Linux: Hanging inside hid_close() #142

New issue

Closed

opened 2020-02-01 05:59:34 -06:00 by vozhyk-com · 3 comments

vozhyk-com commented

2020-02-01 05:59:34 -06:00

(Migrated from github.com)

For catching this hanging need just open device by hid_open() or hid_open_path() and close it by hid_close(), see simple example:

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include "hidapi.h"

int main(int argc, char* argv[])
{
hid_device *handle;
struct hid_device_info *devs, *cur_dev;

 if (hid_init())
      return -1; 
 int i;
 for (i =0 ; i < 10000; i++) { 
      handle = hid_open(0x22c9, 0x0901, NULL);
      if (!handle) {
           printf("unable to open device\n");
           return 1;
      }
      printf("iteration: %d, going to call hid_close() \n", i); 
      hid_close(handle);
      printf("iteration: %d, hid_close() done \n", i); 
 }   
 /* Free static HIDAPI objects. */
 hid_exit();
 return 0;

}

For catching this hanging need just open device by hid_open() or hid_open_path() and close it by hid_close(), see simple example: #include <stdio.h> #include <string.h> #include <stdlib.h> #include "hidapi.h" int main(int argc, char* argv[]) { hid_device *handle; struct hid_device_info *devs, *cur_dev; if (hid_init()) return -1; int i; for (i =0 ; i < 10000; i++) { handle = hid_open(0x22c9, 0x0901, NULL); if (!handle) { printf("unable to open device\n"); return 1; } printf("iteration: %d, going to call hid_close() \n", i); hid_close(handle); printf("iteration: %d, hid_close() done \n", i); } /* Free static HIDAPI objects. */ hid_exit(); return 0; }

Youw commented

2020-09-20 03:11:42 -05:00

(Migrated from github.com)

The fix suggested in #143 kind-of fixes the issue. Kind-of, because it constantly tries not to hit a deadlock, but a race condition is still there, it is a root cause, and we better fix a root cause, not symptoms.

@vozhyk-com would you be able to check a patch I prepared, with your device? https://github.com/libusb/hidapi/tree/libusb-hang

I wasn't able to reproduce the original issue, presumably because I don't have a device that would allow me to do so. I believe there is a race condition, that when hit, may lead to deadlock, when trying to close the device, in case if device submits transfers/interruptions frequently, and device is being closed exactly during successful handling of such transfer. The fix suggested in #143 kind-of fixes the issue. Kind-of, because it constantly tries not to hit a deadlock, but a race condition is still there, it is a root cause, and we better fix a root cause, not symptoms. @vozhyk-com would you be able to check a patch I prepared, with your device? https://github.com/libusb/hidapi/tree/libusb-hang

vozhyk-com commented

2020-09-26 14:35:11 -05:00

(Migrated from github.com)