Memory Management¶

BO management¶

TTM manages (placement, eviction, etc...) all BOs in XE.

BO creation¶

Create a chunk of memory which can be used by the GPU. Placement rules (sysmem or vram region) passed in upon creation. TTM handles placement of BO and can trigger eviction of other BOs to make space for the new BO.

Kernel BOs¶

A kernel BO is created as part of driver load (e.g. uC firmware images, GuC ADS, etc...) or a BO created as part of a user operation which requires a kernel BO (e.g. engine state, memory for page tables, etc...). These BOs are typically mapped in the GGTT (any kernel BOs aside memory for page tables are in the GGTT), are pinned (can’t move or be evicted at runtime), have a vmap (XE can access the memory via xe_map layer) and have contiguous physical memory.

More details of why kernel BOs are pinned and contiguous below.

User BOs¶

A user BO is created via the DRM_IOCTL_XE_GEM_CREATE IOCTL. Once it is created the BO can be mmap’d (via DRM_IOCTL_XE_GEM_MMAP_OFFSET) for user access and it can be bound for GPU access (via DRM_IOCTL_XE_VM_BIND). All user BOs are evictable and user BOs are never pinned by XE. The allocation of the backing store can be defered from creation time until first use which is either mmap, bind, or pagefault.

Private BOs¶

A private BO is a user BO created with a valid VM argument passed into the create IOCTL. If a BO is private it cannot be exported via prime FD and mappings can only be created for the BO within the VM it is tied to. Lastly, the BO dma-resv slots / lock point to the VM’s dma-resv slots / lock (all private BOs to a VM share common dma-resv slots / lock).

External BOs¶

An external BO is a user BO created with a NULL VM argument passed into the create IOCTL. An external BO can be shared with different UMDs / devices via prime FD and the BO can be mapped into multiple VMs. An external BO has its own unique dma-resv slots / lock. An external BO will be in an array of all VMs which has a mapping of the BO. This allows VMs to lookup and lock all external BOs mapped in the VM as needed.

BO placement¶

When a user BO is created, a mask of valid placements is passed indicating which memory regions are considered valid.

The memory region information is available via query uAPI (TODO: add link).

BO validation¶

BO validation (ttm_bo_validate) refers to ensuring a BO has a valid placement. If a BO was swapped to temporary storage, a validation call will trigger a move back to a valid (location where GPU can access BO) placement. Validation of a BO may evict other BOs to make room for the BO being validated.

BO eviction / moving¶

All eviction (or in other words, moving a BO from one memory location to another) is routed through TTM with a callback into XE.

Runtime eviction¶

Runtime evictions refers to during normal operations where TTM decides it needs to move a BO. Typically this is because TTM needs to make room for another BO and the evicted BO is first BO on LRU list that is not locked.

An example of this is a new BO which can only be placed in VRAM but there is not space in VRAM. There could be multiple BOs which have sysmem and VRAM placement rules which currently reside in VRAM, TTM trigger a will move of one (or multiple) of these BO(s) until there is room in VRAM to place the new BO. The evicted BO(s) are valid but still need new bindings before the BO used again (exec or compute mode rebind worker).

Another example would be, TTM can’t find a BO to evict which has another valid placement. In this case TTM will evict one (or multiple) unlocked BO(s) to a temporary unreachable (invalid) placement. The evicted BO(s) are invalid and before next use need to be moved to a valid placement and rebound.

In both cases, moves of these BOs are scheduled behind the fences in the BO’s dma-resv slots.

WW locking tries to ensures if 2 VMs use 51% of the memory forward progress is made on both VMs.

Runtime eviction uses per a GT migration engine (TODO: link to migration engine doc) to do a GPU memcpy from one location to another.

Rebinds after runtime eviction¶

When BOs are moved, every mapping (VMA) of the BO needs to rebound before the BO is used again. Every VMA is added to an evicted list of its VM when the BO is moved. This is safe because of the VM locking structure (TODO: link to VM locking doc). On the next use of a VM (exec or compute mode rebind worker) the evicted VMA list is checked and rebinds are triggered. In the case of faulting VM, the rebind is done in the page fault handler.

Suspend / resume eviction of VRAM¶

During device suspend / resume VRAM may lose power which means the contents of VRAM’s memory is blown away. Thus BOs present in VRAM at the time of suspend must be moved to sysmem in order for their contents to be saved.

A simple TTM call (ttm_resource_manager_evict_all) can move all non-pinned (user) BOs to sysmem. External BOs that are pinned need to be manually evicted with a simple loop + xe_bo_evict call. It gets a little trickier with kernel BOs.

Some kernel BOs are used by the GT migration engine to do moves, thus we can’t move all of the BOs via the GT migration engine. For simplity, use a TTM memcpy (CPU) to move any kernel (pinned) BO on either suspend or resume.

Some kernel BOs need to be restored to the exact same physical location. TTM makes this rather easy but the caveat is the memory must be contiguous. Again for simplity, we enforce that all kernel (pinned) BOs are contiguous and restored to the same physical location.

Pinned external BOs in VRAM are restored on resume via the GPU.

Rebinds after suspend / resume¶

Most kernel BOs have GGTT mappings which must be restored during the resume process. All user BOs are rebound after validation on their next use.

Future work¶

Trim the list of BOs which is saved / restored via TTM memcpy on suspend / resume. All we really need to save / restore via TTM memcpy is the memory required for the GuC to load and the memory for the GT migrate engine to operate.

Do not require kernel BOs to be contiguous in physical memory / restored to the same physical address on resume. In all likelihood the only memory that needs to be restored to the same physical address is memory used for page tables. All of that memory is allocated 1 page at time so the contiguous requirement isn’t needed. Some work on the vmap code would need to be done if kernel BOs are not contiguous too.

Make some kernel BO evictable rather than pinned. An example of this would be engine state, in all likelihood if the dma-slots of these BOs where properly used rather than pinning we could safely evict + rebind these BOs as needed.

Some kernel BOs do not need to be restored on resume (e.g. GuC ADS as that is repopulated on resume), add flag to mark such objects as no save / restore.

GGTT¶

Xe GGTT implements the support for a Global Virtual Address space that is used for resources that are accessible to privileged (i.e. kernel-mode) processes, and not tied to a specific user-level process. For example, the Graphics micro-Controller (GuC) and Display Engine (if present) utilize this Global address space.

The Global GTT (GGTT) translates from the Global virtual address to a physical address that can be accessed by HW. The GGTT is a flat, single-level table.

Xe implements a simplified version of the GGTT specifically managing only a certain range of it that goes from the Write Once Protected Content Memory (WOPCM) Layout to a predefined GUC_GGTT_TOP. This approach avoids complications related to the GuC (Graphics Microcontroller) hardware limitations. The GuC address space is limited on both ends of the GGTT, because the GuC shim HW redirects accesses to those addresses to other HW areas instead of going through the GGTT. On the bottom end, the GuC can’t access offsets below the WOPCM size, while on the top side the limit is fixed at GUC_GGTT_TOP. To keep things simple, instead of checking each object to see if they are accessed by GuC or not, we just exclude those areas from the allocator. Additionally, to simplify the driver load, we use the maximum WOPCM size in this logic instead of the programmed one, so we don’t need to wait until the actual size to be programmed is determined (which requires FW fetch) before initializing the GGTT. These simplifications might waste space in the GGTT (about 20-25 MBs depending on the platform) but we can live with this. Another benefit of this is the GuC bootrom can’t access anything below the WOPCM max size so anything the bootrom needs to access (e.g. a RSA key) needs to be placed in the GGTT above the WOPCM max size. Starting the GGTT allocations above the WOPCM max give us the correct placement for free.

GGTT Internal API¶

struct xe_ggtt¶: Main GGTT struct

Definition:

struct xe_ggtt {
    struct xe_tile *tile;
    u64 size;
#define XE_GGTT_FLAGS_64K BIT(0);
    unsigned int flags;
    struct xe_bo *scratch;
    struct mutex lock;
    u64 __iomem *gsm;
    const struct xe_ggtt_pt_ops *pt_ops;
    struct drm_mm mm;
    unsigned int access_count;
    struct workqueue_struct *wq;
};

Members

tile: Back pointer to tile where this GGTT belongs
size: Total size of this GGTT
flags: Flags for this GGTT Acceptable flags: - XE_GGTT_FLAGS_64K - if PTE size is 64K. Otherwise, regular is 4K.
scratch: Internal object allocation used as a scratch page
lock: Mutex lock to protect GGTT data
gsm: The iomem pointer to the actual location of the translation table located in the GSM for easy PTE manipulation
pt_ops: Page Table operations per platform
mm: The memory manager used to manage individual GGTT allocations
access_count: counts GGTT writes
wq: Dedicated unordered work queue to process node removals

Description

In general, each tile can contains its own Global Graphics Translation Table (GGTT) instance.

struct xe_ggtt_node¶: A node in GGTT.

Definition:

struct xe_ggtt_node {
    struct xe_ggtt *ggtt;
    struct drm_mm_node base;
    struct work_struct delayed_removal_work;
    bool invalidate_on_remove;
};

Members

ggtt: Back pointer to xe_ggtt where this region will be inserted at
base: A drm_mm_node
delayed_removal_work: The work struct for the delayed removal
invalidate_on_remove: If it needs invalidation upon removal

Description

This struct needs to be initialized (only-once) with xe_ggtt_node_init() before any node insertion, reservation, or ‘ballooning’. It will, then, be finalized by either xe_ggtt_node_remove() or xe_ggtt_node_deballoon().

struct xe_ggtt_pt_ops¶: GGTT Page table operations Which can vary from platform to platform.

Definition:

struct xe_ggtt_pt_ops {
    u64 (*pte_encode_bo)(struct xe_bo *bo, u64 bo_offset, u16 pat_index);
    void (*ggtt_set_pte)(struct xe_ggtt *ggtt, u64 addr, u64 pte);
};

Members

pte_encode_bo: Encode PTE address for a given BO
ggtt_set_pte: Directly write into GGTT’s PTE

int xe_ggtt_init_early(struct xe_ggtt *ggtt)¶: Early GGTT initialization

Parameters

struct xe_ggtt *ggtt: the xe_ggtt to be initialized

Description

It allows to create new mappings usable by the GuC. Mappings are not usable by the HW engines, as it doesn’t have scratch nor initial clear done to it yet. That will happen in the regular, non-early GGTT initialization.

Return

0 on success or a negative error code on failure.

void xe_ggtt_node_remove(struct xe_ggtt_node *node, bool invalidate)¶: Remove a xe_ggtt_node from the GGTT

Parameters

struct xe_ggtt_node *node: the xe_ggtt_node to be removed
bool invalidate: if node needs invalidation upon removal

int xe_ggtt_init(struct xe_ggtt *ggtt)¶: Regular non-early GGTT initialization

Parameters

struct xe_ggtt *ggtt: the xe_ggtt to be initialized

Return

0 on success or a negative error code on failure.

int xe_ggtt_node_insert_balloon(struct xe_ggtt_node *node, u64 start, u64 end)¶: prevent allocation of specified GGTT addresses

Parameters

struct xe_ggtt_node *node: the xe_ggtt_node to hold reserved GGTT node
u64 start: the starting GGTT address of the reserved region
u64 end: then end GGTT address of the reserved region

Description

Use xe_ggtt_node_remove_balloon() to release a reserved GGTT node.

Return

0 on success or a negative error code on failure.

void xe_ggtt_node_remove_balloon(struct xe_ggtt_node *node)¶: release a reserved GGTT region

Parameters

struct xe_ggtt_node *node: the xe_ggtt_node with reserved GGTT region

Description

See xe_ggtt_node_insert_balloon() for details.

int xe_ggtt_node_insert_locked(struct xe_ggtt_node *node, u32 size, u32 align, u32 mm_flags)¶: Locked version to insert a xe_ggtt_node into the GGTT

Parameters

struct xe_ggtt_node *node: the xe_ggtt_node to be inserted
u32 size: size of the node
u32 align: alignment constrain of the node
u32 mm_flags: flags to control the node behavior

Description

It cannot be called without first having called xe_ggtt_init() once. To be used in cases where ggtt->lock is already taken.

Return

0 on success or a negative error code on failure.

int xe_ggtt_node_insert(struct xe_ggtt_node *node, u32 size, u32 align)¶: Insert a xe_ggtt_node into the GGTT

Parameters

struct xe_ggtt_node *node: the xe_ggtt_node to be inserted
u32 size: size of the node
u32 align: alignment constrain of the node

Description

It cannot be called without first having called xe_ggtt_init() once.

Return

0 on success or a negative error code on failure.

struct xe_ggtt_node *xe_ggtt_node_init(struct xe_ggtt *ggtt)¶: Initialize xe_ggtt_node struct

Parameters

struct xe_ggtt *ggtt: the xe_ggtt where the new node will later be inserted/reserved.

Description

This function will allocated the struct xe_ggtt_node and return it’s pointer. This struct will then be freed after the node removal upon xe_ggtt_node_remove() or xe_ggtt_node_remove_balloon(). Having xe_ggtt_node struct allocated doesn’t mean that the node is already allocated in GGTT. Only the xe_ggtt_node_insert(), xe_ggtt_node_insert_locked(), xe_ggtt_node_insert_balloon() will ensure the node is inserted or reserved in GGTT.

Return

A pointer to xe_ggtt_node struct on success. An ERR_PTR otherwise.

void xe_ggtt_node_fini(struct xe_ggtt_node *node)¶: Forcebly finalize xe_ggtt_node struct

Parameters

struct xe_ggtt_node *node: the xe_ggtt_node to be freed

Description

If anything went wrong with either xe_ggtt_node_insert(), xe_ggtt_node_insert_locked(), or xe_ggtt_node_insert_balloon(); and this node is not going to be reused, then, this function needs to be called to free the xe_ggtt_node struct

bool xe_ggtt_node_allocated(const struct xe_ggtt_node *node)¶: Check if node is allocated in GGTT

Parameters

const struct xe_ggtt_node *node: the xe_ggtt_node to be inspected

Return

True if allocated, False otherwise.

void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)¶: Map the BO into GGTT

Parameters

struct xe_ggtt *ggtt: the xe_ggtt where node will be mapped
struct xe_bo *bo: the xe_bo to be mapped

int xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo, u64 start, u64 end)¶: Insert BO at a specific GGTT space

Parameters

struct xe_ggtt *ggtt: the xe_ggtt where bo will be inserted
struct xe_bo *bo: the xe_bo to be inserted
u64 start: address where it will be inserted
u64 end: end of the range where it will be inserted

Return

0 on success or a negative error code on failure.

int xe_ggtt_insert_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)¶: Insert BO into GGTT

Parameters

struct xe_ggtt *ggtt: the xe_ggtt where bo will be inserted
struct xe_bo *bo: the xe_bo to be inserted

Return

0 on success or a negative error code on failure.

void xe_ggtt_remove_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)¶: Remove a BO from the GGTT

Parameters

struct xe_ggtt *ggtt: the xe_ggtt where node will be removed
struct xe_bo *bo: the xe_bo to be removed

u64 xe_ggtt_largest_hole(struct xe_ggtt *ggtt, u64 alignment, u64 *spare)¶: Largest GGTT hole

Parameters

struct xe_ggtt *ggtt: the xe_ggtt that will be inspected
u64 alignment: minimum alignment
u64 *spare: If not NULL: in: desired memory size to be spared / out: Adjusted possible spare

Return

size of the largest continuous GGTT region

void xe_ggtt_assign(const struct xe_ggtt_node *node, u16 vfid)¶: assign a GGTT region to the VF

Parameters

const struct xe_ggtt_node *node: the xe_ggtt_node to update
u16 vfid: the VF identifier

Description

This function is used by the PF driver to assign a GGTT region to the VF. In addition to PTE’s VFID bits 11:2 also PRESENT bit 0 is set as on some platforms VFs can’t modify that either.

int xe_ggtt_dump(struct xe_ggtt *ggtt, struct drm_printer *p)¶: Dump GGTT for debug

Parameters

struct xe_ggtt *ggtt: the xe_ggtt to be dumped
struct drm_printer *p: the drm_mm_printer helper handle to be used to dump the information

Return

0 on success or a negative error code on failure.

u64 xe_ggtt_print_holes(struct xe_ggtt *ggtt, u64 alignment, struct drm_printer *p)¶: Print holes

Parameters

struct xe_ggtt *ggtt: the xe_ggtt to be inspected
u64 alignment: min alignment
struct drm_printer *p: the drm_printer

Description

Print GGTT ranges that are available and return total size available.

Return

Total available size.

Pagetable building¶

Below we use the term “page-table” for both page-directories, containing pointers to lower level page-directories or page-tables, and level 0 page-tables that contain only page-table-entries pointing to memory pages.

When inserting an address range in an already existing page-table tree there will typically be a set of page-tables that are shared with other address ranges, and a set that are private to this address range. The set of shared page-tables can be at most two per level, and those can’t be updated immediately because the entries of those page-tables may still be in use by the gpu for other mappings. Therefore when inserting entries into those, we instead stage those insertions by adding insertion data into struct xe_vm_pgtable_update structures. This data, (subtrees for the cpu and page-table-entries for the gpu) is then added in a separate commit step. CPU-data is committed while still under the vm lock, the object lock and for userptr, the notifier lock in read mode. The GPU async data is committed either by the GPU or CPU after fulfilling relevant dependencies. For non-shared page-tables (and, in fact, for shared ones that aren’t existing at the time of staging), we add the data in-place without the special update structures. This private part of the page-table tree will remain disconnected from the vm page-table tree until data is committed to the shared page tables of the vm tree in the commit phase.

The Linux Kernel

Contents

This Page

Memory Management¶

BO management¶

BO creation¶

Kernel BOs¶

User BOs¶

Private BOs¶

External BOs¶

BO placement¶

BO validation¶

BO eviction / moving¶

Runtime eviction¶

Rebinds after runtime eviction¶

Suspend / resume eviction of VRAM¶

Rebinds after suspend / resume¶

Future work¶

GGTT¶

GGTT Internal API¶

Pagetable building¶