platform/kernel/linux-rpi.git
7 years agodrm/i915: Poison the request before emitting commands
Chris Wilson [Sun, 23 Apr 2017 17:06:17 +0000 (18:06 +0100)]
drm/i915: Poison the request before emitting commands

If we poison the request before we emit commands, it should be easier to
spot when we execute an uninitialised request.

References: https://bugs.freedesktop.org/show_bug.cgi?id=100144
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170423170619.7156-2-chris@chris-wilson.co.uk
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
7 years agodrm/i915: Differentiate between sw write location into ring and last hw read
Chris Wilson [Tue, 25 Apr 2017 13:00:49 +0000 (14:00 +0100)]
drm/i915: Differentiate between sw write location into ring and last hw read

We need to keep track of the last location we ask the hw to read up to
(RING_TAIL) separately from our last write location into the ring, so
that in the event of a GPU reset we do not tell the HW to proceed into
a partially written request (which can happen if that request is waiting
for an external signal before being executed).

v2: Refactor intel_ring_reset() (Mika)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100144
Testcase: igt/gem_exec_fence/await-hang
Fixes: 821ed7df6e2a ("drm/i915: Update reset path to fix incomplete requests")
Fixes: d55ac5bf97c6 ("drm/i915: Defer transfer onto execution timeline to actual hw submission")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170425130049.26147-1-chris@chris-wilson.co.uk
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
7 years agodrm/i915: Report request restarts for both execlists/guc
Chris Wilson [Tue, 25 Apr 2017 10:38:35 +0000 (11:38 +0100)]
drm/i915: Report request restarts for both execlists/guc

As we now share the execlist_port[] tracking for both execlists/guc, we
can reset the inflight count on both and report which requests are being
restarted.

Suggested-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michel Thierry <michel.thierry@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170425103835.31871-1-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
7 years agodrm/i915: Include interesting seqno in the missed breadcrumb debug
Chris Wilson [Sun, 23 Apr 2017 17:06:19 +0000 (18:06 +0100)]
drm/i915: Include interesting seqno in the missed breadcrumb debug

Knowing the neighbouring seqno (current on hw, last submitted to hw)
provide some useful breadcrumbs to the debug log.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170423170619.7156-4-chris@chris-wilson.co.uk
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
7 years agodrm/i915: Confirm the request is still active before adding it to the await
Chris Wilson [Sat, 22 Apr 2017 08:15:37 +0000 (09:15 +0100)]
drm/i915: Confirm the request is still active before adding it to the await

Although we do check the completion-status of the request before
actually adding a wait on it (either to its submit fence or its
completion dma-fence), we currently do not check before adding it to the
dependency lists.

In fact, without checking for a completed request we may try to use the
signaler after it has been retired and its dependency tree freed:

[   60.044057] BUG: KASAN: use-after-free in __list_add_valid+0x1d/0xd0 at addr ffff880348c9e6a0
[   60.044118] Read of size 8 by task gem_exec_fence/530
[   60.044164] CPU: 1 PID: 530 Comm: gem_exec_fence Tainted: G            E   4.11.0-rc7+ #46
[   60.044226] Hardware name: ��������������������������������� ���������������������������������/���������������������������������, BIOS RYBDWi35.86A.0246.2
[   60.044290] Call Trace:
[   60.044337]  dump_stack+0x4d/0x6a
[   60.044383]  kasan_object_err+0x21/0x70
[   60.044435]  kasan_report+0x225/0x4e0
[   60.044488]  ? __list_add_valid+0x1d/0xd0
[   60.044534]  ? kasan_kmalloc+0xad/0xe0
[   60.044587]  __asan_load8+0x5e/0x70
[   60.044639]  __list_add_valid+0x1d/0xd0
[   60.044788]  __i915_priotree_add_dependency+0x67/0x130 [i915]
[   60.044895]  i915_gem_request_await_request+0xa8/0x370 [i915]
[   60.044974]  i915_gem_request_await_dma_fence+0x129/0x140 [i915]
[   60.045049]  i915_gem_do_execbuffer.isra.37+0xb0a/0x26b0 [i915]
[   60.045077]  ? save_stack+0xb1/0xd0
[   60.045105]  ? save_stack_trace+0x1b/0x20
[   60.045132]  ? save_stack+0x46/0xd0
[   60.045158]  ? kasan_kmalloc+0xad/0xe0
[   60.045184]  ? __kmalloc+0xd8/0x670
[   60.045229]  ? drm_ioctl+0x359/0x640 [drm]
[   60.045256]  ? SyS_ioctl+0x41/0x70
[   60.045330]  ? i915_vma_move_to_active+0x540/0x540 [i915]
[   60.045360]  ? tty_insert_flip_string_flags+0xa1/0xf0
[   60.045387]  ? tty_flip_buffer_push+0x63/0x70
[   60.045414]  ? remove_wait_queue+0xa9/0xc0
[   60.045441]  ? kasan_unpoison_shadow+0x35/0x50
[   60.045467]  ? kasan_kmalloc+0xad/0xe0
[   60.045494]  ? kasan_check_write+0x14/0x20
[   60.045568]  i915_gem_execbuffer2+0xdb/0x2a0 [i915]
[   60.045616]  drm_ioctl+0x359/0x640 [drm]
[   60.045705]  ? i915_gem_execbuffer+0x5a0/0x5a0 [i915]
[   60.045751]  ? drm_version+0x150/0x150 [drm]
[   60.045778]  ? compat_start_thread+0x60/0x60
[   60.045805]  ? plist_del+0xda/0x1a0
[   60.045833]  do_vfs_ioctl+0x12e/0x910
[   60.045860]  ? ioctl_preallocate+0x130/0x130
[   60.045886]  ? pci_mmcfg_check_reserved+0xc0/0xc0
[   60.045913]  ? vfs_write+0x196/0x240
[   60.045939]  ? __fget_light+0xa7/0xc0
[   60.045965]  SyS_ioctl+0x41/0x70
[   60.045991]  entry_SYSCALL_64_fastpath+0x17/0x98
[   60.046017] RIP: 0033:0x7feb2baefc47
[   60.046042] RSP: 002b:00007fff56d28e58 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
[   60.046075] RAX: ffffffffffffffda RBX: 00007fff56d290a8 RCX: 00007feb2baefc47
[   60.046102] RDX: 00007fff56d29050 RSI: 00000000c0406469 RDI: 0000000000000003
[   60.046129] RBP: 00007fff56d29050 R08: 000055ecc4cd27d0 R09: 00007feb2bda8600
[   60.046154] R10: 0000000000000073 R11: 0000000000000246 R12: 00000000c0406469
[   60.046177] R13: 0000000000000003 R14: 000000000000000f R15: 0000000000000099
[   60.046203] Object at ffff880348c9e680, in cache i915_dependency size: 64
[   60.046225] Allocated:
[   60.046246] PID = 530
[   60.046269]  save_stack_trace+0x1b/0x20
[   60.046292]  save_stack+0x46/0xd0
[   60.046318]  kasan_kmalloc+0xad/0xe0
[   60.046343]  kasan_slab_alloc+0x12/0x20
[   60.046368]  kmem_cache_alloc+0xab/0x650
[   60.046445]  i915_gem_request_await_request+0x88/0x370 [i915]
[   60.046559]  i915_gem_request_await_dma_fence+0x129/0x140 [i915]
[   60.046705]  i915_gem_do_execbuffer.isra.37+0xb0a/0x26b0 [i915]
[   60.046849]  i915_gem_execbuffer2+0xdb/0x2a0 [i915]
[   60.046936]  drm_ioctl+0x359/0x640 [drm]
[   60.046987]  do_vfs_ioctl+0x12e/0x910
[   60.047038]  SyS_ioctl+0x41/0x70
[   60.047090]  entry_SYSCALL_64_fastpath+0x17/0x98
[   60.047139] Freed:
[   60.047179] PID = 530
[   60.047223]  save_stack_trace+0x1b/0x20
[   60.047269]  save_stack+0x46/0xd0
[   60.047317]  kasan_slab_free+0x72/0xc0
[   60.047366]  kmem_cache_free+0x39/0x160
[   60.047512]  i915_gem_request_retire+0x83f/0x930 [i915]
[   60.047657]  i915_gem_request_alloc+0x166/0x600 [i915]
[   60.047799]  i915_gem_do_execbuffer.isra.37+0xad8/0x26b0 [i915]
[   60.047897]  i915_gem_execbuffer2+0xdb/0x2a0 [i915]
[   60.047942]  drm_ioctl+0x359/0x640 [drm]
[   60.047968]  do_vfs_ioctl+0x12e/0x910
[   60.047993]  SyS_ioctl+0x41/0x70
[   60.048019]  entry_SYSCALL_64_fastpath+0x17/0x98
[   60.048044] Memory state around the buggy address:
[   60.048066]  ffff880348c9e580: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc
[   60.048105]  ffff880348c9e600: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc
[   60.048138] >ffff880348c9e680: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
[   60.048170]                                ^
[   60.048191]  ffff880348c9e700: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc
[   60.048225]  ffff880348c9e780: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc

Note to hit the use-after-free requires us to be passed back a request
via a fence-array, that is from explicit fencing accumulated into a
sync-file fence-array.

Fixes: 52e542090701 ("drm/i915/scheduler: Record all dependencies upon request construction")
Testcase: igt/gem_exec_fence/expired-history
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170422081537.6468-1-chris@chris-wilson.co.uk
7 years agodrm/i915: Avoid busy-spinning on VLV_GLTC_PW_STATUS mmio
Chris Wilson [Fri, 21 Apr 2017 13:58:15 +0000 (14:58 +0100)]
drm/i915: Avoid busy-spinning on VLV_GLTC_PW_STATUS mmio

The busy-spin, as the first stage of intel_wait_for_register(), is
currently under suspicion for causing:

[   62.034926] NMI watchdog: Watchdog detected hard LOCKUP on cpu 1
[   62.034928] Modules linked in: i2c_dev i915 intel_gtt drm_kms_helper prime_numbers
[   62.034932] CPU: 1 PID: 183 Comm: kworker/1:2 Not tainted 4.11.0-rc7+ #471
[   62.034933] Hardware name:                  /        , BIOS PYBSWCEL.86A.0027.2015.0507.1758 05/07/2015
[   62.034934] Workqueue: pm pm_runtime_work
[   62.034936] task: ffff880275a04ec0 task.stack: ffffc900002d8000
[   62.034936] RIP: 0010:__intel_wait_for_register_fw+0x77/0x1a0 [i915]
[   62.034937] RSP: 0018:ffffc900002dbc38 EFLAGS: 00000082
[   62.034939] RAX: ffffc90003530094 RBX: 0000000000130094 RCX: 0000000000000001
[   62.034940] RDX: 00000000000000a1 RSI: ffff88027fd15e58 RDI: 0000000000000000
[   62.034941] RBP: ffffc900002dbc78 R08: 0000000000000002 R09: 0000000000000000
[   62.034942] R10: ffffc900002dbc18 R11: ffff880276429dd0 R12: ffff8802707c0000
[   62.034943] R13: 00000000000000a0 R14: 0000000000000000 R15: 00000000fffefc10
[   62.034945] FS:  0000000000000000(0000) GS:ffff88027fd00000(0000) knlGS:0000000000000000
[   62.034945] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   62.034947] CR2: 00007ffd3cd98ff8 CR3: 0000000274c19000 CR4: 00000000001006e0
[   62.034947] Call Trace:
[   62.034948]  intel_wait_for_register+0x77/0x140 [i915]
[   62.034949]  vlv_suspend_complete+0x23/0x5b0 [i915]
[   62.034950]  intel_runtime_suspend+0x16c/0x2a0 [i915]
[   62.034950]  pci_pm_runtime_suspend+0x50/0x180
[   62.034951]  ? pci_pm_runtime_resume+0xa0/0xa0
[   62.034952]  __rpm_callback+0xc5/0x210
[   62.034953]  rpm_callback+0x1f/0x80
[   62.034953]  ? pci_pm_runtime_resume+0xa0/0xa0
[   62.034954]  rpm_suspend+0x118/0x580
[   62.034955]  pm_runtime_work+0x64/0x90
[   62.034956]  process_one_work+0x1bb/0x3e0
[   62.034956]  worker_thread+0x46/0x4f0
[   62.034957]  ? __schedule+0x18b/0x610
[   62.034958]  kthread+0xff/0x140
[   62.034958]  ? process_one_work+0x3e0/0x3e0
[   62.034959]  ? kthread_create_on_node+

and related hard lockups in CI for byt and bsw.

Note this effectively reverts commits 41ce405e6894 and b27366958869
("drm/i915: Convert wait_for(I915_READ(reg)) to intel_wait_for_register()")

v2: Convert bool allow into a u32 mask for clarity and repeat the
comment on vlv rc6 timing to justify the 3ms timeout used for the wait (Ville)

Fixes: 41ce405e6894 ("drm/i915: Convert wait_for(I915_READ(reg)) to intel_wait_for_register()")
Fixes: b27366958869 ("drm/i915: Convert wait_for(I915_READ(reg)) to intel_wait_for_register()")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100718
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Tomi Sarvela <tomi.p.sarvela@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170421135815.11897-1-chris@chris-wilson.co.uk
Tested-by: Tomi Sarvela <tomi.p.sarvela@intel.com>
7 years agodrm/i915: Stop touching hangcheck.seqno from intel_engine_init_global_seqno()
Chris Wilson [Fri, 21 Apr 2017 08:31:13 +0000 (09:31 +0100)]
drm/i915: Stop touching hangcheck.seqno from intel_engine_init_global_seqno()

The hangcheck runs independently to the main flow of seqno through the
driver. However, we have an odd coupling of the seqno reset that is
unwelcome, and if poked at just the right rate can cause spurious hangs
(e.g. gem_exec_whisper) on an apparently idle engine.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170421083113.21321-1-chris@chris-wilson.co.uk
7 years agodrm/i915: Use discardable buffers for rings
Chris Wilson [Thu, 20 Apr 2017 10:17:09 +0000 (11:17 +0100)]
drm/i915: Use discardable buffers for rings

The contents of a ring are only valid between HEAD and TAIL, when the
ring is idle (HEAD == TAIL) we can simply let the pages go under memory
pressure if they are not pinned by an active context. Any new content
will be written after HEAD and so the ring will again be valid between
HEAD and TAIL, everything outside can be discarded.

Note that we take care of ensuring that we do not reset the HEAD
backwards following a GPU hang on an idle ring.

The same precautions are what enable us to use stolen memory for rings.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170420101709.27250-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
7 years agodrm/i915/selftests: Allocate inode/file dynamically
Chris Wilson [Wed, 19 Apr 2017 09:41:17 +0000 (10:41 +0100)]
drm/i915/selftests: Allocate inode/file dynamically

Avoid having too large a stack by creating the fake struct inode/file on
the heap instead.

drivers/gpu/drm/i915/selftests/mock_drm.c: In function 'mock_file':
drivers/gpu/drm/i915/selftests/mock_drm.c:46:1: error: the frame size of 1328 bytes is larger than 1280 bytes [-Werror=frame-larger-than=]
drivers/gpu/drm/i915/selftests/mock_drm.c: In function 'mock_file_free':
drivers/gpu/drm/i915/selftests/mock_drm.c:54:1: error: the frame size of 1312 bytes is larger than 1280 bytes [-Werror=frame-larger-than=]

Reported-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 66d9cb5d805a ("drm/i915: Mock the GEM device for self-testing")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Link: http://patchwork.freedesktop.org/patch/msgid/20170419094143.16922-2-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
7 years agodrm/i915: Avoid format string expansion from engine names
Kees Cook [Tue, 11 Apr 2017 04:56:30 +0000 (21:56 -0700)]
drm/i915: Avoid format string expansion from engine names

While highly unlikely, this makes sure that the string built from
engine names won't be processed as a format string.

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411045630.GA6612@beast
7 years agodrm/i915: Fix GCC 4.4 build issue with __intel_wait_for_register_fw
Tvrtko Ursulin [Tue, 18 Apr 2017 10:52:11 +0000 (11:52 +0100)]
drm/i915: Fix GCC 4.4 build issue with __intel_wait_for_register_fw

Move the BUILD_BUG_ONs for busy-wait duration outside the
_wait_for_atomic macro as discussed on the mailing list.

v2: Simplify the macro by omitting the ret__ local. (Chris Wilson)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Suggested-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Fixes: 1d1a9774e404 ("drm/i915: Extend intel_wait_for_register_fw() with fast timeout")
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170418105211.7089-1-tvrtko.ursulin@linux.intel.com
7 years agodrm/i915: Fix system hang with EI UP masked on Haswell
Mika Kuoppala [Thu, 13 Apr 2017 11:15:27 +0000 (14:15 +0300)]
drm/i915: Fix system hang with EI UP masked on Haswell

Previously with commit a9c1f90c8e17
("drm/i915: Don't mask EI UP interrupt on IVB|SNB") certain,
seemingly unrelated bit (GEN6_PM_RP_UP_EI_EXPIRED) was needed
to be unmasked for IVB and SNB in order to prevent system hang
with chained batchbuffers.

Our CI was seeing incomplete results with tests that used
chained batches and it was found out that HSW needs to have this
same bit unmasked to reliably survive chained batches.

Always unmask GEN6_PM_RP_UP_EI_EXPIRED on Haswell to
prevent system hang with batch chaining.

Testcase: igt/gem_exec_fence/nb-await-default
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100672
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1492082127-29007-1-git-send-email-mika.kuoppala@intel.com
7 years agodrm/i915: Update DRIVER_DATE to 20170418
Daniel Vetter [Tue, 18 Apr 2017 09:18:16 +0000 (11:18 +0200)]
drm/i915: Update DRIVER_DATE to 20170418

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Copy user requested buffers into the error state
Chris Wilson [Sat, 15 Apr 2017 09:39:02 +0000 (10:39 +0100)]
drm/i915: Copy user requested buffers into the error state

Introduce a new execobject.flag (EXEC_OBJECT_CAPTURE) that userspace may
use to indicate that it wants the contents of this buffer preserved in
the error state (/sys/class/drm/cardN/error) following a GPU hang
involving this batch.

Use this at your discretion, the contents of the error state. although
compressed, are allocated with GFP_ATOMIC (i.e. limited) and kept for all
eternity (until the error state is destroyed).

Based on an earlier patch by Ben Widawsky <ben@bwidawsk.net>
Testcase: igt/gem_exec_capture
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ben Widawsky <ben@bwidawsk.net>
Cc: Matt Turner <mattst88@gmail.com>
Acked-by: Ben Widawsky <ben@bwidawsk.net>
Acked-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170415093902.22581-1-chris@chris-wilson.co.uk
7 years agodrm/i915: set "ret" correctly on error paths
Dan Carpenter [Fri, 14 Apr 2017 19:54:25 +0000 (22:54 +0300)]
drm/i915: set "ret" correctly on error paths

If "crtc" is NULL, then my static checker complains that "ret" isn't
initialized on that path.  It doesn't really cause a problem unless
"ret" is somehow set to -EDEADLK which is not likely.

Chris Wilson also noticed another error path where "ret" isn't set
correctly.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170414195425.GA8144@mwanda
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: checking for NULL instead of IS_ERR() in mock selftests
Dan Carpenter [Thu, 13 Apr 2017 19:52:17 +0000 (22:52 +0300)]
drm/i915: checking for NULL instead of IS_ERR() in mock selftests

i915_gem_request_alloc() uses error pointers.  It never returns NULLs.

Fixes: 0daf0113cff6 ("drm/i915: Mock infrastructure for request emission")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170413195217.GA26108@mwanda
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Implement Link Rate fallback on Link training failure
Manasi Navare [Thu, 6 Apr 2017 13:44:19 +0000 (16:44 +0300)]
drm/i915: Implement Link Rate fallback on Link training failure

If link training at a link rate optimal for a particular
mode fails during modeset's atomic commit phase, then we
let the modeset complete and then retry. We save the link rate
value at which link training failed, update the link status property
to "BAD" and use a lower link rate to prune the modes. It will redo
the modeset on the current mode at lower link rate or if the current
mode gets pruned due to lower link constraints then, it will send a
hotplug uevent for userspace to handle it.

This is also required to pass DP CTS tests 4.3.1.3, 4.3.1.4,
4.3.1.6.

This patch is a resend of the original commit id (233ce881dd91fb
"drm/i915: Implement Link Rate fallback on Link training failure")
which got reverted in this commit id (afc1ebf4562a14 Revert
"drm/i915: Implement Link Rate fallback on Link training failure")
due to CI failures.

After investigating the CI failures it was found that these
were essentially the failures which were always there but hidden because
they used to be DRM_DEBUG_KMS messages for link failures so never got
caught by CI. But now this patch actually throws DRM_ERROR if the link
training fails at RBR and 1 lane. So it caught these link train failures.

There were two failures:
1. On SKL 6700k this was because the machine in CI lab is a SKL desktop
without eDP on Port A. But our VBT initialization code in the driver writes
VBT defaults in a way that it always sets DP flag on Port A and this does
not get cleared after parsing the VBT outputs. This has been fixed in
commit id (bb1d132935c2f8 "drm/i915/vbt: split out defaults that are set
when there is no VBT) and (665788572c6410b "drm/i915/vbt: don't propagate
errors from intel_bios_init())

2. On ILK-650 desktop - This was happening because of a bad monitor desktop
combination. I switched the monitor in the CI lab and that helped get rid
of the link failures on ILK system.

v10:
* Rebase on drm-tip and resend after revert
v9:
* Use the trimmed max values of link rate/lane count based on
link train fallback (Daniel Vetter)
v8:
* Set link_status to BAD first and then call mode_valid (Jani Nikula)
v7:
Remove the redundant variable in previous patch itself
v6:
* Obtain link rate index from fallback_link_rate using
the helper intel_dp_link_rate_index (Jani Nikula)
* Include fallback within intel_dp_start_link_train (Jani Nikula)
v5:
* Move set link status to drm core (Daniel Vetter, Jani Nikula)
v4:
* Add fallback support for non DDI platforms too
* Set connector->link status inside set_link_status function
(Jani Nikula)
v3:
* Set link status property to BAd unconditionally (Jani Nikula)
* Dont use two separate variables link_train_failed and link_status
to indicate same thing (Jani Nikula)
v2:
* Squashed a few patches (Jani Nikula)

Acked-by: Tony Cheng <tony.cheng@amd.com>
Acked-by: Harry Wentland <Harry.wentland@amd.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Ville Syrjala <ville.syrjala@linux.intel.com>
Signed-off-by: Manasi Navare <manasi.d.navare@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/16ca48b1e74c618929245e9a085b9e3483c3a16d.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915: Perform link quality check unconditionally during long pulse
Ville Syrjälä [Wed, 12 Apr 2017 19:30:17 +0000 (22:30 +0300)]
drm/i915: Perform link quality check unconditionally during long pulse

Apparently some DP sinks are a little nuts and cause HPD to drop
intermittently during modesets. This happens eg. on an ASUS PB287Q.
In oder to recover from this we can't really use the previous
connector status to determine if the link needs retraining, so let's
just ignore that piece of information and do the retrain
unconditionally. We do of course still check whether the link is
supposed to be running or not.

To actually get read out the EDID and update things properly we
also need to nuke the goto out added by commit 7d23e3c37bb3
("drm/i915: Cleaning up intel_dp_hpd_pulse"). I'm actually not sure
why that was there. Perhaps to avoid an EDID read if the connector
status didn't appear to change, but that sort of thing is quite racy
and would have failed anyway if we failed to keep up with the
hotplugs (if we missed the HPD down in between two HPD ups). And
now that we take this codepath unconditionally we definitely need
to drop the goto as otherwise we would never do the EDID read.

v2: Drop the goto that made us skip EDID reads entirely. Doh!
v3: Rebase due to locking changes
    s/apparely/apparently/ in the comment (Chris)

Cc: stable@vger.kernel.org
Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Reported-by: Palmer Dabbelt <palmer@dabbelt.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99766
References: https://lists.freedesktop.org/archives/intel-gfx/2017-February/119779.html
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170412193017.21029-1-ville.syrjala@linux.intel.com
7 years agodrm/i915/guc: write wopcm related register once during uc init
daniele.ceraolospurio@intel.com [Fri, 7 Apr 2017 00:18:52 +0000 (17:18 -0700)]
drm/i915/guc: write wopcm related register once during uc init

The wopcm registers are write-once, so any write after the first one
will just be ignored. The registers survive a GPU reset but not
always a suspend/resume cycle, so to keep things simple keep the
writes in the intel_uc_init_hw function instead of moving it earlier
to make sure we attempt them every time we try to load GuC.

Cc: Jeff McGee <jeff.mcgee@intel.com>
Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491524332-23860-1-git-send-email-daniele.ceraolospurio@intel.com
7 years agodrm/i915: Fix use after free in lpe_audio_platdev_destroy()
Chris Wilson [Wed, 12 Apr 2017 08:02:51 +0000 (09:02 +0100)]
drm/i915: Fix use after free in lpe_audio_platdev_destroy()

[31908.547136] BUG: KASAN: use-after-free in intel_lpe_audio_teardown+0x78/0xb0 [i915] at addr ffff8801f7788358
[31908.547297] Read of size 8 by task drv_selftest/3781
[31908.547405] CPU: 0 PID: 3781 Comm: drv_selftest Tainted: G    BU  W       4.10.0+ #451
[31908.547553] Hardware name:                  /        , BIOS PYBSWCEL.86A.0027.2015.0507.1758 05/07/2015
[31908.547682] Call Trace:
[31908.547772]  dump_stack+0x68/0x9f
[31908.547857]  kasan_object_err+0x1c/0x70
[31908.547947]  kasan_report_error+0x1f1/0x4f0
[31908.548038]  ? kfree+0xaa/0x170
[31908.548121]  kasan_report+0x34/0x40
[31908.548211]  ? klist_children_get+0x20/0x30
[31908.548472]  ? intel_lpe_audio_teardown+0x78/0xb0 [i915]
[31908.548567]  __asan_load8+0x5e/0x70
[31908.548824]  intel_lpe_audio_teardown+0x78/0xb0 [i915]
[31908.549080]  intel_audio_deinit+0x28/0x80 [i915]
[31908.549315]  i915_driver_unload+0xe4/0x360 [i915]
[31908.549551]  ? i915_driver_load+0x1d70/0x1d70 [i915]
[31908.549651]  ? trace_hardirqs_on+0xd/0x10
[31908.549885]  i915_pci_remove+0x23/0x30 [i915]
[31908.549978]  pci_device_remove+0x5c/0x100
[31908.550069]  device_release_driver_internal+0x1db/0x2e0
[31908.550165]  driver_detach+0x68/0xc0
[31908.550256]  bus_remove_driver+0x8b/0x150
[31908.550346]  driver_unregister+0x3e/0x60
[31908.550439]  pci_unregister_driver+0x1d/0x110
[31908.550531]  ? find_module_all+0x7a/0xa0
[31908.550791]  i915_exit+0x1a/0x87 [i915]
[31908.550881]  SyS_delete_module+0x264/0x2c0
[31908.550971]  ? free_module+0x430/0x430
[31908.551064]  ? trace_hardirqs_off_caller+0x16/0x110
[31908.551159]  ? trace_hardirqs_on_caller+0x16/0x280
[31908.551256]  ? trace_hardirqs_on_thunk+0x1a/0x1c
[31908.551350]  entry_SYSCALL_64_fastpath+0x1c/0xb1
[31908.551440] RIP: 0033:0x7f1d67312ec7
[31908.551520] RSP: 002b:00007ffebe34e888 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0
[31908.551650] RAX: ffffffffffffffda RBX: ffffffff811123f6 RCX: 00007f1d67312ec7
[31908.551743] RDX: 0000000000000000 RSI: 0000000000000800 RDI: 0000560d0af476b8
[31908.551837] RBP: ffff880233d87f98 R08: 0000000000000000 R09: 00007ffebe34e8b8
[31908.551930] R10: 00007f1d68adf8c0 R11: 0000000000000206 R12: 0000000000000000
[31908.552023] R13: 0000560d0af46440 R14: 0000000000000034 R15: 00007ffebe34d860
[31908.552121]  ? trace_hardirqs_off_caller+0x16/0x110
[31908.552217] Object at ffff8801f7788000, in cache kmalloc-2048 size: 2048
[31908.552306] Allocated:
[31908.552377] PID = 3781
[31908.552456]  save_stack_trace+0x16/0x20
[31908.552539]  kasan_kmalloc+0xee/0x190
[31908.552627]  __kmalloc+0xdb/0x1b0
[31908.552713]  platform_device_alloc+0x27/0x90
[31908.552804]  platform_device_register_full+0x36/0x220
[31908.553066]  intel_lpe_audio_init+0x41e/0x570 [i915]
[31908.553320]  intel_audio_init+0xd/0x40 [i915]
[31908.553552]  i915_driver_load+0x13f5/0x1d70 [i915]
[31908.553788]  i915_pci_probe+0x65/0xe0 [i915]
[31908.553881]  pci_device_probe+0xda/0x140
[31908.553969]  driver_probe_device+0x400/0x660
[31908.554058]  __driver_attach+0x11c/0x120
[31908.554147]  bus_for_each_dev+0xe6/0x150
[31908.554237]  driver_attach+0x26/0x30
[31908.554325]  bus_add_driver+0x26b/0x3b0
[31908.554412]  driver_register+0xce/0x190
[31908.554502]  __pci_register_driver+0xaf/0xc0
[31908.554589]  0xffffffffa0550063
[31908.554675]  do_one_initcall+0x8b/0x1e0
[31908.554764]  do_init_module+0x102/0x325
[31908.554852]  load_module+0x3aad/0x45e0
[31908.554944]  SyS_finit_module+0x169/0x1a0
[31908.555033]  entry_SYSCALL_64_fastpath+0x1c/0xb1
[31908.555119] Freed:
[31908.555188] PID = 3781
[31908.555266]  save_stack_trace+0x16/0x20
[31908.555349]  kasan_slab_free+0xb0/0x180
[31908.555436]  kfree+0xaa/0x170
[31908.555520]  platform_device_release+0x76/0x80
[31908.555610]  device_release+0x45/0xe0
[31908.555698]  kobject_put+0x11f/0x260
[31908.555785]  put_device+0x12/0x20
[31908.555871]  platform_device_unregister+0x1b/0x20
[31908.556135]  intel_lpe_audio_teardown+0x5c/0xb0 [i915]
[31908.556390]  intel_audio_deinit+0x28/0x80 [i915]
[31908.556622]  i915_driver_unload+0xe4/0x360 [i915]
[31908.556858]  i915_pci_remove+0x23/0x30 [i915]
[31908.556948]  pci_device_remove+0x5c/0x100
[31908.557037]  device_release_driver_internal+0x1db/0x2e0
[31908.557129]  driver_detach+0x68/0xc0
[31908.557217]  bus_remove_driver+0x8b/0x150
[31908.557304]  driver_unregister+0x3e/0x60
[31908.557394]  pci_unregister_driver+0x1d/0x110
[31908.557653]  i915_exit+0x1a/0x87 [i915]
[31908.557741]  SyS_delete_module+0x264/0x2c0
[31908.557834]  entry_SYSCALL_64_fastpath+0x1c/0xb1
[31908.557919] Memory state around the buggy address:
[31908.558005]  ffff8801f7788200: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[31908.558127]  ffff8801f7788280: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[31908.558255] >ffff8801f7788300: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[31908.558374]                                                     ^
[31908.558467]  ffff8801f7788380: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[31908.558595]  ffff8801f7788400: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb

v2: Just leak the memory (8 bytes) as freeing it ourselves is not safe,
and we need to coordinate a proper fix in platform_device itself.

Fixes: eef57324d926 ("drm/i915: setup bridge for HDMI LPE audio driver")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99952
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Cc: Jerome Anand <jerome.anand@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Takashi Iwai <tiwai@suse.de>
Link: http://patchwork.freedesktop.org/patch/msgid/20170412080251.30648-1-chris@chris-wilson.co.uk
Reviewed-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
7 years agodrm/i915/dp: Validate cached link rate and lane count before retraining
Manasi Navare [Thu, 6 Apr 2017 21:00:12 +0000 (14:00 -0700)]
drm/i915/dp: Validate cached link rate and lane count before retraining

Currently intel_dp_check_link_status() tries to retrain the link if
Clock recovery or Channel EQ for any of the lanes indicated by
intel_dp->lane_count is not set. However these values cached in intel_dp
structure can be stale if link training has failed for these values
during previous modeset. Or these values can get stale since we have
now re read the DPCD registers or it can be 0 in case of connected boot
case.

This patch validates these values against the max link rate and max lane
count values.

This is absolutely required incase the common_rates or max lane count
are now different due to link fallback.

v2:
* Include the FIXME commnet inside the function (Ville Syrjala)
* Remove the redundant parenthesis (Ville Syrjala)

v3 by Jani:
* rebase on the DP refactoring series
* rename intel_dp_link_params_is_valid to intel_dp_link_params_valid
* minor stylistic changes

v4:
* Compare the link rate against max link rate not the
common_rates since common_rates does not account for the
lowered fallback link rate value. (Ville Syrjala)

v5:
* Fixed a warning for unused variable (Manasi)

Cc: Ville Syrjala <ville.syrjala@linux.intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491512412-30016-1-git-send-email-manasi.d.navare@intel.com
7 years agodrm/i915: Wake device for emitting request during selftest
Chris Wilson [Tue, 11 Apr 2017 23:44:27 +0000 (00:44 +0100)]
drm/i915: Wake device for emitting request during selftest

igt_mmap_offset_exhaustion() selftest was using live requests to make an
object busy, but we did not hold a runtime pm wakeref for submitting the
requests. Acquire it to avoid triggering "RPM wakelock ref not held
during HW access" warnings.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411234427.14841-3-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
7 years agodrm/i915: Pretend the engine is always idle when mocking
Chris Wilson [Tue, 11 Apr 2017 23:44:26 +0000 (00:44 +0100)]
drm/i915: Pretend the engine is always idle when mocking

If we have a mock engine and it has no more requests in flight, report
it as idle as there is no hardware to contradict us! Otherwise we
attempt to query the hw that doesn't exist and find that the hw hasn't
set its idle bit and we get upset.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411234427.14841-2-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
7 years agodrm/i915: Add stub mmio read/write routines to mock device
Chris Wilson [Wed, 12 Apr 2017 09:21:43 +0000 (10:21 +0100)]
drm/i915: Add stub mmio read/write routines to mock device

Provide dummy function pointers for the mock device in case we do hit
mmio during testing.

v2: Use ASSIGN_READ/WRITE_MMIO_FUNCS macros

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170412092143.3822-1-chris@chris-wilson.co.uk
7 years agodrm/i915: Treat WC a separate cache domain
Chris Wilson [Wed, 12 Apr 2017 11:01:11 +0000 (12:01 +0100)]
drm/i915: Treat WC a separate cache domain

When discussing a new WC mmap, we based the interface upon the
assumption that GTT was fully coherent. How naive! Commits 3b5724d702ef
("drm/i915: Wait for writes through the GTT to land before reading
back") and ed4596ea992d ("drm/i915/guc: WA to address the Ringbuffer
coherency issue") demonstrate that writes through the GTT are indeed
delayed and may be overtaken by direct WC access. To be safe, if
userspace is mixing WC mmaps with other potential GTT access (pwrite,
GTT mmaps) it should use set_domain(WC).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96563
Testcase: igt/gem_pwrite/small-gtt*
Testcase: igt/drv_selftest/coherency
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170412110111.26626-2-chris@chris-wilson.co.uk
7 years agodrm/i915: Combine write_domain flushes to a single function
Chris Wilson [Wed, 12 Apr 2017 11:01:10 +0000 (12:01 +0100)]
drm/i915: Combine write_domain flushes to a single function

In the next patch, we will introduce a new cache domain for
differentiating between GTT access and direct WC access. This will
require us to include WC in our write_domain flushes. Rather than
duplicate a third function, combine the existing two into one and
flushing WC writes will then be automatically handled as well.

v2: Be smarter and clearer by passing in the write domains to flush (Joonas)
v3: One missed ~ in v2 conversion

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170412110111.26626-1-chris@chris-wilson.co.uk
7 years agodrm/i915: Do not use lock all in hsw_trans_edp_pipe_A_crc_wa
Maarten Lankhorst [Tue, 4 Apr 2017 13:24:57 +0000 (15:24 +0200)]
drm/i915: Do not use lock all in hsw_trans_edp_pipe_A_crc_wa

There is no need to acquire all locks here,
doing a commit after forcing a modeset on the affected crtc
is enough. Any other locks needed will be acquired as needed.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491312297-18673-1-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agoRevert "drm/i915: Lock mode_config.mutex in intel_display_resume."
Maarten Lankhorst [Tue, 4 Apr 2017 13:22:48 +0000 (15:22 +0200)]
Revert "drm/i915: Lock mode_config.mutex in intel_display_resume."

This reverts commit ea49c9acf2db7082f0406bb3a570cc6bad37082b.

mode_config.mutex was originally added to fix WARNs in connector
functions, but now that atomic nonblocking modeset support is
included, we will likely never hold any any lock at all.

The WARN mentioned in commit bbf35e9defb9a6d1 ("drm/i915:
Pass atomic state to intel_audio_codec_enable, v2."), so it's
safe to revert this now.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491312168-18147-1-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Convert intel DVO connector to atomic
Maarten Lankhorst [Mon, 10 Apr 2017 09:07:11 +0000 (11:07 +0200)]
drm/i915: Convert intel DVO connector to atomic

No properties are supported, so just use the helper and reject everything.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491815239-10685-6-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Convert intel_crt connector properties to atomic.
Maarten Lankhorst [Mon, 10 Apr 2017 09:07:10 +0000 (11:07 +0200)]
drm/i915: Convert intel_crt connector properties to atomic.

No properties are supported, so just use the helper and reject
everything.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491815239-10685-5-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Convert intel_dp_mst connector properties to atomic.
Maarten Lankhorst [Mon, 10 Apr 2017 09:07:09 +0000 (11:07 +0200)]
drm/i915: Convert intel_dp_mst connector properties to atomic.

MST doesn't support setting any properties, but it should still
use the atomic helper for setting properties.

Only path and tile properties are supported (read-only).
Those are immutable, and handled by drm core.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491815239-10685-4-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Remove unused dp properties for dp-mst.
Maarten Lankhorst [Mon, 10 Apr 2017 10:51:10 +0000 (12:51 +0200)]
drm/i915: Remove unused dp properties for dp-mst.

Those properties are not hooked up on MST and were ignored. Best not expose them at all.
Without this the next patch fails to start on X.org, because the DP-MST properties could
not be read.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/751b85a0-81cd-09e2-9e60-6d4ddbf1c6ac@linux.intel.com
Testcase: kms_properties
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Convert intel_tv connector properties to atomic, v5.
Maarten Lankhorst [Mon, 10 Apr 2017 09:07:08 +0000 (11:07 +0200)]
drm/i915: Convert intel_tv connector properties to atomic, v5.

intel_tv has properties that are handled in the atomic core, but
needs a modeset to update the properties inside the connector.

The detect(), get_mode() and mode_valid() probe callbacks also
depend on the connector state, which made this a good connector
to convert first. It helped find all the issues when converting
connectors to atomic.

Because of these requirements, connector atomic_check() was added
and connection_mutex is held during probing. The diffstat looks
more favorable now. :)

Changes since v1:
- Add intel_encoder->swap_state to allow updating connector state.
- Add intel_tv->format for detect_mode and mode_valid, updated on atomic commit.
Changes since v2:
- Fix typo in tv_choose_preferred modes function name.
- Assignment of tv properties is done in core, so intel_tv only needs
  a atomic_check function. Thanks Ville!
Changes since v3:
- connection_mutex is now held in mode_valid() and get_modes(),
  this removes the need for caching parts of the connector_state.
Changes since v4:
- Use the new atomic connector check function.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491815239-10685-3-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Remove unused members from intel_tv.c
Maarten Lankhorst [Mon, 10 Apr 2017 09:07:07 +0000 (11:07 +0200)]
drm/i915: Remove unused members from intel_tv.c

They have been unused since 2010, after the code for
intel_tv_save/restore was removed in the below commit:

commit 6443170f6d862a1cc89e61e4bb2410b714b875f4
Author: Eric Anholt <eric@anholt.net>
Date:   Fri Apr 2 15:24:27 2010 -0700

    drm/i915: Remove dead KMS encoder save/restore code.

    This was brought over from UMS, and used for a while until we decided
    that drm_helper_resume_force_mode was easier and more reliable, since
    it didn't require duplicating all the code deleted here.  We just
    forgot to delete all that junk for a while.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491815239-10685-2-git-send-email-maarten.lankhorst@linux.intel.com
[mlankhorst: Add commit blurb based on danvet's feedback.]

7 years agodrm/i915/execlists: Document runtime pm for intel_lrc_irq_handler()
Chris Wilson [Tue, 11 Apr 2017 17:58:50 +0000 (18:58 +0100)]
drm/i915/execlists: Document runtime pm for intel_lrc_irq_handler()

We indirectly hold the runtime-pm for the intel_lrc_irq_handler() by
virtue of dev_priv->gt.awake keeping a wakeref whilst the requests are
busy. As this is not obvious from the code, add a comment.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411175850.2470-1-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
7 years agoMerge airlied/drm-next into drm-intel-next-queued
Daniel Vetter [Wed, 12 Apr 2017 08:07:43 +0000 (10:07 +0200)]
Merge airlied/drm-next into drm-intel-next-queued

Maarten needs both the new connector->atomic_check hook and the
connection_mutex locking changes in the probe helpers to be able to
start merging the connector property conversion to atomic.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
7 years agodrm/i915: Lie and treat all engines as idle if wedged
Chris Wilson [Tue, 11 Apr 2017 19:00:42 +0000 (20:00 +0100)]
drm/i915: Lie and treat all engines as idle if wedged

Similar to commit 8490ae207f1d ("drm/i915: Suppress busy status for
engines if wedged") we also want to report intel_engine_is_idle() as
true as well as the main intel_engines_are_idle(), as we now check that
the engines are idle when overwriting the HWS page. This is not true
whilst we are setting the device as wedged, at least according to our
bookkeeping, so we have to lie to ourselves!

[  383.588601] [drm:i915_reset [i915]] *ERROR* Failed to reset chip: -110
[  383.588685] ------------[ cut here ]------------
[  383.588755] WARNING: CPU: 0 PID: 12 at drivers/gpu/drm/i915/intel_engine_cs.c:226 intel_engine_init_global_seqno+0x222/0x290 [i915]
[  383.588757] WARN_ON(!intel_engine_is_idle(engine))
[  383.588759] Modules linked in: ctr ccm snd_hda_codec_hdmi snd_hda_codec_conexant snd_hda_codec_generic snd_hda_intel snd_hda_codec snd_hda_core arc4 iwldvm mac80211 snd_pcm snd_hwdep snd_seq_midi snd_seq_midi_event rfcomm bnep snd_rawmidi intel_powerclamp coretemp dm_multipath iwlwifi crct10dif_pclmul snd_seq crc32_pclmul ghash_clmulni_intel btusb aesni_intel btrtl btbcm aes_x86_64 crypto_simd cryptd btintel snd_timer glue_helper bluetooth intel_ips snd_seq_device cfg80211 snd soundcore binfmt_misc mei_me mei dm_mirror dm_region_hash dm_log i915 intel_gtt i2c_algo_bit drm_kms_helper cfbfillrect syscopyarea cfbimgblt sysfillrect sysimgblt fb_sys_fops cfbcopyarea prime_numbers ahci libahci drm e1000e
[  383.588851] CPU: 0 PID: 12 Comm: migration/0 Not tainted 4.11.0-rc5+ #207
[  383.588853] Hardware name: LENOVO 514328U/514328U, BIOS 6QET44WW (1.14 ) 04/20/2010
[  383.588855] Call Trace:
[  383.588866]  dump_stack+0x63/0x90
[  383.588871]  __warn+0xc7/0xf0
[  383.588876]  warn_slowpath_fmt+0x4a/0x50
[  383.588883]  ? set_next_entity+0x821/0x910
[  383.588943]  intel_engine_init_global_seqno+0x222/0x290 [i915]
[  383.588998]  __i915_gem_set_wedged_BKL+0xa4/0x190 [i915]
[  383.589003]  ? __switch_to+0x215/0x390
[  383.589008]  multi_cpu_stop+0xbb/0xe0
[  383.589012]  ? cpu_stop_queue_work+0x90/0x90
[  383.589016]  cpu_stopper_thread+0x82/0x110
[  383.589021]  smpboot_thread_fn+0x137/0x190
[  383.589026]  kthread+0xf7/0x130
[  383.589030]  ? sort_range+0x20/0x20
[  383.589034]  ? kthread_park+0x90/0x90
[  383.589040]  ret_from_fork+0x2c/0x40

Fixes: 2ca9faa551c4 ("drm/i915: Assert the engine is idle before overwiting the HWS")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411190042.25662-1-chris@chris-wilson.co.uk
7 years agodrm/i915: Use the engine class to get the context size
Daniele Ceraolo Spurio [Tue, 11 Apr 2017 10:11:12 +0000 (03:11 -0700)]
drm/i915: Use the engine class to get the context size

Technically speaking, the context size is per engine class, not per
instance.

v2: Add MISSING_CASE (Tvrtko)

v3: Rebased

v4: Restore the interface back to hiding the class lookup (Chris)

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491905472-16189-1-git-send-email-oscar.mateo@intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Bail if we do not setup the RCS engine
Chris Wilson [Tue, 11 Apr 2017 16:56:58 +0000 (17:56 +0100)]
drm/i915: Bail if we do not setup the RCS engine

In places, we assume that RCS exists. This has been true forever, but
let us catch this failure during bringup by adding an explicit check
that we do have an RCS engine.

v2: Make use of HAS_ENGINE (Tvrtko)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411165658.23828-1-chris@chris-wilson.co.uk
7 years agodrm/i915/dp: read sink count to a temporary variable first
Jani Nikula [Thu, 6 Apr 2017 13:44:17 +0000 (16:44 +0300)]
drm/i915/dp: read sink count to a temporary variable first

Don't clobber intel_dp->sink_count with the raw value.

Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/37d3222115172922fcd5ab038238359935bd561f.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/dp: use readb and writeb calls for single byte DPCD access
Jani Nikula [Thu, 6 Apr 2017 13:44:16 +0000 (16:44 +0300)]
drm/i915/dp: use readb and writeb calls for single byte DPCD access

This is what we have the readb and writeb variants for. Do some minor
return value and variable cleanup while at it.

Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/fd8a8f110bcfdc73a8c9241e5f9d61f7dd7c9677.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/dp: localize link rate index variable more
Jani Nikula [Thu, 6 Apr 2017 13:44:15 +0000 (16:44 +0300)]
drm/i915/dp: localize link rate index variable more

Localize link_rate_index to the if block, and rename to just index to
reduce indent.

Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/d348d990c96705427b93c1cac8c3e4447d06eebf.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/mst: use max link not sink lane count
Jani Nikula [Thu, 6 Apr 2017 13:44:14 +0000 (16:44 +0300)]
drm/i915/mst: use max link not sink lane count

The source might not support as many lanes as the sink, or the link
training might have failed at higher lane counts. Take these into
account.

Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/cf59530acafaf9258fb643d321ad251b44f34e29.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/dp: add functions for max common link rate and lane count
Jani Nikula [Thu, 6 Apr 2017 13:44:13 +0000 (16:44 +0300)]
drm/i915/dp: add functions for max common link rate and lane count

These are the theoretical maximums common for source and sink. These are
the maximums we should start with. They may be degraded in case of link
training failures, and the dynamic link values are stored separately.

Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/5088aca253c47dfa18251e1adb976aca1718f083.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/dp: don't call the link parameters sink parameters
Jani Nikula [Thu, 6 Apr 2017 13:44:12 +0000 (16:44 +0300)]
drm/i915/dp: don't call the link parameters sink parameters

If we modify these on the fly depending on the link conditions, don't
pretend they are sink properties.

Some link vs. sink confusion still remains, but we'll take care of them
in follow-up patches.

Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/3739b4fac502ebd1c6e075a62c1a195e4094eb16.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/dp: do not limit rate seek when not needed
Jani Nikula [Thu, 6 Apr 2017 13:44:11 +0000 (16:44 +0300)]
drm/i915/dp: do not limit rate seek when not needed

In link training fallback, we're trying to find a rate that we know is
in a sorted array of common link rates. We don't need to limit the array
using the max rate. For test request, the DP CTS doesn't say we should
limit the rate based on earlier fallback. This lets us get rid of
intel_dp_link_rate_index() and use intel_dp_rate_index() instead.

Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/33cab481a3228f31e938b5891a6285d892dcf272.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/dp: cache common rates with sink rates
Jani Nikula [Thu, 6 Apr 2017 13:44:10 +0000 (16:44 +0300)]
drm/i915/dp: cache common rates with sink rates

Now that source rates are static and sink rates are updated whenever
DPCD is updated, we can do and cache the intersection of them whenever
sink rates are updated. This reduces code complexity, as we don't have
to keep calling the functions to intersect. We also get rid of several
common rates arrays on stack.

Limiting the common rates by a max link rate can be done by picking the
first N elements of the cached common rates.

v2: get rid of the local common_rates variable (Manasi)
v3: don't clobber cached eDP rates on short pulse (Ville)

Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/e3b287e8cb6559b1f8fd4e80b78a8d22f1802eb7.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915/dp: use the sink rates array for max sink rates
Jani Nikula [Thu, 6 Apr 2017 13:44:09 +0000 (16:44 +0300)]
drm/i915/dp: use the sink rates array for max sink rates

Looking at DPCD DP_MAX_LINK_RATE may be completely bogus for eDP 1.4
which is allowed to use link rate select method and have 0 in max link
rate. With this change, it makes sense to store the max rate as the
actual rate rather than as a bw code.

Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/3e8baadb406d59f414cab36fed9f0b35d207fde5.1491485983.git.jani.nikula@intel.com
7 years agodrm/i915: Rename intel_engine_cs.exec_id to uabi_id
Chris Wilson [Tue, 11 Apr 2017 12:43:06 +0000 (13:43 +0100)]
drm/i915: Rename intel_engine_cs.exec_id to uabi_id

We want to refer to the index of the engine consistently throughout the
userspace ABI. We already have such an index through the execbuffer
engine specifier, that needs to be able to refer to each engine
specifically, so rename it the index to uabi_id to reflect its
generality beyond execbuf.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411124306.15448-1-chris@chris-wilson.co.uk
7 years agodrm/i915: Split the engine info table in two levels, using class + instance
Oscar Mateo [Mon, 10 Apr 2017 14:34:32 +0000 (07:34 -0700)]
drm/i915: Split the engine info table in two levels, using class + instance

There are some properties that logically belong to the engine class, and some
that belong to the engine instance. Make it explicit.

v2: Commit message (Tvrtko)

v3:
  - Rebased
  - Exec/uabi id should be per instance (Chris)

v4:
  - Rebased
  - Avoid re-ordering fields for smaller diff (Tvrtko)
  - Bug on oob access to the class array (Michal)

v5: Bug on the right thing (Michal)

v6: Rebased

Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491834873-9345-5-git-send-email-oscar.mateo@intel.com
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Generate the engine name based on the instance number
Oscar Mateo [Mon, 10 Apr 2017 14:34:31 +0000 (07:34 -0700)]
drm/i915: Generate the engine name based on the instance number

Not really needed, but makes the next change a little bit more compact.

v2:
  - Use zero-based numbering for engine names: xcs0, xcs1.. xcsN (Tvrtko, Chris)
  - Make sure the mock engine name is null-terminated (Tvrtko, Chris)

v3: Because I'm stupid (Chris)

v4: Verify engine name wasn't truncated (Michal)

v5:
  - Kill the warning in mock engine (Chris)

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491834873-9345-4-git-send-email-oscar.mateo@intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Use the same vfunc for BSD2 ring init
Oscar Mateo [Mon, 10 Apr 2017 14:34:30 +0000 (07:34 -0700)]
drm/i915: Use the same vfunc for BSD2 ring init

If we needed to do something different for the init functions, we could
always look at the engine instance to make the distinction. But, in any
case, the two functions are virtually identical already (please notice
that BSD2_RING is only used from gen8 onwards).

With this, the init functions depends excusively on the engine class
(a fact that we will use soon).

v2: Commit message

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491834873-9345-3-git-send-email-oscar.mateo@intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Classify the engines in class + instance
Daniele Ceraolo Spurio [Mon, 10 Apr 2017 14:34:29 +0000 (07:34 -0700)]
drm/i915: Classify the engines in class + instance

In such a way that vcs and vcs2 are just two different instances (0 and 1)
of the same engine class (VIDEO_DECODE_CLASS).

v2: Align the instance types (Tvrtko)

v3: Don't use enums for bspec-defined stuff (Michal)

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1491834873-9345-2-git-send-email-oscar.mateo@intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Use safer intel_uncore_wait_for_register in ring-init
Chris Wilson [Tue, 11 Apr 2017 10:13:40 +0000 (11:13 +0100)]
drm/i915: Use safer intel_uncore_wait_for_register in ring-init

While we do hold the forcewake for legacy ringbuffer initialisation, we
don't guard our access with the uncore.lock spinlock. In theory, we only
initialise when no others should be accessing the same mmio cachelines,
but in practice be safe as this is an infrequently used path and not
worth risky micro-optimisations.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411101340.31994-5-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
7 years agodrm/i915: Use __intel_uncore_wait_for_register_fw for sandybride_pcode_read
Chris Wilson [Tue, 11 Apr 2017 10:13:39 +0000 (11:13 +0100)]
drm/i915: Use __intel_uncore_wait_for_register_fw for sandybride_pcode_read

Since the sandybridge_pcode_read() may be called from
skl_pcode_request() inside an atomic context (with preempt disabled), we
should avoid hitting any sleeping paths. Currently is being called with
a 500ms timeout, irrespective of being inside an atomic context or not.
This is reduced down to 500us to play nice with the atomic context, and
that appears to be sufficient to keep BAT happy (we have a DRM_ERROR
should it timeout), i.e. we do not see any 500us pcode timeouts for
normal use. So leave it as a pure spin without having to introduce new
code paths to separate atomic/normal contexts.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411101340.31994-4-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
7 years agodrm/i915: Acquire uncore.lock over intel_uncore_wait_for_register()
Chris Wilson [Tue, 11 Apr 2017 10:13:38 +0000 (11:13 +0100)]
drm/i915: Acquire uncore.lock over intel_uncore_wait_for_register()

We acquire the forcewake and use I915_READ_FW instead for the atomic
wait within intel_uncore_wait_for_register. However, this still leaves
us vulnerable to concurrent mmio access to the register, which can cause
system hangs on gen7. The protection is to acquire uncore.lock around
each register, so lets add it back.

v2: Wrap __intel_wait_for_register_fw() to re-use its atomic wait_for
loop and spare adding another for ourselves.
v3: Add might_sleep() annotation

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411101340.31994-3-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
7 years agodrm/i915: Stop sleeping from inside gen6_bsd_submit_request()
Chris Wilson [Tue, 11 Apr 2017 10:13:37 +0000 (11:13 +0100)]
drm/i915: Stop sleeping from inside gen6_bsd_submit_request()

submit_request() is called from an atomic context, it's not allowed to
sleep. We have to be careful in our parameters to
intel_uncore_wait_for_register() to limit ourselves to the atomic wait
loop and not incur the wrath of our warnings.

Fixes: 6976e74b5fa1 ("drm/i915: Don't allow overuse of __intel_wait_for_register_fw()")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170410143807.22725-1-chris@chris-wilson.co.uk
Link: http://patchwork.freedesktop.org/patch/msgid/20170411101340.31994-2-chris@chris-wilson.co.uk
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
7 years agodrm/i915: Stop second guessing the caller for intel_uncore_wait_for_register()
Chris Wilson [Tue, 11 Apr 2017 11:27:05 +0000 (12:27 +0100)]
drm/i915: Stop second guessing the caller for intel_uncore_wait_for_register()

Allow the caller to use the fast_timeout_us to specify how long to wait
within the atomic section, rather than transparently switching to a
sleeping loop for larger values. This is required as some callsites may
need a long wait and are in an atomic section.

v2: Reinforce kerneldoc fast_timeout_us limit with a GEM_BUG_ON

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170411112705.12656-1-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
7 years agoMerge branch 'msm-next' of git://people.freedesktop.org/~robclark/linux into drm...
Dave Airlie [Mon, 10 Apr 2017 21:47:02 +0000 (07:47 +1000)]
Merge branch 'msm-next' of git://people.freedesktop.org/~robclark/linux into drm-next

Noteworthy changes this time:
1) 4k support for newer chips (ganging up hwpipes and mixers)
2) using OPP bindings for gpu
3) more prep work towards per-process pagetables

* 'msm-next' of git://people.freedesktop.org/~robclark/linux: (47 commits)
  msm/drm: gpu: Dynamically locate the clocks from the device tree
  drm/msm: gpu: Use OPP tables if we can
  drm/msm: Hard code the GPU "slow frequency"
  drm/msm: Add MSM_PARAM_GMEM_BASE
  drm/msm: Reference count address spaces
  drm/msm: Make sure to detach the MMU during GPU cleanup
  drm/msm/mdp5: Enable 3D mux in mdp5_ctl
  drm/msm/mdp5: Reset CTL blend registers before configuring them
  drm/msm/mdp5: Assign 'right' mixer to CRTC state
  drm/msm/mdp5: Stage border out on base stage if CRTC has 2 LMs
  drm/msm/mdp5: Stage right side hwpipes on Right-side Layer Mixer
  drm/msm/mdp5: Prepare Layer Mixers for source split
  drm/msm/mdp5: Configure 'right' hwpipe
  drm/msm/mdp5: Assign a 'right hwpipe' to plane state
  drm/msm/mdp5: Create mdp5_hwpipe_mode_set
  drm/msm/mdp5: Add optional 'right' Layer Mixer in CRTC state
  drm/msm/mdp5: Add a CAP for Source Split
  drm/msm/mdp5: Remove mixer/intf pointers from mdp5_ctl
  drm/msm/mdp5: Start using parameters from CRTC state
  drm/msm/mdp5: Add more stuff to CRTC state
  ...

7 years agoMerge tag 'drm-misc-next-2017-04-07' of git://anongit.freedesktop.org/git/drm-misc...
Dave Airlie [Mon, 10 Apr 2017 21:41:10 +0000 (07:41 +1000)]
Merge tag 'drm-misc-next-2017-04-07' of git://anongit.freedesktop.org/git/drm-misc into drm-next

Last drm-misc-next pull req for 4.12

Core changes:
 - fb_helper checkpatch cleanup and simplified _add_one_connector() (Thierry)
 - drm_ioctl and drm_sysfs improved/gained documentation (Daniel)
 - [ABI] Repurpose reserved field in drm_event_vblank for crtc_id (Ander)
 - Plumb acquire ctx through legacy paths to avoid lock_all and legacy_backoff
   (Daniel)
 - Add connector_atomic_check to check conn constraints on modeset (Maarten)
 - Add drm_of_find_panel_or_bridge to remove boilerplate in drivers (Rob)

Driver changes:
 - meson moved to drm-misc (Neil)
 - Added support for Amlogic GX SoCs in dw-hdmi (Neil)
 - Rockchip unbind actually cleans up the things bind initializes (Jeffy)
 - A couple misc fixes in virtio, dw-hdmi

NOTE: this also includes a backmerge of drm-next as well rc5 (we needed vmwgfx
      as well as the new synopsys media formats)

* tag 'drm-misc-next-2017-04-07' of git://anongit.freedesktop.org/git/drm-misc: (77 commits)
  Revert "drm: Don't allow interruptions when opening debugfs/crc"
  drm: Only take cursor locks when the cursor plane exists
  drm/vmwgfx: Fix fbdev emulation using legacy functions
  drm/rockchip: Shutdown all crtcs when unbinding drm
  drm/rockchip: Reorder drm bind/unbind sequence
  drm/rockchip: analogix_dp: Disable clock when unbinding
  drm/rockchip: vop: Unprepare clocks when unbinding
  drm/rockchip: vop: Enable pm domain before vop_initial
  drm/rockchip: cdn-dp: Don't unregister audio dev when unbinding
  drm/rockchip: cdn-dp: Don't try to release firmware when not loaded
  drm: bridge: analogix: Destroy connector & encoder when unbinding
  drm: bridge: analogix: Disable clock when unbinding
  drm: bridge: analogix: Unregister dp aux when unbinding
  drm: bridge: analogix: Detach panel when unbinding analogix dp
  drm: Don't allow interruptions when opening debugfs/crc
  drm/virtio: don't leak bo on drm_gem_object_init failure
  drm: bridge: dw-hdmi: fix input format/encoding from plat_data
  drm: omap: use common OF graph helpers
  drm: convert drivers to use drm_of_find_panel_or_bridge
  drm: convert drivers to use of_graph_get_remote_node
  ...

7 years agoBackmerge tag 'v4.11-rc6' into drm-next
Dave Airlie [Mon, 10 Apr 2017 21:40:42 +0000 (07:40 +1000)]
Backmerge tag 'v4.11-rc6' into drm-next

Linux 4.11-rc6

drm-misc needs 4.11-rc5, may as well fix conflicts with rc6.

7 years agoMerge branch 'drm-next-4.12' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Mon, 10 Apr 2017 21:31:18 +0000 (07:31 +1000)]
Merge branch 'drm-next-4.12' of git://people.freedesktop.org/~agd5f/linux into drm-next

Just some bug fixes and vega10 updates for 4.12.

* 'drm-next-4.12' of git://people.freedesktop.org/~agd5f/linux:
  drm/amdgpu: fix fence memory leak in wait_all_fence V2
  drm/amdgpu: fix "fix 64bit division"
  drm/amd/powerplay: add fan controller table v11 support.
  drm/amd/powerplay: port newest process pptable code for vega10.
  drm/amdgpu: set vm size and block size by individual gmc by default (v3)
  drm/amdgpu: Avoid overflows/divide-by-zero in latency_watermark calculations.
  drm/amdgpu: Make display watermark calculations more accurate
  drm/radeon: fix typo in bandwidth calculation
  drm/radeon: Refuse to migrate a prime BO to VRAM. (v2)
  drm/radeon: Maintain prime import/export refcount for BOs
  drm/amdgpu: Refuse to pin or change acceptable domains of prime BOs to VRAM. (v2)
  drm/amdgpu: Fail fb creation from imported dma-bufs. (v2)
  drm/radeon: Fail fb creation from imported dma-bufs.

7 years agoMerge tag 'drm-intel-testing-2017-04-03' of git://anongit.freedesktop.org/git/drm...
Dave Airlie [Mon, 10 Apr 2017 21:28:01 +0000 (07:28 +1000)]
Merge tag 'drm-intel-testing-2017-04-03' of git://anongit.freedesktop.org/git/drm-intel into drm-next

Last 4.12 feature pile:

GVT updates:
- Add mdev attribute group for per-vgpu info
- Time slice based vGPU scheduling QoS support (Gao Ping)
- Initial KBL support for E3 server (Han Xu)
- other misc.

i915:
- lots and lots of small fixes and improvements all over
- refactor fw_domain code (Chris Wilson)
- improve guc code (Oscar Mateo)
- refactor cursor/sprite code, precompute more for less overhead in
  the critical path (Ville)
- refactor guc/huc fw loading code a bit (Michal Wajdeczko)

* tag 'drm-intel-testing-2017-04-03' of git://anongit.freedesktop.org/git/drm-intel: (121 commits)
  drm/i915: Update DRIVER_DATE to 20170403
  drm/i915: Clear gt.active_requests before checking idle status
  drm/i915/uc: Drop use of MISSING_CASE on trivial enums
  drm/i915: make a few DDI functions static
  drm/i915: Combine reset_all_global_seqno() loops into one
  drm/i915: Remove redudant wait for each engine to idle from seqno wrap
  drm/i915: Wait for all engines to be idle as part of i915_gem_wait_for_idle()
  drm/i915: Move retire-requests into i915_gem_wait_for_idle()
  drm/i915/uc: Move fw path check to fetch_uc_fw()
  drm/i915/huc: Remove unused intel_huc_fini()
  drm/i915/uc: Add intel_uc_fw_fini()
  drm/i915/uc: Add intel_uc_fw_type_repr()
  drm/i915/uc: Move intel_uc_fw_status_repr() to intel_uc.h
  drivers: gpu: drm: i915L intel_lpe_audio: Fix kerneldoc comments
  drm/i915: Suppress busy status for engines if wedged
  drm/i915: Do request retirement before marking engines as wedged
  drm/i915: Drop verbose and archaic "ring" from our internal engine names
  drm/i915: Use a dummy timeline name for a signaled fence
  drm/i915: Ironlake do_idle_maps w/a may be called w/o struct_mutex
  drm/i915/guc: Take enable_guc_loading check out of GEM core code
  ...

7 years agoMerge branch 'drm-next-4.12' of https://github.com/ckhu-mediatek/linux.git-tags into...
Dave Airlie [Mon, 10 Apr 2017 20:32:11 +0000 (06:32 +1000)]
Merge branch 'drm-next-4.12' of https://github.com/ckhu-mediatek/linux.git-tags into drm-next

This series is MT2701 DRM support.

* 'drm-next-4.12' of https://github.com/ckhu-mediatek/linux.git-tags:
  drm/mediatek: add support for Mediatek SoC MT2701
  drm/mediatek: update DSI sub driver flow for sending commands to panel
  drm/mediatek: add non-continuous clock mode and EOT packet control
  drm/mediatek: add dsi transfer function
  drm/mediatek: add dsi interrupt control
  drm/mediatek: cleaning up and refine
  drm/mediatek: update display module connections
  drm/mediatek: add BLS component
  drm/mediatek: add shadow register support
  drm/mediatek: add *driver_data for different hardware settings
  drm/mediatek: add helpers for coverting from the generic components
  dt-bindings: display: mediatek: update supported chips

7 years agodrm/i915: Don't allow overuse of __intel_wait_for_register_fw()
Michal Wajdeczko [Mon, 10 Apr 2017 12:17:47 +0000 (12:17 +0000)]
drm/i915: Don't allow overuse of __intel_wait_for_register_fw()

This function should not be called with long timeouts in atomic context.
Annotate it as might_sleep if timeout is longer than 10us.

v2: fix comment (Michal)

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170410121747.209200-1-michal.wajdeczko@intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Drop const qualifiers from params in wait_for_register()
Michal Wajdeczko [Mon, 10 Apr 2017 09:38:17 +0000 (09:38 +0000)]
drm/i915: Drop const qualifiers from params in wait_for_register()

These params are passed by value, const qualifiers are ignored any way.
While around, unify timeout_ms type from long to int.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170410093817.151280-1-michal.wajdeczko@intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
7 years agodrm/i915: Use drm_i915_private directly from debugfs
Chris Wilson [Fri, 7 Apr 2017 19:42:20 +0000 (20:42 +0100)]
drm/i915: Use drm_i915_private directly from debugfs

The void *data passed to debugfs callbacks is actually the
drm_i915_private pointer, so use it thusly and avoid the to_i915(dev)
indirection.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170407194220.821-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
7 years agoLinux 4.11-rc6
Linus Torvalds [Sun, 9 Apr 2017 16:49:44 +0000 (09:49 -0700)]
Linux 4.11-rc6

7 years agoMerge branch 'for-next' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sun, 9 Apr 2017 16:10:02 +0000 (09:10 -0700)]
Merge branch 'for-next' of git://git.samba.org/sfrench/cifs-2.6

Pull CIFS fixes from Steve French:
 "This is a set of CIFS/SMB3 fixes for stable.

  There is another set of four SMB3 reconnect fixes for stable in
  progress but they are still being reviewed/tested, so didn't want to
  wait any longer to send these five below"

* 'for-next' of git://git.samba.org/sfrench/cifs-2.6:
  Reset TreeId to zero on SMB2 TREE_CONNECT
  CIFS: Fix build failure with smb2
  Introduce cifs_copy_file_range()
  SMB3: Rename clone_range to copychunk_range
  Handle mismatched open calls

7 years agoMerge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm
Linus Torvalds [Sun, 9 Apr 2017 16:05:25 +0000 (09:05 -0700)]
Merge branch 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm

Pull ARM fixes from Russell King:
 "A number of ARM fixes:

   - prevent oopses caused by dma_get_sgtable() and declared DMA
     coherent memory

   - fix boot failure on nommu caused by ID_PFR1 access

   - a number of kprobes fixes from Jon Medhurst and Masami Hiramatsu"

* 'fixes' of git://git.armlinux.org.uk/~rmk/linux-arm:
  ARM: 8665/1: nommu: access ID_PFR1 only if CPUID scheme
  ARM: dma-mapping: disallow dma_get_sgtable() for non-kernel managed memory
  arm: kprobes: Align stack to 8-bytes in test code
  arm: kprobes: Fix the return address of multiple kretprobes
  arm: kprobes: Skip single-stepping in recursing path if possible
  arm: kprobes: Allow to handle reentered kprobe on single-stepping

7 years agoMerge tag 'driver-core-4.11-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 9 Apr 2017 16:03:51 +0000 (09:03 -0700)]
Merge tag 'driver-core-4.11-rc6' of git://git./linux/kernel/git/gregkh/driver-core

Pull driver core fixes from Greg KH:
 "Here are 3 small fixes for 4.11-rc6.

  One resolves a reported issue with sysfs files that NeilBrown found,
  one is a documenatation fix for the stable kernel rules, and the last
  is a small MAINTAINERS file update for kernfs"

* tag 'driver-core-4.11-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  MAINTAINERS: separate out kernfs maintainership
  sysfs: be careful of error returns from ops->show()
  Documentation: stable-kernel-rules: fix stable-tag format

7 years agoMerge tag 'staging-4.11-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 9 Apr 2017 16:02:31 +0000 (09:02 -0700)]
Merge tag 'staging-4.11-rc6' of git://git./linux/kernel/git/gregkh/staging

Pull staging/IIO driver rfixes from Greg KH:
 "Here are a number of small IIO and staging driver fixes for 4.11-rc6.
  Nothing big here, just iio fixes for reported issues, and an ashmem
  fix for a very old bug that has been reported by a number of Android
  vendors"

* tag 'staging-4.11-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: android: ashmem: lseek failed due to no FMODE_LSEEK.
  iio: hid-sensor-attributes: Fix sensor property setting failure.
  iio: accel: hid-sensor-accel-3d: Fix duplicate scan index error
  iio: core: Fix IIO_VAL_FRACTIONAL_LOG2 for negative values
  iio: st_pressure: initialize lps22hb bootime
  iio: bmg160: reset chip when probing
  iio: cros_ec_sensors: Fix return value to get raw and calibbias data.

7 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Sun, 9 Apr 2017 15:26:21 +0000 (08:26 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs

Pull VFS fixes from Al Viro:
 "statx followup fixes and a fix for stack-smashing on alpha"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  alpha: fix stack smashing in old_adjtimex(2)
  statx: Include a mask for stx_attributes in struct statx
  statx: Reserve the top bit of the mask for future struct expansion
  xfs: report crtime and attribute flags to statx
  ext4: Add statx support
  statx: optimize copy of struct statx to userspace
  statx: remove incorrect part of vfs_statx() comment
  statx: reject unknown flags when using NULL path
  Documentation/filesystems: fix documentation for ->getattr()

7 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 8 Apr 2017 18:56:58 +0000 (11:56 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "Here's a pull request for 4.11-rc, fixing a set of issues mostly
  centered around the new scheduling framework. These have been brewing
  for a while, but split up into what we absolutely need in 4.11, and
  what we can defer until 4.12. These are well tested, on both single
  queue and multiqueue setups, and with and without shared tags. They
  fix several hangs that have happened in testing.

  This is obviously larger than I would have preferred at this point in
  time, but I don't think we can shave much off this and still get the
  desired results.

  In detail, this pull request contains:

   - a set of five fixes for NVMe, mostly from Christoph and one from
     Roland.

   - a series from Bart, fixing issues with dm-mq and SCSI shared tags
     and scheduling. Note that one of those patches commit messages may
     read like an optimization, but it is in fact an important fix for
     queue restarts in particular.

   - a series from Omar, most importantly fixing a hang with multiple
     hardware queues when we fail to get a driver tag. Another important
     fix in there is for resizing hardware queues, which nbd does when
     handling multiple sockets for one connection.

   - fixing an imbalance in putting the ctx for hctx request allocations
     from Minchan"

* 'for-linus' of git://git.kernel.dk/linux-block:
  blk-mq: Restart a single queue if tag sets are shared
  dm rq: Avoid that request processing stalls sporadically
  scsi: Avoid that SCSI queues get stuck
  blk-mq: Introduce blk_mq_delay_run_hw_queue()
  blk-mq: remap queues when adding/removing hardware queues
  blk-mq-sched: fix crash in switch error path
  blk-mq-sched: set up scheduler tags when bringing up new queues
  blk-mq-sched: refactor scheduler initialization
  blk-mq: use the right hctx when getting a driver tag fails
  nvmet: fix byte swap in nvmet_parse_io_cmd
  nvmet: fix byte swap in nvmet_execute_write_zeroes
  nvmet: add missing byte swap in nvmet_get_smart_log
  nvme: add missing byte swap in nvme_setup_discard
  nvme: Correct NVMF enum values to match NVMe-oF rev 1.0
  block: do not put mq context in blk_mq_alloc_request_hctx

7 years agoMerge tag 'pinctrl-v4.11-4' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Sat, 8 Apr 2017 18:43:38 +0000 (11:43 -0700)]
Merge tag 'pinctrl-v4.11-4' of git://git./linux/kernel/git/linusw/linux-pinctrl

Pull pin control fix from Linus Walleij:
 "This late fix for pin control is hopefully the last I send this cycle.

  The problem was detected early in the v4.11 release cycle and there
  has been some back and forth on how to solve it. Sadly the proper fix
  arrives late, but at least not too late.

  An issue was detected with pin control on the Freescale i.MX after the
  refactorings for more general group and function handling.

  We now have the proper fix for this"

* tag 'pinctrl-v4.11-4' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: core: Fix pinctrl_register_and_init() with pinctrl_enable()

7 years agoMerge tag 'powerpc-4.11-7' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sat, 8 Apr 2017 18:06:12 +0000 (11:06 -0700)]
Merge tag 'powerpc-4.11-7' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "Some more powerpc fixes for 4.11:

  Headed to stable:

   - disable HFSCR[TM] if TM is not supported, fixes a potential host
     kernel crash triggered by a hostile guest, but only in
     configurations that no one uses

   - don't try to fix up misaligned load-with-reservation instructions

   - fix flush_(d|i)cache_range() called from modules on little endian
     kernels

   - add missing global TLB invalidate if cxl is active

   - fix missing preempt_disable() in crc32c-vpmsum

  And a fix for selftests build changes that went in this release:

   - selftests/powerpc: Fix standalone powerpc build

  Thanks to: Benjamin Herrenschmidt, Frederic Barrat, Oliver O'Halloran,
  Paul Mackerras"

* tag 'powerpc-4.11-7' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/crypto/crc32c-vpmsum: Fix missing preempt_disable()
  powerpc/mm: Add missing global TLB invalidate if cxl is active
  powerpc/64: Fix flush_(d|i)cache_range() called from modules
  powerpc: Don't try to fix up misaligned load-with-reservation instructions
  powerpc: Disable HFSCR[TM] if TM is not supported
  selftests/powerpc: Fix standalone powerpc build

7 years agomm/mempolicy.c: fix error handling in set_mempolicy and mbind.
Chris Salls [Sat, 8 Apr 2017 06:48:11 +0000 (23:48 -0700)]
mm/mempolicy.c: fix error handling in set_mempolicy and mbind.

In the case that compat_get_bitmap fails we do not want to copy the
bitmap to the user as it will contain uninitialized stack data and leak
sensitive data.

Signed-off-by: Chris Salls <salls@cs.ucsb.edu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agosysctl: report EINVAL if value is larger than UINT_MAX for proc_douintvec
Liping Zhang [Fri, 7 Apr 2017 15:51:07 +0000 (23:51 +0800)]
sysctl: report EINVAL if value is larger than UINT_MAX for proc_douintvec

Currently, inputting the following command will succeed but actually the
value will be truncated:

  # echo 0x12ffffffff > /proc/sys/net/ipv4/tcp_notsent_lowat

This is not friendly to the user, so instead, we should report error
when the value is larger than UINT_MAX.

Fixes: e7d316a02f68 ("sysctl: handle error writing UINT_MAX to u32 fields")
Signed-off-by: Liping Zhang <zlpnobody@gmail.com>
Cc: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMAINTAINERS: separate out kernfs maintainership
Tejun Heo [Thu, 23 Mar 2017 17:34:47 +0000 (13:34 -0400)]
MAINTAINERS: separate out kernfs maintainership

Separate out kernfs from driver core and add myself as a
co-maintainer.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
7 years agosysfs: be careful of error returns from ops->show()
NeilBrown [Mon, 3 Apr 2017 01:30:34 +0000 (11:30 +1000)]
sysfs: be careful of error returns from ops->show()

ops->show() can return a negative error code.
Commit 65da3484d9be ("sysfs: correctly handle short reads on PREALLOC attrs.")
(in v4.4) caused this to be stored in an unsigned 'size_t' variable, so errors
would look like large numbers.
As a result, if an error is returned, sysfs_kf_read() will return the
value of 'count', typically 4096.

Commit 17d0774f8068 ("sysfs: correctly handle read offset on PREALLOC attrs")
(in v4.8) extended this error to use the unsigned large 'len' as a size for
memmove().
Consequently, if ->show returns an error, then the first read() on the
sysfs file will return 4096 and could return uninitialized memory to
user-space.
If the application performs a subsequent read, this will trigger a memmove()
with extremely large count, and is likely to crash the machine is bizarre ways.

This bug can currently only be triggered by reading from an md
sysfs attribute declared with __ATTR_PREALLOC() during the
brief period between when mddev_put() deletes an mddev from
the ->all_mddevs list, and when mddev_delayed_delete() - which is
scheduled on a workqueue - completes.
Before this, an error won't be returned by the ->show()
After this, the ->show() won't be called.

I can reproduce it reliably only by putting delay like
usleep_range(500000,700000);
early in mddev_delayed_delete(). Then after creating an
md device md0 run
  echo clear > /sys/block/md0/md/array_state; cat /sys/block/md0/md/array_state

The bug can be triggered without the usleep.

Fixes: 65da3484d9be ("sysfs: correctly handle short reads on PREALLOC attrs.")
Fixes: 17d0774f8068 ("sysfs: correctly handle read offset on PREALLOC attrs")
Cc: stable@vger.kernel.org
Signed-off-by: NeilBrown <neilb@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
7 years agoDocumentation: stable-kernel-rules: fix stable-tag format
Johan Hovold [Mon, 3 Apr 2017 13:53:34 +0000 (15:53 +0200)]
Documentation: stable-kernel-rules: fix stable-tag format

A patch documenting how to specify which kernels a particular fix should
be backported to (seemingly) inadvertently added a minus sign after the
kernel version. This particular stable-tag format had never been used
prior to this patch, and was neither present when the patch in question
was first submitted (it was added in v2 without any comment).

Drop the minus sign to avoid any confusion.

Fixes: fdc81b7910ad ("stable_kernel_rules: Add clause about specification of kernel versions to patch.")
Signed-off-by: Johan Hovold <johan@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
7 years agomsm/drm: gpu: Dynamically locate the clocks from the device tree
Jordan Crouse [Tue, 7 Mar 2017 17:02:56 +0000 (10:02 -0700)]
msm/drm: gpu: Dynamically locate the clocks from the device tree

Instead of using a fixed list of clock names use the clock-names
list in the device tree to discover and get the list of clocks
that we need.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm: gpu: Use OPP tables if we can
Jordan Crouse [Tue, 7 Mar 2017 17:02:55 +0000 (10:02 -0700)]
drm/msm: gpu: Use OPP tables if we can

If a OPP table is defined for the GPU device in the device tree use
that in lieu of the downstream style GPU frequency table. If we do
use the downstream table convert it to a OPP table so that we can
take advantage of the OPP lookup facilities later.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm: Hard code the GPU "slow frequency"
Jordan Crouse [Tue, 7 Mar 2017 17:02:54 +0000 (10:02 -0700)]
drm/msm: Hard code the GPU "slow frequency"

Some A3XX and A4XX GPU targets required that the GPU clock be
programmed to a non zero value when it was disabled so
27Mhz was chosen as the "invalid" frequency.

Even though newer targets do not have the same clock restrictions
we still write 27Mhz on clock disable and expect the clock subsystem
to round down to zero.

For unknown reasons even though the slow clock speed is always
27Mhz and it isn't actually a functional level the legacy device tree
frequency tables always defined it and then did gymnastics to work
around it.

Instead of playing the same silly games just hard code the "slow" clock
speed in the code as 27MHz and save ourselves a bit of infrastructure.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm: Add MSM_PARAM_GMEM_BASE
Jordan Crouse [Tue, 7 Mar 2017 17:02:53 +0000 (10:02 -0700)]
drm/msm: Add MSM_PARAM_GMEM_BASE

User space needs to know where the GMEM whole starts so that they
can set up the addressing correctly.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm: Reference count address spaces
Jordan Crouse [Tue, 7 Mar 2017 17:02:52 +0000 (10:02 -0700)]
drm/msm: Reference count address spaces

There are reasons for a memory object to outlive the file descriptor
that created it and so the address space that a buffer object is
attached to must also outlive the file descriptor. Reference count
the address space so that it can remain viable until all the objects
have released their addresses.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm: Make sure to detach the MMU during GPU cleanup
Jordan Crouse [Mon, 6 Feb 2017 17:39:29 +0000 (10:39 -0700)]
drm/msm: Make sure to detach the MMU during GPU cleanup

We should be detaching the MMU before destroying the address
space. To do this cleanly, the detach has to happen in
adreno_gpu_cleanup() because it needs access to structs
in adreno_gpu.c.  Plus it is better symmetry to have
the attach and detach at the same code level.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Enable 3D mux in mdp5_ctl
Archit Taneja [Thu, 23 Mar 2017 10:28:17 +0000 (15:58 +0530)]
drm/msm/mdp5: Enable 3D mux in mdp5_ctl

3D mux is a small block placed after the DSPPs in MDP5. It can merge
2 LM/DSPP outputs and feed it to a single interface.

Enable 3D Mux if our mdp5_pipeline has 2 active LMs. This check
will need to be made more specific later when we add Dual DSI
support with source split enabled. In that use case, each LM feeds to a
separae INTF, so the 3D mux isn't needed.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Reset CTL blend registers before configuring them
Archit Taneja [Thu, 23 Mar 2017 10:28:16 +0000 (15:58 +0530)]
drm/msm/mdp5: Reset CTL blend registers before configuring them

Assigning LMs dynamically to CRTCs results in REG_MDP5_CTL_LAYER_REGs
and REG_MDP5_CTL_LAYER_EXT_REGs maintaining old values for a LM that
isn't used by our CTL instance anymore.

Clear the ctl's CTL_LAYER_REG and CTL_LAYER_EXT_REGs for all LM
instances. The ones that need to be configured are configured later
in this func.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Assign 'right' mixer to CRTC state
Archit Taneja [Thu, 23 Mar 2017 10:28:15 +0000 (15:58 +0530)]
drm/msm/mdp5: Assign 'right' mixer to CRTC state

Dynamically assign a right mixer to mdp5_crtc_state in the CRTC's
atomic_check path. Assigning the right mixer has some constraints,
i.e, only a few LMs can be paired together. Update mdp5_mixer_assign
to handle these constraints.

Firstly, we need to identify whether we need a right mixer or not.
At the moment, there are 2 scenarios where a right mixer might be
needed:
- If any of the planes connected to this CRTC is too wide (i.e, is
  comprised of 2 hwpipes).
- If the CRTC's mode itself is too wide (i.e, a 4K mode on HDMI).

We implement both these checks in the mdp5_crtc_atomic_check(), and
pass 'need_right_mixer' to mdp5_setup_pipeline.

If a CRTC is already assigned a single mixer, and a new atomic commit
brings in a drm_plane that needs 2 hwpipes, we can successfully commit
this mode without requiring a full modeset, provided that we still use
the previously assigned mixer as the left mixer. If such an assignment
isn't possible, we'd need to do a full modeset. This scenario has been
ignored for now.

The mixer assignment code is a bit messy, considering we have at most
4 LM instances in hardware. This can probably be re-visited later with
simplified logic.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Stage border out on base stage if CRTC has 2 LMs
Archit Taneja [Thu, 23 Mar 2017 10:28:14 +0000 (15:58 +0530)]
drm/msm/mdp5: Stage border out on base stage if CRTC has 2 LMs

If a CRTC comprises of 2 LMs, it is mandatory to enable border out
and assign it to the base stage.

We had to enable border out also when the base plane wasn't fullscreen.
Club these checks and put them in a separate function called
get_start_stage() that returns the starting stage for assigning planes.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Stage right side hwpipes on Right-side Layer Mixer
Archit Taneja [Thu, 23 Mar 2017 10:28:13 +0000 (15:58 +0530)]
drm/msm/mdp5: Stage right side hwpipes on Right-side Layer Mixer

Now that our mdp5_planes can consist of 2 hwpipes, update the
blend_setup() code to stage the right hwpipe to the left and
right LMs

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Prepare Layer Mixers for source split
Archit Taneja [Thu, 23 Mar 2017 10:28:12 +0000 (15:58 +0530)]
drm/msm/mdp5: Prepare Layer Mixers for source split

In order to enable Source Split in HW, we need to add/modify
a few LM register configurations:

- Configure the LM width to be half the mode width, so that
  each LM manages one half of the scanout.
- Tell the 'right' LM that it is configured to be the 'right'
  LM in source split mode.
- Since we now have 2 places where REG_MDP5_LM_BLEND_COLOR_OUT is
  configured, do a read-update-store for the register instead of
  directly writing a value to it.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Configure 'right' hwpipe
Archit Taneja [Thu, 23 Mar 2017 10:28:11 +0000 (15:58 +0530)]
drm/msm/mdp5: Configure 'right' hwpipe

Now that we have a right hwpipe in mdp5_plane_state, configure it
mdp5_plane_mode_set(). The only parameters that vary between the
left and right hwpipes are the src_w, src_img_w, src_x and crtc_x
as we just even chop the fb into left and right halves.

Add a mdp5_plane_right_pipe() which will be used by the crtc code
to set up LM stages.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Assign a 'right hwpipe' to plane state
Archit Taneja [Thu, 23 Mar 2017 10:28:10 +0000 (15:58 +0530)]
drm/msm/mdp5: Assign a 'right hwpipe' to plane state

If the drm_plane has a source width that's greater than the max width
supported by a SSPP (2560 pixels on 8x96), then we assign a 'r_hwpipe'
to it in mdp5_plane_atomic_check().

TODO: There are a few scenarios where the hwpipe assignments aren't
recommended by HW. For example, an assignment which results in a
drm_plane to of two different types of hwpipes (say RGB0 on left
and DMA1 on right) is not recommended.
Also, hwpipes have a priority mapping, where the higher priority pipe
needs to be staged on left LM, and the lower priority needs to be
staged on the right LM. For example, the priority order for VIG pipes
in decreasing order of priority is VIG0, VIG1, VIG2, and VIG3. So, VIG0
on left and VIG1 on right is a correct configuration, but VIG1 on left
and VIG0 on right isn't. These scenarios are ignored for now for the
sake of simplicity.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Create mdp5_hwpipe_mode_set
Archit Taneja [Thu, 23 Mar 2017 10:28:09 +0000 (15:58 +0530)]
drm/msm/mdp5: Create mdp5_hwpipe_mode_set

Refactor mdp5_plane_mode_set to call mdp5_hwpipe_mode_set. The latter
func takes in only the hwpipe and the parameters that need to be
programmed into the hwpipe registers. All the code that calculates these
parameters is left as is in mdp5_plane_mode_set.

In the future, when we let drm_plane be comprised of 2 hwpipes, this func
allow us to configure each pipe without adding redundant code.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Add optional 'right' Layer Mixer in CRTC state
Archit Taneja [Thu, 23 Mar 2017 10:28:08 +0000 (15:58 +0530)]
drm/msm/mdp5: Add optional 'right' Layer Mixer in CRTC state

Add another mdp5_hw_mixer pointer (r_mixer) in mdp5_crtc_state.
This mixer will be used to generate the right half of the scanout.

With Source Split, a SSPP can now be connected to 2 Layer Mixers, but
has to be at the same blend level (stage #) on both Layer Mixers.

A drm_plane that has a lesser width than the max width supported, will
comprise of a single SSPP/hwpipe, staged on both the Layer Mixers at
the same blend level. A plane that is greater than max width will comprise
of 2 SSPPs, with the 'left' SSPP staged on the left LM, and the 'right'
SSPP staged on the right LM at the same blend level.

For now, the drm_plane consists of only one SSPP, therefore, it
needs to be staged on both the LMs in blend_setup() and mdp5_ctl_blend().
We'll extend this logic to support 2 hwpipes per plane later.

The crtc cursor ops (using the LM cursors, not SSPP cursors) simply
return an error if they're called when the right mixer is assigned to
the CRTC state. With source split is enabled, we're expected to only
SSPP cursors.

This commit adds code that configures the right mixer, but the r_mixer
itself isn't assigned at the moment.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Add a CAP for Source Split
Archit Taneja [Thu, 23 Mar 2017 10:28:07 +0000 (15:58 +0530)]
drm/msm/mdp5: Add a CAP for Source Split

Some of the newer MDP5 versions support Source Split of SSPPs. It is a
feature that allows us to route the output of a hwpipe to 2 Layer
Mixers. This is required to achieve the following use cases:

- Dual DSI: For high res DSI panels (such as 2560x1600 etc), a single
  DSI interface doesn't have the bandwidth to drive the required pixel
  clock. We use 2 DSI interfaces to drive the left and right halves
  of the panel (i.e, 1280x1600 each). The MDP5 pipeline here would look
  like:

         LM0 -- DSPP0 -- INTF1 -- DSI1
        /
hwpipe--
        \
         LM1 -- DSPP1 -- INTF2 -- DSI2

  A single hwpipe is used to scan out the left and right halves to DSI1
  and DSI2 respectively. In order to do this, we need to configure the
  2 Layer Mixers in Source Split mode.

- HDMI 4K: In order to support resolutions with width higher than the
  max width supported by a hwpipe, we club 2 hwpipes together:

hwpipe1 --- LM0 -- DSPP0
       -   -             \
         -                -- 3D Mux -- INTF0 -- HDMI
       -   -             /
hwpipe2 --- LM1 -- DSPP1

  hwpipe1 is staged on the 'left' Layer Mixer, and hwpipe2 is staged on
  the 'right' Layer Mixer. An additional block called the '3D Mux' is
  used to merge the output of the 2 DSPPs to a single interface.
  In this use case, it is possible that a 4K surface is downscaled and
  placed completely within one of the halves. In order to support such
  scenarios (and keep the programming simple), Layer Mixers with Source
  Split can be assigned 2 hw pipes per stage. While scanning out, the HW
  takes care of fetching the pixels fom the correct pipe.

Add a MDP cap to tell whether the HW supports source split or not.
Add a MDP LM cap that tells whether a LM instance can operate in
source split mode (and generate the 'left' part of the display
output).

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Remove mixer/intf pointers from mdp5_ctl
Archit Taneja [Thu, 23 Mar 2017 10:28:06 +0000 (15:58 +0530)]
drm/msm/mdp5: Remove mixer/intf pointers from mdp5_ctl

These are a part of CRTC state, it doesn't feel nice to leave them
hanging in mdp5_ctl struct. Pass mdp5_pipeline pointer instead
wherever it is needed.

We still have some params in mdp5_ctl like start_mask etc which
are derivative of atomic state, and should be rolled back if
a commit fails, but it doesn't seem to cause much trouble.

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
7 years agodrm/msm/mdp5: Start using parameters from CRTC state
Archit Taneja [Thu, 23 Mar 2017 10:28:05 +0000 (15:58 +0530)]
drm/msm/mdp5: Start using parameters from CRTC state

In the last few commits, we've been adding params to mdp5_crtc_state, and
assigning them in the atomic_check() funcs. Now it's time to actually
start using them.

Remove the duplicated params from the mdp5_crtc struct, and start using
them in the mdp5_crtc code. The majority of the references to these params
is in code that executes after the atomic swap has occurred, so it's okay
to use crtc->state in them. There are a couple of legacy LM cursor ops that
may not use the updated state, but (I think) it's okay to live with that.

Now that we dynamically allocate a mixer to the CRTC, we can also remove
the static assignment to it in mdp5_crtc_init, and also drop the code that
skipped init-ing WB bound mixers (those will now be rejected by
mdp5_mixer_assign()).

Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>