On 2022-04-05, at 13:28:50 +0200, Florian Westphal wrote: > Jeremy Sowden <jeremy@xxxxxxxxxx> wrote: > > Some bitwise operations are generated in user space when munging > > paylod expressions. During delinearization, user space attempts to > > eliminate these operations. However, it does this before deducing > > the byte-order or the correct length in bits of the operands, which > > means that it doesn't always handle multi-byte host-endian > > operations correctly. Therefore, add support for storing the > > bit-length of the expression, even though the kernel doesn't use it, > > in order to be able to pass it back to user space. > > Can rule udata be used for this, or is that too much work? > The udata infra is already used to store comments and it would not > need kernel changes. It wouldn't be straightforward. Expression udata might make more sense than adding a new bitwise attribute, but that doesn't currently exist. Would it be worth adding? I seem to recall considering something along those lines for passing type information with expressions as a way to implement casting. J.
Attachment:
signature.asc
Description: PGP signature
