Re: [PATCH 21/31] gpu: nova-core: Hopper/Blackwell: add FMC signature extraction

[John Hubbard <jhubbard@nvidia.com>]

On 12/3/25 7:45 AM, Joel Fernandes wrote:
> Hi John,
> 
> On 12/3/2025 12:59 AM, John Hubbard wrote:
>> Add extract_fmc_signatures_static() to parse cryptographic signatures
>> from FMC ELF firmware sections. This extracts the SHA-384 hash, RSA
>> public key, and signature needed for Chain of Trust verification.
>>
>> Also exposes the elf_section() helper from firmware.rs for use by FSP.
>>
>> Signed-off-by: John Hubbard <jhubbard@nvidia.com>
>> ---
>>   drivers/gpu/nova-core/firmware.rs |   4 +-
>>   drivers/gpu/nova-core/fsp.rs      | 104 ++++++++++++++++++++++++++++++
>>   2 files changed, 107 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/nova-core/firmware.rs b/drivers/gpu/nova-core/firmware.rs
>> index 5cbb8be7434f..7f8d62f9ceba 100644
>> --- a/drivers/gpu/nova-core/firmware.rs
>> +++ b/drivers/gpu/nova-core/firmware.rs
>> @@ -23,6 +23,8 @@
>>       },
>>   };
>>   
>> +pub(crate) use elf::elf_section;
>> +
>>   pub(crate) mod booter;
>>   pub(crate) mod fsp;
>>   pub(crate) mod fwsec;
>> @@ -419,7 +421,7 @@ fn elf32_section<'a>(elf: &'a [u8], name: &str) -> Option<&'a [u8]> {
>>       }
>>   
>>       /// Automatically detects ELF32 vs ELF64 based on the ELF header.
>> -    pub(super) fn elf_section<'a>(elf: &'a [u8], name: &str) -> Option<&'a [u8]> {
>> +    pub(crate) fn elf_section<'a>(elf: &'a [u8], name: &str) -> Option<&'a [u8]> {
>>           // Check ELF magic.
>>           if elf.len() < 5 || elf.get(0..4)? != b"\x7fELF" {
>>               return None;
>> diff --git a/drivers/gpu/nova-core/fsp.rs b/drivers/gpu/nova-core/fsp.rs
>> index 389c43bfd538..311b6d4c6011 100644
>> --- a/drivers/gpu/nova-core/fsp.rs
>> +++ b/drivers/gpu/nova-core/fsp.rs
>> @@ -256,4 +256,108 @@ pub(crate) fn wait_secure_boot(
>>           })
>>           .map(|_| ())
>>       }
>> +
>> +    /// Extract FMC firmware signatures for Chain of Trust verification.
>> +    ///
>> +    /// Extracts real cryptographic signatures from FMC ELF32 firmware sections.
>> +    /// Returns signatures in a heap-allocated structure to prevent stack overflow.
>> +    pub(crate) fn extract_fmc_signatures_static(
>> +        dev: &device::Device<device::Bound>,
>> +        fmc_fw_data: &[u8],
>> +    ) -> Result<KBox<FmcSignatures>> {
>> +        dev_dbg!(dev, "FMC firmware size: {} bytes\n", fmc_fw_data.len());
> 
> Let us remove these? I think we discussed [1] that once things are working, we'd
> not want these and can add it on-demand if needed.

Yes, absolutely. Thanks for checking on this, I was having trouble
drawing the line at the right amount of output--I'm sure there is
still too much, now that you point it out.

> 
> [1] https://lore.kernel.org/all/d6c9c7f2-098e-4b55-b754-4287b698fc1c@nvidia.com/
> 
>> +
>> +        // Extract hash section (SHA-384)
>> +        let hash_section = crate::firmware::elf_section(fmc_fw_data, "hash")
>> +            .ok_or(EINVAL)
>> +            .inspect_err(|_| dev_err!(dev, "FMC firmware missing 'hash' section\n"))?;
>> +
>> +        // Extract public key section (RSA public key)
>> +        let pkey_section = crate::firmware::elf_section(fmc_fw_data, "publickey")
>> +            .ok_or(EINVAL)
>> +            .inspect_err(|_| dev_err!(dev, "FMC firmware missing 'publickey' section\n"))?;
>> +
>> +        // Extract signature section (RSA signature)
>> +        let sig_section = crate::firmware::elf_section(fmc_fw_data, "signature")
>> +            .ok_or(EINVAL)
>> +            .inspect_err(|_| dev_err!(dev, "FMC firmware missing 'signature' section\n"))?;
>> +
>> +        dev_dbg!(
>> +            dev,
>> +            "FMC ELF sections: hash={} bytes, pkey={} bytes, sig={} bytes\n",
>> +            hash_section.len(),
>> +            pkey_section.len(),
>> +            sig_section.len()
>> +        );
>> +
> 
> Here as well.

Yes.

> 
>> +        // Validate section sizes - hash must be exactly 48 bytes
>> +        if hash_section.len() != FSP_HASH_SIZE {
>> +            dev_err!(
>> +                dev,
>> +                "FMC hash section size {} != expected {}\n",
>> +                hash_section.len(),
>> +                FSP_HASH_SIZE
>> +            );
>> +            return Err(EINVAL);
>> +        }
>> +
>> +        // Public key and signature can be smaller than the fixed array sizes
>> +        if pkey_section.len() > FSP_PKEY_SIZE * 4 {
>> +            dev_err!(
>> +                dev,
>> +                "FMC publickey section size {} > maximum {}\n",
>> +                pkey_section.len(),
>> +                FSP_PKEY_SIZE * 4
>> +            );
>> +            return Err(EINVAL);
>> +        }
>> +
>> +        if sig_section.len() > FSP_SIG_SIZE * 4 {
>> +            dev_err!(
>> +                dev,
>> +                "FMC signature section size {} > maximum {}\n",
>> +                sig_section.len(),
>> +                FSP_SIG_SIZE * 4
>> +            );
>> +            return Err(EINVAL);
>> +        }
>> +
>> +        // Allocate signature structure on heap to avoid stack overflow
>> +        let mut signatures = KBox::new(FmcSignatures::default(), GFP_KERNEL)?;
>> +
>> +        // Copy hash section directly as bytes (48 bytes exactly)
>> +        // SAFETY: hash384 is a [u32; 12] array (48 bytes), and we create a byte slice of
>> +        // exactly FSP_HASH_SIZE (48) bytes. The pointer is valid and properly aligned.
>> +        let hash_bytes = unsafe {
>> +            core::slice::from_raw_parts_mut(
>> +                signatures.hash384.as_mut_ptr().cast::<u8>(),
>> +                FSP_HASH_SIZE,
>> +            )
>> +        };
>> +        hash_bytes.copy_from_slice(hash_section);
>> +
>> +        // Copy public key section (up to 388 bytes, zero-padded)
>> +        // SAFETY: public_key is a [u32; 96] array (384 bytes), and we create a byte slice of
>> +        // FSP_PKEY_SIZE * 4 bytes. The pointer is valid and properly aligned.
>> +        let pkey_bytes = unsafe {
>> +            core::slice::from_raw_parts_mut(
>> +                signatures.public_key.as_mut_ptr().cast::<u8>(),
>> +                FSP_PKEY_SIZE * 4,
>> +            )
>> +        };
>> +        pkey_bytes[..pkey_section.len()].copy_from_slice(pkey_section);
> 
> Even if this works in practice, I believe it's UB as the `from_raw_parts_mut()`
> should have the entire slice range to be valid memory (see [2]), but
> FSP_PKEY_SIZE * 4 is 388 bytes while public_key is only 384 bytes ([u32; 96]).
> This is vulnerable as the KBox holding the signature may not have the extra
> space even if it does now.
> 
> Can we create a slice with exactly the bytes we need? something like:
> let pkey_bytes = unsafe {
>      core::slice::from_raw_parts_mut(
>          signatures.public_key.as_mut_ptr().cast::<u8>(),
>          pkey_section.len(),
>      )
> };
> pkey_bytes.copy_from_slice(pkey_section);
> 
> Another reason for doing this is, the code is more fragile left as is, as there
> is a risk of unrelated memory leaking into the slice and accessed by new/future
> code.

Sure, I'll go in this direction, thanks for spotting that.


thanks,
-- 
John Hubbard

> 
> [2] "Behavior is undefined if any of the following conditions are violated"
> https://doc.rust-lang.org/std/slice/fn.from_raw_parts_mut.html
> 
> thanks,
> 
>   - Joel
> 
>> +
>> +        // Copy signature section (up to 384 bytes, zero-padded)
>> +        // SAFETY: signature is a [u32; 96] array (384 bytes), and we create a byte slice of
>> +        // FSP_SIG_SIZE * 4 bytes. The pointer is valid and properly aligned.
>> +        let sig_bytes = unsafe {
>> +            core::slice::from_raw_parts_mut(
>> +                signatures.signature.as_mut_ptr().cast::<u8>(),
>> +                FSP_SIG_SIZE * 4,
>> +            )
>> +        };
>> +        sig_bytes[..sig_section.len()].copy_from_slice(sig_section);
>> +
>> +        Ok(signatures)
>> +    }
>>   }
>