Commits on Mar 29, 2022

1. btf: Implement support for bitfield relocations …

   This adds support for relocating bitfields, e.g. rewriting of
   the load offset and bit shifts for accessing the field.
   
   The implementation is mostly following libbpf, but does not
   include the loop for doubling the size of the load as that should
   not be necessary when one assumes that loads must be aligned and
   that a bitfield cannot be larger than the variable type.
   
   Signed-off-by: Jussi Maki <jussi@isovalent.com>
   Signed-off-by: Timo Beckers <timo@isovalent.com>

   

   joamaki authored and ti-mo committed Mar 29, 2022
   Copy the full SHA

   eb92491 View commit details

   Browse the repository at this point in the history

2. btf: check all of coreField in TestCOREFindField …

   The current test doesn't check all fields of coreField. Fix this by
   using go-cmp.

   

   lmb authored and ti-mo committed Mar 29, 2022
   Copy the full SHA

   6b3b15d View commit details

   Browse the repository at this point in the history

3. btf: ignore target when deciding whether to validate a fixup

   

   lmb authored and ti-mo committed Mar 29, 2022
   Copy the full SHA

   e893e37 View commit details

   Browse the repository at this point in the history

4. btf: store coreField.offset in bytes …

   The documentation on coreField.offset states that it is in bytes,
   but the code stores it in bits. Refactor the code to use bytes,
   since that is more natural for the rest of the code.
   
   Change coreFindField to re-initialize coreField variables instead
   of only creating coreFields when returning. This has the benefit
   of implicitly zeroing fields of the local and target variables.
   It also allows us to put helper methods on coreField to slim
   coreFindField down a little bit.

   

   lmb authored and ti-mo committed Mar 29, 2022
   Copy the full SHA

   db7ff9f View commit details

   Browse the repository at this point in the history

5. btf: enforce that bitfield is last accessor in coreFindField …

   Make sure that we're at the end of an accessor when dealing with a
   bitfield. This also allows us to fall through to after the switch
   instead of returning early.

   

   lmb authored and ti-mo committed Mar 29, 2022
   Copy the full SHA

   fd42c74 View commit details

   Browse the repository at this point in the history

6. btf: use alignof instead of sizeof for bitfield loads …

   We currently use the size of the target type to align the load of
   a bitfield. This is OK for integers and enums, since their size
   equals their alignment. It's not correct in general however.
   
   Add alignof to make this explicit.

   

   lmb authored and ti-mo committed Mar 29, 2022
   Copy the full SHA

   c7996e8 View commit details

   Browse the repository at this point in the history

7. btf: redefine coreField.bitfieldOffset …

   coreField.bitfieldOffset is currently defined to count from the start
   of the enclosing composite type. This is incovenient, since the shift
   relocations want to know the offset from the start of the load.
   Redefine coreField.bitfieldOffset to count from the start of the
   field and add coreField.adjustOffsetBits to handle adjusting
   offset and bitfieldOffset while respecting alignment.
   
   Also compute offsets for the local relocation. We don't use this
   information at the moment, but it's what we do for other types
   of relocations.
   
   Finally, move the compatibility behaviour for regular fields accessed
   as bitfield into coreField.sizeBits. Doing the fixup only when
   encountering a bitfield leads to inconsistent behaviour.

   

   lmb authored and ti-mo committed Mar 29, 2022
   Copy the full SHA

   85daa93 View commit details

   Browse the repository at this point in the history