535.43.24

NVIDIA · Jan 31, 2024 · e558660 · e558660
1 parent 2a3b58b
commit e558660
Show file tree

Hide file tree

Showing 267 changed files with 88,989 additions and 82,768 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,7 +2,7 @@
 
 ## Release 535 Entries
 
-### [535.54.03] 2023-06-14
+### [535.43.24] 2024-01-31
 
 ### [535.43.23] 2024-01-24
 
@@ -30,6 +30,7 @@
 
 #### Fixed
 
+- Fixed building main against current centos stream 8 fails, [#550](https://github.com/NVIDIA/open-gpu-kernel-modules/issues/550) by @airlied
 - Fixed console restore with traditional VGA consoles.
 
 #### Added
@@ -58,6 +59,14 @@
 
 ## Release 525 Entries
 
+### [525.147.05] 2023-10-31
+
+#### Fixed
+
+- Fix nvidia_p2p_get_pages(): Fix double-free in register-callback error path, [#557](https://github.com/NVIDIA/open-gpu-kernel-modules/pull/557) by @BrendanCunningham
+
+### [525.125.06] 2023-06-26
+
 ### [525.116.04] 2023-05-09
 
 ### [525.116.03] 2023-04-25

diff --git a/README.md b/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 535.43.23.
+version 535.43.24.
 
 
 ## 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
-535.43.23 driver release.  This can be achieved by installing
+535.43.24 driver release.  This can be achieved by installing
 the NVIDIA GPU driver from the .run file using the `--no-kernel-modules`
 option.  E.g.,
 
@@ -180,15 +180,15 @@ software applications.
 ## Compatible GPUs
 
 The open-gpu-kernel-modules can be used on any Turing or later GPU
-(see the table below). However, in the 535.43.23 release,
+(see the table below). However, in the 535.43.24 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_64/535.43.23/README/kernel_open.html
+https://us.download.nvidia.com/XFree86/Linux-x86_64/535.43.24/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

diff --git a/kernel-open/Kbuild b/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-error -Wno-format-extra-args
 EXTRA_CFLAGS += -D__KERNEL__ -DMODULE -DNVRM
-EXTRA_CFLAGS += -DNV_VERSION_STRING=\"535.43.23\"
+EXTRA_CFLAGS += -DNV_VERSION_STRING=\"535.43.24\"
 
 ifneq ($(SYSSRCHOST1X),)
  EXTRA_CFLAGS += -I$(SYSSRCHOST1X)
@@ -123,6 +123,9 @@ ifneq ($(wildcard /proc/sgi_uv),)
  EXTRA_CFLAGS += -DNV_CONFIG_X86_UV
 endif
 
+ifdef VGX_FORCE_VFIO_PCI_CORE
+ EXTRA_CFLAGS += -DNV_VGPU_FORCE_VFIO_PCI_CORE
+endif
 
 #
 # The conftest.sh script tests various aspects of the target kernel.

diff --git a/kernel-open/common/inc/nv-linux.h b/kernel-open/common/inc/nv-linux.h
@@ -2067,4 +2067,7 @@ typedef enum
 #include <linux/clk-provider.h>
 #endif
 
+#define NV_EXPORT_SYMBOL(symbol)        EXPORT_SYMBOL_GPL(symbol)
+#define NV_CHECK_EXPORT_SYMBOL(symbol)  NV_IS_EXPORT_SYMBOL_PRESENT_##symbol
+
 #endif  /* _NV_LINUX_H_ */
diff --git a/kernel-open/common/inc/nv.h b/kernel-open/common/inc/nv.h
@@ -924,6 +924,7 @@ NV_STATUS  NV_API_CALL  rm_ioctl                 (nvidia_stack_t *, nv_state_t *
 NvBool     NV_API_CALL  rm_isr                   (nvidia_stack_t *, nv_state_t *, NvU32 *);
 void       NV_API_CALL  rm_isr_bh                (nvidia_stack_t *, nv_state_t *);
 void       NV_API_CALL  rm_isr_bh_unlocked       (nvidia_stack_t *, nv_state_t *);
+NvBool     NV_API_CALL  rm_is_msix_allowed       (nvidia_stack_t *, nv_state_t *);
 NV_STATUS  NV_API_CALL  rm_power_management      (nvidia_stack_t *, nv_state_t *, nv_pm_action_t);
 NV_STATUS  NV_API_CALL  rm_stop_user_channels    (nvidia_stack_t *, nv_state_t *);
 NV_STATUS  NV_API_CALL  rm_restart_user_channels (nvidia_stack_t *, nv_state_t *);

diff --git a/kernel-open/common/inc/os-interface.h b/kernel-open/common/inc/os-interface.h
@@ -207,9 +207,13 @@ enum os_pci_req_atomics_type {
     OS_INTF_PCIE_REQ_ATOMICS_128BIT
 };
 NV_STATUS   NV_API_CALL  os_enable_pci_req_atomics   (void *, enum os_pci_req_atomics_type);
+NV_STATUS   NV_API_CALL  os_get_numa_node_memory_usage (NvS32, NvU64 *, NvU64 *);
 NV_STATUS   NV_API_CALL  os_numa_add_gpu_memory      (void *, NvU64, NvU64, NvU32 *);
 NV_STATUS   NV_API_CALL  os_numa_remove_gpu_memory   (void *, NvU64, NvU64, NvU32); 
 NV_STATUS   NV_API_CALL  os_offline_page_at_address(NvU64 address);
+void*       NV_API_CALL  os_get_pid_info(void);
+void        NV_API_CALL  os_put_pid_info(void *pid_info);
+NV_STATUS   NV_API_CALL  os_find_ns_pid(void *pid_info, NvU32 *ns_pid);
 
 extern NvU32 os_page_size;
 extern NvU64 os_page_mask;

diff --git a/kernel-open/conftest.sh b/kernel-open/conftest.sh
@@ -316,7 +316,7 @@ export_symbol_present_conftest() {
     SYMBOL="$1"
     TAB='	'
 
-    if grep -e "${TAB}${SYMBOL}${TAB}.*${TAB}EXPORT_SYMBOL.*\$" \
+    if grep -e "${TAB}${SYMBOL}${TAB}.*${TAB}EXPORT_SYMBOL\(_GPL\)\?\s*\$" \
                "$OUTPUT/Module.symvers" >/dev/null 2>&1; then
         echo "#define NV_IS_EXPORT_SYMBOL_PRESENT_$SYMBOL 1" |
             append_conftest "symbols"
@@ -337,7 +337,7 @@ export_symbol_gpl_conftest() {
     SYMBOL="$1"
     TAB='	'
 
-    if grep -e "${TAB}${SYMBOL}${TAB}.*${TAB}EXPORT_\(UNUSED_\)*SYMBOL_GPL\$" \
+    if grep -e "${TAB}${SYMBOL}${TAB}.*${TAB}EXPORT_\(UNUSED_\)*SYMBOL_GPL\s*\$" \
                "$OUTPUT/Module.symvers" >/dev/null 2>&1; then
         echo "#define NV_IS_EXPORT_SYMBOL_GPL_$SYMBOL 1" |
             append_conftest "symbols"
@@ -4468,6 +4468,24 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_MMU_NOTIFIER_OPS_HAS_INVALIDATE_RANGE" "" "types"
         ;;
 
+        mmu_notifier_ops_arch_invalidate_secondary_tlbs)
+            #
+            # Determine if the mmu_notifier_ops struct has the
+            # 'arch_invalidate_secondary_tlbs' member.
+            #
+            # struct mmu_notifier_ops.invalidate_range was renamed to
+            # arch_invalidate_secondary_tlbs by commit 1af5a8109904
+            # ("mmu_notifiers: rename invalidate_range notifier") due to be
+            # added in v6.6
+           CODE="
+            #include <linux/mmu_notifier.h>
+            int conftest_mmu_notifier_ops_arch_invalidate_secondary_tlbs(void) {
+                return offsetof(struct mmu_notifier_ops, arch_invalidate_secondary_tlbs);
+            }"
+
+            compile_check_conftest "$CODE" "NV_MMU_NOTIFIER_OPS_HAS_ARCH_INVALIDATE_SECONDARY_TLBS" "" "types"
+        ;;
+
         drm_format_num_planes)
             #
             # Determine if drm_format_num_planes() function is present.
@@ -5636,23 +5654,6 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_GPIO_TO_IRQ_PRESENT" "" "functions"
         ;;
 
-        migrate_vma_setup)
-            #
-            # Determine if migrate_vma_setup() function is present
-            #
-            # migrate_vma_setup() function was added by commit
-            # a7d1f22bb74f32cf3cd93f52776007e161f1a738 ("mm: turn migrate_vma
-            # upside down) in v5.4.
-            # (2019-08-20).
-            CODE="
-            #include <linux/migrate.h>
-            int conftest_migrate_vma_setup(void) {
-                migrate_vma_setup();
-            }"
-
-            compile_check_conftest "$CODE" "NV_MIGRATE_VMA_SETUP_PRESENT" "" "functions"
-        ;;
-
         migrate_vma_added_flags)
             #
             # Determine if migrate_vma structure has flags
@@ -5743,23 +5744,25 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_IOASID_GET_PRESENT" "" "functions"
         ;;
 
-        mm_pasid_set)
+        mm_pasid_drop)
+            #
+            # Determine if mm_pasid_drop() function is present
             #
-            # Determine if mm_pasid_set() function is present
+            # Added by commit 701fac40384f ("iommu/sva: Assign a PASID to mm
+            # on PASID allocation and free it on mm exit") in v5.18.
+            # Moved to linux/iommu.h in commit cd3891158a77 ("iommu/sva: Move
+            # PASID helpers to sva code") in v6.4.
             #
-            # mm_pasid_set() function was added by commit
-            # 701fac40384f07197b106136012804c3cae0b3de (iommu/sva: Assign a
-            # PASID to mm on PASID allocation and free it on mm exit) in v5.18.
-            # (2022-02-15).
             CODE="
             #if defined(NV_LINUX_SCHED_MM_H_PRESENT)
             #include <linux/sched/mm.h>
             #endif
-            void conftest_mm_pasid_set(void) {
-                mm_pasid_set();
+            #include <linux/iommu.h>
+            void conftest_mm_pasid_drop(void) {
+                mm_pasid_drop();
             }"
 
-            compile_check_conftest "$CODE" "NV_MM_PASID_SET_PRESENT" "" "functions"
+            compile_check_conftest "$CODE" "NV_MM_PASID_DROP_PRESENT" "" "functions"
         ;;
 
         drm_crtc_state_has_no_vblank)
@@ -6279,6 +6282,21 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_MEMORY_FAILURE_MF_SW_SIMULATED_DEFINED" "" "types"
         ;;
 
+        crypto_tfm_ctx_aligned)
+            # Determine if 'crypto_tfm_ctx_aligned' is defined.
+            #
+            # Removed by commit 25c74a39e0f6 ("crypto: hmac - remove unnecessary
+            # alignment logic") in v6.7.
+            #
+            CODE="
+            #include <crypto/algapi.h>
+            void conftest_crypto_tfm_ctx_aligned(void) {
+                  (void)crypto_tfm_ctx_aligned();
+            }"
+
+            compile_check_conftest "$CODE" "NV_CRYPTO_TFM_CTX_ALIGNED_PRESENT" "" "functions"
+        ;;
+
         crypto)
             #
             # Determine if we support various crypto functions.
@@ -6341,6 +6359,22 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_MEMPOLICY_HAS_HOME_NODE" "" "types"
         ;;
 
+        mpol_preferred_many_present)
+            #
+            # Determine if MPOL_PREFERRED_MANY enum is present or not
+            #
+            # Added by commit b27abaccf8e8b ("mm/mempolicy: add
+            # MPOL_PREFERRED_MANY for multiple preferred nodes") in
+            # v5.15
+            #
+            CODE="
+            #include <linux/mempolicy.h>
+            int mpol_preferred_many = MPOL_PREFERRED_MANY;
+            "
+
+            compile_check_conftest "$CODE" "NV_MPOL_PREFERRED_MANY_PRESENT" "" "types"
+        ;;
+
         mmu_interval_notifier)
             #
             # Determine if mmu_interval_notifier struct is present or not
@@ -6356,6 +6390,21 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_MMU_INTERVAL_NOTIFIER" "" "types"
         ;;
 
+        drm_unlocked_ioctl_flag_present)
+            # Determine if DRM_UNLOCKED IOCTL flag is present.
+            #
+            # DRM_UNLOCKED was removed by commit 2798ffcc1d6a ("drm: Remove
+            # locking for legacy ioctls and DRM_UNLOCKED") in Linux
+            # next-20231208.
+            CODE="
+            #if defined(NV_DRM_DRM_IOCTL_H_PRESENT)
+            #include <drm/drm_ioctl.h>
+            #endif
+            int flags = DRM_UNLOCKED;"
+
+            compile_check_conftest "$CODE" "NV_DRM_UNLOCKED_IOCTL_FLAG_PRESENT" "" "types"
+        ;;
+
         # When adding a new conftest entry, please use the correct format for
         # specifying the relevant upstream Linux kernel commit.
         #
@@ -6680,18 +6729,9 @@ case "$5" in
                 VFIO_PCI_CORE_PRESENT=1
             fi
 
-            # When this sanity check is run via nvidia-installer, it sets ARCH as aarch64.
-            # But, when it is run via Kbuild, ARCH is set as arm64
-            if [ "$ARCH" = "aarch64" ]; then
-                ARCH="arm64"
-            fi
-
             if [ "$VFIO_IOMMU_PRESENT" != "0" ] && [ "$KVM_PRESENT" != "0" ] ; then
-
-                # On x86_64, vGPU requires MDEV framework to be present.
-                # On aarch64, vGPU requires MDEV or vfio-pci-core framework to be present.
-                if ([ "$ARCH" = "arm64" ] && ([ "$VFIO_MDEV_PRESENT" != "0" ] || [ "$VFIO_PCI_CORE_PRESENT" != "0" ])) ||
-                   ([ "$ARCH" = "x86_64" ] && [ "$VFIO_MDEV_PRESENT" != "0" ];) then
+                # vGPU requires either MDEV or vfio-pci-core framework to be present.
+                if [ "$VFIO_MDEV_PRESENT" != "0" ] || [ "$VFIO_PCI_CORE_PRESENT" != "0" ]; then
                     exit 0
                 fi
             fi
@@ -6702,14 +6742,10 @@ case "$5" in
                 echo "CONFIG_VFIO_IOMMU_TYPE1";
             fi
 
-            if [ "$ARCH" = "arm64" ] && [ "$VFIO_MDEV_PRESENT" = "0" ] && [ "$VFIO_PCI_CORE_PRESENT" = "0" ]; then
+            if [ "$VFIO_MDEV_PRESENT" = "0" ] && [ "$VFIO_PCI_CORE_PRESENT" = "0" ]; then
                 echo "either CONFIG_VFIO_MDEV or CONFIG_VFIO_PCI_CORE";
             fi
 
-            if [ "$ARCH" = "x86_64" ] && [ "$VFIO_MDEV_PRESENT" = "0" ]; then
-                echo "CONFIG_VFIO_MDEV";
-            fi
-
             if [ "$KVM_PRESENT" = "0" ]; then
                 echo "CONFIG_KVM";
             fi

diff --git a/kernel-open/nvidia-drm/nvidia-drm-drv.c b/kernel-open/nvidia-drm/nvidia-drm-drv.c
@@ -1312,9 +1312,21 @@ static const struct drm_ioctl_desc nv_drm_ioctls[] = {
                       DRM_RENDER_ALLOW|DRM_UNLOCKED),
 #endif
 
+    /*
+     * DRM_UNLOCKED is implicit for all non-legacy DRM driver IOCTLs since Linux
+     * v4.10 commit fa5386459f06 "drm: Used DRM_LEGACY for all legacy functions"
+     * (Linux v4.4 commit ea487835e887 "drm: Enforce unlocked ioctl operation
+     * for kms driver ioctls" previously did it only for drivers that set the
+     * DRM_MODESET flag), so this will race with SET_CLIENT_CAP. Linux v4.11
+     * commit dcf727ab5d17 "drm: setclientcap doesn't need the drm BKL" also
+     * removed locking from SET_CLIENT_CAP so there is no use attempting to lock
+     * manually. The latter commit acknowledges that this can expose userspace
+     * to inconsistent behavior when racing with itself, but accepts that risk.
+     */
     DRM_IOCTL_DEF_DRV(NVIDIA_GET_CLIENT_CAPABILITY,
                       nv_drm_get_client_capability_ioctl,
                       0),
+
 #if defined(NV_DRM_ATOMIC_MODESET_AVAILABLE)
     DRM_IOCTL_DEF_DRV(NVIDIA_GET_CRTC_CRC32,
                       nv_drm_get_crtc_crc32_ioctl,

diff --git a/kernel-open/nvidia-drm/nvidia-drm-gem-nvkms-memory.c b/kernel-open/nvidia-drm/nvidia-drm-gem-nvkms-memory.c
@@ -243,6 +243,15 @@ static int __nv_drm_nvkms_gem_obj_init(
     NvU64 *pages = NULL;
     NvU32 numPages = 0;
 
+    if ((size % PAGE_SIZE) != 0) {
+        NV_DRM_DEV_LOG_ERR(
+            nv_dev,
+            "NvKmsKapiMemory 0x%p size should be in a multiple of page size to "
+            "create a gem object",
+            pMemory);
+        return -EINVAL;
+    }
+
     nv_nvkms_memory->pPhysicalAddress = NULL;
     nv_nvkms_memory->pWriteCombinedIORemapAddress = NULL;
     nv_nvkms_memory->physically_mapped = false;

diff --git a/kernel-open/nvidia-drm/nvidia-drm-helper.h b/kernel-open/nvidia-drm/nvidia-drm-helper.h
@@ -582,6 +582,19 @@ static inline int nv_drm_format_num_planes(uint32_t format)
 
 #endif /* defined(NV_DRM_FORMAT_MODIFIERS_PRESENT) */
 
+/*
+ * DRM_UNLOCKED was removed with linux-next commit 2798ffcc1d6a ("drm: Remove
+ * locking for legacy ioctls and DRM_UNLOCKED"), but it was previously made
+ * implicit for all non-legacy DRM driver IOCTLs since Linux v4.10 commit
+ * fa5386459f06 "drm: Used DRM_LEGACY for all legacy functions" (Linux v4.4
+ * commit ea487835e887 "drm: Enforce unlocked ioctl operation for kms driver
+ * ioctls" previously did it only for drivers that set the DRM_MODESET flag), so
+ * it was effectively a no-op anyway.
+ */
+#if !defined(NV_DRM_UNLOCKED_IOCTL_FLAG_PRESENT)
+#define DRM_UNLOCKED 0
+#endif
+
 /*
  * drm_vma_offset_exact_lookup_locked() were added
  * by kernel commit 2225cfe46bcc which was Signed-off-by:

diff --git a/kernel-open/nvidia-drm/nvidia-drm.Kbuild b/kernel-open/nvidia-drm/nvidia-drm.Kbuild
@@ -133,3 +133,4 @@ NV_CONFTEST_TYPE_COMPILE_TESTS += drm_connector_lookup
 NV_CONFTEST_TYPE_COMPILE_TESTS += drm_connector_put
 NV_CONFTEST_TYPE_COMPILE_TESTS += vm_area_struct_has_const_vm_flags
 NV_CONFTEST_TYPE_COMPILE_TESTS += drm_driver_has_dumb_destroy
+NV_CONFTEST_TYPE_COMPILE_TESTS += drm_unlocked_ioctl_flag_present