Skip to contentSkip to conversation

drm/vkms: Introduce basic support for configfs

PATCH In review Rodrigo Siqueira 3 participants 7 messages Jul 1, 2019 – Jul 16, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover drm/vkms: Introduce basic support for configfs

This patchset introduces the support for configfs in vkms by adding a primary structure for handling the vkms subsystem and exposing connectors as a use case. This series allows enabling/disabling virtual and writeback connectors on the fly. The first patch of this series reworks the initialization and cleanup code of each type of connector, with this change, the second patch adds the configfs support for vkms. It is important to highlight that this patchset depends on
https://patchwork.freedesktop.org/series/61738/.

After applying this series, the user can utilize these features with the following steps:

1. Load vkms without parameter

modprobe vkms

2. Mount a configfs filesystem

mount -t configfs none /mnt/

After that, the vkms subsystem will look like this:

vkms/
|__connectors

    |__Virtual
        |__ enable

The connectors directories have information related to connectors, and as can be seen, the virtual connector is enabled by default. Inside a connector directory (e.g., Virtual) has an attribute named ‘enable’ which is used to enable and disable the target connector. For example, the Virtual connector has the enable attribute set to 1. If the user wants to enable the writeback connector it is required to use the mkdir command, as follows:

cd /mnt/vkms/connectors
mkdir Writeback

After the above command, the writeback connector will be enabled, and the user could see the following tree:

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable

If the user wants to remove the writeback connector, it is required to use the command rmdir, for example

rmdir Writeback

Another way to enable and disable a connector it is by using the enable attribute, for example, we can disable the Virtual connector with:

echo 0 > /mnt/vkms/connectors/Virtual/enable

And enable it again with:

echo 1 > /mnt/vkms/connectors/Virtual/enable

It is important to highlight that configfs 'obey' the parameters used during the vkms load and does not allow users to remove a connector directory if it was load via module parameter. For example:

modprobe vkms enable_writeback=1

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable

If the user tries to remove the Writeback connector with “rmdir Writeback”, the operation will be not permitted because the Writeback connector was loaded with the modules. However, the user may disable the writeback connector with:

echo 0 > /mnt/vkms/connectors/Writeback/enable

Rodrigo Siqueira (2):
drm/vkms: Add enable/disable functions per connector drm/vkms: Introduce configfs for enabling/disabling connectors

 drivers/gpu/drm/vkms/Makefile         |   3 +-
 drivers/gpu/drm/vkms/vkms_configfs.c  | 229 ++++++++++++++++++++++++++
 drivers/gpu/drm/vkms/vkms_drv.c       |   6 +
 drivers/gpu/drm/vkms/vkms_drv.h       |  17 ++
 drivers/gpu/drm/vkms/vkms_output.c    |  84 ++++++----
 drivers/gpu/drm/vkms/vkms_writeback.c |  31 +++-
 6 files changed, 332 insertions(+), 38 deletions(-)
 create mode 100644 drivers/gpu/drm/vkms/vkms_configfs.c
Signature · 10 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH 1/2 drm/vkms: Add enable/disable functions per connector

Currently, virtual and writeback connectors have the code responsible for initialization and cleanup spread around different places in vkms. This patch creates an enable and disable function per connector which allows vkms to hotplug connectors easily.

---
 drivers/gpu/drm/vkms/vkms_drv.h       |  5 ++
 drivers/gpu/drm/vkms/vkms_output.c    | 84 +++++++++++++++++----------
 drivers/gpu/drm/vkms/vkms_writeback.c | 31 ++++++++--
 3 files changed, 83 insertions(+), 37 deletions(-)
drivers/gpu/drm/vkms/vkms_drv.h +5 -0
index 9ff2cd4ebf81..a1ca5c658355 100644
@@ -152,7 +152,12 @@ int vkms_verify_crc_source(struct drm_crtc *crtc, const char *source_name,
 /* Composer Support */
 void vkms_composer_worker(struct work_struct *work);
 
+/* Virtual connector */
+int enable_virtual_connector(struct vkms_device *vkmsdev);
+void disable_virtual_connector(struct vkms_device *vkmsdev);
+
 /* Writeback */
 int enable_writeback_connector(struct vkms_device *vkmsdev);
+void disable_writeback_connector(struct vkms_device *connector);
 
 #endif /* _VKMS_DRV_H_ */
drivers/gpu/drm/vkms/vkms_output.c +52 -32

118 lines not on this page · raw message

drivers/gpu/drm/vkms/vkms_writeback.c +26 -5
index 34dad37a0236..6a3f37d60c1d 100644
@@ -125,17 +125,38 @@ static const struct drm_connector_helper_funcs vkms_wb_conn_helper_funcs = {
 	.atomic_commit = vkms_wb_atomic_commit,
 };
 
+void disable_writeback_connector(struct vkms_device *vkmsdev)
+{
+	struct vkms_output *output = &vkmsdev->output;
+
+	output->composer_enabled = false;
+	drm_connector_unregister(&output->wb_connector.base);
+	drm_connector_cleanup(&output->wb_connector.base);
+	drm_encoder_cleanup(&output->wb_connector.encoder);
+}
+
 int enable_writeback_connector(struct vkms_device *vkmsdev)
 {
 	struct drm_writeback_connector *wb = &vkmsdev->output.wb_connector;
+	int ret;
 
 	vkmsdev->output.wb_connector.encoder.possible_crtcs = 1;
 	drm_connector_helper_add(&wb->base, &vkms_wb_conn_helper_funcs);
 
-	return drm_writeback_connector_init(&vkmsdev->drm, wb,
-					    &vkms_wb_connector_funcs,
-					    &vkms_wb_encoder_helper_funcs,
-					    vkms_wb_formats,
-					    ARRAY_SIZE(vkms_wb_formats));
+	ret = drm_writeback_connector_init(&vkmsdev->drm, wb,
+					   &vkms_wb_connector_funcs,
+					   &vkms_wb_encoder_helper_funcs,
+					   vkms_wb_formats,
+					   ARRAY_SIZE(vkms_wb_formats));
+	if (!ret) {
+		vkmsdev->output.composer_enabled = true;
+		DRM_INFO("Writeback connector enabled");
+	} else {
+		DRM_ERROR("Failed to init writeback connector\n");
+	}
+
+	drm_connector_register(&wb->base);
+
+	return ret;
 }
 
Signature · 5 lines
2.21.0
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#2 lore raw

PATCH 2/2 drm/vkms: Introduce configfs for enabling/disabling connectors

In reply to Rodrigo Siqueira

This patch introduces an implementation of vkms subsystems through configfs; we want to be able to reconfigure vkms instance without having to reload the module. This commit adds the primary support for configfs by allowing vkms to expose the ability to enable/disable its connectors.

---
 drivers/gpu/drm/vkms/Makefile        |   3 +-
 drivers/gpu/drm/vkms/vkms_configfs.c | 229 +++++++++++++++++++++++++++
 drivers/gpu/drm/vkms/vkms_drv.c      |   6 +
 drivers/gpu/drm/vkms/vkms_drv.h      |  12 ++
 4 files changed, 249 insertions(+), 1 deletion(-)
 create mode 100644 drivers/gpu/drm/vkms/vkms_configfs.c
drivers/gpu/drm/vkms/Makefile +2 -1
index 333d3cead0e3..f90c016cd9fe 100644
@@ -6,6 +6,7 @@ vkms-y := \
 	vkms_crtc.o \
 	vkms_gem.o \
 	vkms_composer.o \
-	vkms_writeback.o
+	vkms_writeback.o \
+	vkms_configfs.o
 
 obj-$(CONFIG_DRM_VKMS) += vkms.o
drivers/gpu/drm/vkms/vkms_configfs.c new file +229 -0

229 lines not on this page · raw message

drivers/gpu/drm/vkms/vkms_drv.c +6 -0
index 152d7de24a76..a930a5a52ce4 100644
@@ -166,6 +166,10 @@ static int __init vkms_init(void)
 	if (ret)
 		goto out_fini;
 
+	ret = vkms_configfs_init(vkms_device);
+	if (ret)
+		DRM_ERROR("Could not initialize configfs");
+
 	ret = drm_dev_register(&vkms_device->drm, 0);
 	if (ret)
 		goto out_fini;
@@ -194,6 +198,8 @@ static void __exit vkms_exit(void)
 	drm_dev_put(&vkms_device->drm);
 
 	kfree(vkms_device);
+
+	vkms_configfs_exit();
 }
 
 module_init(vkms_init);
drivers/gpu/drm/vkms/vkms_drv.h +12 -0
index a1ca5c658355..c811bc192606 100644
@@ -9,6 +9,7 @@
 #include <drm/drm_encoder.h>
 #include <drm/drm_writeback.h>
 #include <linux/hrtimer.h>
+#include <linux/configfs.h>
 
 #define XRES_MIN    20
 #define YRES_MIN    20
@@ -83,10 +84,17 @@ struct vkms_output {
 	spinlock_t composer_lock;
 };
 
+#define MAX_CONN_CONFIGFS 10
+
+struct vkms_config_state {
+	struct config_group *connectors[MAX_CONN_CONFIGFS];
+};
+
 struct vkms_device {
 	struct drm_device drm;
 	struct platform_device *platform;
 	struct vkms_output output;
+	struct vkms_config_state config_state;
 };
 
 struct vkms_gem_object {
@@ -160,4 +168,8 @@ void disable_virtual_connector(struct vkms_device *vkmsdev);
 int enable_writeback_connector(struct vkms_device *vkmsdev);
 void disable_writeback_connector(struct vkms_device *connector);
 
+/* Configfs */
+int vkms_configfs_init(struct vkms_device *vkmsdev);
+void vkms_configfs_exit(void);
+
 #endif /* _VKMS_DRV_H_ */
Signature · 1 line
2.21.0
#3 lore raw
In reply to Rodrigo Siqueira
Quoted 75 lines · Rodrigo Siqueira

This patchset introduces the support for configfs in vkms by adding a primary structure for handling the vkms subsystem and exposing connectors as a use case. This series allows enabling/disabling virtual and writeback connectors on the fly. The first patch of this series reworks the initialization and cleanup code of each type of connector, with this change, the second patch adds the configfs support for vkms. It is important to highlight that this patchset depends on
https://patchwork.freedesktop.org/series/61738/.

After applying this series, the user can utilize these features with the following steps:

1. Load vkms without parameter

modprobe vkms

2. Mount a configfs filesystem

mount -t configfs none /mnt/

After that, the vkms subsystem will look like this:

vkms/
|__connectors

    |__Virtual
        |__ enable

The connectors directories have information related to connectors, and as can be seen, the virtual connector is enabled by default. Inside a connector directory (e.g., Virtual) has an attribute named ‘enable’ which is used to enable and disable the target connector. For example, the Virtual connector has the enable attribute set to 1. If the user wants to enable the writeback connector it is required to use the mkdir command, as follows:

cd /mnt/vkms/connectors
mkdir Writeback

After the above command, the writeback connector will be enabled, and the user could see the following tree:

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable

If the user wants to remove the writeback connector, it is required to use the command rmdir, for example

rmdir Writeback

Another way to enable and disable a connector it is by using the enable attribute, for example, we can disable the Virtual connector with:

echo 0 > /mnt/vkms/connectors/Virtual/enable

And enable it again with:

echo 1 > /mnt/vkms/connectors/Virtual/enable

It is important to highlight that configfs 'obey' the parameters used during the vkms load and does not allow users to remove a connector directory if it was load via module parameter. For example:

modprobe vkms enable_writeback=1

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable
Quoted 6 lines

If the user tries to remove the Writeback connector with “rmdir Writeback”, the operation will be not permitted because the Writeback connector was loaded with the modules. However, the user may disable the writeback connector with:

echo 0 > /mnt/vkms/connectors/Writeback/enable

I guess I should have put a warning into that task that step one is designing the interface. Here's the fundamental thoughts:

- The _only_ thing we can hotplug after drm_dev_register() is a drm_connector. That's an interesting use-case, but atm not really supported by the vkms codebase. So we can't just enable/disable writeback like this. We also can't change _anything_ else in the drm driver like this.

- The other bit we want is support multiple vkms instances, to simulate multi-gpus and fun stuff like that.

- Therefore vkms configs should be at the drm_device level, so a directory under configfs/vkms/ represents an entire device.

- We need a special config item to control drm_dev_register/drm_dev_unregister. While a drm_device is registers, all other config items need to fail if userspace tries to change them. Maybe this should be a top-level "enable" property.

- Every time enable goes from 0 -> 1 we need to create a completely new vkms instance. The old one might still be around, waiting for the last few references to disappear.

- enable going from 1 -> 0 needs to be treated like a physical hotunplug, i.e. not drm_dev_unregister but drm_dev_unplug. We also need to annotate all the vkms code with drm_dev_enter/exit() as the kerneldoc of drm_dev_unplug explains.

- rmdir should be treated like enable going from 1 -> 0. Or maybe we should disable enable every going from 1 -> 0, would propably simplify everything.

- The initial instance created at module load also neeeds to be removable like this.

Once we have all this, then can we start to convert driver module options over to configs and add cool features. But lots of infrastructure needed first.

Also, we probably want some nasty testcases which races an rmdir in configfs against userspace still doing ioctl calls against vkms. This is ideal for validation the hotunplug infrastructure we have in drm.

An alternative is adding connector hotplugging. But I think before we do that we need to have support for more crtc and more connectors as static module options. So maybe not a good starting point for configfs.

The above text should probably be added to the vkms.rst todo item ... -Daniel

Quoted 22 lines

Rodrigo Siqueira (2):
drm/vkms: Add enable/disable functions per connector drm/vkms: Introduce configfs for enabling/disabling connectors

 drivers/gpu/drm/vkms/Makefile         |   3 +-
 drivers/gpu/drm/vkms/vkms_configfs.c  | 229 ++++++++++++++++++++++++++
 drivers/gpu/drm/vkms/vkms_drv.c       |   6 +
 drivers/gpu/drm/vkms/vkms_drv.h       |  17 ++
 drivers/gpu/drm/vkms/vkms_output.c    |  84 ++++++----
 drivers/gpu/drm/vkms/vkms_writeback.c |  31 +++-
 6 files changed, 332 insertions(+), 38 deletions(-)
 create mode 100644 drivers/gpu/drm/vkms/vkms_configfs.c
Signature · 6 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
#4 lore raw
Quoted 76 lines · Daniel Vetter
Quoted 75 lines · Rodrigo Siqueira

This patchset introduces the support for configfs in vkms by adding a primary structure for handling the vkms subsystem and exposing connectors as a use case. This series allows enabling/disabling virtual and writeback connectors on the fly. The first patch of this series reworks the initialization and cleanup code of each type of connector, with this change, the second patch adds the configfs support for vkms. It is important to highlight that this patchset depends on
https://patchwork.freedesktop.org/series/61738/.

After applying this series, the user can utilize these features with the following steps:

1. Load vkms without parameter

modprobe vkms

2. Mount a configfs filesystem

mount -t configfs none /mnt/

After that, the vkms subsystem will look like this:

vkms/
|__connectors

    |__Virtual
        |__ enable

The connectors directories have information related to connectors, and as can be seen, the virtual connector is enabled by default. Inside a connector directory (e.g., Virtual) has an attribute named ‘enable’ which is used to enable and disable the target connector. For example, the Virtual connector has the enable attribute set to 1. If the user wants to enable the writeback connector it is required to use the mkdir command, as follows:

cd /mnt/vkms/connectors
mkdir Writeback

After the above command, the writeback connector will be enabled, and the user could see the following tree:

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable

If the user wants to remove the writeback connector, it is required to use the command rmdir, for example

rmdir Writeback

Another way to enable and disable a connector it is by using the enable attribute, for example, we can disable the Virtual connector with:

echo 0 > /mnt/vkms/connectors/Virtual/enable

And enable it again with:

echo 1 > /mnt/vkms/connectors/Virtual/enable

It is important to highlight that configfs 'obey' the parameters used during the vkms load and does not allow users to remove a connector directory if it was load via module parameter. For example:

modprobe vkms enable_writeback=1

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable
Quoted 6 lines
Quoted 6 lines

If the user tries to remove the Writeback connector with “rmdir Writeback”, the operation will be not permitted because the Writeback connector was loaded with the modules. However, the user may disable the writeback connector with:

echo 0 > /mnt/vkms/connectors/Writeback/enable

Thanks for detail this issue, I just have some few questions inline.

Quoted 8 lines

I guess I should have put a warning into that task that step one is designing the interface. Here's the fundamental thoughts:

- The _only_ thing we can hotplug after drm_dev_register() is a drm_connector. That's an interesting use-case, but atm not really supported by the vkms codebase. So we can't just enable/disable writeback like this. We also can't change _anything_ else in the drm driver like this.

In the first patch of this series, I tried to decouple enable/disable for virtual and writeback connectors; I tried to take advantage of drm_connector_[register/unregister] in each connector. Can we use the first patch or it doesn't make sense?

I did not understand why writeback connectors should not be registered and unregister by calling drm_connector_[register/unregister], is it a writeback or vkms limitation? Could you detail why we cannot change connectors as I did?

Additionally, below you said "enable going from 1 -> 0, needs to be treated like a physical hotunplug", do you mean that we first have to add support for drm_dev_plug and drm_dev_unplug in vkms?

Quoted 2 lines

- The other bit we want is support multiple vkms instances, to simulate multi-gpus and fun stuff like that.

Do you mean something like this:

configfs/vkms/instance1
|_enable_device
|_more_stuff
configfs/vkms/instance2
|_enable_device
|_more_stuff
configfs/vkms/instanceN
|_enable_device
|_more_stuff

Will each instance work like a totally independent device? What is the main benefit of this? I can think about some use case for testing prime, but I cannot see other things.

Quoted 28 lines

- Therefore vkms configs should be at the drm_device level, so a directory under configfs/vkms/ represents an entire device.

- We need a special config item to control drm_dev_register/drm_dev_unregister. While a drm_device is registers, all other config items need to fail if userspace tries to change them. Maybe this should be a top-level "enable" property.

- Every time enable goes from 0 -> 1 we need to create a completely new vkms instance. The old one might still be around, waiting for the last few references to disappear.

- enable going from 1 -> 0 needs to be treated like a physical hotunplug, i.e. not drm_dev_unregister but drm_dev_unplug. We also need to annotate all the vkms code with drm_dev_enter/exit() as the kerneldoc of drm_dev_unplug explains.

- rmdir should be treated like enable going from 1 -> 0. Or maybe we should disable enable every going from 1 -> 0, would propably simplify everything.

- The initial instance created at module load also neeeds to be removable like this.

Once we have all this, then can we start to convert driver module options over to configs and add cool features. But lots of infrastructure needed first.

Quoted 7 lines

Also, we probably want some nasty testcases which races an rmdir in configfs against userspace still doing ioctl calls against vkms. This is ideal for validation the hotunplug infrastructure we have in drm.

An alternative is adding connector hotplugging. But I think before we do that we need to have support for more crtc and more connectors as static module options. So maybe not a good starting point for configfs.

So, probably the first set of tasks should be:

1. Enable multiple CRTC via module parameters. For example:
modprobe vkms crtcs=13
2. Enable multiple connectors via module parameters. For example:
modprobe vkms crtcs=3 connector=3 // 3 virtual connectors per crtc

Thanks again,

Quoted 30 lines

The above text should probably be added to the vkms.rst todo item ... -Daniel

Quoted 22 lines

Rodrigo Siqueira (2):
drm/vkms: Add enable/disable functions per connector drm/vkms: Introduce configfs for enabling/disabling connectors

 drivers/gpu/drm/vkms/Makefile         |   3 +-
 drivers/gpu/drm/vkms/vkms_configfs.c  | 229 ++++++++++++++++++++++++++
 drivers/gpu/drm/vkms/vkms_drv.c       |   6 +
 drivers/gpu/drm/vkms/vkms_drv.h       |  17 ++
 drivers/gpu/drm/vkms/vkms_output.c    |  84 ++++++----
 drivers/gpu/drm/vkms/vkms_writeback.c |  31 +++-
 6 files changed, 332 insertions(+), 38 deletions(-)
 create mode 100644 drivers/gpu/drm/vkms/vkms_configfs.c
Signature · 6 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Signature · 2 lines
Rodrigo Siqueira
https://siqueira.tech

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#5 lore raw
Quoted 121 lines · Rodrigo Siqueira
Quoted 94 lines · Daniel Vetter
Quoted 93 lines · Rodrigo Siqueira

This patchset introduces the support for configfs in vkms by adding a
primary structure for handling the vkms subsystem and exposing connectors as a use case. This series allows enabling/disabling virtual
and writeback connectors on the fly. The first patch of this series
reworks the initialization and cleanup code of each type of connector,
with this change, the second patch adds the configfs support for vkms.
It is important to highlight that this patchset depends on
https://patchwork.freedesktop.org/series/61738/.

After applying this series, the user can utilize these features with the
following steps:

1. Load vkms without parameter

modprobe vkms

2. Mount a configfs filesystem

mount -t configfs none /mnt/

After that, the vkms subsystem will look like this:

vkms/
|__connectors

    |__Virtual
        |__ enable

The connectors directories have information related to connectors, and
as can be seen, the virtual connector is enabled by default. Inside a
connector directory (e.g., Virtual) has an attribute named ‘enable’
which is used to enable and disable the target connector. For example,
the Virtual connector has the enable attribute set to 1. If the user
wants to enable the writeback connector it is required to use the mkdir
command, as follows:

cd /mnt/vkms/connectors
mkdir Writeback

After the above command, the writeback connector will be enabled, and
the user could see the following tree:

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable

If the user wants to remove the writeback connector, it is required to
use the command rmdir, for example

rmdir Writeback

Another way to enable and disable a connector it is by using the enable
attribute, for example, we can disable the Virtual connector with:

echo 0 > /mnt/vkms/connectors/Virtual/enable

And enable it again with:

echo 1 > /mnt/vkms/connectors/Virtual/enable

It is important to highlight that configfs 'obey' the parameters used
during the vkms load and does not allow users to remove a connector
directory if it was load via module parameter. For example:

modprobe vkms enable_writeback=1

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable
Quoted 8 lines
Quoted 8 lines

If the user tries to remove the Writeback connector with “rmdir Writeback”, the operation will be not permitted because the Writeback
connector was loaded with the modules. However, the user may disable the
writeback connector with:

echo 0 > /mnt/vkms/connectors/Writeback/enable

Thanks for detail this issue, I just have some few questions inline.

Quoted 9 lines

I guess I should have put a warning into that task that step one is designing the interface. Here's the fundamental thoughts:

- The _only_ thing we can hotplug after drm_dev_register() is a drm_connector. That's an interesting use-case, but atm not really supported by the vkms codebase. So we can't just enable/disable writeback like this. We also can't change _anything_ else in the
drm
driver like this.

In the first patch of this series, I tried to decouple enable/disable for virtual and writeback connectors; I tried to take advantage of drm_connector_[register/unregister] in each connector. Can we use the first patch or it doesn't make sense?

Quoted 6 lines

I did not understand why writeback connectors should not be registered
and unregister by calling drm_connector_[register/unregister], is it a
writeback or vkms limitation? Could you detail why we cannot change connectors as I did?

Hi,

I guess, some more stuff should happen during the hotplug, like drm_kms_helper_hotplug_event()?

Quoted 11 lines

Additionally, below you said "enable going from 1 -> 0, needs to be treated like a physical hotunplug", do you mean that we first have to add support for drm_dev_plug and drm_dev_unplug in vkms?

Quoted 3 lines

- The other bit we want is support multiple vkms instances, to simulate
multi-gpus and fun stuff like that.

Do you mean something like this:

Quoted 9 lines

configfs/vkms/instance1

Quoted 2 lines

_enable_device
_more_stuff

configfs/vkms/instance2

Quoted 2 lines

_enable_device
_more_stuff

configfs/vkms/instanceN

Quoted 2 lines

_enable_device
_more_stuff

I think it would be a good idea. This way the creation of new device could look like this:

mkdir -p instanceN/connector/virtual0
echo "virtual" > instanceN/connector/virtual0/type echo 1 > instanceN/connector/virtual0/enable mkdir -p instanceN/crtc/crtc0
...
echo 1 > instanceN/enable

Once the last command is executed, the whole instanceN/ becomes immutable, except for
- instanceN/enable, so we can later disable it
- instanceN/connector, where we can create new connectors, it would be
treated as a hotplug.

Quoted 5 lines

Will each instance work like a totally independent device? What is the
main benefit of this? I can think about some use case for testing prime, but I cannot see other things.

We can test that userspace always select the right device to display.

Quoted 51 lines
Quoted 45 lines

- Therefore vkms configs should be at the drm_device level, so a directory under configfs/vkms/ represents an entire device.

- We need a special config item to control drm_dev_register/drm_dev_unregister. While a drm_device is
registers,
all other config items need to fail if userspace tries to change
them.
Maybe this should be a top-level "enable" property.

- Every time enable goes from 0 -> 1 we need to create a completely new
vkms instance. The old one might still be around, waiting for the
last
few references to disappear.

- enable going from 1 -> 0 needs to be treated like a physical hotunplug,
i.e. not drm_dev_unregister but drm_dev_unplug. We also need to
annotate
all the vkms code with drm_dev_enter/exit() as the kerneldoc of drm_dev_unplug explains.

- rmdir should be treated like enable going from 1 -> 0. Or maybe we
should disable enable every going from 1 -> 0, would propably
simplify
everything.

- The initial instance created at module load also neeeds to be removable
like this.

Once we have all this, then can we start to convert driver module options
over to configs and add cool features. But lots of infrastructure needed
first.

Also, we probably want some nasty testcases which races an rmdir in configfs against userspace still doing ioctl calls against vkms. This is
ideal for validation the hotunplug infrastructure we have in drm.

Quoted 5 lines

An alternative is adding connector hotplugging. But I think before we do
that we need to have support for more crtc and more connectors as static
module options. So maybe not a good starting point for configfs.

Quoted 6 lines

So, probably the first set of tasks should be:

1. Enable multiple CRTC via module parameters. For example:
modprobe vkms crtcs=13
2. Enable multiple connectors via module parameters. For example:
modprobe vkms crtcs=3 connector=3 // 3 virtual connectors per crtc

But do we want to have those parameters as module options, if we then plan to replace those with configfs?

Quoted 39 lines

Thanks again,

Quoted 19 lines

The above text should probably be added to the vkms.rst todo item
...
-Daniel

Quoted 15 lines

Rodrigo Siqueira (2):
drm/vkms: Add enable/disable functions per connector drm/vkms: Introduce configfs for enabling/disabling connectors

 drivers/gpu/drm/vkms/Makefile         |   3 +-
 drivers/gpu/drm/vkms/vkms_configfs.c  | 229
++++++++++++++++++++++++++
 drivers/gpu/drm/vkms/vkms_drv.c       |   6 +
 drivers/gpu/drm/vkms/vkms_drv.h       |  17 ++
 drivers/gpu/drm/vkms/vkms_output.c    |  84 ++++++----
 drivers/gpu/drm/vkms/vkms_writeback.c |  31 +++-
 6 files changed, 332 insertions(+), 38 deletions(-)
 create mode 100644 drivers/gpu/drm/vkms/vkms_configfs.c
Quoted 12 lines
Quoted 7 lines
Signature · 6 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Signature · 2 lines
Oleg Vasilev <[email protected]>
Intel Corporation

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#6 lore raw
Quoted 117 lines · Oleg
Quoted 116 lines · Rodrigo Siqueira
Quoted 94 lines · Daniel Vetter
Quoted 93 lines · Rodrigo Siqueira

This patchset introduces the support for configfs in vkms by adding a
primary structure for handling the vkms subsystem and exposing connectors as a use case. This series allows enabling/disabling virtual
and writeback connectors on the fly. The first patch of this series
reworks the initialization and cleanup code of each type of connector,
with this change, the second patch adds the configfs support for vkms.
It is important to highlight that this patchset depends on
https://patchwork.freedesktop.org/series/61738/.

After applying this series, the user can utilize these features with the
following steps:

1. Load vkms without parameter

modprobe vkms

2. Mount a configfs filesystem

mount -t configfs none /mnt/

After that, the vkms subsystem will look like this:

vkms/
|__connectors

    |__Virtual
        |__ enable

The connectors directories have information related to connectors, and
as can be seen, the virtual connector is enabled by default. Inside a
connector directory (e.g., Virtual) has an attribute named ‘enable’
which is used to enable and disable the target connector. For example,
the Virtual connector has the enable attribute set to 1. If the user
wants to enable the writeback connector it is required to use the mkdir
command, as follows:

cd /mnt/vkms/connectors
mkdir Writeback

After the above command, the writeback connector will be enabled, and
the user could see the following tree:

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable

If the user wants to remove the writeback connector, it is required to
use the command rmdir, for example

rmdir Writeback

Another way to enable and disable a connector it is by using the enable
attribute, for example, we can disable the Virtual connector with:

echo 0 > /mnt/vkms/connectors/Virtual/enable

And enable it again with:

echo 1 > /mnt/vkms/connectors/Virtual/enable

It is important to highlight that configfs 'obey' the parameters used
during the vkms load and does not allow users to remove a connector
directory if it was load via module parameter. For example:

modprobe vkms enable_writeback=1

vkms/
|__connectors

    |__Virtual
    |   |__ enable
    |__Writeback
        |__ enable
Quoted 8 lines
Quoted 8 lines

If the user tries to remove the Writeback connector with “rmdir Writeback”, the operation will be not permitted because the Writeback
connector was loaded with the modules. However, the user may disable the
writeback connector with:

echo 0 > /mnt/vkms/connectors/Writeback/enable

Thanks for detail this issue, I just have some few questions inline.

Quoted 9 lines

I guess I should have put a warning into that task that step one is designing the interface. Here's the fundamental thoughts:

- The _only_ thing we can hotplug after drm_dev_register() is a drm_connector. That's an interesting use-case, but atm not really supported by the vkms codebase. So we can't just enable/disable writeback like this. We also can't change _anything_ else in the
drm
driver like this.

Quoted 4 lines
Quoted 4 lines

In the first patch of this series, I tried to decouple enable/disable for virtual and writeback connectors; I tried to take advantage of drm_connector_[register/unregister] in each connector. Can we use the first patch or it doesn't make sense?

No, because you also add the drm_encoder. That cannot be hotadded/removed right now. So you'd need to add a drm_encoder preemptively for all writeback connectors you might need, which doesn't make much sense.

Quoted 11 lines
Quoted 6 lines

I did not understand why writeback connectors should not be registered
and unregister by calling drm_connector_[register/unregister], is it a
writeback or vkms limitation? Could you detail why we cannot change connectors as I did?

Hi,

I guess, some more stuff should happen during the hotplug, like drm_kms_helper_hotplug_event()?

writeback is generally a SoC feature. I don't think anyone expects that you can hotplug a writeback connector.

Quoted 19 lines
Quoted 19 lines

Additionally, below you said "enable going from 1 -> 0, needs to be treated like a physical hotunplug", do you mean that we first have to add support for drm_dev_plug and drm_dev_unplug in vkms?

Quoted 3 lines

- The other bit we want is support multiple vkms instances, to simulate
multi-gpus and fun stuff like that.

Do you mean something like this:

configfs/vkms/instance1

Quoted 2 lines

_enable_device
_more_stuff

configfs/vkms/instance2

Quoted 2 lines

_enable_device
_more_stuff

configfs/vkms/instanceN

Quoted 8 lines
Quoted 2 lines
Quoted 2 lines

_enable_device
_more_stuff

I think it would be a good idea. This way the creation of new device could look like this:

mkdir -p instanceN/connector/virtual0
echo "virtual" > instanceN/connector/virtual0/type

virtual should be a top-level property. I'm not aware of any real driver where vblank works on some connector/crtc, but not on another one.

Quoted 1 line

echo 1 > instanceN/connector/virtual0/enable

For the initial connectors those don't need separate enable properties, they get enabled with the global enable knob.

Quoted 9 lines

mkdir -p instanceN/crtc/crtc0
...
echo 1 > instanceN/enable

Once the last command is executed, the whole instanceN/ becomes immutable, except for
- instanceN/enable, so we can later disable it
- instanceN/connector, where we can create new connectors, it would be
treated as a hotplug.

For hotplugged connectors we need a separate enable knob. But we're still _very_ far away from making that possible I think.

Quoted 6 lines
Quoted 4 lines

Will each instance work like a totally independent device? What is the
main benefit of this? I can think about some use case for testing prime, but I cannot see other things.

We can test that userspace always select the right device to display.

Also hotunplug behaviour. We can take out an entire drm_device, and make sure userspace and the kernel copes correctly. This is _really_ hard to get right on both sides.

And it's a real thing with stuff like usb display link.

Quoted 52 lines
Quoted 52 lines
Quoted 44 lines

- Therefore vkms configs should be at the drm_device level, so a directory under configfs/vkms/ represents an entire device.

- We need a special config item to control drm_dev_register/drm_dev_unregister. While a drm_device is
registers,
all other config items need to fail if userspace tries to change
them.
Maybe this should be a top-level "enable" property.

- Every time enable goes from 0 -> 1 we need to create a completely new
vkms instance. The old one might still be around, waiting for the
last
few references to disappear.

- enable going from 1 -> 0 needs to be treated like a physical hotunplug,
i.e. not drm_dev_unregister but drm_dev_unplug. We also need to
annotate
all the vkms code with drm_dev_enter/exit() as the kerneldoc of drm_dev_unplug explains.

- rmdir should be treated like enable going from 1 -> 0. Or maybe we
should disable enable every going from 1 -> 0, would propably
simplify
everything.

- The initial instance created at module load also neeeds to be removable
like this.

Once we have all this, then can we start to convert driver module options
over to configs and add cool features. But lots of infrastructure needed
first.

Also, we probably want some nasty testcases which races an rmdir in configfs against userspace still doing ioctl calls against vkms. This is
ideal for validation the hotunplug infrastructure we have in drm.

Quoted 5 lines

An alternative is adding connector hotplugging. But I think before we do
that we need to have support for more crtc and more connectors as static
module options. So maybe not a good starting point for configfs.

So, probably the first set of tasks should be:

Quoted 7 lines
Quoted 4 lines

1. Enable multiple CRTC via module parameters. For example:
modprobe vkms crtcs=13
2. Enable multiple connectors via module parameters. For example:
modprobe vkms crtcs=3 connector=3 // 3 virtual connectors per crtc

But do we want to have those parameters as module options, if we then plan to replace those with configfs?

Imo we should phase out module options, or at least not add more. Then you can just have a small script which sets up the vkms device you want, and then run the tests. I'd go further even, and say we should have default_device=0 knob to start out with nothing when you load the driver. -Daniel

Quoted 50 lines
Quoted 40 lines

Thanks again,

Quoted 19 lines

The above text should probably be added to the vkms.rst todo item
...
-Daniel

Quoted 15 lines

Rodrigo Siqueira (2):
drm/vkms: Add enable/disable functions per connector drm/vkms: Introduce configfs for enabling/disabling connectors

 drivers/gpu/drm/vkms/Makefile         |   3 +-
 drivers/gpu/drm/vkms/vkms_configfs.c  | 229
++++++++++++++++++++++++++
 drivers/gpu/drm/vkms/vkms_drv.c       |   6 +
 drivers/gpu/drm/vkms/vkms_drv.h       |  17 ++
 drivers/gpu/drm/vkms/vkms_output.c    |  84 ++++++----
 drivers/gpu/drm/vkms/vkms_writeback.c |  31 +++-
 6 files changed, 332 insertions(+), 38 deletions(-)
 create mode 100644 drivers/gpu/drm/vkms/vkms_configfs.c
Quoted 12 lines
Quoted 7 lines
Signature · 6 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Signature · 9 lines
Oleg Vasilev <[email protected]>
Intel Corporation



_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch