• State Resolved
  • Locked Locked
  • Replies 1 reply
  • Subscribers 98 subscribers
  • Views 4556 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Original question:

[FAQ] TDA4VM: Are there any known bugs and patches that I should use in my GPU driver?

[FAQ] TDA4VL-Q1: What are the GPU driver bug fixes for SDK 8.6 or earlier?

Erick Narvaez
Mastermind 36307 points

Part Number: TDA4VL-Q1

I am going into production on SDK 8.6 or earlier, but I have experienced a GPU driver bug. What is the latest GPU driver I can take to ensure I have all of the latest driver fixes and how do I install that driver into my system?

  • 0
    TI__Mastermind 36307 points

    Updated 7/25/2024

    The GPU driver experiences bugs, as seen by the related question. Further bugs are also found in testing as users go to production and identify issues within their GPU applications. Customers who wish to stay on an older released SDK will need to take the latest GPU driver with all the known bug fixes to ensure the issue they are facing is not already known.

    The GPU driver version for the previous SDKs up-to SDK 7.0 are as follows (SDK 9.x are shown for reference only):

    SDK GPU Driver Version
    7.0 1.10
    7.3 1.13
    8.4 1.15
    9.0 23.1
    9.1 23.2
    9.2 23.3

    All of the known GPU driver bugs will be fixed on the GPU 1.15 driver. Users can do a simple upgrade to the 1.15 driver version who are 1.13 or 1.10. Below are the necessary libraries and drivers required for the latest GPU 1.15 patched driver and patches to the kernel module.

    Installation instructions can be found at this FAQ:

    https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1316761/faq-tda4vl-q1-how-do-i-update-the-gpu-driver-on-tda4x-devices

    Latest 1.15 Drivers

    UM Libraries: 1.15-umlibs-latest-patches_7-25-2024.tar.gz

    The following patch information is to keep track of the version of these binary libraries and does not provide an explanation of the patches. The patches address internal bugs in the GPU closed-source driver, and the specifics will not help the average user understand the bugs. If the patch details are required, please contact your local TI contact to request that information.

    Patch Number Patch Name
    1 PP131820_CL6516292_Limit_tiles_in_flight_1.15
    2 PP135344_CL6268155_CL6268772_GLES3_Disallow_unsafe_glMapBufferRange_1.15
    3 PP135344_CL6324052_Fix_draw_batch_render_corruption_1.15
    4 PP135344_CL6228347_GrowSparseBuffer_changes_1.15
    5 PP135344_CL6386875_CL6387822_Fix_HWR_caused_by_PDSRW_fault_1.15_v2
    6 PP135344_CL6138204_Fix_use_of_incorrect_3D_buffers_1.15
    7 PP135344_CL6146431_Fix_TA_kick_occasionally_being_missed_1.15
    8 PP135747_CL6324217_CL6324233_GLES3_Make_PDS_DOUTU_data_for_pixel_event_programs_static
    9 PP134522_CL6384712_Firmware_Add_missing_TFBC_cache_maintenance_1.15
    10 PP160112_CL6589013

    KM Driver Patches:

    Branch: git.ti.com/.../

    Patch Name Patch File Patch Description
    Cache Coherency Patch https://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/791/6283.CL6529585_5F00_Enable_5F00_cached_5F00_mappings_5F00_in_5F00_KM_5F00_on_5F00_ARM64_5F00_for_5F00_DDK_5F00_1.15_5F00_with_5F00_snooping_5F00_mode_5F00_update.patch

    On TI platforms with MSMC (SnoopFilter) and UMA systems with PIPT caches when device mappings are un-cached in KM there is a risk of pre-fetching of cache lines by the Linux Kernel leading to incorrect view of memory from the GPU when Linux direct map (cached) being snooped by the device causing invalid data read.

    As these TI platforms have an one-way coherency using ACE-Lite where the GPU can snoop from CPU caches, the KM mappings to be cached. Making all allocations cached solves the above issue due to the nature of PIPT and implicit cache maintenance performed by the system. If the CPU cache is being correctly invalidated in response to GPU writes (as our testing has shown) then the DDK would not require any cache maintenance at all, hence every device memory allocation can be CPU-cached.
    Revert Previous Hack No patch file, manually run this command: "git revert c901804e8221d477983a6f7224a9cdc6e832f050" commit 4e3acbb09195b3486d429e96f900cb4f4fd969f6 (HEAD -> linuxws/dunfell/k5.10/1.15.6133109_unified_fw_pagesize)
    Author: Erick Narvaez <e-narvaez@ti.com>
    Date:   Mon Jan 22 16:20:21 2024 -0600

        Revert "HACK: server: Make CCB allocations incoherent"
        
        This reverts commit c901804e8221d477983a6f7224a9cdc6e832f050.