550.40.55

CHANGELOG.md

@@ -2,6 +2,8 @@
 
 ## Release 550 Entries
 
+### [550.40.55] 2024-03-07
+
 ### [550.40.53] 2024-02-28
 
 #### Added

README.md

@@ -1,7 +1,7 @@
 # NVIDIA Linux Open GPU Kernel Module Source
 
 This is the source release of the NVIDIA Linux open GPU kernel modules,
-version 550.40.53.
+version 550.40.55.
 
 
 ## How to Build
@@ -17,7 +17,7 @@ as root:
 
 Note that the kernel modules built here must be used with GSP
 firmware and user-space NVIDIA GPU driver components from a corresponding
-550.40.53 driver release.  This can be achieved by installing
+550.40.55 driver release.  This can be achieved by installing
 the NVIDIA GPU driver from the .run file using the `--no-kernel-modules`
 option.  E.g.,
 
@@ -188,7 +188,7 @@ encountered specific to them.
 For details on feature support and limitations, see the NVIDIA GPU driver
 end user README here:
 
-https://us.download.nvidia.com/XFree86/Linux-x86_64/550.40.53/README/kernel_open.html
+https://us.download.nvidia.com/XFree86/Linux-x86_64/550.40.55/README/kernel_open.html
 
 For vGPU support, please refer to the README.vgpu packaged in the vGPU Host
 Package for more details.
@@ -653,6 +653,7 @@ Subsystem Device ID.
 | NVIDIA T400 4GB                                 | 1FF2 103C 1613 |
 | NVIDIA T400 4GB                                 | 1FF2 103C 8A80 |
 | NVIDIA T400 4GB                                 | 1FF2 10DE 1613 |
+| NVIDIA T400E                                    | 1FF2 10DE 18FF |
 | NVIDIA T400 4GB                                 | 1FF2 17AA 1613 |
 | NVIDIA T400E                                    | 1FF2 17AA 18FF |
 | Quadro T1000                                    | 1FF9           |
@@ -812,6 +813,7 @@ Subsystem Device ID.
 | NVIDIA RTX A2000 12GB                           | 2571 10DE 1611 |
 | NVIDIA RTX A2000 12GB                           | 2571 17AA 1611 |
 | NVIDIA GeForce RTX 3050                         | 2582           |
+| NVIDIA GeForce RTX 3050                         | 2584           |
 | NVIDIA GeForce RTX 3050 Ti Laptop GPU           | 25A0           |
 | NVIDIA GeForce RTX 3050Ti Laptop GPU            | 25A0 103C 8928 |
 | NVIDIA GeForce RTX 3050Ti Laptop GPU            | 25A0 103C 89F9 |
@@ -905,6 +907,9 @@ Subsystem Device ID.
 | NVIDIA RTX 2000 Ada Generation                  | 28B0 10DE 1870 |
 | NVIDIA RTX 2000 Ada Generation                  | 28B0 17AA 1870 |
 | NVIDIA RTX 2000 Ada Generation Laptop GPU       | 28B8           |
+| NVIDIA RTX 1000 Ada Generation Laptop GPU       | 28B9           |
+| NVIDIA RTX 500 Ada Generation Laptop GPU        | 28BA           |
+| NVIDIA RTX 500 Ada Generation Laptop GPU        | 28BB           |
 | NVIDIA GeForce RTX 4060 Laptop GPU              | 28E0           |
 | NVIDIA GeForce RTX 4050 Laptop GPU              | 28E1           |
 | NVIDIA RTX 2000 Ada Generation Embedded GPU     | 28F8           |

kernel-open/Kbuild

@@ -72,7 +72,7 @@ EXTRA_CFLAGS += -I$(src)/common/inc
 EXTRA_CFLAGS += -I$(src)
 EXTRA_CFLAGS += -Wall $(DEFINES) $(INCLUDES) -Wno-cast-qual -Wno-format-extra-args
 EXTRA_CFLAGS += -D__KERNEL__ -DMODULE -DNVRM
-EXTRA_CFLAGS += -DNV_VERSION_STRING=\"550.40.53\"
+EXTRA_CFLAGS += -DNV_VERSION_STRING=\"550.40.55\"
 
 ifneq ($(SYSSRCHOST1X),)
  EXTRA_CFLAGS += -I$(SYSSRCHOST1X)

kernel-open/conftest.sh

@@ -3096,6 +3096,22 @@ compile_test() {
 
         ;;
 
+        foll_longterm_present)
+            #
+            # Determine if FOLL_LONGTERM enum is present or not
+            #
+            # Added by commit 932f4a630a69 ("mm/gup: replace
+            # get_user_pages_longterm() with FOLL_LONGTERM") in
+            # v5.2
+            #
+            CODE="
+            #include <linux/mm.h>
+            int foll_longterm = FOLL_LONGTERM;
+            "
+
+            compile_check_conftest "$CODE" "NV_FOLL_LONGTERM_PRESENT" "" "types"
+        ;;
+
         vfio_pin_pages_has_vfio_device_arg)
             #
             # Determine if vfio_pin_pages() kABI accepts "struct vfio_device *"

kernel-open/nvidia-drm/nvidia-drm-drv.c

@@ -1903,8 +1903,33 @@ void nv_drm_remove_devices(void)
  */
 void nv_drm_suspend_resume(NvBool suspend)
 {
+    static DEFINE_MUTEX(nv_drm_suspend_mutex);
+    static NvU32 nv_drm_suspend_count = 0;
+    struct nv_drm_device *nv_dev;
+
+    mutex_lock(&nv_drm_suspend_mutex);
+
+    /*
+     * Count the number of times the driver is asked to suspend. Suspend all DRM
+     * devices on the first suspend call and resume them on the last resume
+     * call.  This is necessary because the kernel may call nvkms_suspend()
+     * simultaneously for each GPU, but NVKMS itself also suspends all GPUs on
+     * the first call.
+     */
+    if (suspend) {
+        if (nv_drm_suspend_count++ > 0) {
+            goto done;
+        }
+    } else {
+        BUG_ON(nv_drm_suspend_count == 0);
+
+        if (--nv_drm_suspend_count > 0) {
+            goto done;
+        }
+    }
+
 #if defined(NV_DRM_ATOMIC_MODESET_AVAILABLE)
-    struct nv_drm_device *nv_dev = dev_list;
+    nv_dev = dev_list;
 
     /*
      * NVKMS shuts down all heads on suspend. Update DRM state accordingly.
@@ -1930,6 +1955,9 @@ void nv_drm_suspend_resume(NvBool suspend)
         }
     }
 #endif /* NV_DRM_ATOMIC_MODESET_AVAILABLE */
+
+done:
+    mutex_unlock(&nv_drm_suspend_mutex);
 }
 
 #endif /* NV_DRM_AVAILABLE */

kernel-open/nvidia-modeset/nvidia-modeset-linux.c

@@ -56,7 +56,11 @@
 #include "nv-lock.h"
 #include "nv-chardev-numbers.h"
 
-#if !defined(CONFIG_RETPOLINE)
+/*
+ * Commit aefb2f2e619b ("x86/bugs: Rename CONFIG_RETPOLINE =>
+ * CONFIG_MITIGATION_RETPOLINE) in v6.8 renamed CONFIG_RETPOLINE.
+ */
+#if !defined(CONFIG_RETPOLINE) && !defined(CONFIG_MITIGATION_RETPOLINE)
 #include "nv-retpoline.h"
 #endif
 
@@ -499,8 +503,9 @@ nvkms_event_queue_changed(nvkms_per_open_handle_t *pOpenKernel,
 
 static void nvkms_suspend(NvU32 gpuId)
 {
+    nvKmsKapiSuspendResume(NV_TRUE /* suspend */);
+
     if (gpuId == 0) {
-        nvKmsKapiSuspendResume(NV_TRUE /* suspend */);
         nvkms_write_lock_pm_lock();
     }
 
@@ -517,8 +522,9 @@ static void nvkms_resume(NvU32 gpuId)
 
     if (gpuId == 0) {
         nvkms_write_unlock_pm_lock();
-        nvKmsKapiSuspendResume(NV_FALSE /* suspend */);
     }
+
+    nvKmsKapiSuspendResume(NV_FALSE /* suspend */);
 }
 
 

kernel-open/nvidia/nv.c

@@ -57,7 +57,11 @@
 #include "nv-dmabuf.h"
 #include "nv-caps-imex.h"
 
-#if !defined(CONFIG_RETPOLINE)
+/*
+ * Commit aefb2f2e619b ("x86/bugs: Rename CONFIG_RETPOLINE =>
+ * CONFIG_MITIGATION_RETPOLINE) in v6.8 renamed CONFIG_RETPOLINE.
+ */
+#if !defined(CONFIG_RETPOLINE) && !defined(CONFIG_MITIGATION_RETPOLINE)
 #include "nv-retpoline.h"
 #endif
 

kernel-open/nvidia/nvidia.Kbuild

@@ -250,6 +250,7 @@ NV_CONFTEST_TYPE_COMPILE_TESTS += num_registered_fb
 NV_CONFTEST_TYPE_COMPILE_TESTS += pci_driver_has_driver_managed_dma
 NV_CONFTEST_TYPE_COMPILE_TESTS += vm_area_struct_has_const_vm_flags
 NV_CONFTEST_TYPE_COMPILE_TESTS += memory_failure_has_trapno_arg
+NV_CONFTEST_TYPE_COMPILE_TESTS += foll_longterm_present
 
 NV_CONFTEST_GENERIC_COMPILE_TESTS += dom0_kernel_present
 NV_CONFTEST_GENERIC_COMPILE_TESTS += nvidia_vgpu_kvm_build

kernel-open/nvidia/os-mlock.c

@@ -26,6 +26,12 @@
 #include "os-interface.h"
 #include "nv-linux.h"
 
+#if defined(NVCPU_FAMILY_X86) && defined(NV_FOLL_LONGTERM_PRESENT) && \
+    (defined(NV_PIN_USER_PAGES_HAS_ARGS_VMAS) ||                      \
+     defined(NV_GET_USER_PAGES_HAS_ARGS_FLAGS_VMAS))
+#define NV_NUM_PIN_PAGES_PER_ITERATION 0x80000
+#endif
+
 static inline int nv_follow_pfn(struct vm_area_struct *vma,
                                 unsigned long address,
                                 unsigned long *pfn)
@@ -163,9 +169,15 @@ NV_STATUS NV_API_CALL os_lock_user_pages(
     NV_STATUS rmStatus;
     struct mm_struct *mm = current->mm;
     struct page **user_pages;
-    NvU64 i, pinned;
+    NvU64 i;
+    NvU64 npages = page_count;
+    NvU64 pinned = 0;
     unsigned int gup_flags = DRF_VAL(_LOCK_USER_PAGES, _FLAGS, _WRITE, flags) ? FOLL_WRITE : 0;
-    int ret;
+    long ret;
+
+#if defined(NVCPU_FAMILY_X86) && defined(NV_FOLL_LONGTERM_PRESENT)
+    gup_flags |= FOLL_LONGTERM;
+#endif
 
     if (!NV_MAY_SLEEP())
     {
@@ -185,16 +197,51 @@ NV_STATUS NV_API_CALL os_lock_user_pages(
 
     nv_mmap_read_lock(mm);
     ret = NV_PIN_USER_PAGES((unsigned long)address,
-                            page_count, gup_flags, user_pages);
-    nv_mmap_read_unlock(mm);
-    pinned = ret;
-
-    if (ret < 0)
+                            npages, gup_flags, user_pages);
+    if (ret > 0)
     {
-        os_free_mem(user_pages);
-        return NV_ERR_INVALID_ADDRESS;
+        pinned = ret;
     }
-    else if (pinned < page_count)
+#if defined(NVCPU_FAMILY_X86) && defined(NV_FOLL_LONGTERM_PRESENT) && \
+    (defined(NV_PIN_USER_PAGES_HAS_ARGS_VMAS) ||                      \
+     defined(NV_GET_USER_PAGES_HAS_ARGS_FLAGS_VMAS))
+    //
+    // NV_PIN_USER_PAGES() passes in NULL for the vmas parameter (if required)
+    // in pin_user_pages() (or get_user_pages() if pin_user_pages() does not
+    // exist). For kernels which do not contain the commit 52650c8b466b
+    // (mm/gup: remove the vma allocation from gup_longterm_locked()), if
+    // FOLL_LONGTERM is passed in, this results in the kernel trying to kcalloc
+    // the vmas array, and since the limit for kcalloc is 4 MB, it results in
+    // NV_PIN_USER_PAGES() failing with ENOMEM if more than
+    // NV_NUM_PIN_PAGES_PER_ITERATION pages are requested on 64-bit systems.
+    //
+    // As a workaround, if we requested more than
+    // NV_NUM_PIN_PAGES_PER_ITERATION pages and failed with ENOMEM, try again
+    // with multiple calls of NV_NUM_PIN_PAGES_PER_ITERATION pages at a time.
+    //
+    else if ((ret == -ENOMEM) &&
+             (page_count > NV_NUM_PIN_PAGES_PER_ITERATION))
+    {
+        for (pinned = 0; pinned < page_count; pinned += ret)
+        {
+            npages = page_count - pinned;
+            if (npages > NV_NUM_PIN_PAGES_PER_ITERATION)
+            {
+                npages = NV_NUM_PIN_PAGES_PER_ITERATION;
+            }
+
+            ret = NV_PIN_USER_PAGES(((unsigned long) address) + (pinned * PAGE_SIZE),
+                                    npages, gup_flags, &user_pages[pinned]);
+            if (ret <= 0)
+            {
+                break;
+            }
+        }
+    }
+#endif
+    nv_mmap_read_unlock(mm);
+
+    if (pinned < page_count)
     {
         for (i = 0; i < pinned; i++)
             NV_UNPIN_USER_PAGE(user_pages[i]);

src/common/displayport/inc/dp_connectorimpl.h

@@ -348,6 +348,9 @@ namespace DisplayPort
         //
         bool        bPowerDownPhyBeforeD3;
 
+        // Force DSC on sink irrespective of LT status
+        bool        bForceDscOnSink;
+
         //
         // Reset the MSTM_CTRL registers on branch device irrespective of
         // IRQ VECTOR register having stale message. Certain branch devices

src/common/displayport/inc/dp_linkconfig.h

@@ -294,8 +294,8 @@ namespace DisplayPort
             else
             {
                 // if FEC is not enabled, link overhead comprises only of
-                // 0.05% downspread.
-                return rate - 5 * rate/ 1000;
+                // 0.6% downspread.
+                return rate - 6 * rate/ 1000;
 
             }
         }

src/common/displayport/inc/dp_regkeydatabase.h

@@ -79,6 +79,11 @@
 //
 #define NV_DP_REGKEY_MST_PCON_CAPS_READ_DISABLED       "DP_BUG_4388987_WAR"
 
+//
+// Bug 4459839 : This regkey will enable DSC irrespective of LT status.
+//
+#define NV_DP_REGKEY_FORCE_DSC_ON_SINK                 "DP_FORCE_DSC_ON_SINK"
+
 //
 // Data Base used to store all the regkey values.
 // The actual data base is declared statically in dp_evoadapter.cpp.
@@ -113,6 +118,7 @@ struct DP_REGKEY_DATABASE
     bool  bPowerDownPhyBeforeD3;
     bool  bReassessMaxLink;
     bool  bMSTPCONCapsReadDisabled;
+    bool  bForceDscOnSink;
 };
 
 #endif //INCLUDED_DP_REGKEYDATABASE_H

src/common/displayport/src/dp_connectorimpl.cpp

@@ -174,6 +174,7 @@ void ConnectorImpl::applyRegkeyOverrides(const DP_REGKEY_DATABASE& dpRegkeyDatab
     this->bDscMstCapBug3143315          = dpRegkeyDatabase.bDscMstCapBug3143315;
     this->bPowerDownPhyBeforeD3         = dpRegkeyDatabase.bPowerDownPhyBeforeD3;
     this->bReassessMaxLink              = dpRegkeyDatabase.bReassessMaxLink;
+    this->bForceDscOnSink               = dpRegkeyDatabase.bForceDscOnSink;
 }
 
 void ConnectorImpl::setPolicyModesetOrderMitigation(bool enabled)
@@ -3129,7 +3130,7 @@ bool ConnectorImpl::notifyAttachBegin(Group *                target,       // Gr
 
     // if LT is successful, see if panel supports DSC and if so, set DSC enabled/disabled
     // according to the mode requested.
-    if(bLinkTrainingStatus)
+    if(bLinkTrainingStatus || bForceDscOnSink)
     {
         for (Device * dev = target->enumDevices(0); dev; dev = target->enumDevices(dev))
         {
@@ -4631,6 +4632,11 @@ bool ConnectorImpl::trainLinkOptimized(LinkConfiguration lConfig)
                 }
             }
 
+            //
+            // There is no point in fallback here since we are link training  
+            // to loweset link config that can support the mode.
+            //
+            lowestSelected.policy.setSkipFallBack(true);
             bLinkTrainingSuccessful = train(lowestSelected, false);
             //
             // If LT failed, check if skipLT was marked. If so, clear the flag and
@@ -4648,16 +4654,37 @@ bool ConnectorImpl::trainLinkOptimized(LinkConfiguration lConfig)
             }
             if (!bLinkTrainingSuccessful)
             {
-                // Try fall back to max link config and if that fails try original assessed link configuration
+                // If optimized link config fails, try max link config with fallback. 
                 if (!train(getMaxLinkConfig(), false))
                 {
+                    //
+                    // Note here that if highest link config fails and a lower  
+                    // link config passes, link training will be returned as 
+                    // failure but activeLinkConfig will be set to that passing config.
+                    // 
                     if (!willLinkSupportModeSST(activeLinkConfig, groupAttached->lastModesetInfo))
                     {
+                        //
+                        // If none of the link configs pass LT or a fall back link config passed LT 
+                        // but cannot support the mode, then we will force the optimized link config
+                        // on the link and mark LT as fail.
+                        //
                         train(lowestSelected, true);
-
-                        // Mark link training as failed since we forced it
                         bLinkTrainingSuccessful = false;
                     }
+                    else
+                    {
+                        //
+                        // If a fallback link config pass LT and can support 
+                        // the mode, mark LT as pass.
+                        //
+                        bLinkTrainingSuccessful = true;
+                    }
+                }
+                else
+                {
+                    // If LT passes at max link config, mark LT as pass.
+                    bLinkTrainingSuccessful = true;
                 }
             }
         }