For a while, we've had the problem of i2c bus access not grabbing
a runtime PM ref when it's being used in userspace by i2c-dev, resulting
in nouveau spamming the kernel log with errors if anything attempts to
access the i2c bus while the GPU is in runtime suspend. An example:
[ 130.078386] nouveau 0000:01:00.0: i2c: aux 000d: begin idle timeout ffffffff
Since the GPU is in runtime suspend, the MMIO region that the i2c bus is
on isn't accessible. On x86, the standard behavior for accessing an
unavailable MMIO region is to just return ~0.
Except, that turned out to be a lie. While computers with a clean
concious will return ~0 in this scenario, some machines will actually
completely hang a CPU on certian bad MMIO accesses. This was witnessed
with someone's Lenovo ThinkPad P50, where sensors-detect attempting to
access the i2c bus while the GPU was suspended would result in a CPU
hang:
CPU: 5 PID: 12438 Comm: sensors-detect Not tainted 5.0.0-0.rc4.git3.1.fc30.x86_64 #1
Hardware name: LENOVO 20EQS64N17/20EQS64N17, BIOS N1EET74W (1.47 ) 11/21/2017
RIP: 0010:ioread32+0x2b/0x30
Code: 81 ff ff ff 03 00 77 20 48 81 ff 00 00 01 00 76 05 0f b7 d7 ed c3
48 c7 c6 e1 0c 36 96 e8 2d ff ff ff b8 ff ff ff ff c3 8b 07 <c3> 0f 1f
40 00 49 89 f0 48 81 fe ff ff 03 00 76 04 40 88 3e c3 48
RSP: 0018:ffffaac3c5007b48 EFLAGS: 00000292 ORIG_RAX: ffffffffffffff13
RAX: 0000000001111000 RBX: 0000000001111000 RCX: 0000043017a97186
RDX: 0000000000000aaa RSI: 0000000000000005 RDI: ffffaac3c400e4e4
RBP: ffff9e6443902c00 R08: ffffaac3c400e4e4 R09: ffffaac3c5007be7
R10: 0000000000000004 R11: 0000000000000001 R12: ffff9e6445dd0000
R13: 000000000000e4e4 R14: 00000000000003c4 R15: 0000000000000000
FS: 00007f253155a740(0000) GS:ffff9e644f600000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00005630d1500358 CR3: 0000000417c44006 CR4: 00000000003606e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
g94_i2c_aux_xfer+0x326/0x850 [nouveau]
nvkm_i2c_aux_i2c_xfer+0x9e/0x140 [nouveau]
__i2c_transfer+0x14b/0x620
i2c_smbus_xfer_emulated+0x159/0x680
? _raw_spin_unlock_irqrestore+0x1/0x60
? rt_mutex_slowlock.constprop.0+0x13d/0x1e0
? __lock_is_held+0x59/0xa0
__i2c_smbus_xfer+0x138/0x5a0
i2c_smbus_xfer+0x4f/0x80
i2cdev_ioctl_smbus+0x162/0x2d0 [i2c_dev]
i2cdev_ioctl+0x1db/0x2c0 [i2c_dev]
do_vfs_ioctl+0x408/0x750
ksys_ioctl+0x5e/0x90
__x64_sys_ioctl+0x16/0x20
do_syscall_64+0x60/0x1e0
entry_SYSCALL_64_after_hwframe+0x49/0xbe
RIP: 0033:0x7f25317f546b
Code: 0f 1e fa 48 8b 05 1d da 0c 00 64 c7 00 26 00 00 00 48 c7 c0 ff ff
ff ff c3 66 0f 1f 44 00 00 f3 0f 1e fa b8 10 00 00 00 0f 05 <48> 3d 01
f0 ff ff 73 01 c3 48 8b 0d ed d9 0c 00 f7 d8 64 89 01 48
RSP: 002b:00007ffc88caab68 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
RAX: ffffffffffffffda RBX: 00005630d0fe7260 RCX: 00007f25317f546b
RDX: 00005630d1598e80 RSI: 0000000000000720 RDI: 0000000000000003
RBP: 00005630d155b968 R08: 0000000000000001 R09: 00005630d15a1da0
R10: 0000000000000070 R11: 0000000000000246 R12: 00005630d1598e80
R13: 00005630d12f3d28 R14: 0000000000000720 R15: 00005630d12f3ce0
watchdog: BUG: soft lockup - CPU#5 stuck for 23s! [sensors-detect:12438]
Yikes! While I wanted to try to make it so that accessing an i2c bus on
nouveau would wake up the GPU as needed, airlied pointed out that pretty
much any usecase for userspace accessing an i2c bus on a GPU (mainly for
the DDC brightness control that some displays have) is going to only be
useful while there's at least one display enabled on the GPU anyway, and
the GPU never sleeps while there's displays running.
Since teaching the i2c bus to wake up the GPU on userspace accesses is a
good deal more difficult than it might seem, mostly due to the fact that
we have to use the i2c bus during runtime resume of the GPU, we instead
opt for the easiest solution: don't let userspace access i2c busses on
the GPU at all while it's in runtime suspend.
Changes since v1:
* Also disable i2c busses that run over DP AUX
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Some GPU units are capable of supporting "replayable" page faults, where
the execution unit will wait for SW to fixup GPU page tables rather than
triggering a channel-fatal fault.
This feature isn't useful (it's harmful, even) unless something like HMM
is being used to manage events appearing in the replayable fault buffer,
so, it's disabled by default.
This commit allows a client to request it be enabled.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Host methods exist to do at least some of what we need, but we are not
currently pushing replay/cancels through a channel like UVM does as it's
not clear whether it's necessary in our case (UVM also updates PTEs with
the GPU).
UVM also pushes a software method for fault cancels on Pascal, seemingly
because the host methods don't appear to be sufficient. If/when we want
to push the replay/cancel on the GPU, we can re-purpose the cancellation
code here to implement that swmthd.
Keep it simple for now, until we figure out exactly what we need here.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This provides a somewhat more direct method of manipulating the GPU page
tables, which will be required to support SVM.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This will be used to support a privileged client providing PTEs directly,
without a memory object to use as a reference.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
NVKM is currently responsible for managing the allocation of a client's
GPU address-space, but there's various use-cases (ie. HMM address-space
mirroring) where giving a client more direct control is desirable.
This commit allows for a VMM to be created where the area allocated for
NVKM is limited to a client-specified window, the remainder of address-
space is controlled directly by the client.
Leaving a window is necessary to support various internal requirements,
but also to support existing allocation interfaces as not all of the HW
is capable of working with a HMM allocation.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Turing has its SEC2 instance in an alternate location, and this avoids
needing to duplicate the code here for it.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Will be using this in upcoming changes to avoid the need for entirely
new subdevs to deal with Turing register moves.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
GF117 appears to use the same register as GK104 (but still with the
general Fermi readout mechanism).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108980
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
New registers.
Currently uncertain how exactly to mask fault buffer interrupts. This will
likely be corrected at around the same time as the new MC interrupt stuff
has been properly figured out and implemented.
For the moment, it shouldn't matter too much.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Things are a bit different here on Turing, and will require further changes
yet once I've investigated them more thoroughly.
For now though, the existing GP100 code is compatible enough with one small
hack to forward on fault buffer interrupts.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
The token will also contain runlist ID on Turing, so instead expose it as
an opaque value from NVKM so the client doesn't need to care.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This is needed for Turing, but we're supposed to wait for completion after
re-writing the value on older GPUs anyway.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Various structures are accessed by the GPU through BAR2 for some reason
on newer GPUs. This commit makes it more convenient to handle.
Will be used for GP100- fault buffers, and GV100- fault method buffers.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
High pixel clocks are required to use a 40 TMDS divider instead of 10,
and even low ones may optionally use scrambling depending on device
support.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Similar to commit 0bf8bf50ed ("module: Remove
const attribute from alias for MODULE_DEVICE_TABLE")
Fixes many -Wduplicate-decl-specifier warnings due to the combination of
const typeof() of already const variables.
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
