Re: [PATCH 1/8] rust: firmware: add request_into_buf()

["Alexandre Courbot" <acourbot@nvidia.com>]

On Thu Jun 11, 2026 at 3:24 AM JST, Timur Tabi wrote:
> On Wed, 2026-06-10 at 19:03 +0100, Gary Guo wrote:
>> > +pub fn request_into_buf(name: &CStr, dev: &Device, buf: &mut [u8]) -> Result<usize> {
>> 
>> It looks like this API design is very prone to misuse. Patch 5 of this series
>> for example does not check for the returned size.
>
> I was debating changing it to use request_partial_firmware_into_buf, which would "encourage" users
> to verify the return value.
>
>> Also, I think people might just write `buf[..size]` and then it'll panic because
>> `size` can be larger.
>
> Why would it panic if the buffer is larger than the size of the file?  The kernel API requires you
> to know the size of the file before using request_firmware_into_buf(). If the file on disk is larger
> than the buffer, the kernel returns -EFBIG.  If the file on this is smaller, then it is loader
> completely and the returned Result<usize> is the number of bytes actually loaded from disk.
>
> This has always been a problem for Nouveau because the GSP binary is 90MB in size, and
> request_firmware() allocates a contiguous block of memory for it, only for us to memcpy it to a
> vmalloc'd buffer.

IIUC `request_firmware` allocates its memory using `vmalloc`. Unless the
firmware is built-in, in which case it returns a pointer to its
location in `.rodata`. That last option is probably the reason for the
unconditional copy in Nouveau.