nvkm_gsp_radix3_sg(9)

💅 nvkm_gsp_radix3_sg — like, the kernel API manual one

okay so bestie, welcome to nvkm_gsp_radix3_sg(9) 💖 this is the page where we build a radix3 table out of a scatter/gather list, and honestly? it's giving main character energy. stay with me, i promise it slaps.

---

✨ NAME ✨

nvkm_gsp_radix3_sg — build a radix3 table from a S/G list 🎀

---

🌸 SYNOPSIS 🌸

int nvkm_gsp_radix3_sg (struct nvkm_gsp *gsp , struct sg_table *sgt ,
u64 size , struct nvkm_gsp_radix3 *rx3 );

like, that's the whole vibe check right there, no cap ✨

---

🦋 ARGUMENTS 🦋

• gsp — the gsp pointer 🫶 that's it, that's the descriptor
• sgt — the S/G list we're gonna traverse, walk it like a runway 💅
• size — how big the image is, in bytes. yes, actual bytes
• rx3 — the radix3 array we're updating. bestie is getting a glow-up in place ✨

---

👑 DESCRIPTION 👑

okay so the GSP uses a three-level page table, and it's literally called radix3, and its whole job is mapping the firmware 🔥 each 64-bit "pointer" in the table is one of two things, and you gotta know which:

• the bus address of an entry in the next table — that's for levels 0 and 1
• the bus address of the next page in the GSP firmware image itself

💜 Level 0

level 0 contains a single entry, in one page, that points to the first page of level 1. one entry. iconic minimalism 🎀

💜 Level 1

level 1, since it's also only one page in size, contains up to 512 entries, one for each page in level 2. slay, a whole 512 of them.

💜 Level 2

level 2 can be up to 512 pages in size, and each of those entries points to the next page of the firmware image. and since there can be up to 512*512 pages, that caps the firmware at:

512*512*GSP_PAGE_SIZE = 1GB

that's the ceiling, bestie. 1GB. no notes. 💀

🧚 the whole memory situation

internally, the GSP has its window into system memory, but the base physical address of the aperture is not 0 — like, at all. in fact, it varies depending on the GPU architecture 🙈 and since the GPU is a PCI device, this window is accessed via DMA, so it's bound by IOMMU translation. the end result? GSP-RM must translate the bus addresses in the table into GSP physical addresses. all of this should happen transparently, so don't stress about it 🌈

⭐ returns

returns 0 on success, or a negative error code 😭 (we love a clear signal)

also, see kgspCreateRadix3_IMPL — that's the reference, iykyk 👑

---

🍓 SEE ALSO 🍓

the kernel file, first and foremost:

./drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c

and then the whole friend group, fr fr:

• nvkm_gsp_mem_ctor(9)
• struct registry_list_entry(9)
• add_registry(9)
• build_registry(9)
• clean_registry(9)
• strip(9)
• r535_gsp_rpc_set_registry(9)
• r535_gsp_msg_run_cpu_sequencer(9)
• struct rpc_ucode_libos_print_v1e_08(9)
• r535_gsp_msg_libos_print(9)
• create_debugfs(9)
• r535_gsp_libos_debugfs_init(9)
• create_pte_array(9)
• r535_gsp_libos_init(9)
• r535_debugfs_shutdown(9)
• is_empty(9)
• r535_gsp_copy_log(9)
• r535_gsp_retain_logging(9)
• r535_gsp_libos_debugfs_fini(9)

---

💖 and that's the page!! this one came out of ./drivers/gpu/drm/nouveau/nvkm... and it's dated October 2026, so mark your calendars bestie 🗓️✨ stay hydrated, read your page tables, and remember: 512*512*GSP_PAGE_SIZE = 1GB. periodt. 💅🫶