allocate_timestamps_buffers(9)

💫 allocate_timestamps_buffers(9) — Kernel API Manual, but make it make sense 🎀

okay so bestie, welcome to allocate_timestamps_buffers(9) — the Kernel API Manual entry that's literally all about setting up your timestamp buffers before anybody gets to peek at them. no cap, it's giving "prep work that saves you later" energy, and honestly? iconic. 🦋

---

✨ NAME ✨

allocate_timestamps_buffers — allocate timestamps buffers 💅

so this function? it will allocate the ts buffer that will later on be mapped to the user, so the user can actually read the timestamp. that's the whole point, bestie. ⭐

in addition (and this part is lowkey important), it'll allocate an extra buffer for registration management. why? because we cannot fail during registration for an out-of-memory situation 😭 — so we'll prepare a pool which will be used as user interrupt nodes, and instead of dynamically allocating nodes while registration, we'll pick the node from this pool. the way that saves you from an OOM disaster mid-registration? obsessed. 🔥

in addition (!), it'll add the node to the mapping hash, which will be used to map the user ts buffer to the internal kernel ts buffer. so yeah, everybody gets matched up properly. 💖

---

🌸 SYNOPSIS 🌸

int allocate_timestamps_buffers (struct hl_fpriv *hpriv , struct
hl_mem_in *args , u64 *handle );

like, yes, that's the whole signature, and yes, it returns an int. periodt. 👑

---

🎀 ARGUMENTS 🎀

• hpriv — pointer to the private data of the fd 🫶 (the fd's whole personal lore lives here)
• args — ioctl input 🌈 (this is what came in through the ioctl)
• handle — user timestamp buffer handle as an output ✨ (it comes back out to you, bestie!)

---

🧚 SEE ALSO 🧚

Kernel file ./drivers/accel/habanalabs/common/memory.c — that's where the source lives, go read it iykyk 💜

and then the whole friend group of related symbols, in exactly the order they were introduced at the party:

• dma_map_host_va(9)
• dma_unmap_host_va(9)
• dram_pg_pool_do_release(9)
• free_phys_pg_pack(9)
• free_device_memory(9)
• clear_va_list_locked(9)
• print_va_list_locked(9)
• merge_va_blocks_locked(9)
• add_va_block_locked(9)
• add_va_block(9)
• is_hint_crossing_range(9)
• get_va_block(9)
• hl_get_va_range_type(9)
• hl_unreserve_va_block(9)
• init_phys_pg_pack_from_userptr(9)
• map_phys_pg_pack(9)
• unmap_phys_pg_pack(9)
• map_device_va(9)
• unmap_device_va(9)
• hl_hw_block_mmap(9)
• export_dmabuf_from_addr(9)
• hl_pin_host_memory(9)
• hl_userptr_delete_list(9)
• hl_userptr_is_pinned(9)
• va_range_init(9)
• va_range_fini(9)
• vm_ctx_init_with_ranges(9)
• hl_vm_ctx_fini(9)
• hl_vm_init(9)
• hl_vm_fini(9)
• hl_hw_block_mem_init(9)
• hl_hw_block_mem_fini(9)

---

bestie reminder: the pool situation isn't just cute architecture — it's there because registration can't fail on out-of-memory. if you're poking at this code, that constraint is the whole vibe check. 🫶

---

and that's the page!! mwah 💋 this one comes from ./drivers/accel/habanalabs/common, dated October 2026 — so like, future-era kernel lore, slay. go forth and allocate your timestamps, bestie, you've got this 🔥👑🌸