Re: [PATCH] usb:serial: Add Fintek F81532/534 driver

[Ji-Ze Hong (Peter Hong)]

Hi Alan,

One Thousand Gnomes 於 2016/7/29 下午 08:48 寫道:

O

+static int f81534_set_normal_register(struct usb_device *dev, u16 reg, u8 data)
+{
+   size_t count = F81534_USB_MAX_RETRY;
+   int status;
+   u8 *tmp;
+
+   tmp = kmalloc(sizeof(u8), GFP_KERNEL);
+   if (!tmp)
+   return -ENOMEM;


You end up doing huge numbers of tiny allocation and frees in some of the
code paths. I think it would be better to allocate them at a higher level
as they are not that cheap on CPU time.


+static int f81534_read_data(struct usb_serial *usbserial, u32 address,
+   size_t size, unsigned char *buf)
+{


Is a particularly good example - you do 4 mallocs plus two per byte of
data.



I'll re-factor the newest V9 patch with your suggestion. To malloc a
byte within usb_serial privates, and make a mutex to protect it.

I'll send it as V10 when I tested it.

Thanks for your suggestion.
--
With Best Regards,
Peter Hong
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] usb:serial: Add Fintek F81532/534 driver

[One Thousand Gnomes]

O
> +static int f81534_set_normal_register(struct usb_device *dev, u16 reg, u8 
> data)
> +{
> + size_t count = F81534_USB_MAX_RETRY;
> + int status;
> + u8 *tmp;
> +
> + tmp = kmalloc(sizeof(u8), GFP_KERNEL);
> + if (!tmp)
> + return -ENOMEM;

You end up doing huge numbers of tiny allocation and frees in some of the
code paths. I think it would be better to allocate them at a higher level
as they are not that cheap on CPU time.

> +static int f81534_read_data(struct usb_serial *usbserial, u32 address,
> + size_t size, unsigned char *buf)
> +{

Is a particularly good example - you do 4 mallocs plus two per byte of
data.

Alan
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] usb:serial: Add Fintek F81532/534 driver

[Peter Hung]

Hi,

Ji-Ze Hong (Peter Hong) 於 2016/5/31 上午 09:33 寫道:

This driver is for Fintek F81532/F81534 USB to Serial Ports IC.



Sorry, I forgot to change the mail title for "PATCH V9". I'll resend a
patch with title "PATCH V9".

Thanks
--
With Best Regards,
Peter Hung
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] usb:serial: Add Fintek F81532/534 driver

[Ji-Ze Hong (Peter Hong)]

This driver is for Fintek F81532/F81534 USB to Serial Ports IC.

F81532 spec:
https://drive.google.com/file/d/0B8vRwwYO7aMFOTRRMmhWQVNvajQ/view?
usp=sharing

F81534 spec:
https://drive.google.com/file/d/0B8vRwwYO7aMFV29pQWJqbVBNc00/view?
usp=sharing

Features:
1. F81532 is 1-to-2 & F81534 is 1-to-4 serial ports IC
2. Support Baudrate from B50 to B115200.

Signed-off-by: Ji-Ze Hong (Peter Hong) 
---
Changelog:
v9
1. Remove lots of code to make more generic for F81532/534. e.g.,
   high baud rate support, RS485/422 mode switch, most of GPIO
   control and internal storage write functional.
2. Change f81534_tiocmget() MSR delay from schedule_timeout_killable()
   to wait_for_completion_killable_timeout(). This IC will delayed
   receiving MSR change when doing loop-back test e.g., BurnInTest.
   We'll reset completion "msr_done" in f81534_update_mctrl(). If we
   changed MCR, the next f81534_tiocmget() will delay for 20ms or
   continue with new MSR arrived.
3. Fix for non-zero buffer allocated in f81534_setup_ports(). It'll
   make device malfunctional with incorrect tx length for other ports.

v8
1. Remove driver mode GPIOLIB & RS485 control support, the driver will
   only load GPIO/UART Mode when driver attach() & port_probe().
2. Add more documents for 3 generation IC with f81534_calc_num_ports().
3. Simplify the GPIO register structure "f81534_pin_control".
4. Change all counter type from int to size_t.
5. Change some failed message with failed: "status code" and remove all
   exclamation mark in messages.
6. Change all save blocks to block0 due to the driver is only used 1
   block (block0) to save data.
7. Change read MSR in open() instead of port_probe().
8. use GFP_ATOMIC kmalloc mode in write().
9. Maintain old style with 1 read URBs and 4 write URBs like mxuports.c
   I had tested with submit 4 read URBs, but it'll make some port freeze
   when doing BurnInTest Port test.

v7
1. Make all gpiolib function with #ifdef CONFIG_GPIOLIB marco block.
   Due to F81532/534 could run without gpiolib, we implements
   f81534_prepare_gpio()/f81534_release_gpio() always success without
   CONFIG_GPIOLIB.
2. Fix crash when receiving MSR change on driver load/unload. It's cause
   by f81534_process_read_urb() get read URB callback data, but port
   private data is not init complete or released. We solve with 2
   modifications.

   1. add null pointer check with f81534_process_read_urb(). We'll skip
  this report when port_priv = NULL.
   2. when "one" port f81534_port_remove() is called, kill the port-0
  read URB before kfree port_priv.

v6
1. Re-implement the write()/resume() function. Due to this device cant be
   suitable with generic write(), we'll do the submit write URB when
   write()/received tx empty/set_termios()/resume()
2. Logic/Phy Port mapping rewrite in f81534_port_probe() &
   f81534_phy_to_logic_port().
3. Introduced "Port Hide" function. Some customer use F81532 reference
   design for HW layout, but really use F81534 IC. We'll check
   F81534_PORT_CONF_DISABLE_PORT flag with in uart mode field to do
   port hide with port not used. It can be avoid end-user to use not
   layouted port.
4. The 4x3 output-only open-drain pins for F81532/534 is designed for
   control outer devices (with our EVB for examples, the 4 sets of pins
   are designed to control transceiver mode). So we decide to implement
   with gpiolib interface.
5. Add device vendor id with 0x2c42

v5
1. Change f81534_port_disable/enable() from H/W mode to S/W mode
   It'll skip all rx data when port is not opened.
2. Some function modifier add with static (Thanks for Paul Bolle)
3. It's will direct return when count=0 in f81534_write() to
   reduce spin_lock usage.

v4
1. clearify f81534_process_read_urb() with
   f81534_process_per_serial_block(). (referenced from mxuport.c)
2. We limited f81534_write() max tx kfifo with 124Bytes.
   Original subsystem is designed for auto tranmiting fifo data
   if available. But we must wait for tx_empty for next tx data
   (H/W design).

   With this kfifo size limit, we can use generic subsystem api with
   f81534_write(). When usb_serial_generic_write_start() called after
   first write URB complete, the fifo will no data. The generic
   subsystem of write will go to idle state. Until we received
   TX_EMPTY and release write spinlock, the fifo will fill max
   124Bytes by following f81534_write().

v3
1. Migrate read, write and some routine from custom code to usbserial
   subsystem callback function.
2. Use more defines to replece magic numbers to make it meaningful
3. Make more comments as document in source code.

v2
1. v1 version submit to staging tree, but