Re: [BUG v3.6.2, 3.7-rc1] Unable to rmmod any hid special driver

On Thu, 1 Nov 2012, Benjamin Tissoires wrote:

quoted

quoted

quoted

 ------------[ cut here ]------------
 WARNING: at lib/list_debug.c:59 __list_del_entry+0xa4/0xd0()
 Hardware name: 7470BN2
 list_del corruption. prev->next should be ffff880036f97178, but was 6b6b6b6b6b6b6b6b
 Modules linked in: af_packet rfcomm bnep tun iptable_mangle xt_DSCP nf_conntrack_ipv6 nf_defrag_ipv6 ip6table_filter ip6_tables xt
e nf_conntrack iptable_filter ip_tables x_tables cpufreq_conservative cpufreq_userspace cpufreq_powersave acpi_cpufreq mperf dm_mod
m_intel kvm iwldvm mac80211 snd_hda_codec_conexant microcode usbhid sg iwlwifi pcspkr snd_hda_intel cfg80211 lpc_ich i2c_i801 mfd_c
imer snd_page_alloc thinkpad_acpi rfkill wmi snd ac soundcore battery tpm_tis tpm tpm_bios autofs4 uhci_hcd ehci_hcd usbcore usb_co
button edd fan processor ata_generic thermal thermal_sys [last unloaded: hid_cypress]
 Pid: 4725, comm: hexdump Not tainted 3.7.0-rc3-00008-g35fd3dc #157
 Call Trace:
  [<ffffffff812f95c4>] ? __list_del_entry+0xa4/0xd0
  [<ffffffff812f95c4>] ? __list_del_entry+0xa4/0xd0
  [<ffffffff810452fa>] warn_slowpath_common+0x7a/0xb0
  [<ffffffff810453d1>] warn_slowpath_fmt+0x41/0x50
  [<ffffffff812f95c4>] __list_del_entry+0xa4/0xd0
  [<ffffffff812f9601>] list_del+0x11/0x40
  [<ffffffff81454a23>] hidraw_release+0x33/0x50
  [<ffffffff8118cd0a>] __fput+0xca/0x2b0
  [<ffffffff8118cf49>] ____fput+0x9/0x10
  [<ffffffff8106b7a1>] task_work_run+0xb1/0xe0
  [<ffffffff8104c81e>] do_exit+0x1ee/0x4c0
  [<ffffffff810adcad>] ? trace_hardirqs_on_caller+0x12d/0x1b0
  [<ffffffff8104cb37>] do_group_exit+0x47/0xc0
  [<ffffffff8104cbc2>] sys_exit_group+0x12/0x20
  [<ffffffff8158a529>] system_call_fastpath+0x16/0x1b
 ---[ end trace 3bec13493e086171 ]---

This is weird, on my setup (kernel 3.6.2 + Henrik's patches + Ratan's
patch + your fix), I don't have this warning.

My reproduction scenario was

- run hexdump on hidraw device driven by a special driver (hid-cypress in
  this case)
- let the device produce a few events
- rmmod hid-cypress while hexdump is still running

well, with the very same test (except I'm testing against
hid-multitouch), the warning doesn't show up on my configuration...

Perhaps it happens when the removed hidraw is actually the last one 
existing in the system?

-- 
Jiri Kosina
SUSE Labs