gpu:nvgpu: fix the nvs mmap issues

- As part of mmap call handler mmap_sem is being acquired which is
causing an BUG: scheduling while atomic:. This is happening because
mmap callback gets called with mmap_sem held. The mmap file handler
is called via call_mmap() via mmap_region() via do_mmap(), and the
caller of do_mmap() must hold down_write(&current->mm->mmap_sem).
To fix this issue removing the mmap_sem locking from nvs mmap handler.

- If remap_vmalloc_range() is used to map the pages to userspace then
allocated pages should be mapped into virtually contiguous space by
passing VM_USERMAP flag to vmap(). Passing VM_USERMAP to vmap() call
if API nvgpu_dma_alloc_flags_sys() is called with flag
NVGPU_DMA_VM_USERMAP_ADDRESS.

Bug 3884011

Change-Id: I64264f6f6e0dd75b1f828dc58355d740e6ef5ccb
Signed-off-by: prsethi <prsethi@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2820781
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/common/nvs/nvs_sched_ctrl.c

@@ -507,7 +507,8 @@ int nvgpu_nvs_buffer_alloc(struct nvgpu_nvs_domain_ctrl_fifo *sched_ctrl,
 	(void)memset(buf, 0, sizeof(*buf));
 	buf->g = g;
 
-	err = nvgpu_dma_alloc_map_sys(system_vm, bytes, &buf->mem);
+	err = nvgpu_dma_alloc_map_flags_sys(system_vm,
+		NVGPU_DMA_VM_USERMAP_ADDRESS, bytes, &buf->mem);
 	if (err != 0) {
 		nvgpu_err(g, "failed to allocate memory for dma");
 		goto fail;

drivers/gpu/nvgpu/include/nvgpu/dma.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2021, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2017-2022, NVIDIA CORPORATION.  All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -62,6 +62,13 @@ struct nvgpu_mem;
  */
 #define NVGPU_DMA_PHYSICALLY_ADDRESSED	BIT64(2)
 
+/**
+ * If remap_vmalloc_range() is used to map the pages to userspace then allocated
+ * pages should be mapped into virtually contiguous space by passing VM_USERMAP
+ * flag to vmap().
+ */
+#define NVGPU_DMA_VM_USERMAP_ADDRESS	BIT64(3)
+
 
 /**
  * nvgpu_iommuable - Check if GPU is behind IOMMU

drivers/gpu/nvgpu/os/linux/linux-dma.c

@@ -260,6 +260,7 @@ int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
 	gfp_t gfps = GFP_KERNEL|__GFP_ZERO;
 	dma_addr_t iova;
 	unsigned long dma_attrs = 0;
+	unsigned long vma_flags = 0;
 	void *alloc_ret;
 	int err;
 
@@ -309,8 +310,18 @@ int nvgpu_dma_alloc_flags_sys(struct gk20a *g, unsigned long flags,
 
 	/* Map the page list from the non-contiguous allocation */
 	if (nvgpu_nvlink_non_contig(g, flags)) {
+		/*
+		 * If remap_vmalloc_range() is going to be used to map cpu_va to
+		 * userspace then VM_USERMAP flag should be passed to vmap() to
+		 * get appropriate cpu_va.
+		 */
+		if (flags & NVGPU_DMA_VM_USERMAP_ADDRESS) {
+			vma_flags |= VM_USERMAP;
+		}
+
 		mem->cpu_va = vmap(mem->priv.pages, size >> PAGE_SHIFT,
-				   0, PAGE_KERNEL);
+			   vma_flags, PAGE_KERNEL);
+
 		if (!mem->cpu_va) {
 			err = -ENOMEM;
 			goto fail_free_sgt;
@@ -365,12 +376,6 @@ int nvgpu_dma_mmap_sys(struct gk20a *g, struct vm_area_struct *vma, struct nvgpu
 		return -EINVAL;
 	}
 
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0)
-	mmap_write_lock(vma->vm_mm);
-#else
-	down_write(&vma->vm_mm->mmap_sem);
-#endif
-
 	vma_exists = find_vma_intersection(vma->vm_mm, vma->vm_start, vma->vm_end);
 	if (vma_exists != NULL) {
 		err = -EEXIST;
@@ -395,14 +400,9 @@ int nvgpu_dma_mmap_sys(struct gk20a *g, struct vm_area_struct *vma, struct nvgpu
 	}
 
 done:
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(5, 8, 0)
-	mmap_write_unlock(vma->vm_mm);
-#else
-	up_write(&vma->vm_mm->mmap_sem);
-#endif
 
 	if (err != 0) {
-		nvgpu_err(g, "failed to map mem into userspace vma");
+		nvgpu_err(g, "failed to map mem into userspace vma %d", err);
 	}
 
 	return err;