570.148.08

Author: mmaneetsingh

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 570.144.
+version 570.148.08.
 
 
 ## 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
-570.144 driver release.  This can be achieved by installing
+570.148.08 driver release.  This can be achieved by installing
 the NVIDIA GPU driver from the .run file using the `--no-kernel-modules`
 option.  E.g.,
 
@@ -185,7 +185,7 @@ table below).
 For details on feature support and limitations, see the NVIDIA GPU driver
 end user README here:
 
-https://us.download.nvidia.com/XFree86/Linux-x86_64/570.144/README/kernel_open.html
+https://us.download.nvidia.com/XFree86/Linux-x86_64/570.148.08/README/kernel_open.html
 
 For vGPU support, please refer to the README.vgpu packaged in the vGPU Host
 Package for more details.
@@ -964,6 +964,7 @@ Subsystem Device ID.
 | NVIDIA GeForce RTX 5080 Laptop GPU                      | 2C19           |
 | NVIDIA GeForce RTX 5090 Laptop GPU                      | 2C58           |
 | NVIDIA GeForce RTX 5080 Laptop GPU                      | 2C59           |
+| NVIDIA GeForce RTX 5060 Ti                              | 2D04           |
 | NVIDIA GeForce RTX 5070                                 | 2F04           |
 | NVIDIA GeForce RTX 5070 Ti Laptop GPU                   | 2F18           |
 | NVIDIA GeForce RTX 5070 Ti Laptop GPU                   | 2F58           |

kernel-open/Kbuild

@@ -86,7 +86,7 @@ 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=\"570.144\"
+EXTRA_CFLAGS += -DNV_VERSION_STRING=\"570.148.08\"
 
 ifneq ($(SYSSRCHOST1X),)
  EXTRA_CFLAGS += -I$(SYSSRCHOST1X)

kernel-open/conftest.sh

@@ -3132,6 +3132,21 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_FOLL_LONGTERM_PRESENT" "" "types"
         ;;
 
+        has_enum_pidtype_tgid)
+            # Determine if PIDTYPE_TGID is present in the kernel as an enum
+            #
+            # Added by commit 6883f81aac6f ("pid: Implement PIDTYPE_TGID")
+            # in v4.19
+            #
+            CODE="
+            #include <linux/pid.h>
+
+            enum pid_type type = PIDTYPE_TGID;
+            "
+
+            compile_check_conftest "$CODE" "NV_HAS_ENUM_PIDTYPE_TGID" "" "types"
+        ;;
+
         vfio_pin_pages_has_vfio_device_arg)
             #
             # Determine if vfio_pin_pages() kABI accepts "struct vfio_device *"

kernel-open/nvidia-uvm/uvm_blackwell_host.c

@@ -254,3 +254,31 @@ void uvm_hal_blackwell_host_tlb_invalidate_test(uvm_push_t *push,
                                    HWVALUE(C96F, MEM_OP_D, TLB_INVALIDATE_PDB_ADDR_HI, pdb_hi));
     }
 }
+
+uvm_access_counter_clear_op_t
+uvm_hal_blackwell_access_counter_query_clear_op_gb100(uvm_parent_gpu_t *parent_gpu,
+                                                      uvm_access_counter_buffer_entry_t **buffer_entries,
+                                                      NvU32 num_entries)
+{
+    if (parent_gpu->rm_info.accessCntrBufferCount > 1) {
+        NvU32 i;
+
+        for (i = 0; i < num_entries; i++) {
+            const uvm_access_counter_buffer_entry_t *entry = buffer_entries[i];
+
+            // The LSb identifies the die ID.
+            if ((entry->tag & 0x1) == 1)
+                return UVM_ACCESS_COUNTER_CLEAR_OP_ALL;
+        }
+    }
+
+    return UVM_ACCESS_COUNTER_CLEAR_OP_TARGETED;
+}
+
+uvm_access_counter_clear_op_t
+uvm_hal_blackwell_access_counter_query_clear_op_gb20x(uvm_parent_gpu_t *parent_gpu,
+                                                      uvm_access_counter_buffer_entry_t **buffer_entries,
+                                                      NvU32 num_entries)
+{
+    return UVM_ACCESS_COUNTER_CLEAR_OP_TARGETED;
+}

kernel-open/nvidia-uvm/uvm_gpu.c

@@ -1197,6 +1197,8 @@ static NV_STATUS alloc_parent_gpu(const NvProcessorUuid *gpu_uuid,
     uvm_sema_init(&parent_gpu->isr.replayable_faults.service_lock, 1, UVM_LOCK_ORDER_ISR);
     uvm_sema_init(&parent_gpu->isr.non_replayable_faults.service_lock, 1, UVM_LOCK_ORDER_ISR);
     uvm_mutex_init(&parent_gpu->access_counters_enablement_lock, UVM_LOCK_ORDER_ACCESS_COUNTERS);
+    uvm_mutex_init(&parent_gpu->access_counters_clear_tracker_lock, UVM_LOCK_ACCESS_COUNTERS_CLEAR_OPS);
+    uvm_tracker_init(&parent_gpu->access_counters_clear_tracker);
     uvm_spin_lock_irqsave_init(&parent_gpu->isr.interrupts_lock, UVM_LOCK_ORDER_LEAF);
     uvm_spin_lock_init(&parent_gpu->instance_ptr_table_lock, UVM_LOCK_ORDER_LEAF);
     uvm_rb_tree_init(&parent_gpu->instance_ptr_table);
@@ -1214,6 +1216,7 @@ static NV_STATUS alloc_parent_gpu(const NvProcessorUuid *gpu_uuid,
     return NV_OK;
 
 cleanup:
+    uvm_tracker_deinit(&parent_gpu->access_counters_clear_tracker);
     uvm_kvfree(parent_gpu);
 
     return status;
@@ -1644,19 +1647,12 @@ static void sync_parent_gpu_trackers(uvm_parent_gpu_t *parent_gpu,
 
     // Sync the access counter clear tracker too.
     if (parent_gpu->access_counters_supported && parent_gpu->access_counter_buffer) {
-        NvU32 notif_buf_index;
-        for (notif_buf_index = 0; notif_buf_index < parent_gpu->rm_info.accessCntrBufferCount; notif_buf_index++) {
-            uvm_access_counter_buffer_t *access_counters = &parent_gpu->access_counter_buffer[notif_buf_index];
+        uvm_mutex_lock(&parent_gpu->access_counters_clear_tracker_lock);
+        status = uvm_tracker_wait(&parent_gpu->access_counters_clear_tracker);
+        uvm_mutex_unlock(&parent_gpu->access_counters_clear_tracker_lock);
 
-            if (access_counters->rm_info.accessCntrBufferHandle != 0) {
-                uvm_access_counters_isr_lock(access_counters);
-                status = uvm_tracker_wait(&access_counters->clear_tracker);
-                uvm_access_counters_isr_unlock(access_counters);
-
-                if (status != NV_OK)
-                    UVM_ASSERT(status == uvm_global_get_status());
-            }
-        }
+        if (status != NV_OK)
+            UVM_ASSERT(status == uvm_global_get_status());
     }
 }
 
@@ -1787,6 +1783,8 @@ static void uvm_parent_gpu_destroy(nv_kref_t *nv_kref)
     for_each_sub_processor_index(sub_processor_index)
         UVM_ASSERT(!parent_gpu->gpus[sub_processor_index]);
 
+    uvm_tracker_deinit(&parent_gpu->access_counters_clear_tracker);
+
     uvm_kvfree(parent_gpu);
 }
 
@@ -2881,6 +2879,9 @@ static NV_STATUS gpu_retain_by_uuid_locked(const NvProcessorUuid *gpu_uuid,
     if (status != NV_OK)
         goto error_unregister;
 
+    if (gpu_info->accessCntrBufferCount > 1)
+        gpu_info->accessCntrBufferCount = 1;
+
     if (parent_gpu != NULL) {
         // If the UUID has been seen before, and if SMC is enabled, then check
         // if this specific partition has been seen previously. The UUID-based

kernel-open/nvidia-uvm/uvm_gpu.h

@@ -522,10 +522,6 @@ struct uvm_access_counter_buffer_struct
     // PCIe
     NvU32 cached_put;
 
-    // Tracker used to aggregate access counters clear operations, needed for
-    // GPU removal
-    uvm_tracker_t clear_tracker;
-
     // Current access counter configuration. During normal operation this
     // information is computed once during GPU initialization. However, tests
     // may override it to try different configuration values.
@@ -1205,6 +1201,11 @@ struct uvm_parent_gpu_struct
     uvm_access_counter_buffer_t *access_counter_buffer;
     uvm_mutex_t access_counters_enablement_lock;
 
+    // Tracker used to aggregate access counters clear operations, needed for
+    // GPU removal. It is only used when supports_access_counters is set.
+    uvm_tracker_t access_counters_clear_tracker;
+    uvm_mutex_t access_counters_clear_tracker_lock;
+
     // Number of uTLBs per GPC. This information is only valid on Pascal+ GPUs.
     NvU32 utlb_per_gpc_count;