gpu: nvgpu: modify alloc_as ioctl to accept mem size

- Modify NVGPU_GPU_IOCTL_ALLOC_AS and struct nvgpu_alloc_as_args to
accept start address and size of user memory. This allows configurable
address space allocation.
- Modify gk20a_as_alloc_share() and gk20a_vm_alloc_share() to receive
va_range_start and va_range_end values.
- gk20a_vm_alloc_share() initializes vm with low_hole = va_range_start,
and user vma size = (va_range_end - va_range_start).
- Modify nvgpu_as_alloc_space_args and nvgpu_as_free_space_args to
accept 64 bit number of pages.

Bug 2043269
JIRA NVGPU-5302

Change-Id: I243995adf5b7e0e84d6b36abe3b35a5ccabd7a37
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2385496
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: Sami Kiminki <skiminki@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: Sami Kiminki <skiminki@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/common/mm/as.c

@@ -50,7 +50,9 @@ static void release_as_share_id(struct gk20a_as_share *as_share)
 
 /* address space interfaces for the gk20a module */
 static int gk20a_vm_alloc_share(struct gk20a_as_share *as_share,
-				u32 big_page_size, u32 flags)
+				u32 big_page_size, u32 flags,
+				u64 va_range_start, u64 va_range_end,
+				u64 va_range_split)
 {
 	struct gk20a_as *as = as_share->as;
 	struct gk20a *g = gk20a_from_as(as);
@@ -58,6 +60,10 @@ static int gk20a_vm_alloc_share(struct gk20a_as_share *as_share,
 	struct vm_gk20a *vm;
 	char name[NVGPU_VM_NAME_LEN];
 	char *p;
+	u64 user_size;
+	u64 kernel_size = mm->channel.kernel_size;
+	u64 pde_size, pde_size_mask;
+	bool big_pages;
 	const bool userspace_managed =
 		(flags & NVGPU_AS_ALLOC_USERSPACE_MANAGED) != 0U;
 	const bool unified_va =
@@ -67,6 +73,7 @@ static int gk20a_vm_alloc_share(struct gk20a_as_share *as_share,
 	nvgpu_log_fn(g, " ");
 
 	if (big_page_size == 0U) {
+		big_pages = false;
 		big_page_size = g->ops.mm.gmmu.get_default_big_page_size();
 	} else {
 		if (!is_power_of_2(big_page_size)) {
@@ -77,21 +84,60 @@ static int gk20a_vm_alloc_share(struct gk20a_as_share *as_share,
 		     nvgpu_mm_get_available_big_page_sizes(g)) == 0U) {
 			return -EINVAL;
 		}
+		big_pages = true;
 	}
 
+	pde_size = BIT64(nvgpu_vm_pde_coverage_bit_count(g, big_page_size));
+	pde_size_mask = nvgpu_safe_sub_u64(pde_size, U64(1));
+
+	if ((va_range_start == 0ULL) ||
+		((va_range_start & pde_size_mask) != 0ULL)) {
+		return -EINVAL;
+	}
+
+	if ((va_range_end == 0ULL) ||
+		((va_range_end & pde_size_mask) != 0ULL)) {
+		return -EINVAL;
+	}
+
+	if (va_range_start >= va_range_end) {
+		return -EINVAL;
+	}
+
+	user_size = nvgpu_safe_sub_u64(va_range_end, va_range_start);
+
+	if (unified_va || !big_pages) {
+		if (va_range_split != 0ULL) {
+			return -EINVAL;
+		}
+	} else {
+		/* non-unified VA: split required */
+		if ((va_range_split == 0ULL) ||
+			((va_range_split & pde_size_mask) != 0ULL)) {
+			return -EINVAL;
+		}
+
+		/* non-unified VA: split range checks */
+		if ((va_range_split <= va_range_start) ||
+		    (va_range_split >= va_range_end)) {
+			return -EINVAL;
+		}
+	}
+
+	nvgpu_log_info(g,
+		"vm: low_hole=0x%llx, user_size=0x%llx, kernel_size=0x%llx",
+		va_range_start, user_size, kernel_size);
+
 	p = strncpy(name, "as_", sizeof("as_"));
 	(void) nvgpu_strnadd_u32(p, nvgpu_safe_cast_s32_to_u32(as_share->id),
 					sizeof(name) - sizeof("as_"), 10U);
 
 	vm = nvgpu_vm_init(g, big_page_size,
-			U64(big_page_size) << U64(10),
+			   va_range_start,
-			nvgpu_safe_sub_u64(mm->channel.user_size,
+			   user_size,
-				nvgpu_safe_sub_u64(mm->channel.kernel_size,
+			   kernel_size,
-				U64(big_page_size) << U64(10))),
+			   va_range_split,
-			mm->channel.kernel_size,
+			   big_pages, userspace_managed, unified_va, name);
-			nvgpu_gmmu_va_small_page_limit(),
-			!mm->disable_bigpage,
-			userspace_managed, unified_va, name);
 	if (vm == NULL) {
 		return -ENOMEM;
 	}
@@ -104,7 +150,8 @@ static int gk20a_vm_alloc_share(struct gk20a_as_share *as_share,
 }
 
 int gk20a_as_alloc_share(struct gk20a *g,
-			 u32 big_page_size, u32 flags,
+			u32 big_page_size, u32 flags, u64 va_range_start,
+			u64 va_range_end, u64 va_range_split,
 			struct gk20a_as_share **out)
 {
 	struct gk20a_as_share *as_share;
@@ -130,7 +177,8 @@ int gk20a_as_alloc_share(struct gk20a *g,
 	if (err != 0) {
 		goto failed;
 	}
-	err = gk20a_vm_alloc_share(as_share, big_page_size, flags);
+	err = gk20a_vm_alloc_share(as_share, big_page_size, flags,
+		va_range_start, va_range_end, va_range_split);
 	gk20a_idle(g);
 
 	if (err != 0) {

drivers/gpu/nvgpu/common/mm/vm_area.c

@@ -133,19 +133,19 @@ static int nvgpu_vm_area_alloc_get_pagesize_index(struct vm_gk20a *vm,
 }
 
 static int nvgpu_vm_area_alloc_memory(struct nvgpu_allocator *vma, u64 our_addr,
-			u32 pages, u32 page_size, u32 flags,
+			u64 pages, u32 page_size, u32 flags,
 			u64 *vaddr_start_ptr)
 {
 	u64 vaddr_start = 0;
 
 	if ((flags & NVGPU_VM_AREA_ALLOC_FIXED_OFFSET) != 0U) {
 		vaddr_start = nvgpu_alloc_fixed(vma, our_addr,
-						(u64)pages *
+						pages *
 						(u64)page_size,
 						page_size);
 	} else {
 		vaddr_start = nvgpu_alloc_pte(vma,
-					      (u64)pages *
+					      pages *
 					      (u64)page_size,
 					      page_size);
 	}
@@ -190,7 +190,7 @@ static int nvgpu_vm_area_alloc_gmmu_map(struct vm_gk20a *vm,
 	return 0;
 }
 
-int nvgpu_vm_area_alloc(struct vm_gk20a *vm, u32 pages, u32 page_size,
+int nvgpu_vm_area_alloc(struct vm_gk20a *vm, u64 pages, u32 page_size,
 			u64 *addr, u32 flags)
 {
 	struct gk20a *g = vm->mm->g;
@@ -211,7 +211,7 @@ int nvgpu_vm_area_alloc(struct vm_gk20a *vm, u32 pages, u32 page_size,
 	}
 
 	nvgpu_log(g, gpu_dbg_map,
-		  "ADD vm_area: pgsz=%#-8x pages=%-9u a/o=%#-14llx flags=0x%x",
+		  "ADD vm_area: pgsz=%#-8x pages=%-9llu a/o=%#-14llx flags=0x%x",
 		  page_size, pages, our_addr, flags);
 
 	if (nvgpu_vm_area_alloc_get_pagesize_index(vm, &pgsz_idx,
@@ -232,7 +232,7 @@ int nvgpu_vm_area_alloc(struct vm_gk20a *vm, u32 pages, u32 page_size,
 
 	vm_area->flags = flags;
 	vm_area->addr = vaddr_start;
-	vm_area->size = (u64)page_size * (u64)pages;
+	vm_area->size = (u64)page_size * pages;
 	vm_area->pgsz_idx = pgsz_idx;
 	nvgpu_init_list_node(&vm_area->buffer_list_head);
 	nvgpu_init_list_node(&vm_area->vm_area_list);

drivers/gpu/nvgpu/include/nvgpu/as.h

@@ -1,7 +1,7 @@
 /*
  * GK20A Address Spaces
  *
- * Copyright (c) 2011-2019, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2011-2020, 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"),
@@ -97,8 +97,17 @@ int gk20a_vm_release_share(struct gk20a_as_share *as_share);
  * @brief Allocate an AS share.
  *
  * @param g [in]			The GPU
- * @param big_page_size [in]	Big page size to use for the VM
+ * @param big_page_size [in]		Big page size to use for the VM,
+ *					set 0 for no big pages
  * @param flags [in]			NVGPU_AS_ALLOC_* flags
+ * @param va_range_start [in]		Requested user managed memory start
+ *					address, used to map buffers, save data
+ *					should be aligned by PDE
+ * @param va_range_end [in]		Requested user managed va end address
+ *					should be aligned by PDE
+ * @param va_range_split [in]		Requested small/big page split
+ *					should be aligned by PDE,
+ *					ignored if UNIFIED_VA is set
  * @params out [out]			The resulting, allocated, gk20a_as_share
  *					structure
  *
@@ -110,7 +119,9 @@ int gk20a_vm_release_share(struct gk20a_as_share *as_share);
  * @return 0 in case of success, < 0 in case of failure.
  */
 int gk20a_as_alloc_share(struct gk20a *g, u32 big_page_size,
-			 u32 flags, struct gk20a_as_share **out);
+			u32 flags, u64 va_range_start,
+			u64 va_range_end, u64 va_range_split,
+			struct gk20a_as_share **out);
 
 /**
  * @brief Retrieve the instance of gk20a from a gk20a_as instance.

drivers/gpu/nvgpu/include/nvgpu/vm_area.h

@@ -117,7 +117,7 @@ nvgpu_vm_area_from_vm_area_list(struct nvgpu_list_node *node)
  * @return		Zero, for successful allocation.
  *			Suitable error code for failures.
  */
-int nvgpu_vm_area_alloc(struct vm_gk20a *vm, u32 pages, u32 page_size,
+int nvgpu_vm_area_alloc(struct vm_gk20a *vm, u64 pages, u32 page_size,
 			u64 *addr, u32 flags);
 
 /**

drivers/gpu/nvgpu/os/linux/ioctl_as.c

@@ -21,6 +21,7 @@
 #include <uapi/linux/nvgpu.h>
 
 #include <nvgpu/gmmu.h>
+#include <nvgpu/mm.h>
 #include <nvgpu/vm_area.h>
 #include <nvgpu/log2.h>
 #include <nvgpu/gk20a.h>
@@ -305,15 +306,24 @@ int gk20a_as_dev_open(struct inode *inode, struct file *filp)
 {
 	struct gk20a_as_share *as_share;
 	struct gk20a *g;
+	struct mm_gk20a *mm;
 	int err;
 	struct nvgpu_cdev *cdev;
+	u32 big_page_size;
 
 	cdev = container_of(inode->i_cdev, struct nvgpu_cdev, cdev);
 	g = nvgpu_get_gk20a_from_cdev(cdev);
+	mm = &g->mm;
+	big_page_size = g->ops.mm.gmmu.get_default_big_page_size();
 
 	nvgpu_log_fn(g, " ");
 
-	err = gk20a_as_alloc_share(g, 0, 0, &as_share);
+	err = gk20a_as_alloc_share(g,
+		big_page_size,
+		NVGPU_AS_ALLOC_UNIFIED_VA,
+		U64(big_page_size) << U64(10),
+		mm->channel.user_size,
+		0ULL, &as_share);
 	if (err) {
 		nvgpu_log_fn(g, "failed to alloc share");
 		return err;

drivers/gpu/nvgpu/os/linux/ioctl_ctrl.c

@@ -594,6 +594,9 @@ static int gk20a_ctrl_alloc_as(
 	err = gk20a_as_alloc_share(g, args->big_page_size,
 				   gk20a_as_translate_as_alloc_flags(g,
 					   args->flags),
+				   args->va_range_start,
+				   args->va_range_end,
+				   args->va_range_split,
 				   &as_share);
 	if (err)
 		goto clean_up_file;

include/uapi/linux/nvgpu.h

@@ -392,7 +392,7 @@ struct nvgpu_gpu_mark_compressible_write_args {
 };
 
 struct nvgpu_alloc_as_args {
-	__u32 big_page_size;
+	__u32 big_page_size;	/* zero for no big pages for this VA */
 	__s32 as_fd;
 
 /*
@@ -409,8 +409,12 @@ struct nvgpu_alloc_as_args {
 #define NVGPU_GPU_IOCTL_ALLOC_AS_FLAGS_USERSPACE_MANAGED	(1 << 0)
 #define NVGPU_GPU_IOCTL_ALLOC_AS_FLAGS_UNIFIED_VA	 	(1 << 1)
 	__u32 flags;
-
 	__u32 reserved;		/* must be zero */
+	__u64 va_range_start;	/* in: starting VA (aligned by PDE) */
+	__u64 va_range_end;	/* in: ending VA (aligned by PDE) */
+	__u64 va_range_split;	/* in: small/big page split (aligned by PDE,
+				 * must be zero if UNIFIED_VA is set) */
+	__u32 padding[6];
 };
 
 struct nvgpu_gpu_open_tsg_args {
@@ -2037,14 +2041,14 @@ struct nvgpu32_as_alloc_space_args {
 };
 
 struct nvgpu_as_alloc_space_args {
-	__u32 pages;     /* in, pages */
+	__u64 pages;     /* in, pages */
 	__u32 page_size; /* in, bytes */
 	__u32 flags;     /* in */
-	__u32 padding;     /* in */
 	union {
 		__u64 offset; /* inout, byte address valid iff _FIXED_OFFSET */
 		__u64 align;  /* in, alignment multiple (0:={1 or n/a}) */
 	} o_a;
+	__u32 padding[2];     /* in */
 };
 
 /*
@@ -2055,8 +2059,9 @@ struct nvgpu_as_alloc_space_args {
  */
 struct nvgpu_as_free_space_args {
 	__u64 offset; /* in, byte address */
-	__u32 pages;     /* in, pages */
+	__u64 pages;     /* in, pages */
 	__u32 page_size; /* in, bytes */
+	__u32 padding[3];
 };
 
 /*

userspace/units/gr/setup/nvgpu-gr-setup.c

@@ -172,7 +172,10 @@ static int gr_test_setup_allocate_ch_tsg(struct unit_module *m,
 		goto ch_cleanup;
 	}
 
-	err = gk20a_as_alloc_share(g, 0, 0, &as_share);
+	err = gk20a_as_alloc_share(g,
+		0U, NVGPU_AS_ALLOC_UNIFIED_VA,
+		U64(SZ_4K) << U64(10),
+		(1ULL << 37), 0ULL, &as_share);
 	if (err != 0) {
 		unit_err(m, "failed vm memory alloc\n");
 		goto tsg_unbind;

userspace/units/mm/as/as.c

@@ -58,24 +58,28 @@ static int global_id_count;
 /* Parameters to test standard cases of allocation */
 static struct test_parameters test_64k_user_managed = {
 	.big_page_size = SZ_64K,
+	.small_big_split = (SZ_1G * 56ULL),
 	.flags = NVGPU_AS_ALLOC_USERSPACE_MANAGED,
 	.expected_error = 0
 };
 
 static struct test_parameters test_0k_user_managed = {
 	.big_page_size = 0,
+	.small_big_split = 0,
 	.flags = NVGPU_AS_ALLOC_USERSPACE_MANAGED,
 	.expected_error = 0
 };
 
 static struct test_parameters test_64k_unified_va = {
 	.big_page_size = SZ_64K,
+	.small_big_split = 0,
 	.flags = NVGPU_AS_ALLOC_UNIFIED_VA,
 	.expected_error = 0
 };
 
 static struct test_parameters test_64k_unified_va_enabled = {
 	.big_page_size = SZ_64K,
+	.small_big_split = 0,
 	.flags = 0,
 	.expected_error = 0,
 	.unify_address_spaces_flag = true
@@ -83,12 +87,14 @@ static struct test_parameters test_64k_unified_va_enabled = {
 
 static struct test_parameters test_einval_user_managed = {
 	.big_page_size = 1,
+	.small_big_split = (SZ_1G * 56ULL),
 	.flags = NVGPU_AS_ALLOC_USERSPACE_MANAGED,
 	.expected_error = -EINVAL
 };
 
 static struct test_parameters test_notp2_user_managed = {
 	.big_page_size = SZ_64K-1,
+	.small_big_split = (SZ_1G * 56ULL),
 	.flags = NVGPU_AS_ALLOC_USERSPACE_MANAGED,
 	.expected_error = -EINVAL
 };
@@ -96,6 +102,7 @@ static struct test_parameters test_notp2_user_managed = {
 /* Parameters to test corner cases and error handling */
 static struct test_parameters test_64k_user_managed_as_fail = {
 	.big_page_size = SZ_64K,
+	.small_big_split = (SZ_1G * 56ULL),
 	.flags = 0,
 	.expected_error = -ENOMEM,
 	.special_case = SPECIAL_CASE_AS_MALLOC_FAIL
@@ -103,6 +110,7 @@ static struct test_parameters test_64k_user_managed_as_fail = {
 
 static struct test_parameters test_64k_user_managed_vm_fail = {
 	.big_page_size = SZ_64K,
+	.small_big_split = (SZ_1G * 56ULL),
 	.flags = 0,
 	.expected_error = -ENOMEM,
 	.special_case = SPECIAL_CASE_VM_INIT_FAIL
@@ -110,6 +118,7 @@ static struct test_parameters test_64k_user_managed_vm_fail = {
 
 static struct test_parameters test_64k_user_managed_busy_fail_1 = {
 	.big_page_size = SZ_64K,
+	.small_big_split = (SZ_1G * 56ULL),
 	.flags = 0,
 	.expected_error = -ENODEV,
 	.special_case = SPECIAL_CASE_GK20A_BUSY_ALLOC
@@ -117,6 +126,7 @@ static struct test_parameters test_64k_user_managed_busy_fail_1 = {
 
 static struct test_parameters test_64k_user_managed_busy_fail_2 = {
 	.big_page_size = SZ_64K,
+	.small_big_split = (SZ_1G * 56ULL),
 	.flags = 0,
 	.expected_error = 0,
 	.special_case = SPECIAL_CASE_GK20A_BUSY_RELEASE
@@ -173,7 +183,7 @@ int test_init_mm(struct unit_module *m, struct gk20a *g, void *args)
 	 * Before ref_init calls to gk20a_as_alloc_share should immediately
 	 * fail.
 	 */
-	err = gk20a_as_alloc_share(g, 0, 0, NULL);
+	err = gk20a_as_alloc_share(g, 0, 0, 0, 0, 0, NULL);
 	if (err != -ENODEV) {
 		unit_return_fail(m, "gk20a_as_alloc_share did not fail as expected err=%d\n",
 				err);
@@ -213,7 +223,8 @@ int test_as_alloc_share(struct unit_module *m, struct gk20a *g, void *args)
 	}
 
 	err = gk20a_as_alloc_share(g, params->big_page_size,
-			 params->flags, &out);
+			params->flags, (SZ_64K << 10), (1ULL << 37),
+			params->small_big_split, &out);
 
 	if (params->unify_address_spaces_flag) {
 		nvgpu_set_enabled(g, NVGPU_MM_UNIFY_ADDRESS_SPACES, false);
@@ -264,7 +275,8 @@ int test_gk20a_from_as(struct unit_module *m, struct gk20a *g, void *args)
 	int err;
 
 	err = gk20a_as_alloc_share(g, SZ_64K, NVGPU_AS_ALLOC_USERSPACE_MANAGED,
-			 &out);
+			(SZ_64K << 10), (1ULL << 37),
+			nvgpu_gmmu_va_small_page_limit(), &out);
 	if (err != 0) {
 		unit_return_fail(m, "gk20a_as_alloc_share failed err=%d\n",
 			err);

userspace/units/mm/as/as.h

@@ -55,6 +55,11 @@ struct test_parameters {
 	 */
 	int big_page_size;
 
+	/**
+	 * Address for small big page vma split
+	 */
+	unsigned long long small_big_split;
+
 	/**
 	 * Flags to use when calling gk20a_as_alloc_share. Should be one of the
 	 * NVGPU_AS_ALLOC_* flag defined in as.h.