The hardware needs some time to process the information received in the
ExecList Submission Port, and expects us to not write anything more until
it has 'acknowledged' this new submission by sending an IDLE_ACTIVE or
PREEMPTED CSB event.
If we do not follow this, the driver could write new data into the ELSP
before HW had finishing fetching the previous one, putting us in
'undefined behaviour' space.
This seems to be the problem causing the spurious PREEMPTED & COMPLETE
events after a COMPLETE like the one below:
[] vcs0: sw rd pointer = 2, hw wr pointer = 0, current 'head' = 3.
[] vcs0: Execlist CSB[0]: 0x00000018 _ 0x00000007
[] vcs0: Execlist CSB[1]: 0x00000001 _ 0x00000000
[] vcs0: Execlist CSB[2]: 0x00000018 _ 0x00000007 <<< COMPLETE
[] vcs0: Execlist CSB[3]: 0x00000012 _ 0x00000007 <<< PREEMPTED & COMPLETE
[] vcs0: Execlist CSB[4]: 0x00008002 _ 0x00000006
[] vcs0: Execlist CSB[5]: 0x00000014 _ 0x00000006
The ELSP writes that lead to this CSB sequence show that the HW hadn't
started executing the previous execlist (the one with only ctx 0x6) by the
time the new one was submitted; this is a bit more clear in the data
show in the EXECLIST_STATUS register at the time of the ELSP write.
[] vcs0: ELSP[0] = 0x0_0 [execlist1] - status_reg = 0x0_302
[] vcs0: ELSP[1] = 0x6_fedb2119 [execlist0] - status_reg = 0x0_8302
[] vcs0: ELSP[2] = 0x7_fedaf119 [execlist1] - status_reg = 0x0_8308
[] vcs0: ELSP[3] = 0x6_fedb2119 [execlist0] - status_reg = 0x7_8308
Note that having to wait for this ack does not disable lite-restores,
although it may reduce their numbers.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102035
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/<20171118003038.7935-1-michel.thierry@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171120123458.23242-4-chris@chris-wilson.co.uk
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Now that we have this stored in the device info, we can drop it from perf
part of the driver.
Note that this requires to init perf after we've computed the frequency,
hence why we move i915_perf_init() from i915_driver_init_early() to after
intel_device_info_runtime_init().
v2: Use div_u64 (Chris)
v3: Drop u64 divs by switching to kHz (Chris/Ville)
Move i915_perf_fini to i915_driver_cleanup_hw (Matthew)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171113181902.12411-2-lionel.g.landwerlin@intel.com
As the request will, in the following patch, implicitly invoke a
context-switch on construction, we should precede that with a GPU TLB
invalidation. Also, even before using GGTT, we always want to invalidate
the TLBs for any updates (as well as the ppgtt invalidates that are
unconditionally applied by execbuf). Since we almost always require the
TLB invalidate, do it unconditionally on request allocation and so we can
remove it from all other paths.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171120102002.22254-1-chris@chris-wilson.co.uk
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Rodrigo gave a persuasive argument for keeping workarounds: that they
serve as a good guide for the bring up of the next generation. Not only
do workarounds persist into the early revisions, they show where the
workarounds were previously added to the code flow and sometimes the old
workarounds have an explanation that give insight into their wider
implications.
Based on his suggestion, document the policy that we want to keep the
workarounds from the current generation to guide the next. Older
preproduction workarounds we still want to remove to keep the code
clean.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20171117102635.8689-1-chris@chris-wilson.co.uk
Acked-by: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
The watermarks it should calculate against are the old optimal watermarks.
The currently active crtc watermarks are pure fiction, and are invalid in
case of a nonblocking modeset, page flip enabling/disabling planes or any
other reason.
When the crtc is disabled or during a modeset the intermediate watermarks
don't need to be programmed separately, and could be directly assigned
to the optimal watermarks.
CXSR must always be disabled in the intermediate case for modesets,
else we get a WARN for vblank wait timeout.
Also rename crtc_state to new_crtc_state, to distinguish it from the old
state.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171115163157.14372-2-maarten.lankhorst@linux.intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
The watermarks it should calculate against are the old optimal watermarks.
The currently active crtc watermarks are pure fiction, and are invalid in
case of a nonblocking modeset, page flip enabling/disabling planes or any
other reason.
When the crtc is disabled or during a modeset the intermediate watermarks
don't need to be programmed separately, and could be directly assigned
to the optimal watermarks.
CXSR must always be disabled in the intermediate case for modesets, else
we get a WARN for vblank wait timeout.
Also rename crtc_state to new_crtc_state, to distinguish it from the old state.
Changes since v1:
- Use intel_atomic_get_old_crtc_state. (ville)
Changes since v2:
- Always unset cxsr during modeset.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171115163157.14372-1-maarten.lankhorst@linux.intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
The firmware may have set up the pipe correctly, but the FIFO
underrun and CRC interrupts are likely not enabled.
This resulted in debugfs_test.read_all_entries failing on haswell,
because of a timeout when reading the crc debugfs entry.
Solve this by enabling FIFO underrun reporting after the initial
fastset, which lets interrupts be generated as expected.
Changes since v1:
- Always enable CPU FIFO underrun reporting for >GEN2,
and handle GEN2 correctly.
Changes since v2:
- Remove unneeded HAS_DDI, simplify GEN2 case.
Changes since v3:
- Use intel_crtc_pch_transcoder to determine pch transcoder for underruns. (Ville)
- Remove crtc->config dereference in intel_crtc_pch_transcoder. (Ville)
Testcase: debugfs_test.read_all_entries
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171113144043.58658-1-maarten.lankhorst@linux.intel.com
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
The first test aims to check guc_init_doorbell_hw, changing the existing
guc clients and doorbells state before calling it.
The second test tries to create as many clients as it is currently possible
(currently limited to max number of doorbells) and exercise the doorbell
alloc/dealloc code.
Since our usage mode require very few clients/doorbells, this code has
been exercised very lightly and it's good to have a simple test for it.
As reference, this test already helped identify the bug fixed by
commit 7f1ea2ac30 ("drm/i915/guc: Fix doorbell id selection").
v2: Extend number of clients; check for client allocation failure when
number of doorbells is exceeded; validate client properties; reuse
guc_init_doorbell_hw (Chris).
v3: guc_init_doorbell_hw test added per Chris suggestion.
v4: Try to explain why guc_init_doorbell_hw exist and comment some
details in the subtest.
v5: Remove redundant pr_info at the beginning of each subtest (Chris);
rebase (s/i915_guc_client/intel_guc_client/).
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20171116220632.1909-1-michel.thierry@intel.com
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Spec describe all values in MHz. We handle our
clocks in KHz. This includes the best_dco_centrality that was
forgot in the same unity as spec. Consequently we couldn't
get a good divider for high frequenies. Hence HDMI 2.0 wasn't
working.
Spec tells 999999 for initial best_dco_centrality meaning the
max value in MHz.
Since we convert dco from MHz to KHz we also need to convert
this initial best_doc_centrality to 999999000 or 999999999
or even better, to the max that its variable allow.
This patch also replaces the use of "* KHz(1)" with the values
directly on KHz to avoid future confusion.
v2: Use U32_MAX instead of random 99999 as spec tells. (Ville).
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Shashank Sharma <shashank.sharma@intel.com>
Cc: Mika Kahola <mika.kahola@intel.com>
Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: James Ausmus <james.ausmus@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171114234223.10600-1-rodrigo.vivi@intel.com
We check whether the multiplies will overflow prior to calling
kmalloc_array so that we can respond with -EINVAL for the invalid user
arguments rather than treating it as an -ENOMEM that would otherwise
occur. However, as Dan Carpenter pointed out, we did an addition on the
unsigned int prior to passing to kmalloc_array where it would be
promoted to size_t for the calculation, thereby allowing it to overflow
and underallocate.
v2: buffer_count is currently limited to INT_MAX because we treat it as
signaled variable for LUT_HANDLE in eb_lookup_vma
v3: Move common checks for eb1/eb2 into the same function
v4: Put the check back for nfence*sizeof(user_fence) overflow
v5: access_ok uses ULONG_MAX but kvmalloc_array uses SIZE_MAX
v6: size_t and unsigned long are not type-equivalent on 32b
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171116105059.25142-1-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
When we call intel_engine_cancel_signaling() to stop reporting when
a request is completed via an asynchronous signal, we remove that request
from the breadcrumb wait queue. However, we may be concurrently
processing that request in the signaler itself, the actual operations on
the request's node itself are serialised but we do not actually clear the
waiter after removing it from the tree allowing both parties to attempt
to do so and corrupting the rbtree. (Previously removing from the
breadcrumb wait queue could only be done on behalf of i915_wait_request,
so this race could not happen).
Reported-by: "He, Bo" <bo.he@intel.com>
Fixes: 9eb143bbec ("drm/i915: Allow a request to be cancelled")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: "He, Bo" <bo.he@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171115121458.24655-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Previously the performance is improved through the workload auditing
and shadowing ahead of vGPU scheduling, however, there is the case that
more requests are allocated in submit_context before the previous request
is added, the timeline will hold its seqno which is later.
This patch is to move the request alloc to dispatch_workload function,
where is the same place as request is added.
It will fix the issue of kernel BUG for (timeline->seqno != request->fence.seqno)
check when add_request.
Fixes: 89ea20b930 ("drm/i915/gvt: Factor out scan and shadow from workload dispatch")
Signed-off-by: Chuanxiao Dong <chuanxiao.dong@intel.com>
Signed-off-by: fred gao <fred.gao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Currently every vgpu share a common gvt opregion memory, but
it is freed at vgpu destroy, then the later vgpu doesn't have
opregion memory once the first vgpu is destroyed. This cause
guest function failure like reboot, second or later boot.
This patch allocate and init virt opregion memory for each
vgpu, so this memory could be freed at vgpu destroy.
Signed-off-by: Xiong Zhang <xiong.y.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
mmio_read_from_hw() let vgpu could read hw reg, if vgpu's workload
is running on hw, things is good. Otherwise vgpu will get other
vgpu's reg val, it is unsafe.
This patch limit such hw access to active vgpu. If vgpu isn't
running on hw, the reg read of this vgpu will get the last active
val which saved at schedule_out.
v2: ring timestamp is walking continuously even if the ring is idle.
so read hw directly. (Zhenyu)
Signed-off-by: Xiong Zhang <xiong.y.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
This reverts commit b20d09886fd1b74cd2255d846029a049e524db14.
This caused windows driver boot errors for invalid page address.
Revert for now.
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Cc: Zhi Wang <zhi.a.wang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Our vGPU doesn't have a device ROM, we need follow the PCI spec to
report this info to drivers. Otherwise, we would see below errors.
Inspecting possible rom at 0xfe049000 (vd=8086:1912 bdf=00:10.0)
qemu-system-x86_64: vfio-pci: Cannot read device rom at 00000000-0000-0000-0000-000000000001
Device option ROM contents are probably invalid (check dmesg).
Skip option ROM probe with rombar=0, or load from file with romfile=No option rom signature (got 4860)
I will also send a improvement patch to PCI subsystem related to PCI ROM.
But no idea to omit below error, since no pattern to detect vbios shadow
without touch its content.
0000:00:10.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0x0000
Signed-off-by: Changbin Du <changbin.du@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
gvt_vgpu_err means something goes wrong. We need the error propagates to
kernel message by default.
Signed-off-by: Changbin Du <changbin.du@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
I have seen the cmd parser dump partial odd info. Stop that and only dump
the full verbose info when debug enabled.
Signed-off-by: Changbin Du <changbin.du@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Use I915_WRITE_FW instead of I915_WRITE to reduce overhead.
The overall mmio switch latency lowers from ~600us to ~180us.
Signed-off-by: Changbin Du <changbin.du@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
This new debugfs entry is used to figure out which registers of vGPU
is different to host. It is a useful tool for new platform enabling
and debugging. When read this entry, all the diff mmio are recognized
and sorted by mmio offset. Besides, the bit positions of different
value are listed in 'Diff' column. Here is a show:
$ sudo cat ./mmio_diff
Offset HW vGPU Diff
00002030 000025f8 00000000 3-8,10,13
00002034 012025f8 00000000 3-8,10,13,21,24
00002038 027fb000 00000000 12-13,15-22,25
0000203c 00003000 00000000 12-13
00002054 0000000a 00000040 1,3,6
00002074 012025f8 00000000 3-8,10,13,21,24
00002080 fffe6000 00000000 13-14,17-31
000020a8 fffffeff ffffffff 8
000020d4 00000004 00000000 2
....
00145974 eb42718c 010c11b0 2-5,13-14,17-19,22,25,27,29-31
00145978 0000002f 0000002a 0,2
0014597c 0000002f 0000002a 0,2
00145980 0000002b 00000028 0-1
00145984 a5a87c9e b27d20c0 1-4,6,10-12,14,16,18,20,22-26,28
001459c0 88390000 883c0000 16,18
00146200 88350000 883a0000 16-19
Total: 72432, Diff: 901
v3: fix a typo.
v2: add mmio_hw_access_pre/post().
Signed-off-by: Changbin Du <changbin.du@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
