AM335x (SDK 8.0) gadgetfs

Simon Defayette5

After I use gadgetfs in my program, I see often the "Used by" count go to 2147483647 (0x7FFFFFFF) and then it's unsusable... I have to reboot each time the target. Does anyone knows why?

root@am335x-evm:~# lsmod

Module Size Used by
gadgetfs 15921 2147483647
musb_dsps 6979 0
musb_hdrc 30544 1 musb_dsps
udc_core 6884 2 musb_hdrc,gadgetfs
phy_am335x 1975 2
usb_common 1786 3 musb_dsps,musb_hdrc,udc_core
phy_am335x_control 2226 1 phy_am335x
musb_am335x 999 0

Thanks,

Simon Defayette

over 10 years ago

0 Bin Liu over 10 years ago

TI__Guru**** 168311 points

Simon,

Can you please give more detail how to reproduce the issue? I use www.linux-usb.org/.../usb.c to test gadgetfs but got spinlock lockup issue in kernel and not able to see the issue you reported.

0 Bin Liu over 10 years ago in reply to Bin Liu

TI__Guru**** 168311 points

Never mind. I am able to see the issue. I will keep you posted.

0 Simon Defayette5 over 10 years ago in reply to Bin Liu

Intellectual 365 points

I tried what it was proposed here :
bugzilla.kernel.org/show_bug.cgi

At the end I have the rigth refcount, but I can't open the file the second time I run my program. It tells me that the device is busy.

0 Bin Liu over 10 years ago in reply to Simon Defayette5

TI__Guru**** 168311 points

Simon,

Is it possible for you to try Felipe's kernel repo, which has latest USB drivers?

git://git.kernel.org/pub/scm/linux/kernel/git/balbi/usb.git, branch: testing/next.

I tested it after applied the refcount patch, it seems to me the kernel is ok, but the user space application usb.c is broken - the gadget exits during emuneration every time connecting to host.

0 Simon Defayette5 over 10 years ago in reply to Bin Liu

Intellectual 365 points

Bin,

I added the .owner = THIS_MODULES to dev_init_operations and ep_config_operations, but I'm not sure why it was removed before.

Also, I wasn't able to open the /dev/gadget/musb-hdrc file more than once without rebooting the target... this modifications (that is reincluded in the kernel 3.17) make this works :

2014-07-21 Merge tag 'usb-for-v3.17' of git://git.kernel.org/pub/scm/linux/kernel/git/balbi/usb into usb-next Greg Kroah-Hartman 1 -0/+2147

in function dev_release (struct inode *inode, struct file *fd) in driver/usb/gadget/inode.c

kfree (dev->buf);
dev->buf = NULL;

+ /* other endpoints were all decoupled from this device */
+ spin_lock_irq(&dev->lock);
+ dev->state = STATE_DEV_DISABLED;
+ spin_unlock_irq(&dev->lock);

put_dev (dev);

0 Bin Liu over 10 years ago in reply to Simon Defayette5

TI__Guru**** 168311 points

Simon,

You added those changes into SDK8.0 kernel? Does gadgetfs work now with the changes or you still have any issue? Is the transfer stop issue you mentioned in the earlier thread still happening?

0 Simon Defayette5 over 10 years ago in reply to Bin Liu

Intellectual 365 points

Now it is working.

1. I can open/close musb_hdrc multiple times without restarting the target. (the missing STATE_DEV_DISABLED)

2. The counter issue is fixed if we add those .owner = THIS_MODULES lines (why it was removed?)

3. The transfert stop issue that I saw in SDK 5.5.1 is not present in 8.0.0... the one I saw in 8.0.0 was my mistake.

0 Bin Liu over 10 years ago in reply to Simon Defayette5

TI__Guru**** 168311 points

Simon,

Simon Defayette5 said:
2. The counter issue is fixed if we add those .owner = THIS_MODULES lines (why it was removed

As in the link you provided - bugzilla.kernel.org/show_bug.cgi, it was removed by Al Viro's patch 3273097 "gadgetfs: don't bother with fops->owner".

0 Simon Defayette5 over 10 years ago in reply to Bin Liu

Intellectual 365 points

Yes,

I saw that comment, but it does not explain why it was not working... or why shouldn't we let those THIS_MODULE there. Seems to be related to the filesystem system calls, but I did not go deeper for now.

Processors

Processors forum

AM335x (SDK 8.0) gadgetfs