适用于 Linux 4.10 内核的 Nvidia 驱动补丁
4.10 内核于上周发布了第一个 rc 版。
除了大量的驱动更新，当然喜闻乐见地又把 Nvidia 闭源驱动搞挂了。
虽然 Nvidia 的开发论坛上早有人发布了 for 4.10 内核的补丁，但是那个补丁似乎只适用于 pre-rc，也就是 rc0 版本。
而 rc1 的发布以后又修改了 CPU Hotplug 相关的几个函数，因此那个补丁并不适用。
为了编译通过，上周我只是简单粗暴地删除了 Nvidia 驱动文件中跟被删除的 CPU Hotplug 函数相关的一些代码。但是这样的做法实在是过于 dirty。
好在，这星期 Nvidia 开发论坛上终于放出了适用于 4.10 内核的正式补丁。
跟以往一样，只在最新的 370.26 驱动上通过编译测试，其他版本不保证。
--- a/kernel/common/inc/nv-linux.h 2016-12-09 02:17:49.000000000 +0100
+++ b/kernel/common/inc/nv-linux.h 2016-12-31 10:55:17.637374373 +0100
@@ -294,7 +294,8 @@
extern int nv_pat_mode;
-#if defined(CONFIG_HOTPLUG_CPU)
+//#if defined(CONFIG_HOTPLUG_CPU)
+#if 0
#define NV_ENABLE_HOTPLUG_CPU
#include <linux/cpu.h> /* CPU hotplug support */
#include <linux/notifier.h> /* struct notifier_block, etc */
--- a/kernel/common/inc/nv-mm.h 2016-12-09 02:17:49.000000000 +0100
+++ b/kernel/common/inc/nv-mm.h 2016-12-15 11:51:32.638165272 +0100
@@ -83,7 +83,7 @@
if (force)
flags |= FOLL_FORCE;
- return get_user_pages_remote(tsk, mm, start, nr_pages, flags, pages, vmas);
+ return get_user_pages_remote(tsk, mm, start, nr_pages, flags, pages, vmas, NULL);
}
#endif
#else
Only in NVIDIA-Linux-x86_64-375.26.patched/kernel: nv_compiler.h
--- a/kernel/nvidia/nv-p2p.c 2016-12-09 02:17:49.000000000 +0100
+++ b/kernel/nvidia/nv-p2p.c 2016-12-15 11:51:32.638165272 +0100
@@ -146,7 +146,7 @@
int nvidia_p2p_get_pages(
uint64_t p2p_token,
uint32_t va_space,
- uint64_t virtual_address,
+ uint64_t address,
uint64_t length,
struct nvidia_p2p_page_table **page_table,
void (*free_callback)(void * data),
@@ -211,7 +211,7 @@
}
status = rm_p2p_get_pages(sp, p2p_token, va_space,
- virtual_address, length, physical_addresses, wreqmb_h,
+ address, length, physical_addresses, wreqmb_h,
rreqmb_h, &entries, &gpu_uuid, *page_table,
free_callback, data);
if (status != NV_OK)
@@ -286,7 +286,7 @@
if (bGetPages)
{
- rm_p2p_put_pages(sp, p2p_token, va_space, virtual_address,
+ rm_p2p_put_pages(sp, p2p_token, va_space, address,
gpu_uuid, *page_table);
}
@@ -329,7 +329,7 @@
int nvidia_p2p_put_pages(
uint64_t p2p_token,
uint32_t va_space,
- uint64_t virtual_address,
+ uint64_t address,
struct nvidia_p2p_page_table *page_table
)
{
@@ -343,7 +343,7 @@
return rc;
}
- status = rm_p2p_put_pages(sp, p2p_token, va_space, virtual_address,
+ status = rm_p2p_put_pages(sp, p2p_token, va_space, address,
page_table->gpu_uuid, page_table);
if (status == NV_OK)
nvidia_p2p_free_page_table(page_table);
--- a/kernel/nvidia/nv-pat.c 2016-12-09 02:17:49.000000000 +0100
+++ b/kernel/nvidia/nv-pat.c 2016-12-31 10:45:47.380405916 +0100
@@ -217,7 +217,7 @@
else
NV_SMP_CALL_FUNCTION(nv_setup_pat_entries, hcpu, 1);
break;
- case CPU_DOWN_PREPARE:
+ case CPU_DOWN_PREPARE_FROZEN:
if (cpu == (NvUPtr)hcpu)
nv_restore_pat_entries(NULL);
else
--- a/kernel/nvidia-drm/nvidia-drm-fence.c 2016-12-09 02:13:39.000000000 +0100
+++ b/kernel/nvidia-drm/nvidia-drm-fence.c 2016-12-15 11:51:32.638165272 +0100
@@ -31,7 +31,7 @@
#if defined(NV_DRM_DRIVER_HAS_GEM_PRIME_RES_OBJ)
struct nv_fence {
- struct fence base;
+ struct dma_fence base;
spinlock_t lock;
struct nvidia_drm_device *nv_dev;
@@ -51,7 +51,7 @@
static const char *nvidia_drm_gem_prime_fence_op_get_driver_name
(
- struct fence *fence
+ struct dma_fence *fence
)
{
return "NVIDIA";
@@ -59,7 +59,7 @@
static const char *nvidia_drm_gem_prime_fence_op_get_timeline_name
(
- struct fence *fence
+ struct dma_fence *fence
)
{
return "nvidia.prime";
@@ -67,7 +67,7 @@
static bool nvidia_drm_gem_prime_fence_op_signaled
(
- struct fence *fence
+ struct dma_fence *fence
)
{
struct nv_fence *nv_fence = container_of(fence, struct nv_fence, base);
@@ -99,7 +99,7 @@
static bool nvidia_drm_gem_prime_fence_op_enable_signaling
(
- struct fence *fence
+ struct dma_fence *fence
)
{
bool ret = true;
@@ -107,7 +107,7 @@
struct nvidia_drm_gem_object *nv_gem = nv_fence->nv_gem;
struct nvidia_drm_device *nv_dev = nv_fence->nv_dev;
- if (fence_is_signaled(fence))
+ if (dma_fence_is_signaled(fence))
{
return false;
}
@@ -132,7 +132,7 @@
}
nv_gem->fenceContext.softFence = fence;
- fence_get(fence);
+ dma_fence_get(fence);
unlock_struct_mutex:
mutex_unlock(&nv_dev->dev->struct_mutex);
@@ -142,7 +142,7 @@
static void nvidia_drm_gem_prime_fence_op_release
(
- struct fence *fence
+ struct dma_fence *fence
)
{
struct nv_fence *nv_fence = container_of(fence, struct nv_fence, base);
@@ -151,7 +151,7 @@
static signed long nvidia_drm_gem_prime_fence_op_wait
(
- struct fence *fence,
+ struct dma_fence *fence,
bool intr,
signed long timeout
)
@@ -166,12 +166,12 @@
* that it should never get hit during normal operation, but not so long
* that the system becomes unresponsive.
*/
- return fence_default_wait(fence, intr,
+ return dma_fence_default_wait(fence, intr,
(timeout == MAX_SCHEDULE_TIMEOUT) ?
msecs_to_jiffies(96) : timeout);
}
-static const struct fence_ops nvidia_drm_gem_prime_fence_ops = {
+static const struct dma_fence_ops nvidia_drm_gem_prime_fence_ops = {
.get_driver_name = nvidia_drm_gem_prime_fence_op_get_driver_name,
.get_timeline_name = nvidia_drm_gem_prime_fence_op_get_timeline_name,
.signaled = nvidia_drm_gem_prime_fence_op_signaled,
@@ -281,7 +281,7 @@
bool force
)
{
- struct fence *fence = nv_gem->fenceContext.softFence;
+ struct dma_fence *fence = nv_gem->fenceContext.softFence;
WARN_ON(!mutex_is_locked(&nv_dev->dev->struct_mutex));
@@ -297,10 +297,10 @@
if (force || nv_fence_ready_to_signal(nv_fence))
{
- fence_signal(&nv_fence->base);
+ dma_fence_signal(&nv_fence->base);
nv_gem->fenceContext.softFence = NULL;
- fence_put(&nv_fence->base);
+ dma_fence_put(&nv_fence->base);
nvKms->disableChannelEvent(nv_dev->pDevice,
nv_gem->fenceContext.cb);
@@ -316,7 +316,7 @@
nv_fence = container_of(fence, struct nv_fence, base);
- fence_signal(&nv_fence->base);
+ dma_fence_signal(&nv_fence->base);
}
}
@@ -509,7 +509,7 @@
* fence_context_alloc() cannot fail, so we do not need to check a return
* value.
*/
- nv_gem->fenceContext.context = fence_context_alloc(1);
+ nv_gem->fenceContext.context = dma_fence_context_alloc(1);
ret = nvidia_drm_gem_prime_fence_import_semaphore(
nv_dev, nv_gem, p->index,
@@ -666,13 +666,13 @@
nv_fence->nv_gem = nv_gem;
spin_lock_init(&nv_fence->lock);
- fence_init(&nv_fence->base, &nvidia_drm_gem_prime_fence_ops,
+ dma_fence_init(&nv_fence->base, &nvidia_drm_gem_prime_fence_ops,
&nv_fence->lock, nv_gem->fenceContext.context,
p->sem_thresh);
reservation_object_add_excl_fence(&nv_gem->fenceContext.resv,
&nv_fence->base);
- fence_put(&nv_fence->base); /* Reservation object has reference */
+ dma_fence_put(&nv_fence->base); /* Reservation object has reference */
ret = 0;
--- a/kernel/nvidia-drm/nvidia-drm-gem.h 2016-12-09 02:13:39.000000000 +0100
+++ b/kernel/nvidia-drm/nvidia-drm-gem.h 2016-12-15 11:51:32.638165272 +0100
@@ -98,7 +98,7 @@
/* Software signaling structures */
struct NvKmsKapiChannelEvent *cb;
struct nvidia_drm_gem_prime_soft_fence_event_args *cbArgs;
- struct fence *softFence; /* Fence for software signaling */
+ struct dma_fence *softFence; /* Fence for software signaling */
} fenceContext;
#endif
};
--- a/kernel/nvidia-drm/nvidia-drm-modeset.c 2016-12-09 02:13:39.000000000 +0100
+++ b/kernel/nvidia-drm/nvidia-drm-modeset.c 2016-12-15 11:51:32.638165272 +0100
@@ -69,8 +69,7 @@
void nvidia_drm_atomic_state_free(struct drm_atomic_state *state)
{
- struct nvidia_drm_atomic_state *nv_state =
- to_nv_atomic_state(state);
+ struct nvidia_drm_atomic_state *nv_state = to_nv_atomic_state(state);
drm_atomic_state_default_release(state);
nvidia_drm_free(nv_state);
}
@@ -645,7 +644,7 @@
wake_up_all(&nv_dev->pending_commit_queue);
- drm_atomic_state_free(state);
+ drm_atomic_state_put(state);
#if !defined(NV_DRM_MODE_CONFIG_FUNCS_HAS_ATOMIC_STATE_ALLOC)
nvidia_drm_free(requested_config);
@@ -983,7 +982,7 @@
* drm_atomic_commit().
*/
if (ret != 0) {
- drm_atomic_state_free(state);
+ drm_atomic_state_put(state);
}
drm_modeset_unlock_all(dev);
--- a/kernel/nvidia-drm/nvidia-drm-priv.h 2016-12-09 02:13:39.000000000 +0100
+++ b/kernel/nvidia-drm/nvidia-drm-priv.h 2016-12-15 11:51:32.638165272 +0100
@@ -34,7 +34,7 @@
#endif
#if defined(NV_DRM_DRIVER_HAS_GEM_PRIME_RES_OBJ)
-#include <linux/fence.h>
+#include <linux/dma-fence.h>
#include <linux/reservation.h>
#endif
--- a/kernel/nvidia-uvm/uvm8.c 2016-12-09 02:17:46.000000000 +0100
+++ b/kernel/nvidia-uvm/uvm8.c 2016-12-15 11:51:32.638165272 +0100
@@ -101,7 +101,7 @@
// so we force it to fail instead.
static int uvm_vm_fault_sigbus(struct vm_area_struct *vma, struct vm_fault *vmf)
{
- UVM_DBG_PRINT_RL("Fault to address 0x%p in disabled vma\n", vmf->virtual_address);
+ UVM_DBG_PRINT_RL("Fault to address 0x%p in disabled vma\n", vmf->address);
vmf->page = NULL;
return VM_FAULT_SIGBUS;
}
@@ -315,7 +315,7 @@
{
uvm_va_space_t *va_space = uvm_va_space_get(vma->vm_file);
uvm_va_block_t *va_block;
- NvU64 fault_addr = (NvU64)(uintptr_t)vmf->virtual_address;
+ NvU64 fault_addr = (NvU64)(uintptr_t)vmf->address;
bool is_write = vmf->flags & FAULT_FLAG_WRITE;
NV_STATUS status = uvm_global_get_status();
bool tools_enabled;
--- a/kernel/nvidia-uvm/uvm8_test.c 2016-12-09 02:17:46.000000000 +0100
+++ b/kernel/nvidia-uvm/uvm8_test.c 2016-12-15 11:51:32.639165272 +0100
@@ -103,7 +103,7 @@
return NV_ERR_INVALID_STATE;
}
-static NV_STATUS uvm8_test_get_kernel_virtual_address(
+static NV_STATUS uvm8_test_get_kernel_address(
UVM_TEST_GET_KERNEL_VIRTUAL_ADDRESS_PARAMS *params,
struct file *filp)
{
@@ -173,7 +173,7 @@
UVM_ROUTE_CMD_STACK(UVM_TEST_RANGE_GROUP_RANGE_COUNT, uvm8_test_range_group_range_count);
UVM_ROUTE_CMD_STACK(UVM_TEST_GET_PREFETCH_FAULTS_REENABLE_LAPSE, uvm8_test_get_prefetch_faults_reenable_lapse);
UVM_ROUTE_CMD_STACK(UVM_TEST_SET_PREFETCH_FAULTS_REENABLE_LAPSE, uvm8_test_set_prefetch_faults_reenable_lapse);
- UVM_ROUTE_CMD_STACK(UVM_TEST_GET_KERNEL_VIRTUAL_ADDRESS, uvm8_test_get_kernel_virtual_address);
+ UVM_ROUTE_CMD_STACK(UVM_TEST_GET_KERNEL_VIRTUAL_ADDRESS, uvm8_test_get_kernel_address);
UVM_ROUTE_CMD_STACK(UVM_TEST_PMA_ALLOC_FREE, uvm8_test_pma_alloc_free);
UVM_ROUTE_CMD_STACK(UVM_TEST_PMM_ALLOC_FREE_ROOT, uvm8_test_pmm_alloc_free_root);
UVM_ROUTE_CMD_STACK(UVM_TEST_PMM_INJECT_PMA_EVICT_ERROR, uvm8_test_pmm_inject_pma_evict_error);
--- a/kernel/nvidia-uvm/uvm_lite.c 2016-12-09 02:17:46.000000000 +0100
+++ b/kernel/nvidia-uvm/uvm_lite.c 2016-12-15 11:51:32.639165272 +0100
@@ -1333,7 +1333,7 @@
#if defined(NV_VM_OPERATIONS_STRUCT_HAS_FAULT)
int _fault(struct vm_area_struct *vma, struct vm_fault *vmf)
{
- unsigned long vaddr = (unsigned long)vmf->virtual_address;
+ unsigned long vaddr = (unsigned long)vmf->address;
struct page *page = NULL;
int retval;目前在 4.10-rc2 内核上也测试通过。