[alsa-devel] ALSA: hda - Fix pointer leftover issues in hdac_acomp
Greg KH
gregkh at linuxfoundation.org
Fri Dec 22 09:35:36 CET 2017
On Fri, Dec 22, 2017 at 09:11:50AM +0100, Takashi Iwai wrote:
> On Fri, 22 Dec 2017 09:06:02 +0100,
> Greg KH wrote:
> >
> > On Fri, Dec 22, 2017 at 09:00:28AM +0100, Takashi Iwai wrote:
> > > On Fri, 22 Dec 2017 07:16:26 +0100,
> > > 岡本 幸大 wrote:
> > > >
> > > > Hello
> > > >
> > > > I run into an Oops while executing "cat /proc/iomem" on the latest 4.4 kernel.
> > > > See the log below:
> > > >
> > > > [ 55.945264] BUG: unable to handle kernel NULL pointer dereference at 0000000000000020
> > > > [ 55.945292] IP: [<ffffffff8f10d00e>] r_show+0x34/0xb4
> > > > [ 55.945324] PGD d7a39067 PUD cb3e9067 PMD 0
> > > > [ 55.945342] Oops: 0000 [#1] PREEMPT SMP
> > > > [ 55.945358] Modules linked in: snd_hda_codec_hdmi(E) dell_led(E) input_leds(E) joydev(E) snd_hda_codec_realtek(E) snd_hda_codec_generic(E) i915(E) snd_hda_intel(E) snd_hda_codec(E) snd_hda_core(E) snd_hwdep(E) snd_pcm(E) snd_seq_midi(E) snd_seq_midi_event(E) drm_kms_helper(E) snd_rawmidi(E) snd_seq(E) acpi_als(E) kfifo_buf(E) intel_rapl(E) snd_seq_device(E) snd_timer(E) drm(E) snd(E) industrialio(E) x86_pkg_temp_thermal(E) dell_wmi(E) sparse_keymap(E) dcdbas(E) intel_powerclamp(E) coretemp(E) serio_raw(E) wmi(E) mei_me(E) i2c_algo_bit(E) parport_pc(E) soundcore(E) fb_sys_fops(E) syscopyarea(E) intel_lpss_acpi(E) sysfillrect(E) mei(E) ppdev(E) intel_lpss(E) sysimgblt(E) mac_hid(E) 8250_fintek(E) i2c_hid(E) lp(E) parport(E) hid_generic(E) usbhid(E) hid(E) psmouse(E) r8169(E) mii(E) ahci(E) libahci(E) fjes(E)
> > > > [ 55.945630] CPU: 0 PID: 1781 Comm: cat Tainted: G E 4.4.92-cip11-pc-platform #1
> > > > [ 55.945655] Hardware name: Dell Inc. OptiPlex 3050/0W0CHX, BIOS 1.5.4 07/14/2017
> > > > [ 55.945679] task: ffff8800cb060400 ti: ffff8800c9910000 task.ti: ffff8800c9910000
> > > > [ 55.945701] RIP: 0010:[<ffffffff8f10d00e>] [<ffffffff8f10d00e>] r_show+0x34/0xb4
> > > > [ 55.945726] RSP: 0018:ffff8800c9913da8 EFLAGS: 00010297
> > > > [ 55.945743] RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000001000
> > > > [ 55.945765] RDX: ffffffff8fe42130 RSI: ffff8800d89844a8 RDI: ffff8800d9250dc0
> > > > [ 55.945786] RBP: ffff8800c9913dc8 R08: 0000000000000008 R09: 0000000000000001
> > > > [ 55.945807] R10: 0000000000000000 R11: 0000000000000201 R12: ffff8800d89844a8
> > > > [ 55.945828] R13: ffff8800d9250dc0 R14: ffff8800d9250dc0 R15: ffff880118719500
> > > > [ 55.945851] FS: 00007f12f3dce740(0000) GS:ffff88011dc00000(0000) knlGS:0000000000000000
> > > > [ 55.945876] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > [ 55.945894] CR2: 0000000000000020 CR3: 000000003f7e9000 CR4: 00000000003406f0
> > > > [ 55.945916] Stack:
> > > > [ 55.945923] ffffffff8fa2ffa0 0000000000000000 0000000000010000 ffff8800c9913f28
> > > > [ 55.945950] ffff8800c9913e30 ffffffff8f22e009 ffff8800d89844a8 000000000000063c
> > > > [ 55.945976] ffff8800d9250e00 0000000001ae0000 000000000000002d 000000000000002e
> > > > [ 55.946013] Call Trace:
> > > > [ 55.946025] [<ffffffff8f22e009>] seq_read+0x24b/0x317
> > > > [ 55.946043] [<ffffffff8f263c1a>] proc_reg_read+0x48/0x67
> > > > [ 55.946061] [<ffffffff8f263bd2>] ? proc_reg_write+0x67/0x67
> > > > [ 55.946080] [<ffffffff8f21011a>] __vfs_read+0x26/0xba
> > > > [ 55.946097] [<ffffffff8f37cb49>] ? security_file_permission+0x96/0xa3
> > > > [ 55.946118] [<ffffffff8f21070c>] ? rw_verify_area+0x7e/0xd2
> > > > [ 55.946135] [<ffffffff8f2107f8>] vfs_read+0x98/0x123
> > > > [ 55.946151] [<ffffffff8f21131d>] SyS_read+0x4e/0x89
> > > > [ 55.946169] [<ffffffff8f84dbae>] entry_SYSCALL_64_fastpath+0x12/0x71
> > > > [ 55.946188] Code: 41 55 41 54 53 50 49 89 fd 48 8b 97 80 00 00 00 49 89 f4 48 89 f0 48 81 7a 08 00 00 01 00 45 19 c0 31 db 41 83 e0 fc 41 83 c0 08 <48> 8b 40 20 48 39 d0 74 07 ff c3 83 fb 05 75 f0 49 8b 7d 78 ba
> > > > [ 55.946310] RIP [<ffffffff8f10d00e>] r_show+0x34/0xb4
> > > > [ 55.946328] RSP <ffff8800c9913da8>
> > > > [ 55.946339] CR2: 0000000000000020
> > > >
> > > > I found out that the problem is in sound/hda/hdac_i915.c.
> > > >
> > > > When the i915 component binding failed in snd_hdac_i915_init(), the memory used
> > > > for "acomp" was released and "bus->audio_component" was cleared. However
> > > > "hadc_acomp" was not. "hdac_acomp" is later used in snd_hdac_i915_register_notifier(),
> > > > which leads to the Oops.
> > > >
> > > > In my case, the pointer left over in "hdac_acomp" is re-used for the iomem control
> > > > structure, causing the above Oops.
> > > >
> > > > The following commits, already upstream, fix the above issue by clearing "hdac_acomp".
> > > >
> > > > faafd03d23c913633d2ef7e6ffebdce01b164409
> > > > (ALSA: hda - Clear the leftover component assignment at snd_hdac_i915_exit())
> > > >
> > > > 97cc2ed27e5a168cf423f67c3bc7c6cc41d12f82
> > > > (ALSA: hda - Fix yet another i915 pointer leftover in error path)
> > > >
> > > > The commit 97cc2ed27e5a168cf423f67c3bc7c6cc41d12f82 depends on the following commit,
> > > > just a change in log message.
> > > >
> > > > bed2e98e1f4db8b827df507abc30be7b11b0613d
> > > > (ALSA: hda - Degrade i915 binding failure message)
> > > >
> > > > With the patches above applied, a WARNING still occurrs if "hdac_acomp" has a NULL pointer
> > > > in snd_hdac_i915_register_notifier().
> > >
> > > Yes, these are fine to backport. However...
> > >
> > >
> > > > int snd_hdac_i915_register_notifier(const struct i915_audio_component_audio_ops *aops)
> > > > {
> > > > if (WARN_ON(!hdac_acomp)) <--
> > > > return -ENODEV;
> > > >
> > > > This WARNING is also fixed upstream with the following commit, whose main purpose is
> > > > to support old Intel PCH devices, but it also suppress the WARNING.
> > > >
> > > > 6603249dcdbb6aab0b726bdf372d6f20c0d2d611
> > > > (ALSA: hda - Enable audio component for old Intel PCH devices)
> > >
> > > ... this isn't. It'll lead to other regressions that have been
> > > addressed later commits. Either backport only that check, or take a
> > > significant risk of other breakage.
> >
> > Ick, ok, want me to drop this one and keep the others? Or just drop all
> > of the ones i just queued up?
>
> Please just drop this one for now.
>
> And, either we need to queue more fixes or make a partial fix just for
> papering over this issue.
Now dropped, thanks.
greg k-h
More information about the Alsa-devel
mailing list