| Message ID | 20221228163102.468-1-mario.limonciello@amd.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Recover from failure to probe GPU | expand |
Patches 1-10 are: Reviewed-by: Alex Deucher <alexander.deucher@amd.com> On Wed, Dec 28, 2022 at 11:31 AM Mario Limonciello <mario.limonciello@amd.com> wrote: > > One of the first thing that KMS drivers do during initialization is > destroy the system firmware framebuffer by means of > `drm_aperture_remove_conflicting_pci_framebuffers` > > This means that if for any reason the GPU failed to probe the user > will be stuck with at best a screen frozen at the last thing that > was shown before the KMS driver continued it's probe. > > The problem is most pronounced when new GPU support is introduced > because users will need to have a recent linux-firmware snapshot > on their system when they boot a kernel with matching support. > > However the problem is further exaggerated in the case of amdgpu because > it has migrated to "IP discovery" where amdgpu will attempt to load > on "ALL" AMD GPUs even if the driver is missing support for IP blocks > contained in that GPU. > > IP discovery requires some probing and isn't run until after the > framebuffer has been destroyed. > > This means a situation can occur where a user purchases a new GPU not > yet supported by a distribution and when booting the installer it will > "freeze" even if the distribution doesn't have the matching kernel support > for those IP blocks. > > The perfect example of this is Ubuntu 22.10 and the new dGPUs just > launched by AMD. The installation media ships with kernel 5.19 (which > has IP discovery) but the amdgpu support for those IP blocks landed in > kernel 6.0. The matching linux-firmware was released after 22.10's launch. > The screen will freeze without nomodeset. Even if a user manages to install > and then upgrades to kernel 6.0 after install they'll still have the > problem of missing firmware, and the same experience. > > This is quite jarring for users, particularly if they don't know > that they have to use "nomodeset" to install. > > To help the situation make changes to GPU discovery: > 1) Delay releasing the firmware framebuffer until after IP discovery has > completed. This will help the situation of an older kernel that doesn't > yet support the IP blocks probing a new GPU. > 2) Request loading all PSP, VCN, SDMA, MES and GC microcode into memory > during IP discovery. This will help the situation of new enough kernel for > the IP discovery phase to otherwise pass but missing microcode from > linux-firmware.git. > > Not all requested firmware will be loaded during IP discovery as some of it > will require larger driver architecture changes. For example SMU firmware > isn't loaded on certain products, but that's not known until later on when > the early_init phase of the SMU load occurs. > > v1->v2: > * Take the suggestion from v1 thread to delay the framebuffer release until > ip discovery is done. This patch is CC to stable to that older stable > kernels with IP discovery won't try to probe unknown IP. > * Drop changes to drm aperature. > * Fetch SDMA, VCN, MES, GC and PSP microcode during IP discovery. > > Mario Limonciello (11): > drm/amd: Delay removal of the firmware framebuffer > drm/amd: Add a legacy mapping to "amdgpu_ucode_ip_version_decode" > drm/amd: Convert SMUv11 microcode init to use > `amdgpu_ucode_ip_version_decode` > drm/amd: Convert SMU v13 to use `amdgpu_ucode_ip_version_decode` > drm/amd: Request SDMA microcode during IP discovery > drm/amd: Request VCN microcode during IP discovery > drm/amd: Request MES microcode during IP discovery > drm/amd: Request GFX9 microcode during IP discovery > drm/amd: Request GFX10 microcode during IP discovery > drm/amd: Request GFX11 microcode during IP discovery > drm/amd: Request PSP microcode during IP discovery > > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 + > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 590 +++++++++++++++++- > drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 6 - > drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 2 - > drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c | 9 +- > drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +- > drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c | 208 ++++++ > drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 85 +-- > drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 180 +----- > drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 64 +- > drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 143 +---- > drivers/gpu/drm/amd/amdgpu/mes_v10_1.c | 28 - > drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 25 +- > drivers/gpu/drm/amd/amdgpu/psp_v10_0.c | 106 +--- > drivers/gpu/drm/amd/amdgpu/psp_v11_0.c | 165 +---- > drivers/gpu/drm/amd/amdgpu/psp_v12_0.c | 102 +-- > drivers/gpu/drm/amd/amdgpu/psp_v13_0.c | 82 --- > drivers/gpu/drm/amd/amdgpu/psp_v13_0_4.c | 36 -- > drivers/gpu/drm/amd/amdgpu/psp_v3_1.c | 36 -- > drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 61 +- > drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 42 +- > drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 65 +- > drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c | 30 +- > .../gpu/drm/amd/pm/swsmu/smu11/smu_v11_0.c | 35 +- > .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 12 +- > 25 files changed, 919 insertions(+), 1203 deletions(-) > > > base-commit: de9a71e391a92841582ca3008e7b127a0b8ccf41 > -- > 2.34.1 >
On 12/28/2022 10:00 PM, Mario Limonciello wrote: > One of the first thing that KMS drivers do during initialization is > destroy the system firmware framebuffer by means of > `drm_aperture_remove_conflicting_pci_framebuffers` > > This means that if for any reason the GPU failed to probe the user > will be stuck with at best a screen frozen at the last thing that > was shown before the KMS driver continued it's probe. > > The problem is most pronounced when new GPU support is introduced > because users will need to have a recent linux-firmware snapshot > on their system when they boot a kernel with matching support. > > However the problem is further exaggerated in the case of amdgpu because > it has migrated to "IP discovery" where amdgpu will attempt to load > on "ALL" AMD GPUs even if the driver is missing support for IP blocks > contained in that GPU. > > IP discovery requires some probing and isn't run until after the > framebuffer has been destroyed. > > This means a situation can occur where a user purchases a new GPU not > yet supported by a distribution and when booting the installer it will > "freeze" even if the distribution doesn't have the matching kernel support > for those IP blocks. > > The perfect example of this is Ubuntu 22.10 and the new dGPUs just > launched by AMD. The installation media ships with kernel 5.19 (which > has IP discovery) but the amdgpu support for those IP blocks landed in > kernel 6.0. The matching linux-firmware was released after 22.10's launch. > The screen will freeze without nomodeset. Even if a user manages to install > and then upgrades to kernel 6.0 after install they'll still have the > problem of missing firmware, and the same experience. > > This is quite jarring for users, particularly if they don't know > that they have to use "nomodeset" to install. > > To help the situation make changes to GPU discovery: > 1) Delay releasing the firmware framebuffer until after IP discovery has > completed. This will help the situation of an older kernel that doesn't > yet support the IP blocks probing a new GPU. > 2) Request loading all PSP, VCN, SDMA, MES and GC microcode into memory > during IP discovery. This will help the situation of new enough kernel for > the IP discovery phase to otherwise pass but missing microcode from > linux-firmware.git. > > Not all requested firmware will be loaded during IP discovery as some of it > will require larger driver architecture changes. For example SMU firmware > isn't loaded on certain products, but that's not known until later on when > the early_init phase of the SMU load occurs. > > v1->v2: > * Take the suggestion from v1 thread to delay the framebuffer release until > ip discovery is done. This patch is CC to stable to that older stable > kernels with IP discovery won't try to probe unknown IP. > * Drop changes to drm aperature. > * Fetch SDMA, VCN, MES, GC and PSP microcode during IP discovery. > What is the gain here in just checking if firmware files are available? It can fail anywhere during sw_init and it's the same situation. Restricting IP FWs to IP specific files looks better to me than centralizing and creating interdependencies. Thanks, Lijo > Mario Limonciello (11): > drm/amd: Delay removal of the firmware framebuffer > drm/amd: Add a legacy mapping to "amdgpu_ucode_ip_version_decode" > drm/amd: Convert SMUv11 microcode init to use > `amdgpu_ucode_ip_version_decode` > drm/amd: Convert SMU v13 to use `amdgpu_ucode_ip_version_decode` > drm/amd: Request SDMA microcode during IP discovery > drm/amd: Request VCN microcode during IP discovery > drm/amd: Request MES microcode during IP discovery > drm/amd: Request GFX9 microcode during IP discovery > drm/amd: Request GFX10 microcode during IP discovery > drm/amd: Request GFX11 microcode during IP discovery > drm/amd: Request PSP microcode during IP discovery > > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 + > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 590 +++++++++++++++++- > drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 6 - > drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 2 - > drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c | 9 +- > drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +- > drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c | 208 ++++++ > drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 85 +-- > drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 180 +----- > drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 64 +- > drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 143 +---- > drivers/gpu/drm/amd/amdgpu/mes_v10_1.c | 28 - > drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 25 +- > drivers/gpu/drm/amd/amdgpu/psp_v10_0.c | 106 +--- > drivers/gpu/drm/amd/amdgpu/psp_v11_0.c | 165 +---- > drivers/gpu/drm/amd/amdgpu/psp_v12_0.c | 102 +-- > drivers/gpu/drm/amd/amdgpu/psp_v13_0.c | 82 --- > drivers/gpu/drm/amd/amdgpu/psp_v13_0_4.c | 36 -- > drivers/gpu/drm/amd/amdgpu/psp_v3_1.c | 36 -- > drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 61 +- > drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 42 +- > drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 65 +- > drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c | 30 +- > .../gpu/drm/amd/pm/swsmu/smu11/smu_v11_0.c | 35 +- > .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 12 +- > 25 files changed, 919 insertions(+), 1203 deletions(-) > > > base-commit: de9a71e391a92841582ca3008e7b127a0b8ccf41
On Tue, Jan 3, 2023 at 5:10 AM Lazar, Lijo <lijo.lazar@amd.com> wrote: > > > > On 12/28/2022 10:00 PM, Mario Limonciello wrote: > > One of the first thing that KMS drivers do during initialization is > > destroy the system firmware framebuffer by means of > > `drm_aperture_remove_conflicting_pci_framebuffers` > > > > This means that if for any reason the GPU failed to probe the user > > will be stuck with at best a screen frozen at the last thing that > > was shown before the KMS driver continued it's probe. > > > > The problem is most pronounced when new GPU support is introduced > > because users will need to have a recent linux-firmware snapshot > > on their system when they boot a kernel with matching support. > > > > However the problem is further exaggerated in the case of amdgpu because > > it has migrated to "IP discovery" where amdgpu will attempt to load > > on "ALL" AMD GPUs even if the driver is missing support for IP blocks > > contained in that GPU. > > > > IP discovery requires some probing and isn't run until after the > > framebuffer has been destroyed. > > > > This means a situation can occur where a user purchases a new GPU not > > yet supported by a distribution and when booting the installer it will > > "freeze" even if the distribution doesn't have the matching kernel support > > for those IP blocks. > > > > The perfect example of this is Ubuntu 22.10 and the new dGPUs just > > launched by AMD. The installation media ships with kernel 5.19 (which > > has IP discovery) but the amdgpu support for those IP blocks landed in > > kernel 6.0. The matching linux-firmware was released after 22.10's launch. > > The screen will freeze without nomodeset. Even if a user manages to install > > and then upgrades to kernel 6.0 after install they'll still have the > > problem of missing firmware, and the same experience. > > > > This is quite jarring for users, particularly if they don't know > > that they have to use "nomodeset" to install. > > > > To help the situation make changes to GPU discovery: > > 1) Delay releasing the firmware framebuffer until after IP discovery has > > completed. This will help the situation of an older kernel that doesn't > > yet support the IP blocks probing a new GPU. > > 2) Request loading all PSP, VCN, SDMA, MES and GC microcode into memory > > during IP discovery. This will help the situation of new enough kernel for > > the IP discovery phase to otherwise pass but missing microcode from > > linux-firmware.git. > > > > Not all requested firmware will be loaded during IP discovery as some of it > > will require larger driver architecture changes. For example SMU firmware > > isn't loaded on certain products, but that's not known until later on when > > the early_init phase of the SMU load occurs. > > > > v1->v2: > > * Take the suggestion from v1 thread to delay the framebuffer release until > > ip discovery is done. This patch is CC to stable to that older stable > > kernels with IP discovery won't try to probe unknown IP. > > * Drop changes to drm aperature. > > * Fetch SDMA, VCN, MES, GC and PSP microcode during IP discovery. > > > > What is the gain here in just checking if firmware files are available? > It can fail anywhere during sw_init and it's the same situation. Other failures are presumably a bug or hardware issue. The missing firmware would be a common issue when chips are first launched. Thinking about it a bit more, another option might be to move the calls to request_firmware() into the IP specific early_init() functions and then move the drm_aperture release after early_init(). That would keep the firmware handling in the IPs and should still happen early enough that we haven't messed with the hardware yet. Alex > > Restricting IP FWs to IP specific files looks better to me than > centralizing and creating interdependencies. > > Thanks, > Lijo > > > Mario Limonciello (11): > > drm/amd: Delay removal of the firmware framebuffer > > drm/amd: Add a legacy mapping to "amdgpu_ucode_ip_version_decode" > > drm/amd: Convert SMUv11 microcode init to use > > `amdgpu_ucode_ip_version_decode` > > drm/amd: Convert SMU v13 to use `amdgpu_ucode_ip_version_decode` > > drm/amd: Request SDMA microcode during IP discovery > > drm/amd: Request VCN microcode during IP discovery > > drm/amd: Request MES microcode during IP discovery > > drm/amd: Request GFX9 microcode during IP discovery > > drm/amd: Request GFX10 microcode during IP discovery > > drm/amd: Request GFX11 microcode during IP discovery > > drm/amd: Request PSP microcode during IP discovery > > > > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 + > > drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 590 +++++++++++++++++- > > drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 6 - > > drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 2 - > > drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c | 9 +- > > drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +- > > drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c | 208 ++++++ > > drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 85 +-- > > drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 180 +----- > > drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 64 +- > > drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 143 +---- > > drivers/gpu/drm/amd/amdgpu/mes_v10_1.c | 28 - > > drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 25 +- > > drivers/gpu/drm/amd/amdgpu/psp_v10_0.c | 106 +--- > > drivers/gpu/drm/amd/amdgpu/psp_v11_0.c | 165 +---- > > drivers/gpu/drm/amd/amdgpu/psp_v12_0.c | 102 +-- > > drivers/gpu/drm/amd/amdgpu/psp_v13_0.c | 82 --- > > drivers/gpu/drm/amd/amdgpu/psp_v13_0_4.c | 36 -- > > drivers/gpu/drm/amd/amdgpu/psp_v3_1.c | 36 -- > > drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 61 +- > > drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 42 +- > > drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 65 +- > > drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c | 30 +- > > .../gpu/drm/amd/pm/swsmu/smu11/smu_v11_0.c | 35 +- > > .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 12 +- > > 25 files changed, 919 insertions(+), 1203 deletions(-) > > > > > > base-commit: de9a71e391a92841582ca3008e7b127a0b8ccf41
One of the first thing that KMS drivers do during initialization is destroy the system firmware framebuffer by means of `drm_aperture_remove_conflicting_pci_framebuffers` This means that if for any reason the GPU failed to probe the user will be stuck with at best a screen frozen at the last thing that was shown before the KMS driver continued it's probe. The problem is most pronounced when new GPU support is introduced because users will need to have a recent linux-firmware snapshot on their system when they boot a kernel with matching support. However the problem is further exaggerated in the case of amdgpu because it has migrated to "IP discovery" where amdgpu will attempt to load on "ALL" AMD GPUs even if the driver is missing support for IP blocks contained in that GPU. IP discovery requires some probing and isn't run until after the framebuffer has been destroyed. This means a situation can occur where a user purchases a new GPU not yet supported by a distribution and when booting the installer it will "freeze" even if the distribution doesn't have the matching kernel support for those IP blocks. The perfect example of this is Ubuntu 22.10 and the new dGPUs just launched by AMD. The installation media ships with kernel 5.19 (which has IP discovery) but the amdgpu support for those IP blocks landed in kernel 6.0. The matching linux-firmware was released after 22.10's launch. The screen will freeze without nomodeset. Even if a user manages to install and then upgrades to kernel 6.0 after install they'll still have the problem of missing firmware, and the same experience. This is quite jarring for users, particularly if they don't know that they have to use "nomodeset" to install. To help the situation make changes to GPU discovery: 1) Delay releasing the firmware framebuffer until after IP discovery has completed. This will help the situation of an older kernel that doesn't yet support the IP blocks probing a new GPU. 2) Request loading all PSP, VCN, SDMA, MES and GC microcode into memory during IP discovery. This will help the situation of new enough kernel for the IP discovery phase to otherwise pass but missing microcode from linux-firmware.git. Not all requested firmware will be loaded during IP discovery as some of it will require larger driver architecture changes. For example SMU firmware isn't loaded on certain products, but that's not known until later on when the early_init phase of the SMU load occurs. v1->v2: * Take the suggestion from v1 thread to delay the framebuffer release until ip discovery is done. This patch is CC to stable to that older stable kernels with IP discovery won't try to probe unknown IP. * Drop changes to drm aperature. * Fetch SDMA, VCN, MES, GC and PSP microcode during IP discovery. Mario Limonciello (11): drm/amd: Delay removal of the firmware framebuffer drm/amd: Add a legacy mapping to "amdgpu_ucode_ip_version_decode" drm/amd: Convert SMUv11 microcode init to use `amdgpu_ucode_ip_version_decode` drm/amd: Convert SMU v13 to use `amdgpu_ucode_ip_version_decode` drm/amd: Request SDMA microcode during IP discovery drm/amd: Request VCN microcode during IP discovery drm/amd: Request MES microcode during IP discovery drm/amd: Request GFX9 microcode during IP discovery drm/amd: Request GFX10 microcode during IP discovery drm/amd: Request GFX11 microcode during IP discovery drm/amd: Request PSP microcode during IP discovery drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 + drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 590 +++++++++++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 6 - drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 2 - drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c | 9 +- drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c | 208 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 85 +-- drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 180 +----- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 64 +- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 143 +---- drivers/gpu/drm/amd/amdgpu/mes_v10_1.c | 28 - drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 25 +- drivers/gpu/drm/amd/amdgpu/psp_v10_0.c | 106 +--- drivers/gpu/drm/amd/amdgpu/psp_v11_0.c | 165 +---- drivers/gpu/drm/amd/amdgpu/psp_v12_0.c | 102 +-- drivers/gpu/drm/amd/amdgpu/psp_v13_0.c | 82 --- drivers/gpu/drm/amd/amdgpu/psp_v13_0_4.c | 36 -- drivers/gpu/drm/amd/amdgpu/psp_v3_1.c | 36 -- drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 61 +- drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c | 42 +- drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 65 +- drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c | 30 +- .../gpu/drm/amd/pm/swsmu/smu11/smu_v11_0.c | 35 +- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 12 +- 25 files changed, 919 insertions(+), 1203 deletions(-) base-commit: de9a71e391a92841582ca3008e7b127a0b8ccf41