_SYMBOL_GPL(i3c_masterdev_type);

No need to export the symbol, removing the static and adding the definition to internal.h should work just fine (i3c.o contains both master.o and device.o).

Quoted 3 lines
drivers/i3c/master.c +0 -0
 static int i3c_bus_set_mode(struct i3c_bus *i3cbus, enum i3c_bus_mode mode,
 			    unsigned long max_i2c_scl_rate)
#3 lore raw

Re: [RFC v2 1/4] i3c: master: export i3c_masterdev_type

Quoted 32 lines · On Mon, 17 Feb 2020 15:56:23 +0100 Boris Brezillon…
Quoted 29 lines · On Wed, 29 Jan 2020 13:17:32 +0100 Vitor Soares <[email protected]>…

Exporte i3c_masterdev_type so i3cdev module can verify if an i3c device is a master.

---
 drivers/i3c/internals.h | 1 +
 drivers/i3c/master.c    | 3 ++-
 2 files changed, 3 insertions(+), 1 deletion(-)
drivers/i3c/internals.h +1 -0
index 86b7b44..bc062e8 100644
@@ -11,6 +11,7 @@
 #include <linux/i3c/master.h>
 
 extern struct bus_type i3c_bus_type;
+extern const struct device_type i3c_masterdev_type;
 
 void i3c_bus_normaluse_lock(struct i3c_bus *bus);
 void i3c_bus_normaluse_unlock(struct i3c_bus *bus);
drivers/i3c/master.c +0 -0
index 7f8f896..8a0ba34 100644
@@ -523,9 +523,10 @@ static void i3c_masterdev_release(struct device *dev)
 	of_node_put(dev->of_node);
 }
 
Quoted 9 lines

drivers/i3c/master.c static void i3c_masterdev_release(struct device *dev) · line 529

Quoted 5 lines
drivers/i3c/master.c +2 -1
-static const struct device_type i3c_masterdev_type = {
+const struct device_type i3c_masterdev_type = {
 	.groups	= i3c_masterdev_groups,
 };
+EXPORT_SYMBOL_GPL(i3c_masterdev_type);  

No need to export the symbol, removing the static and adding the definition to internal.h should work just fine (i3c.o contains both master.o and device.o).

Hm, my bad. Looks like i3cdev is a separate module/driver. If that's the case, it should not have direct access to internals.h. I see 2 options here:

1/ make the i3cdev logic part of the core
2/ provide helpers to find devices by type

But maybe none of that is needed if you let userspace bind i3c devices to the i3cdev driver.

Quoted 5 lines
Quoted 4 lines
drivers/i3c/master.c +0 -0
 static int i3c_bus_set_mode(struct i3c_bus *i3cbus, enum i3c_bus_mode mode,
 			    unsigned long max_i2c_scl_rate)  
#13 lore raw
In reply to Boris Brezillon
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver. The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Arnd

#11 lore raw

Re: [RFC v2 4/4] i3c: add i3cdev module to expose i3c dev in /dev

In reply to Vitor Soares
Quoted 11 lines · On Wed, 29 Jan 2020 13:17:35 +0100 Vitor Soares <[email protected]>…
include/uapi/linux/i3c/i3cdev.h new file +5 -0
new file mode 100644
index 0000000..0897313
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+/*
+ * Copyright (c) 2019 Synopsys, Inc. and/or its affiliates.
+ *
+ * Author: Vitor Soares <[email protected]>

include/uapi/linux/i3c/i3cdev.h line 15

Quoted 10 lines
include/uapi/linux/i3c/i3cdev.h new file +10 -0
+ */
+
+#ifndef _UAPI_I3C_DEV_H_
+#define _UAPI_I3C_DEV_H_
+
+#include <linux/types.h>
+#include <linux/ioctl.h>
+
+/* IOCTL commands */
+#define I3C_DEV_IOC_MAGIC	0x07

I guess you already made sure there was no collision with other magic values.

Quoted 11 lines
include/uapi/linux/i3c/i3cdev.h new file +11 -0
+
+/**
+ * struct i3c_ioc_priv_xfer - I3C SDR ioctl private transfer
+ * @data: Holds pointer to userspace buffer with transmit data.
+ * @len: Length of data buffer buffers, in bytes.
+ * @rnw: encodes the transfer direction. true for a read, false for a write
+ */
+struct i3c_ioc_priv_xfer {
+	__u64 data;
+	__u16 len;
+	__u8 rnw;

include/uapi/linux/i3c/i3cdev.h line 36

Quoted 10 lines
include/uapi/linux/i3c/i3cdev.h new file +10 -0
+	__u8 pad[5];
+};
+
+
+#define I3C_PRIV_XFER_SIZE(N)	\
+	((((sizeof(struct i3c_ioc_priv_xfer)) * (N)) < (1 << _IOC_SIZEBITS)) \
+	? ((sizeof(struct i3c_ioc_priv_xfer)) * (N)) : 0)
+
+#define I3C_IOC_PRIV_XFER(N)	\
+	_IOC(_IOC_READ|_IOC_WRITE, I3C_DEV_IOC_MAGIC, 30, I3C_PRIV_XFER_SIZE(N))

Any reason for starting at 30 instead of 0x0 or 0x1?

Also, this ioctl definition is a bit unusual. Most of the time, when we want to pass an array of elements we pass a struct that contains the number of entries in this array, and a pointer to the array itself.

struct i3cdev_priv_xfers {
	__u64 nxfers;
	__u64 xfers; /*Use u64_to_user_ptr() to get the __user pointer*/
};
Quoted 2 lines
include/uapi/linux/i3c/i3cdev.h new file +2 -0
+
+#endif
#23 lore raw
In reply to Boris Brezillon

Hi Boris,

From: Boris Brezillon <[email protected]>
Date: Mon, Feb 17, 2020 at 14:51:41

Quoted 17 lines

Hello Vitor,

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

My idea was to expose every device to user-space by default so we can test them without a driver (more or less the i2c use case) but as we agreed during the i3c subsystem only expose devices that doesn't have device driver.
I considered to have a uevent but to expose the devices by default it would required something generic, what I didn't figure out and tend to follow the i2c-dev module.

With this current approach even if a device has a driver we can unbind it through the Sysfs and have access from user space which I found useful for debug.

Quoted 4 lines

Regards,

Boris

#14 lore raw
In reply to Arnd Bergmann
Quoted 26 lines · On Mon, 17 Feb 2020 16:06:45 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 3 lines

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver.

Hm, isn't that what udev is supposed to do anyway? Remember that I3C devices expose a manufacturer and part-id (which are similar to the USB vendor and product ids), so deciding when an I3C device should be bound to the i3cdev driver should be fairly easy, and that's a per-device decision anyway.

Quoted 3 lines

The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Well, I'm more worried about the extra churn this auto-binding logic might create for the common 'on-demand driver loading' use case. At first, there's no driver matching a specific device, but userspace might load one based on the uevents it receives. With the current approach, that means we'd first have to unbind the device before loading the driver. AFAICT, no other subsystem does that.

#24 lore raw
In reply to Vitor Soares
Quoted 31 lines · On Mon, 17 Feb 2020 15:32:55 +0000 Vitor Soares <[email protected]>…

Hi Boris,

From: Boris Brezillon <[email protected]>
Date: Mon, Feb 17, 2020 at 14:51:41

Quoted 17 lines

Hello Vitor,

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 4 lines

My idea was to expose every device to user-space by default so we can test them without a driver (more or less the i2c use case) but as we agreed during the i3c subsystem only expose devices that doesn't have device driver.

Those that do not have a driver *yet*. What if i3cdev is compiled-in and other I3C drivers are enabled as modules? When the device is discovered at boot time is will be automatically bound to the i3cdev driver since no matching drivers are available at that point. But the end user probably expects this device to be attached to the in kernel driver.

Quoted 3 lines

I considered to have a uevent but to expose the devices by default it would required something generic, what I didn't figure out and tend to follow the i2c-dev module.

Well, I3C and I2C/SPI are quite different in this regard. I2C dev exposes the whole bus, and SPI devs don't have a standard way to uniquely identify the device connected on the bus (unless it has a dedicated compatible for DT-based boards). In that case it might make sense to auto-bind all orphan devs to the default spidev driver, though I'm not entirely sure it's really necessary since that's probably a per-board decision, and having a udev rule matching the bus/CS would make sense too.

Quoted 4 lines

With this current approach even if a device has a driver we can unbind it through the Sysfs and have access from user space which I found useful for debug.

That's my point. We're making the default 'on-demand driver loading' use case more complicated to ease the less common 'userspace driver' use case.

#15 lore raw
In reply to Boris Brezillon

Hi,

From: Boris Brezillon <[email protected]>
Date: Mon, Feb 17, 2020 at 15:36:22

Quoted 43 lines
Quoted 26 lines · On Mon, 17 Feb 2020 16:06:45 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 3 lines

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver.

Hm, isn't that what udev is supposed to do anyway? Remember that I3C devices expose a manufacturer and part-id (which are similar to the USB vendor and product ids), so deciding when an I3C device should be bound to the i3cdev driver should be fairly easy, and that's a per-device decision anyway.

Quoted 3 lines

The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Quoted 6 lines

Well, I'm more worried about the extra churn this auto-binding logic might create for the common 'on-demand driver loading' use case. At first, there's no driver matching a specific device, but userspace might load one based on the uevents it receives. With the current approach, that means we'd first have to unbind the device before loading the driver. AFAICT, no other subsystem does that.

I'm about to finish v3 (today or tomorrow) and I think I fixed all concerns rise during v2. I would like you to see that version before any change.

Best regards,
Vitor Soares

#16 lore raw
Quoted 55 lines · Vitor Soares

Hi,

From: Boris Brezillon <[email protected]>
Date: Mon, Feb 17, 2020 at 15:36:22

Quoted 43 lines
Quoted 26 lines · On Mon, 17 Feb 2020 16:06:45 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 3 lines

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver.

Hm, isn't that what udev is supposed to do anyway? Remember that I3C devices expose a manufacturer and part-id (which are similar to the USB vendor and product ids), so deciding when an I3C device should be bound to the i3cdev driver should be fairly easy, and that's a per-device decision anyway.

Quoted 3 lines

The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Quoted 6 lines

Well, I'm more worried about the extra churn this auto-binding logic might create for the common 'on-demand driver loading' use case. At first, there's no driver matching a specific device, but userspace might load one based on the uevents it receives. With the current approach, that means we'd first have to unbind the device before loading the driver. AFAICT, no other subsystem does that.

Quoted 3 lines

I'm about to finish v3 (today or tomorrow) and I think I fixed all concerns rise during v2. I would like you to see that version before any change.

Why are there so many "RFC" series here? I treat "RFC" as "I don't really like this patch but I'm throwing it out there for others to look at if they care". No RFC should ever go on beyond a v1 as obviously you think this is good enough by now, right?

Also, I almost never review RFC patches, we have enough "real" patches to review as it is :)

thanks,

greg k-h

#17 lore raw

From: gregkh <[email protected]>
Date: Mon, Feb 17, 2020 at 16:03:45

Quoted 56 lines · Vitor Soares
Quoted 55 lines

Hi,

From: Boris Brezillon <[email protected]>
Date: Mon, Feb 17, 2020 at 15:36:22

Quoted 49 lines
Quoted 26 lines · On Mon, 17 Feb 2020 16:06:45 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 3 lines

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver.

Hm, isn't that what udev is supposed to do anyway? Remember that I3C devices expose a manufacturer and part-id (which are similar to the USB vendor and product ids), so deciding when an I3C device should be bound to the i3cdev driver should be fairly easy, and that's a per-device decision anyway.

Quoted 3 lines

The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Well, I'm more worried about the extra churn this auto-binding logic might create for the common 'on-demand driver loading' use case. At first, there's no driver matching a specific device, but userspace might load one based on the uevents it receives. With the current approach, that means we'd first have to unbind the device before loading the driver. AFAICT, no other subsystem does that.

Quoted 7 lines
Quoted 3 lines

I'm about to finish v3 (today or tomorrow) and I think I fixed all concerns rise during v2. I would like you to see that version before any change.

Why are there so many "RFC" series here? I treat "RFC" as "I don't really like this patch but I'm throwing it out there for others to look at if they care". No RFC should ever go on beyond a v1

My bad ☹.

Quoted 2 lines

as obviously you
think this is good enough by now, right?

Yes and I really appreciate the feedback that leads me to this v3.

Quoted 7 lines

Also, I almost never review RFC patches, we have enough "real" patches to review as it is :)

thanks,

greg k-h

Thanks for your advice.

Best regards,
Vitor Soares

#21 lore raw
In reply to Boris Brezillon
Quoted 41 lines · Boris Brezillon
Quoted 26 lines · On Mon, 17 Feb 2020 16:06:45 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 3 lines

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver.

Hm, isn't that what udev is supposed to do anyway? Remember that I3C devices expose a manufacturer and part-id (which are similar to the USB vendor and product ids), so deciding when an I3C device should be bound to the i3cdev driver should be fairly easy, and that's a per-device decision anyway.

Quoted 3 lines

The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Quoted 6 lines

Well, I'm more worried about the extra churn this auto-binding logic might create for the common 'on-demand driver loading' use case. At first, there's no driver matching a specific device, but userspace might load one based on the uevents it receives. With the current approach, that means we'd first have to unbind the device before loading the driver. AFAICT, no other subsystem does that.

As I understand it, this is handled by the patches: when a new device shows up, this triggers the creation of the userspace interface and also the event that leads to the kernel driver to get loaded. If there is a kernel driver for the device, that should still load and bind to the device, at which point the user space interface will go away again.

This may waste CPU cycles for first creating and then destroying the user space interface, but I don't see how it requires extra work. If it does require manual configuration or unbinding, that would indeed be a bad design.

Arnd

#18 lore raw
In reply to Vitor Soares
Quoted 54 lines · On Mon, 17 Feb 2020 15:55:08 +0000 Vitor Soares <[email protected]>…

Hi,

From: Boris Brezillon <[email protected]>
Date: Mon, Feb 17, 2020 at 15:36:22

Quoted 49 lines
Quoted 26 lines · On Mon, 17 Feb 2020 16:06:45 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 3 lines

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver.

Hm, isn't that what udev is supposed to do anyway? Remember that I3C devices expose a manufacturer and part-id (which are similar to the USB vendor and product ids), so deciding when an I3C device should be bound to the i3cdev driver should be fairly easy, and that's a per-device decision anyway.

Quoted 3 lines

The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Well, I'm more worried about the extra churn this auto-binding logic might create for the common 'on-demand driver loading' use case. At first, there's no driver matching a specific device, but userspace might load one based on the uevents it receives. With the current approach, that means we'd first have to unbind the device before loading the driver. AFAICT, no other subsystem does that.

Okay, I have clearly not read the code carefully enough. I thought you were declaring a new i3c_device_driver and were manually binding all orphan devices to this driver. Looks like the solution is more subtle than that, and i3cdevs are actually subdevices that are automatically created/removed when the I3C device is unbound/bound. That means the 'on-demand driver loading' logic is not impacted by this new layer. I'm still not convinced this is needed (I expect i3cdev to be used mostly for experiment, and in that case, having a udev rule, or manually binding the device to the i3cdev driver shouldn't be a problem). I'm also not sure what happens if the device is still used when i3cdev_detach() is called, can transfers still be done after the device is attached to its in-kernel driver?

#19 lore raw
Quoted 11 lines · Boris Brezillon

On Mon, 17 Feb 2020 15:55:08 +0000 Vitor Soares <[email protected]> wrote:

Okay, I have clearly not read the code carefully enough. I thought you were declaring a new i3c_device_driver and were manually binding all orphan devices to this driver. Looks like the solution is more subtle than that, and i3cdevs are actually subdevices that are automatically created/removed when the I3C device is unbound/bound. That means the 'on-demand driver loading' logic is not impacted by this new layer. I'm still not convinced this is needed (I expect i3cdev to be used mostly for experiment, and in that case, having a udev rule, or manually binding the device to the i3cdev driver shouldn't be a problem).

I'm fairly sure it's not needed, other approaches could be used to provide user space access, but it's not clear if any other way is better either. It also took me a while to figure out what is going on when I read the code.

One thought that I had was that this could be better integrated into the core, with user space being there implicitly through sysfs rather than a /dev file.

Quoted 3 lines

I'm also not sure what happens if the device is still used when i3cdev_detach() is called, can transfers still be done after the device is attached to its in-kernel driver?

I think this is still an open issue that I also pointed out. The driver binding/unbinding and user space access definitely needs to be properly serialized, whichever method is used to implement the user access.

Arnd

#22 lore raw
In reply to Arnd Bergmann
Quoted 50 lines · On Mon, 17 Feb 2020 17:19:57 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 41 lines · Boris Brezillon
Quoted 26 lines · On Mon, 17 Feb 2020 16:06:45 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 15 lines · Boris Brezillon

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Quoted 3 lines

I think that would require manual configuration to ensure that the correct set of devices get bound to either the userspace driver or an in-kernel driver.

Hm, isn't that what udev is supposed to do anyway? Remember that I3C devices expose a manufacturer and part-id (which are similar to the USB vendor and product ids), so deciding when an I3C device should be bound to the i3cdev driver should be fairly easy, and that's a per-device decision anyway.

Quoted 3 lines

The method from the current patch series is more complicated, but it means that any device can be accessed by the user space driver as long as it's not already owned by a kernel driver.

Quoted 6 lines

Well, I'm more worried about the extra churn this auto-binding logic might create for the common 'on-demand driver loading' use case. At first, there's no driver matching a specific device, but userspace might load one based on the uevents it receives. With the current approach, that means we'd first have to unbind the device before loading the driver. AFAICT, no other subsystem does that.

Quoted 5 lines

As I understand it, this is handled by the patches: when a new device shows up, this triggers the creation of the userspace interface and also the event that leads to the kernel driver to get loaded. If there is a kernel driver for the device, that should still load and bind to the device, at which point the user space interface will go away again.

Yep, that's what I figured after having a closer look at the code.

Quoted 5 lines

This may waste CPU cycles for first creating and then destroying the user space interface, but I don't see how it requires extra work. If it does require manual configuration or unbinding, that would indeed be a bad design.

To be honest, I had something less invasive in mind. Something closer to what spidev provides (a driver that can expose I3C devices to userspace when explicitly requested). I see now that the USB subsystem does something similar to what's done here, but I'm wondering if it's really worth it in the I3C case. As I said in my previous reply, I expect i3cdev to be used when experimenting or when kernel-space driver is not an option (licensing/security issues).

#20 lore raw
In reply to Arnd Bergmann
Quoted 14 lines · On Mon, 17 Feb 2020 17:31:08 +0100 Arnd Bergmann <[email protected]> wrote:
Quoted 11 lines · Boris Brezillon

On Mon, 17 Feb 2020 15:55:08 +0000 Vitor Soares <[email protected]> wrote:

Okay, I have clearly not read the code carefully enough. I thought you were declaring a new i3c_device_driver and were manually binding all orphan devices to this driver. Looks like the solution is more subtle than that, and i3cdevs are actually subdevices that are automatically created/removed when the I3C device is unbound/bound. That means the 'on-demand driver loading' logic is not impacted by this new layer. I'm still not convinced this is needed (I expect i3cdev to be used mostly for experiment, and in that case, having a udev rule, or manually binding the device to the i3cdev driver shouldn't be a problem).

Quoted 8 lines

I'm fairly sure it's not needed, other approaches could be used to provide user space access, but it's not clear if any other way is better either. It also took me a while to figure out what is going on when I read the code.

One thought that I had was that this could be better integrated into the core, with user space being there implicitly through sysfs rather than a /dev file.

Hm, doing I3C transfers through sysfs sounds a bit odd. sysfs entries are most of the time exposing human readable/writable stuff (plain text data, that is).

Quoted 9 lines
Quoted 4 lines

I'm also not sure what happens if the device is still used when i3cdev_detach() is called, can transfers still be done after the device is attached to its in-kernel driver?

I think this is still an open issue that I also pointed out. The driver binding/unbinding and user space access definitely needs to be properly serialized, whichever method is used to implement the user access.

Well, going for the spidev approach would solve that and make the whole implementation more straightforward and less invasive (no notifier, no need to expose internal device types to this i3cdev driver, no concurrency issues, ...). We can always revisit this solution if it proves to be to limited and we feel the need for a kernel-driven i3cdev auto-binding logic, but I doubt we'll ever be in that position since udev can handle that for us, and if we start having actual userspace drivers (which is not the case yet), I expect distros to add the relevant udev rules to take care of this auto-bind.

#25 lore raw
In reply to Boris Brezillon
Quoted 17 lines · On Mon, 17 Feb 2020 15:51:41 +0100 Boris Brezillon…

Hello Vitor,

Sorry for taking so long to reply, and thanks for working on that topic.

Quoted 10 lines · On Wed, 29 Jan 2020 13:17:31 +0100 Vitor Soares <[email protected]>…

For today there is no way to use i3c devices from user space and the introduction of such API will help developers during the i3c device or i3c host controllers development.

The i3cdev module is highly based on i2c-dev and yet I tried to address the concerns raised in [1].

NOTES:
- The i3cdev dynamically request an unused major number.

Quoted 8 lines
Quoted 2 lines

- The i3c devices are dynamically exposed/removed from dev/ folder based on if they have a device driver bound to it.

May I ask why you need to automatically bind devices to the i3cdev driver when they don't have a driver matching the device id loaded/compiled-in? If we get the i3c subsystem to generate proper uevents we should be able to load the i3cdev module and bind the device to this driver using a udev rule.

Hm, looks like we already send a proper MODALIAS [1], so we can actually implement this auto-bind to i3cdev in udev (fall back on i3cdev load+bind if no matching module is found).

[1]https://elixir.bootlin.com/linux/latest/source/drivers/i3c/master.c#L269