drm/vc4: Add a debugfs entry to disable/enable the load tracker
Subsystems (3)
PATCH v2 1/1 drm/vc4: Add a debugfs entry to disable/enable the load tracker
In order to test whether the load tracker is working as expected, we need the ability to compare the commit result with the underrun indication. With the load tracker always enabled, commits that are expected to trigger an underrun are always rejected, so userspace cannot get the actual underrun indication from the hardware.
Add a debugfs entry to disable/enable the load tracker, so that a DRM commit expected to trigger an underrun can go through with the load tracker disabled. The underrun indication is then available to userspace and can be checked against the commit result with the load tracker enabled.
---
Changes since v1:
* Moved all the debugfs-related functions and structure to vc4_debugfs.c.
drivers/gpu/drm/vc4/vc4_debugfs.c | 58 +++++++++++++++++++++++++++++++ drivers/gpu/drm/vc4/vc4_drv.c | 2 ++ drivers/gpu/drm/vc4/vc4_drv.h | 2 ++ drivers/gpu/drm/vc4/vc4_kms.c | 6 +++- 4 files changed, 67 insertions(+), 1 deletion(-)
drivers/gpu/drm/vc4/vc4_debugfs.c +58 -0
@@ -32,9 +32,67 @@ static const struct drm_info_list vc4_debugfs_list[] = {#define VC4_DEBUGFS_ENTRIES ARRAY_SIZE(vc4_debugfs_list) +static int vc4_debugfs_load_tracker_get (struct seq_file *m,void *data) +{ +struct drm_device *dev = m->private; +struct vc4_dev *vc4 =to_vc4_dev (dev); + +if (vc4->load_tracker_enabled) +seq_printf (m,"enabled\n" ); +else +seq_printf (m,"disabled\n" ); + +return 0 ; +} + +static ssize_t vc4_debugfs_load_tracker_set (struct file *file, +const char __user *ubuf, +size_t len,loff_t *offp) +{ +struct seq_file *m = file->private_data; +struct drm_device *dev = m->private; +struct vc4_dev *vc4 =to_vc4_dev (dev); +char buf[32 ] = {}; + +if (len >=sizeof (buf)) +return -EINVAL ; + +if (copy_from_user (buf, ubuf, len)) +return -EFAULT ; + +if (!strncasecmp (buf,"enable" ,6 )) + vc4->load_tracker_enabled = true; +else if (!strncasecmp (buf,"disable" ,7 )) + vc4->load_tracker_enabled = false; +else +return -EINVAL ; + +return len; +} + +static int vc4_debugfs_load_tracker_open (struct inode *inode,struct file *file) +{ +return single_open (file, vc4_debugfs_load_tracker_get, inode->i_private); +} + +static const struct file_operations vc4_debugfs_load_tracker_fops = { + .owner =THIS_MODULE , + .open = vc4_debugfs_load_tracker_open, + .read = seq_read, + .write = vc4_debugfs_load_tracker_set, +}; +int vc4_debugfs_init (struct drm_minor *minor) { +struct dentry *dentry; + + dentry =debugfs_create_file ("load_tracker" ,S_IRUGO |S_IWUSR , + minor->debugfs_root, minor->dev, + &vc4_debugfs_load_tracker_fops); +if (!dentry) +return -ENOMEM ; +return drm_debugfs_create_files (vc4_debugfs_list,VC4_DEBUGFS_ENTRIES , minor->debugfs_root, minor); }
drivers/gpu/drm/vc4/vc4_drv.c +2 -0
@@ -290,6 +290,8 @@ static int vc4_drm_bind(struct device *dev)drm_fbdev_generic_setup (drm,32 ); + vc4->load_tracker_enabled = true; +return 0 ; unbind_all:
drivers/gpu/drm/vc4/vc4_drv.h +2 -0
@@ -188,6 +188,8 @@ struct vc4_dev {int power_refcount; +bool load_tracker_enabled; +/* Mutex controlling the power refcount. */ struct mutex power_lock;
drivers/gpu/drm/vc4/vc4_kms.c +5 -1
@@ -473,6 +473,7 @@ static const struct drm_private_state_funcs vc4_load_tracker_state_funcs = {static int vc4_atomic_check (struct drm_device *dev,struct drm_atomic_state *state) { +struct vc4_dev *vc4 =to_vc4_dev (dev);int ret; ret =vc4_ctm_atomic_check (dev, state);
@@ -483,7 +484,10 @@ vc4_atomic_check(struct drm_device *dev, struct drm_atomic_state *state)if (ret)return ret; -return vc4_load_tracker_atomic_check (state); +if (vc4->load_tracker_enabled) +return vc4_load_tracker_atomic_check (state); + +return 0 ; }static const struct drm_mode_config_funcs vc4_mode_funcs = {
Signature · 6 lines
2.19.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 82 lines · On Fri, 30 Nov 2018 17:11:04 +0100 Paul Kocialkowski…
In order to test whether the load tracker is working as expected, we need the ability to compare the commit result with the underrun indication. With the load tracker always enabled, commits that are expected to trigger an underrun are always rejected, so userspace cannot get the actual underrun indication from the hardware.
Add a debugfs entry to disable/enable the load tracker, so that a DRM commit expected to trigger an underrun can go through with the load tracker disabled. The underrun indication is then available to userspace and can be checked against the commit result with the load tracker enabled.
Signed-off-by Paul Kocialkowski <[email protected]>---
Changes since v1:
* Moved all the debugfs-related functions and structure tovc4_debugfs.c.drivers/gpu/drm/vc4/vc4_debugfs.c | 58 +++++++++++++++++++++++++++++++ drivers/gpu/drm/vc4/vc4_drv.c | 2 ++ drivers/gpu/drm/vc4/vc4_drv.h | 2 ++ drivers/gpu/drm/vc4/vc4_kms.c | 6 +++- 4 files changed, 67 insertions(+), 1 deletion(-)drivers/gpu/drm/vc4/vc4_debugfs.c +50
-0@@ -32,9 +32,67 @@ static const struct drm_info_list vc4_debugfs_list[] = {#define VC4_DEBUGFS_ENTRIES ARRAY_SIZE(vc4_debugfs_list) +static int vc4_debugfs_load_tracker_get (struct seq_file *m,void *data) +{ +struct drm_device *dev = m->private; +struct vc4_dev *vc4 =to_vc4_dev (dev); + +if (vc4->load_tracker_enabled) +seq_printf (m,"enabled\n" ); +else +seq_printf (m,"disabled\n" ); + +return 0 ; +} + +static ssize_t vc4_debugfs_load_tracker_set (struct file *file, +const char __user *ubuf, +size_t len,loff_t *offp) +{ +struct seq_file *m = file->private_data; +struct drm_device *dev = m->private; +struct vc4_dev *vc4 =to_vc4_dev (dev); +char buf[32 ] = {}; + +if (len >=sizeof (buf)) +return -EINVAL ; + +if (copy_from_user (buf, ubuf, len)) +return -EFAULT ; + +if (!strncasecmp (buf,"enable" ,6 )) + vc4->load_tracker_enabled = true; +else if (!strncasecmp (buf,"disable" ,7 )) + vc4->load_tracker_enabled = false; +else +return -EINVAL ; + +return len; +} + +static int vc4_debugfs_load_tracker_open (struct inode *inode,struct file *file) +{ +return single_open (file, vc4_debugfs_load_tracker_get, inode->i_private); +} + +static const struct file_operations vc4_debugfs_load_tracker_fops = { + .owner =THIS_MODULE , + .open = vc4_debugfs_load_tracker_open, + .read = seq_read, + .write = vc4_debugfs_load_tracker_set, +}; +
drivers/gpu/drm/vc4/vc4_debugfs.c static const struct drm_info_list vc4_debugfs_list[] = { · line 94
Quoted 10 lines
drivers/gpu/drm/vc4/vc4_debugfs.c +7
-0int vc4_debugfs_init (struct drm_minor *minor) { +struct dentry *dentry; + + dentry =debugfs_create_file ("load_tracker" ,S_IRUGO |S_IWUSR , + minor->debugfs_root, minor->dev, + &vc4_debugfs_load_tracker_fops); +if (!dentry) +return -ENOMEM ;
Hm, maybe you could use debugfs_create_bool() instead of re-implementing it. The values won't be enabled/disabled though (IIRC, it's Y or N).
Quoted 54 lines
drivers/gpu/drm/vc4/vc4_debugfs.c +1
-0+return drm_debugfs_create_files (vc4_debugfs_list,VC4_DEBUGFS_ENTRIES , minor->debugfs_root, minor); }drivers/gpu/drm/vc4/vc4_drv.c +2
-0@@ -290,6 +290,8 @@ static int vc4_drm_bind(struct device *dev)drm_fbdev_generic_setup (drm,32 ); + vc4->load_tracker_enabled = true; +return 0 ; unbind_all:drivers/gpu/drm/vc4/vc4_drv.h +2
-0@@ -188,6 +188,8 @@ struct vc4_dev {int power_refcount; +bool load_tracker_enabled; +/* Mutex controlling the power refcount. */ struct mutex power_lock;drivers/gpu/drm/vc4/vc4_kms.c +5
-1@@ -473,6 +473,7 @@ static const struct drm_private_state_funcs vc4_load_tracker_state_funcs = {static int vc4_atomic_check (struct drm_device *dev,struct drm_atomic_state *state) { +struct vc4_dev *vc4 =to_vc4_dev (dev);int ret; ret =vc4_ctm_atomic_check (dev, state);@@ -483,7 +484,10 @@ vc4_atomic_check(struct drm_device *dev, struct drm_atomic_state *state)if (ret)return ret; -return vc4_load_tracker_atomic_check (state); +if (vc4->load_tracker_enabled) +return vc4_load_tracker_atomic_check (state); + +return 0 ; }static const struct drm_mode_config_funcs vc4_mode_funcs = {
Hi Boris,
Quoted 91 lines · Boris Brezillon
Quoted 88 lines · On Fri, 30 Nov 2018 17:11:04 +0100 Paul Kocialkowski…
In order to test whether the load tracker is working as expected, we need the ability to compare the commit result with the underrun indication. With the load tracker always enabled, commits that are expected to trigger an underrun are always rejected, so userspace cannot get the actual underrun indication from the hardware.
Add a debugfs entry to disable/enable the load tracker, so that a DRM commit expected to trigger an underrun can go through with the load tracker disabled. The underrun indication is then available to userspace and can be checked against the commit result with the load tracker enabled.
Signed-off-by Paul Kocialkowski <[email protected]>---
Changes since v1:
* Moved all the debugfs-related functions and structure tovc4_debugfs.c.drivers/gpu/drm/vc4/vc4_debugfs.c | 58 +++++++++++++++++++++++++++++++ drivers/gpu/drm/vc4/vc4_drv.c | 2 ++ drivers/gpu/drm/vc4/vc4_drv.h | 2 ++ drivers/gpu/drm/vc4/vc4_kms.c | 6 +++- 4 files changed, 67 insertions(+), 1 deletion(-)drivers/gpu/drm/vc4/vc4_debugfs.c +53
-0@@ -32,9 +32,67 @@ static const struct drm_info_list vc4_debugfs_list[] = {#define VC4_DEBUGFS_ENTRIES ARRAY_SIZE(vc4_debugfs_list) +static int vc4_debugfs_load_tracker_get (struct seq_file *m,void *data) +{ +struct drm_device *dev = m->private; +struct vc4_dev *vc4 =to_vc4_dev (dev); + +if (vc4->load_tracker_enabled) +seq_printf (m,"enabled\n" ); +else +seq_printf (m,"disabled\n" ); + +return 0 ; +} + +static ssize_t vc4_debugfs_load_tracker_set (struct file *file, +const char __user *ubuf, +size_t len,loff_t *offp) +{ +struct seq_file *m = file->private_data; +struct drm_device *dev = m->private; +struct vc4_dev *vc4 =to_vc4_dev (dev); +char buf[32 ] = {}; + +if (len >=sizeof (buf)) +return -EINVAL ; + +if (copy_from_user (buf, ubuf, len)) +return -EFAULT ; + +if (!strncasecmp (buf,"enable" ,6 )) + vc4->load_tracker_enabled = true; +else if (!strncasecmp (buf,"disable" ,7 )) + vc4->load_tracker_enabled = false; +else +return -EINVAL ; + +return len; +} + +static int vc4_debugfs_load_tracker_open (struct inode *inode,struct file *file) +{ +return single_open (file, vc4_debugfs_load_tracker_get, inode->i_private); +} + +static const struct file_operations vc4_debugfs_load_tracker_fops = { + .owner =THIS_MODULE , + .open = vc4_debugfs_load_tracker_open, + .read = seq_read, + .write = vc4_debugfs_load_tracker_set, +}; +int vc4_debugfs_init (struct drm_minor *minor) { +struct dentry *dentry; + + dentry = debugfs_create_file("load_tracker", S_IRUGO | S_IWUSR,
Quoted 8 lines
drivers/gpu/drm/vc4/vc4_debugfs.c static const struct drm_info_list vc4_debugfs_list[] = { · line 94
Quoted 4 lines
drivers/gpu/drm/vc4/vc4_debugfs.c +4
-0+ minor->debugfs_root, minor->dev, + &vc4_debugfs_load_tracker_fops); +if (!dentry) +return -ENOMEM ;Hm, maybe you could use
debugfs_create_bool()instead of re-implementing it. The values won't be enabled/disabled though (IIRC, it's Y or N).
Oh, I didn't know about that! It definitely makes more sense yes.
Thanks,
Paul
Quoted 54 lines
Quoted 54 lines
drivers/gpu/drm/vc4/vc4_debugfs.c +1
-0+return drm_debugfs_create_files (vc4_debugfs_list,VC4_DEBUGFS_ENTRIES , minor->debugfs_root, minor); }drivers/gpu/drm/vc4/vc4_drv.c +2
-0@@ -290,6 +290,8 @@ static int vc4_drm_bind(struct device *dev)drm_fbdev_generic_setup (drm,32 ); + vc4->load_tracker_enabled = true; +return 0 ; unbind_all:drivers/gpu/drm/vc4/vc4_drv.h +2
-0@@ -188,6 +188,8 @@ struct vc4_dev {int power_refcount; +bool load_tracker_enabled; +/* Mutex controlling the power refcount. */ struct mutex power_lock;drivers/gpu/drm/vc4/vc4_kms.c +5
-1@@ -473,6 +473,7 @@ static const struct drm_private_state_funcs vc4_load_tracker_state_funcs = {static int vc4_atomic_check (struct drm_device *dev,struct drm_atomic_state *state) { +struct vc4_dev *vc4 =to_vc4_dev (dev);int ret; ret =vc4_ctm_atomic_check (dev, state);@@ -483,7 +484,10 @@ vc4_atomic_check(struct drm_device *dev, struct drm_atomic_state *state)if (ret)return ret; -return vc4_load_tracker_atomic_check (state); +if (vc4->load_tracker_enabled) +return vc4_load_tracker_atomic_check (state); + +return 0 ; }static const struct drm_mode_config_funcs vc4_mode_funcs = {
Quoted 6 lines · Paul Kocialkowski
In order to test whether the load tracker is working as expected, we need the ability to compare the commit result with the underrun indication. With the load tracker always enabled, commits that are expected to trigger an underrun are always rejected, so userspace cannot get the actual underrun indication from the hardware.
Quoted 7 lines
Add a debugfs entry to disable/enable the load tracker, so that a DRM commit expected to trigger an underrun can go through with the load tracker disabled. The underrun indication is then available to userspace and can be checked against the commit result with the load tracker enabled.
Signed-off-by Paul Kocialkowski <[email protected]>
Given that the load tracker is going to be conservative and say things will underrun even when they might not in practice, will this actually be useful for automated testing? Or is the intent to make it easier to tune the load tracker by disabling it so that you can experiment freely?
Hi Eric,
Quoted 14 lines · Eric Anholt
Quoted 12 lines · Paul Kocialkowski
In order to test whether the load tracker is working as expected, we need the ability to compare the commit result with the underrun indication. With the load tracker always enabled, commits that are expected to trigger an underrun are always rejected, so userspace cannot get the actual underrun indication from the hardware.
Add a debugfs entry to disable/enable the load tracker, so that a DRM commit expected to trigger an underrun can go through with the load tracker disabled. The underrun indication is then available to userspace and can be checked against the commit result with the load tracker enabled.
Quoted 6 lines
Quoted 1 line
Signed-off-by Paul Kocialkowski <[email protected]>Given that the load tracker is going to be conservative and say things will underrun even when they might not in practice, will this actually be useful for automated testing? Or is the intent to make it easier to tune the load tracker by disabling it so that you can experiment freely?
The intent here is to be able to compare the load tracker's result with the underrun indication from the HVS.
The idea is to have the following testing scheme:
1. enable the load tracker
2. try a commit and get the errno if it fails
3. disable the load tracker
4. do the commit
5. wait for vblank and get the underrun indication
6. check that the underrun occurs when errno == -ENOSPC
This way, we can find out when the load tracker gave us a false positive and refine the load calculation.
Cheers,
Paul
Quoted 14 lines · On Fri, 30 Nov 2018 12:30:52 -0800 Eric Anholt <[email protected]> wrote:
Quoted 12 lines · Paul Kocialkowski
In order to test whether the load tracker is working as expected, we need the ability to compare the commit result with the underrun indication. With the load tracker always enabled, commits that are expected to trigger an underrun are always rejected, so userspace cannot get the actual underrun indication from the hardware.
Add a debugfs entry to disable/enable the load tracker, so that a DRM commit expected to trigger an underrun can go through with the load tracker disabled. The underrun indication is then available to userspace and can be checked against the commit result with the load tracker enabled.
Quoted 6 lines
Quoted 1 line
Signed-off-by Paul Kocialkowski <[email protected]>Given that the load tracker is going to be conservative and say things will underrun even when they might not in practice, will this actually be useful for automated testing? Or is the intent to make it easier to tune the load tracker by disabling it so that you can experiment freely?
Yes, that's one goal, though I'm not sure IGT is supposed to contain such debugging tools. But the main benefit is being able to track regressions in the load tracking algo that makes it more (too?) conservative. I think people won't like this sort of regressions. The idea would be to settle on an acceptable load tracking algo (maybe after refining the proposed one), record the results (both good and too conservative predictions) and use that as a reference for the IGT test.
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 24 lines · Boris Brezillon
Quoted 14 lines · On Fri, 30 Nov 2018 12:30:52 -0800 Eric Anholt <[email protected]> wrote:
Quoted 12 lines · Paul Kocialkowski
In order to test whether the load tracker is working as expected, we need the ability to compare the commit result with the underrun indication. With the load tracker always enabled, commits that are expected to trigger an underrun are always rejected, so userspace cannot get the actual underrun indication from the hardware.
Add a debugfs entry to disable/enable the load tracker, so that a DRM commit expected to trigger an underrun can go through with the load tracker disabled. The underrun indication is then available to userspace and can be checked against the commit result with the load tracker enabled.
Quoted 6 lines
Quoted 1 line
Signed-off-by Paul Kocialkowski <[email protected]>Given that the load tracker is going to be conservative and say things will underrun even when they might not in practice, will this actually be useful for automated testing? Or is the intent to make it easier to tune the load tracker by disabling it so that you can experiment freely?
Quoted 8 lines
Yes, that's one goal, though I'm not sure IGT is supposed to contain such debugging tools. But the main benefit is being able to track regressions in the load tracking algo that makes it more (too?) conservative. I think people won't like this sort of regressions. The idea would be to settle on an acceptable load tracking algo (maybe after refining the proposed one), record the results (both good and too conservative predictions) and use that as a reference for the IGT test.
Yeah, I think I'm sold on it at this point -- having a tool that isn't an automated regression test, but an automated thing that can help a developer see how accurate the estimate is, would be useful and is worth a bit of kernel code to support.