kwin/src/plugins/platforms/drm/drm_pipeline.cpp

565 lines
18 KiB
C++
Raw Normal View History

/*
KWin - the KDE window manager
This file is part of the KDE project.
SPDX-FileCopyrightText: 2021 Xaver Hugl <xaver.hugl@gmail.com>
SPDX-License-Identifier: GPL-2.0-or-later
*/
#include "drm_pipeline.h"
#include <errno.h>
#include "logging.h"
#include "drm_gpu.h"
#include "drm_object_connector.h"
#include "drm_object_crtc.h"
#include "drm_object_plane.h"
#include "drm_buffer.h"
#include "cursor.h"
#include "session.h"
#include "abstract_output.h"
#include "drm_backend.h"
#if HAVE_GBM
#include <gbm.h>
#include "egl_gbm_backend.h"
#include "drm_buffer_gbm.h"
#endif
namespace KWin
{
DrmPipeline::DrmPipeline(DrmGpu *gpu, DrmConnector *conn, DrmCrtc *crtc, DrmPlane *primaryPlane)
: m_pageflipUserData(nullptr)
, m_gpu(gpu)
, m_connector(conn)
, m_crtc(crtc)
, m_primaryPlane(primaryPlane)
{
if (m_gpu->atomicModeSetting()) {
m_connector->findCurrentMode(m_crtc->queryCurrentMode());
m_connector->setPending(DrmConnector::PropertyIndex::CrtcId, m_crtc->id());
auto mode = m_connector->currentMode();
m_crtc->setPending(DrmCrtc::PropertyIndex::Active, 1);
m_crtc->setPendingBlob(DrmCrtc::PropertyIndex::ModeId, &mode.mode, sizeof(drmModeModeInfo));
}
m_allObjects << m_connector << m_crtc;
if (m_primaryPlane) {
m_primaryPlane->setPending(DrmPlane::PropertyIndex::CrtcId, m_crtc->id());
m_primaryPlane->set(QPoint(0, 0), sourceSize(), QPoint(0, 0), m_connector->currentMode().size);
m_allObjects << m_primaryPlane;
}
}
DrmPipeline::~DrmPipeline()
{
}
bool DrmPipeline::test(const QVector<DrmPipeline*> &pipelines)
{
if (m_gpu->atomicModeSetting()) {
return checkTestBuffer() && atomicTest(pipelines);
} else {
return m_legacyNeedsModeset ? modeset(0) : true;
}
}
bool DrmPipeline::test()
{
return test(m_gpu->pipelines());
}
bool DrmPipeline::present(const QSharedPointer<DrmBuffer> &buffer)
{
m_primaryBuffer = buffer;
if (m_gpu->useEglStreams() && m_gpu->eglBackend() != nullptr && m_gpu == m_gpu->platform()->primaryGpu()) {
// EglStreamBackend queues normal page flips through EGL,
// modesets etc are performed through DRM-KMS
bool needsCommit = std::any_of(m_allObjects.constBegin(), m_allObjects.constEnd(), [](auto obj){return obj->needsCommit();});
if (!needsCommit) {
return true;
}
}
if (m_gpu->atomicModeSetting()) {
if (!atomicCommit()) {
// update properties and try again
updateProperties();
if (!atomicCommit()) {
qCWarning(KWIN_DRM) << "Atomic present failed!" << strerror(errno);
printDebugInfo();
return false;
}
}
} else {
if (!presentLegacy()) {
qCWarning(KWIN_DRM) << "Present failed!" << strerror(errno);
return false;
}
}
return true;
}
bool DrmPipeline::atomicCommit()
{
drmModeAtomicReq *req = drmModeAtomicAlloc();
if (!req) {
qCDebug(KWIN_DRM) << "Failed to allocate drmModeAtomicReq!" << strerror(errno);
return false;
}
bool result = doAtomicCommit(req, 0, false);
drmModeAtomicFree(req);
return result;
}
bool DrmPipeline::atomicTest(const QVector<DrmPipeline*> &pipelines)
{
drmModeAtomicReq *req = drmModeAtomicAlloc();
if (!req) {
qCDebug(KWIN_DRM) << "Failed to allocate drmModeAtomicReq!" << strerror(errno);
return false;
}
uint32_t flags = 0;
bool result = true;
// the whole batch needs to be tested in one atomic commit to consider not-yet-applied changes of other pipelines
for (const auto &pipeline : pipelines) {
if (pipeline != this) {
result &= pipeline->populateAtomicValues(req, flags);
if (!result) {
break;
}
}
}
if (result) {
result &= doAtomicCommit(req, flags, true);
} else {
if (m_oldTestBuffer) {
m_primaryBuffer = m_oldTestBuffer;
m_oldTestBuffer = nullptr;
}
for (const auto &obj : qAsConst(m_allObjects)) {
obj->rollbackPending();
}
}
drmModeAtomicFree(req);
return result;
}
bool DrmPipeline::doAtomicCommit(drmModeAtomicReq *req, uint32_t flags, bool testOnly)
{
bool result = populateAtomicValues(req, flags);
// test
if (result && drmModeAtomicCommit(m_gpu->fd(), req, (flags & (~DRM_MODE_PAGE_FLIP_EVENT)) | DRM_MODE_ATOMIC_TEST_ONLY, m_pageflipUserData) != 0) {
qCWarning(KWIN_DRM) << "Atomic test failed!" << strerror(errno);
printDebugInfo();
result = false;
}
// commit
if (!testOnly && result && drmModeAtomicCommit(m_gpu->fd(), req, flags, m_pageflipUserData) != 0) {
qCCritical(KWIN_DRM) << "Atomic commit failed! This never should've happened!" << strerror(errno);
printDebugInfo();
result = false;
}
if (result) {
m_oldTestBuffer = nullptr;
for (const auto &obj : qAsConst(m_allObjects)) {
obj->commitPending();
}
if (!testOnly) {
for (const auto &obj : qAsConst(m_allObjects)) {
obj->commit();
}
m_primaryPlane->setNext(m_primaryBuffer);
}
} else {
if (m_oldTestBuffer) {
m_primaryBuffer = m_oldTestBuffer;
m_oldTestBuffer = nullptr;
}
for (const auto &obj : qAsConst(m_allObjects)) {
obj->rollbackPending();
}
}
return result;
}
bool DrmPipeline::populateAtomicValues(drmModeAtomicReq *req, uint32_t &flags)
{
bool usesEglStreams = m_gpu->useEglStreams() && m_gpu->eglBackend() != nullptr && m_gpu == m_gpu->platform()->primaryGpu();
if (!usesEglStreams && m_active) {
flags |= DRM_MODE_PAGE_FLIP_EVENT;
}
bool needsModeset = std::any_of(m_allObjects.constBegin(), m_allObjects.constEnd(), [](auto obj){return obj->needsModeset();});
if (needsModeset) {
flags |= DRM_MODE_ATOMIC_ALLOW_MODESET;
} else {
flags |= DRM_MODE_ATOMIC_NONBLOCK;
}
m_lastFlags = flags;
auto modeSize = m_connector->currentMode().size;
m_primaryPlane->set(QPoint(0, 0), m_primaryBuffer ? m_primaryBuffer->size() : modeSize, QPoint(0, 0), modeSize);
m_primaryPlane->setBuffer(m_active ? m_primaryBuffer.get() : nullptr);
for (const auto &obj : qAsConst(m_allObjects)) {
if (!obj->atomicPopulate(req)) {
return false;
}
}
return true;
}
bool DrmPipeline::presentLegacy()
{
if ((!currentBuffer() || currentBuffer()->needsModeChange(m_primaryBuffer.get())) && !modeset(m_connector->currentModeIndex())) {
return false;
}
m_lastFlags = DRM_MODE_PAGE_FLIP_EVENT;
m_crtc->setNext(m_primaryBuffer);
if (drmModePageFlip(m_gpu->fd(), m_crtc->id(), m_primaryBuffer ? m_primaryBuffer->bufferId() : 0, DRM_MODE_PAGE_FLIP_EVENT, m_pageflipUserData) != 0) {
qCWarning(KWIN_DRM) << "Page flip failed:" << strerror(errno) << m_primaryBuffer;
return false;
}
return true;
}
bool DrmPipeline::modeset(int modeIndex)
{
int oldModeIndex = m_connector->currentModeIndex();
m_connector->setModeIndex(modeIndex);
auto mode = m_connector->currentMode().mode;
if (m_gpu->atomicModeSetting()) {
m_crtc->setPendingBlob(DrmCrtc::PropertyIndex::ModeId, &mode, sizeof(drmModeModeInfo));
if (m_connector->hasOverscan()) {
m_connector->setOverscan(m_connector->overscan(), m_connector->currentMode().size);
}
bool works = test();
// hardware rotation could fail in some modes, try again with soft rotation if possible
if (!works
&& transformation() != DrmPlane::Transformations(DrmPlane::Transformation::Rotate0)
&& setPendingTransformation(DrmPlane::Transformation::Rotate0)) {
// values are reset on the failing test, set them again
m_crtc->setPendingBlob(DrmCrtc::PropertyIndex::ModeId, &mode, sizeof(drmModeModeInfo));
if (m_connector->hasOverscan()) {
m_connector->setOverscan(m_connector->overscan(), m_connector->currentMode().size);
}
works = test();
}
if (!works) {
qCDebug(KWIN_DRM) << "Modeset failed!" << strerror(errno);
m_connector->setModeIndex(oldModeIndex);
return false;
}
} else {
uint32_t connId = m_connector->id();
if (!checkTestBuffer() || drmModeSetCrtc(m_gpu->fd(), m_crtc->id(), m_primaryBuffer->bufferId(), 0, 0, &connId, 1, &mode) != 0) {
qCWarning(KWIN_DRM) << "Modeset failed!" << strerror(errno);
m_connector->setModeIndex(oldModeIndex);
m_primaryBuffer = m_oldTestBuffer;
return false;
}
m_oldTestBuffer = nullptr;
m_legacyNeedsModeset = false;
}
return true;
}
bool DrmPipeline::checkTestBuffer()
{
if (m_primaryBuffer && m_primaryBuffer->size() == sourceSize()) {
return true;
}
#if HAVE_GBM
auto backend = m_gpu->eglBackend();
if (backend && m_pageflipUserData) {
auto buffer = backend->renderTestFrame(m_pageflipUserData);
if (buffer && buffer->bufferId()) {
m_oldTestBuffer = m_primaryBuffer;
m_primaryBuffer = buffer;
return true;
} else {
return false;
}
}
// we either don't have a DrmOutput or we're using QPainter
QSharedPointer<DrmBuffer> buffer;
if (backend && m_gpu->gbmDevice()) {
gbm_bo *bo = gbm_bo_create(m_gpu->gbmDevice(), sourceSize().width(), sourceSize().height(), GBM_FORMAT_XRGB8888, GBM_BO_USE_SCANOUT | GBM_BO_USE_RENDERING);
if (!bo) {
return false;
}
buffer = QSharedPointer<DrmGbmBuffer>::create(m_gpu, bo, nullptr);
} else {
buffer = QSharedPointer<DrmDumbBuffer>::create(m_gpu, sourceSize());
}
#else
auto buffer = QSharedPointer<DrmDumbBuffer>::create(m_gpu, sourceSize());
#endif
if (buffer && buffer->bufferId()) {
m_oldTestBuffer = m_primaryBuffer;
m_primaryBuffer = buffer;
return true;
}
return false;
}
bool DrmPipeline::setCursor(const QSharedPointer<DrmDumbBuffer> &buffer)
{
const QSize &s = buffer ? buffer->size() : QSize(64, 64);
if (drmModeSetCursor(m_gpu->fd(), m_crtc->id(), buffer ? buffer->handle() : 0, s.width(), s.height()) != 0) {
qCWarning(KWIN_DRM) << "Could not set cursor:" << strerror(errno);
return false;
}
m_cursor.buffer = buffer;
return true;
}
bool DrmPipeline::moveCursor(QPoint pos)
{
if (drmModeMoveCursor(m_gpu->fd(), m_crtc->id(), pos.x(), pos.y()) != 0) {
m_cursor.pos = pos;
return false;
}
return true;
}
bool DrmPipeline::setActive(bool active)
{
// disable the cursor before the primary plane to circumvent a crash in amdgpu
if (m_active && !active) {
if (drmModeSetCursor(m_gpu->fd(), m_crtc->id(), 0, 0, 0) != 0) {
qCWarning(KWIN_DRM) << "Could not set cursor:" << strerror(errno);
}
}
bool success = false;
auto mode = m_connector->currentMode().mode;
bool oldActive = m_active;
m_active = active;
if (m_gpu->atomicModeSetting()) {
m_connector->setPending(DrmConnector::PropertyIndex::CrtcId, active ? m_crtc->id() : 0);
m_crtc->setPending(DrmCrtc::PropertyIndex::Active, active);
m_crtc->setPendingBlob(DrmCrtc::PropertyIndex::ModeId, active ? &mode : nullptr, sizeof(drmModeModeInfo));
m_primaryPlane->setPending(DrmPlane::PropertyIndex::CrtcId, active ? m_crtc->id() : 0);
if (active) {
success = test();
if (!success) {
updateProperties();
success = test();
}
} else {
// immediately commit if disabling as there will be no present
success = atomicCommit();
}
} else {
auto dpmsProp = m_connector->getProp(DrmConnector::PropertyIndex::Dpms);
if (!dpmsProp) {
qCWarning(KWIN_DRM) << "Setting active failed: dpms property missing!";
} else {
success = drmModeConnectorSetProperty(m_gpu->fd(), m_connector->id(), dpmsProp->propId(), active ? DRM_MODE_DPMS_ON : DRM_MODE_DPMS_OFF) == 0;
}
}
if (!success) {
m_active = oldActive;
qCWarning(KWIN_DRM) << "Setting active to" << active << "failed" << strerror(errno);
}
if (m_active) {
// enable cursor (again)
setCursor(m_cursor.buffer);
}
return success;
}
bool DrmPipeline::setGammaRamp(const GammaRamp &ramp)
{
// There are old Intel iGPUs that don't have full support for setting
// the gamma ramp with AMS -> fall back to legacy without the property
if (m_gpu->atomicModeSetting() && m_crtc->getProp(DrmCrtc::PropertyIndex::Gamma_LUT)) {
struct drm_color_lut *gamma = new drm_color_lut[ramp.size()];
for (uint32_t i = 0; i < ramp.size(); i++) {
gamma[i].red = ramp.red()[i];
gamma[i].green = ramp.green()[i];
gamma[i].blue = ramp.blue()[i];
}
bool result = m_crtc->setPendingBlob(DrmCrtc::PropertyIndex::Gamma_LUT, gamma, ramp.size() * sizeof(drm_color_lut));
delete[] gamma;
if (!result) {
qCWarning(KWIN_DRM) << "Could not create gamma LUT property blob" << strerror(errno);
return false;
}
if (!test()) {
qCWarning(KWIN_DRM) << "Setting gamma failed!" << strerror(errno);
return false;
}
} else {
uint16_t *red = const_cast<uint16_t*>(ramp.red());
uint16_t *green = const_cast<uint16_t*>(ramp.green());
uint16_t *blue = const_cast<uint16_t*>(ramp.blue());
if (drmModeCrtcSetGamma(m_gpu->fd(), m_crtc->id(), ramp.size(), red, green, blue) != 0) {
qCWarning(KWIN_DRM) << "setting gamma failed!" << strerror(errno);
return false;
}
}
return true;
}
bool DrmPipeline::setTransformation(const DrmPlane::Transformations &transformation)
{
return setPendingTransformation(transformation) && test();
}
bool DrmPipeline::setPendingTransformation(const DrmPlane::Transformations &transformation)
{
if (this->transformation() == transformation) {
return true;
}
if (!m_gpu->atomicModeSetting()) {
return false;
}
if (!m_primaryPlane->setTransformation(transformation)) {
return false;
}
return true;
}
bool DrmPipeline::setSyncMode(RenderLoopPrivate::SyncMode syncMode)
{
auto vrrProp = m_crtc->getProp(DrmCrtc::PropertyIndex::VrrEnabled);
if (!vrrProp || !m_connector->vrrCapable()) {
return syncMode == RenderLoopPrivate::SyncMode::Fixed;
}
bool vrr = syncMode == RenderLoopPrivate::SyncMode::Adaptive;
if (vrrProp->pending() == vrr) {
return true;
}
if (m_gpu->atomicModeSetting()) {
vrrProp->setPending(vrr);
return test();
} else {
return drmModeObjectSetProperty(m_gpu->fd(), m_crtc->id(), DRM_MODE_OBJECT_CRTC, vrrProp->propId(), vrr) == 0;
}
}
bool DrmPipeline::setOverscan(uint32_t overscan)
{
if (overscan > 100 || (overscan != 0 && !m_connector->hasOverscan())) {
return false;
}
m_connector->setOverscan(overscan, m_connector->currentMode().size);
return test();
}
QSize DrmPipeline::sourceSize() const
{
auto mode = m_connector->currentMode();
if (transformation() & (DrmPlane::Transformation::Rotate90 | DrmPlane::Transformation::Rotate270)) {
return mode.size.transposed();
}
return mode.size;
}
DrmPlane::Transformations DrmPipeline::transformation() const
{
return m_primaryPlane ? m_primaryPlane->transformation() : DrmPlane::Transformation::Rotate0;
}
bool DrmPipeline::isActive() const
{
return m_active;
}
bool DrmPipeline::isCursorVisible() const
{
return m_cursor.buffer && QRect(m_cursor.pos, m_cursor.buffer->size()).intersects(QRect(QPoint(0, 0), m_connector->currentMode().size));
}
QPoint DrmPipeline::cursorPos() const
{
return m_cursor.pos;
}
DrmConnector *DrmPipeline::connector() const
{
return m_connector;
}
DrmCrtc *DrmPipeline::crtc() const
{
return m_crtc;
}
DrmPlane *DrmPipeline::primaryPlane() const
{
return m_primaryPlane;
}
DrmBuffer *DrmPipeline::currentBuffer() const
{
return m_primaryPlane ? m_primaryPlane->current().get() : m_crtc->current().get();
}
void DrmPipeline::pageFlipped()
{
m_crtc->flipBuffer();
if (m_primaryPlane) {
m_primaryPlane->flipBuffer();
}
}
void DrmPipeline::setUserData(DrmOutput *data)
{
m_pageflipUserData = data;
}
void DrmPipeline::updateProperties()
{
for (const auto &obj : qAsConst(m_allObjects)) {
obj->updateProperties();
}
}
static void printProps(DrmObject *object)
{
auto list = object->properties();
for (const auto &prop : list) {
if (prop) {
if (prop->isImmutable() || !prop->needsCommit()) {
qCWarning(KWIN_DRM).nospace() << "\t" << prop->name() << ": " << prop->current();
} else {
qCWarning(KWIN_DRM).nospace() << "\t" << prop->name() << ": " << prop->current() << "->" << prop->pending();
}
}
}
}
void DrmPipeline::printDebugInfo() const
{
if (m_lastFlags == 0) {
qCWarning(KWIN_DRM) << "Flags: none";
} else {
qCWarning(KWIN_DRM) << "Flags:";
if (m_lastFlags & DRM_MODE_PAGE_FLIP_EVENT) {
qCWarning(KWIN_DRM) << "\t DRM_MODE_PAGE_FLIP_EVENT";
}
if (m_lastFlags & DRM_MODE_ATOMIC_ALLOW_MODESET) {
qCWarning(KWIN_DRM) << "\t DRM_MODE_ATOMIC_ALLOW_MODESET";
}
if (m_lastFlags & DRM_MODE_PAGE_FLIP_ASYNC) {
qCWarning(KWIN_DRM) << "\t DRM_MODE_PAGE_FLIP_ASYNC";
}
}
qCWarning(KWIN_DRM) << "Drm objects:";
qCWarning(KWIN_DRM) << "connector" << m_connector->id();
auto list = m_connector->properties();
printProps(m_connector);
qCWarning(KWIN_DRM) << "crtc" << m_crtc->id();
printProps(m_crtc);
if (m_primaryPlane) {
qCWarning(KWIN_DRM) << "primary plane" << m_primaryPlane->id();
printProps(m_primaryPlane);
}
}
}