Skip to contentSkip to conversation

ACPI: scan: Honor _DEP for Intel CVS devices

PATCH Applied Miguel Vadillo 4 participants 7 messages Jun 1 – Sep 13

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

PATCH 1/1 ACPI: scan: Honor _DEP for Intel CVS devices

CVS (Computer Vision Sensing) is an ACPI-enumerated device that sits inline in the CSI-2 path between the camera sensor and Intel IPU. On platforms where CVS is present, the camera sensor's ACPI node declares a _DEP dependency on the CVS device.

The CVS driver must be fully initialized before camera sensor drivers probe, because CVS controls the CSI-2 link ownership handshake (via GPIO REQ/RESP), the MIPI/CSI-2 lane configuration, and the camera power domain. Without CVS ready, the sensor driver can bind but the CSI-2 stream will not function correctly.

The CVS driver calls acpi_dev_clear_dependencies() at the end of its probe() to unblock waiting consumers once it is ready.

Move the CVS HIDs from acpi_ignore_dep_ids[] to acpi_honor_dep_ids[] so that camera sensor enumeration is deferred until the CVS driver has finished probing, matching the behavior already in place for IVSC.

---
Note: this patch has a functional dependency on the CVS driver and its IPU bridge support, which are queued in media-next for v7.2. Without those drivers, no code will call acpi_dev_clear_dependencies() for the CVS HIDs and camera sensors will remain deferred indefinitely.

The CVS driver series (v5) can be found at:
https://lore.kernel.org/linux-media/[email protected]/

 drivers/acpi/scan.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/acpi/scan.c +3 -2
index 530547cda8b2..1463eb3a4f6e 100644
@@ -848,8 +848,6 @@ static bool acpi_info_matches_ids(struct acpi_device_info *info,
 static const char * const acpi_ignore_dep_ids[] = {
 	"PNP0D80", /* Windows-compatible System Power Management Controller */
 	"INT33BD", /* Intel Baytrail Mailbox Device */
-	"INTC10DE", /* Intel CVS LNL */
-	"INTC10E0", /* Intel CVS ARL */
 	"LATT2021", /* Lattice FW Update Client Driver */
 	NULL
 };
@@ -861,6 +859,9 @@ static const char * const acpi_honor_dep_ids[] = {
 	"INTC1095", /* IVSC (ADL) driver must be loaded to allow i2c access to camera sensors */
 	"INTC100A", /* IVSC (RPL) driver must be loaded to allow i2c access to camera sensors */
 	"INTC10CF", /* IVSC (MTL) driver must be loaded to allow i2c access to camera sensors */
+	"INTC10DE", /* CVS (LNL) driver must be loaded to allow camera streaming */
+	"INTC10E0", /* CVS (ARL) driver must be loaded to allow camera streaming */
+	"INTC10E1", /* CVS (PTL) driver must be loaded to allow camera streaming */
 	"RSCV0001", /* RISC-V PLIC */
 	"RSCV0002", /* RISC-V APLIC */
 	"RSCV0005", /* RISC-V SBI MPXY MBOX */
Signature · 1 line
2.43.0
Reviewed-by on v1 1/1
#1 lore raw

Hi Miguel, Rafael,

Thanks for the patch.

Quoted 11 lines · Miguel Vadillo

CVS (Computer Vision Sensing) is an ACPI-enumerated device that sits inline in the CSI-2 path between the camera sensor and Intel IPU. On platforms where CVS is present, the camera sensor's ACPI node declares a _DEP dependency on the CVS device.

The CVS driver must be fully initialized before camera sensor drivers probe, because CVS controls the CSI-2 link ownership handshake (via GPIO REQ/RESP), the MIPI/CSI-2 lane configuration, and the camera power domain. Without CVS ready, the sensor driver can bind but the CSI-2 stream will not function correctly.

Quoted 8 lines

The CVS driver calls acpi_dev_clear_dependencies() at the end of its probe() to unblock waiting consumers once it is ready.

Move the CVS HIDs from acpi_ignore_dep_ids[] to acpi_honor_dep_ids[] so that camera sensor enumeration is deferred until the CVS driver has finished probing, matching the behavior already in place for IVSC.

Rafael: do you think this could still make it to 7.2? Otherwise cameras might not work on a some laptops where they did use to.

The default configuration is appropriate in some laptops so camera used to work without the CVS driver but now that we have the CVS driver merged and the CVS device is part of the software node / ACPI graph, probing the camera sensor driver will in fact not take place at all (and obviously the camera won't work in this case).

Signature · 3 lines
Kind regards,

Sakari Ailus
maintainer Applied
#2 lore raw
Quoted 28 lines · Sakari Ailus

Hi Miguel, Rafael,

Thanks for the patch.

Quoted 11 lines · Miguel Vadillo

CVS (Computer Vision Sensing) is an ACPI-enumerated device that sits inline in the CSI-2 path between the camera sensor and Intel IPU. On platforms where CVS is present, the camera sensor's ACPI node declares a _DEP dependency on the CVS device.

The CVS driver must be fully initialized before camera sensor drivers probe, because CVS controls the CSI-2 link ownership handshake (via GPIO REQ/RESP), the MIPI/CSI-2 lane configuration, and the camera power domain. Without CVS ready, the sensor driver can bind but the CSI-2 stream will not function correctly.

Quoted 8 lines

The CVS driver calls acpi_dev_clear_dependencies() at the end of its probe() to unblock waiting consumers once it is ready.

Move the CVS HIDs from acpi_ignore_dep_ids[] to acpi_honor_dep_ids[] so that camera sensor enumeration is deferred until the CVS driver has finished probing, matching the behavior already in place for IVSC.

Quoted 8 lines

Rafael: do you think this could still make it to 7.2? Otherwise cameras might not work on a some laptops where they did use to.

The default configuration is appropriate in some laptops so camera used to work without the CVS driver but now that we have the CVS driver merged and the CVS device is part of the software node / ACPI graph, probing the camera sensor driver will in fact not take place at all (and obviously the camera won't work in this case).

Applied as 7.2 material, thanks!

#3 lore raw
In reply to Miguel Vadillo
Quoted 11 lines · Miguel Vadillo

CVS (Computer Vision Sensing) is an ACPI-enumerated device that sits inline in the CSI-2 path between the camera sensor and Intel IPU. On platforms where CVS is present, the camera sensor's ACPI node declares a _DEP dependency on the CVS device.

The CVS driver must be fully initialized before camera sensor drivers probe, because CVS controls the CSI-2 link ownership handshake (via GPIO REQ/RESP), the MIPI/CSI-2 lane configuration, and the camera power domain. Without CVS ready, the sensor driver can bind but the CSI-2 stream will not function correctly.

Quoted 6 lines

The CVS driver calls acpi_dev_clear_dependencies() at the end of its probe() to unblock waiting consumers once it is ready.

Move the CVS HIDs from acpi_ignore_dep_ids[] to acpi_honor_dep_ids[] so that camera sensor enumeration is deferred until the CVS driver has finished probing, matching the behavior already in place for IVSC.

TWIMC, there is a bug report where one or two people claim that this change caused a regression; the report is quite verbose (guess LLM assisted) and hard to grasp for me as outsider; requests for clarifications from my side were not really fruitful. Not sure what to make of this, but I thought I'd let you know; for details see:

https://bugzilla.kernel.org/show_bug.cgi?id=221988

Ciao, Thorsten

Quoted 36 lines

---
Note: this patch has a functional dependency on the CVS driver and its IPU bridge support, which are queued in media-next for v7.2. Without those drivers, no code will call acpi_dev_clear_dependencies() for the CVS HIDs and camera sensors will remain deferred indefinitely.

The CVS driver series (v5) can be found at:
https://lore.kernel.org/linux-media/[email protected]/

 drivers/acpi/scan.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/acpi/scan.c +3 -2
index 530547cda8b2..1463eb3a4f6e 100644
@@ -848,8 +848,6 @@ static bool acpi_info_matches_ids(struct acpi_device_info *info,
 static const char * const acpi_ignore_dep_ids[] = {
 	"PNP0D80", /* Windows-compatible System Power Management Controller */
 	"INT33BD", /* Intel Baytrail Mailbox Device */
-	"INTC10DE", /* Intel CVS LNL */
-	"INTC10E0", /* Intel CVS ARL */
 	"LATT2021", /* Lattice FW Update Client Driver */
 	NULL
 };
@@ -861,6 +859,9 @@ static const char * const acpi_honor_dep_ids[] = {
 	"INTC1095", /* IVSC (ADL) driver must be loaded to allow i2c access to camera sensors */
 	"INTC100A", /* IVSC (RPL) driver must be loaded to allow i2c access to camera sensors */
 	"INTC10CF", /* IVSC (MTL) driver must be loaded to allow i2c access to camera sensors */
+	"INTC10DE", /* CVS (LNL) driver must be loaded to allow camera streaming */
+	"INTC10E0", /* CVS (ARL) driver must be loaded to allow camera streaming */
+	"INTC10E1", /* CVS (PTL) driver must be loaded to allow camera streaming */
 	"RSCV0001", /* RISC-V PLIC */
 	"RSCV0002", /* RISC-V APLIC */
 	"RSCV0005", /* RISC-V SBI MPXY MBOX */
#4 lore raw
Quoted 19 lines · Thorsten Leemhuis
Quoted 11 lines · Miguel Vadillo

CVS (Computer Vision Sensing) is an ACPI-enumerated device that sits inline in the CSI-2 path between the camera sensor and Intel IPU. On platforms where CVS is present, the camera sensor's ACPI node declares a _DEP dependency on the CVS device.

The CVS driver must be fully initialized before camera sensor drivers probe, because CVS controls the CSI-2 link ownership handshake (via GPIO REQ/RESP), the MIPI/CSI-2 lane configuration, and the camera power domain. Without CVS ready, the sensor driver can bind but the CSI-2 stream will not function correctly.

Quoted 6 lines

The CVS driver calls acpi_dev_clear_dependencies() at the end of its probe() to unblock waiting consumers once it is ready.

Move the CVS HIDs from acpi_ignore_dep_ids[] to acpi_honor_dep_ids[] so that camera sensor enumeration is deferred until the CVS driver has finished probing, matching the behavior already in place for IVSC.

Quoted 7 lines

TWIMC, there is a bug report where one or two people claim that this change caused a regression; the report is quite verbose (guess LLM assisted) and hard to grasp for me as outsider; requests for clarifications from my side were not really fruitful. Not sure what to make of this, but I thought I'd let you know; for details see:

https://bugzilla.kernel.org/show_bug.cgi?id=221988

For completeness, a report that is more straight forward to read (at least for me) was now posted here:
https://lore.kernel.org/all/[email protected]/

Cia, Thorsten

Quoted 38 lines
Quoted 36 lines

---
Note: this patch has a functional dependency on the CVS driver and its IPU bridge support, which are queued in media-next for v7.2. Without those drivers, no code will call acpi_dev_clear_dependencies() for the CVS HIDs and camera sensors will remain deferred indefinitely.

The CVS driver series (v5) can be found at:
https://lore.kernel.org/linux-media/[email protected]/

 drivers/acpi/scan.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/acpi/scan.c +3 -2
index 530547cda8b2..1463eb3a4f6e 100644
@@ -848,8 +848,6 @@ static bool acpi_info_matches_ids(struct acpi_device_info *info,
 static const char * const acpi_ignore_dep_ids[] = {
 	"PNP0D80", /* Windows-compatible System Power Management Controller */
 	"INT33BD", /* Intel Baytrail Mailbox Device */
-	"INTC10DE", /* Intel CVS LNL */
-	"INTC10E0", /* Intel CVS ARL */
 	"LATT2021", /* Lattice FW Update Client Driver */
 	NULL
 };
@@ -861,6 +859,9 @@ static const char * const acpi_honor_dep_ids[] = {
 	"INTC1095", /* IVSC (ADL) driver must be loaded to allow i2c access to camera sensors */
 	"INTC100A", /* IVSC (RPL) driver must be loaded to allow i2c access to camera sensors */
 	"INTC10CF", /* IVSC (MTL) driver must be loaded to allow i2c access to camera sensors */
+	"INTC10DE", /* CVS (LNL) driver must be loaded to allow camera streaming */
+	"INTC10E0", /* CVS (ARL) driver must be loaded to allow camera streaming */
+	"INTC10E1", /* CVS (PTL) driver must be loaded to allow camera streaming */
 	"RSCV0001", /* RISC-V PLIC */
 	"RSCV0002", /* RISC-V APLIC */
 	"RSCV0005", /* RISC-V SBI MPXY MBOX */
#5 lore raw
Quoted 27 lines · Thorsten Leemhuis
Quoted 25 lines · Thorsten Leemhuis
Quoted 11 lines · Miguel Vadillo

CVS (Computer Vision Sensing) is an ACPI-enumerated device that sits inline in the CSI-2 path between the camera sensor and Intel IPU. On platforms where CVS is present, the camera sensor's ACPI node declares a _DEP dependency on the CVS device.

The CVS driver must be fully initialized before camera sensor drivers probe, because CVS controls the CSI-2 link ownership handshake (via GPIO REQ/RESP), the MIPI/CSI-2 lane configuration, and the camera power domain. Without CVS ready, the sensor driver can bind but the CSI-2 stream will not function correctly.

Quoted 6 lines

The CVS driver calls acpi_dev_clear_dependencies() at the end of its probe() to unblock waiting consumers once it is ready.

Move the CVS HIDs from acpi_ignore_dep_ids[] to acpi_honor_dep_ids[] so that camera sensor enumeration is deferred until the CVS driver has finished probing, matching the behavior already in place for IVSC.

TWIMC, there is a bug report where one or two people claim that this change caused a regression; the report is quite verbose (guess LLM assisted) and hard to grasp for me as outsider; requests for clarifications from my side were not really fruitful. Not sure what to make of this, but I thought I'd let you know; for details see:

Quoted 5 lines
Quoted 1 line

https://bugzilla.kernel.org/show_bug.cgi?id=221988

For completeness, a report that is more straight forward to read (at least for me) was now posted here:
https://lore.kernel.org/all/[email protected]/

This doesn't seem to be about the same commit though as it mentions ipu-bridge.c which is not touched by the patch in question here.

Thanks!

#6 lore raw
In reply to Thorsten Leemhuis

Hi Thorsten,

Quoted 26 lines · Thorsten Leemhuis
Quoted 25 lines · Thorsten Leemhuis
Quoted 11 lines · Miguel Vadillo

CVS (Computer Vision Sensing) is an ACPI-enumerated device that sits inline in the CSI-2 path between the camera sensor and Intel IPU. On platforms where CVS is present, the camera sensor's ACPI node declares a _DEP dependency on the CVS device.

The CVS driver must be fully initialized before camera sensor drivers probe, because CVS controls the CSI-2 link ownership handshake (via GPIO REQ/RESP), the MIPI/CSI-2 lane configuration, and the camera power domain. Without CVS ready, the sensor driver can bind but the CSI-2 stream will not function correctly.

Quoted 6 lines

The CVS driver calls acpi_dev_clear_dependencies() at the end of its probe() to unblock waiting consumers once it is ready.

Move the CVS HIDs from acpi_ignore_dep_ids[] to acpi_honor_dep_ids[] so that camera sensor enumeration is deferred until the CVS driver has finished probing, matching the behavior already in place for IVSC.

TWIMC, there is a bug report where one or two people claim that this change caused a regression; the report is quite verbose (guess LLM assisted) and hard to grasp for me as outsider; requests for clarifications from my side were not really fruitful. Not sure what to make of this, but I thought I'd let you know; for details see:

Quoted 5 lines
Quoted 1 line

https://bugzilla.kernel.org/show_bug.cgi?id=221988

For completeness, a report that is more straight forward to read (at least for me) was now posted here:
https://lore.kernel.org/all/[email protected]/

Thanks for the link, I've replied to the bug report.

Signature · 3 lines
Kind regards,

Sakari Ailus