blob: 02f1c3b4bc5db32d77130a6fd9609982f2921f24 [file] [log] [blame]
/*
*
* (C) COPYRIGHT 2010, 2012-2019 ARM Limited. All rights reserved.
*
* This program is free software and is provided to you under the terms of the
* GNU General Public License version 2 as published by the Free Software
* Foundation, and any use by you of this program is subject to the terms
* of such GNU licence.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, you can access it online at
* http://www.gnu.org/licenses/gpl-2.0.html.
*
* SPDX-License-Identifier: GPL-2.0
*
*/
/**
* @file mali_kbase_mem_linux.h
* Base kernel memory APIs, Linux implementation.
*/
#ifndef _KBASE_MEM_LINUX_H_
#define _KBASE_MEM_LINUX_H_
/** A HWC dump mapping */
struct kbase_hwc_dma_mapping {
void *cpu_va;
dma_addr_t dma_pa;
size_t size;
};
/**
* kbase_mem_alloc - Create a new allocation for GPU
*
* @kctx: The kernel context
* @va_pages: The number of pages of virtual address space to reserve
* @commit_pages: The number of physical pages to allocate upfront
* @extent: The number of extra pages to allocate on each GPU fault which
* grows the region.
* @flags: bitmask of BASE_MEM_* flags to convey special requirements &
* properties for the new allocation.
* @gpu_va: Start address of the memory region which was allocated from GPU
* virtual address space.
*
* Return: 0 on success or error code
*/
struct kbase_va_region *kbase_mem_alloc(struct kbase_context *kctx,
u64 va_pages, u64 commit_pages, u64 extent, u64 *flags,
u64 *gpu_va);
/**
* kbase_mem_query - Query properties of a GPU memory region
*
* @kctx: The kernel context
* @gpu_addr: A GPU address contained within the memory region
* @query: The type of query, from KBASE_MEM_QUERY_* flags, which could be
* regarding the amount of backing physical memory allocated so far
* for the region or the size of the region or the flags associated
* with the region.
* @out: Pointer to the location to store the result of query.
*
* Return: 0 on success or error code
*/
int kbase_mem_query(struct kbase_context *kctx, u64 gpu_addr, u64 query,
u64 *const out);
/**
* kbase_mem_import - Import the external memory for use by the GPU
*
* @kctx: The kernel context
* @type: Type of external memory
* @phandle: Handle to the external memory interpreted as per the type.
* @padding: Amount of extra VA pages to append to the imported buffer
* @gpu_va: GPU address assigned to the imported external memory
* @va_pages: Size of the memory region reserved from the GPU address space
* @flags: bitmask of BASE_MEM_* flags to convey special requirements &
* properties for the new allocation representing the external
* memory.
* Return: 0 on success or error code
*/
int kbase_mem_import(struct kbase_context *kctx, enum base_mem_import_type type,
void __user *phandle, u32 padding, u64 *gpu_va, u64 *va_pages,
u64 *flags);
/**
* kbase_mem_alias - Create a new allocation for GPU, aliasing one or more
* memory regions
*
* @kctx: The kernel context
* @flags: bitmask of BASE_MEM_* flags.
* @stride: Bytes between start of each memory region
* @nents: The number of regions to pack together into the alias
* @ai: Pointer to the struct containing the memory aliasing info
* @num_pages: Number of pages the alias will cover
*
* Return: 0 on failure or otherwise the GPU VA for the alias
*/
u64 kbase_mem_alias(struct kbase_context *kctx, u64 *flags, u64 stride, u64 nents, struct base_mem_aliasing_info *ai, u64 *num_pages);
/**
* kbase_mem_flags_change - Change the flags for a memory region
*
* @kctx: The kernel context
* @gpu_addr: A GPU address contained within the memory region to modify.
* @flags: The new flags to set
* @mask: Mask of the flags, from BASE_MEM_*, to modify.
*
* Return: 0 on success or error code
*/
int kbase_mem_flags_change(struct kbase_context *kctx, u64 gpu_addr, unsigned int flags, unsigned int mask);
/**
* kbase_mem_commit - Change the physical backing size of a region
*
* @kctx: The kernel context
* @gpu_addr: Handle to the memory region
* @new_pages: Number of physical pages to back the region with
*
* Return: 0 on success or error code
*/
int kbase_mem_commit(struct kbase_context *kctx, u64 gpu_addr, u64 new_pages);
/**
* kbase_context_mmap - Memory map method, gets invoked when mmap system call is
* issued on device file /dev/malixx.
* @kctx: The kernel context
* @vma: Pointer to the struct containing the info where the GPU allocation
* will be mapped in virtual address space of CPU.
*
* Return: 0 on success or error code
*/
int kbase_context_mmap(struct kbase_context *kctx, struct vm_area_struct *vma);
/**
* kbase_mem_evictable_init - Initialize the Ephemeral memory eviction
* mechanism.
* @kctx: The kbase context to initialize.
*
* Return: Zero on success or -errno on failure.
*/
int kbase_mem_evictable_init(struct kbase_context *kctx);
/**
* kbase_mem_evictable_deinit - De-initialize the Ephemeral memory eviction
* mechanism.
* @kctx: The kbase context to de-initialize.
*/
void kbase_mem_evictable_deinit(struct kbase_context *kctx);
/**
* kbase_mem_grow_gpu_mapping - Grow the GPU mapping of an allocation
* @kctx: Context the region belongs to
* @reg: The GPU region
* @new_pages: The number of pages after the grow
* @old_pages: The number of pages before the grow
*
* Return: 0 on success, -errno on error.
*
* Expand the GPU mapping to encompass the new psychical pages which have
* been added to the allocation.
*
* Note: Caller must be holding the region lock.
*/
int kbase_mem_grow_gpu_mapping(struct kbase_context *kctx,
struct kbase_va_region *reg,
u64 new_pages, u64 old_pages);
/**
* kbase_mem_evictable_make - Make a physical allocation eligible for eviction
* @gpu_alloc: The physical allocation to make evictable
*
* Return: 0 on success, -errno on error.
*
* Take the provided region and make all the physical pages within it
* reclaimable by the kernel, updating the per-process VM stats as well.
* Remove any CPU mappings (as these can't be removed in the shrinker callback
* as mmap_sem might already be taken) but leave the GPU mapping intact as
* and until the shrinker reclaims the allocation.
*
* Note: Must be called with the region lock of the containing context.
*/
int kbase_mem_evictable_make(struct kbase_mem_phy_alloc *gpu_alloc);
/**
* kbase_mem_evictable_unmake - Remove a physical allocations eligibility for
* eviction.
* @alloc: The physical allocation to remove eviction eligibility from.
*
* Return: True if the allocation had its backing restored and false if
* it hasn't.
*
* Make the physical pages in the region no longer reclaimable and update the
* per-process stats, if the shrinker has already evicted the memory then
* re-allocate it if the region is still alive.
*
* Note: Must be called with the region lock of the containing context.
*/
bool kbase_mem_evictable_unmake(struct kbase_mem_phy_alloc *alloc);
struct kbase_vmap_struct {
off_t offset_in_page;
struct kbase_mem_phy_alloc *cpu_alloc;
struct kbase_mem_phy_alloc *gpu_alloc;
struct tagged_addr *cpu_pages;
struct tagged_addr *gpu_pages;
void *addr;
size_t size;
bool sync_needed;
};
/**
* kbase_vmap_prot - Map a GPU VA range into the kernel safely, only if the
* requested access permissions are supported
* @kctx: Context the VA range belongs to
* @gpu_addr: Start address of VA range
* @size: Size of VA range
* @prot_request: Flags indicating how the caller will then access the memory
* @map: Structure to be given to kbase_vunmap() on freeing
*
* Return: Kernel-accessible CPU pointer to the VA range, or NULL on error
*
* Map a GPU VA Range into the kernel. The VA range must be contained within a
* GPU memory region. Appropriate CPU cache-flushing operations are made as
* required, dependent on the CPU mapping for the memory region.
*
* This is safer than using kmap() on the pages directly,
* because the pages here are refcounted to prevent freeing (and hence reuse
* elsewhere in the system) until an kbase_vunmap()
*
* The flags in @prot_request should use KBASE_REG_{CPU,GPU}_{RD,WR}, to check
* whether the region should allow the intended access, and return an error if
* disallowed. This is essential for security of imported memory, particularly
* a user buf from SHM mapped into the process as RO. In that case, write
* access must be checked if the intention is for kernel to write to the
* memory.
*
* The checks are also there to help catch access errors on memory where
* security is not a concern: imported memory that is always RW, and memory
* that was allocated and owned by the process attached to @kctx. In this case,
* it helps to identify memory that was was mapped with the wrong access type.
*
* Note: KBASE_REG_GPU_{RD,WR} flags are currently supported for legacy cases
* where either the security of memory is solely dependent on those flags, or
* when userspace code was expecting only the GPU to access the memory (e.g. HW
* workarounds).
*
* All cache maintenance operations shall be ignored if the
* memory region has been imported.
*
*/
void *kbase_vmap_prot(struct kbase_context *kctx, u64 gpu_addr, size_t size,
unsigned long prot_request, struct kbase_vmap_struct *map);
/**
* kbase_vmap - Map a GPU VA range into the kernel safely
* @kctx: Context the VA range belongs to
* @gpu_addr: Start address of VA range
* @size: Size of VA range
* @map: Structure to be given to kbase_vunmap() on freeing
*
* Return: Kernel-accessible CPU pointer to the VA range, or NULL on error
*
* Map a GPU VA Range into the kernel. The VA range must be contained within a
* GPU memory region. Appropriate CPU cache-flushing operations are made as
* required, dependent on the CPU mapping for the memory region.
*
* This is safer than using kmap() on the pages directly,
* because the pages here are refcounted to prevent freeing (and hence reuse
* elsewhere in the system) until an kbase_vunmap()
*
* kbase_vmap_prot() should be used in preference, since kbase_vmap() makes no
* checks to ensure the security of e.g. imported user bufs from RO SHM.
*
* Note: All cache maintenance operations shall be ignored if the memory region
* has been imported.
*/
void *kbase_vmap(struct kbase_context *kctx, u64 gpu_addr, size_t size,
struct kbase_vmap_struct *map);
/**
* kbase_vunmap - Unmap a GPU VA range from the kernel
* @kctx: Context the VA range belongs to
* @map: Structure describing the mapping from the corresponding kbase_vmap()
* call
*
* Unmaps a GPU VA range from the kernel, given its @map structure obtained
* from kbase_vmap(). Appropriate CPU cache-flushing operations are made as
* required, dependent on the CPU mapping for the memory region.
*
* The reference taken on pages during kbase_vmap() is released.
*
* Note: All cache maintenance operations shall be ignored if the memory region
* has been imported.
*/
void kbase_vunmap(struct kbase_context *kctx, struct kbase_vmap_struct *map);
extern const struct vm_operations_struct kbase_vm_ops;
/**
* kbase_sync_mem_regions - Perform the cache maintenance for the kernel mode
* CPU mapping.
* @kctx: Context the CPU mapping belongs to.
* @map: Structure describing the CPU mapping, setup previously by the
* kbase_vmap() call.
* @dest: Indicates the type of maintenance required (i.e. flush or invalidate)
*
* Note: The caller shall ensure that CPU mapping is not revoked & remains
* active whilst the maintenance is in progress.
*/
void kbase_sync_mem_regions(struct kbase_context *kctx,
struct kbase_vmap_struct *map, enum kbase_sync_type dest);
/**
* kbase_mem_shrink_cpu_mapping - Shrink the CPU mapping(s) of an allocation
* @kctx: Context the region belongs to
* @reg: The GPU region
* @new_pages: The number of pages after the shrink
* @old_pages: The number of pages before the shrink
*
* Shrink (or completely remove) all CPU mappings which reference the shrunk
* part of the allocation.
*/
void kbase_mem_shrink_cpu_mapping(struct kbase_context *kctx,
struct kbase_va_region *reg,
u64 new_pages, u64 old_pages);
/**
* kbase_mem_shrink_gpu_mapping - Shrink the GPU mapping of an allocation
* @kctx: Context the region belongs to
* @reg: The GPU region or NULL if there isn't one
* @new_pages: The number of pages after the shrink
* @old_pages: The number of pages before the shrink
*
* Return: 0 on success, negative -errno on error
*
* Unmap the shrunk pages from the GPU mapping. Note that the size of the region
* itself is unmodified as we still need to reserve the VA, only the page tables
* will be modified by this function.
*/
int kbase_mem_shrink_gpu_mapping(struct kbase_context *kctx,
struct kbase_va_region *reg,
u64 new_pages, u64 old_pages);
/**
* kbase_phy_alloc_mapping_term - Terminate the kernel side mapping of a
* physical allocation
* @kctx: The kernel base context associated with the mapping
* @alloc: Pointer to the allocation to terminate
*
* This function will unmap the kernel mapping, and free any structures used to
* track it.
*/
void kbase_phy_alloc_mapping_term(struct kbase_context *kctx,
struct kbase_mem_phy_alloc *alloc);
/**
* kbase_phy_alloc_mapping_get - Get a kernel-side CPU pointer to the permanent
* mapping of a physical allocation
* @kctx: The kernel base context @gpu_addr will be looked up in
* @gpu_addr: The gpu address to lookup for the kernel-side CPU mapping
* @out_kern_mapping: Pointer to storage for a struct kbase_vmap_struct pointer
* which will be used for a call to
* kbase_phy_alloc_mapping_put()
*
* Return: Pointer to a kernel-side accessible location that directly
* corresponds to @gpu_addr, or NULL on failure
*
* Looks up @gpu_addr to retrieve the CPU pointer that can be used to access
* that location kernel-side. Only certain kinds of memory have a permanent
* kernel mapping, refer to the internal functions
* kbase_reg_needs_kernel_mapping() and kbase_phy_alloc_mapping_init() for more
* information.
*
* If this function succeeds, a CPU access to the returned pointer will access
* the actual location represented by @gpu_addr. That is, the return value does
* not require any offset added to it to access the location specified in
* @gpu_addr
*
* The client must take care to either apply any necessary sync operations when
* accessing the data, or ensure that the enclosing region was coherent with
* the GPU, or uncached in the CPU.
*
* The refcount on the physical allocations backing the region are taken, so
* that they do not disappear whilst the client is accessing it. Once the
* client has finished accessing the memory, it must be released with a call to
* kbase_phy_alloc_mapping_put()
*
* Whilst this is expected to execute quickly (the mapping was already setup
* when the physical allocation was created), the call is not IRQ-safe due to
* the region lookup involved.
*
* An error code may indicate that:
* - a userside process has freed the allocation, and so @gpu_addr is no longer
* valid
* - the region containing @gpu_addr does not support a permanent kernel mapping
*/
void *kbase_phy_alloc_mapping_get(struct kbase_context *kctx, u64 gpu_addr,
struct kbase_vmap_struct **out_kern_mapping);
/**
* kbase_phy_alloc_mapping_put - Put a reference to the kernel-side mapping of a
* physical allocation
* @kctx: The kernel base context associated with the mapping
* @kern_mapping: Pointer to a struct kbase_phy_alloc_mapping pointer obtained
* from a call to kbase_phy_alloc_mapping_get()
*
* Releases the reference to the allocations backing @kern_mapping that was
* obtained through a call to kbase_phy_alloc_mapping_get(). This must be used
* when the client no longer needs to access the kernel-side CPU pointer.
*
* If this was the last reference on the underlying physical allocations, they
* will go through the normal allocation free steps, which also includes an
* unmap of the permanent kernel mapping for those allocations.
*
* Due to these operations, the function is not IRQ-safe. However it is
* expected to execute quickly in the normal case, i.e. when the region holding
* the physical allocation is still present.
*/
void kbase_phy_alloc_mapping_put(struct kbase_context *kctx,
struct kbase_vmap_struct *kern_mapping);
/**
* kbase_get_cache_line_alignment - Return cache line alignment
*
* Helper function to return the maximum cache line alignment considering
* both CPU and GPU cache sizes.
*
* Return: CPU and GPU cache line alignment, in bytes.
*
* @kbdev: Device pointer.
*/
u32 kbase_get_cache_line_alignment(struct kbase_device *kbdev);
#if (KERNEL_VERSION(4, 20, 0) > LINUX_VERSION_CODE)
static inline vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma,
unsigned long addr, unsigned long pfn, pgprot_t pgprot)
{
int err;
#if ((KERNEL_VERSION(4, 4, 147) >= LINUX_VERSION_CODE) || \
((KERNEL_VERSION(4, 6, 0) > LINUX_VERSION_CODE) && \
(KERNEL_VERSION(4, 5, 0) <= LINUX_VERSION_CODE)))
if (pgprot_val(pgprot) != pgprot_val(vma->vm_page_prot))
return VM_FAULT_SIGBUS;
err = vm_insert_pfn(vma, addr, pfn);
#else
err = vm_insert_pfn_prot(vma, addr, pfn, pgprot);
#endif
if (unlikely(err == -ENOMEM))
return VM_FAULT_OOM;
if (unlikely(err < 0 && err != -EBUSY))
return VM_FAULT_SIGBUS;
return VM_FAULT_NOPAGE;
}
#endif
#endif /* _KBASE_MEM_LINUX_H_ */