* [PATCH v4 0/2] drm: lima: devfreq and cooling device support
@ 2020-03-19 20:34 Martin Blumenstingl
2020-03-19 20:34 ` [PATCH v4 1/2] dt-bindings: gpu: mali-utgard: Add the #cooling-cells property Martin Blumenstingl
2020-03-19 20:34 ` [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support Martin Blumenstingl
0 siblings, 2 replies; 8+ messages in thread
From: Martin Blumenstingl @ 2020-03-19 20:34 UTC (permalink / raw)
To: yuq825, dri-devel
Cc: tomeu.vizoso, linux-pm, airlied, linux-kernel, steven.price,
linux-rockchip, wens, alyssa.rosenzweig, linux-amlogic,
Martin Blumenstingl
This is my attempt at adding devfreq (and cooling device) support to
the lima driver.
Test results from a Meson8m2 board:
TEST #1: glmark2-es2-drm --off-screen in an infinite loop while cycling
through all available frequencies using the userspace governor
From : To
: 182142857 318750000 425000000 510000000 637500000 time(ms)
182142857: 0 1274 1274 1273 1279 5399468
318750000: 1274 0 1274 1273 1272 5114700
425000000: 1276 1274 0 1272 1271 5122008
510000000: 1909 1273 1273 0 636 5274292
* 637500000: 640 1272 1272 1273 0 5186796
Total transition : 24834
TEST #2: glmark2-es2-drm --off-screen in an infinite loop with the
simple_ondemand governor
From : To
: 182142857 318750000 425000000 510000000 637500000 time(ms)
182142857: 0 0 0 0 203 318328
318750000: 53 0 0 0 21 56044
425000000: 27 18 0 0 2 34172
510000000: 27 6 14 0 1 41348
* 637500000: 95 50 33 48 0 2085312
Changes since RFC v3 at [2]:
- fix devfreq init error handling and allow lima_devfreq_fini to be
called multiple times (thanks Qiang)
- switch from atomic counter to a simple int which is only accessed
under the devfreq spinlock (thanks Qiang)
- union lock areas in same function (thanks Qiang)
- select DEVFREQ_GOV_SIMPLE_ONDEMAND like panfrost does (thanks Qiang)
- move lima_devfreq struct to lima_devfreq.h (thanks Qiang)
- fix duplicate variable in lima_sched_pipe_task_done
- only call dev_pm_opp_of_remove_table if dev_pm_opp_of_add_table
succeeded previously
- update copyright year to 2020
- rebased on top of drm-misc-next
- dropped RFC status
Changes since RFC v2 at [1]:
- added #cooling-cells to the dt-bindings (new patch #1)
- skip devfreq initialization when the operating-points-v2 property is
absent
- call dev_pm_opp_set_regulators() so devfreq will actually manage the
mali-supply regulator
- rebased on top of drm-misc-next-2020-02-21
Changes since RFC v1 at [0]:
- added lock to protect the statistics as these can be written
concurrently for example when the GP and PP IRQ are firing at the
same time. Thanks to Qiang Yu for the suggestion!
- updated the copyright notice of lima_devfreq.c to indicate that the
code is derived from panfrost_devfreq.c. Thanks to Chen-Yu Tsai for
the suggestion!
- I did not unify the code with panfrost yet because I don't know where
to put the result. any suggestion is welcome though!
[0] https://patchwork.freedesktop.org/series/70967/
[1] https://patchwork.kernel.org/cover/11311293/
[2] https://patchwork.kernel.org/cover/11398365/
Martin Blumenstingl (2):
dt-bindings: gpu: mali-utgard: Add the #cooling-cells property
drm/lima: Add optional devfreq and cooling device support
.../bindings/gpu/arm,mali-utgard.yaml | 4 +
drivers/gpu/drm/lima/Kconfig | 2 +
drivers/gpu/drm/lima/Makefile | 3 +-
drivers/gpu/drm/lima/lima_devfreq.c | 234 ++++++++++++++++++
drivers/gpu/drm/lima/lima_devfreq.h | 41 +++
drivers/gpu/drm/lima/lima_device.c | 4 +
drivers/gpu/drm/lima/lima_device.h | 3 +
drivers/gpu/drm/lima/lima_drv.c | 14 +-
drivers/gpu/drm/lima/lima_sched.c | 7 +
drivers/gpu/drm/lima/lima_sched.h | 3 +
10 files changed, 312 insertions(+), 3 deletions(-)
create mode 100644 drivers/gpu/drm/lima/lima_devfreq.c
create mode 100644 drivers/gpu/drm/lima/lima_devfreq.h
--
2.25.2
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 8+ messages in thread
* [PATCH v4 1/2] dt-bindings: gpu: mali-utgard: Add the #cooling-cells property
2020-03-19 20:34 [PATCH v4 0/2] drm: lima: devfreq and cooling device support Martin Blumenstingl
@ 2020-03-19 20:34 ` Martin Blumenstingl
2020-03-22 2:23 ` Qiang Yu
2020-03-19 20:34 ` [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support Martin Blumenstingl
1 sibling, 1 reply; 8+ messages in thread
From: Martin Blumenstingl @ 2020-03-19 20:34 UTC (permalink / raw)
To: yuq825, dri-devel
Cc: tomeu.vizoso, linux-pm, airlied, linux-kernel, steven.price,
linux-rockchip, wens, alyssa.rosenzweig, linux-amlogic,
Martin Blumenstingl
The GPU can be one of the big heat sources on a SoC. Allow the
"#cooling-cells" property to be specified for ARM Mali Utgard GPUs so
the GPU clock speeds (and voltages) can be reduced to prevent a SoC from
overheating.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
---
Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml b/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml
index afde81be3c29..33548ca2a759 100644
--- a/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml
+++ b/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml
@@ -107,6 +107,9 @@ properties:
operating-points-v2: true
+ "#cooling-cells":
+ const: 2
+
required:
- compatible
- reg
@@ -162,6 +165,7 @@ examples:
clocks = <&ccu 1>, <&ccu 2>;
clock-names = "bus", "core";
resets = <&ccu 1>;
+ #cooling-cells = <2>;
};
...
--
2.25.2
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support
2020-03-19 20:34 [PATCH v4 0/2] drm: lima: devfreq and cooling device support Martin Blumenstingl
2020-03-19 20:34 ` [PATCH v4 1/2] dt-bindings: gpu: mali-utgard: Add the #cooling-cells property Martin Blumenstingl
@ 2020-03-19 20:34 ` Martin Blumenstingl
2020-03-22 2:24 ` Qiang Yu
1 sibling, 1 reply; 8+ messages in thread
From: Martin Blumenstingl @ 2020-03-19 20:34 UTC (permalink / raw)
To: yuq825, dri-devel
Cc: tomeu.vizoso, linux-pm, airlied, linux-kernel, steven.price,
linux-rockchip, wens, alyssa.rosenzweig, linux-amlogic,
Martin Blumenstingl
Most platforms with a Mali-400 or Mali-450 GPU also have support for
changing the GPU clock frequency. Add devfreq support so the GPU clock
rate is updated based on the actual GPU usage when the
"operating-points-v2" property is present in the board.dts.
The actual devfreq code is taken from panfrost_devfreq.c and modified so
it matches what the lima hardware needs:
- a call to dev_pm_opp_set_clkname() during initialization because there
are two clocks on Mali-4x0 IPs. "core" is the one that actually clocks
the GPU so we need to control it using devfreq.
- locking when reading or writing the devfreq statistics because (unlike
than panfrost) we have multiple PP and GP IRQs which may finish jobs
concurrently.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
---
drivers/gpu/drm/lima/Kconfig | 2 +
drivers/gpu/drm/lima/Makefile | 3 +-
drivers/gpu/drm/lima/lima_devfreq.c | 234 ++++++++++++++++++++++++++++
drivers/gpu/drm/lima/lima_devfreq.h | 41 +++++
drivers/gpu/drm/lima/lima_device.c | 4 +
drivers/gpu/drm/lima/lima_device.h | 3 +
drivers/gpu/drm/lima/lima_drv.c | 14 +-
drivers/gpu/drm/lima/lima_sched.c | 7 +
drivers/gpu/drm/lima/lima_sched.h | 3 +
9 files changed, 308 insertions(+), 3 deletions(-)
create mode 100644 drivers/gpu/drm/lima/lima_devfreq.c
create mode 100644 drivers/gpu/drm/lima/lima_devfreq.h
diff --git a/drivers/gpu/drm/lima/Kconfig b/drivers/gpu/drm/lima/Kconfig
index d589f09d04d9..fa1d4f5df31e 100644
--- a/drivers/gpu/drm/lima/Kconfig
+++ b/drivers/gpu/drm/lima/Kconfig
@@ -10,5 +10,7 @@ config DRM_LIMA
depends on OF
select DRM_SCHED
select DRM_GEM_SHMEM_HELPER
+ select PM_DEVFREQ
+ select DEVFREQ_GOV_SIMPLE_ONDEMAND
help
DRM driver for ARM Mali 400/450 GPUs.
diff --git a/drivers/gpu/drm/lima/Makefile b/drivers/gpu/drm/lima/Makefile
index a85444b0a1d4..5e5c29875e9c 100644
--- a/drivers/gpu/drm/lima/Makefile
+++ b/drivers/gpu/drm/lima/Makefile
@@ -14,6 +14,7 @@ lima-y := \
lima_sched.o \
lima_ctx.o \
lima_dlbu.o \
- lima_bcast.o
+ lima_bcast.o \
+ lima_devfreq.o
obj-$(CONFIG_DRM_LIMA) += lima.o
diff --git a/drivers/gpu/drm/lima/lima_devfreq.c b/drivers/gpu/drm/lima/lima_devfreq.c
new file mode 100644
index 000000000000..8c4d21d07529
--- /dev/null
+++ b/drivers/gpu/drm/lima/lima_devfreq.c
@@ -0,0 +1,234 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2020 Martin Blumenstingl <martin.blumenstingl@googlemail.com>
+ *
+ * Based on panfrost_devfreq.c:
+ * Copyright 2019 Collabora ltd.
+ */
+#include <linux/clk.h>
+#include <linux/devfreq.h>
+#include <linux/devfreq_cooling.h>
+#include <linux/device.h>
+#include <linux/platform_device.h>
+#include <linux/pm_opp.h>
+#include <linux/property.h>
+
+#include "lima_device.h"
+#include "lima_devfreq.h"
+
+static void lima_devfreq_update_utilization(struct lima_devfreq *devfreq)
+{
+ ktime_t now, last;
+
+ now = ktime_get();
+ last = devfreq->time_last_update;
+
+ if (devfreq->busy_count > 0)
+ devfreq->busy_time += ktime_sub(now, last);
+ else
+ devfreq->idle_time += ktime_sub(now, last);
+
+ devfreq->time_last_update = now;
+}
+
+static int lima_devfreq_target(struct device *dev, unsigned long *freq,
+ u32 flags)
+{
+ struct dev_pm_opp *opp;
+ int err;
+
+ opp = devfreq_recommended_opp(dev, freq, flags);
+ if (IS_ERR(opp))
+ return PTR_ERR(opp);
+ dev_pm_opp_put(opp);
+
+ err = dev_pm_opp_set_rate(dev, *freq);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static void lima_devfreq_reset(struct lima_devfreq *devfreq)
+{
+ devfreq->busy_time = 0;
+ devfreq->idle_time = 0;
+ devfreq->time_last_update = ktime_get();
+}
+
+static int lima_devfreq_get_dev_status(struct device *dev,
+ struct devfreq_dev_status *status)
+{
+ struct lima_device *ldev = dev_get_drvdata(dev);
+ struct lima_devfreq *devfreq = &ldev->devfreq;
+ unsigned long irqflags;
+
+ status->current_frequency = clk_get_rate(ldev->clk_gpu);
+
+ spin_lock_irqsave(&devfreq->lock, irqflags);
+
+ lima_devfreq_update_utilization(devfreq);
+
+ status->total_time = ktime_to_ns(ktime_add(devfreq->busy_time,
+ devfreq->idle_time));
+ status->busy_time = ktime_to_ns(devfreq->busy_time);
+
+ lima_devfreq_reset(devfreq);
+
+ spin_unlock_irqrestore(&devfreq->lock, irqflags);
+
+ dev_dbg(ldev->dev, "busy %lu total %lu %lu %% freq %lu MHz\n",
+ status->busy_time, status->total_time,
+ status->busy_time / (status->total_time / 100),
+ status->current_frequency / 1000 / 1000);
+
+ return 0;
+}
+
+static struct devfreq_dev_profile lima_devfreq_profile = {
+ .polling_ms = 50, /* ~3 frames */
+ .target = lima_devfreq_target,
+ .get_dev_status = lima_devfreq_get_dev_status,
+};
+
+void lima_devfreq_fini(struct lima_device *ldev)
+{
+ struct lima_devfreq *devfreq = &ldev->devfreq;
+
+ if (devfreq->cooling) {
+ devfreq_cooling_unregister(devfreq->cooling);
+ devfreq->cooling = NULL;
+ }
+
+ if (devfreq->devfreq) {
+ devm_devfreq_remove_device(&ldev->pdev->dev,
+ devfreq->devfreq);
+ devfreq->devfreq = NULL;
+ }
+
+ if (devfreq->opp_of_table_added) {
+ dev_pm_opp_of_remove_table(&ldev->pdev->dev);
+ devfreq->opp_of_table_added = false;
+ }
+
+ if (devfreq->regulators_opp_table) {
+ dev_pm_opp_put_regulators(devfreq->regulators_opp_table);
+ devfreq->regulators_opp_table = NULL;
+ }
+
+ if (devfreq->clkname_opp_table) {
+ dev_pm_opp_put_clkname(devfreq->clkname_opp_table);
+ devfreq->clkname_opp_table = NULL;
+ }
+}
+
+int lima_devfreq_init(struct lima_device *ldev)
+{
+ struct thermal_cooling_device *cooling;
+ struct device *dev = &ldev->pdev->dev;
+ struct opp_table *opp_table;
+ struct devfreq *devfreq;
+ struct lima_devfreq *ldevfreq = &ldev->devfreq;
+ struct dev_pm_opp *opp;
+ unsigned long cur_freq;
+ int ret;
+
+ if (!device_property_present(dev, "operating-points-v2"))
+ /* Optional, continue without devfreq */
+ return 0;
+
+ spin_lock_init(&ldevfreq->lock);
+
+ opp_table = dev_pm_opp_set_clkname(dev, "core");
+ if (IS_ERR(opp_table)) {
+ ret = PTR_ERR(opp_table);
+ goto err_fini;
+ }
+
+ ldevfreq->clkname_opp_table = opp_table;
+
+ opp_table = dev_pm_opp_set_regulators(dev,
+ (const char *[]){ "mali" },
+ 1);
+ if (IS_ERR(opp_table)) {
+ ret = PTR_ERR(opp_table);
+
+ /* Continue if the optional regulator is missing */
+ if (ret != -ENODEV)
+ goto err_fini;
+ } else {
+ ldevfreq->regulators_opp_table = opp_table;
+ }
+
+ ret = dev_pm_opp_of_add_table(dev);
+ if (ret)
+ goto err_fini;
+ ldevfreq->opp_of_table_added = true;
+
+ lima_devfreq_reset(ldevfreq);
+
+ cur_freq = clk_get_rate(ldev->clk_gpu);
+
+ opp = devfreq_recommended_opp(dev, &cur_freq, 0);
+ if (IS_ERR(opp)) {
+ ret = PTR_ERR(opp);
+ goto err_fini;
+ }
+
+ lima_devfreq_profile.initial_freq = cur_freq;
+ dev_pm_opp_put(opp);
+
+ devfreq = devm_devfreq_add_device(dev, &lima_devfreq_profile,
+ DEVFREQ_GOV_SIMPLE_ONDEMAND, NULL);
+ if (IS_ERR(devfreq)) {
+ dev_err(dev, "Couldn't initialize GPU devfreq\n");
+ ret = PTR_ERR(devfreq);
+ goto err_fini;
+ }
+
+ ldevfreq->devfreq = devfreq;
+
+ cooling = of_devfreq_cooling_register(dev->of_node, devfreq);
+ if (IS_ERR(cooling))
+ dev_info(dev, "Failed to register cooling device\n");
+ else
+ ldevfreq->cooling = cooling;
+
+ return 0;
+
+err_fini:
+ lima_devfreq_fini(ldev);
+ return ret;
+}
+
+void lima_devfreq_record_busy(struct lima_devfreq *devfreq)
+{
+ unsigned long irqflags;
+
+ if (!devfreq->devfreq)
+ return;
+
+ spin_lock_irqsave(&devfreq->lock, irqflags);
+
+ lima_devfreq_update_utilization(devfreq);
+
+ devfreq->busy_count++;
+
+ spin_unlock_irqrestore(&devfreq->lock, irqflags);
+}
+
+void lima_devfreq_record_idle(struct lima_devfreq *devfreq)
+{
+ unsigned long irqflags;
+
+ if (!devfreq->devfreq)
+ return;
+
+ spin_lock_irqsave(&devfreq->lock, irqflags);
+
+ lima_devfreq_update_utilization(devfreq);
+
+ WARN_ON(--devfreq->busy_count < 0);
+
+ spin_unlock_irqrestore(&devfreq->lock, irqflags);
+}
diff --git a/drivers/gpu/drm/lima/lima_devfreq.h b/drivers/gpu/drm/lima/lima_devfreq.h
new file mode 100644
index 000000000000..8d71ba9fb22a
--- /dev/null
+++ b/drivers/gpu/drm/lima/lima_devfreq.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright 2020 Martin Blumenstingl <martin.blumenstingl@googlemail.com> */
+
+#ifndef __LIMA_DEVFREQ_H__
+#define __LIMA_DEVFREQ_H__
+
+#include <linux/spinlock.h>
+#include <linux/ktime.h>
+
+struct devfreq;
+struct opp_table;
+struct thermal_cooling_device;
+
+struct lima_device;
+
+struct lima_devfreq {
+ struct devfreq *devfreq;
+ struct opp_table *clkname_opp_table;
+ struct opp_table *regulators_opp_table;
+ struct thermal_cooling_device *cooling;
+ bool opp_of_table_added;
+
+ ktime_t busy_time;
+ ktime_t idle_time;
+ ktime_t time_last_update;
+ int busy_count;
+ /*
+ * Protect busy_time, idle_time, time_last_update and busy_count
+ * because these can be updated concurrently, for example by the GP
+ * and PP interrupts.
+ */
+ spinlock_t lock;
+};
+
+int lima_devfreq_init(struct lima_device *ldev);
+void lima_devfreq_fini(struct lima_device *ldev);
+
+void lima_devfreq_record_busy(struct lima_devfreq *devfreq);
+void lima_devfreq_record_idle(struct lima_devfreq *devfreq);
+
+#endif
diff --git a/drivers/gpu/drm/lima/lima_device.c b/drivers/gpu/drm/lima/lima_device.c
index 19829b543024..7f1f7a1c03e5 100644
--- a/drivers/gpu/drm/lima/lima_device.c
+++ b/drivers/gpu/drm/lima/lima_device.c
@@ -214,6 +214,8 @@ static int lima_init_gp_pipe(struct lima_device *dev)
struct lima_sched_pipe *pipe = dev->pipe + lima_pipe_gp;
int err;
+ pipe->ldev = dev;
+
err = lima_sched_pipe_init(pipe, "gp");
if (err)
return err;
@@ -244,6 +246,8 @@ static int lima_init_pp_pipe(struct lima_device *dev)
struct lima_sched_pipe *pipe = dev->pipe + lima_pipe_pp;
int err, i;
+ pipe->ldev = dev;
+
err = lima_sched_pipe_init(pipe, "pp");
if (err)
return err;
diff --git a/drivers/gpu/drm/lima/lima_device.h b/drivers/gpu/drm/lima/lima_device.h
index 31158d86271c..d6c0f8b8f7e8 100644
--- a/drivers/gpu/drm/lima/lima_device.h
+++ b/drivers/gpu/drm/lima/lima_device.h
@@ -8,6 +8,7 @@
#include <linux/delay.h>
#include "lima_sched.h"
+#include "lima_devfreq.h"
enum lima_gpu_id {
lima_gpu_mali400 = 0,
@@ -94,6 +95,8 @@ struct lima_device {
u32 *dlbu_cpu;
dma_addr_t dlbu_dma;
+
+ struct lima_devfreq devfreq;
};
static inline struct lima_device *
diff --git a/drivers/gpu/drm/lima/lima_drv.c b/drivers/gpu/drm/lima/lima_drv.c
index 2daac64d8955..d9f64e52ae9e 100644
--- a/drivers/gpu/drm/lima/lima_drv.c
+++ b/drivers/gpu/drm/lima/lima_drv.c
@@ -10,6 +10,7 @@
#include <drm/drm_prime.h>
#include <drm/lima_drm.h>
+#include "lima_device.h"
#include "lima_drv.h"
#include "lima_gem.h"
#include "lima_vm.h"
@@ -306,18 +307,26 @@ static int lima_pdev_probe(struct platform_device *pdev)
if (err)
goto err_out1;
+ err = lima_devfreq_init(ldev);
+ if (err) {
+ dev_err(&pdev->dev, "Fatal error during devfreq init\n");
+ goto err_out2;
+ }
+
/*
* Register the DRM device with the core and the connectors with
* sysfs.
*/
err = drm_dev_register(ddev, 0);
if (err < 0)
- goto err_out2;
+ goto err_out3;
return 0;
-err_out2:
+err_out3:
lima_device_fini(ldev);
+err_out2:
+ lima_devfreq_fini(ldev);
err_out1:
drm_dev_put(ddev);
err_out0:
@@ -331,6 +340,7 @@ static int lima_pdev_remove(struct platform_device *pdev)
struct drm_device *ddev = ldev->ddev;
drm_dev_unregister(ddev);
+ lima_devfreq_fini(ldev);
lima_device_fini(ldev);
drm_dev_put(ddev);
lima_sched_slab_fini();
diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index 3886999b4533..d937d626b892 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -5,6 +5,7 @@
#include <linux/slab.h>
#include <linux/xarray.h>
+#include "lima_devfreq.h"
#include "lima_drv.h"
#include "lima_sched.h"
#include "lima_vm.h"
@@ -214,6 +215,8 @@ static struct dma_fence *lima_sched_run_job(struct drm_sched_job *job)
*/
ret = dma_fence_get(task->fence);
+ lima_devfreq_record_busy(&pipe->ldev->devfreq);
+
pipe->current_task = task;
/* this is needed for MMU to work correctly, otherwise GP/PP
@@ -285,6 +288,8 @@ static void lima_sched_timedout_job(struct drm_sched_job *job)
pipe->current_vm = NULL;
pipe->current_task = NULL;
+ lima_devfreq_record_idle(&pipe->ldev->devfreq);
+
drm_sched_resubmit_jobs(&pipe->base);
drm_sched_start(&pipe->base, true);
}
@@ -364,5 +369,7 @@ void lima_sched_pipe_task_done(struct lima_sched_pipe *pipe)
} else {
pipe->task_fini(pipe);
dma_fence_signal(task->fence);
+
+ lima_devfreq_record_idle(&pipe->ldev->devfreq);
}
}
diff --git a/drivers/gpu/drm/lima/lima_sched.h b/drivers/gpu/drm/lima/lima_sched.h
index d64393fb50a9..19bbc5214cf2 100644
--- a/drivers/gpu/drm/lima/lima_sched.h
+++ b/drivers/gpu/drm/lima/lima_sched.h
@@ -6,6 +6,7 @@
#include <drm/gpu_scheduler.h>
+struct lima_device;
struct lima_vm;
struct lima_sched_task {
@@ -44,6 +45,8 @@ struct lima_sched_pipe {
u32 fence_seqno;
spinlock_t fence_lock;
+ struct lima_device *ldev;
+
struct lima_sched_task *current_task;
struct lima_vm *current_vm;
--
2.25.2
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: [PATCH v4 1/2] dt-bindings: gpu: mali-utgard: Add the #cooling-cells property
2020-03-19 20:34 ` [PATCH v4 1/2] dt-bindings: gpu: mali-utgard: Add the #cooling-cells property Martin Blumenstingl
@ 2020-03-22 2:23 ` Qiang Yu
0 siblings, 0 replies; 8+ messages in thread
From: Qiang Yu @ 2020-03-22 2:23 UTC (permalink / raw)
To: Martin Blumenstingl
Cc: Tomeu Vizoso, linux-pm, David Airlie, Linux Kernel Mailing List,
dri-devel, Steven Price, linux-rockchip, Chen-Yu Tsai,
Alyssa Rosenzweig, open list:ARM/Amlogic Meson...
Looks good for me, patch is:
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Regards,
Qiang
On Fri, Mar 20, 2020 at 4:35 AM Martin Blumenstingl
<martin.blumenstingl@googlemail.com> wrote:
>
> The GPU can be one of the big heat sources on a SoC. Allow the
> "#cooling-cells" property to be specified for ARM Mali Utgard GPUs so
> the GPU clock speeds (and voltages) can be reduced to prevent a SoC from
> overheating.
>
> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> ---
> Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml b/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml
> index afde81be3c29..33548ca2a759 100644
> --- a/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml
> +++ b/Documentation/devicetree/bindings/gpu/arm,mali-utgard.yaml
> @@ -107,6 +107,9 @@ properties:
>
> operating-points-v2: true
>
> + "#cooling-cells":
> + const: 2
> +
> required:
> - compatible
> - reg
> @@ -162,6 +165,7 @@ examples:
> clocks = <&ccu 1>, <&ccu 2>;
> clock-names = "bus", "core";
> resets = <&ccu 1>;
> + #cooling-cells = <2>;
> };
>
> ...
> --
> 2.25.2
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support
2020-03-19 20:34 ` [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support Martin Blumenstingl
@ 2020-03-22 2:24 ` Qiang Yu
2020-03-28 8:40 ` Qiang Yu
0 siblings, 1 reply; 8+ messages in thread
From: Qiang Yu @ 2020-03-22 2:24 UTC (permalink / raw)
To: Martin Blumenstingl
Cc: Tomeu Vizoso, linux-pm, David Airlie, Linux Kernel Mailing List,
dri-devel, Steven Price, linux-rockchip, Chen-Yu Tsai,
Alyssa Rosenzweig, open list:ARM/Amlogic Meson...
Looks good for me, patch is:
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Regards,
Qiang
On Fri, Mar 20, 2020 at 4:35 AM Martin Blumenstingl
<martin.blumenstingl@googlemail.com> wrote:
>
> Most platforms with a Mali-400 or Mali-450 GPU also have support for
> changing the GPU clock frequency. Add devfreq support so the GPU clock
> rate is updated based on the actual GPU usage when the
> "operating-points-v2" property is present in the board.dts.
>
> The actual devfreq code is taken from panfrost_devfreq.c and modified so
> it matches what the lima hardware needs:
> - a call to dev_pm_opp_set_clkname() during initialization because there
> are two clocks on Mali-4x0 IPs. "core" is the one that actually clocks
> the GPU so we need to control it using devfreq.
> - locking when reading or writing the devfreq statistics because (unlike
> than panfrost) we have multiple PP and GP IRQs which may finish jobs
> concurrently.
>
> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> ---
> drivers/gpu/drm/lima/Kconfig | 2 +
> drivers/gpu/drm/lima/Makefile | 3 +-
> drivers/gpu/drm/lima/lima_devfreq.c | 234 ++++++++++++++++++++++++++++
> drivers/gpu/drm/lima/lima_devfreq.h | 41 +++++
> drivers/gpu/drm/lima/lima_device.c | 4 +
> drivers/gpu/drm/lima/lima_device.h | 3 +
> drivers/gpu/drm/lima/lima_drv.c | 14 +-
> drivers/gpu/drm/lima/lima_sched.c | 7 +
> drivers/gpu/drm/lima/lima_sched.h | 3 +
> 9 files changed, 308 insertions(+), 3 deletions(-)
> create mode 100644 drivers/gpu/drm/lima/lima_devfreq.c
> create mode 100644 drivers/gpu/drm/lima/lima_devfreq.h
>
> diff --git a/drivers/gpu/drm/lima/Kconfig b/drivers/gpu/drm/lima/Kconfig
> index d589f09d04d9..fa1d4f5df31e 100644
> --- a/drivers/gpu/drm/lima/Kconfig
> +++ b/drivers/gpu/drm/lima/Kconfig
> @@ -10,5 +10,7 @@ config DRM_LIMA
> depends on OF
> select DRM_SCHED
> select DRM_GEM_SHMEM_HELPER
> + select PM_DEVFREQ
> + select DEVFREQ_GOV_SIMPLE_ONDEMAND
> help
> DRM driver for ARM Mali 400/450 GPUs.
> diff --git a/drivers/gpu/drm/lima/Makefile b/drivers/gpu/drm/lima/Makefile
> index a85444b0a1d4..5e5c29875e9c 100644
> --- a/drivers/gpu/drm/lima/Makefile
> +++ b/drivers/gpu/drm/lima/Makefile
> @@ -14,6 +14,7 @@ lima-y := \
> lima_sched.o \
> lima_ctx.o \
> lima_dlbu.o \
> - lima_bcast.o
> + lima_bcast.o \
> + lima_devfreq.o
>
> obj-$(CONFIG_DRM_LIMA) += lima.o
> diff --git a/drivers/gpu/drm/lima/lima_devfreq.c b/drivers/gpu/drm/lima/lima_devfreq.c
> new file mode 100644
> index 000000000000..8c4d21d07529
> --- /dev/null
> +++ b/drivers/gpu/drm/lima/lima_devfreq.c
> @@ -0,0 +1,234 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2020 Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> + *
> + * Based on panfrost_devfreq.c:
> + * Copyright 2019 Collabora ltd.
> + */
> +#include <linux/clk.h>
> +#include <linux/devfreq.h>
> +#include <linux/devfreq_cooling.h>
> +#include <linux/device.h>
> +#include <linux/platform_device.h>
> +#include <linux/pm_opp.h>
> +#include <linux/property.h>
> +
> +#include "lima_device.h"
> +#include "lima_devfreq.h"
> +
> +static void lima_devfreq_update_utilization(struct lima_devfreq *devfreq)
> +{
> + ktime_t now, last;
> +
> + now = ktime_get();
> + last = devfreq->time_last_update;
> +
> + if (devfreq->busy_count > 0)
> + devfreq->busy_time += ktime_sub(now, last);
> + else
> + devfreq->idle_time += ktime_sub(now, last);
> +
> + devfreq->time_last_update = now;
> +}
> +
> +static int lima_devfreq_target(struct device *dev, unsigned long *freq,
> + u32 flags)
> +{
> + struct dev_pm_opp *opp;
> + int err;
> +
> + opp = devfreq_recommended_opp(dev, freq, flags);
> + if (IS_ERR(opp))
> + return PTR_ERR(opp);
> + dev_pm_opp_put(opp);
> +
> + err = dev_pm_opp_set_rate(dev, *freq);
> + if (err)
> + return err;
> +
> + return 0;
> +}
> +
> +static void lima_devfreq_reset(struct lima_devfreq *devfreq)
> +{
> + devfreq->busy_time = 0;
> + devfreq->idle_time = 0;
> + devfreq->time_last_update = ktime_get();
> +}
> +
> +static int lima_devfreq_get_dev_status(struct device *dev,
> + struct devfreq_dev_status *status)
> +{
> + struct lima_device *ldev = dev_get_drvdata(dev);
> + struct lima_devfreq *devfreq = &ldev->devfreq;
> + unsigned long irqflags;
> +
> + status->current_frequency = clk_get_rate(ldev->clk_gpu);
> +
> + spin_lock_irqsave(&devfreq->lock, irqflags);
> +
> + lima_devfreq_update_utilization(devfreq);
> +
> + status->total_time = ktime_to_ns(ktime_add(devfreq->busy_time,
> + devfreq->idle_time));
> + status->busy_time = ktime_to_ns(devfreq->busy_time);
> +
> + lima_devfreq_reset(devfreq);
> +
> + spin_unlock_irqrestore(&devfreq->lock, irqflags);
> +
> + dev_dbg(ldev->dev, "busy %lu total %lu %lu %% freq %lu MHz\n",
> + status->busy_time, status->total_time,
> + status->busy_time / (status->total_time / 100),
> + status->current_frequency / 1000 / 1000);
> +
> + return 0;
> +}
> +
> +static struct devfreq_dev_profile lima_devfreq_profile = {
> + .polling_ms = 50, /* ~3 frames */
> + .target = lima_devfreq_target,
> + .get_dev_status = lima_devfreq_get_dev_status,
> +};
> +
> +void lima_devfreq_fini(struct lima_device *ldev)
> +{
> + struct lima_devfreq *devfreq = &ldev->devfreq;
> +
> + if (devfreq->cooling) {
> + devfreq_cooling_unregister(devfreq->cooling);
> + devfreq->cooling = NULL;
> + }
> +
> + if (devfreq->devfreq) {
> + devm_devfreq_remove_device(&ldev->pdev->dev,
> + devfreq->devfreq);
> + devfreq->devfreq = NULL;
> + }
> +
> + if (devfreq->opp_of_table_added) {
> + dev_pm_opp_of_remove_table(&ldev->pdev->dev);
> + devfreq->opp_of_table_added = false;
> + }
> +
> + if (devfreq->regulators_opp_table) {
> + dev_pm_opp_put_regulators(devfreq->regulators_opp_table);
> + devfreq->regulators_opp_table = NULL;
> + }
> +
> + if (devfreq->clkname_opp_table) {
> + dev_pm_opp_put_clkname(devfreq->clkname_opp_table);
> + devfreq->clkname_opp_table = NULL;
> + }
> +}
> +
> +int lima_devfreq_init(struct lima_device *ldev)
> +{
> + struct thermal_cooling_device *cooling;
> + struct device *dev = &ldev->pdev->dev;
> + struct opp_table *opp_table;
> + struct devfreq *devfreq;
> + struct lima_devfreq *ldevfreq = &ldev->devfreq;
> + struct dev_pm_opp *opp;
> + unsigned long cur_freq;
> + int ret;
> +
> + if (!device_property_present(dev, "operating-points-v2"))
> + /* Optional, continue without devfreq */
> + return 0;
> +
> + spin_lock_init(&ldevfreq->lock);
> +
> + opp_table = dev_pm_opp_set_clkname(dev, "core");
> + if (IS_ERR(opp_table)) {
> + ret = PTR_ERR(opp_table);
> + goto err_fini;
> + }
> +
> + ldevfreq->clkname_opp_table = opp_table;
> +
> + opp_table = dev_pm_opp_set_regulators(dev,
> + (const char *[]){ "mali" },
> + 1);
> + if (IS_ERR(opp_table)) {
> + ret = PTR_ERR(opp_table);
> +
> + /* Continue if the optional regulator is missing */
> + if (ret != -ENODEV)
> + goto err_fini;
> + } else {
> + ldevfreq->regulators_opp_table = opp_table;
> + }
> +
> + ret = dev_pm_opp_of_add_table(dev);
> + if (ret)
> + goto err_fini;
> + ldevfreq->opp_of_table_added = true;
> +
> + lima_devfreq_reset(ldevfreq);
> +
> + cur_freq = clk_get_rate(ldev->clk_gpu);
> +
> + opp = devfreq_recommended_opp(dev, &cur_freq, 0);
> + if (IS_ERR(opp)) {
> + ret = PTR_ERR(opp);
> + goto err_fini;
> + }
> +
> + lima_devfreq_profile.initial_freq = cur_freq;
> + dev_pm_opp_put(opp);
> +
> + devfreq = devm_devfreq_add_device(dev, &lima_devfreq_profile,
> + DEVFREQ_GOV_SIMPLE_ONDEMAND, NULL);
> + if (IS_ERR(devfreq)) {
> + dev_err(dev, "Couldn't initialize GPU devfreq\n");
> + ret = PTR_ERR(devfreq);
> + goto err_fini;
> + }
> +
> + ldevfreq->devfreq = devfreq;
> +
> + cooling = of_devfreq_cooling_register(dev->of_node, devfreq);
> + if (IS_ERR(cooling))
> + dev_info(dev, "Failed to register cooling device\n");
> + else
> + ldevfreq->cooling = cooling;
> +
> + return 0;
> +
> +err_fini:
> + lima_devfreq_fini(ldev);
> + return ret;
> +}
> +
> +void lima_devfreq_record_busy(struct lima_devfreq *devfreq)
> +{
> + unsigned long irqflags;
> +
> + if (!devfreq->devfreq)
> + return;
> +
> + spin_lock_irqsave(&devfreq->lock, irqflags);
> +
> + lima_devfreq_update_utilization(devfreq);
> +
> + devfreq->busy_count++;
> +
> + spin_unlock_irqrestore(&devfreq->lock, irqflags);
> +}
> +
> +void lima_devfreq_record_idle(struct lima_devfreq *devfreq)
> +{
> + unsigned long irqflags;
> +
> + if (!devfreq->devfreq)
> + return;
> +
> + spin_lock_irqsave(&devfreq->lock, irqflags);
> +
> + lima_devfreq_update_utilization(devfreq);
> +
> + WARN_ON(--devfreq->busy_count < 0);
> +
> + spin_unlock_irqrestore(&devfreq->lock, irqflags);
> +}
> diff --git a/drivers/gpu/drm/lima/lima_devfreq.h b/drivers/gpu/drm/lima/lima_devfreq.h
> new file mode 100644
> index 000000000000..8d71ba9fb22a
> --- /dev/null
> +++ b/drivers/gpu/drm/lima/lima_devfreq.h
> @@ -0,0 +1,41 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/* Copyright 2020 Martin Blumenstingl <martin.blumenstingl@googlemail.com> */
> +
> +#ifndef __LIMA_DEVFREQ_H__
> +#define __LIMA_DEVFREQ_H__
> +
> +#include <linux/spinlock.h>
> +#include <linux/ktime.h>
> +
> +struct devfreq;
> +struct opp_table;
> +struct thermal_cooling_device;
> +
> +struct lima_device;
> +
> +struct lima_devfreq {
> + struct devfreq *devfreq;
> + struct opp_table *clkname_opp_table;
> + struct opp_table *regulators_opp_table;
> + struct thermal_cooling_device *cooling;
> + bool opp_of_table_added;
> +
> + ktime_t busy_time;
> + ktime_t idle_time;
> + ktime_t time_last_update;
> + int busy_count;
> + /*
> + * Protect busy_time, idle_time, time_last_update and busy_count
> + * because these can be updated concurrently, for example by the GP
> + * and PP interrupts.
> + */
> + spinlock_t lock;
> +};
> +
> +int lima_devfreq_init(struct lima_device *ldev);
> +void lima_devfreq_fini(struct lima_device *ldev);
> +
> +void lima_devfreq_record_busy(struct lima_devfreq *devfreq);
> +void lima_devfreq_record_idle(struct lima_devfreq *devfreq);
> +
> +#endif
> diff --git a/drivers/gpu/drm/lima/lima_device.c b/drivers/gpu/drm/lima/lima_device.c
> index 19829b543024..7f1f7a1c03e5 100644
> --- a/drivers/gpu/drm/lima/lima_device.c
> +++ b/drivers/gpu/drm/lima/lima_device.c
> @@ -214,6 +214,8 @@ static int lima_init_gp_pipe(struct lima_device *dev)
> struct lima_sched_pipe *pipe = dev->pipe + lima_pipe_gp;
> int err;
>
> + pipe->ldev = dev;
> +
> err = lima_sched_pipe_init(pipe, "gp");
> if (err)
> return err;
> @@ -244,6 +246,8 @@ static int lima_init_pp_pipe(struct lima_device *dev)
> struct lima_sched_pipe *pipe = dev->pipe + lima_pipe_pp;
> int err, i;
>
> + pipe->ldev = dev;
> +
> err = lima_sched_pipe_init(pipe, "pp");
> if (err)
> return err;
> diff --git a/drivers/gpu/drm/lima/lima_device.h b/drivers/gpu/drm/lima/lima_device.h
> index 31158d86271c..d6c0f8b8f7e8 100644
> --- a/drivers/gpu/drm/lima/lima_device.h
> +++ b/drivers/gpu/drm/lima/lima_device.h
> @@ -8,6 +8,7 @@
> #include <linux/delay.h>
>
> #include "lima_sched.h"
> +#include "lima_devfreq.h"
>
> enum lima_gpu_id {
> lima_gpu_mali400 = 0,
> @@ -94,6 +95,8 @@ struct lima_device {
>
> u32 *dlbu_cpu;
> dma_addr_t dlbu_dma;
> +
> + struct lima_devfreq devfreq;
> };
>
> static inline struct lima_device *
> diff --git a/drivers/gpu/drm/lima/lima_drv.c b/drivers/gpu/drm/lima/lima_drv.c
> index 2daac64d8955..d9f64e52ae9e 100644
> --- a/drivers/gpu/drm/lima/lima_drv.c
> +++ b/drivers/gpu/drm/lima/lima_drv.c
> @@ -10,6 +10,7 @@
> #include <drm/drm_prime.h>
> #include <drm/lima_drm.h>
>
> +#include "lima_device.h"
> #include "lima_drv.h"
> #include "lima_gem.h"
> #include "lima_vm.h"
> @@ -306,18 +307,26 @@ static int lima_pdev_probe(struct platform_device *pdev)
> if (err)
> goto err_out1;
>
> + err = lima_devfreq_init(ldev);
> + if (err) {
> + dev_err(&pdev->dev, "Fatal error during devfreq init\n");
> + goto err_out2;
> + }
> +
> /*
> * Register the DRM device with the core and the connectors with
> * sysfs.
> */
> err = drm_dev_register(ddev, 0);
> if (err < 0)
> - goto err_out2;
> + goto err_out3;
>
> return 0;
>
> -err_out2:
> +err_out3:
> lima_device_fini(ldev);
> +err_out2:
> + lima_devfreq_fini(ldev);
> err_out1:
> drm_dev_put(ddev);
> err_out0:
> @@ -331,6 +340,7 @@ static int lima_pdev_remove(struct platform_device *pdev)
> struct drm_device *ddev = ldev->ddev;
>
> drm_dev_unregister(ddev);
> + lima_devfreq_fini(ldev);
> lima_device_fini(ldev);
> drm_dev_put(ddev);
> lima_sched_slab_fini();
> diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
> index 3886999b4533..d937d626b892 100644
> --- a/drivers/gpu/drm/lima/lima_sched.c
> +++ b/drivers/gpu/drm/lima/lima_sched.c
> @@ -5,6 +5,7 @@
> #include <linux/slab.h>
> #include <linux/xarray.h>
>
> +#include "lima_devfreq.h"
> #include "lima_drv.h"
> #include "lima_sched.h"
> #include "lima_vm.h"
> @@ -214,6 +215,8 @@ static struct dma_fence *lima_sched_run_job(struct drm_sched_job *job)
> */
> ret = dma_fence_get(task->fence);
>
> + lima_devfreq_record_busy(&pipe->ldev->devfreq);
> +
> pipe->current_task = task;
>
> /* this is needed for MMU to work correctly, otherwise GP/PP
> @@ -285,6 +288,8 @@ static void lima_sched_timedout_job(struct drm_sched_job *job)
> pipe->current_vm = NULL;
> pipe->current_task = NULL;
>
> + lima_devfreq_record_idle(&pipe->ldev->devfreq);
> +
> drm_sched_resubmit_jobs(&pipe->base);
> drm_sched_start(&pipe->base, true);
> }
> @@ -364,5 +369,7 @@ void lima_sched_pipe_task_done(struct lima_sched_pipe *pipe)
> } else {
> pipe->task_fini(pipe);
> dma_fence_signal(task->fence);
> +
> + lima_devfreq_record_idle(&pipe->ldev->devfreq);
> }
> }
> diff --git a/drivers/gpu/drm/lima/lima_sched.h b/drivers/gpu/drm/lima/lima_sched.h
> index d64393fb50a9..19bbc5214cf2 100644
> --- a/drivers/gpu/drm/lima/lima_sched.h
> +++ b/drivers/gpu/drm/lima/lima_sched.h
> @@ -6,6 +6,7 @@
>
> #include <drm/gpu_scheduler.h>
>
> +struct lima_device;
> struct lima_vm;
>
> struct lima_sched_task {
> @@ -44,6 +45,8 @@ struct lima_sched_pipe {
> u32 fence_seqno;
> spinlock_t fence_lock;
>
> + struct lima_device *ldev;
> +
> struct lima_sched_task *current_task;
> struct lima_vm *current_vm;
>
> --
> 2.25.2
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support
2020-03-22 2:24 ` Qiang Yu
@ 2020-03-28 8:40 ` Qiang Yu
2020-03-28 10:20 ` Martin Blumenstingl
0 siblings, 1 reply; 8+ messages in thread
From: Qiang Yu @ 2020-03-28 8:40 UTC (permalink / raw)
To: Martin Blumenstingl
Cc: Tomeu Vizoso, linux-pm, David Airlie, Linux Kernel Mailing List,
dri-devel, Steven Price, linux-rockchip, Chen-Yu Tsai,
Alyssa Rosenzweig, open list:ARM/Amlogic Meson...
Applied to drm-misc-next.
On Sun, Mar 22, 2020 at 10:24 AM Qiang Yu <yuq825@gmail.com> wrote:
>
> Looks good for me, patch is:
> Reviewed-by: Qiang Yu <yuq825@gmail.com>
>
> Regards,
> Qiang
>
> On Fri, Mar 20, 2020 at 4:35 AM Martin Blumenstingl
> <martin.blumenstingl@googlemail.com> wrote:
> >
> > Most platforms with a Mali-400 or Mali-450 GPU also have support for
> > changing the GPU clock frequency. Add devfreq support so the GPU clock
> > rate is updated based on the actual GPU usage when the
> > "operating-points-v2" property is present in the board.dts.
> >
> > The actual devfreq code is taken from panfrost_devfreq.c and modified so
> > it matches what the lima hardware needs:
> > - a call to dev_pm_opp_set_clkname() during initialization because there
> > are two clocks on Mali-4x0 IPs. "core" is the one that actually clocks
> > the GPU so we need to control it using devfreq.
> > - locking when reading or writing the devfreq statistics because (unlike
> > than panfrost) we have multiple PP and GP IRQs which may finish jobs
> > concurrently.
> >
> > Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> > ---
> > drivers/gpu/drm/lima/Kconfig | 2 +
> > drivers/gpu/drm/lima/Makefile | 3 +-
> > drivers/gpu/drm/lima/lima_devfreq.c | 234 ++++++++++++++++++++++++++++
> > drivers/gpu/drm/lima/lima_devfreq.h | 41 +++++
> > drivers/gpu/drm/lima/lima_device.c | 4 +
> > drivers/gpu/drm/lima/lima_device.h | 3 +
> > drivers/gpu/drm/lima/lima_drv.c | 14 +-
> > drivers/gpu/drm/lima/lima_sched.c | 7 +
> > drivers/gpu/drm/lima/lima_sched.h | 3 +
> > 9 files changed, 308 insertions(+), 3 deletions(-)
> > create mode 100644 drivers/gpu/drm/lima/lima_devfreq.c
> > create mode 100644 drivers/gpu/drm/lima/lima_devfreq.h
> >
> > diff --git a/drivers/gpu/drm/lima/Kconfig b/drivers/gpu/drm/lima/Kconfig
> > index d589f09d04d9..fa1d4f5df31e 100644
> > --- a/drivers/gpu/drm/lima/Kconfig
> > +++ b/drivers/gpu/drm/lima/Kconfig
> > @@ -10,5 +10,7 @@ config DRM_LIMA
> > depends on OF
> > select DRM_SCHED
> > select DRM_GEM_SHMEM_HELPER
> > + select PM_DEVFREQ
> > + select DEVFREQ_GOV_SIMPLE_ONDEMAND
> > help
> > DRM driver for ARM Mali 400/450 GPUs.
> > diff --git a/drivers/gpu/drm/lima/Makefile b/drivers/gpu/drm/lima/Makefile
> > index a85444b0a1d4..5e5c29875e9c 100644
> > --- a/drivers/gpu/drm/lima/Makefile
> > +++ b/drivers/gpu/drm/lima/Makefile
> > @@ -14,6 +14,7 @@ lima-y := \
> > lima_sched.o \
> > lima_ctx.o \
> > lima_dlbu.o \
> > - lima_bcast.o
> > + lima_bcast.o \
> > + lima_devfreq.o
> >
> > obj-$(CONFIG_DRM_LIMA) += lima.o
> > diff --git a/drivers/gpu/drm/lima/lima_devfreq.c b/drivers/gpu/drm/lima/lima_devfreq.c
> > new file mode 100644
> > index 000000000000..8c4d21d07529
> > --- /dev/null
> > +++ b/drivers/gpu/drm/lima/lima_devfreq.c
> > @@ -0,0 +1,234 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * Copyright 2020 Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> > + *
> > + * Based on panfrost_devfreq.c:
> > + * Copyright 2019 Collabora ltd.
> > + */
> > +#include <linux/clk.h>
> > +#include <linux/devfreq.h>
> > +#include <linux/devfreq_cooling.h>
> > +#include <linux/device.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/pm_opp.h>
> > +#include <linux/property.h>
> > +
> > +#include "lima_device.h"
> > +#include "lima_devfreq.h"
> > +
> > +static void lima_devfreq_update_utilization(struct lima_devfreq *devfreq)
> > +{
> > + ktime_t now, last;
> > +
> > + now = ktime_get();
> > + last = devfreq->time_last_update;
> > +
> > + if (devfreq->busy_count > 0)
> > + devfreq->busy_time += ktime_sub(now, last);
> > + else
> > + devfreq->idle_time += ktime_sub(now, last);
> > +
> > + devfreq->time_last_update = now;
> > +}
> > +
> > +static int lima_devfreq_target(struct device *dev, unsigned long *freq,
> > + u32 flags)
> > +{
> > + struct dev_pm_opp *opp;
> > + int err;
> > +
> > + opp = devfreq_recommended_opp(dev, freq, flags);
> > + if (IS_ERR(opp))
> > + return PTR_ERR(opp);
> > + dev_pm_opp_put(opp);
> > +
> > + err = dev_pm_opp_set_rate(dev, *freq);
> > + if (err)
> > + return err;
> > +
> > + return 0;
> > +}
> > +
> > +static void lima_devfreq_reset(struct lima_devfreq *devfreq)
> > +{
> > + devfreq->busy_time = 0;
> > + devfreq->idle_time = 0;
> > + devfreq->time_last_update = ktime_get();
> > +}
> > +
> > +static int lima_devfreq_get_dev_status(struct device *dev,
> > + struct devfreq_dev_status *status)
> > +{
> > + struct lima_device *ldev = dev_get_drvdata(dev);
> > + struct lima_devfreq *devfreq = &ldev->devfreq;
> > + unsigned long irqflags;
> > +
> > + status->current_frequency = clk_get_rate(ldev->clk_gpu);
> > +
> > + spin_lock_irqsave(&devfreq->lock, irqflags);
> > +
> > + lima_devfreq_update_utilization(devfreq);
> > +
> > + status->total_time = ktime_to_ns(ktime_add(devfreq->busy_time,
> > + devfreq->idle_time));
> > + status->busy_time = ktime_to_ns(devfreq->busy_time);
> > +
> > + lima_devfreq_reset(devfreq);
> > +
> > + spin_unlock_irqrestore(&devfreq->lock, irqflags);
> > +
> > + dev_dbg(ldev->dev, "busy %lu total %lu %lu %% freq %lu MHz\n",
> > + status->busy_time, status->total_time,
> > + status->busy_time / (status->total_time / 100),
> > + status->current_frequency / 1000 / 1000);
> > +
> > + return 0;
> > +}
> > +
> > +static struct devfreq_dev_profile lima_devfreq_profile = {
> > + .polling_ms = 50, /* ~3 frames */
> > + .target = lima_devfreq_target,
> > + .get_dev_status = lima_devfreq_get_dev_status,
> > +};
> > +
> > +void lima_devfreq_fini(struct lima_device *ldev)
> > +{
> > + struct lima_devfreq *devfreq = &ldev->devfreq;
> > +
> > + if (devfreq->cooling) {
> > + devfreq_cooling_unregister(devfreq->cooling);
> > + devfreq->cooling = NULL;
> > + }
> > +
> > + if (devfreq->devfreq) {
> > + devm_devfreq_remove_device(&ldev->pdev->dev,
> > + devfreq->devfreq);
> > + devfreq->devfreq = NULL;
> > + }
> > +
> > + if (devfreq->opp_of_table_added) {
> > + dev_pm_opp_of_remove_table(&ldev->pdev->dev);
> > + devfreq->opp_of_table_added = false;
> > + }
> > +
> > + if (devfreq->regulators_opp_table) {
> > + dev_pm_opp_put_regulators(devfreq->regulators_opp_table);
> > + devfreq->regulators_opp_table = NULL;
> > + }
> > +
> > + if (devfreq->clkname_opp_table) {
> > + dev_pm_opp_put_clkname(devfreq->clkname_opp_table);
> > + devfreq->clkname_opp_table = NULL;
> > + }
> > +}
> > +
> > +int lima_devfreq_init(struct lima_device *ldev)
> > +{
> > + struct thermal_cooling_device *cooling;
> > + struct device *dev = &ldev->pdev->dev;
> > + struct opp_table *opp_table;
> > + struct devfreq *devfreq;
> > + struct lima_devfreq *ldevfreq = &ldev->devfreq;
> > + struct dev_pm_opp *opp;
> > + unsigned long cur_freq;
> > + int ret;
> > +
> > + if (!device_property_present(dev, "operating-points-v2"))
> > + /* Optional, continue without devfreq */
> > + return 0;
> > +
> > + spin_lock_init(&ldevfreq->lock);
> > +
> > + opp_table = dev_pm_opp_set_clkname(dev, "core");
> > + if (IS_ERR(opp_table)) {
> > + ret = PTR_ERR(opp_table);
> > + goto err_fini;
> > + }
> > +
> > + ldevfreq->clkname_opp_table = opp_table;
> > +
> > + opp_table = dev_pm_opp_set_regulators(dev,
> > + (const char *[]){ "mali" },
> > + 1);
> > + if (IS_ERR(opp_table)) {
> > + ret = PTR_ERR(opp_table);
> > +
> > + /* Continue if the optional regulator is missing */
> > + if (ret != -ENODEV)
> > + goto err_fini;
> > + } else {
> > + ldevfreq->regulators_opp_table = opp_table;
> > + }
> > +
> > + ret = dev_pm_opp_of_add_table(dev);
> > + if (ret)
> > + goto err_fini;
> > + ldevfreq->opp_of_table_added = true;
> > +
> > + lima_devfreq_reset(ldevfreq);
> > +
> > + cur_freq = clk_get_rate(ldev->clk_gpu);
> > +
> > + opp = devfreq_recommended_opp(dev, &cur_freq, 0);
> > + if (IS_ERR(opp)) {
> > + ret = PTR_ERR(opp);
> > + goto err_fini;
> > + }
> > +
> > + lima_devfreq_profile.initial_freq = cur_freq;
> > + dev_pm_opp_put(opp);
> > +
> > + devfreq = devm_devfreq_add_device(dev, &lima_devfreq_profile,
> > + DEVFREQ_GOV_SIMPLE_ONDEMAND, NULL);
> > + if (IS_ERR(devfreq)) {
> > + dev_err(dev, "Couldn't initialize GPU devfreq\n");
> > + ret = PTR_ERR(devfreq);
> > + goto err_fini;
> > + }
> > +
> > + ldevfreq->devfreq = devfreq;
> > +
> > + cooling = of_devfreq_cooling_register(dev->of_node, devfreq);
> > + if (IS_ERR(cooling))
> > + dev_info(dev, "Failed to register cooling device\n");
> > + else
> > + ldevfreq->cooling = cooling;
> > +
> > + return 0;
> > +
> > +err_fini:
> > + lima_devfreq_fini(ldev);
> > + return ret;
> > +}
> > +
> > +void lima_devfreq_record_busy(struct lima_devfreq *devfreq)
> > +{
> > + unsigned long irqflags;
> > +
> > + if (!devfreq->devfreq)
> > + return;
> > +
> > + spin_lock_irqsave(&devfreq->lock, irqflags);
> > +
> > + lima_devfreq_update_utilization(devfreq);
> > +
> > + devfreq->busy_count++;
> > +
> > + spin_unlock_irqrestore(&devfreq->lock, irqflags);
> > +}
> > +
> > +void lima_devfreq_record_idle(struct lima_devfreq *devfreq)
> > +{
> > + unsigned long irqflags;
> > +
> > + if (!devfreq->devfreq)
> > + return;
> > +
> > + spin_lock_irqsave(&devfreq->lock, irqflags);
> > +
> > + lima_devfreq_update_utilization(devfreq);
> > +
> > + WARN_ON(--devfreq->busy_count < 0);
> > +
> > + spin_unlock_irqrestore(&devfreq->lock, irqflags);
> > +}
> > diff --git a/drivers/gpu/drm/lima/lima_devfreq.h b/drivers/gpu/drm/lima/lima_devfreq.h
> > new file mode 100644
> > index 000000000000..8d71ba9fb22a
> > --- /dev/null
> > +++ b/drivers/gpu/drm/lima/lima_devfreq.h
> > @@ -0,0 +1,41 @@
> > +/* SPDX-License-Identifier: GPL-2.0 */
> > +/* Copyright 2020 Martin Blumenstingl <martin.blumenstingl@googlemail.com> */
> > +
> > +#ifndef __LIMA_DEVFREQ_H__
> > +#define __LIMA_DEVFREQ_H__
> > +
> > +#include <linux/spinlock.h>
> > +#include <linux/ktime.h>
> > +
> > +struct devfreq;
> > +struct opp_table;
> > +struct thermal_cooling_device;
> > +
> > +struct lima_device;
> > +
> > +struct lima_devfreq {
> > + struct devfreq *devfreq;
> > + struct opp_table *clkname_opp_table;
> > + struct opp_table *regulators_opp_table;
> > + struct thermal_cooling_device *cooling;
> > + bool opp_of_table_added;
> > +
> > + ktime_t busy_time;
> > + ktime_t idle_time;
> > + ktime_t time_last_update;
> > + int busy_count;
> > + /*
> > + * Protect busy_time, idle_time, time_last_update and busy_count
> > + * because these can be updated concurrently, for example by the GP
> > + * and PP interrupts.
> > + */
> > + spinlock_t lock;
> > +};
> > +
> > +int lima_devfreq_init(struct lima_device *ldev);
> > +void lima_devfreq_fini(struct lima_device *ldev);
> > +
> > +void lima_devfreq_record_busy(struct lima_devfreq *devfreq);
> > +void lima_devfreq_record_idle(struct lima_devfreq *devfreq);
> > +
> > +#endif
> > diff --git a/drivers/gpu/drm/lima/lima_device.c b/drivers/gpu/drm/lima/lima_device.c
> > index 19829b543024..7f1f7a1c03e5 100644
> > --- a/drivers/gpu/drm/lima/lima_device.c
> > +++ b/drivers/gpu/drm/lima/lima_device.c
> > @@ -214,6 +214,8 @@ static int lima_init_gp_pipe(struct lima_device *dev)
> > struct lima_sched_pipe *pipe = dev->pipe + lima_pipe_gp;
> > int err;
> >
> > + pipe->ldev = dev;
> > +
> > err = lima_sched_pipe_init(pipe, "gp");
> > if (err)
> > return err;
> > @@ -244,6 +246,8 @@ static int lima_init_pp_pipe(struct lima_device *dev)
> > struct lima_sched_pipe *pipe = dev->pipe + lima_pipe_pp;
> > int err, i;
> >
> > + pipe->ldev = dev;
> > +
> > err = lima_sched_pipe_init(pipe, "pp");
> > if (err)
> > return err;
> > diff --git a/drivers/gpu/drm/lima/lima_device.h b/drivers/gpu/drm/lima/lima_device.h
> > index 31158d86271c..d6c0f8b8f7e8 100644
> > --- a/drivers/gpu/drm/lima/lima_device.h
> > +++ b/drivers/gpu/drm/lima/lima_device.h
> > @@ -8,6 +8,7 @@
> > #include <linux/delay.h>
> >
> > #include "lima_sched.h"
> > +#include "lima_devfreq.h"
> >
> > enum lima_gpu_id {
> > lima_gpu_mali400 = 0,
> > @@ -94,6 +95,8 @@ struct lima_device {
> >
> > u32 *dlbu_cpu;
> > dma_addr_t dlbu_dma;
> > +
> > + struct lima_devfreq devfreq;
> > };
> >
> > static inline struct lima_device *
> > diff --git a/drivers/gpu/drm/lima/lima_drv.c b/drivers/gpu/drm/lima/lima_drv.c
> > index 2daac64d8955..d9f64e52ae9e 100644
> > --- a/drivers/gpu/drm/lima/lima_drv.c
> > +++ b/drivers/gpu/drm/lima/lima_drv.c
> > @@ -10,6 +10,7 @@
> > #include <drm/drm_prime.h>
> > #include <drm/lima_drm.h>
> >
> > +#include "lima_device.h"
> > #include "lima_drv.h"
> > #include "lima_gem.h"
> > #include "lima_vm.h"
> > @@ -306,18 +307,26 @@ static int lima_pdev_probe(struct platform_device *pdev)
> > if (err)
> > goto err_out1;
> >
> > + err = lima_devfreq_init(ldev);
> > + if (err) {
> > + dev_err(&pdev->dev, "Fatal error during devfreq init\n");
> > + goto err_out2;
> > + }
> > +
> > /*
> > * Register the DRM device with the core and the connectors with
> > * sysfs.
> > */
> > err = drm_dev_register(ddev, 0);
> > if (err < 0)
> > - goto err_out2;
> > + goto err_out3;
> >
> > return 0;
> >
> > -err_out2:
> > +err_out3:
> > lima_device_fini(ldev);
> > +err_out2:
> > + lima_devfreq_fini(ldev);
> > err_out1:
> > drm_dev_put(ddev);
> > err_out0:
> > @@ -331,6 +340,7 @@ static int lima_pdev_remove(struct platform_device *pdev)
> > struct drm_device *ddev = ldev->ddev;
> >
> > drm_dev_unregister(ddev);
> > + lima_devfreq_fini(ldev);
> > lima_device_fini(ldev);
> > drm_dev_put(ddev);
> > lima_sched_slab_fini();
> > diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
> > index 3886999b4533..d937d626b892 100644
> > --- a/drivers/gpu/drm/lima/lima_sched.c
> > +++ b/drivers/gpu/drm/lima/lima_sched.c
> > @@ -5,6 +5,7 @@
> > #include <linux/slab.h>
> > #include <linux/xarray.h>
> >
> > +#include "lima_devfreq.h"
> > #include "lima_drv.h"
> > #include "lima_sched.h"
> > #include "lima_vm.h"
> > @@ -214,6 +215,8 @@ static struct dma_fence *lima_sched_run_job(struct drm_sched_job *job)
> > */
> > ret = dma_fence_get(task->fence);
> >
> > + lima_devfreq_record_busy(&pipe->ldev->devfreq);
> > +
> > pipe->current_task = task;
> >
> > /* this is needed for MMU to work correctly, otherwise GP/PP
> > @@ -285,6 +288,8 @@ static void lima_sched_timedout_job(struct drm_sched_job *job)
> > pipe->current_vm = NULL;
> > pipe->current_task = NULL;
> >
> > + lima_devfreq_record_idle(&pipe->ldev->devfreq);
> > +
> > drm_sched_resubmit_jobs(&pipe->base);
> > drm_sched_start(&pipe->base, true);
> > }
> > @@ -364,5 +369,7 @@ void lima_sched_pipe_task_done(struct lima_sched_pipe *pipe)
> > } else {
> > pipe->task_fini(pipe);
> > dma_fence_signal(task->fence);
> > +
> > + lima_devfreq_record_idle(&pipe->ldev->devfreq);
> > }
> > }
> > diff --git a/drivers/gpu/drm/lima/lima_sched.h b/drivers/gpu/drm/lima/lima_sched.h
> > index d64393fb50a9..19bbc5214cf2 100644
> > --- a/drivers/gpu/drm/lima/lima_sched.h
> > +++ b/drivers/gpu/drm/lima/lima_sched.h
> > @@ -6,6 +6,7 @@
> >
> > #include <drm/gpu_scheduler.h>
> >
> > +struct lima_device;
> > struct lima_vm;
> >
> > struct lima_sched_task {
> > @@ -44,6 +45,8 @@ struct lima_sched_pipe {
> > u32 fence_seqno;
> > spinlock_t fence_lock;
> >
> > + struct lima_device *ldev;
> > +
> > struct lima_sched_task *current_task;
> > struct lima_vm *current_vm;
> >
> > --
> > 2.25.2
> >
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support
2020-03-28 8:40 ` Qiang Yu
@ 2020-03-28 10:20 ` Martin Blumenstingl
2020-03-29 9:57 ` Qiang Yu
0 siblings, 1 reply; 8+ messages in thread
From: Martin Blumenstingl @ 2020-03-28 10:20 UTC (permalink / raw)
To: Qiang Yu
Cc: Tomeu Vizoso, linux-pm, David Airlie, Linux Kernel Mailing List,
dri-devel, Steven Price, linux-rockchip, Chen-Yu Tsai,
Alyssa Rosenzweig, open list:ARM/Amlogic Meson...
On Sat, Mar 28, 2020 at 9:40 AM Qiang Yu <yuq825@gmail.com> wrote:
>
> Applied to drm-misc-next.
thank you!
regarding patch #1 - can you apply this as well?
patch #1 just takes this midgard change [0] and ports it to utgard
Thank you!
Martin
[0] https://cgit.freedesktop.org/drm/drm-misc/commit/Documentation/devicetree/bindings/gpu?id=982c0500fd1a8012c31d3c9dd8de285129904656
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support
2020-03-28 10:20 ` Martin Blumenstingl
@ 2020-03-29 9:57 ` Qiang Yu
0 siblings, 0 replies; 8+ messages in thread
From: Qiang Yu @ 2020-03-29 9:57 UTC (permalink / raw)
To: Martin Blumenstingl
Cc: Tomeu Vizoso, linux-pm, David Airlie, Linux Kernel Mailing List,
dri-devel, Steven Price, linux-rockchip, Chen-Yu Tsai,
Alyssa Rosenzweig, open list:ARM/Amlogic Meson...
I'm not the maintainer of patch 1 file, so please contact:
- Rob Herring <robh@kernel.org>
- Maxime Ripard <maxime.ripard@free-electrons.com>
- Heiko Stuebner <heiko@sntech.de>
to review and apply patch 1.
Regards,
Qiang
On Sat, Mar 28, 2020 at 6:20 PM Martin Blumenstingl
<martin.blumenstingl@googlemail.com> wrote:
>
> On Sat, Mar 28, 2020 at 9:40 AM Qiang Yu <yuq825@gmail.com> wrote:
> >
> > Applied to drm-misc-next.
> thank you!
>
> regarding patch #1 - can you apply this as well?
> patch #1 just takes this midgard change [0] and ports it to utgard
>
>
> Thank you!
> Martin
>
>
> [0] https://cgit.freedesktop.org/drm/drm-misc/commit/Documentation/devicetree/bindings/gpu?id=982c0500fd1a8012c31d3c9dd8de285129904656
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2020-03-30 7:12 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-03-19 20:34 [PATCH v4 0/2] drm: lima: devfreq and cooling device support Martin Blumenstingl
2020-03-19 20:34 ` [PATCH v4 1/2] dt-bindings: gpu: mali-utgard: Add the #cooling-cells property Martin Blumenstingl
2020-03-22 2:23 ` Qiang Yu
2020-03-19 20:34 ` [PATCH v4 2/2] drm/lima: Add optional devfreq and cooling device support Martin Blumenstingl
2020-03-22 2:24 ` Qiang Yu
2020-03-28 8:40 ` Qiang Yu
2020-03-28 10:20 ` Martin Blumenstingl
2020-03-29 9:57 ` Qiang Yu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).