520.56.06

Author: aritger

CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## Release 520 Entries
 
+### [520.56.06] 2022-10-12
+
+#### Added
+
+- Introduce support for GeForce RTX 4090 GPUs.
+
 ### [520.61.05] 2022-10-10
 
 #### 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 520.61.05.
+version 520.56.06.
 
 
 ## How to Build
@@ -17,7 +17,7 @@ as root:
 
 Note that the kernel modules built here must be used with gsp.bin
 firmware and user-space NVIDIA GPU driver components from a corresponding
-520.61.05 driver release.  This can be achieved by installing
+520.56.06 driver release.  This can be achieved by installing
 the NVIDIA GPU driver from the .run file using the `--no-kernel-modules`
 option.  E.g.,
 
@@ -167,15 +167,15 @@ for the target kernel.
 ## Compatible GPUs
 
 The open-gpu-kernel-modules can be used on any Turing or later GPU
-(see the table below). However, in the 520.61.05 release,
+(see the table below). However, in the 520.56.06 release,
 GeForce and Workstation support is still considered alpha-quality.
 
 To enable use of the open kernel modules on GeForce and Workstation GPUs,
 set the "NVreg_OpenRmEnableUnsupportedGpus" nvidia.ko kernel module
 parameter to 1. For more details, see the NVIDIA GPU driver end user
 README here:
 
-https://us.download.nvidia.com/XFree86/Linux-x86_64/520.61.05/README/kernel_open.html
+https://us.download.nvidia.com/XFree86/Linux-x86_64/520.56.06/README/kernel_open.html
 
 In the below table, if three IDs are listed, the first is the PCI Device 
 ID, the second is the PCI Subsystem Vendor ID, and the third is the PCI
@@ -685,6 +685,7 @@ Subsystem Device ID.
 | NVIDIA GeForce RTX 3090 Ti                      | 2203           |
 | NVIDIA GeForce RTX 3090                         | 2204           |
 | NVIDIA GeForce RTX 3080                         | 2206           |
+| NVIDIA GeForce RTX 3070 Ti                      | 2207           |
 | NVIDIA GeForce RTX 3080 Ti                      | 2208           |
 | NVIDIA GeForce RTX 3080                         | 220A           |
 | NVIDIA CMP 90HX                                 | 220D           |
@@ -709,6 +710,7 @@ Subsystem Device ID.
 | NVIDIA A10                                      | 2236 10DE 1482 |
 | NVIDIA A10G                                     | 2237 10DE 152F |
 | NVIDIA A10M                                     | 2238 10DE 1677 |
+| NVIDIA H100 PCIe                                | 2331 10DE 1626 |
 | NVIDIA GeForce RTX 3060 Ti                      | 2414           |
 | NVIDIA GeForce RTX 3080 Ti Laptop GPU           | 2420           |
 | NVIDIA RTX A5500 Laptop GPU                     | 2438           |
@@ -736,6 +738,7 @@ Subsystem Device ID.
 | NVIDIA RTX A3000 12GB Laptop GPU                | 24B9           |
 | NVIDIA RTX A4500 Laptop GPU                     | 24BA           |
 | NVIDIA RTX A3000 12GB Laptop GPU                | 24BB           |
+| NVIDIA GeForce RTX 3060 Ti                      | 24C9           |
 | NVIDIA GeForce RTX 3080 Laptop GPU              | 24DC           |
 | NVIDIA GeForce RTX 3070 Laptop GPU              | 24DD           |
 | NVIDIA GeForce RTX 3070 Ti Laptop GPU           | 24E0           |
@@ -751,6 +754,7 @@ Subsystem Device ID.
 | NVIDIA RTX A2000                                | 2531 103C 151D |
 | NVIDIA RTX A2000                                | 2531 10DE 151D |
 | NVIDIA RTX A2000                                | 2531 17AA 151D |
+| NVIDIA GeForce RTX 3060                         | 2544           |
 | NVIDIA GeForce RTX 3060 Laptop GPU              | 2560           |
 | NVIDIA GeForce RTX 3050 Ti Laptop GPU           | 2563           |
 | NVIDIA RTX A2000 12GB                           | 2571 1028 1611 |

kernel-open/Kbuild

@@ -72,7 +72,7 @@ EXTRA_CFLAGS += -I$(src)/common/inc
 EXTRA_CFLAGS += -I$(src)
 EXTRA_CFLAGS += -Wall -MD $(DEFINES) $(INCLUDES) -Wno-cast-qual -Wno-error -Wno-format-extra-args
 EXTRA_CFLAGS += -D__KERNEL__ -DMODULE -DNVRM
-EXTRA_CFLAGS += -DNV_VERSION_STRING=\"520.61.05\"
+EXTRA_CFLAGS += -DNV_VERSION_STRING=\"520.56.06\"
 
 EXTRA_CFLAGS += -Wno-unused-function
 

kernel-open/common/inc/nv-pgprot.h

@@ -78,8 +78,13 @@ static inline pgprot_t pgprot_modify_writecombine(pgprot_t old_prot)
 
 #define NV_PGPROT_UNCACHED_DEVICE(old_prot)     pgprot_noncached(old_prot)
 #if defined(NVCPU_AARCH64)
+#if defined(NV_MT_DEVICE_GRE_PRESENT)
+#define NV_PROT_WRITE_COMBINED_DEVICE   (PROT_DEFAULT | PTE_PXN | PTE_UXN |   \
+                                         PTE_ATTRINDX(MT_DEVICE_GRE))
+#else
 #define NV_PROT_WRITE_COMBINED_DEVICE   (PROT_DEFAULT | PTE_PXN | PTE_UXN |   \
                                          PTE_ATTRINDX(MT_DEVICE_nGnRE))
+#endif
 #define NV_PGPROT_WRITE_COMBINED_DEVICE(old_prot)                             \
     __pgprot_modify(old_prot, PTE_ATTRINDX_MASK, NV_PROT_WRITE_COMBINED_DEVICE)
 #define NV_PGPROT_WRITE_COMBINED(old_prot)      NV_PGPROT_UNCACHED(old_prot)

kernel-open/common/inc/nv.h

@@ -636,33 +636,27 @@ typedef enum
 #define NV_GET_NV_STATE(pGpu) \
     (nv_state_t *)((pGpu) ? (pGpu)->pOsGpuInfo : NULL)
 
-static inline NvBool IS_REG_OFFSET(nv_state_t *nv, NvU64 offset, NvU64 length)
-{
-    return ((offset >= nv->regs->cpu_address) &&
-            ((offset + (length - 1)) <= (nv->regs->cpu_address + (nv->regs->size - 1))));
-}
-
-static inline NvBool IS_FB_OFFSET(nv_state_t *nv, NvU64 offset, NvU64 length)
-{
-    return  ((nv->fb) && (offset >= nv->fb->cpu_address) &&
-             ((offset + (length - 1)) <= (nv->fb->cpu_address + (nv->fb->size - 1))));
-}
-
-static inline NvBool IS_UD_OFFSET(nv_state_t *nv, NvU64 offset, NvU64 length)
-{
-    return ((nv->ud.cpu_address != 0) && (nv->ud.size != 0) &&
-            (offset >= nv->ud.cpu_address) &&
-            ((offset + (length - 1)) <= (nv->ud.cpu_address + (nv->ud.size - 1))));
-}
-
-static inline NvBool IS_IMEM_OFFSET(nv_state_t *nv, NvU64 offset, NvU64 length)
-{
-    return ((nv->bars[NV_GPU_BAR_INDEX_IMEM].cpu_address != 0) &&
-            (nv->bars[NV_GPU_BAR_INDEX_IMEM].size != 0) &&
-            (offset >= nv->bars[NV_GPU_BAR_INDEX_IMEM].cpu_address) &&
-            ((offset + (length - 1)) <= (nv->bars[NV_GPU_BAR_INDEX_IMEM].cpu_address +
-                                         (nv->bars[NV_GPU_BAR_INDEX_IMEM].size - 1))));
-}
+#define IS_REG_OFFSET(nv, offset, length)                                       \
+    (((offset) >= (nv)->regs->cpu_address) &&                                   \
+    (((offset) + ((length)-1)) <=                                               \
+        (nv)->regs->cpu_address + ((nv)->regs->size-1)))
+
+#define IS_FB_OFFSET(nv, offset, length)                                        \
+    (((nv)->fb) && ((offset) >= (nv)->fb->cpu_address) &&                       \
+    (((offset) + ((length)-1)) <= (nv)->fb->cpu_address + ((nv)->fb->size-1)))
+
+#define IS_UD_OFFSET(nv, offset, length)                                        \
+    (((nv)->ud.cpu_address != 0) && ((nv)->ud.size != 0) &&                     \
+    ((offset) >= (nv)->ud.cpu_address) &&                                       \
+    (((offset) + ((length)-1)) <= (nv)->ud.cpu_address + ((nv)->ud.size-1)))
+
+#define IS_IMEM_OFFSET(nv, offset, length)                                      \
+    (((nv)->bars[NV_GPU_BAR_INDEX_IMEM].cpu_address != 0) &&                    \
+     ((nv)->bars[NV_GPU_BAR_INDEX_IMEM].size != 0) &&                           \
+     ((offset) >= (nv)->bars[NV_GPU_BAR_INDEX_IMEM].cpu_address) &&             \
+     (((offset) + ((length) - 1)) <=                                            \
+        (nv)->bars[NV_GPU_BAR_INDEX_IMEM].cpu_address +                         \
+            ((nv)->bars[NV_GPU_BAR_INDEX_IMEM].size - 1)))
 
 #define NV_RM_MAX_MSIX_LINES  8
 

kernel-open/conftest.sh

@@ -2334,6 +2334,23 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_PCI_DEV_HAS_ATS_ENABLED" "" "types"
         ;;
 
+        mt_device_gre)
+            #
+            # Determine if MT_DEVICE_GRE flag is present.
+            #
+            # MT_DEVICE_GRE flag is removed by commit 58cc6b72a21274
+            # ("arm64: mm: Remove unused support for Device-GRE memory type") in v5.14-rc1
+            # (2021-06-01).
+            #
+            CODE="
+            #include <asm/memory.h>
+            unsigned int conftest_mt_device_gre(void) {
+                return MT_DEVICE_GRE;
+            }"
+
+            compile_check_conftest "$CODE" "NV_MT_DEVICE_GRE_PRESENT" "" "types"
+        ;;
+
         get_user_pages)
             #
             # Conftest for get_user_pages()

kernel-open/nvidia/nv-mmap.c

@@ -431,7 +431,7 @@ static int nvidia_mmap_numa(
     const nv_alloc_mapping_context_t *mmap_context)
 {
     NvU64 start, addr;
-    NvU64 pages;
+    unsigned int pages;
     NvU64 i;
 
     pages = NV_VMA_SIZE(vma) >> PAGE_SHIFT;

kernel-open/nvidia/nv.c

@@ -1429,9 +1429,6 @@ static int nv_open_device(nv_state_t *nv, nvidia_stack_t *sp)
         return -ENODEV;
     }
 
-    if (unlikely(NV_ATOMIC_READ(nvl->usage_count) >= NV_S32_MAX))
-        return -EMFILE;
-
     if ( ! (nv->flags & NV_FLAG_OPEN))
     {
         /* Sanity check: !NV_FLAG_OPEN requires usage_count == 0 */

kernel-open/nvidia/nvidia.Kbuild

@@ -243,6 +243,7 @@ NV_CONFTEST_TYPE_COMPILE_TESTS += vmalloc_has_pgprot_t_arg
 NV_CONFTEST_TYPE_COMPILE_TESTS += mm_has_mmap_lock
 NV_CONFTEST_TYPE_COMPILE_TESTS += pci_channel_state
 NV_CONFTEST_TYPE_COMPILE_TESTS += pci_dev_has_ats_enabled
+NV_CONFTEST_TYPE_COMPILE_TESTS += mt_device_gre
 NV_CONFTEST_TYPE_COMPILE_TESTS += remove_memory_has_nid_arg
 NV_CONFTEST_TYPE_COMPILE_TESTS += add_memory_driver_managed_has_mhp_flags_arg
 NV_CONFTEST_TYPE_COMPILE_TESTS += num_registered_fb

kernel-open/nvidia/nvlink_linux.c

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2015-2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-FileCopyrightText: Copyright (c) 2015-2019 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  * SPDX-License-Identifier: MIT
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -207,6 +207,8 @@ static int nvlink_fops_release(struct inode *inode, struct file *filp)
 
     nvlink_print(NVLINK_DBG_INFO, "nvlink driver close\n");
 
+    WARN_ON(private == NULL);
+
     mutex_lock(&nvlink_drvctx.lock);
 
     if (private->capability_fds.fabric_mgmt > 0)