VM_FRAGMENT_SIZE (INT)(9)

💅 vm_fragment_size (int) — a page from the Kernel API Manual ✨

okay so bestie, welcome to vm_fragment_size (int)(9), your new fave module param for the amdgpu driver. it's giving "one tiny knob, big energy" and honestly? iconic. let's get into it 🎀

✨ NAME ✨

vm_fragment_size (int) — the little setting that lets you override the VM fragment size in bits 🦋

🌸 DESCRIPTION 🌸

so like, this param overrides the VM fragment size, measured in bits. the values go 4, 5, etc. — and in case your brain needs the cheat sheet: 4 = 64K, 9 = 2M 🔥

the default is -1, which lowkey means automatic for each asic — like, the driver just handles it per-asic and you don't have to think about it. main character energy of defaults, honestly 👑

ngl, if you're not sure, leave it at -1 and let the asic do its thing. no cap 💀

👑 SEE ALSO 👑

fr fr, there's a whole squad of other kernel params hanging out in the kernel file ./drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c — go say hi to all of them 💖

• vramlimit (int)(9)
• vis_vramlimit (int)(9)
• gartsize (uint)(9)
• gttsize (int)(9)
• moverate (int)(9)
• audio (int)(9)
• disp_priority (int)(9)
• hw_i2c (int)(9)
• pcie_gen2 (int)(9)
• msi (int)(9)
• svm_default_granularity (uint)(9)
• lockup_timeout (string)(9)
• dpm (int)(9)
• fw_load_type (int)(9)
• aspm (int)(9)
• runpm (int)(9)
• ip_block_mask (uint)(9)
• bapm (int)(9)
• deep_color (int)(9)
• vm_size (int)(9)
• vm_block_size (int)(9)
• vm_fault_stop (int)(9)
• vm_update_mode (int)(9)
• exp_hw_support (int)(9)
• dc (int)(9)
• sched_jobs (int)(9)
• sched_hw_submission (int)(9)
• ppfeaturemask (hexint)(9)
• forcelongtraining (uint)(9)
• pcie_gen_cap (uint)(9)
• pcie_lane_cap (uint)(9)
• cg_mask (ullong)(9)
• pg_mask (uint)(9)
• sdma_phase_quantum (uint)(9)
• disable_cu (charp)(9)
• virtual_display (charp)(9)
• lbpw (int)(9)
• gpu_recovery (int)(9)
• emu_mode (int)(9)
• ras_enable (int)(9)
• ras_mask (uint)(9)
• timeout_fatal_disable (bool)(9)
• timeout_period (uint)(9)
• si_support (int)(9)
• cik_support (int)(9)
• smu_memory_pool_size (uint)(9)
• async_gfx_ring (int)(9)
• mcbp (int)(9)
• discovery (int)(9)
• mes_log_enable (int)(9)
• uni_mes (int)(9)
• noretry (int)(9)
• force_asic_type (int)(9)
• use_xgmi_p2p (int)(9)
• sched_policy (int)(9)
• hws_max_conc_proc (int)(9)
• cwsr_enable (int)(9)
• max_num_of_queues_per_device (int)(9)
• send_sigterm (int)(9)
• halt_if_hws_hang (int)(9)
• hws_gws_support(bool)(9)
• queue_preemption_timeout_ms (int)(9)
• debug_evictions(bool)(9)
• no_system_mem_limit(bool)(9)
• no_queue_eviction_on_vm_fault (int)(9)
• mtype_local (int)(9)
• pcie_p2p (bool)(9)
• dcfeaturemask (uint)(9)
• dcdebugmask (uint)(9)
• abmlevel (uint)(9)
• damageclips (int)(9)
• tmz (int)(9)
• freesync_video (uint)(9)
• reset_method (int)(9)
• bad_page_threshold (int) Bad page threshold is specifies the(9)
• vcnfw_log (int)(9)
• sg_display (int)(9)
• umsch_mm (int)(9)
• umsch_mm_fwlog (int)(9)
• smu_pptable_id (int)(9)
• partition_mode (int)(9)
• enforce_isolation (int)(9)
• modeset (int)(9)
• seamless (int)(9)
• debug_mask (uint)(9)
• agp (int)(9)
• wbrf (int)(9)
• rebar (int)(9)
• user_queue (int)(9)
• ptl (int)(9)

that's the whole page, bestie — go forth and fragment responsibly 🫶

mwah 💋 this one's from ./drivers/gpu/drm/amd/amdgpu, dated October 2026 — stamped, sealed, slayed 🎀✨