| 2018-07-20 03:38:36 |
Hui Wang |
bug |
|
|
added bug |
| 2018-07-20 03:39:30 |
Hui Wang |
bug |
|
|
added subscriber Canonical Hardware Enablement |
| 2018-07-20 03:39:37 |
Hui Wang |
tags |
|
originate-from-1775331 somerville |
|
| 2018-07-20 03:40:12 |
Hui Wang |
linux (Ubuntu): importance |
Undecided |
Critical |
|
| 2018-07-20 04:00:07 |
Ubuntu Kernel Bot |
linux (Ubuntu): status |
New |
Incomplete |
|
| 2018-07-20 06:40:39 |
Hui Wang |
description |
Steps:
1. Install manifest and boot into OS
2. Connect an external monitor via HDMI port
3. Check sound setting
Expected results: System should able to detect HDMI audio output
Actual results: System can't detect HDMI audio output
Failure rate: 2/2 |
This bug comes from the OEM project, and the engineer Jim Qu from AMD took
almost 2 weeks to investigate this problem, then he worte these two patches to
fix this problem and now the patches are merged to sound repository.
According to Jim Qu's investigation, the root cause of this problem is:
1. there is two GPU on the system. iGPU has a audio codec for HDMI output,
and dGPU is without audio codec it is only for offload rendering.
2. under runtime pm, when dGPU suspend, amdgpu will also call vgaswitchroo
driver to power off audio which vgaswitchroo client is VGA_SWITCHEROO_DIS.
3. In current HDA audio driver, the iGPU's audio will register to vgaswitchroo
as VGA_SWITCHEROO_DIS, therefore, the audio will be selected and powered off
by runtime pm.
[Impact]
On the Dell Latitude 5495, If we plug a monitor with audio capability to the
HDMI/DP ports, we can't find HDMI/DP audio sinks as expected, then we can't
play any sound through HDMI/DP audio.
[Fix]
With these two patches, the driver will not always set vgaswicheroo clients of
HDA audio as VGA_SWITCHEROO_DIS, it will set it to _DIS or _IGD with the help
of callback function of DRM drivers. So on this machine, the vgaswicheroo client
will be set _IGD, then it will not be powered off when discrete gpu is powered off.
[Test Case]
We tested plug/unplug detection and playback through HDMI/DP audio, everything works
well.
[Regression Potential]
Very low, without these two patches, the vgaswitchroo client of audio will be
set to _DIS unconditionally, it did not expose any problem because in the past,
all the HDMI/DP audio codecs are in the discreate GPU. But on Latitude 5495, the
HDMI/DP audio codec is in the integrated GPU, so we need to change the driver to
let DRM driver decide if it is _DIS or _IGD, it will not introduce regression for
old mahcines on old machines, the client will be set to _DIS as before.
And we have tested these two patches on some old machines with two gpus like A+A
, I+A and I+N, all of them worked well as before. |
|
| 2018-07-30 10:37:36 |
Stefan Bader |
bug task added |
|
linux-oem (Ubuntu) |
|
| 2018-07-30 10:37:45 |
Stefan Bader |
nominated for series |
|
Ubuntu Bionic |
|
| 2018-07-30 10:37:45 |
Stefan Bader |
bug task added |
|
linux (Ubuntu Bionic) |
|
| 2018-07-30 10:37:45 |
Stefan Bader |
bug task added |
|
linux-oem (Ubuntu Bionic) |
|
| 2018-07-31 00:18:44 |
Hui Wang |
description |
This bug comes from the OEM project, and the engineer Jim Qu from AMD took
almost 2 weeks to investigate this problem, then he worte these two patches to
fix this problem and now the patches are merged to sound repository.
According to Jim Qu's investigation, the root cause of this problem is:
1. there is two GPU on the system. iGPU has a audio codec for HDMI output,
and dGPU is without audio codec it is only for offload rendering.
2. under runtime pm, when dGPU suspend, amdgpu will also call vgaswitchroo
driver to power off audio which vgaswitchroo client is VGA_SWITCHEROO_DIS.
3. In current HDA audio driver, the iGPU's audio will register to vgaswitchroo
as VGA_SWITCHEROO_DIS, therefore, the audio will be selected and powered off
by runtime pm.
[Impact]
On the Dell Latitude 5495, If we plug a monitor with audio capability to the
HDMI/DP ports, we can't find HDMI/DP audio sinks as expected, then we can't
play any sound through HDMI/DP audio.
[Fix]
With these two patches, the driver will not always set vgaswicheroo clients of
HDA audio as VGA_SWITCHEROO_DIS, it will set it to _DIS or _IGD with the help
of callback function of DRM drivers. So on this machine, the vgaswicheroo client
will be set _IGD, then it will not be powered off when discrete gpu is powered off.
[Test Case]
We tested plug/unplug detection and playback through HDMI/DP audio, everything works
well.
[Regression Potential]
Very low, without these two patches, the vgaswitchroo client of audio will be
set to _DIS unconditionally, it did not expose any problem because in the past,
all the HDMI/DP audio codecs are in the discreate GPU. But on Latitude 5495, the
HDMI/DP audio codec is in the integrated GPU, so we need to change the driver to
let DRM driver decide if it is _DIS or _IGD, it will not introduce regression for
old mahcines on old machines, the client will be set to _DIS as before.
And we have tested these two patches on some old machines with two gpus like A+A
, I+A and I+N, all of them worked well as before. |
In the v2:
the patchset is also sent to OEM-B since the OEM project is waiting for the patchset.
The v1 introduced a building error, my investigation is shown as below:
The declaration of "enum vga_switcheroo_client_id (*get_client_id)(struct pci_dev *pdev);"
and the implementation of "static int nouveau_dsm_get_client_id(struct pci_dev *pdev)"
exist in the kernel (bionic, artful) for a long time, but "-Werror=incompatible-pointer-types"
didn't report it as an error, after applied these 2 patches, it is reported as an error,
that is because the patch touches the definition of "enum vga_switcheroo_client_id",
before applying these 2 patches, the vga_switcheroo_client_id includes a negative numer:
VGA_SWITCHEROO_UNKNOWN_ID = -1, but after applying the patches, there is no negative number
anymore in the "enum vga_swicheroo_client_id".
I guess if the definition of enum includes negative number, the c compiler treat it as int type,
that is why artful and bionic did not expose building error before.
To fix the building error, I cherry-picked 4 more patches [3/6-6/6], and the purpose of those
4 patches is straightforward.
In the v1:
This bug comes from the OEM project, and the engineer Jim Qu from AMD took
almost 2 weeks to investigate this problem, then he worte these two patches to
fix this problem and now the patches are merged to sound repository.
According to Jim Qu's investigation, the root cause of this problem is:
1. there is two GPU on the system. iGPU has a audio codec for HDMI output,
and dGPU is without audio codec it is only for offload rendering.
2. under runtime pm, when dGPU suspend, amdgpu will also call vgaswitchroo
driver to power off audio which vgaswitchroo client is VGA_SWITCHEROO_DIS.
3. In current HDA audio driver, the iGPU's audio will register to vgaswitchroo
as VGA_SWITCHEROO_DIS, therefore, the audio will be selected and powered off
by runtime pm.
[Impact]
On the Dell Latitude 5495, If we plug a monitor with audio capability to the
HDMI/DP ports, we can't find HDMI/DP audio sinks as expected, then we can't
play any sound through HDMI/DP audio.
[Fix]
With these 6 patches, the driver will not always set vgaswicheroo clients of
HDA audio as VGA_SWITCHEROO_DIS, it will set it to _DIS or _IGD with the help
of callback function of DRM drivers. So on this machine, the vgaswicheroo client
will be set _IGD, then it will not be powered off when discrete gpu is powered off.
[Test Case]
We tested plug/unplug detection and playback through HDMI/DP audio, everything works
well.
[Regression Potential]
Very low, without these 6 patches, the vgaswitchroo client of audio will be
set to _DIS unconditionally, it did not expose any problem because in the past,
all the HDMI/DP audio codecs are in the discreate GPU. But on Latitude 5495, the
HDMI/DP audio codec is in the integrated GPU, so we need to change the driver to
let DRM driver decide if it is _DIS or _IGD, it will not introduce regression for
old mahcines on old machines, the client will be set to _DIS as before.
And we have tested these two patches on some old machines with two gpus like A+A
, I+A and I+N, all of them worked well as before. |
|
| 2018-07-31 00:25:05 |
Hui Wang |
description |
In the v2:
the patchset is also sent to OEM-B since the OEM project is waiting for the patchset.
The v1 introduced a building error, my investigation is shown as below:
The declaration of "enum vga_switcheroo_client_id (*get_client_id)(struct pci_dev *pdev);"
and the implementation of "static int nouveau_dsm_get_client_id(struct pci_dev *pdev)"
exist in the kernel (bionic, artful) for a long time, but "-Werror=incompatible-pointer-types"
didn't report it as an error, after applied these 2 patches, it is reported as an error,
that is because the patch touches the definition of "enum vga_switcheroo_client_id",
before applying these 2 patches, the vga_switcheroo_client_id includes a negative numer:
VGA_SWITCHEROO_UNKNOWN_ID = -1, but after applying the patches, there is no negative number
anymore in the "enum vga_swicheroo_client_id".
I guess if the definition of enum includes negative number, the c compiler treat it as int type,
that is why artful and bionic did not expose building error before.
To fix the building error, I cherry-picked 4 more patches [3/6-6/6], and the purpose of those
4 patches is straightforward.
In the v1:
This bug comes from the OEM project, and the engineer Jim Qu from AMD took
almost 2 weeks to investigate this problem, then he worte these two patches to
fix this problem and now the patches are merged to sound repository.
According to Jim Qu's investigation, the root cause of this problem is:
1. there is two GPU on the system. iGPU has a audio codec for HDMI output,
and dGPU is without audio codec it is only for offload rendering.
2. under runtime pm, when dGPU suspend, amdgpu will also call vgaswitchroo
driver to power off audio which vgaswitchroo client is VGA_SWITCHEROO_DIS.
3. In current HDA audio driver, the iGPU's audio will register to vgaswitchroo
as VGA_SWITCHEROO_DIS, therefore, the audio will be selected and powered off
by runtime pm.
[Impact]
On the Dell Latitude 5495, If we plug a monitor with audio capability to the
HDMI/DP ports, we can't find HDMI/DP audio sinks as expected, then we can't
play any sound through HDMI/DP audio.
[Fix]
With these 6 patches, the driver will not always set vgaswicheroo clients of
HDA audio as VGA_SWITCHEROO_DIS, it will set it to _DIS or _IGD with the help
of callback function of DRM drivers. So on this machine, the vgaswicheroo client
will be set _IGD, then it will not be powered off when discrete gpu is powered off.
[Test Case]
We tested plug/unplug detection and playback through HDMI/DP audio, everything works
well.
[Regression Potential]
Very low, without these 6 patches, the vgaswitchroo client of audio will be
set to _DIS unconditionally, it did not expose any problem because in the past,
all the HDMI/DP audio codecs are in the discreate GPU. But on Latitude 5495, the
HDMI/DP audio codec is in the integrated GPU, so we need to change the driver to
let DRM driver decide if it is _DIS or _IGD, it will not introduce regression for
old mahcines on old machines, the client will be set to _DIS as before.
And we have tested these two patches on some old machines with two gpus like A+A
, I+A and I+N, all of them worked well as before. |
In the v2:
the patchset is also sent to OEM-B since the OEM project is waiting for the patchset.
The v1 introduced a building error, my investigation is shown as below:
The declaration of "enum vga_switcheroo_client_id (*get_client_id)(struct pci_dev *pdev);"and the implementation of "static int nouveau_dsm_get_client_id(struct pci_dev *pdev)"
exist in the kernel (bionic, artful) for a long time, but "-Werror=incompatible-pointer-types" didn't report it as an error, after applied these 2 patches, it is reported as an error,
that is because the patch touches the definition of "enum vga_switcheroo_client_id", before applying these 2 patches, the vga_switcheroo_client_id includes a negative numer:VGA_SWITCHEROO_UNKNOWN_ID = -1, but after applying the patches, there is no negative number anymore in the "enum vga_swicheroo_client_id".
I guess if the definition of enum includes negative number, the c compiler treat it as int type, that is why artful and bionic did not expose building error before.
To fix the building error, I cherry-picked 4 more patches [3/6-6/6], and the purpose of those 4 patches is straightforward.
In the v1:
This bug comes from the OEM project, and the engineer Jim Qu from AMD took almost 2 weeks to investigate this problem, then he worte two patches to fix this problem and now the patches are merged to sound repository.
According to Jim Qu's investigation, the root cause of this problem is:
1. there is two GPU on the system. iGPU has a audio codec for HDMI output, and dGPU is without audio codec it is only for offload rendering.
2. under runtime pm, when dGPU suspend, amdgpu will also call vgaswitchroo driver to power off audio which vgaswitchroo client is VGA_SWITCHEROO_DIS.
3. In current HDA audio driver, the iGPU's audio will register to vgaswitchroo as VGA_SWITCHEROO_DIS, therefore, the audio will be selected and powered off by runtime pm.
[Impact]
On the Dell Latitude 5495, If we plug a monitor with audio capability to the HDMI/DP ports, we can't find HDMI/DP audio sinks as expected, then we can't play any sound through HDMI/DP audio.
[Fix]
With these 6 patches, the driver will not always set vgaswicheroo clients of HDA audio as VGA_SWITCHEROO_DIS, it will set it to _DIS or _IGD with the help of callback function of DRM drivers. So on this machine, the vgaswicheroo client will be set _IGD, then it will not be powered off when discrete gpu is powered off.
[Test Case]
We tested plug/unplug detection and playback through HDMI/DP audio, everything works well.
[Regression Potential]
Very low, without these 6 patches, the vgaswitchroo client of audio will be set to _DIS unconditionally, it did not expose any problem because in the past, all the HDMI/DP audio codecs are in the discreate GPU. But on Latitude 5495, the HDMI/DP audio codec is in the integrated GPU, so we need to change the driver to let DRM driver decide if it is _DIS or _IGD, it will not introduce regression for
old mahcines on old machines, the client will be set to _DIS as before.
And we have tested these two patches on some old machines with two gpus like A+A, I+A and I+N, all of them worked well as before. |
|
| 2018-07-31 13:58:24 |
Kleber Sacilotto de Souza |
linux (Ubuntu Bionic): status |
New |
Fix Committed |
|
| 2018-08-01 13:09:49 |
Seth Forshee |
linux (Ubuntu): status |
Incomplete |
Fix Committed |
|
| 2018-08-07 14:01:44 |
Brad Figg |
tags |
originate-from-1775331 somerville |
originate-from-1775331 somerville verification-needed-bionic |
|
| 2018-08-16 13:27:56 |
Launchpad Janitor |
linux (Ubuntu): status |
Fix Committed |
Fix Released |
|
| 2018-08-16 13:27:56 |
Launchpad Janitor |
cve linked |
|
2018-12233 |
|
| 2018-08-16 13:27:56 |
Launchpad Janitor |
cve linked |
|
2018-13094 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
linux (Ubuntu Bionic): status |
Fix Committed |
Fix Released |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-10323 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-10840 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-10881 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-1108 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-11412 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-11506 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-12232 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-12904 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-13405 |
|
| 2018-08-23 13:23:09 |
Launchpad Janitor |
cve linked |
|
2018-13406 |
|
| 2018-08-27 07:00:56 |
Launchpad Janitor |
linux-oem (Ubuntu Bionic): status |
New |
Fix Released |
|
| 2018-08-29 17:30:18 |
Anthony Wong |
linux-oem (Ubuntu): status |
New |
Invalid |
|
| 2018-08-29 17:30:21 |
Anthony Wong |
hwe-next: status |
New |
Fix Released |
|
| 2018-09-10 11:05:46 |
Launchpad Janitor |
linux-oem (Ubuntu): status |
Invalid |
Fix Released |
|
| 2018-09-14 00:38:02 |
Hui Wang |
tags |
originate-from-1775331 somerville verification-needed-bionic |
originate-from-1775331 somerville verification-done-bionic |
|
| 2019-07-24 20:18:38 |
Brad Figg |
tags |
originate-from-1775331 somerville verification-done-bionic |
cscc originate-from-1775331 somerville verification-done-bionic |
|
