Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20220118163754.nfy53mfjpazgw2a2@eve> <YecV+rh/4rzygUbx@google.com>
 <CACeCKaeHAV1RLovgMt43uFtHioOeKNrqEbaPq8ZtKNiCS_tTsQ@mail.gmail.com>
 <CACeCKac0BctZae4n2CiAnpD4J-Dn+h2ROkx7CEVA9EmnobbNUw@mail.gmail.com>
 <CAFTm+6APx0PkRgp+7LLEOi=2E-7ZSgYdQ824U1XB+q1wMPNg3g@mail.gmail.com>
 <CACeCKaeAaS3QcLwvcPNYVtkKxaBViij53TBjOXvwcpKQk+NDbA@mail.gmail.com>
 <CAFTm+6BC--tNjbez_f_A_ckK7gjkbBMWHYSExxQcp9+u60Z2WQ@mail.gmail.com>
 <CACeCKaeAUy8JCO9hv7XjeQA_P_At9SN_Cuw2v=YD01gfwvGXFQ@mail.gmail.com> <20220119023752.ad34u6hgjpkpk4dw@eve>
In-Reply-To: <20220119023752.ad34u6hgjpkpk4dw@eve>
From:   Prashant Malani <pmalani@chromium.org>
Date:   Wed, 19 Jan 2022 10:24:38 -0800
Message-ID: <CACeCKafA4eDNgEbACjorCHBdrLaG6YviRS+gA=Xr-XD+GVaxZw@mail.gmail.com>
Subject: Re: Null pointer dereference in cros-ec-typec
To:     Alyssa Ross <hi@alyssa.is>
Cc:     "Mr. Chromebox" <mrchromebox@gmail.com>,
        Benson Leung <bleung@google.com>,
        Benson Leung <bleung@chromium.org>,
        linux-kernel@vger.kernel.org,
        Tim Wawrzynczak <twawrzynczak@chromium.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Thanks Alyssa,

It looks like the right fix here should go in coreboot.

I'll wait for a response from Matt regarding whether those EC commands
are supported on the eve EC firmware image Mr.Chromebox releases. I
have a fix but I can't test it since :
- I am not sure how the Mr.Chromebox eve BIOS is compiled.
- I don't have an eve.

On Tue, Jan 18, 2022 at 6:37 PM Alyssa Ross <hi@alyssa.is> wrote:
>
> On Tue, Jan 18, 2022 at 04:35:45PM -0800, Prashant Malani wrote:
> > There seem to be some differences.
> > The Mr.Chromebox GOOG0004 device doesn't have any connectors listed.
> > See here for a BIOS snippet from voxel:
> > Scope (\_SB.PCI0.LPCB.EC0.CREC)
> >     {
> >         Device (USBC)
> >         {
> >             Name (_HID, "GOOG0014")  // _HID: Hardware ID
> >             Name (_DDN, "ChromeOS EC Embedded Controller USB Type-C
> > Control")  // _DDN: DOS Device Name
> >             Device (CON0)
> >             {
> >                 ...
> >             }
> >
> >             Device (CON1)
> >             {
> >                ...
> >             }
> >         }
> >     }
> >
> > Does the EC on this device support EC_CMD_GET_PD_PORT_CAPS , or
> > EC_CMD_USB_PD_PORTS ?
> > It doesn't look like the EC is returning anything for these commands,
> > which are used to populate GOOG0014.
> >
> > The CREC device looks about right:
> >
> > Device (CREC)
> >             {
> >                 Name (_HID, "GOOG0004")  // _HID: Hardware ID
> >                 Name (_UID, One)  // _UID: Unique ID
> >                 Name (_DDN, "EC Command Device")  // _DDN: DOS Device N=
ame
> >                 Name (_PRW, Package (0x02)  // _PRW: Power Resources fo=
r Wake
> >                 {
> >                    ....
> >                 })
> >              }
> >
> > 2 observations:
> > - We probably shouldn't be generating a GOOG0014 device at all if the
> > EC_CMD_GET_PD_PORT_CAPS and EC_CMD_USB_PD_PORTS commands aren't
> > supported by the EC. I can work with coreboot to make that change
> > - Is the order of probing for some reason causing the GOOG0014 child
> > device to not be linked to the GOOG0004 device? Alyssa, does the
> > following diff help:
>
> With the diff applied, I no longer see the Oops from cros_ec_typec, and
> I can reboot normally.  But, it looks like the driver is deferred
> indefinitely =E2=80=94 I added a debug print after the check, and it was =
never
> triggered.
>
> > diff --git a/drivers/platform/chrome/cros_ec_typec.c
> > b/drivers/platform/chrome/cros_ec_typec.c
> > index 5de0bfb0bc4d..7059912b75c1 100644
> > --- a/drivers/platform/chrome/cros_ec_typec.c
> > +++ b/drivers/platform/chrome/cros_ec_typec.c
> > @@ -1076,6 +1076,10 @@ static int cros_typec_probe(struct platform_devi=
ce *pdev)
> >
> >         typec->dev =3D dev;
> >         typec->ec =3D dev_get_drvdata(pdev->dev.parent);
> > +
> > +       if (!typec->ec)
> > +               return -EPROBE_DEFER;
> > +
> >         platform_set_drvdata(pdev, typec);
> >
> >         ret =3D cros_typec_get_cmd_version(typec);
> >
> >
> > On Tue, Jan 18, 2022 at 2:34 PM Mr. Chromebox <mrchromebox@gmail.com> w=
rote:
> > >
> > > On Tue, Jan 18, 2022 at 4:16 PM Prashant Malani <pmalani@chromium.org=
> wrote:
> > > >
> > > > Hi Matt,
> > > >
> > > > On Tue, Jan 18, 2022 at 2:04 PM Mr. Chromebox <mrchromebox@gmail.co=
m> wrote:
> > > > >
> > > > > hi Prashant,
> > > > >
> > > > > my releases track upstream coreboot; my most recent release was b=
ased
> > > > > on coreboot 4.14 (I'm behind on getting a 4.15-based release out)=
.
> > > > >
> > > > > A quick perusal of the source for src/ec/google/chromeec/ doesn't=
 show
> > > > > any recent changes to the location of the GOOG0014 ACPI device. T=
he
> > > > > most recent change was 2 years ago (so, landed in the 4.12 releas=
e),
> > > > > which moved the USB-C child device to its present location: under
> > > > > \_SB.PCI0.LPCB.EC0.CREC
> > > > >
> > > > > ref: https://github.com/coreboot/coreboot/commit/eec30f7beae074c3=
f80a182cc2950ed8e4f0a640
> > > > >
> > > > > prior to that, it was located under  \_SB.PCI0.LPCB.EC0.
> > > > >
> > > > > I also dumped/disassembled the ACPI from a recent build to confir=
m the above.
> > > >
> > > > Is it possible to share the disassembled ACPI tables? We can then
> > > > compare it to the ones on shipping Chromebooks to identify a
> > > > discrepancy.
> > > > If the GOOG0014 device is correctly listed as a child of the EC dev=
ice
> > > > (GOOG0004), then the kernel ACPI framework should be setting
> > > > GOOG0004 as a parent (and dev_get_drvdata(pdev->dev.parent) shouldn=
't
> > > > return NULL).
> > >
> > > as the GOOG0014 device is runtime-generated, it's located in the SSDT=
:
> > >
> > > External (_SB_.PCI0.LPCB.EC0_.CREC, DeviceObj)
> > > ...
> > > Scope (\_SB.PCI0.LPCB.EC0.CREC)
> > > {
> > >     Device (USBC)
> > >     {
> > >         Name (_HID, "GOOG0014")  // _HID: Hardware ID
> > >         Name (_DDN, "ChromeOS EC Embedded Controller USB Type-C
> > > Control")  // _DDN: DOS Device Name
> > >     }
> > > }
> > >
> > > GOOG0004 is defined in the DSDT, under EC0:
> > >
> > > Device (CREC)
> > > {
> > >     Name (_HID, "GOOG0004")  // _HID: Hardware ID
> > >     Name (_UID, One)  // _UID: Unique ID
> > >     Name (_DDN, "EC Command Device")  // _DDN: DOS Device Name
> > >     Name (_PRW, Package (0x02)  // _PRW: Power Resources for Wake
> > >     {
> > >         0x70,
> > >         0x05
> > >     })
> > > ...
> > > }
> > >
> > > -Matt
> > >
> > > > > regards,
> > > > > Matt / MrChromebox
> > > > >
> > > > > On Tue, Jan 18, 2022 at 2:12 PM Prashant Malani <pmalani@chromium=
.org> wrote:
> > > > > >
> > > > > > (+Mr.Chromebox team; using the address listed in
> > > > > > https://mrchromebox.tech/#support )
> > > > > >
> > > > > > Hi Team Mr.Chromebox,
> > > > > >
> > > > > > Could you kindly provide some more detail regarding how the GOO=
G0014
> > > > > > Type C ACPI device is set up in the Mr Chromebox BIOS for Chrom=
ebooks
> > > > > > (the driver expects it to be embedded in the GOOG0004 EC device=
)?
> > > > > > We want to enable Alyssa and other developers using the Mr.Chro=
mebox
> > > > > > BIOS to have a functional cros-ec-typec driver, so would like t=
o help
> > > > > > ensure that the device is set up correctly in ACPI.
> > > > > >
> > > > > > Thanks!
> > > > > >
> > > > > > -Prashant
> > > > > >
> > > > > > On Tue, Jan 18, 2022 at 11:49 AM Prashant Malani <pmalani@chrom=
ium.org> wrote:
> > > > > > >
> > > > > > > Hi Benson and Alyssa,
> > > > > > >
> > > > > > > On Tue, Jan 18, 2022 at 11:33 AM Benson Leung <bleung@google.=
com> wrote:
> > > > > > > >
> > > > > > > > Hi Alyssa,
> > > > > > > >
> > > > > > > > Thanks for reaching out.
> > > > > > > >
> > > > > > > > On Tue, Jan 18, 2022 at 04:37:54PM +0000, Alyssa Ross wrote=
:
> > > > > > > > > My distribution recently enabled the Chrome OS EC Type C =
control driver
> > > > > > > > > in its kernel builds.  On my Google Pixelbook i7 (eve), t=
he driver reports
> > > > > > > > > a null pointer dereference at boot.  From what I can tell=
, this happens
> > > > > > > > > because typec->ec is set to NULL in cros_typec_probe.  Ot=
her drivers,
> > > > > > > > > like cros-usbpd-notify, appear to be set up to handle thi=
s case.  As a
> > > > > > > > > result of this bug, I'm no longer able to reboot my compu=
ter, because
> > > > > > > > > udevd hangs while trying to do something with the device =
whose driver
> > > > > > > > > isn't working.
> > > > > > > > >
> > > > > > > >
> > > > > > > > I've copied Prashant, who's the author of the typec driver =
as well as
> > > > > > > > cros-usbpd-notify.
> > > > > > > >
> > > > > > > > Prashant, any thoughts on a more graceful failure out of th=
e typec driver's
> > > > > > > > probe in case there's no ec object?
> > > > > > >
> > > > > > > We can add a NULL check and just abort the driver probe if th=
e pointer is
> > > > > > > not valid (the driver is useless without that pointer anyway)=
.
> > > > > > >
> > > > > > > A note: The NULL check makes sense on older drivers like cros=
-usbpd-notify since
> > > > > > > they can exist in ACPI configurations where they are *not* em=
bedded
> > > > > > > inside the GOOG0004
> > > > > > > EC device (on older Chromebooks). That is not the case for th=
e EC Type C device.
> > > > > > >
> > > > > > > This raises another issue: the custom BIOS from Mr. Chromebox=
 is
> > > > > > > likely not setting
> > > > > > > up the EC Type C ACPI (GOOG0014) device correctly; it *must* =
be
> > > > > > > embedded inside the overall
> > > > > > > EC device (GOOG0004). If this is not being done, then the GOO=
G0014
> > > > > > > device should not
> > > > > > > be added to the ACPI tables at all.
> > > > > > >
> > > > > > > I would like to understand whether the above was intentional =
from the
> > > > > > > Mr. Chromebox BIOS developers;
> > > > > > > otherwise we are letting an incorrect ACPI configuration just=
 fail
> > > > > > > with a probe error.
> > > > > > >
> > > > > > > Thanks,
> > > > > > >
> > > > > > > -Prashant
> > > > > > >
> > > > > > > >
> > > > > > > > > Here's the full Oops.  I was able to reproduce the issue =
with every
> > > > > > > > > kernel I tried, from 5.10 to mainline.
> > > > > > > > >
> > > > > > > > > cros-usbpd-notify-acpi GOOG0003:00: Couldn't get Chrome E=
C device pointer.
> > > > > > > > > input: Intel Virtual Buttons as /devices/pci0000:00/0000:=
00:1f.0/PNP0C09:00/INT33D6:00/input/input14
> > > > > > > > > BUG: kernel NULL pointer dereference, address: 0000000000=
0000d8
> > > > > > > > > #PF: supervisor read access in kernel mode
> > > > > > > > > #PF: error_code(0x0000) - not-present page
> > > > > > > > > PGD 0 P4D 0
> > > > > > > > > Oops: 0000 [#1] SMP PTI
> > > > > > > > > CPU: 1 PID: 561 Comm: systemd-udevd Not tainted 5.15.12 #=
4
> > > > > > > > > Hardware name: Google Eve/Eve, BIOS MrChromebox-4.14 08/0=
6/2021
> > > > > > > >
> > > > > > > >
> > > > > > > > Ah, here's the problem. It looks like this is a custom bios=
 from Mr Chromebox,
> > > > > > > > so this is not a bios combination we validate at Google.
> > > > > > > >
> > > > > > > > Thank you for the report. We'll look into fixing this and m=
arking the fix
> > > > > > > > for stable kernels so that it goes back to 5.10.
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > >
> > > > > > > > Benson
> > > > > > > >
> > > > > > > > > RIP: 0010:__mutex_lock+0x59/0x8c0
> > > > > > > > > Code: 53 48 89 cb 48 83 ec 70 89 75 9c be 3d 02 00 00 4c =
89 45 90 e8 18 47 33 ff e8 e3 e2 ff ff 44 8b 35 a4 85 e8 02 45 85 f6 75 0a =
<4d> 3b 6d 68 0f 85 bf 07 00 00 65 ff 05 b6 5b 23 75 ff 75 90 4d 8d
> > > > > > > > > RSP: 0018:ffffb44580a4bb50 EFLAGS: 00010246
> > > > > > > > > RAX: 0000000000000000 RBX: 0000000000000000 RCX: 00000000=
00000001
> > > > > > > > > RDX: 0000000000000000 RSI: ffffffff8bf91320 RDI: ffff922c=
bba50e20
> > > > > > > > > RBP: ffffb44580a4bbf0 R08: 0000000000000000 R09: ffff922c=
5bac8140
> > > > > > > > > R10: ffffb44580a4bc10 R11: 0000000000000000 R12: 00000000=
00000000
> > > > > > > > > R13: 0000000000000070 R14: 0000000000000000 R15: 00000000=
00000001
> > > > > > > > > FS:  00007f55338d6b40(0000) GS:ffff922fae200000(0000) knl=
GS:0000000000000000
> > > > > > > > > CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > > > > > > CR2: 00000000000000d8 CR3: 000000011bbb2006 CR4: 00000000=
003706e0
> > > > > > > > > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 00000000=
00000000
> > > > > > > > > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 00000000=
00000400
> > > > > > > > > Call Trace:
> > > > > > > > >  <TASK>
> > > > > > > > >  ? fs_reclaim_acquire+0x4d/0xd0
> > > > > > > > >  ? lock_is_held_type+0xaa/0x120
> > > > > > > > >  ? cros_ec_cmd_xfer_status+0x1f/0x110
> > > > > > > > >  ? lock_is_held_type+0xaa/0x120
> > > > > > > > >  ? cros_ec_cmd_xfer_status+0x1f/0x110
> > > > > > > > >  cros_ec_cmd_xfer_status+0x1f/0x110
> > > > > > > > >  cros_typec_ec_command+0x91/0x1c0 [cros_ec_typec]
> > > > > > > > >  cros_typec_probe+0x7f/0x5a8 [cros_ec_typec]
> > > > > > > > >  platform_probe+0x3f/0x90
> > > > > > > > >  really_probe+0x1f5/0x3f0
> > > > > > > > >  __driver_probe_device+0xfe/0x180
> > > > > > > > >  driver_probe_device+0x1e/0x90
> > > > > > > > >  __driver_attach+0xc4/0x1d0
> > > > > > > > >  ? __device_attach_driver+0xe0/0xe0
> > > > > > > > >  ? __device_attach_driver+0xe0/0xe0
> > > > > > > > >  bus_for_each_dev+0x67/0x90
> > > > > > > > >  bus_add_driver+0x12e/0x1f0
> > > > > > > > >  driver_register+0x8f/0xe0
> > > > > > > > >  ? 0xffffffffc04ec000
> > > > > > > > >  do_one_initcall+0x67/0x320
> > > > > > > > >  ? rcu_read_lock_sched_held+0x3f/0x80
> > > > > > > > >  ? trace_kmalloc+0x38/0xe0
> > > > > > > > >  ? kmem_cache_alloc_trace+0x17c/0x2b0
> > > > > > > > >  do_init_module+0x5c/0x270
> > > > > > > > >  __do_sys_finit_module+0x95/0xe0
> > > > > > > > >  do_syscall_64+0x3b/0x90
> > > > > > > > >  entry_SYSCALL_64_after_hwframe+0x44/0xae
> > > > > > > > > RIP: 0033:0x7f55344b1f3d
> > > > > > > > > Code: 5b 41 5c c3 66 0f 1f 84 00 00 00 00 00 f3 0f 1e fa =
48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 =
<48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d bb ee 0e 00 f7 d8 64 89 01 48
> > > > > > > > > RSP: 002b:00007fff187f1388 EFLAGS: 00000246 ORIG_RAX: 000=
0000000000139
> > > > > > > > > RAX: ffffffffffffffda RBX: 000055a53acbe6e0 RCX: 00007f55=
344b1f3d
> > > > > > > > > RDX: 0000000000000000 RSI: 00007f553461732c RDI: 00000000=
0000000e
> > > > > > > > > RBP: 0000000000020000 R08: 0000000000000000 R09: 00000000=
00000002
> > > > > > > > > R10: 000000000000000e R11: 0000000000000246 R12: 00007f55=
3461732c
> > > > > > > > > R13: 000055a53ad94010 R14: 0000000000000007 R15: 000055a5=
3ad95690
> > > > > > > > >  </TASK>
> > > > > > > > > Modules linked in: fjes(+) cros_ec_typec(+) typec intel_v=
btn(+) cros_usbpd_notify sparse_keymap soc_button_array int3403_thermal int=
340x_thermal_zone int3400_thermal acpi_thermal_rel cros_kbd_led_backlight z=
ram ip_tables i915 hid_multitouch i2c_algo_bit ttm crct10dif_pclmul crc32_p=
clmul crc32c_intel drm_kms_helper nvme ghash_clmulni_intel sdhci_pci cqhci =
cec nvme_core sdhci serio_raw drm mmc_core i2c_hid_acpi i2c_hid video pinct=
rl_sunrisepoint fuse
> > > > > > > > > CR2: 00000000000000d8
> > > > > > > > > ---[ end trace 4a12c4896d70352b ]---
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > > --
> > > > > > > > Benson Leung
> > > > > > > > Staff Software Engineer
> > > > > > > > Chrome OS Kernel
> > > > > > > > Google Inc.
> > > > > > > > bleung@google.com
> > > > > > > > Chromium OS Project
> > > > > > > > bleung@chromium.org