This patch adds a generic "post_cap_deletion" step that is called by
finalise_slot. Previous to this, the only caps which had actions
required at this stage were IRQHandlerCaps -- it was required that the
IRQ bitmap be updated after the cap itself was removed (as the
invariants state that for any existing IRQHandlerCap, the corresponding
bit in the IRQ bitmap must be set).
By genericising this, we add the capacity for new, arch-specific post
cap deletion actions to occur in the future.
Colloquially known as "invert-fastpath".
Update verification efforts on ARM for the following seL4 changes:
- scheduling decisions done in possibleSwitchTo are moved to the
scheduler
- possibleSwitchTo only checks whether the candidate is valid for a
fast switch, not its priority, accepting possible candidates
immmediately as a switch-to scheduler action
- the scheduler checks the candidate against the current thread and
against the bitmaps before making a decision
- attemptSwitchTo and switchIfRequiredTo are gone
- scheduler is now more complicated, and numerous proofs related to it
are rewritten from scratch
- fast path now checks ready queues via the scheduler bitmaps
- L2 scheduler bitmap order reversed for better cache locality
Many iterations between the kernel and verification teams were needed
to get this right.
* This is trivial/irrelevant since getActiveIRQ ignores its argument
in ARM, but it makes a bit more sense to have it being this way,
and it is consistent with the equivalent function in InfoFlowC.
The things that usually go wrong:
- wp fall through: add +, e.g.
apply (wp select_wp) -> apply (wp select_wp)+
- precondition: you can remove most hoare_pre, but wpc still needs it, and
sometimes the wp instance relies on being able to fit a rule to the
current non-schematic precondition. In that case, use "including no_pre"
to switch off the automatic hoare_pre application.
- very rarely there is a schematic postcondition that interferes with the
new trivial cleanup rules, because the rest of the script assumes some
specific state afterwards (shouldn't happen in a reasonable proof, but
not all proofs are reasonable..). In that case, (wp_once ...)+ should
emulate the old behaviour precisely.
* tcb_context rephrasing to (tcb_context o tcb_arch) and respectively
for set operations
* unfolding of reserved_irq for trivially solving most lemmas
* Changes to the inductive definition of integrity_obj to account for
tcb_arch and tcb_context new location
* Changes to the tcb examples in ExampleSystem to include tcb_arch
* Rephrasing of domain_sep_inv to accommodate the ReservedIRQ case
* Mostly rephrasing of tcb_context to (some form of) (tcb_context o tcb_arch)
* Trivial unfolding of handle_reserved_irq for hoare rules
* Examples in Example_Valid_State.thy were updated
* Nothing remarkable, mostly rephrasing of tcb_context and ReservedIRQ
handling
* Fun fact, some proofs are now shorter
tags: [VER-623][SELFOUR-413]
These changes to the automatons are required by:
SELFOUR-242: invert bitfield scheduler and optimise fast path
Details:
When we enter the kernel, the domain time left (ksDomainTime) is never zero.
If we entered on a timer interrupt, we may decrement it to zero before the
scheduler runs. If we do so, we set the scheduler state to choose_new_thread.
When choosing a new thread, the scheduler switches to a new domain if the
present one is required, and sets the new domain time left from domain_list
(ksDomSchedule).
When entering the kernel on a non-interrupt event, we never touch the domain
time left, which trivially preserves the new constraints.
To prove these, we had to ban a transition from kernel entry to kernel being
preempted when handling an interrupt event in InfoFlow. This is fine, as by
design handling interrupts is not meant to be preempted by interrupts.
To finish the proof of refinement to C, the specification for checkPrio
needed strengthening: the checkPrio spec now takes a machine word
argument. In the spec, priorities are still stored as 8-bit quantities,
however. Once the spec was strenthened, it was possible to remove some
redundant checks and mask operations from the C code.
A thread's maximum controlled priority (MCP) determines the maximum
thread priority or MCP it can assign to another thread (or itself).
This commit should at least remove merge conflict markers, and the idea
is that at least refine, crefine, drefine, and infoflow (with sorrys)
build. Subsequent commits may be required to fix build issues that I
have not picked up.
UserOp_IF had its own way of extracting the XN bit from page tables.
This is now unified with the existing functions in ADT_AI, which also
means that the proof for XN bit equality is basically the same as for
pt_rights and pt_lift.
The fact that the C infoflow adt refines the abstract infoflow adt now only requires that given user operation is nonempty and not sane (nonempty and doesn't return an interrupt).
Also added some more general lemmas about fw_sim and refinement to lib/Simulation.thy.