drm/i915: Introduce i915_address_space.mutex

Add a mutex into struct i915_address_space to be used while operating on
the vma and their lists for a particular vm. As this may be called from
the shrinker, we taint the mutex with fs_reclaim so that from the start
lockdep warns us if we are caught holding the mutex across an
allocation. (With such small steps we will eventually rid ourselves of
struct_mutex recursion!)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20180711073608.20286-2-chris@chris-wilson.co.uk

drm/i915: use the ICL stolen memory

Now that our stolen memory is already reserved by the x86 subsystem
(since commit "x86/gpu: reserve ICL's graphics stolen memory"), make
use of it.

Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: x86@kernel.org
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180504203252.28048-2-paulo.r.zanoni@intel.com

x86/gpu: reserve ICL's graphics stolen memory

ICL changes the registers and addresses to 64 bits.

I also briefly looked at implementing an u64 version of the PCI config
read functions, but I concluded this wouldn't be trivial, so it's not
worth doing it for a single user that can't have any racing problems
while reading the register in two separate operations.

v2:
 - Scrub the development (non-public) changelog (Joonas).
 - Remove the i915.ko bits so this can be easily backported in order
   to properly avoid stolen memory even on machines without i915.ko
   (Joonas).
 - CC stable for the reasons above.

Issue: VIZ-9250
CC: stable@vger.kernel.org
Cc: Ingo Molnar <mingo@kernel.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: x86@kernel.org
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Fixes: 412310019a20 ("drm/i915/icl: Add initial Icelake definitions.")
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180504203252.28048-1-paulo.r.zanoni@intel.com

drm/i915: Unwind HW init after GVT setup failure

Following intel_gvt_init() failure, we missed unwinding our setup
leaving pointers dangling past the module unload. For our example, the
pm_qos:

[  441.057615] top: 000000006b3baf1c, n: 0000000054d8ef33, p: 0000000097cdf1a2
               prev: 0000000054d8ef33, n: 0000000097cdf1a2, p: 000000006b3baf1c
               next: 0000000097cdf1a2, n: 000000006de8fc8b, p: 0000000081087253
[  441.057627] WARNING: CPU: 4 PID: 9277 at lib/plist.c:42 plist_check_prev_next+0x2d/0x40
[  441.057628] Modules linked in: i915(+) vgem snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic x86_pkg_temp_thermal intel_powerclamp coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel snd_hda_codec snd_hwdep snd_hda_core e1000e snd_pcm mei_me mei prime_numbers [last unloaded: i915]
[  441.057652] CPU: 4 PID: 9277 Comm: drv_selftest Tainted: G     U            4.18.0-rc4-CI-CI_DRM_4464+ #1
[  441.057653] Hardware name: System manufacturer System Product Name/Z170 PRO GAMING, BIOS 3402 04/26/2017
[  441.057656] RIP: 0010:plist_check_prev_next+0x2d/0x40
[  441.057657] Code: 08 48 39 f0 74 2b 49 89 f0 48 8b 4f 08 50 ff 32 52 48 89 fe 41 ff 70 08 48 8b 17 48 c7 c7 d8 ae 14 82 4d 8b 08 e8 63 0e 76 ff <0f> 0b 48 83 c4 20 c3 48 39 10 75 d0 f3 c3 0f 1f 44 00 00 41 54 55
[  441.057717] RSP: 0018:ffffc900003a3a68 EFLAGS: 00010082
[  441.057720] RAX: 0000000000000000 RBX: ffff8802193978c0 RCX: 0000000000000002
[  441.057721] RDX: 0000000080000002 RSI: ffffffff820c65a4 RDI: 00000000ffffffff
[  441.057722] RBP: ffff8802193978c0 R08: 0000000000000000 R09: 0000000000000001
[  441.057724] R10: ffffc900003a3a70 R11: 0000000000000000 R12: ffffffff82243de0
[  441.057725] R13: ffffffff82243de0 R14: ffff88021a6c78c0 R15: 0000000077359400
[  441.057726] FS:  00007fc23a4a9980(0000) GS:ffff880236d00000(0000) knlGS:0000000000000000
[  441.057728] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  441.057729] CR2: 0000563e4503d038 CR3: 0000000138f86005 CR4: 00000000003606e0
[  441.057730] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  441.057731] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  441.057732] Call Trace:
[  441.057736]  plist_check_list+0x2e/0x40
[  441.057738]  plist_add+0x23/0x130
[  441.057743]  pm_qos_update_target+0x1bd/0x2f0
[  441.057771]  i915_driver_load+0xec4/0x1060 [i915]
[  441.057775]  ? trace_hardirqs_on_caller+0xe0/0x1b0
[  441.057800]  i915_pci_probe+0x29/0x90 [i915]
[  441.057804]  pci_device_probe+0xa1/0x130
[  441.057807]  driver_probe_device+0x306/0x480
[  441.057810]  __driver_attach+0xdb/0x100
[  441.057812]  ? driver_probe_device+0x480/0x480
[  441.057813]  ? driver_probe_device+0x480/0x480
[  441.057816]  bus_for_each_dev+0x74/0xc0
[  441.057819]  bus_add_driver+0x15f/0x250
[  441.057821]  ? 0xffffffffa0696000
[  441.057823]  driver_register+0x56/0xe0
[  441.057825]  ? 0xffffffffa0696000
[  441.057827]  do_one_initcall+0x58/0x370
[  441.057830]  ? do_init_module+0x1d/0x1ea
[  441.057832]  ? rcu_read_lock_sched_held+0x6f/0x80
[  441.057834]  ? kmem_cache_alloc_trace+0x282/0x2e0
[  441.057838]  do_init_module+0x56/0x1ea
[  441.057841]  load_module+0x2435/0x2b20
[  441.057852]  ? __se_sys_finit_module+0xd3/0xf0
[  441.057854]  __se_sys_finit_module+0xd3/0xf0
[  441.057861]  do_syscall_64+0x55/0x190
[  441.057863]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  441.057865] RIP: 0033:0x7fc239d75839
[  441.057866] Code: 00 f3 c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 1f f6 2c 00 f7 d8 64 89 01 48
[  441.057927] RSP: 002b:00007fffb7825d38 EFLAGS: 00000246 ORIG_RAX: 0000000000000139
[  441.057930] RAX: ffffffffffffffda RBX: 0000563e45035dd0 RCX: 00007fc239d75839
[  441.057931] RDX: 0000000000000000 RSI: 0000563e4502f8a0 RDI: 0000000000000004
[  441.057932] RBP: 0000563e4502f8a0 R08: 0000000000000004 R09: 0000000000000000
[  441.057933] R10: 00007fffb7825ea0 R11: 0000000000000246 R12: 0000000000000000
[  441.057934] R13: 0000563e4502f690 R14: 0000000000000000 R15: 000000000000003f
[  441.057940] irq event stamp: 231338
[  441.057943] hardirqs last  enabled at (231337): [<ffffffff8193e3fc>] _raw_spin_unlock_irqrestore+0x4c/0x60
[  441.057944] hardirqs last disabled at (231338): [<ffffffff8193e26d>] _raw_spin_lock_irqsave+0xd/0x50
[  441.057947] softirqs last  enabled at (231024): [<ffffffff81c0034f>] __do_softirq+0x34f/0x505
[  441.057949] softirqs last disabled at (231005): [<ffffffff8108c7b9>] irq_exit+0xa9/0xc0
[  441.057951] WARNING: CPU: 4 PID: 9277 at lib/plist.c:42 plist_check_prev_next+0x2d/0x40

v2: Add a load failure point to intel_gvt_init() so that we always
exercise this path in future.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107129
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180710143821.1889-1-chris@chris-wilson.co.uk

drm/i915: Cleanup modesetting on load-error path

After handling a critical failure initialising GEM we need to unwind the
modesetting setup.

Testcase: igt/drv_module_reload/basic-reload-inject
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180710094421.16223-2-chris@chris-wilson.co.uk
Reviewed-by: Matthew Auld <matthew.auld@intel.com>

drm/i915: Flush the residual parking on emergency shutdown

On unwinding following a critical failure inside GEM init, we also need
to be sure to flush the workers before unloading the module.

Testcase: igt/drv_module_reload/basic-reload-inject
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180710094421.16223-1-chris@chris-wilson.co.uk

drm/i915: Tidy i915_gem_suspend()

In the next patch, we will make a fairly minor change to flush
outstanding resets before suspend. In order to keep churn to a minimum
in that functional patch, we fix up the comments and coding style now.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709130208.11730-7-chris@chris-wilson.co.uk

drm/i915: Only reset hangcheck at the start of an activity cycle

Across a reset, the seqno (and thus hangcheck) should restart and the
hangcheck naturally progress, for when it does not, we want to declare an
emergency. Currently, we only detect if reset and reinit fails, but we
do not detect if the call to reinit succeeds but the HW is fried - as we
are resetting hangcheck on initialisation the engine. Remove that and
rely on the natural progress to reset the hangcheck timer.

References: e21b141376f9 ("drm/i915: Mark the hangcheck as idle when unparking the engines")
References: 1fd00c0faeec ("drm/i915: Declare the driver wedged if hangcheck makes no progress")
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: https://patchwork.freedesktop.org/patch/msgid/20180709130208.11730-2-chris@chris-wilson.co.uk

drm/i915/selftests: Filter out both physical address swizzles

In our swizzling selftests, we cannot predict the physical address of
the target page (at least not simply!) and so skip bit17 swizzles.
However, there are two bit17 swizzle modes and we only skipped one, with
the second being observed on the lab gdg causing the test to fail,
as soon as we hit a page with bit17 set in its address.

Testcase: igt/drv_selftest/live_objects #gdg
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709194915.5789-1-chris@chris-wilson.co.uk

drm/i915/selftests: Constrain mock_gtt tests to fit within RAM

Be pessimistic and presume that we actually allocate every page we
exercise via the mock_gtt (e.g. for gvt). In which case we have to keep
our working set under the available physical memory to prevent oom.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180710080424.7821-1-chris@chris-wilson.co.uk

drm/i915: Remove function details from device error messages

Error messages are intended to be addressed to the user; be clear,
succinct, instructive and unambiguous. Adding the function name to
that message does not add any information the user requires and in
the process makes the message less clear.

E.g.

[  245.539711] i915 0000:00:02.0: [drm:i915_gem_init [i915]] Failed to initialize GPU, declaring it wedged!

becomes

[  245.539711] i915 0000:00:02.0: Failed to initialize GPU, declaring it wedged!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709134858.12446-1-chris@chris-wilson.co.uk

drm/i915: Update DRIVER_DATE to 20180709

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drm/i915/selftests: Prevent background reaping of active objects

igt_mmap_offset_exhaustion() wants to test what happens when the mmap
space is filled with zombie objects, objects discarded by userspace but
still active on the GPU. As they are only protected by the active
reference, we have to be certain that active reference is kept while we
peek into our dangling pointer. That active reference should not be
freed until we retire, but we do that retirement from a background
thread. This leaves us with a subtle timing problem, exacerbated and
highlighted by KASAN:

<3>[  132.380399] BUG: KASAN: use-after-free in drm_gem_create_mmap_offset+0x8c/0xd0
<3>[  132.380430] Read of size 8 at addr ffff8801e13245f8 by task drv_selftest/5822

<4>[  132.380470] CPU: 0 PID: 5822 Comm: drv_selftest Tainted: G     U            4.18.0-rc3-g7ae7763aa2be-kasan_48+ #1
<4>[  132.380473] Hardware name: Dell Inc. XPS 8300  /0Y2MRG, BIOS A06 10/17/2011
<4>[  132.380475] Call Trace:
<4>[  132.380481]  dump_stack+0x7c/0xbb
<4>[  132.380487]  print_address_description+0x65/0x270
<4>[  132.380493]  kasan_report+0x25b/0x380
<4>[  132.380497]  ? drm_gem_create_mmap_offset+0x8c/0xd0
<4>[  132.380503]  drm_gem_create_mmap_offset+0x8c/0xd0
<4>[  132.380584]  i915_gem_object_create_mmap_offset+0x6d/0x100 [i915]
<4>[  132.380650]  igt_mmap_offset_exhaustion+0x462/0x940 [i915]
<4>[  132.380714]  ? i915_gem_close_object+0x740/0x740 [i915]
<4>[  132.380784]  ? igt_gem_huge+0x269/0x3d0 [i915]
<4>[  132.380865]  __i915_subtests+0x5a/0x160 [i915]
<4>[  132.380936]  __run_selftests+0x1a2/0x2f0 [i915]
<4>[  132.381008]  i915_live_selftests+0x4e/0x80 [i915]
<4>[  132.381071]  i915_pci_probe+0xd8/0x1b0 [i915]
<4>[  132.381077]  pci_device_probe+0x1c5/0x3a0
<4>[  132.381087]  driver_probe_device+0x6b6/0xcb0
<4>[  132.381094]  __driver_attach+0x22d/0x2c0
<4>[  132.381100]  ? driver_probe_device+0xcb0/0xcb0
<4>[  132.381103]  bus_for_each_dev+0x113/0x1a0
<4>[  132.381108]  ? check_flags.part.24+0x450/0x450
<4>[  132.381112]  ? subsys_dev_iter_exit+0x10/0x10
<4>[  132.381123]  bus_add_driver+0x38b/0x6e0
<4>[  132.381131]  driver_register+0x189/0x400
<4>[  132.381136]  ? 0xffffffffc12d8000
<4>[  132.381140]  do_one_initcall+0xa0/0x4c0
<4>[  132.381145]  ? initcall_blacklisted+0x180/0x180
<4>[  132.381152]  ? do_init_module+0x4a/0x54c
<4>[  132.381156]  ? rcu_lockdep_current_cpu_online+0xdc/0x130
<4>[  132.381161]  ? kasan_unpoison_shadow+0x30/0x40
<4>[  132.381169]  do_init_module+0x1b5/0x54c
<4>[  132.381177]  load_module+0x619e/0x9b70
<4>[  132.381202]  ? module_frob_arch_sections+0x20/0x20
<4>[  132.381211]  ? vfs_read+0x257/0x2f0
<4>[  132.381214]  ? vfs_read+0x257/0x2f0
<4>[  132.381221]  ? kernel_read+0x8b/0x130
<4>[  132.381231]  ? copy_strings_kernel+0x120/0x120
<4>[  132.381244]  ? __se_sys_finit_module+0x17c/0x1a0
<4>[  132.381248]  __se_sys_finit_module+0x17c/0x1a0
<4>[  132.381252]  ? __ia32_sys_init_module+0xa0/0xa0
<4>[  132.381261]  ? __se_sys_newstat+0x77/0xd0
<4>[  132.381265]  ? cp_new_stat+0x590/0x590
<4>[  132.381269]  ? kmem_cache_free+0x2f0/0x340
<4>[  132.381285]  do_syscall_64+0x97/0x400
<4>[  132.381292]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
<4>[  132.381295] RIP: 0033:0x7eff4af46839
<4>[  132.381297] Code: 00 f3 c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 1f f6 2c 00 f7 d8 64 89 01 48
<4>[  132.381426] RSP: 002b:00007ffcd84f4cf8 EFLAGS: 00000246 ORIG_RAX: 0000000000000139
<4>[  132.381432] RAX: ffffffffffffffda RBX: 000055dfdeb429a0 RCX: 00007eff4af46839
<4>[  132.381435] RDX: 0000000000000000 RSI: 000055dfdeb43670 RDI: 0000000000000004
<4>[  132.381437] RBP: 000055dfdeb43670 R08: 0000000000000004 R09: 0000000000000000
<4>[  132.381440] R10: 00007ffcd84f4e60 R11: 0000000000000246 R12: 0000000000000000
<4>[  132.381442] R13: 000055dfdeb3bec0 R14: 0000000000000000 R15: 000000000000003b

<3>[  132.381466] Allocated by task 5822:
<4>[  132.381485]  kmem_cache_alloc+0xdf/0x2e0
<4>[  132.381546]  i915_gem_object_create_internal+0x24/0x1e0 [i915]
<4>[  132.381609]  igt_mmap_offset_exhaustion+0x257/0x940 [i915]
<4>[  132.381677]  __i915_subtests+0x5a/0x160 [i915]
<4>[  132.381742]  __run_selftests+0x1a2/0x2f0 [i915]
<4>[  132.381806]  i915_live_selftests+0x4e/0x80 [i915]
<4>[  132.381865]  i915_pci_probe+0xd8/0x1b0 [i915]
<4>[  132.381868]  pci_device_probe+0x1c5/0x3a0
<4>[  132.381871]  driver_probe_device+0x6b6/0xcb0
<4>[  132.381874]  __driver_attach+0x22d/0x2c0
<4>[  132.381877]  bus_for_each_dev+0x113/0x1a0
<4>[  132.381880]  bus_add_driver+0x38b/0x6e0
<4>[  132.381884]  driver_register+0x189/0x400
<4>[  132.381886]  do_one_initcall+0xa0/0x4c0
<4>[  132.381889]  do_init_module+0x1b5/0x54c
<4>[  132.381892]  load_module+0x619e/0x9b70
<4>[  132.381895]  __se_sys_finit_module+0x17c/0x1a0
<4>[  132.381898]  do_syscall_64+0x97/0x400
<4>[  132.381901]  entry_SYSCALL_64_after_hwframe+0x49/0xbe

<3>[  132.381914] Freed by task 150:
<4>[  132.381931]  kmem_cache_free+0xb7/0x340
<4>[  132.381995]  __i915_gem_free_objects+0x875/0xf50 [i915]
<4>[  132.382054]  __i915_gem_free_work+0x69/0xb0 [i915]
<4>[  132.382058]  process_one_work+0x78b/0x1740
<4>[  132.382061]  worker_thread+0x82/0xb80
<4>[  132.382064]  kthread+0x30c/0x3d0
<4>[  132.382067]  ret_from_fork+0x3a/0x50

<3>[  132.382081] The buggy address belongs to the object at ffff8801e1324500
                   which belongs to the cache drm_i915_gem_object of size 1168
<3>[  132.382133] The buggy address is located 248 bytes inside of
                   1168-byte region [ffff8801e1324500, ffff8801e1324990)
<3>[  132.382179] The buggy address belongs to the page:
<0>[  132.382202] page:ffffea000784c800 count:1 mapcount:0 mapping:ffff8801dedf6500 index:0xffff8801e1323ec0 compound_mapcount: 0
<0>[  132.382251] flags: 0x8000000000008100(slab|head)
<1>[  132.382274] raw: 8000000000008100 ffff8801d6317440 ffff8801d6317440 ffff8801dedf6500
<1>[  132.382307] raw: ffff8801e1323ec0 0000000000140013 00000001ffffffff 0000000000000000
<1>[  132.382339] page dumped because: kasan: bad access detected

<3>[  132.382373] Memory state around the buggy address:
<3>[  132.382395]  ffff8801e1324480: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
<3>[  132.382426]  ffff8801e1324500: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
<3>[  132.382457] >ffff8801e1324580: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
<3>[  132.382488]                                                                 ^
<3>[  132.382517]  ffff8801e1324600: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
<3>[  132.382548]  ffff8801e1324680: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb

This patch tricks the system into running without the background retire
thread, until after we finish the test. The only reaping should then be
performed by the mmap offset routine to reclaim the space as required.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709130208.11730-1-chris@chris-wilson.co.uk

drm/i915/selftests: Replace wait-on-timeout with explicit timeout

In igt_flush_test() we install a background timer in order to ensure
that the wait completes within a certain time. We can now tell the wait
that it has to complete within a timeout, and so no longer need the
background timer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709122044.7028-3-chris@chris-wilson.co.uk

drm/i915: Provide a timeout to i915_gem_wait_for_idle() on setup

With a broken GPU we expect it to fail during the initial
GPU setup where do a couple of context switches to record the defaults.
This is a task that takes a few milliseconds even on the slowest of
devices, but we may have to wait 60s for hangcheck to give in and
declare the machine inoperable. In this a case where any gpu hang is
unacceptable, both from a timeliness and practical standpoint.

We can therefore set a timeout on our wait-for-idle that is shorter than
the hangcheck (which may be up to 60s for a declaring a wedged driver)
and so detect the broken GPU much more quickly during driver load (and
so prevent stalling userspace for ages).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709122044.7028-2-chris@chris-wilson.co.uk

drm/i915: Provide a timeout to i915_gem_wait_for_idle()

Usually we have no idea about the upper bound we need to wait to catch
up with userspace when idling the device, but in a few situations we
know the system was idle beforehand and can provide a short timeout in
order to very quickly catch a failure, long before hangcheck kicks in.

In the following patches, we will use the timeout to curtain two overly
long waits, where we know we can expect the GPU to complete within a
reasonable time or declare it broken.

In particular, with a broken GPU we expect it to fail during the initial
GPU setup where do a couple of context switches to record the defaults.
This is a task that takes a few milliseconds even on the slowest of
devices, but we may have to wait 60s for hangcheck to give in and
declare the machine inoperable. In this a case where any gpu hang is
unacceptable, both from a timeliness and practical standpoint.

The other improvement is that in selftests, we do not need to arm an
independent timer to inject a wedge, as we can just limit the timeout on
the wait directly.

v2: Include the timeout parameter in the trace.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180709122044.7028-1-chris@chris-wilson.co.uk

drm/i915/selftests: Magic numbers for old Y-tiling

i915g has a slightly different tiling layout, and so requires a
different reference swizzle pattern.

Testcase: igt/drv_selftests/live_objects #gdg
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180707100405.817-1-chris@chris-wilson.co.uk

drm/i915: Replace nested subclassing with explicit subclasses

In the next patch, we will want a third distinct class of timeline that
may overlap with the current pair of client and engine timeline classes.
Rather than use the ad hoc markup of SINGLE_DEPTH_NESTING, initialise
the different timeline classes with an explicit subclass.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706210710.16251-1-chris@chris-wilson.co.uk

drm/i915/selftests: Avoid warning if runtime pm is disabled

Inside the mock GEM device, we try to grab the runtime pm for the fake
device to prevent it from ever suspending. However, if CONFIG_PM is not
set, trying to obtain the wakref returns an error which we WARN about.
Suppress the expected warning.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706205947.11209-1-chris@chris-wilson.co.uk

drm/i915/selftests: Provide full mb() around clflush

clflush is an unserialised instruction and the IA manual strongly advises
you to serialise it with a mb. To be cautious, apply one before and one
after, so that it is serialised with both writes and reads without
worrying too much about the required direction.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706174926.4712-1-chris@chris-wilson.co.uk

drm/i915: Track the last-active inside the i915_vma

Using a VMA on more than one timeline concurrently is the exception
rather than the rule (using it concurrently on multiple engines). As we
expect to only use one active tracker, store the most recently used
tracker inside the i915_vma itself and only fallback to the rbtree if
we need a second or more concurrent active trackers.

v2: Comments on how we overwrite any existing last_active cache.
v3: __list_del_entry() before list_replace_init() is confusing and, much
more important, entirely redundant.
v4: Note that both last_active and the rbtree may be simultaneously
tracking this timeline, albeit with different requests, and so the vma
may be retired twice for the same timeline.
v5: No, that list_del is required!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706123157.9645-1-chris@chris-wilson.co.uk

drm/i915: Track vma activity per fence.context, not per engine

In the next patch, we will want to be able to use more flexible request
timelines that can hop between engines. From the vma pov, we can then
not rely on the binding of this request to an engine and so can not
ensure that different requests are ordered through a per-engine
timeline, and so we must track activity of all timelines. (We track
activity on the vma itself to prevent unbinding from HW before the HW
has finished accessing it.)

v2: Switch to a rbtree for 32b safety (since using u64 as a radixtree
index is fraught with aliasing of unsigned longs).
v3: s/lookup_active/active_instance/ because we can never agree on names

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706103947.15919-5-chris@chris-wilson.co.uk

drm/i915: Move i915_vma_move_to_active() to i915_vma.c

i915_vma_move_to_active() has grown beyond its execbuf origins, and
should take its rightful place in i915_vma.c as a method for i915_vma!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706103947.15919-4-chris@chris-wilson.co.uk

drm/i915: Start returning an error from i915_vma_move_to_active()

Handling such a late error in request construction is tricky, but to
accommodate future patches which may allocate here, we potentially could
err. To handle the error after already adjusting global state to track
the new request, we must finish and submit the request. But we don't
want to use the request as not everything is being tracked by it, so we
opt to cancel the commands inside the request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706103947.15919-3-chris@chris-wilson.co.uk

drm/i915: Export i915_request_skip()

In the next patch, we will want to start skipping requests on failing to
complete their payloads. So export the utility function current used to
make requests inoperable following a failed gpu reset.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706103947.15919-2-chris@chris-wilson.co.uk

drm/i915: Refactor export_fence() after i915_vma_move_to_active()

Currently all callers are responsible for adding the vma to the active
timeline and then exporting its fence. Combine the two operations into
i915_vma_move_to_active() to move all the extra handling from the
callers to the single site.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706103947.15919-1-chris@chris-wilson.co.uk

drm/i915/selftests: Fixup missing MI_MEM_VIRTUAL for live_hangcheck

We always want to use a virtual address (i.e. use the GTT) for
MI_STORE_DWORD_IMM, but forgot the ever so important flag in
live_hangcheck for gen3.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706142323.25699-2-chris@chris-wilson.co.uk

drm/i915/selftests: Replace magic 1<<22 with MI_USE_GGTT/MI_MEM_VIRTUAL

Replace the magic bit with the proper symbolic name for instructing
MI_STORE_DWORD_IMM to use a virtual address (on gen3) or the global GTT
address (still virtual!) on gen4+.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706142323.25699-1-chris@chris-wilson.co.uk
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

drm/i915/selftests: Limit live_gtt allocation test to fit within RAM

Limit the GTT size we try and allocate to ensure that it fits within RAM
and does not trigger the oomkiller indiscriminately.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706125338.24432-1-chris@chris-wilson.co.uk

drm/i915/gtt: Control cache domain of dma_map_page() directly

We already maually control the CPU cache for our page table directories,
so we can tell the dma mapper to skip doing it as well.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706122611.4142-2-chris@chris-wilson.co.uk

drm/i915/gtt: Suppress warnings for dma_map_page

As we propagate back the error to the caller for them to handle, we do
not need the lowest level spitting out a redundant warning upon an
allocation failure inside dma_map_page().

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706122611.4142-1-chris@chris-wilson.co.uk

drm/i915: Flush the WCB following a WC write

If we have just completed a WC write, we must ensure that the WCB (Write
Combining Buffer) is flushed out to main memory before we can expect to
see the results. This is especially important when mixing WC with GTT as
the physical paths are different and cachelines are not naturally flushed.

Testcase: igt/drv_selftests/live_coherency #gdg
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706115402.18547-1-chris@chris-wilson.co.uk

drm/i915/selftests: Skip live_execlists if the GPU is terminally wedged

If the GPU is irrecoverably wedged, we can not execute any requests
making testing execlists (request execution) pointless. Skip!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706114510.18467-1-chris@chris-wilson.co.uk

drm/i915/selftests: Skip live context execution test without logical contexts

If the HW (or driver) doesn't support logical contexts, don't pretend we
gain anything from trying to execute GPU commands with them. At best it
reports -ENODEV, which is an unhelpful failure that we should just skip.

v2: Be more specific and check the driver/engine caps for logical (HW)
context support.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706101923.28548-1-chris@chris-wilson.co.uk

drm/i915: Record logical context support in driver caps

Avoid looking at the magical engines[RCS] to decide if the HW and driver
supports logical contexts, and instead record that knowledge during
initialisation.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706101442.21279-1-chris@chris-wilson.co.uk

drm/i915/ddi: Simplify get_encoder_power_domains()

We can simplify the encoder's get_power_domains() hook by calling it
only if the encoder is active. That way the hook can return its power
domains unconditionally without checking the active state by calling
encoder::get_hw_state(). This get_hw_state() query is in fact
redundant since it's already done by intel_modeset_readout_hw_state()
setting the encoder's crtc or leaving it NULL accordingly. Let's use
this fact to decide if the encoder is active.

While at it clarify the comment in intel_ddi_get_power_domains() about
primary vs. fake MST encoders and make sure we never do an incorrect
encoder->dig_port cast for fake MST encoders.

Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705122654.17072-1-imre.deak@intel.com

drm/i915: Remove support for legacy debugfs crc interface

This interface is deprecated, and has been replaced by the upstream
drm crc interface.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Tomi Sarvela <tomi.p.sarvela@intel.com>
Cc: Petri Latvala <petri.latvala@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628072303.14175-1-maarten.lankhorst@linux.intel.com

drm/i915/selftests: Skip over live context testing when wedged

If the GPU is terminally wedged we cannot submit any requests into a
context, completely unfulfilling our purpose of doing so. As this
expectedly fails, skip over the test.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-9-chris@chris-wilson.co.uk

drm/i915/selftests: Skip huge pages live tests if wedged

We test the GPU handling of huge pages by submitting requests that write
into a huge page, but if the GPU is irrecoverably wedged we cannot
submit any requests. As the test expectedly fails, skip over it.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-8-chris@chris-wilson.co.uk

drm/i915/selftests: Skip live eviction tests when wedged

If the GPU is irrecoverably wedged, we cannot submit any requests and so
cannot make the GTT busy in order to test evicting active objects. As
this expectedly fails, skip over the test.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-7-chris@chris-wilson.co.uk

drm/i915/selftests: Skip workaround tests when wedged

If the GPU is irrecoverably wedged, we cannot submit any request and
therefore cannot query the register state of the context (which is done
using the GPU command stream). So skip over the test as it expectedly
fails.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-6-chris@chris-wilson.co.uk

drm/i915/selftests: Skip all request selftests when wedged

If the GPU is irrecoverably wedge, we cannot submit any request and so
all of the request selftests will expectedly fail. Skip over them.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-5-chris@chris-wilson.co.uk

drm/i915/selftests: Skip making an object busy if the GPU is wedged

If the GPU is wedged, we cannot make the object busy as trying to
submit a request will generate -EIO. Skip to the end of the test.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-4-chris@chris-wilson.co.uk

drm/i915/selftests: Skip using the GPU if wedged

If the GPU is irrecoverably broken, we can not use it to dirty memory
and check for cache coherency with the CPU. All we can do is simply skip
over the GPU subtests and focus on the CPU domains (WC, WB) cache
management.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107127
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-3-chris@chris-wilson.co.uk

drm/i915/selftests: Destroy partial tiling vma after use

As we keep VMA around until the object is destroyed, when testing
partial tiling we instantiate many, many VMA (as the object is huge
allowing for many different partial regions). We test elsewhere our
handling of populating large objects with a full set of VMA and checking
we can retrieve them afterwards, but in this test we incur the cost of
flushing all VMA after every GTT write, dramatically slowing down the
test.

References: https://bugs.freedesktop.org/show_bug.cgi?id=107130
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-2-chris@chris-wilson.co.uk

drm/i915: Squelch very verbose error logging

Having found the error causing the IGT test to fail, downgrade the
verbose logging so that we stop flooding the syslogs as we deliberately
provoke it many thousands of time during selftests.

References: 10195b1e4411 ("drm/i915: Show vma allocator stack when in doubt")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180706065332.15214-1-chris@chris-wilson.co.uk

drm/i915/icl: Define AUX lane registers for Port A/B

This patch defines AUX lane registers for PORT_PCS_DW1,
PORT_TX_DW2, PORT_TX_DW4, PORT_TX_DW5 used during
dsi enabling.

v2: Review comments from Jani N:
    - Define _ICL_PORT_PCS_DW1_AUX_A for consistency
    - Three spaces for bitfield definition.

Signed-off-by: Madhav Chauhan <madhav.chauhan@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530798591-2077-8-git-send-email-madhav.chauhan@intel.com

drm/i915/icl: Power down unused DSI lanes

To save power, unused lanes should be powered
down using the bitfield of PORT_CL_DW10.

v2: Review comments from Jani N
    - Put default label next to case 4
    - Include the shifts in the macros

Signed-off-by: Madhav Chauhan <madhav.chauhan@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530798591-2077-7-git-send-email-madhav.chauhan@intel.com

drm/i915/icl: Define PORT_CL_DW_10 register

This register used to power down individual lanes for
DDI/DSI ports. Bitfields to power up/down various
combinations of lanes are also added in this patch.

v2: Review comments from Jani N
    - Use override instead of "override" for bitfields
    - Define mask for override bitfield
    - Define PWR_DOWN_LN* macros shifted in place
v3: Correct PWR_DOWN_LN_MASK value (Jani N)

Signed-off-by: Madhav Chauhan <madhav.chauhan@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530798591-2077-6-git-send-email-madhav.chauhan@intel.com

drm/i915/icl: Enable DSI IO power

This patch configures mode of operation for DSI
and enable DDI IO power by configuring power well.

v2: Use for_each_dsi_port() for power get (Jani N)

Signed-off-by: Madhav Chauhan <madhav.chauhan@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530798591-2077-5-git-send-email-madhav.chauhan@intel.com

drm/i915/icl: Define DSI mode ctl register

This patch defines DSI IO mode control register and it's bits
used while enabling IO power for DSI.

Signed-off-by: Madhav Chauhan <madhav.chauhan@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530798591-2077-4-git-send-email-madhav.chauhan@intel.com

drm/i915/icl: Program DSI Escape clock Divider

Escape Clock is used for LP communication across the DSI
Link. To achieve the constant frequency of the escape clock
from the variable DPLL frequency output, a variable divider(M)
is needed. This patch programs the same.

v2: (Jani N) Don't end line with "(".

Signed-off-by: Madhav Chauhan <madhav.chauhan@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1530798591-2077-3-git-send-email-madhav.chauhan@intel.com

drm/i915/dsi: update some of the platform based checks

Use the more customary order of latest platform first, and don't bother
with an if in the last branch.

Cc: Madhav Chauhan <madhav.chauhan@intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Madhav Chauhan <madhav.chauhan@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705132509.12881-3-jani.nikula@intel.com

drm/i915/dsi: use vlv and bxt prefixes for the global DSI functions

Avoid confusion with the functions to be added for the new ICL or gen 11
DSI implementation by renaming the current DSI functions. While at it,
permutate the words in the function names to make them all start with
"vlv_dsi" or "vlv_dsi_pll" etc.

Reduce the platform abstractions in the PLL file while at it, moving the
checks to vlv_dsi.c instead, where we typically already have the
necessary if ladders.

Leave the static functions as-is for now; they could be renamed later if
needed.

No functional changes.

v2: use "gen7" prefix.

v3: use "vlv" and "bxt" prefixes, reduce the abstractions.

References: https://patchwork.freedesktop.org/series/44823/
Cc: Madhav Chauhan <madhav.chauhan@intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Madhav Chauhan <madhav.chauhan@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705132509.12881-2-jani.nikula@intel.com

drm/i915/dsi: rename the current DSI files based on first platform

Starting from ICL or gen 11 we have a new DSI block which requires
completely different programming from the current implementation. Having
them in the same file would be confusing. Rename the current DSI and DSI
PLL implementation files as vlv_dsi.c and vlv_dsi_pll.c.

No functional changes.

v2: use "gen7" prefix.

v3: use "vlv" prefix.

References: https://patchwork.freedesktop.org/series/44823/
Cc: Madhav Chauhan <madhav.chauhan@intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Madhav Chauhan <madhav.chauhan@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705132509.12881-1-jani.nikula@intel.com

drm/i915/selftests: Fail hangcheck testing if the GPU is wedged

If the GPU is irrecoverably wedged on startup, it means that it failed
on initialisation and we have already tried to reset it but failed. We
can ignore all further testing, as it is already dead. Failing early,
prevents us from slowly failing in our endeavours later and timing out.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705150214.28316-1-chris@chris-wilson.co.uk

drm/i915/selftests: Detect unknown swizzling correctly

i915_gem_detect_bit_6_swizzle() tries to hide unknown swizzling from
userspace (and ourselves) leaving us with the only clue inside
i915->quirks & QUIRK_PIN_SWIZZLED_PAGES. If we see this bit set, it
means that we really have no clue as to what the swizzle pattern is
being used in any one page and so cannot compute what the reference
value should be in our tiling selftests. We have to skip the test.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107133
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705171523.18462-1-chris@chris-wilson.co.uk

drm/i915: Remove pointless if-else from sdvo code

The return value is a bool so we can just return the result of
the biwise AND. The compiler will take care of the rest.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180621174658.18823-1-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915/selftests: Fixup recursive MI_BB_START for gen3

There's no magic bit0 in MI_BB_START for gen3, it's the same dword length
parameter as elsewhere and needs to be zero.

v2: Same bug in both live_requests and live_hanghcheck.

References: https://bugs.freedesktop.org/show_bug.cgi?id=107132
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705154756.5533-1-chris@chris-wilson.co.uk
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

drm/i915: Mark expected switch fall-throughs

In preparation to enabling -Wimplicit-fallthrough, mark switch cases
where we are expecting to fall through.

Addresses-Coverity-ID: 141432
Addresses-Coverity-ID: 141433
Addresses-Coverity-ID: 141434
Addresses-Coverity-ID: 141435
Addresses-Coverity-ID: 141436
Addresses-Coverity-ID: 1357360
Addresses-Coverity-ID: 1357403
Addresses-Coverity-ID: 1357433
Addresses-Coverity-ID: 1392622
Addresses-Coverity-ID: 1415273
Addresses-Coverity-ID: 1435752
Addresses-Coverity-ID: 1441500
Addresses-Coverity-ID: 1454596
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628223541.GA17665@embeddedor.com

drm/i915/icl: Define register for DSI PLL

This patch adds the new registers and corresponding bit definitions
which will be used for programming/enable DSI PLL.

v2: Review comments from Jani N
    - Fix spaces while defining ICL_ESC_CLK_DIV_MASK
    - Define shift and mask for bitfields.

Signed-off-by: Madhav Chauhan <madhav.chauhan@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530795727-28644-2-git-send-email-madhav.chauhan@intel.com

drm/i915/selftests: Replace open-coded i915_address_space_init()

Use i915_address_space_init() rather than open-code it inside
mock_ppgtt() as we will forget to keep it in sync.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705065653.20449-3-chris@chris-wilson.co.uk

drm/i915/selftests: Use full release for local ppgtt allocation

We can now use the full release mechanism (i915_ppgtt_put) for our local
ppgtt allocation in igt_ppgtt_alloc.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705065653.20449-2-chris@chris-wilson.co.uk

drm/i915: Remove defunct i915->vm_list

No longer used and can be removed. One less global that currently
demands struct_mutex protection.

References: e9e7dc4144cd ("drm/i915/gtt: Make gen6 page directories evictable")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705065653.20449-1-chris@chris-wilson.co.uk

drm/i915/gtt: Pull global wc page stash under its own locking

Currently, the wc-stash used for providing flushed WC pages ready for
constructing the page directories is assumed to be protected by the
struct_mutex. However, we want to remove this global lock and so must
install a replacement global lock for accessing the global wc-stash (the
per-vm stash continues to be guarded by the vm).

We need to push ahead on this patch due to an oversight in hastily
removing the struct_mutex guard around the igt_ppgtt_alloc selftest. No
matter, it will prove very useful (i.e. will be required) in the near
future.

v2: Restore the onstack stash so that we can drop the vm->mutex in
future across the allocation.
v3: Restore the lost pagevec_init of the onstack allocation, and repaint
function names.
v4: Reorder init so that we don't try and use i915_address_space before
it is ininitialised.

Fixes: 1f6f00238abf ("drm/i915/selftests: Drop struct_mutex around lowlevel pggtt allocation")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180704185518.4193-1-chris@chris-wilson.co.uk

drm/i915: Unmask and enable master error interrupt on gen2/3

For whatever reason we only unmask and enable the master error
interrut on gen4. With the EIR handling fixed let's do that
on gen2/3 as well.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180611200258.27121-4-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>

drm/i915: Fix pre-ILK error interrupt ack

Adjust the EIR clearing to cope with the edge triggered IIR
on i965/g4x. To guarantee an edge in the ISR master error bit
we temporarily mask everything in EMR. As some of the EIR bits
can't even be directly cleared we also borrow a trick from
i915_clear_error_registers() and permanently mask any bit that
remains high. No real thought given to how we might unmask them
again once the cause for the error has been clered. I suppose
on pre-g4x GPU reset will reinitialize EMR from scratch.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180611200258.27121-3-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>

drm/i915: Fix hotplug irq ack on i965/g4x

Just like with PIPESTAT, the edge triggered IIR on i965/g4x
also causes problems for hotplug interrupts. To make sure
we don't get the IIR port interrupt bit stuck low with the
ISR bit high we must force an edge in ISR. Unfortunately
we can't borrow the PIPESTAT trick and toggle the enable
bits in PORT_HOTPLUG_EN as that act itself generates hotplug
interrupts. Instead we just have to loop until we've cleared
PORT_HOTPLUG_STAT, or we just give up and WARN.

v2: Don't frob with PORT_HOTPLUG_EN

Cc: stable@vger.kernel.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180614175625.1615-1-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>

drm/i915/selftests: Drop struct_mutex around lowlevel pggtt allocation

For a ppgtt that we are constructing, there is no struct_mutex
dependence so skip it. In the process, also ping the scheduler
frequently to try and avoid the NMI watchdog.

v2: gen6 requires struct_mutex to clean up (currently)

Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
References: https://bugs.freedesktop.org/show_bug.cgi?id=107094
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180703135331.12265-1-chris@chris-wilson.co.uk

drm/i915/selftests: Release the struct_mutex to free the objects

live_gtt is a very slow test to run, simply because it tries to allocate
and use as much as the 48b address space as possibly can and in the
process will try to own all of the system memory. This leads to resource
exhaustion and CPU starvation; the latter impacts us when the NMI
watchdog declares a task hung due to a mutex contention with ourselves.
This we can prevent by releasing the struct_mutex and forcing our
i915/rcu workers to run, and in particular flushing the freed object
worker that is the cause for concern.

References: https://bugs.freedesktop.org/show_bug.cgi?id=107094
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180703101829.7360-1-chris@chris-wilson.co.uk

drm/i915: Wait for PSR exit before checking for vblank evasion

The PIPEDSL freezes on PSR entry and if PSR hasn't fully exited, then
the pipe_update_start call schedules itself out to check back later.

On ChromeOS-4.4 kernel, which is fairly up-to-date w.r.t drm/i915 but
lags w.r.t core kernel code, hot plugging an external display triggers
tons of "potential atomic update errors" in the dmesg, on *pipe A*. A
closer analysis reveals that we try to read the scanline 3 times and
eventually timeout, b/c PSR hasn't exited fully leading to a PIPEDSL
stuck @ 1599. This issue is not seen on upstream kernels, b/c for *some*
reason we loop inside intel_pipe_update start for ~2+ msec which in this
case is more than enough to exit PSR fully, hence an *unstuck* PIPEDSL
counter, hence no error. On the other hand, the ChromeOS kernel spends
~1.1 msec looping inside intel_pipe_update_start and hence errors out
b/c the source is still in PSR.

Regardless, we should wait for PSR exit (if PSR is disabled, we incur
a ~1-2 usec penalty) before reading the PIPEDSL, b/c if we haven't
fully exited PSR, then checking for vblank evasion isn't actually
applicable.

v4: Comment explaining psr_wait after enabling VBL interrupts (DK)

v5: CAN_PSR() to handle platforms that don't support PSR.

v6: Handle local_irq_disable on early return (Chris)

Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Signed-off-by: Tarun Vyas <tarun.vyas@intel.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180627200250.1515-2-tarun.vyas@intel.com

drm/i915/psr: Lockless version of psr_wait_for_idle

This is a lockless version of the exisiting psr_wait_for_idle().
We want to wait for PSR to idle out inside intel_pipe_update_start.
At the time of a pipe update, we should never race with any psr
enable or disable code, which is a part of crtc enable/disable.
The follow up patch will use this lockless wait inside pipe_update_
start to wait for PSR to idle out before checking for vblank evasion.
We need to keep the wait in pipe_update_start to as less as it can be.
So,we can live and flourish w/o taking any psr locks at all.

Even if psr is never enabled, psr2_enabled will be false and this
function will wait for PSR1 to idle out, which should just return
immediately, so a very short (~1-2 usec) wait for cases where PSR
is disabled.

v2: Add comment to explain the 25msec timeout (DK)

v3: Rename psr_wait_for_idle to __psr_wait_for_idle_locked to avoid
    naming conflicts and propagate err (if any) to the caller (Chris)

v5: Form a series with the next patch

v7: Better explain the need for lockless wait and increase the max
    timeout to handle refresh rates < 60 Hz (Daniel Vetter)

v8: Rebase

Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Signed-off-by: Tarun Vyas <tarun.vyas@intel.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180627200250.1515-1-tarun.vyas@intel.com

drm/i915: Remove unnecessary check for unsupported modifiers for NV12

There is already a check to allow only RGB8888 formats with CCS
modifiers.

Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628061854.6430-1-dhinakaran.pandiyan@intel.com

drm/i915/psr: Add psr1 live status

Prints live state of psr1.Extending the existing
PSR2 live state function to cover psr1.

Tested on KBL with psr2 and psr1 panel.

v2: rebase
v3: DK
    Rename psr2_live_status to psr_source_status.
v4: DK
    Move EDP_PSR_STATUS_STATE_SHIFT below EDP_PSR_STATUS_STATE_MASK.
    Pass seq to psr_source_status, handle source status prints in
    psr_source_status.
v5: Fixed CI warning messages
v6:
    Remove extra space in the title before the colon.(DK)
    Rebase. (Jani)
v7: Use tabs for indenting the values.(Jani)
v8: Addressed dk's review comments.

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Signed-off-by: Vathsala Nagaraju <vathsala.nagaraju@intel.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530086910-15914-1-git-send-email-vathsala.nagaraju@intel.com

drm/i915: Try GGTT mmapping whole object as partial

If the whole object is already pinned by HW for use as scanout, we will
fail to move it to the mappable region and so must resort to using a
partial VMA covering the whole object.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104513
Fixes: aa136d9d72c2 ("drm/i915: Convert partial ggtt vma to full ggtt if it spans the entire object")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Reviewed-by: Matthew Auld <matthew.william.auld@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180630090509.469-1-chris@chris-wilson.co.uk

drm/i915: abstract and document register picking macros

Try to describe what the pick variants do, and which to prefer. No
functional changes.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180629102039.2435-1-jani.nikula@intel.com

drm/i915/guc: Print CTL params passed to Guc

While debugging we may want to examine params passed to GuC.

v2: drop #ifdef DEBUG_GUC - Michal

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com> #1
Cc: Michal Winiarski <michal.winiarski@intel.com>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180618111821.47088-1-michal.wajdeczko@intel.com

drm/i915/selftests: Attach the fence to the object when making busy

make_obj_busy() makes a dummy busy object, but didn't attach the fence
to the reservation object, so it would not have registered as busy. For
completeness, attach the dummy request as the exclusive fence and mark
the object as written (in i915_vma_move_to_active)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180629133717.11761-2-chris@chris-wilson.co.uk

drm/i915/selftests: Mark up write into scratch vma

We correctly attach the exclusive fetch for the scratch object when
emitting a request that writes into it, but for completeness we should
also declared the write to i915_vma_move_to_active()

Reported-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180629133717.11761-1-chris@chris-wilson.co.uk

drm/i915: Remove delayed FBC activation.

The only time we should start FBC is when we have waited a vblank
after the atomic update. We've already forced a vblank wait by doing
wait_for_flip_done before intel_post_plane_update(), so we don't need
to wait a second time before enabling.

Removing the worker simplifies the code and removes possible race
conditions, like happening in 103167.

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103167
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180625163758.10871-2-maarten.lankhorst@linux.intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

drm/i915: Block enabling FBC until flips have been completed

There is a small race window in which FBC can be enabled after
pre_plane_update is called, but before the page flip has been
queued or completed.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103167
Link: https://patchwork.freedesktop.org/patch/msgid/20180625163758.10871-1-maarten.lankhorst@linux.intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

drm/i915/execlists: Direct submission of new requests (avoid tasklet/ksoftirqd)

Back in commit 27af5eea54d1 ("drm/i915: Move execlists irq handler to a
bottom half"), we came to the conclusion that running our CSB processing
and ELSP submission from inside the irq handler was a bad idea. A really
bad idea as we could impose nearly 1s latency on other users of the
system, on average! Deferring our work to a tasklet allowed us to do the
processing with irqs enabled, reducing the impact to an average of about
50us.

We have since eradicated the use of forcewaked mmio from inside the CSB
processing and ELSP submission, bringing the impact down to around 5us
(on Kabylake); an order of magnitude better than our measurements 2
years ago on Broadwell and only about 2x worse on average than the
gem_syslatency on an unladen system.

In this iteration of the tasklet-vs-direct submission debate, we seek a
compromise where by we submit new requests immediately to the HW but
defer processing the CS interrupt onto a tasklet. We gain the advantage
of low-latency and ksoftirqd avoidance when waking up the HW, while
avoiding the system-wide starvation of our CS irq-storms.

Comparing the impact on the maximum latency observed (that is the time
stolen from an RT process) over a 120s interval, repeated several times
(using gem_syslatency, similar to RT's cyclictest) while the system is
fully laden with i915 nops, we see that direct submission an actually
improve the worse case.

Maximum latency in microseconds of a third party RT thread
(gem_syslatency -t 120 -f 2)
  x Always using tasklets (a couple of >1000us outliers removed)
  + Only using tasklets from CS irq, direct submission of requests
+------------------------------------------------------------------------+
|          +                                                             |
|          +                                                             |
|          +                                                             |
|          +       +                                                     |
|          + +     +                                                     |
|       +  + +     +  x     x     x                                      |
|      +++ + +     +  x  x  x  x  x  x                                   |
|      +++ + ++  + +  *x x  x  x  x  x                                   |
|      +++ + ++  + *  *x x  *  x  x  x                                   |
|    + +++ + ++  * * +*xxx  *  x  x  xx                                  |
|    * +++ + ++++* *x+**xx+ *  x  x xxxx x                               |
|   **x++++*++**+*x*x****x+ * +x xx xxxx x          x                    |
|x* ******+***************++*+***xxxxxx* xx*x     xxx +                x+|
|             |__________MA___________|                                  |
|      |______M__A________|                                              |
+------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x 118            91           186           124     125.28814     16.279137
+ 120            92           187           109     112.00833     13.458617
Difference at 95.0% confidence
-13.2798 +/- 3.79219
-10.5994% +/- 3.02677%
(Student's t, pooled s = 14.9237)

However the mean latency is adversely affected:

Mean latency in microseconds of a third party RT thread
(gem_syslatency -t 120 -f 1)
  x Always using tasklets
  + Only using tasklets from CS irq, direct submission of requests
+------------------------------------------------------------------------+
|           xxxxxx                                        +   ++         |
|           xxxxxx                                        +   ++         |
|           xxxxxx                                      + +++ ++         |
|           xxxxxxx                                     +++++ ++         |
|           xxxxxxx                                     +++++ ++         |
|           xxxxxxx                                     +++++ +++        |
|           xxxxxxx                                   + ++++++++++       |
|           xxxxxxxx                                 ++ ++++++++++       |
|           xxxxxxxx                                 ++ ++++++++++       |
|          xxxxxxxxxx                                +++++++++++++++     |
|         xxxxxxxxxxx    x                           +++++++++++++++     |
|x       xxxxxxxxxxxxx   x           +            + ++++++++++++++++++  +|
|           |__A__|                                                      |
|                                                      |____A___|        |
+------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x 120         3.506         3.727         3.631     3.6321417    0.02773109
+ 120         3.834         4.149         4.039     4.0375167   0.041221676
Difference at 95.0% confidence
0.405375 +/- 0.00888913
11.1608% +/- 0.244735%
(Student's t, pooled s = 0.03513)

However, since the mean latency corresponds to the amount of irqsoff
processing we have to do for a CS interrupt, we only need to speed that
up to benefit not just system latency but our own throughput.

v2: Remember to defer submissions when under reset.
v4: Only use direct submission for new requests
v5: Be aware that with mixing direct tasklet evaluation and deferred
tasklets, we may end up idling before running the deferred tasklet.
v6: Remove the redudant likely() from tasklet_is_enabled(), restrict the
annotation to reset_in_progress().
v7: Take the full timeline.lock when enabling perf_pmu stats as the
tasklet is no longer a valid guard. A consequence is that the stats are
now only valid for engines also using the timeline.lock to process
state.

Testcase: igt/gem_exec_latency/*rthog*
References: 27af5eea54d1 ("drm/i915: Move execlists irq handler to a bottom half")
Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-9-chris@chris-wilson.co.uk

drm/i915/execlists: Trust the CSB

Now that we use the CSB stored in the CPU friendly HWSP, we do not need
to track interrupts for when the mmio CSB registers are valid and can
just check where we read up to last from the cached HWSP. This means we
can forgo the atomic bit tracking from interrupt, and in the next patch
it means we can check the CSB at any time.

v2: Change the splitting inside reset_prepare, we only want to lose
testing the interrupt in this patch, the next patch requires the change
in locking

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-8-chris@chris-wilson.co.uk

drm/i915/execlists: Stop storing the CSB read pointer in the mmio register

As we now never read back our current head position from the CSB
pointers register, and the HW itself doesn't use it to prevent
overwriting unread CSB entries, we do not need to keep updating the
register. As it turns out this register is not listed as being shadowed,
and so requires forcewake -- but we haven't been taking forcewake around
it so the writes has probably been regularly dropped. Fortuitously, we
only read the value after a reset where it did not matter, and zero was
the right answer (well, close enough).

Mika pointed out that this was how we used to do it (accidentally!)
before he fixed it in commit cc53699b25b5 ("drm/i915: Use masked write
for Context Status Buffer Pointer").

References: cc53699b25b5 ("drm/i915: Use masked write for Context Status Buffer Pointer")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-7-chris@chris-wilson.co.uk

drm/i915/execlists: Reset CSB write pointer after reset

On HW reset, the HW clears the write pointer (to 0). But since it also
writes its first CSB entry to slot 0, we need to reset the write pointer
back to the element before (so the first entry we read is 0).

This is required for the next patch, where we trust the CSB completely!

v2: Use _MASKED_FIELD
v3: Store the reset value, so that we differentiate between mmio/hwsp
transparently and without pretense.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-6-chris@chris-wilson.co.uk

drm/i915/execlists: Unify CSB access pointers

Following the removal of the last workarounds, the only CSB mmio access
is for the old vGPU interface. The mmio registers presented by vGPU do
not require forcewake and can be treated as ordinary volatile memory,
i.e. they behave just like the HWSP access just at a different location.
We can reduce the CSB access to a set of read/write/buffer pointers and
treat the various paths identically and not worry about forcewake.
(Forcewake is nightmare for worstcase latency, and we want to process
this all with irqsoff -- no latency allowed!)

v2: Comments, comments, comments. Well, 2 bonus comments.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-5-chris@chris-wilson.co.uk

drm/i915/execlists: Process one CSB update at a time

In the next patch, we will process the CSB events directly from the
submission path, rather than only after a CS interrupt. Hence, we will
no longer have the need for a loop until the has-interrupt bit is clear,
and in the meantime can remove that small optimisation.

v2: Tvrtko pointed out it was safer to unconditionally kick the tasklet
after each irq, when assuming that the tasklet is called for each irq.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-4-chris@chris-wilson.co.uk

drm/i915/execlists: Pull CSB reset under the timeline.lock

In the following patch, we will process the CSB events under the
timeline.lock and not serialised by the tasklet. This also means that we
will need to protect access to common variables such as
execlists->csb_head with the timeline.lock during reset.

v2: Move sync_irq to avoid deadlocks between taking timeline.lock from
our interrupt handler.
v3: Kill off the synchronize_hardirq as it raises more questions than
answered; now we use the timeline.lock entirely for CSB serialisation
between the irq and elsewhere, we don't need to be so heavy handed with
flushing
v4: Treat request cancellation (wedging after failed reset) similarly

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-3-chris@chris-wilson.co.uk

drm/i915/execlists: Pull submit after dequeue under timeline lock

In the next patch, we will begin processing the CSB from inside the
submission path (underneath an irqsoff section, and even from inside
interrupt handlers). This means that updating the execlists->port[] will
no longer be serialised by the tasklet but needs to be locked by the
engine->timeline.lock instead. Pull dequeue and submit under the same
lock for protection. (An alternate future plan is to keep the in/out
arrays separate for concurrent processing and reduced lock coverage.)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-2-chris@chris-wilson.co.uk

drm/i915: Drop posting reads to flush master interrupts

We do not need to do a posting read of our uncached mmio write to
re-enable the master interrupt lines after handling an interrupt, so
don't. This saves us a slow UC read before we can process the interrupt,
most noticeable in execlists where any stalls imposes extra latency on
GPU command execution.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjala <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628201211.13837-1-chris@chris-wilson.co.uk

drm/i915/uc: Fetch GuC/HuC firmwares from guc/huc specific init

We're fetching GuC/HuC firmwares directly from uc level during
init_early stage but this breaks guc/huc struct isolation and
also strict SW-only initialization rule for init_early. Move fw
fetching to init phase and do it separately per guc/huc struct.

v2: don't forget to move wopcm_init - Michele
v3: fetch in init_misc phase - Michal

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com> #2
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628141522.62788-2-michal.wajdeczko@intel.com

drm/i915/guc: Use intel_guc_init_misc to hide GuC internals

We will add more init steps to misc phase and there is no need
to expose them separately for use in uc_init_misc function.

Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628141522.62788-1-michal.wajdeczko@intel.com

drm/i915: Only signal from interrupt when requested

Avoid calling dma_fence_signal() from inside the interrupt if we haven't
enabled signaling on the request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180627201304.15817-4-chris@chris-wilson.co.uk

drm/i915: Move the irq_counter inside the spinlock

Rather than have multiple locked instructions inside the notify_ring()
irq handler, move them inside the spinlock and reduce their intrinsic
locking.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180627201304.15817-3-chris@chris-wilson.co.uk

drm/i915: Only trigger missed-seqno checking next to boundary

If we have more interrupts pending (because we know there are more
breadcrumb signals before the completion), then we do not need to
trigger an irq_seqno_barrier or even wakeup the task on this interrupt
as there will be another. To allow some margin of error (we are trying
to work around incoherent seqno after all), we wakeup the breadcrumb
before the target as well as on the target.

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: https://patchwork.freedesktop.org/patch/msgid/20180627201304.15817-2-chris@chris-wilson.co.uk

drm/i915: Reduce spinlock hold time during notify_ring() interrupt

By taking advantage of the RCU protection of the task struct, we can find
the appropriate signaler under the spinlock and then release the spinlock
before waking the task and signaling the fence.

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: https://patchwork.freedesktop.org/patch/msgid/20180627201304.15817-1-chris@chris-wilson.co.uk

drm/i915: Show vma allocator stack when in doubt

At the moment, gem_exec_gttfill fails with a sporadic EBUSY due to us
wanting to unbind a pinned batch. Let's dump who first bound that vma to
see if that helps us identify who still unexpectedly has it pinned.

v2: We cannot allocate inside the printer (as it may be on an fs-reclaim
path), so hope for the best and build the string on the stack
v3: stack depth of 16 routinely overflows a 512 character string, limit
it to 12 to avoid unsightly truncation.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180628132206.8329-1-chris@chris-wilson.co.uk

drm/i915: Replace drm_dev_unref with drm_dev_put

This patch unifies the naming of DRM functions for reference counting
of struct drm_device. The resulting code is more aligned with the rest
of the Linux kernel interfaces.

Signed-off-by: Thomas Zimmermann <tdz@users.sourceforge.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20180618110154.30462-6-tdz@users.sourceforge.net

drm/i915: Replace drm_gem_object_unreference_unlocked with put function

This patch unifies the naming of DRM functions for reference counting
of struct drm_gem_object. The resulting code is more aligned with the
rest of the Linux kernel interfaces.

Signed-off-by: Thomas Zimmermann <tdz@users.sourceforge.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20180618110154.30462-5-tdz@users.sourceforge.net

drm/i915: Replace __drm_gem_object_unreference with __drm_gem_object_put

This patch unifies the naming of DRM functions for reference counting
of struct drm_gem_object. The resulting code is more aligned with the
rest of the Linux kernel interfaces.

Signed-off-by: Thomas Zimmermann <tdz@users.sourceforge.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20180618110154.30462-4-tdz@users.sourceforge.net