[meta-ti][scarthgap][RFC 1/6] ti-sgx-ddk-um: UM side driver depends on KM side driver

[meta-ti][scarthgap][RFC 1/6] ti-sgx-ddk-um: UM side driver depends on KM side driver

[Andrew Davis]

This also matches what is done for ti-img-rogue-umlibs.

Signed-off-by: Andrew Davis <afd@ti.com>
---
 .../powervr-umlibs/ti-sgx-ddk-um_1.17.4948957.bb              | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/meta-ti-bsp/recipes-graphics/powervr-umlibs/ti-sgx-ddk-um_1.17.4948957.bb b/meta-ti-bsp/recipes-graphics/powervr-umlibs/ti-sgx-ddk-um_1.17.4948957.bb
index b260abdf..b6f5534e 100644
--- a/meta-ti-bsp/recipes-graphics/powervr-umlibs/ti-sgx-ddk-um_1.17.4948957.bb
+++ b/meta-ti-bsp/recipes-graphics/powervr-umlibs/ti-sgx-ddk-um_1.17.4948957.bb
@@ -40,9 +40,7 @@ TARGET_PRODUCT:ti43x = "ti437x_linux"
 TARGET_PRODUCT:omap-a15 = "ti572x_linux"
 TARGET_PRODUCT:am65xx = "ti654x_linux"
 
-RDEPENDS:${PN} += "libdrm"
-
-RRECOMMENDS:${PN} += "ti-sgx-ddk-km"
+RDEPENDS:${PN} += "libdrm ti-sgx-ddk-km"
 
 S = "${WORKDIR}/git"
 
-- 
2.39.2

[meta-ti][scarthgap][RFC 2/6] ti-bsp: Do not remove gpu MACHINE_FEATURE conditionally on BSP version

[Andrew Davis]

Setting or removing the gpu MACHINE_FEATURE does not actually change
if the GPU driver is built or not. That is done based on the setting
of the BSP_{SGX,ROGUE}_{DRIVER,UMLIBS}_{PROVIDER,VERSION} vars. These
are default empty for BSP versions without supported GPU drivers, which
will correctly cause a fallback to SW rendering when the GPU driver is
not available.

Signed-off-by: Andrew Davis <afd@ti.com>
---
 meta-ti-bsp/conf/machine/include/ti-bsp.inc | 8 --------
 1 file changed, 8 deletions(-)

diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
index 262aa884..a5d9a51f 100644
--- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
+++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
@@ -18,10 +18,6 @@ BSP_KERNEL_VERSION:bsp-mainline = "%"
 BSP_BOOTLOADER_PROVIDER:bsp-mainline = "u-boot-ti-mainline"
 BSP_BOOTLOADER_VERSION:bsp-mainline = "%"
 
-# GPU support requires out-of-tree SGX and Rogue drivers not available
-# in mainline, usually present in TI staging or derivative like BB.org
-MACHINE_FEATURES:remove:bsp-mainline = "gpu"
-
 # GC320 support requires out-of-tree drivers not available in mainline
 MACHINE_FEATURES:remove:bsp-mainline = "gc320"
 
@@ -34,10 +30,6 @@ BSP_KERNEL_VERSION:bsp-next = "%"
 BSP_BOOTLOADER_PROVIDER:bsp-next = "u-boot-ti-next"
 BSP_BOOTLOADER_VERSION:bsp-next = "%"
 
-# GPU support requires out-of-tree SGX and Rogue drivers not available
-# in next, usually present in TI staging or derivative like BB.org
-MACHINE_FEATURES:remove:bsp-next = "gpu"
-
 # GC320 support requires out-of-tree drivers not available in next
 MACHINE_FEATURES:remove:bsp-next = "gc320"
 
-- 
2.39.2

[meta-ti][scarthgap][RFC 3/6] ti-bsp: Use same version for both UM and KM driver

[Andrew Davis]

The DRIVER and matching UMLIBS must always match, no need to define a
version number for both. Keep this simple and remove the extra defines.

Signed-off-by: Andrew Davis <afd@ti.com>
---
 meta-beagle/conf/machine/include/beagle-bsp.inc |  4 ----
 meta-ti-bsp/conf/machine/include/ti-bsp.inc     | 10 ++--------
 2 files changed, 2 insertions(+), 12 deletions(-)

diff --git a/meta-beagle/conf/machine/include/beagle-bsp.inc b/meta-beagle/conf/machine/include/beagle-bsp.inc
index d180c297..448e2086 100644
--- a/meta-beagle/conf/machine/include/beagle-bsp.inc
+++ b/meta-beagle/conf/machine/include/beagle-bsp.inc
@@ -18,10 +18,8 @@ KERNEL_DEVICETREE_PREFIX:bsp-bb_org-6_12 = ""
 
 BSP_SGX_DRIVER_PROVIDER:bsp-bb_org-6_12 = "ti-sgx-ddk-km"
 BSP_SGX_DRIVER_VERSION:bsp-bb_org-6_12 = "1.17%"
-BSP_SGX_UMLIBS_VERSION:bsp-bb_org-6_12 = "1.17%"
 BSP_ROGUE_DRIVER_PROVIDER:bsp-bb_org-6_12 = "ti-img-rogue-driver"
 BSP_ROGUE_DRIVER_VERSION:bsp-bb_org-6_12 = "24%"
-BSP_ROGUE_UMLIBS_VERSION:bsp-bb_org-6_12 = "24%"
 BSP_MESA_PVR_VERSION:bsp-bb_org-6_12 = "24%"
 
 # GC320 support requires out-of-tree drivers not yet available in bb.org 6.12
@@ -41,10 +39,8 @@ KERNEL_DEVICETREE_PREFIX:bsp-bb_org-6_6 = ""
 
 BSP_SGX_DRIVER_PROVIDER:bsp-bb_org-6_6 = "ti-sgx-ddk-km"
 BSP_SGX_DRIVER_VERSION:bsp-bb_org-6_6 = "1.17%"
-BSP_SGX_UMLIBS_VERSION:bsp-bb_org-6_6 = "1.17%"
 BSP_ROGUE_DRIVER_PROVIDER:bsp-bb_org-6_6 = "ti-img-rogue-driver"
 BSP_ROGUE_DRIVER_VERSION:bsp-bb_org-6_6 = "24%"
-BSP_ROGUE_UMLIBS_VERSION:bsp-bb_org-6_6 = "24%"
 BSP_MESA_PVR_VERSION:bsp-bb_org-6_6 = "24%"
 
 # GC320 support requires out-of-tree drivers not yet available in bb.org 6.6
diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
index a5d9a51f..dc0473de 100644
--- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
+++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
@@ -44,10 +44,8 @@ BSP_BOOTLOADER_VERSION:bsp-ti-6_12 = "2025.01%"
 
 BSP_SGX_DRIVER_PROVIDER:bsp-ti-6_12 = "ti-sgx-ddk-km"
 BSP_SGX_DRIVER_VERSION:bsp-ti-6_12 = "1.17%"
-BSP_SGX_UMLIBS_VERSION:bsp-ti-6_12 = "1.17%"
 BSP_ROGUE_DRIVER_PROVIDER:bsp-ti-6_12 = "ti-img-rogue-driver"
 BSP_ROGUE_DRIVER_VERSION:bsp-ti-6_12 = "24%"
-BSP_ROGUE_UMLIBS_VERSION:bsp-ti-6_12 = "24%"
 BSP_MESA_PVR_VERSION:bsp-ti-6_12 = "24%"
 
 # ==========
@@ -61,10 +59,8 @@ BSP_BOOTLOADER_VERSION:bsp-ti-6_6 = "2024.04%"
 
 BSP_SGX_DRIVER_PROVIDER:bsp-ti-6_6 = "ti-sgx-ddk-km"
 BSP_SGX_DRIVER_VERSION:bsp-ti-6_6 = "1.17%"
-BSP_SGX_UMLIBS_VERSION:bsp-ti-6_6 = "1.17%"
 BSP_ROGUE_DRIVER_PROVIDER:bsp-ti-6_6 = "ti-img-rogue-driver"
 BSP_ROGUE_DRIVER_VERSION:bsp-ti-6_6 = "24%"
-BSP_ROGUE_UMLIBS_VERSION:bsp-ti-6_6 = "24%"
 BSP_MESA_PVR_VERSION:bsp-ti-6_6 = "24%"
 
 # GC320 support requires out-of-tree drivers not yet available in 6.6
@@ -84,8 +80,6 @@ BSP_SGX_DRIVER_PROVIDER ?= ""
 BSP_SGX_DRIVER_VERSION ?= ""
 BSP_ROGUE_DRIVER_PROVIDER ?= ""
 BSP_ROGUE_DRIVER_VERSION ?= ""
-BSP_SGX_UMLIBS_VERSION ?= ""
-BSP_ROGUE_UMLIBS_VERSION ?= ""
 BSP_MESA_PVR_VERSION ?= ""
 
 # ==========
@@ -101,7 +95,7 @@ PREFERRED_VERSION_${BSP_BOOTLOADER_PROVIDER} ?= "${BSP_BOOTLOADER_VERSION}"
 # GPU provider gets set in machine configs, as some machines are headless
 # Select default preferred versions here
 PREFERRED_VERSION_ti-sgx-ddk-km ?= "${BSP_SGX_DRIVER_VERSION}"
-PREFERRED_VERSION_ti-sgx-ddk-um ?= "${BSP_SGX_UMLIBS_VERSION}"
+PREFERRED_VERSION_ti-sgx-ddk-um ?= "${BSP_SGX_DRIVER_VERSION}"
 PREFERRED_VERSION_ti-img-rogue-driver ?= "${BSP_ROGUE_DRIVER_VERSION}"
-PREFERRED_VERSION_ti-img-rogue-umlibs ?= "${BSP_ROGUE_UMLIBS_VERSION}"
+PREFERRED_VERSION_ti-img-rogue-umlibs ?= "${BSP_ROGUE_DRIVER_VERSION}"
 PREFERRED_VERSION_mesa-pvr ?= "${BSP_MESA_PVR_VERSION}"
-- 
2.39.2

[meta-ti][scarthgap][RFC 4/6] mesa-pvr: Remove need for PVR/SGX display aliases

[Andrew Davis]

Update mesa-pvr which includes a default set of display controller
aliases for the supported display controllers. This means we do not
have to manually set the alias for each device here in the integration
layer. Remove that here.

Signed-off-by: Andrew Davis <afd@ti.com>
---
 meta-ti-bsp/conf/machine/include/omap-a15.inc        | 1 -
 meta-ti-bsp/conf/machine/include/ti33x.inc           | 1 -
 meta-ti-bsp/conf/machine/include/ti43x.inc           | 1 -
 meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb | 7 +------
 4 files changed, 1 insertion(+), 9 deletions(-)

diff --git a/meta-ti-bsp/conf/machine/include/omap-a15.inc b/meta-ti-bsp/conf/machine/include/omap-a15.inc
index 2179476f..9058b461 100644
--- a/meta-ti-bsp/conf/machine/include/omap-a15.inc
+++ b/meta-ti-bsp/conf/machine/include/omap-a15.inc
@@ -9,7 +9,6 @@ require conf/machine/include/ti-bsp.inc
 # Graphics providers and variables
 require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_SGX_DRIVER_PROVIDER}"
-PVR_DISPLAY_CONTROLLER_ALIAS ?= "omapdrm"
 
 KERNEL_IMAGETYPE = "zImage"
 KERNEL_IMAGETYPES = "zImage uImage"
diff --git a/meta-ti-bsp/conf/machine/include/ti33x.inc b/meta-ti-bsp/conf/machine/include/ti33x.inc
index c47165e4..5d7375e5 100644
--- a/meta-ti-bsp/conf/machine/include/ti33x.inc
+++ b/meta-ti-bsp/conf/machine/include/ti33x.inc
@@ -9,7 +9,6 @@ require conf/machine/include/ti-bsp.inc
 # Graphics providers and variables
 require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_SGX_DRIVER_PROVIDER}"
-PVR_DISPLAY_CONTROLLER_ALIAS ?= "tilcdc"
 
 KERNEL_IMAGETYPE = "zImage"
 
diff --git a/meta-ti-bsp/conf/machine/include/ti43x.inc b/meta-ti-bsp/conf/machine/include/ti43x.inc
index a23b6518..7dd5b1f2 100644
--- a/meta-ti-bsp/conf/machine/include/ti43x.inc
+++ b/meta-ti-bsp/conf/machine/include/ti43x.inc
@@ -9,7 +9,6 @@ require conf/machine/include/ti-bsp.inc
 # Graphics providers and variables
 require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_SGX_DRIVER_PROVIDER}"
-PVR_DISPLAY_CONTROLLER_ALIAS ?= "omapdrm"
 
 KERNEL_IMAGETYPE = "zImage"
 
diff --git a/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb b/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
index e6bd5fde..6a4b296f 100644
--- a/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
+++ b/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
@@ -28,14 +28,9 @@ PACKAGECONFIG:append = " \
     ${@bb.utils.contains('PREFERRED_PROVIDER_virtual/gpudriver', 'ti-sgx-ddk-km', 'sgx', '', d)} \
 "
 
-SRCREV = "82e6a9293c476267417c5b6b906b01fb73a34e38"
+SRCREV = "68af6a102c2298569e77d1aa8bccc1ff61438b3e"
 PV = "24.0.1"
 
-PVR_DISPLAY_CONTROLLER_ALIAS ??= "tidss"
-PACKAGECONFIG[pvr] = "-Dgallium-pvr-alias=${PVR_DISPLAY_CONTROLLER_ALIAS},"
-PACKAGECONFIG[sgx] = "-Dgallium-sgx-alias=${PVR_DISPLAY_CONTROLLER_ALIAS},"
-
-
 PACKAGE_ARCH = "${MACHINE_ARCH}"
 
 GALLIUMDRIVERS:append = "${@bb.utils.contains('PACKAGECONFIG', 'pvr', ',pvr', '', d)}"
-- 
2.39.2

[meta-ti][scarthgap][RFC 5/6] mesa-pvr: Use PREFERRED_PROVIDER directly for GPU selection

[Andrew Davis]

Now that display controller aliases are not set using PACKAGECONFIG,
instead of using PREFERRED_PROVIDER to add either "sgx" or "pvr" to
PACKAGECONFIG, followed by reading it back out, simply use
PREFERRED_PROVIDER directly.

Signed-off-by: Andrew Davis <afd@ti.com>
---
 meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb b/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
index 6a4b296f..40226f65 100644
--- a/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
+++ b/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
@@ -23,20 +23,15 @@ SRC_URI = " \
 
 S = "${WORKDIR}/git"
 
-PACKAGECONFIG:append = " \
-    ${@bb.utils.contains('PREFERRED_PROVIDER_virtual/gpudriver', 'ti-img-rogue-driver', 'pvr', '', d)} \
-    ${@bb.utils.contains('PREFERRED_PROVIDER_virtual/gpudriver', 'ti-sgx-ddk-km', 'sgx', '', d)} \
-"
-
 SRCREV = "68af6a102c2298569e77d1aa8bccc1ff61438b3e"
 PV = "24.0.1"
 
 PACKAGE_ARCH = "${MACHINE_ARCH}"
 
-GALLIUMDRIVERS:append = "${@bb.utils.contains('PACKAGECONFIG', 'pvr', ',pvr', '', d)}"
-GALLIUMDRIVERS:append = "${@bb.utils.contains('PACKAGECONFIG', 'sgx', ',sgx', '', d)}"
+GALLIUMDRIVERS:append = "${@bb.utils.contains('PREFERRED_PROVIDER_virtual/gpudriver', 'ti-img-rogue-driver', ',pvr', '', d)}"
+GALLIUMDRIVERS:append = "${@bb.utils.contains('PREFERRED_PROVIDER_virtual/gpudriver', 'ti-sgx-ddk-km', ',sgx', '', d)}"
 
-VULKAN_DRIVERS:append = "${@bb.utils.contains('PACKAGECONFIG', 'pvr', ',pvr', '', d)}"
+VULKAN_DRIVERS:append = "${@bb.utils.contains('PREFERRED_PROVIDER_virtual/gpudriver', 'ti-img-rogue-driver', ',pvr', '', d)}"
 
 do_install:append () {
     # remove pvr custom pkgconfig
-- 
2.39.2

[meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Andrew Davis]

TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
SW rendering just like the normal mesa package. The benefit is using the
same version across all supported devices is more consistent and only
one version of Mesa is needed for all TI SoCs.

Signed-off-by: Andrew Davis <afd@ti.com>
---
 meta-ti-bsp/conf/machine/beagleplay-ti.conf          | 1 -
 meta-ti-bsp/conf/machine/include/am62pxx.inc         | 1 -
 meta-ti-bsp/conf/machine/include/am62xx.inc          | 1 -
 meta-ti-bsp/conf/machine/include/am65xx.inc          | 1 -
 meta-ti-bsp/conf/machine/include/am68.inc            | 1 -
 meta-ti-bsp/conf/machine/include/am69.inc            | 1 -
 meta-ti-bsp/conf/machine/include/j721e.inc           | 1 -
 meta-ti-bsp/conf/machine/include/j721s2.inc          | 1 -
 meta-ti-bsp/conf/machine/include/j722s.inc           | 1 -
 meta-ti-bsp/conf/machine/include/j742s2.inc          | 1 -
 meta-ti-bsp/conf/machine/include/j784s4.inc          | 1 -
 meta-ti-bsp/conf/machine/include/mesa-pvr.inc        | 8 --------
 meta-ti-bsp/conf/machine/include/omap-a15.inc        | 1 -
 meta-ti-bsp/conf/machine/include/ti-bsp.inc          | 8 ++++++++
 meta-ti-bsp/conf/machine/include/ti33x.inc           | 1 -
 meta-ti-bsp/conf/machine/include/ti43x.inc           | 1 -
 meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb | 1 +
 17 files changed, 9 insertions(+), 22 deletions(-)
 delete mode 100644 meta-ti-bsp/conf/machine/include/mesa-pvr.inc

diff --git a/meta-ti-bsp/conf/machine/beagleplay-ti.conf b/meta-ti-bsp/conf/machine/beagleplay-ti.conf
index 792c91b2..8a11a188 100644
--- a/meta-ti-bsp/conf/machine/beagleplay-ti.conf
+++ b/meta-ti-bsp/conf/machine/beagleplay-ti.conf
@@ -11,7 +11,6 @@ MACHINE_FEATURES += "screen gpu"
 
 SERIAL_CONSOLES = "115200;ttyS2"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "lite"
diff --git a/meta-ti-bsp/conf/machine/include/am62pxx.inc b/meta-ti-bsp/conf/machine/include/am62pxx.inc
index d59f607d..05fc6433 100644
--- a/meta-ti-bsp/conf/machine/include/am62pxx.inc
+++ b/meta-ti-bsp/conf/machine/include/am62pxx.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":am62pxx"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "lite"
diff --git a/meta-ti-bsp/conf/machine/include/am62xx.inc b/meta-ti-bsp/conf/machine/include/am62xx.inc
index 74054a78..16bfa3d3 100644
--- a/meta-ti-bsp/conf/machine/include/am62xx.inc
+++ b/meta-ti-bsp/conf/machine/include/am62xx.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":am62xx"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 # Bitmap image tarball for early splashscreen
diff --git a/meta-ti-bsp/conf/machine/include/am65xx.inc b/meta-ti-bsp/conf/machine/include/am65xx.inc
index d70bb22d..6b482227 100644
--- a/meta-ti-bsp/conf/machine/include/am65xx.inc
+++ b/meta-ti-bsp/conf/machine/include/am65xx.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":am65xx"
 
 MACHINE_FEATURES += "screen touchscreen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_SGX_DRIVER_PROVIDER}"
 
 KERNEL_DEVICETREE_PREFIX = "ti/k3-am654"
diff --git a/meta-ti-bsp/conf/machine/include/am68.inc b/meta-ti-bsp/conf/machine/include/am68.inc
index b6d18fe0..a4ac12f3 100644
--- a/meta-ti-bsp/conf/machine/include/am68.inc
+++ b/meta-ti-bsp/conf/machine/include/am68.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":j721s2"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "generic"
diff --git a/meta-ti-bsp/conf/machine/include/am69.inc b/meta-ti-bsp/conf/machine/include/am69.inc
index 759984ea..c98dad7b 100644
--- a/meta-ti-bsp/conf/machine/include/am69.inc
+++ b/meta-ti-bsp/conf/machine/include/am69.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":j784s4"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "j784s4"
diff --git a/meta-ti-bsp/conf/machine/include/j721e.inc b/meta-ti-bsp/conf/machine/include/j721e.inc
index f9fdac0c..a892736a 100644
--- a/meta-ti-bsp/conf/machine/include/j721e.inc
+++ b/meta-ti-bsp/conf/machine/include/j721e.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":j721e"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 # Deploy SYSFW and any variants
diff --git a/meta-ti-bsp/conf/machine/include/j721s2.inc b/meta-ti-bsp/conf/machine/include/j721s2.inc
index b6d18fe0..a4ac12f3 100644
--- a/meta-ti-bsp/conf/machine/include/j721s2.inc
+++ b/meta-ti-bsp/conf/machine/include/j721s2.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":j721s2"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "generic"
diff --git a/meta-ti-bsp/conf/machine/include/j722s.inc b/meta-ti-bsp/conf/machine/include/j722s.inc
index 8a02441b..fa860f79 100644
--- a/meta-ti-bsp/conf/machine/include/j722s.inc
+++ b/meta-ti-bsp/conf/machine/include/j722s.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":j722s"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "lite"
diff --git a/meta-ti-bsp/conf/machine/include/j742s2.inc b/meta-ti-bsp/conf/machine/include/j742s2.inc
index d2de1d0d..7ede2de5 100644
--- a/meta-ti-bsp/conf/machine/include/j742s2.inc
+++ b/meta-ti-bsp/conf/machine/include/j742s2.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":j742s2"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "j784s4"
diff --git a/meta-ti-bsp/conf/machine/include/j784s4.inc b/meta-ti-bsp/conf/machine/include/j784s4.inc
index 759984ea..c98dad7b 100644
--- a/meta-ti-bsp/conf/machine/include/j784s4.inc
+++ b/meta-ti-bsp/conf/machine/include/j784s4.inc
@@ -5,7 +5,6 @@ SOC_FAMILY:append = ":j784s4"
 
 MACHINE_FEATURES += "screen gpu"
 
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_ROGUE_DRIVER_PROVIDER}"
 
 TFA_BOARD = "j784s4"
diff --git a/meta-ti-bsp/conf/machine/include/mesa-pvr.inc b/meta-ti-bsp/conf/machine/include/mesa-pvr.inc
deleted file mode 100644
index d5581315..00000000
--- a/meta-ti-bsp/conf/machine/include/mesa-pvr.inc
+++ /dev/null
@@ -1,8 +0,0 @@
-PREFERRED_PROVIDER_virtual/mesa ?= "mesa-pvr"
-PREFERRED_PROVIDER_virtual/egl ?= "mesa-pvr"
-PREFERRED_PROVIDER_virtual/libgles1 ?= "mesa-pvr"
-PREFERRED_PROVIDER_virtual/libgles2 ?= "mesa-pvr"
-PREFERRED_PROVIDER_virtual/libgles3 ?= "mesa-pvr"
-PREFERRED_PROVIDER_virtual/libgbm ?= "mesa-pvr"
-PREFERRED_PROVIDER_virtual/libglx ?= "mesa-pvr"
-PREFERRED_PROVIDER_virtual/libgl ?= "mesa-pvr"
diff --git a/meta-ti-bsp/conf/machine/include/omap-a15.inc b/meta-ti-bsp/conf/machine/include/omap-a15.inc
index 9058b461..c48f89ae 100644
--- a/meta-ti-bsp/conf/machine/include/omap-a15.inc
+++ b/meta-ti-bsp/conf/machine/include/omap-a15.inc
@@ -7,7 +7,6 @@ require conf/machine/include/arm/armv7a/tune-cortexa15.inc
 require conf/machine/include/ti-bsp.inc
 
 # Graphics providers and variables
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_SGX_DRIVER_PROVIDER}"
 
 KERNEL_IMAGETYPE = "zImage"
diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
index dc0473de..37a59135 100644
--- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
+++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
@@ -99,3 +99,11 @@ PREFERRED_VERSION_ti-sgx-ddk-um ?= "${BSP_SGX_DRIVER_VERSION}"
 PREFERRED_VERSION_ti-img-rogue-driver ?= "${BSP_ROGUE_DRIVER_VERSION}"
 PREFERRED_VERSION_ti-img-rogue-umlibs ?= "${BSP_ROGUE_DRIVER_VERSION}"
 PREFERRED_VERSION_mesa-pvr ?= "${BSP_MESA_PVR_VERSION}"
+PREFERRED_PROVIDER_virtual/mesa ?= "mesa-pvr"
+PREFERRED_PROVIDER_virtual/egl ?= "mesa-pvr"
+PREFERRED_PROVIDER_virtual/libgles1 ?= "mesa-pvr"
+PREFERRED_PROVIDER_virtual/libgles2 ?= "mesa-pvr"
+PREFERRED_PROVIDER_virtual/libgles3 ?= "mesa-pvr"
+PREFERRED_PROVIDER_virtual/libgbm ?= "mesa-pvr"
+PREFERRED_PROVIDER_virtual/libglx ?= "mesa-pvr"
+PREFERRED_PROVIDER_virtual/libgl ?= "mesa-pvr"
diff --git a/meta-ti-bsp/conf/machine/include/ti33x.inc b/meta-ti-bsp/conf/machine/include/ti33x.inc
index 5d7375e5..1301e2e7 100644
--- a/meta-ti-bsp/conf/machine/include/ti33x.inc
+++ b/meta-ti-bsp/conf/machine/include/ti33x.inc
@@ -7,7 +7,6 @@ require conf/machine/include/arm/armv7a/tune-cortexa8.inc
 require conf/machine/include/ti-bsp.inc
 
 # Graphics providers and variables
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_SGX_DRIVER_PROVIDER}"
 
 KERNEL_IMAGETYPE = "zImage"
diff --git a/meta-ti-bsp/conf/machine/include/ti43x.inc b/meta-ti-bsp/conf/machine/include/ti43x.inc
index 7dd5b1f2..2b061d53 100644
--- a/meta-ti-bsp/conf/machine/include/ti43x.inc
+++ b/meta-ti-bsp/conf/machine/include/ti43x.inc
@@ -7,7 +7,6 @@ require conf/machine/include/arm/armv7a/tune-cortexa9.inc
 require conf/machine/include/ti-bsp.inc
 
 # Graphics providers and variables
-require conf/machine/include/mesa-pvr.inc
 PREFERRED_PROVIDER_virtual/gpudriver ?= "${BSP_SGX_DRIVER_PROVIDER}"
 
 KERNEL_IMAGETYPE = "zImage"
diff --git a/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb b/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
index 40226f65..a432dcdd 100644
--- a/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
+++ b/meta-ti-bsp/recipes-graphics/mesa/mesa-pvr_24.0.1.bb
@@ -41,4 +41,5 @@ do_install:append () {
 FILES:${PN}-dev += "${datadir}/mesa/wayland-drm.xml"
 FILES:mesa-vulkan-drivers += "${libdir}/libpvr_mesa_wsi.so"
 
+PREFERRED_PROVIDER_virtual/gpudriver ?= ""
 RRECOMMENDS:mesa-megadriver:append:class-target = " ${@d.getVar('PREFERRED_PROVIDER_virtual/gpudriver')}"
-- 
2.39.2

Re: [meta-ti][scarthgap][RFC 3/6] ti-bsp: Use same version for both UM and KM driver

[Ryan Eatmon]

On 9/9/2025 10:10 AM, Andrew Davis wrote:
> The DRIVER and matching UMLIBS must always match, no need to define a
> version number for both. Keep this simple and remove the extra defines.
> 
> Signed-off-by: Andrew Davis <afd@ti.com>
> ---
>   meta-beagle/conf/machine/include/beagle-bsp.inc |  4 ----
>   meta-ti-bsp/conf/machine/include/ti-bsp.inc     | 10 ++--------
>   2 files changed, 2 insertions(+), 12 deletions(-)
> 
> diff --git a/meta-beagle/conf/machine/include/beagle-bsp.inc b/meta-beagle/conf/machine/include/beagle-bsp.inc
> index d180c297..448e2086 100644
> --- a/meta-beagle/conf/machine/include/beagle-bsp.inc
> +++ b/meta-beagle/conf/machine/include/beagle-bsp.inc
> @@ -18,10 +18,8 @@ KERNEL_DEVICETREE_PREFIX:bsp-bb_org-6_12 = ""
>   
>   BSP_SGX_DRIVER_PROVIDER:bsp-bb_org-6_12 = "ti-sgx-ddk-km"
>   BSP_SGX_DRIVER_VERSION:bsp-bb_org-6_12 = "1.17%"
> -BSP_SGX_UMLIBS_VERSION:bsp-bb_org-6_12 = "1.17%"
>   BSP_ROGUE_DRIVER_PROVIDER:bsp-bb_org-6_12 = "ti-img-rogue-driver"
>   BSP_ROGUE_DRIVER_VERSION:bsp-bb_org-6_12 = "24%"
> -BSP_ROGUE_UMLIBS_VERSION:bsp-bb_org-6_12 = "24%"
>   BSP_MESA_PVR_VERSION:bsp-bb_org-6_12 = "24%"
>   
>   # GC320 support requires out-of-tree drivers not yet available in bb.org 6.12
> @@ -41,10 +39,8 @@ KERNEL_DEVICETREE_PREFIX:bsp-bb_org-6_6 = ""
>   
>   BSP_SGX_DRIVER_PROVIDER:bsp-bb_org-6_6 = "ti-sgx-ddk-km"
>   BSP_SGX_DRIVER_VERSION:bsp-bb_org-6_6 = "1.17%"
> -BSP_SGX_UMLIBS_VERSION:bsp-bb_org-6_6 = "1.17%"
>   BSP_ROGUE_DRIVER_PROVIDER:bsp-bb_org-6_6 = "ti-img-rogue-driver"
>   BSP_ROGUE_DRIVER_VERSION:bsp-bb_org-6_6 = "24%"
> -BSP_ROGUE_UMLIBS_VERSION:bsp-bb_org-6_6 = "24%"
>   BSP_MESA_PVR_VERSION:bsp-bb_org-6_6 = "24%"
>   
>   # GC320 support requires out-of-tree drivers not yet available in bb.org 6.6
> diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> index a5d9a51f..dc0473de 100644
> --- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> +++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> @@ -44,10 +44,8 @@ BSP_BOOTLOADER_VERSION:bsp-ti-6_12 = "2025.01%"
>   
>   BSP_SGX_DRIVER_PROVIDER:bsp-ti-6_12 = "ti-sgx-ddk-km"
>   BSP_SGX_DRIVER_VERSION:bsp-ti-6_12 = "1.17%"
> -BSP_SGX_UMLIBS_VERSION:bsp-ti-6_12 = "1.17%"
>   BSP_ROGUE_DRIVER_PROVIDER:bsp-ti-6_12 = "ti-img-rogue-driver"
>   BSP_ROGUE_DRIVER_VERSION:bsp-ti-6_12 = "24%"
> -BSP_ROGUE_UMLIBS_VERSION:bsp-ti-6_12 = "24%"
>   BSP_MESA_PVR_VERSION:bsp-ti-6_12 = "24%"
>   
>   # ==========
> @@ -61,10 +59,8 @@ BSP_BOOTLOADER_VERSION:bsp-ti-6_6 = "2024.04%"
>   
>   BSP_SGX_DRIVER_PROVIDER:bsp-ti-6_6 = "ti-sgx-ddk-km"
>   BSP_SGX_DRIVER_VERSION:bsp-ti-6_6 = "1.17%"
> -BSP_SGX_UMLIBS_VERSION:bsp-ti-6_6 = "1.17%"
>   BSP_ROGUE_DRIVER_PROVIDER:bsp-ti-6_6 = "ti-img-rogue-driver"
>   BSP_ROGUE_DRIVER_VERSION:bsp-ti-6_6 = "24%"
> -BSP_ROGUE_UMLIBS_VERSION:bsp-ti-6_6 = "24%"
>   BSP_MESA_PVR_VERSION:bsp-ti-6_6 = "24%"
>   
>   # GC320 support requires out-of-tree drivers not yet available in 6.6
> @@ -84,8 +80,6 @@ BSP_SGX_DRIVER_PROVIDER ?= ""
>   BSP_SGX_DRIVER_VERSION ?= ""
>   BSP_ROGUE_DRIVER_PROVIDER ?= ""
>   BSP_ROGUE_DRIVER_VERSION ?= ""
> -BSP_SGX_UMLIBS_VERSION ?= ""
> -BSP_ROGUE_UMLIBS_VERSION ?= ""
>   BSP_MESA_PVR_VERSION ?= ""
>   
>   # ==========
> @@ -101,7 +95,7 @@ PREFERRED_VERSION_${BSP_BOOTLOADER_PROVIDER} ?= "${BSP_BOOTLOADER_VERSION}"
>   # GPU provider gets set in machine configs, as some machines are headless
>   # Select default preferred versions here
>   PREFERRED_VERSION_ti-sgx-ddk-km ?= "${BSP_SGX_DRIVER_VERSION}"
> -PREFERRED_VERSION_ti-sgx-ddk-um ?= "${BSP_SGX_UMLIBS_VERSION}"
> +PREFERRED_VERSION_ti-sgx-ddk-um ?= "${BSP_SGX_DRIVER_VERSION}"

Since we are simplifying, do we want to take it further and rename the 
XXX_DRIVER_VERSION to just XXX_VERSION?

BSP_ROGUE_DRIVER_VERSION -> BSP_ROGUE_VERSION
BSP_SGX_DRIVER_VERSION -> BSP_SGX_VERSION

>   PREFERRED_VERSION_ti-img-rogue-driver ?= "${BSP_ROGUE_DRIVER_VERSION}"
> -PREFERRED_VERSION_ti-img-rogue-umlibs ?= "${BSP_ROGUE_UMLIBS_VERSION}"
> +PREFERRED_VERSION_ti-img-rogue-umlibs ?= "${BSP_ROGUE_DRIVER_VERSION}"
>   PREFERRED_VERSION_mesa-pvr ?= "${BSP_MESA_PVR_VERSION}"

-- 
Ryan Eatmon                reatmon@ti.com
-----------------------------------------
Texas Instruments, Inc.  -  LCPD  -  MGTS

Re: [meta-ti][scarthgap][RFC 2/6] ti-bsp: Do not remove gpu MACHINE_FEATURE conditionally on BSP version

[Denys Dmytriyenko]

On Tue, Sep 09, 2025 at 10:10:24AM -0500, Andrew Davis via lists.yoctoproject.org wrote:
> Setting or removing the gpu MACHINE_FEATURE does not actually change
> if the GPU driver is built or not. That is done based on the setting
> of the BSP_{SGX,ROGUE}_{DRIVER,UMLIBS}_{PROVIDER,VERSION} vars. These
> are default empty for BSP versions without supported GPU drivers, which
> will correctly cause a fallback to SW rendering when the GPU driver is
> not available.

While cleaning the remaining uses of "gpu" MACHINE_FEATURE, why not remove it 
completely from all the machine configs?


> Signed-off-by: Andrew Davis <afd@ti.com>
> ---
>  meta-ti-bsp/conf/machine/include/ti-bsp.inc | 8 --------
>  1 file changed, 8 deletions(-)
> 
> diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> index 262aa884..a5d9a51f 100644
> --- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> +++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> @@ -18,10 +18,6 @@ BSP_KERNEL_VERSION:bsp-mainline = "%"
>  BSP_BOOTLOADER_PROVIDER:bsp-mainline = "u-boot-ti-mainline"
>  BSP_BOOTLOADER_VERSION:bsp-mainline = "%"
>  
> -# GPU support requires out-of-tree SGX and Rogue drivers not available
> -# in mainline, usually present in TI staging or derivative like BB.org
> -MACHINE_FEATURES:remove:bsp-mainline = "gpu"
> -
>  # GC320 support requires out-of-tree drivers not available in mainline
>  MACHINE_FEATURES:remove:bsp-mainline = "gc320"
>  
> @@ -34,10 +30,6 @@ BSP_KERNEL_VERSION:bsp-next = "%"
>  BSP_BOOTLOADER_PROVIDER:bsp-next = "u-boot-ti-next"
>  BSP_BOOTLOADER_VERSION:bsp-next = "%"
>  
> -# GPU support requires out-of-tree SGX and Rogue drivers not available
> -# in next, usually present in TI staging or derivative like BB.org
> -MACHINE_FEATURES:remove:bsp-next = "gpu"
> -
>  # GC320 support requires out-of-tree drivers not available in next
>  MACHINE_FEATURES:remove:bsp-next = "gc320"
>  
> -- 
> 2.39.2

Re: [meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Randolph Sapp]

On Tue Sep 9, 2025 at 10:10 AM CDT, Andrew Davis wrote:
> TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
> SW rendering just like the normal mesa package. The benefit is using the
> same version across all supported devices is more consistent and only
> one version of Mesa is needed for all TI SoCs.
>
> Signed-off-by: Andrew Davis <afd@ti.com>
> ---

[snip]

Originally this was done to mitigate damage in the event where there is an
issue with mesa. I'm not entirely sure we want to move everything over to
pvr-mesa as they still hijack some of the common dri code for their own goofy
stuff.

My opinion is we should use vanilla software where we can. Especially
considering we'll need to be able to access vanilla mesa for AM62 later if I get
my way.

- Randolph

Re: [meta-ti][scarthgap][RFC 3/6] ti-bsp: Use same version for both UM and KM driver

[Andrew Davis]

On 9/9/25 3:42 PM, Ryan Eatmon wrote:
> 
> 
> On 9/9/2025 10:10 AM, Andrew Davis wrote:
>> The DRIVER and matching UMLIBS must always match, no need to define a
>> version number for both. Keep this simple and remove the extra defines.
>>
>> Signed-off-by: Andrew Davis <afd@ti.com>
>> ---
>>   meta-beagle/conf/machine/include/beagle-bsp.inc |  4 ----
>>   meta-ti-bsp/conf/machine/include/ti-bsp.inc     | 10 ++--------
>>   2 files changed, 2 insertions(+), 12 deletions(-)
>>
>> diff --git a/meta-beagle/conf/machine/include/beagle-bsp.inc b/meta-beagle/conf/machine/include/beagle-bsp.inc
>> index d180c297..448e2086 100644
>> --- a/meta-beagle/conf/machine/include/beagle-bsp.inc
>> +++ b/meta-beagle/conf/machine/include/beagle-bsp.inc
>> @@ -18,10 +18,8 @@ KERNEL_DEVICETREE_PREFIX:bsp-bb_org-6_12 = ""
>>   BSP_SGX_DRIVER_PROVIDER:bsp-bb_org-6_12 = "ti-sgx-ddk-km"
>>   BSP_SGX_DRIVER_VERSION:bsp-bb_org-6_12 = "1.17%"
>> -BSP_SGX_UMLIBS_VERSION:bsp-bb_org-6_12 = "1.17%"
>>   BSP_ROGUE_DRIVER_PROVIDER:bsp-bb_org-6_12 = "ti-img-rogue-driver"
>>   BSP_ROGUE_DRIVER_VERSION:bsp-bb_org-6_12 = "24%"
>> -BSP_ROGUE_UMLIBS_VERSION:bsp-bb_org-6_12 = "24%"
>>   BSP_MESA_PVR_VERSION:bsp-bb_org-6_12 = "24%"
>>   # GC320 support requires out-of-tree drivers not yet available in bb.org 6.12
>> @@ -41,10 +39,8 @@ KERNEL_DEVICETREE_PREFIX:bsp-bb_org-6_6 = ""
>>   BSP_SGX_DRIVER_PROVIDER:bsp-bb_org-6_6 = "ti-sgx-ddk-km"
>>   BSP_SGX_DRIVER_VERSION:bsp-bb_org-6_6 = "1.17%"
>> -BSP_SGX_UMLIBS_VERSION:bsp-bb_org-6_6 = "1.17%"
>>   BSP_ROGUE_DRIVER_PROVIDER:bsp-bb_org-6_6 = "ti-img-rogue-driver"
>>   BSP_ROGUE_DRIVER_VERSION:bsp-bb_org-6_6 = "24%"
>> -BSP_ROGUE_UMLIBS_VERSION:bsp-bb_org-6_6 = "24%"
>>   BSP_MESA_PVR_VERSION:bsp-bb_org-6_6 = "24%"
>>   # GC320 support requires out-of-tree drivers not yet available in bb.org 6.6
>> diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>> index a5d9a51f..dc0473de 100644
>> --- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>> +++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>> @@ -44,10 +44,8 @@ BSP_BOOTLOADER_VERSION:bsp-ti-6_12 = "2025.01%"
>>   BSP_SGX_DRIVER_PROVIDER:bsp-ti-6_12 = "ti-sgx-ddk-km"
>>   BSP_SGX_DRIVER_VERSION:bsp-ti-6_12 = "1.17%"
>> -BSP_SGX_UMLIBS_VERSION:bsp-ti-6_12 = "1.17%"
>>   BSP_ROGUE_DRIVER_PROVIDER:bsp-ti-6_12 = "ti-img-rogue-driver"
>>   BSP_ROGUE_DRIVER_VERSION:bsp-ti-6_12 = "24%"
>> -BSP_ROGUE_UMLIBS_VERSION:bsp-ti-6_12 = "24%"
>>   BSP_MESA_PVR_VERSION:bsp-ti-6_12 = "24%"
>>   # ==========
>> @@ -61,10 +59,8 @@ BSP_BOOTLOADER_VERSION:bsp-ti-6_6 = "2024.04%"
>>   BSP_SGX_DRIVER_PROVIDER:bsp-ti-6_6 = "ti-sgx-ddk-km"
>>   BSP_SGX_DRIVER_VERSION:bsp-ti-6_6 = "1.17%"
>> -BSP_SGX_UMLIBS_VERSION:bsp-ti-6_6 = "1.17%"
>>   BSP_ROGUE_DRIVER_PROVIDER:bsp-ti-6_6 = "ti-img-rogue-driver"
>>   BSP_ROGUE_DRIVER_VERSION:bsp-ti-6_6 = "24%"
>> -BSP_ROGUE_UMLIBS_VERSION:bsp-ti-6_6 = "24%"
>>   BSP_MESA_PVR_VERSION:bsp-ti-6_6 = "24%"
>>   # GC320 support requires out-of-tree drivers not yet available in 6.6
>> @@ -84,8 +80,6 @@ BSP_SGX_DRIVER_PROVIDER ?= ""
>>   BSP_SGX_DRIVER_VERSION ?= ""
>>   BSP_ROGUE_DRIVER_PROVIDER ?= ""
>>   BSP_ROGUE_DRIVER_VERSION ?= ""
>> -BSP_SGX_UMLIBS_VERSION ?= ""
>> -BSP_ROGUE_UMLIBS_VERSION ?= ""
>>   BSP_MESA_PVR_VERSION ?= ""
>>   # ==========
>> @@ -101,7 +95,7 @@ PREFERRED_VERSION_${BSP_BOOTLOADER_PROVIDER} ?= "${BSP_BOOTLOADER_VERSION}"
>>   # GPU provider gets set in machine configs, as some machines are headless
>>   # Select default preferred versions here
>>   PREFERRED_VERSION_ti-sgx-ddk-km ?= "${BSP_SGX_DRIVER_VERSION}"
>> -PREFERRED_VERSION_ti-sgx-ddk-um ?= "${BSP_SGX_UMLIBS_VERSION}"
>> +PREFERRED_VERSION_ti-sgx-ddk-um ?= "${BSP_SGX_DRIVER_VERSION}"
> 
> Since we are simplifying, do we want to take it further and rename the XXX_DRIVER_VERSION to just XXX_VERSION?
> 
> BSP_ROGUE_DRIVER_VERSION -> BSP_ROGUE_VERSION
> BSP_SGX_DRIVER_VERSION -> BSP_SGX_VERSION

My only worry would be to me this looks like it should contain the
version of the GPU itself. Including "driver" makes it clear this
is for the SW version. No huge preference either way.

Andrew

> 
>>   PREFERRED_VERSION_ti-img-rogue-driver ?= "${BSP_ROGUE_DRIVER_VERSION}"
>> -PREFERRED_VERSION_ti-img-rogue-umlibs ?= "${BSP_ROGUE_UMLIBS_VERSION}"
>> +PREFERRED_VERSION_ti-img-rogue-umlibs ?= "${BSP_ROGUE_DRIVER_VERSION}"
>>   PREFERRED_VERSION_mesa-pvr ?= "${BSP_MESA_PVR_VERSION}"
> 

Re: [meta-ti][scarthgap][RFC 2/6] ti-bsp: Do not remove gpu MACHINE_FEATURE conditionally on BSP version

[Andrew Davis]

On 9/9/25 3:44 PM, Denys Dmytriyenko wrote:
> On Tue, Sep 09, 2025 at 10:10:24AM -0500, Andrew Davis via lists.yoctoproject.org wrote:
>> Setting or removing the gpu MACHINE_FEATURE does not actually change
>> if the GPU driver is built or not. That is done based on the setting
>> of the BSP_{SGX,ROGUE}_{DRIVER,UMLIBS}_{PROVIDER,VERSION} vars. These
>> are default empty for BSP versions without supported GPU drivers, which
>> will correctly cause a fallback to SW rendering when the GPU driver is
>> not available.
> 
> While cleaning the remaining uses of "gpu" MACHINE_FEATURE, why not remove it
> completely from all the machine configs?
> 

I thought about that, but it could still be useful to key off of that for
other reasons, or even maybe turn the "gpu" MACHINE_FEATURE into "rgx" and
"sgx" so we can later switch on that in the mesa-pvr driver.

Andrew

> 
>> Signed-off-by: Andrew Davis <afd@ti.com>
>> ---
>>   meta-ti-bsp/conf/machine/include/ti-bsp.inc | 8 --------
>>   1 file changed, 8 deletions(-)
>>
>> diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>> index 262aa884..a5d9a51f 100644
>> --- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>> +++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>> @@ -18,10 +18,6 @@ BSP_KERNEL_VERSION:bsp-mainline = "%"
>>   BSP_BOOTLOADER_PROVIDER:bsp-mainline = "u-boot-ti-mainline"
>>   BSP_BOOTLOADER_VERSION:bsp-mainline = "%"
>>   
>> -# GPU support requires out-of-tree SGX and Rogue drivers not available
>> -# in mainline, usually present in TI staging or derivative like BB.org
>> -MACHINE_FEATURES:remove:bsp-mainline = "gpu"
>> -
>>   # GC320 support requires out-of-tree drivers not available in mainline
>>   MACHINE_FEATURES:remove:bsp-mainline = "gc320"
>>   
>> @@ -34,10 +30,6 @@ BSP_KERNEL_VERSION:bsp-next = "%"
>>   BSP_BOOTLOADER_PROVIDER:bsp-next = "u-boot-ti-next"
>>   BSP_BOOTLOADER_VERSION:bsp-next = "%"
>>   
>> -# GPU support requires out-of-tree SGX and Rogue drivers not available
>> -# in next, usually present in TI staging or derivative like BB.org
>> -MACHINE_FEATURES:remove:bsp-next = "gpu"
>> -
>>   # GC320 support requires out-of-tree drivers not available in next
>>   MACHINE_FEATURES:remove:bsp-next = "gc320"
>>   
>> -- 
>> 2.39.2

Re: [meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Andrew Davis]

On 9/9/25 3:52 PM, Randolph Sapp wrote:
> On Tue Sep 9, 2025 at 10:10 AM CDT, Andrew Davis wrote:
>> TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
>> SW rendering just like the normal mesa package. The benefit is using the
>> same version across all supported devices is more consistent and only
>> one version of Mesa is needed for all TI SoCs.
>>
>> Signed-off-by: Andrew Davis <afd@ti.com>
>> ---
> 
> [snip]
> 
> Originally this was done to mitigate damage in the event where there is an
> issue with mesa. I'm not entirely sure we want to move everything over to
> pvr-mesa as they still hijack some of the common dri code for their own goofy
> stuff.
> 

Fair point, would be nice if the PVR support didn't touch the core/common DRI
at all. Having this the common Mesa for all boards would help in identifying
if they broke anything in the common/sw rendering side though, right now we
would only find those bugs if one was both using a device with GPU *and* it
falls back to SW.

> My opinion is we should use vanilla software where we can. Especially
> considering we'll need to be able to access vanilla mesa for AM62 later if I get
> my way.
> 

Could you expand on that? Outside of the eventual switch to the upstream
PVR driver, what would we need vanilla mesa for on AM62?

Andrew

Re: [meta-ti][scarthgap][RFC 2/6] ti-bsp: Do not remove gpu MACHINE_FEATURE conditionally on BSP version

[Denys Dmytriyenko]

On Fri, Sep 12, 2025 at 11:31:19AM -0500, Andrew Davis wrote:
> On 9/9/25 3:44 PM, Denys Dmytriyenko wrote:
> >On Tue, Sep 09, 2025 at 10:10:24AM -0500, Andrew Davis via lists.yoctoproject.org wrote:
> >>Setting or removing the gpu MACHINE_FEATURE does not actually change
> >>if the GPU driver is built or not. That is done based on the setting
> >>of the BSP_{SGX,ROGUE}_{DRIVER,UMLIBS}_{PROVIDER,VERSION} vars. These
> >>are default empty for BSP versions without supported GPU drivers, which
> >>will correctly cause a fallback to SW rendering when the GPU driver is
> >>not available.
> >
> >While cleaning the remaining uses of "gpu" MACHINE_FEATURE, why not remove it
> >completely from all the machine configs?
> >
> 
> I thought about that, but it could still be useful to key off of that for
> other reasons, or even maybe turn the "gpu" MACHINE_FEATURE into "rgx" and
> "sgx" so we can later switch on that in the mesa-pvr driver.

Heh, this MACHINE_FEATURE used to be called "sgx" back in the day. When Rogue 
was being added, I briefly had a separate "rgx" flag, but eventually combined 
both into "gpu" one...


> >>Signed-off-by: Andrew Davis <afd@ti.com>
> >>---
> >>  meta-ti-bsp/conf/machine/include/ti-bsp.inc | 8 --------
> >>  1 file changed, 8 deletions(-)
> >>
> >>diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> >>index 262aa884..a5d9a51f 100644
> >>--- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> >>+++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
> >>@@ -18,10 +18,6 @@ BSP_KERNEL_VERSION:bsp-mainline = "%"
> >>  BSP_BOOTLOADER_PROVIDER:bsp-mainline = "u-boot-ti-mainline"
> >>  BSP_BOOTLOADER_VERSION:bsp-mainline = "%"
> >>-# GPU support requires out-of-tree SGX and Rogue drivers not available
> >>-# in mainline, usually present in TI staging or derivative like BB.org
> >>-MACHINE_FEATURES:remove:bsp-mainline = "gpu"
> >>-
> >>  # GC320 support requires out-of-tree drivers not available in mainline
> >>  MACHINE_FEATURES:remove:bsp-mainline = "gc320"
> >>@@ -34,10 +30,6 @@ BSP_KERNEL_VERSION:bsp-next = "%"
> >>  BSP_BOOTLOADER_PROVIDER:bsp-next = "u-boot-ti-next"
> >>  BSP_BOOTLOADER_VERSION:bsp-next = "%"
> >>-# GPU support requires out-of-tree SGX and Rogue drivers not available
> >>-# in next, usually present in TI staging or derivative like BB.org
> >>-MACHINE_FEATURES:remove:bsp-next = "gpu"
> >>-
> >>  # GC320 support requires out-of-tree drivers not available in next
> >>  MACHINE_FEATURES:remove:bsp-next = "gc320"
> >>-- 
> >>2.39.2

Re: [meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Randolph Sapp]

On Fri Sep 12, 2025 at 11:38 AM CDT, Andrew Davis wrote:
> On 9/9/25 3:52 PM, Randolph Sapp wrote:
>> On Tue Sep 9, 2025 at 10:10 AM CDT, Andrew Davis wrote:
>>> TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
>>> SW rendering just like the normal mesa package. The benefit is using the
>>> same version across all supported devices is more consistent and only
>>> one version of Mesa is needed for all TI SoCs.
>>>
>>> Signed-off-by: Andrew Davis <afd@ti.com>
>>> ---
>> 
>> [snip]
>> 
>> Originally this was done to mitigate damage in the event where there is an
>> issue with mesa. I'm not entirely sure we want to move everything over to
>> pvr-mesa as they still hijack some of the common dri code for their own goofy
>> stuff.
>> 
>
> Fair point, would be nice if the PVR support didn't touch the core/common DRI
> at all. Having this the common Mesa for all boards would help in identifying
> if they broke anything in the common/sw rendering side though, right now we
> would only find those bugs if one was both using a device with GPU *and* it
> falls back to SW.
>
>> My opinion is we should use vanilla software where we can. Especially
>> considering we'll need to be able to access vanilla mesa for AM62 later if I get
>> my way.
>> 
>
> Could you expand on that? Outside of the eventual switch to the upstream
> PVR driver, what would we need vanilla mesa for on AM62?
>
> Andrew

Ah, not much to expand on really. Upstream driver needs vanilla mesa, and I'd
like to avoid shipping hacked libraries on devices that don't need em.

- Randolph

Re: [meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Andrew Davis]

On 9/17/25 1:22 PM, Randolph Sapp wrote:
> On Fri Sep 12, 2025 at 11:38 AM CDT, Andrew Davis wrote:
>> On 9/9/25 3:52 PM, Randolph Sapp wrote:
>>> On Tue Sep 9, 2025 at 10:10 AM CDT, Andrew Davis wrote:
>>>> TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
>>>> SW rendering just like the normal mesa package. The benefit is using the
>>>> same version across all supported devices is more consistent and only
>>>> one version of Mesa is needed for all TI SoCs.
>>>>
>>>> Signed-off-by: Andrew Davis <afd@ti.com>
>>>> ---
>>>
>>> [snip]
>>>
>>> Originally this was done to mitigate damage in the event where there is an
>>> issue with mesa. I'm not entirely sure we want to move everything over to
>>> pvr-mesa as they still hijack some of the common dri code for their own goofy
>>> stuff.
>>>
>>
>> Fair point, would be nice if the PVR support didn't touch the core/common DRI
>> at all. Having this the common Mesa for all boards would help in identifying
>> if they broke anything in the common/sw rendering side though, right now we
>> would only find those bugs if one was both using a device with GPU *and* it
>> falls back to SW.
>>
>>> My opinion is we should use vanilla software where we can. Especially
>>> considering we'll need to be able to access vanilla mesa for AM62 later if I get
>>> my way.
>>>
>>
>> Could you expand on that? Outside of the eventual switch to the upstream
>> PVR driver, what would we need vanilla mesa for on AM62?
>>
>> Andrew
> 
> Ah, not much to expand on really. Upstream driver needs vanilla mesa, and I'd
> like to avoid shipping hacked libraries on devices that don't need em.
> 
> - Randolph

What is the timeline looking like for that? This series is for scarthgap and
I'd guess the upstream driver won't be usable in time for the next LTS either.
We can always switch back to vanilla when it can be used on these devices.

Andrew

Re: [meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Randolph Sapp]

On Wed Sep 17, 2025 at 1:40 PM CDT, Andrew Davis wrote:
> On 9/17/25 1:22 PM, Randolph Sapp wrote:
>> On Fri Sep 12, 2025 at 11:38 AM CDT, Andrew Davis wrote:
>>> On 9/9/25 3:52 PM, Randolph Sapp wrote:
>>>> On Tue Sep 9, 2025 at 10:10 AM CDT, Andrew Davis wrote:
>>>>> TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
>>>>> SW rendering just like the normal mesa package. The benefit is using the
>>>>> same version across all supported devices is more consistent and only
>>>>> one version of Mesa is needed for all TI SoCs.
>>>>>
>>>>> Signed-off-by: Andrew Davis <afd@ti.com>
>>>>> ---
>>>>
>>>> [snip]
>>>>
>>>> Originally this was done to mitigate damage in the event where there is an
>>>> issue with mesa. I'm not entirely sure we want to move everything over to
>>>> pvr-mesa as they still hijack some of the common dri code for their own goofy
>>>> stuff.
>>>>
>>>
>>> Fair point, would be nice if the PVR support didn't touch the core/common DRI
>>> at all. Having this the common Mesa for all boards would help in identifying
>>> if they broke anything in the common/sw rendering side though, right now we
>>> would only find those bugs if one was both using a device with GPU *and* it
>>> falls back to SW.
>>>
>>>> My opinion is we should use vanilla software where we can. Especially
>>>> considering we'll need to be able to access vanilla mesa for AM62 later if I get
>>>> my way.
>>>>
>>>
>>> Could you expand on that? Outside of the eventual switch to the upstream
>>> PVR driver, what would we need vanilla mesa for on AM62?
>>>
>>> Andrew
>> 
>> Ah, not much to expand on really. Upstream driver needs vanilla mesa, and I'd
>> like to avoid shipping hacked libraries on devices that don't need em.
>> 
>> - Randolph
>
> What is the timeline looking like for that? This series is for scarthgap and
> I'd guess the upstream driver won't be usable in time for the next LTS either.
> We can always switch back to vanilla when it can be used on these devices.
>
> Andrew

That's still ignoring my last comment. I don't think this is necessary, and
bringing up timelines to apply pressure here isn't useful. Merging an
unnecessary patch now just to remove it later is goofy.

- Randolph

Re: [meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Andrew Davis]

On 9/17/25 2:12 PM, Randolph Sapp wrote:
> On Wed Sep 17, 2025 at 1:40 PM CDT, Andrew Davis wrote:
>> On 9/17/25 1:22 PM, Randolph Sapp wrote:
>>> On Fri Sep 12, 2025 at 11:38 AM CDT, Andrew Davis wrote:
>>>> On 9/9/25 3:52 PM, Randolph Sapp wrote:
>>>>> On Tue Sep 9, 2025 at 10:10 AM CDT, Andrew Davis wrote:
>>>>>> TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
>>>>>> SW rendering just like the normal mesa package. The benefit is using the
>>>>>> same version across all supported devices is more consistent and only
>>>>>> one version of Mesa is needed for all TI SoCs.
>>>>>>
>>>>>> Signed-off-by: Andrew Davis <afd@ti.com>
>>>>>> ---
>>>>>
>>>>> [snip]
>>>>>
>>>>> Originally this was done to mitigate damage in the event where there is an
>>>>> issue with mesa. I'm not entirely sure we want to move everything over to
>>>>> pvr-mesa as they still hijack some of the common dri code for their own goofy
>>>>> stuff.
>>>>>
>>>>
>>>> Fair point, would be nice if the PVR support didn't touch the core/common DRI
>>>> at all. Having this the common Mesa for all boards would help in identifying
>>>> if they broke anything in the common/sw rendering side though, right now we
>>>> would only find those bugs if one was both using a device with GPU *and* it
>>>> falls back to SW.
>>>>
>>>>> My opinion is we should use vanilla software where we can. Especially
>>>>> considering we'll need to be able to access vanilla mesa for AM62 later if I get
>>>>> my way.
>>>>>
>>>>
>>>> Could you expand on that? Outside of the eventual switch to the upstream
>>>> PVR driver, what would we need vanilla mesa for on AM62?
>>>>
>>>> Andrew
>>>
>>> Ah, not much to expand on really. Upstream driver needs vanilla mesa, and I'd
>>> like to avoid shipping hacked libraries on devices that don't need em.
>>>
>>> - Randolph
>>
>> What is the timeline looking like for that? This series is for scarthgap and
>> I'd guess the upstream driver won't be usable in time for the next LTS either.
>> We can always switch back to vanilla when it can be used on these devices.
>>
>> Andrew
> 
> That's still ignoring my last comment. I don't think this is necessary, and
> bringing up timelines to apply pressure here isn't useful. Merging an
> unnecessary patch now just to remove it later is goofy.
> 
> - Randolph

Oh I wasn't trying to pressure you, in fact this patch was meant to
help you out specifically, if you don't want it then I'll drop it
from v2, no worries :)

Andrew

Re: [meta-ti][scarthgap][RFC 6/6] mesa-pvr: Use mesa-pvr for all TI SoCs

[Randolph Sapp]

On Wed Sep 17, 2025 at 2:15 PM CDT, Andrew Davis wrote:
> On 9/17/25 2:12 PM, Randolph Sapp wrote:
>> On Wed Sep 17, 2025 at 1:40 PM CDT, Andrew Davis wrote:
>>> On 9/17/25 1:22 PM, Randolph Sapp wrote:
>>>> On Fri Sep 12, 2025 at 11:38 AM CDT, Andrew Davis wrote:
>>>>> On 9/9/25 3:52 PM, Randolph Sapp wrote:
>>>>>> On Tue Sep 9, 2025 at 10:10 AM CDT, Andrew Davis wrote:
>>>>>>> TI SoCs without GPUs can still use mesa-pvr as it will simply fallback to
>>>>>>> SW rendering just like the normal mesa package. The benefit is using the
>>>>>>> same version across all supported devices is more consistent and only
>>>>>>> one version of Mesa is needed for all TI SoCs.
>>>>>>>
>>>>>>> Signed-off-by: Andrew Davis <afd@ti.com>
>>>>>>> ---
>>>>>>
>>>>>> [snip]
>>>>>>
>>>>>> Originally this was done to mitigate damage in the event where there is an
>>>>>> issue with mesa. I'm not entirely sure we want to move everything over to
>>>>>> pvr-mesa as they still hijack some of the common dri code for their own goofy
>>>>>> stuff.
>>>>>>
>>>>>
>>>>> Fair point, would be nice if the PVR support didn't touch the core/common DRI
>>>>> at all. Having this the common Mesa for all boards would help in identifying
>>>>> if they broke anything in the common/sw rendering side though, right now we
>>>>> would only find those bugs if one was both using a device with GPU *and* it
>>>>> falls back to SW.
>>>>>
>>>>>> My opinion is we should use vanilla software where we can. Especially
>>>>>> considering we'll need to be able to access vanilla mesa for AM62 later if I get
>>>>>> my way.
>>>>>>
>>>>>
>>>>> Could you expand on that? Outside of the eventual switch to the upstream
>>>>> PVR driver, what would we need vanilla mesa for on AM62?
>>>>>
>>>>> Andrew
>>>>
>>>> Ah, not much to expand on really. Upstream driver needs vanilla mesa, and I'd
>>>> like to avoid shipping hacked libraries on devices that don't need em.
>>>>
>>>> - Randolph
>>>
>>> What is the timeline looking like for that? This series is for scarthgap and
>>> I'd guess the upstream driver won't be usable in time for the next LTS either.
>>> We can always switch back to vanilla when it can be used on these devices.
>>>
>>> Andrew
>> 
>> That's still ignoring my last comment. I don't think this is necessary, and
>> bringing up timelines to apply pressure here isn't useful. Merging an
>> unnecessary patch now just to remove it later is goofy.
>> 
>> - Randolph
>
> Oh I wasn't trying to pressure you, in fact this patch was meant to
> help you out specifically, if you don't want it then I'll drop it
> from v2, no worries :)
>
> Andrew

Sorry if that was a bit aggressive. First day back and things are already
getting tense. Fun times.

- Randolph

Re: [meta-ti][scarthgap][RFC 2/6] ti-bsp: Do not remove gpu MACHINE_FEATURE conditionally on BSP version

[Andrew Davis]

On 9/12/25 1:24 PM, Denys Dmytriyenko wrote:
> On Fri, Sep 12, 2025 at 11:31:19AM -0500, Andrew Davis wrote:
>> On 9/9/25 3:44 PM, Denys Dmytriyenko wrote:
>>> On Tue, Sep 09, 2025 at 10:10:24AM -0500, Andrew Davis via lists.yoctoproject.org wrote:
>>>> Setting or removing the gpu MACHINE_FEATURE does not actually change
>>>> if the GPU driver is built or not. That is done based on the setting
>>>> of the BSP_{SGX,ROGUE}_{DRIVER,UMLIBS}_{PROVIDER,VERSION} vars. These
>>>> are default empty for BSP versions without supported GPU drivers, which
>>>> will correctly cause a fallback to SW rendering when the GPU driver is
>>>> not available.
>>>
>>> While cleaning the remaining uses of "gpu" MACHINE_FEATURE, why not remove it
>>> completely from all the machine configs?
>>>
>>
>> I thought about that, but it could still be useful to key off of that for
>> other reasons, or even maybe turn the "gpu" MACHINE_FEATURE into "rgx" and
>> "sgx" so we can later switch on that in the mesa-pvr driver.
> 
> Heh, this MACHINE_FEATURE used to be called "sgx" back in the day. When Rogue
> was being added, I briefly had a separate "rgx" flag, but eventually combined
> both into "gpu" one...
> 

Okay so then I have two options, I can switch back to "sgx" and "rgx" for
some later use, or I can just drop the MACHINE_FEATURE completely, any
preference?

> 
>>>> Signed-off-by: Andrew Davis <afd@ti.com>
>>>> ---
>>>>   meta-ti-bsp/conf/machine/include/ti-bsp.inc | 8 --------
>>>>   1 file changed, 8 deletions(-)
>>>>
>>>> diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>>>> index 262aa884..a5d9a51f 100644
>>>> --- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>>>> +++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>>>> @@ -18,10 +18,6 @@ BSP_KERNEL_VERSION:bsp-mainline = "%"
>>>>   BSP_BOOTLOADER_PROVIDER:bsp-mainline = "u-boot-ti-mainline"
>>>>   BSP_BOOTLOADER_VERSION:bsp-mainline = "%"
>>>> -# GPU support requires out-of-tree SGX and Rogue drivers not available
>>>> -# in mainline, usually present in TI staging or derivative like BB.org
>>>> -MACHINE_FEATURES:remove:bsp-mainline = "gpu"
>>>> -
>>>>   # GC320 support requires out-of-tree drivers not available in mainline
>>>>   MACHINE_FEATURES:remove:bsp-mainline = "gc320"
>>>> @@ -34,10 +30,6 @@ BSP_KERNEL_VERSION:bsp-next = "%"
>>>>   BSP_BOOTLOADER_PROVIDER:bsp-next = "u-boot-ti-next"
>>>>   BSP_BOOTLOADER_VERSION:bsp-next = "%"
>>>> -# GPU support requires out-of-tree SGX and Rogue drivers not available
>>>> -# in next, usually present in TI staging or derivative like BB.org
>>>> -MACHINE_FEATURES:remove:bsp-next = "gpu"
>>>> -
>>>>   # GC320 support requires out-of-tree drivers not available in next
>>>>   MACHINE_FEATURES:remove:bsp-next = "gc320"
>>>> -- 
>>>> 2.39.2

Re: [meta-ti][scarthgap][RFC 2/6] ti-bsp: Do not remove gpu MACHINE_FEATURE conditionally on BSP version

[Randolph Sapp]

On Wed Sep 17, 2025 at 4:54 PM CDT, Andrew Davis wrote:
> On 9/12/25 1:24 PM, Denys Dmytriyenko wrote:
>> On Fri, Sep 12, 2025 at 11:31:19AM -0500, Andrew Davis wrote:
>>> On 9/9/25 3:44 PM, Denys Dmytriyenko wrote:
>>>> On Tue, Sep 09, 2025 at 10:10:24AM -0500, Andrew Davis via lists.yoctoproject.org wrote:
>>>>> Setting or removing the gpu MACHINE_FEATURE does not actually change
>>>>> if the GPU driver is built or not. That is done based on the setting
>>>>> of the BSP_{SGX,ROGUE}_{DRIVER,UMLIBS}_{PROVIDER,VERSION} vars. These
>>>>> are default empty for BSP versions without supported GPU drivers, which
>>>>> will correctly cause a fallback to SW rendering when the GPU driver is
>>>>> not available.
>>>>
>>>> While cleaning the remaining uses of "gpu" MACHINE_FEATURE, why not remove it
>>>> completely from all the machine configs?
>>>>
>>>
>>> I thought about that, but it could still be useful to key off of that for
>>> other reasons, or even maybe turn the "gpu" MACHINE_FEATURE into "rgx" and
>>> "sgx" so we can later switch on that in the mesa-pvr driver.
>> 
>> Heh, this MACHINE_FEATURE used to be called "sgx" back in the day. When Rogue
>> was being added, I briefly had a separate "rgx" flag, but eventually combined
>> both into "gpu" one...
>> 
>
> Okay so then I have two options, I can switch back to "sgx" and "rgx" for
> some later use, or I can just drop the MACHINE_FEATURE completely, any
> preference?

I'm partial to just dropping it entirely. If we don't intend external layers to
read it, then it shouldn't be a MACHINE_FEATURE. We're already tracking that
information in other variables anyway.

- Randolph

>>>>> Signed-off-by: Andrew Davis <afd@ti.com>
>>>>> ---
>>>>>   meta-ti-bsp/conf/machine/include/ti-bsp.inc | 8 --------
>>>>>   1 file changed, 8 deletions(-)
>>>>>
>>>>> diff --git a/meta-ti-bsp/conf/machine/include/ti-bsp.inc b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>>>>> index 262aa884..a5d9a51f 100644
>>>>> --- a/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>>>>> +++ b/meta-ti-bsp/conf/machine/include/ti-bsp.inc
>>>>> @@ -18,10 +18,6 @@ BSP_KERNEL_VERSION:bsp-mainline = "%"
>>>>>   BSP_BOOTLOADER_PROVIDER:bsp-mainline = "u-boot-ti-mainline"
>>>>>   BSP_BOOTLOADER_VERSION:bsp-mainline = "%"
>>>>> -# GPU support requires out-of-tree SGX and Rogue drivers not available
>>>>> -# in mainline, usually present in TI staging or derivative like BB.org
>>>>> -MACHINE_FEATURES:remove:bsp-mainline = "gpu"
>>>>> -
>>>>>   # GC320 support requires out-of-tree drivers not available in mainline
>>>>>   MACHINE_FEATURES:remove:bsp-mainline = "gc320"
>>>>> @@ -34,10 +30,6 @@ BSP_KERNEL_VERSION:bsp-next = "%"
>>>>>   BSP_BOOTLOADER_PROVIDER:bsp-next = "u-boot-ti-next"
>>>>>   BSP_BOOTLOADER_VERSION:bsp-next = "%"
>>>>> -# GPU support requires out-of-tree SGX and Rogue drivers not available
>>>>> -# in next, usually present in TI staging or derivative like BB.org
>>>>> -MACHINE_FEATURES:remove:bsp-next = "gpu"
>>>>> -
>>>>>   # GC320 support requires out-of-tree drivers not available in next
>>>>>   MACHINE_FEATURES:remove:bsp-next = "gc320"
>>>>> -- 
>>>>> 2.39.2