xref: /linux/drivers/gpu/nova-core/firmware/gsp.rs (revision e10dcb9d654177270b48119fa79f3924aef9ff48)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 use kernel::{
4     device,
5     dma::{
6         Coherent,
7         CoherentBox,
8         DataDirection,
9         DmaAddress, //
10     },
11     prelude::*,
12     scatterlist::{
13         Owned,
14         SGTable, //
15     },
16 };
17 
18 use crate::{
19     firmware::riscv::RiscvFirmware,
20     gpu::{
21         Architecture,
22         Chipset, //
23     },
24     gsp::GSP_PAGE_SIZE,
25     num::FromSafeCast,
26 };
27 
28 /// Ad-hoc and temporary module to extract sections from ELF images.
29 ///
30 /// Some firmware images are currently packaged as ELF files, where sections names are used as keys
31 /// to specific and related bits of data. Future firmware versions are scheduled to move away from
32 /// that scheme before nova-core becomes stable, which means this module will eventually be
33 /// removed.
34 mod elf {
35     use kernel::{
36         bindings,
37         prelude::*,
38         transmute::FromBytes, //
39     };
40 
41     /// Newtype to provide a [`FromBytes`] implementation.
42     #[repr(transparent)]
43     struct Elf64Hdr(bindings::elf64_hdr);
44     // SAFETY: all bit patterns are valid for this type, and it doesn't use interior mutability.
45     unsafe impl FromBytes for Elf64Hdr {}
46 
47     #[repr(transparent)]
48     struct Elf64SHdr(bindings::elf64_shdr);
49     // SAFETY: all bit patterns are valid for this type, and it doesn't use interior mutability.
50     unsafe impl FromBytes for Elf64SHdr {}
51 
52     /// Tries to extract section with name `name` from the ELF64 image `elf`, and returns it.
53     pub(super) fn elf64_section<'a, 'b>(elf: &'a [u8], name: &'b str) -> Option<&'a [u8]> {
54         let hdr = &elf
55             .get(0..size_of::<bindings::elf64_hdr>())
56             .and_then(Elf64Hdr::from_bytes)?
57             .0;
58 
59         // Get all the section headers.
60         let mut shdr = {
61             let shdr_num = usize::from(hdr.e_shnum);
62             let shdr_start = usize::try_from(hdr.e_shoff).ok()?;
63             let shdr_end = shdr_num
64                 .checked_mul(size_of::<Elf64SHdr>())
65                 .and_then(|v| v.checked_add(shdr_start))?;
66 
67             elf.get(shdr_start..shdr_end)
68                 .map(|slice| slice.chunks_exact(size_of::<Elf64SHdr>()))?
69         };
70 
71         // Get the strings table.
72         let strhdr = shdr
73             .clone()
74             .nth(usize::from(hdr.e_shstrndx))
75             .and_then(Elf64SHdr::from_bytes)?;
76 
77         // Find the section which name matches `name` and return it.
78         shdr.find(|&sh| {
79             let Some(hdr) = Elf64SHdr::from_bytes(sh) else {
80                 return false;
81             };
82 
83             let Some(name_idx) = strhdr
84                 .0
85                 .sh_offset
86                 .checked_add(u64::from(hdr.0.sh_name))
87                 .and_then(|idx| usize::try_from(idx).ok())
88             else {
89                 return false;
90             };
91 
92             // Get the start of the name.
93             elf.get(name_idx..)
94                 .and_then(|nstr| CStr::from_bytes_until_nul(nstr).ok())
95                 // Convert into str.
96                 .and_then(|c_str| c_str.to_str().ok())
97                 // Check that the name matches.
98                 .map(|str| str == name)
99                 .unwrap_or(false)
100         })
101         // Return the slice containing the section.
102         .and_then(|sh| {
103             let hdr = Elf64SHdr::from_bytes(sh)?;
104             let start = usize::try_from(hdr.0.sh_offset).ok()?;
105             let end = usize::try_from(hdr.0.sh_size)
106                 .ok()
107                 .and_then(|sh_size| start.checked_add(sh_size))?;
108 
109             elf.get(start..end)
110         })
111     }
112 }
113 
114 /// GSP firmware with 3-level radix page tables for the GSP bootloader.
115 ///
116 /// The bootloader expects firmware to be mapped starting at address 0 in GSP's virtual address
117 /// space:
118 ///
119 /// ```text
120 /// Level 0:  1 page, 1 entry         -> points to first level 1 page
121 /// Level 1:  Multiple pages/entries  -> each entry points to a level 2 page
122 /// Level 2:  Multiple pages/entries  -> each entry points to a firmware page
123 /// ```
124 ///
125 /// Each page is 4KB, each entry is 8 bytes (64-bit DMA address).
126 /// Also known as "Radix3" firmware.
127 #[pin_data]
128 pub(crate) struct GspFirmware {
129     /// The GSP firmware inside a [`VVec`], device-mapped via a SG table.
130     #[pin]
131     fw: SGTable<Owned<VVec<u8>>>,
132     /// Level 2 page table whose entries contain DMA addresses of firmware pages.
133     #[pin]
134     level2: SGTable<Owned<VVec<u8>>>,
135     /// Level 1 page table whose entries contain DMA addresses of level 2 pages.
136     #[pin]
137     level1: SGTable<Owned<VVec<u8>>>,
138     /// Level 0 page table (single 4KB page) with one entry: DMA address of first level 1 page.
139     level0: Coherent<[u64]>,
140     /// Size in bytes of the firmware contained in [`Self::fw`].
141     pub(crate) size: usize,
142     /// Device-mapped GSP signatures matching the GPU's [`Chipset`].
143     pub(crate) signatures: Coherent<[u8]>,
144     /// GSP bootloader, verifies the GSP firmware before loading and running it.
145     pub(crate) bootloader: RiscvFirmware,
146 }
147 
148 impl GspFirmware {
149     /// Loads the GSP firmware binaries, map them into `dev`'s address-space, and creates the page
150     /// tables expected by the GSP bootloader to load it.
151     pub(crate) fn new<'a>(
152         dev: &'a device::Device<device::Bound>,
153         chipset: Chipset,
154         ver: &'a str,
155     ) -> impl PinInit<Self, Error> + 'a {
156         pin_init::pin_init_scope(move || {
157             let firmware = super::request_firmware(dev, chipset, "gsp", ver)?;
158 
159             let fw_section = elf::elf64_section(firmware.data(), ".fwimage").ok_or(EINVAL)?;
160 
161             let size = fw_section.len();
162 
163             // Move the firmware into a vmalloc'd vector and map it into the device address
164             // space.
165             let fw_vvec = VVec::with_capacity(fw_section.len(), GFP_KERNEL)
166                 .and_then(|mut v| {
167                     v.extend_from_slice(fw_section, GFP_KERNEL)?;
168                     Ok(v)
169                 })
170                 .map_err(|_| ENOMEM)?;
171 
172             Ok(try_pin_init!(Self {
173                 fw <- SGTable::new(dev, fw_vvec, DataDirection::ToDevice, GFP_KERNEL),
174                 level2 <- {
175                     // Allocate the level 2 page table, map the firmware onto it, and map it into
176                     // the device address space.
177                     VVec::<u8>::with_capacity(
178                         fw.iter().count() * core::mem::size_of::<u64>(),
179                         GFP_KERNEL,
180                     )
181                     .map_err(|_| ENOMEM)
182                     .and_then(|level2| map_into_lvl(&fw, level2))
183                     .map(|level2| SGTable::new(dev, level2, DataDirection::ToDevice, GFP_KERNEL))?
184                 },
185                 level1 <- {
186                     // Allocate the level 1 page table, map the level 2 page table onto it, and map
187                     // it into the device address space.
188                     VVec::<u8>::with_capacity(
189                         level2.iter().count() * core::mem::size_of::<u64>(),
190                         GFP_KERNEL,
191                     )
192                     .map_err(|_| ENOMEM)
193                     .and_then(|level1| map_into_lvl(&level2, level1))
194                     .map(|level1| SGTable::new(dev, level1, DataDirection::ToDevice, GFP_KERNEL))?
195                 },
196                 level0: {
197                     // Allocate the level 0 page table as a device-visible DMA object, and map the
198                     // level 1 page table onto it.
199 
200                     // Fill level 1 page entry.
201                     let level1_entry = level1.iter().next().ok_or(EINVAL)?;
202                     let level1_entry_addr = level1_entry.dma_address();
203 
204                     // Create level 0 page table data and fill its first entry with the level 1
205                     // table.
206                     let mut level0 = CoherentBox::<[u64]>::zeroed_slice(
207                         dev,
208                         GSP_PAGE_SIZE / size_of::<u64>(),
209                         GFP_KERNEL
210                     )?;
211                     level0[0] = level1_entry_addr.to_le();
212 
213                     level0.into()
214                 },
215                 size,
216                 signatures: {
217                     let sigs_section = match chipset.arch() {
218                         Architecture::Turing
219                             if matches!(chipset, Chipset::TU116 | Chipset::TU117) =>
220                         {
221                             ".fwsignature_tu11x"
222                         }
223                         Architecture::Turing => ".fwsignature_tu10x",
224                         // GA100 uses the same firmware as Turing
225                         Architecture::Ampere if chipset == Chipset::GA100 => ".fwsignature_tu10x",
226                         Architecture::Ampere => ".fwsignature_ga10x",
227                         Architecture::Ada => ".fwsignature_ad10x",
228                     };
229 
230                     elf::elf64_section(firmware.data(), sigs_section)
231                         .ok_or(EINVAL)
232                         .and_then(|data| Coherent::from_slice(dev, data, GFP_KERNEL))?
233                 },
234                 bootloader: {
235                     let bl = super::request_firmware(dev, chipset, "bootloader", ver)?;
236 
237                     RiscvFirmware::new(dev, &bl)?
238                 },
239             }))
240         })
241     }
242 
243     /// Returns the DMA handle of the radix3 level 0 page table.
244     pub(crate) fn radix3_dma_handle(&self) -> DmaAddress {
245         self.level0.dma_handle()
246     }
247 }
248 
249 /// Build a page table from a scatter-gather list.
250 ///
251 /// Takes each DMA-mapped region from `sg_table` and writes page table entries
252 /// for all 4KB pages within that region. For example, a 16KB SG entry becomes
253 /// 4 consecutive page table entries.
254 fn map_into_lvl(sg_table: &SGTable<Owned<VVec<u8>>>, mut dst: VVec<u8>) -> Result<VVec<u8>> {
255     for sg_entry in sg_table.iter() {
256         // Number of pages we need to map.
257         let num_pages = usize::from_safe_cast(sg_entry.dma_len()).div_ceil(GSP_PAGE_SIZE);
258 
259         for i in 0..num_pages {
260             let entry = sg_entry.dma_address()
261                 + (u64::from_safe_cast(i) * u64::from_safe_cast(GSP_PAGE_SIZE));
262             dst.extend_from_slice(&entry.to_le_bytes(), GFP_KERNEL)?;
263         }
264     }
265 
266     Ok(dst)
267 }
268