If the only thing that is changing is SAGV vs. no SAGV but
the number of active planes and the total data rates end up
unchanged we currently bail out of intel_bw_atomic_check()
early and forget to actually compute the new WGV point
mask and thus won't actually enable/disable SAGV as requested.
This ends up poorly if we end up running with SAGV enabled
when we shouldn't. Usually ends up in underruns.
To fix this let's go through the QGV point mask computation
if either the data rates/number of planes, or the state
of SAGV is changing.
v2: Check more carefully if things are changing to avoid
the extra calculations/debugs from introducing unwanted
overhead
Cc: stable@vger.kernel.org
Reviewed-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> #v1
Fixes: 20f505f225 ("drm/i915: Restrict qgv points which don't have enough bandwidth.")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220218064039.12834-3-ville.syrjala@linux.intel.com
We don't want any RMWs in the part of the commit that happens
under vblank evasion. Eventually we want to use the DSB to
handle that and it can't read registers at all. Also reads
are just slowing us down needlessly.
Let's move the whole PIPE_CHICKEN stuff out from the critical
section since we don't have anything there that needs to be
syncrhonized with other plane/pipe registers. If we ever need
to add such things then we have to move it back, but without
doing any reads.
TODO: should look into eliminating the RMW anyway...
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220202111616.1579-1-ville.syrjala@linux.intel.com
Reviewed-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
A new programming step was added to combo and TC PLL sequences.
If override_AFC_startup is set in VBT, driver should overwrite
AFC_startup value to 0x0 or 0x7 in PLL's div0 register.
The current understating is that only TGL needs this and all other
display 12 and newer platforms will have a older VBT or a newer VBT
with override_AFC_startup set to 0 but in any case there is a
drm_warn_on_once() to let us know if this is not true.
v2:
- specification updated, now AFC can be override to 0x0 or 0x7
- not using a union for div0 (Imre)
- following previous wrong vbt naming: bits instead of bytes (Imre)
BSpec: 49204
BSpec: 20122
BSpec: 49968
BSpec: 71360
Cc: Imre Deak <imre.deak@intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220216134059.25348-1-jose.souza@intel.com
BIOS may leave a TypeC PHY in a connected state even though the
corresponding port is disabled. This will prevent any hotplug events
from being signalled (after the monitor deasserts and then reasserts its
HPD) until the PHY is disconnected and so the driver will not detect a
connected sink. Rebooting with the PHY in the connected state also
results in a system hang.
Fix the above by disconnecting TypeC PHYs on disabled ports.
Before commit 64851a32c4 the PHY connected state was read out even
for disabled ports and later the PHY got disconnected as a side effect
of a tc_port_lock/unlock() sequence (during connector probing), hence
recovering the port's hotplug functionality.
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5014
Fixes: 64851a32c4 ("drm/i915/tc: Add a mode for the TypeC PHY's disconnected state")
Cc: <stable@vger.kernel.org> # v5.16+
Cc: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220217152237.670220-1-imre.deak@intel.com
No reason the high level intel_update_crtc() needs to know
that there is something magical about the commit order of
planes between different platforms. So let's hide that
detail even better.
In order to keep to somewhat consistent naming between
things we shall call this intel_crtc_planes_update_arm()
to match the plane->update_arm() vfunc naming convention.
And let's rename the noarm counterpart to
intel_crtc_planes_update_noarm() to more clearly associate
it with the plane->update_noarm() vfunc.
v2: Change the naming convention a bit
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220216232806.6194-2-ville.syrjala@linux.intel.com
Registers that belong to the shared render/compute reset domain need to
be placed on an engine workaround list to ensure that they are properly
re-applied whenever any RCS or CCS engine is reset, even if the
registers do not belong to a specific engine's MMIO range. We have a
number of workarounds today that are incorrectly implemented on the 'gt'
workaround list and need to be moved accordingly. We also have one
workaround (Wa_22012532006) that is incorrectly implemented on the
context workaround list, even though the register it is adjusting is not
part of the RCS engine's context image; it must also be moved.
We'll have some workaround refactoring coming in the near future that
deals with registers in the reset domain in a more clear way. But in
the meantime, we should just move these workarounds to
rcs_engine_wa_init() to place them on the RCS engine's workaround list.
All production DG2 platforms will have an RCS engine (it's never fused
off) so these registers will be properly restored after a domain reset
triggered via an RCS engine _or_ a CCS engine.
Cc: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.s@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220215235531.2236399-1-matthew.d.roper@intel.com
Clang warns:
drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_packet_manager_v9.c:267:3:
error: implicit conversion from enumeration type 'enum
mes_map_queues_extended_engine_sel_enum' to different enumeration type
'enum mes_unmap_queues_extended_engine_sel_enum'
[-Werror,-Wenum-conversion]
extended_engine_sel__mes_map_queues__sdma0_to_7_sel :
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1 error generated.
Use 'extended_engine_sel__mes_unmap_queues__sdma0_to_7_sel' to eliminate
the warning, which is the same numeric value of the proper type.
Fixes: 009e9a1585 ("drm/amdkfd: navi2x requires extended engines to map and unmap sdma queues")
Link: https://github.com/ClangBuiltLinux/linux/issues/1596
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Clang build fails with
amdgpu_ras.c:2416:7: error: variable 'ras_obj' is used uninitialized
whenever 'if' condition is true
if (adev->in_suspend || amdgpu_in_reset(adev)) {
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
amdgpu_ras.c:2453:6: note: uninitialized use occurs here
if (ras_obj->ras_cb)
^~~~~~~
There is a logic error in the error handler's labels.
ex/ The sysfs: is the last goto label in the normal code but
is the middle of error handler. Rework the error handler.
cleanup: is the first error, so it's handler should be last.
interrupt: is the second error, it's handler is next. interrupt:
handles the failure of amdgpu_ras_interrupt_add_hander() by
calling amdgpu_ras_interrupt_remove_handler(). This is wrong,
remove() assumes the interrupt has been setup, not torn down by
add(). Change the goto label to cleanup.
sysfs is the last error, it's handler should be first. sysfs:
handles the failure of amdgpu_ras_sysfs_create() by calling
amdgpu_ras_sysfs_remove(). But when the create() fails there
is nothing added so there is nothing to remove. This error
handler is not needed. Remove the error handler and change
goto label to interrupt.
Fixes: b293e891b0 ("drm/amdgpu: add helper function to do common ras_late_init/fini (v3)")
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Tom Rix <trix@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Needed by umr to detect if ip discovered ASIC is an APU or not.
(v2): Remove asic type from packet it's not strictly needed
(v3): Correct comment
Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The `program_aspm` callback is already guarded for aspm, but the
`enable_aspm` callback doesn't follow the module parameter.
Update it to use the helper `amdgpu_device_should_use_aspm`.
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Evaluating `pcie_aspm_enabled` as part of driver probe has the implication
that if one PCIe bridge with an AMD GPU connected doesn't support ASPM
then none of them do. This is an invalid assumption as the PCIe core will
configure ASPM for individual PCIe bridges.
Create a new helper function that can be called by individual dGPUs to
react to the `amdgpu_aspm` module parameter without having negative results
for other dGPUs on the PCIe bus.
Suggested-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
1. Define amdgpu_ras_block_late_init_default in amdgpu_ras.c as
.ras_late_init common function, which is called when
.ras_late_init of ras block isn't initialized.
2. Remove the code of using amdgpu_ras_block_late_init to
initialize .ras_late_init in ras blocks.
Signed-off-by: yipechai <YiPeng.Chai@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
1. Move calling ras block instance members from module internal
function to the top calling xxx_ras_late_init.
2. Module internal function calls can only use parameter variables
of xxx_ras_late_init instead of ras block instance members.
Signed-off-by: yipechai <YiPeng.Chai@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Modify .ras_late_init function pointer parameter so that
it can remove redundant intermediate calls in some ras blocks.
Signed-off-by: yipechai <YiPeng.Chai@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
- set DC version
- add construct/destroy dc clock management function
- register dcn interrupt handler
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drm/i915 adds some extra cflags, namely -Wall, which causes instances of
-Wformat-security to appear when building with clang, even though this
warning is turned off kernel-wide in the main Makefile:
drivers/gpu/drm/i915/gt/intel_gt.c:983:2: error: format string is not a string literal (potentially insecure) [-Werror,-Wformat-security]
GEM_TRACE("ERROR\n");
^~~~~~~~~~~~~~~~~~~~
./drivers/gpu/drm/i915/i915_gem.h:76:24: note: expanded from macro 'GEM_TRACE'
#define GEM_TRACE(...) trace_printk(__VA_ARGS__)
^~~~~~~~~~~~~~~~~~~~~~~~~
./include/linux/kernel.h:369:3: note: expanded from macro 'trace_printk'
do_trace_printk(fmt, ##__VA_ARGS__); \
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
./include/linux/kernel.h:383:30: note: expanded from macro 'do_trace_printk'
__trace_bprintk(_THIS_IP_, trace_printk_fmt, ##args); \
^~~~~~~~~~~~~~~~
drivers/gpu/drm/i915/gt/intel_gt.c:983:2: note: treat the string as an argument to avoid this
This does not happen with GCC because it does not enable
-Wformat-security with -Wall. Disable -Wformat-security within the i915
Makefile so that these warnings do not show up with clang.
Signed-off-by: Tong Zhang <ztong0001@gmail.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20220214195821.29809-1-ztong0001@gmail.com
