deriveCap_ret_t
deriveCap(cte_t *slot, cap_t cap)
{
deriveCap_ret_t ret;
if (isArchCap(cap)) {
return Arch_deriveCap(slot, cap);
}
switch (cap_get_capType(cap)) {
case cap_zombie_cap:
ret.status = EXCEPTION_NONE;
ret.cap = cap_null_cap_new();
break;
case cap_irq_control_cap:
ret.status = EXCEPTION_NONE;
ret.cap = cap_null_cap_new();
break;
case cap_reply_cap:
ret.status = EXCEPTION_NONE;
ret.cap = cap_null_cap_new();
break;
default:
ret.status = EXCEPTION_NONE;
ret.cap = cap;
}
return ret;
}
exception_t
decodeSetPriority(cap_t cap, unsigned int length, word_t *buffer)
{
prio_t newPrio;
if (length < 1) {
userError("TCB SetPriority: Truncated message.");
current_syscall_error.type = seL4_TruncatedMessage;
return EXCEPTION_SYSCALL_ERROR;
}
newPrio = getSyscallArg(0, buffer);
/* assuming here seL4_MaxPrio is of form 2^n - 1 */
newPrio = newPrio & MASK(8);
if (newPrio > ksCurThread->tcbPriority) {
userError("TCB SetPriority: Requested priority %d too high (max %d).",
(int)newPrio, (int)ksCurThread->tcbPriority);
current_syscall_error.type = seL4_IllegalOperation;
return EXCEPTION_SYSCALL_ERROR;
}
setThreadState(ksCurThread, ThreadState_Restart);
return invokeTCB_ThreadControl(
TCB_PTR(cap_thread_cap_get_capTCBPtr(cap)), NULL,
0, newPrio,
cap_null_cap_new(), NULL,
cap_null_cap_new(), NULL,
0, cap_null_cap_new(),
NULL, thread_control_update_priority);
}
deriveCap_ret_t
Arch_deriveCap(cte_t *slot, cap_t cap)
{
deriveCap_ret_t ret;
switch (cap_get_capType(cap)) {
case cap_page_table_cap:
if (cap_page_table_cap_get_capPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving an unmapped PT cap");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_page_directory_cap:
if (cap_page_directory_cap_get_capPDIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a PD cap without an assigned ASID");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
/* This is a deviation from haskell, which has only
* one frame cap type on ARM */
case cap_small_frame_cap:
ret.cap = cap_small_frame_cap_set_capFMappedASID(cap, asidInvalid);
ret.status = EXCEPTION_NONE;
return ret;
case cap_frame_cap:
ret.cap = cap_frame_cap_set_capFMappedASID(cap, asidInvalid);
ret.status = EXCEPTION_NONE;
return ret;
case cap_asid_control_cap:
case cap_asid_pool_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
default:
/* This assert has no equivalent in haskell,
* as the options are restricted by type */
fail("Invalid arch cap");
}
}
void
doReplyTransfer(tcb_t *sender, tcb_t *receiver, cte_t *slot)
{
assert(thread_state_get_tsType(receiver->tcbState) ==
ThreadState_BlockedOnReply);
if (likely(fault_get_faultType(receiver->tcbFault) == fault_null_fault)) {
doIPCTransfer(sender, NULL, 0, true, receiver, false);
/** GHOSTUPD: "(True, gs_set_assn cteDeleteOne_'proc (ucast cap_reply_cap))" */
setThreadState(receiver, ThreadState_Running);
attemptSwitchTo(receiver);
} else {
bool_t restart;
/** GHOSTUPD: "(True, gs_set_assn cteDeleteOne_'proc (ucast cap_reply_cap))" */
restart = handleFaultReply(receiver, sender);
fault_null_fault_ptr_new(&receiver->tcbFault);
if (restart) {
setThreadState(receiver, ThreadState_Restart);
attemptSwitchTo(receiver);
} else {
setThreadState(receiver, ThreadState_Inactive);
}
}
finaliseCap(slot->cap, true, true);
slot->cap = cap_null_cap_new();
}
cap_t CONST Arch_updateCapData(bool_t preserve, word_t data, cap_t cap)
{
/* Avoid a switch statement with just a 'default' case as the C parser does not like this */
#ifdef CONFIG_IOMMU
switch (cap_get_capType(cap)) {
case cap_io_space_cap: {
io_space_capdata_t w = { { data } };
uint16_t PCIDevice = io_space_capdata_get_PCIDevice(w);
uint16_t domainID = io_space_capdata_get_domainID(w);
if (!preserve && cap_io_space_cap_get_capPCIDevice(cap) == 0 &&
domainID >= x86KSFirstValidIODomain &&
domainID != 0 &&
domainID <= MASK(x86KSnumIODomainIDBits)) {
return cap_io_space_cap_new(domainID, PCIDevice);
} else {
return cap_null_cap_new();
}
}
default:
return cap;
}
#endif
return cap;
}
void
doReplyTransfer(tcb_t *sender, tcb_t *receiver, cte_t *slot)
{
assert(thread_state_get_tsType(receiver->tcbState) ==
ThreadState_BlockedOnReply);
if (likely(fault_get_faultType(receiver->tcbFault) == fault_null_fault)) {
doIPCTransfer(sender, NULL, 0, true, receiver, false);
setThreadState(receiver, ThreadState_Running);
attemptSwitchTo(receiver);
} else {
bool_t restart;
restart = handleFaultReply(receiver, sender);
fault_null_fault_ptr_new(&receiver->tcbFault);
if (restart) {
setThreadState(receiver, ThreadState_Restart);
attemptSwitchTo(receiver);
} else {
setThreadState(receiver, ThreadState_Inactive);
}
}
if (cap_reply_cap_get_capInCDT(slot->cap)) {
cte_t *replySlot = TCB_PTR_CTE_PTR(receiver, tcbReply);
assert(cap_get_capType(replySlot->cap) == cap_reply_cap);
assert(cap_reply_cap_get_capInCDT(replySlot->cap));
cdtRemove(replySlot);
cdtRemove(slot);
slot->cap = cap_null_cap_new();
replySlot->cap = cap_reply_cap_new(false, true, TCB_REF(NULL));
} else {
deleteCallerCap(sender);
}
}
BOOT_CODE cap_t
create_root_cnode(void)
{
pptr_t pptr;
cap_t cap;
/* write the number of root CNode slots to global state */
ndks_boot.slot_pos_max = BIT(CONFIG_ROOT_CNODE_SIZE_BITS);
/* create an empty root CNode */
pptr = alloc_region(CONFIG_ROOT_CNODE_SIZE_BITS + CTE_SIZE_BITS);
if (!pptr) {
printf("Kernel init failing: could not create root cnode\n");
return cap_null_cap_new();
}
memzero(CTE_PTR(pptr), 1U << (CONFIG_ROOT_CNODE_SIZE_BITS + CTE_SIZE_BITS));
cap =
cap_cnode_cap_new(
CONFIG_ROOT_CNODE_SIZE_BITS, /* radix */
wordBits - CONFIG_ROOT_CNODE_SIZE_BITS, /* guard size */
0, /* guard */
pptr /* pptr */
);
/* write the root CNode cap into the root CNode */
write_slot(SLOT_PTR(pptr, BI_CAP_IT_CNODE), cap);
return cap;
}
exception_t
decodeSetIPCBuffer(cap_t cap, unsigned int length, cte_t* slot,
extra_caps_t extraCaps, word_t *buffer)
{
cptr_t cptr_bufferPtr;
cap_t bufferCap;
cte_t *bufferSlot;
if (length < 1 || extraCaps.excaprefs[0] == NULL) {
userError("TCB SetIPCBuffer: Truncated message.");
current_syscall_error.type = seL4_TruncatedMessage;
return EXCEPTION_SYSCALL_ERROR;
}
cptr_bufferPtr = getSyscallArg(0, buffer);
bufferSlot = extraCaps.excaprefs[0];
bufferCap = extraCaps.excaprefs[0]->cap;
if (cptr_bufferPtr == 0) {
bufferSlot = NULL;
} else {
exception_t e;
deriveCap_ret_t dc_ret;
dc_ret = deriveCap(bufferSlot, bufferCap);
if (dc_ret.status != EXCEPTION_NONE) {
return dc_ret.status;
}
bufferCap = dc_ret.cap;
e = checkValidIPCBuffer(cptr_bufferPtr, bufferCap);
if (e != EXCEPTION_NONE) {
return e;
}
}
setThreadState(ksCurThread, ThreadState_Restart);
return invokeTCB_ThreadControl(
TCB_PTR(cap_thread_cap_get_capTCBPtr(cap)), slot,
0,
0, /* used to be prioInvalid, but it doesn't matter */
cap_null_cap_new(), NULL,
cap_null_cap_new(), NULL,
cptr_bufferPtr, bufferCap,
bufferSlot, thread_control_update_ipc_buffer);
}
cap_t CONST Arch_updateCapData(bool_t preserve, word_t data, cap_t cap)
{
switch (cap_get_capType(cap)) {
#ifdef CONFIG_IOMMU
case cap_io_space_cap: {
io_space_capdata_t w = { { data } };
uint16_t PCIDevice = io_space_capdata_get_PCIDevice(w);
uint16_t domainID = io_space_capdata_get_domainID(w);
vtd_cte_t *vtd_context_table = (vtd_cte_t*)vtd_rte_get_ctp(ia32KSvtdRootTable[get_pci_bus(data)]);
if (!preserve && cap_io_space_cap_get_capPCIDevice(cap) == 0 &&
vtd_rte_get_present(ia32KSvtdRootTable[get_pci_bus(data)]) &&
(!vtd_cte_get_present(vtd_context_table[PCIDevice & 0xff]) || vtd_cte_get_translation_type(vtd_context_table[PCIDevice & 0xff]) != 2) &&
domainID <= MASK(ia32KSnumIODomainIDBits)) {
return cap_io_space_cap_new(domainID, PCIDevice);
} else {
return cap_null_cap_new();
}
}
#endif
case cap_io_port_cap: {
io_port_capdata_t w = { .words = { data } };
uint16_t firstPort = io_port_capdata_get_firstPort(w);
uint16_t lastPort = io_port_capdata_get_lastPort(w);
uint16_t capFirstPort = cap_io_port_cap_get_capIOPortFirstPort(cap);
uint16_t capLastPort = cap_io_port_cap_get_capIOPortLastPort(cap);
assert(capFirstPort <= capLastPort);
/* Ensure input data is ordered correctly. */
if (firstPort > lastPort) {
return cap_null_cap_new();
}
/* Allow the update if the new cap has range no larger than the old
* cap. */
if ((firstPort >= capFirstPort) && (lastPort <= capLastPort)) {
return cap_io_port_cap_new(firstPort, lastPort);
} else {
return cap_null_cap_new();
}
}
default:
return cap;
}
}
void
ipcCancel(tcb_t *tptr)
{
thread_state_t *state = &tptr->tcbState;
switch (thread_state_ptr_get_tsType(state)) {
case ThreadState_BlockedOnSend:
case ThreadState_BlockedOnReceive: {
/* blockedIPCCancel state */
endpoint_t *epptr;
tcb_queue_t queue;
epptr = EP_PTR(thread_state_ptr_get_blockingIPCEndpoint(state));
/* Haskell error "blockedIPCCancel: endpoint must not be idle" */
assert(endpoint_ptr_get_state(epptr) != EPState_Idle);
/* Dequeue TCB */
queue = ep_ptr_get_queue(epptr);
queue = tcbEPDequeue(tptr, queue);
ep_ptr_set_queue(epptr, queue);
if (!queue.head) {
endpoint_ptr_set_state(epptr, EPState_Idle);
}
setThreadState(tptr, ThreadState_Inactive);
break;
}
case ThreadState_BlockedOnAsyncEvent:
asyncIPCCancel(tptr,
AEP_PTR(thread_state_ptr_get_blockingIPCEndpoint(state)));
break;
case ThreadState_BlockedOnReply: {
cte_t *slot, *callerCap;
fault_null_fault_ptr_new(&tptr->tcbFault);
/* Get the reply cap slot */
slot = TCB_PTR_CTE_PTR(tptr, tcbReply);
callerCap = CTE_PTR(cap_reply_cap_get_capCallerSlot(slot->cap));
if (callerCap) {
finaliseCap(callerCap->cap, true, true);
callerCap->cap = cap_null_cap_new();
}
cap_reply_cap_ptr_set_capCallerSlot(&slot->cap, CTE_REF(NULL));
break;
}
}
}
void
deleteCallerCap(tcb_t *receiver)
{
cte_t *callerSlot;
callerSlot = TCB_PTR_CTE_PTR(receiver, tcbCaller);
if (cap_get_capType(callerSlot->cap) == cap_reply_cap) {
finaliseCap(callerSlot->cap, true, true);
callerSlot->cap = cap_null_cap_new();
}
}
BOOT_CODE cap_t
create_ipcbuf_frame(cap_t root_cnode_cap, cap_t pd_cap, vptr_t vptr)
{
cap_t cap;
pptr_t pptr;
/* allocate the IPC buffer frame */
pptr = alloc_region(PAGE_BITS);
if (!pptr) {
printf("Kernel init failing: could not create ipc buffer frame\n");
return cap_null_cap_new();
}
clearMemory((void*)pptr, PAGE_BITS);
/* create a cap of it and write it into the root CNode */
cap = create_mapped_it_frame_cap(pd_cap, pptr, vptr, false, false);
write_slot(SLOT_PTR(pptr_of_cap(root_cnode_cap), BI_CAP_IT_IPCBUF), cap);
return cap;
}
BOOT_CODE cap_t
create_it_asid_pool(cap_t root_cnode_cap)
{
pptr_t ap_pptr;
cap_t ap_cap;
/* create ASID pool */
ap_pptr = alloc_region(ASID_POOL_SIZE_BITS);
if (!ap_pptr) {
printf("Kernel init failed: failed to create initial thread asid pool\n");
return cap_null_cap_new();
}
memzero(ASID_POOL_PTR(ap_pptr), 1 << ASID_POOL_SIZE_BITS);
ap_cap = cap_asid_pool_cap_new(IT_ASID >> asidLowBits, ap_pptr);
write_slot(SLOT_PTR(pptr_of_cap(root_cnode_cap), BI_CAP_IT_ASID_POOL), ap_cap);
/* create ASID control cap */
write_slot(
SLOT_PTR(pptr_of_cap(root_cnode_cap), BI_CAP_ASID_CTRL),
cap_asid_control_cap_new()
);
return ap_cap;
}
/* Create an address space for the initial thread.
* This includes page directory and page tables */
BOOT_CODE static cap_t
create_it_address_space(cap_t root_cnode_cap, v_region_t it_v_reg)
{
cap_t vspace_cap;
vptr_t vptr;
pptr_t pptr;
slot_pos_t slot_pos_before;
slot_pos_t slot_pos_after;
slot_pos_before = ndks_boot.slot_pos_cur;
if (PDPT_BITS == 0) {
cap_t pd_cap;
pptr_t pd_pptr;
/* just create single PD obj and cap */
pd_pptr = alloc_region(PD_SIZE_BITS);
if (!pd_pptr) {
return cap_null_cap_new();
}
memzero(PDE_PTR(pd_pptr), 1 << PD_SIZE_BITS);
copyGlobalMappings(PDE_PTR(pd_pptr));
pd_cap = create_it_page_directory_cap(cap_null_cap_new(), pd_pptr, 0, IT_ASID);
if (!provide_cap(root_cnode_cap, pd_cap)) {
return cap_null_cap_new();
}
write_slot(SLOT_PTR(pptr_of_cap(root_cnode_cap), BI_CAP_IT_VSPACE), pd_cap);
vspace_cap = pd_cap;
} else {
cap_t pdpt_cap;
pptr_t pdpt_pptr;
unsigned int i;
/* create a PDPT obj and cap */
pdpt_pptr = alloc_region(PDPT_SIZE_BITS);
if (!pdpt_pptr) {
return cap_null_cap_new();
}
memzero(PDPTE_PTR(pdpt_pptr), 1 << PDPT_SIZE_BITS);
pdpt_cap = cap_pdpt_cap_new(
true, /* capPDPTISMapped */
IT_ASID, /* capPDPTMappedASID */
pdpt_pptr /* capPDPTBasePtr */
);
/* create all PD objs and caps necessary to cover userland image. For simplicity
* to ensure we also cover the kernel window we create all PDs */
for (i = 0; i < BIT(PDPT_BITS); i++) {
/* The compiler is under the mistaken belief here that this shift could be
* undefined. However, in the case that it would be undefined this code path
* is not reachable because PDPT_BITS == 0 (see if statement at the top of
* this function), so to work around it we must both put in a redundant
* if statement AND place the shift in a variable. While the variable
* will get compiled away it prevents the compiler from evaluating
* the 1 << 32 as a constant when it shouldn't
* tl;dr gcc evaluates constants even if code is unreachable */
int shift = (PD_BITS + PT_BITS + PAGE_BITS);
if (shift != 32) {
vptr = i << shift;
} else {
return cap_null_cap_new();
}
pptr = alloc_region(PD_SIZE_BITS);
if (!pptr) {
return cap_null_cap_new();
}
memzero(PDE_PTR(pptr), 1 << PD_SIZE_BITS);
if (!provide_cap(root_cnode_cap,
create_it_page_directory_cap(pdpt_cap, pptr, vptr, IT_ASID))
) {
return cap_null_cap_new();
}
}
/* now that PDs exist we can copy the global mappings */
copyGlobalMappings(PDPTE_PTR(pdpt_pptr));
write_slot(SLOT_PTR(pptr_of_cap(root_cnode_cap), BI_CAP_IT_VSPACE), pdpt_cap);
vspace_cap = pdpt_cap;
}
slot_pos_after = ndks_boot.slot_pos_cur;
ndks_boot.bi_frame->ui_pd_caps = (slot_region_t) {
slot_pos_before, slot_pos_after
};
/* create all PT objs and caps necessary to cover userland image */
slot_pos_before = ndks_boot.slot_pos_cur;
for (vptr = ROUND_DOWN(it_v_reg.start, PT_BITS + PAGE_BITS);
vptr < it_v_reg.end;
vptr += BIT(PT_BITS + PAGE_BITS)) {
pptr = alloc_region(PT_SIZE_BITS);
if (!pptr) {
return cap_null_cap_new();
}
memzero(PTE_PTR(pptr), 1 << PT_SIZE_BITS);
if (!provide_cap(root_cnode_cap,
create_it_page_table_cap(vspace_cap, pptr, vptr, IT_ASID))
) {
return cap_null_cap_new();
}
}
slot_pos_after = ndks_boot.slot_pos_cur;
ndks_boot.bi_frame->ui_pt_caps = (slot_region_t) {
slot_pos_before, slot_pos_after
};
return vspace_cap;
}
deriveCap_ret_t Arch_deriveCap(cte_t* slot, cap_t cap)
{
deriveCap_ret_t ret;
switch (cap_get_capType(cap)) {
case cap_page_table_cap:
if (cap_page_table_cap_get_capPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving an unmapped PT cap");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_page_directory_cap:
if (cap_page_directory_cap_get_capPDIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a PD cap without an assigned ASID");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_frame_cap:
#ifdef CONFIG_IOMMU
cap = cap_frame_cap_set_capFIsIOSpace(cap, 0);
#endif
ret.cap = cap_frame_cap_set_capFMappedASID(cap, asidInvalid);
ret.status = EXCEPTION_NONE;
return ret;
case cap_asid_control_cap:
case cap_asid_pool_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
case cap_io_port_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
#ifdef CONFIG_IOMMU
case cap_io_space_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
case cap_io_page_table_cap:
if (cap_io_page_table_cap_get_capIOPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
#endif
default:
/* This assert has no equivalent in haskell,
* as the options are restricted by type */
fail("Invalid arch cap type");
}
}
void fastpath_reply_recv(word_t cptr, word_t msgInfo)
{
seL4_MessageInfo_t info;
cap_t ep_cap;
endpoint_t *ep_ptr;
word_t length;
cte_t *callerSlot;
cap_t callerCap;
tcb_t *caller;
word_t badge;
tcb_t *endpointTail;
word_t fault_type;
cap_t newVTable;
vspace_root_t *cap_pd;
pde_t stored_hw_asid;
dom_t dom;
/* Get message info and length */
info = messageInfoFromWord_raw(msgInfo);
length = seL4_MessageInfo_get_length(info);
fault_type = seL4_Fault_get_seL4_FaultType(NODE_STATE(ksCurThread)->tcbFault);
/* Check there's no extra caps, the length is ok and there's no
* saved fault. */
if (unlikely(fastpath_mi_check(msgInfo) ||
fault_type != seL4_Fault_NullFault)) {
slowpath(SysReplyRecv);
}
/* Lookup the cap */
ep_cap = lookup_fp(TCB_PTR_CTE_PTR(NODE_STATE(ksCurThread), tcbCTable)->cap,
cptr);
/* Check it's an endpoint */
if (unlikely(!cap_capType_equals(ep_cap, cap_endpoint_cap) ||
!cap_endpoint_cap_get_capCanReceive(ep_cap))) {
slowpath(SysReplyRecv);
}
/* Check there is nothing waiting on the notification */
if (NODE_STATE(ksCurThread)->tcbBoundNotification &&
notification_ptr_get_state(NODE_STATE(ksCurThread)->tcbBoundNotification) == NtfnState_Active) {
slowpath(SysReplyRecv);
}
/* Get the endpoint address */
ep_ptr = EP_PTR(cap_endpoint_cap_get_capEPPtr(ep_cap));
/* Check that there's not a thread waiting to send */
if (unlikely(endpoint_ptr_get_state(ep_ptr) == EPState_Send)) {
slowpath(SysReplyRecv);
}
/* Only reply if the reply cap is valid. */
callerSlot = TCB_PTR_CTE_PTR(NODE_STATE(ksCurThread), tcbCaller);
callerCap = callerSlot->cap;
if (unlikely(!fastpath_reply_cap_check(callerCap))) {
slowpath(SysReplyRecv);
}
/* Determine who the caller is. */
caller = TCB_PTR(cap_reply_cap_get_capTCBPtr(callerCap));
/* ensure we are not single stepping the caller in ia32 */
#if defined(CONFIG_HARDWARE_DEBUG_API) && defined(CONFIG_ARCH_IA32)
if (caller->tcbArch.tcbContext.breakpointState.single_step_enabled) {
slowpath(SysReplyRecv);
}
#endif
/* Check that the caller has not faulted, in which case a fault
reply is generated instead. */
fault_type = seL4_Fault_get_seL4_FaultType(caller->tcbFault);
if (unlikely(fault_type != seL4_Fault_NullFault)) {
slowpath(SysReplyRecv);
}
/* Get destination thread.*/
newVTable = TCB_PTR_CTE_PTR(caller, tcbVTable)->cap;
/* Get vspace root. */
cap_pd = cap_vtable_cap_get_vspace_root_fp(newVTable);
/* Ensure that the destination has a valid MMU. */
if (unlikely(! isValidVTableRoot_fp(newVTable))) {
slowpath(SysReplyRecv);
}
#ifdef CONFIG_ARCH_AARCH32
/* Get HWASID. */
stored_hw_asid = cap_pd[PD_ASID_SLOT];
#endif
#ifdef CONFIG_ARCH_X86_64
stored_hw_asid.words[0] = cap_pml4_cap_get_capPML4MappedASID(newVTable);
#endif
#ifdef CONFIG_ARCH_AARCH64
stored_hw_asid.words[0] = cap_page_global_directory_cap_get_capPGDMappedASID(newVTable);
#endif
#ifdef CONFIG_ARCH_RISCV
stored_hw_asid.words[0] = cap_page_table_cap_get_capPTMappedASID(newVTable);
#endif
/* Ensure the original caller can be scheduled directly. */
dom = maxDom ? ksCurDomain : 0;
if (unlikely(!isHighestPrio(dom, caller->tcbPriority))) {
slowpath(SysReplyRecv);
}
#ifdef CONFIG_ARCH_AARCH32
/* Ensure the HWASID is valid. */
if (unlikely(!pde_pde_invalid_get_stored_asid_valid(stored_hw_asid))) {
slowpath(SysReplyRecv);
}
#endif
/* Ensure the original caller is in the current domain and can be scheduled directly. */
if (unlikely(caller->tcbDomain != ksCurDomain && maxDom)) {
slowpath(SysReplyRecv);
}
#ifdef ENABLE_SMP_SUPPORT
/* Ensure both threads have the same affinity */
if (unlikely(NODE_STATE(ksCurThread)->tcbAffinity != caller->tcbAffinity)) {
slowpath(SysReplyRecv);
}
#endif /* ENABLE_SMP_SUPPORT */
/*
* --- POINT OF NO RETURN ---
*
* At this stage, we have committed to performing the IPC.
*/
#ifdef CONFIG_BENCHMARK_TRACK_KERNEL_ENTRIES
ksKernelEntry.is_fastpath = true;
#endif
/* Set thread state to BlockedOnReceive */
thread_state_ptr_mset_blockingObject_tsType(
&NODE_STATE(ksCurThread)->tcbState, (word_t)ep_ptr, ThreadState_BlockedOnReceive);
thread_state_ptr_set_blockingIPCCanGrant(&NODE_STATE(ksCurThread)->tcbState,
cap_endpoint_cap_get_capCanGrant(ep_cap));;
/* Place the thread in the endpoint queue */
endpointTail = endpoint_ptr_get_epQueue_tail_fp(ep_ptr);
if (likely(!endpointTail)) {
NODE_STATE(ksCurThread)->tcbEPPrev = NULL;
NODE_STATE(ksCurThread)->tcbEPNext = NULL;
/* Set head/tail of queue and endpoint state. */
endpoint_ptr_set_epQueue_head_np(ep_ptr, TCB_REF(NODE_STATE(ksCurThread)));
endpoint_ptr_mset_epQueue_tail_state(ep_ptr, TCB_REF(NODE_STATE(ksCurThread)),
EPState_Recv);
} else {
/* Append current thread onto the queue. */
endpointTail->tcbEPNext = NODE_STATE(ksCurThread);
NODE_STATE(ksCurThread)->tcbEPPrev = endpointTail;
NODE_STATE(ksCurThread)->tcbEPNext = NULL;
/* Update tail of queue. */
endpoint_ptr_mset_epQueue_tail_state(ep_ptr, TCB_REF(NODE_STATE(ksCurThread)),
EPState_Recv);
}
/* Delete the reply cap. */
mdb_node_ptr_mset_mdbNext_mdbRevocable_mdbFirstBadged(
&CTE_PTR(mdb_node_get_mdbPrev(callerSlot->cteMDBNode))->cteMDBNode,
0, 1, 1);
callerSlot->cap = cap_null_cap_new();
callerSlot->cteMDBNode = nullMDBNode;
/* I know there's no fault, so straight to the transfer. */
/* Replies don't have a badge. */
badge = 0;
fastpath_copy_mrs(length, NODE_STATE(ksCurThread), caller);
/* Dest thread is set Running, but not queued. */
thread_state_ptr_set_tsType_np(&caller->tcbState,
ThreadState_Running);
switchToThread_fp(caller, cap_pd, stored_hw_asid);
msgInfo = wordFromMessageInfo(seL4_MessageInfo_set_capsUnwrapped(info, 0));
fastpath_restore(badge, msgInfo, NODE_STATE(ksCurThread));
}
exception_t
decodeSetSpace(cap_t cap, unsigned int length, cte_t* slot,
extra_caps_t extraCaps, word_t *buffer)
{
cptr_t faultEP;
word_t cRootData, vRootData;
cte_t *cRootSlot, *vRootSlot;
cap_t cRootCap, vRootCap;
deriveCap_ret_t dc_ret;
if (length < 3 || extraCaps.excaprefs[0] == NULL
|| extraCaps.excaprefs[1] == NULL) {
userError("TCB SetSpace: Truncated message.");
current_syscall_error.type = seL4_TruncatedMessage;
return EXCEPTION_SYSCALL_ERROR;
}
faultEP = getSyscallArg(0, buffer);
cRootData = getSyscallArg(1, buffer);
vRootData = getSyscallArg(2, buffer);
cRootSlot = extraCaps.excaprefs[0];
cRootCap = extraCaps.excaprefs[0]->cap;
vRootSlot = extraCaps.excaprefs[1];
vRootCap = extraCaps.excaprefs[1]->cap;
if (slotCapLongRunningDelete(
TCB_PTR_CTE_PTR(cap_thread_cap_get_capTCBPtr(cap), tcbCTable)) ||
slotCapLongRunningDelete(
TCB_PTR_CTE_PTR(cap_thread_cap_get_capTCBPtr(cap), tcbVTable))) {
userError("TCB SetSpace: CSpace or VSpace currently being deleted.");
current_syscall_error.type = seL4_IllegalOperation;
return EXCEPTION_SYSCALL_ERROR;
}
if (cRootData != 0) {
cRootCap = updateCapData(false, cRootData, cRootCap);
}
dc_ret = deriveCap(cRootSlot, cRootCap);
if (dc_ret.status != EXCEPTION_NONE) {
return dc_ret.status;
}
cRootCap = dc_ret.cap;
if (cap_get_capType(cRootCap) != cap_cnode_cap &&
(!config_set(CONFIG_ALLOW_NULL_CSPACE) ||
cap_get_capType(cRootCap) != cap_null_cap)) {
userError("TCB SetSpace: Invalid CNode cap.");
current_syscall_error.type = seL4_IllegalOperation;
return EXCEPTION_SYSCALL_ERROR;
}
if (vRootData != 0) {
vRootCap = updateCapData(false, vRootData, vRootCap);
}
dc_ret = deriveCap(vRootSlot, vRootCap);
if (dc_ret.status != EXCEPTION_NONE) {
return dc_ret.status;
}
vRootCap = dc_ret.cap;
if (!isValidVTableRoot(vRootCap)) {
userError("TCB SetSpace: Invalid VSpace cap.");
current_syscall_error.type = seL4_IllegalOperation;
return EXCEPTION_SYSCALL_ERROR;
}
setThreadState(ksCurThread, ThreadState_Restart);
return invokeTCB_ThreadControl(
TCB_PTR(cap_thread_cap_get_capTCBPtr(cap)), slot,
faultEP,
0, /* used to be prioInvalid, but it doesn't matter */
cRootCap, cRootSlot,
vRootCap, vRootSlot,
0, cap_null_cap_new(), NULL, thread_control_update_space);
}
deriveCap_ret_t
Arch_deriveCap(cte_t *slot, cap_t cap)
{
deriveCap_ret_t ret;
switch (cap_get_capType(cap)) {
case cap_page_global_directory_cap:
if (cap_page_global_directory_cap_get_capPGDIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a PDG cap without an assigned ASID");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_page_upper_directory_cap:
if (cap_page_upper_directory_cap_get_capPUDIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a PUD cap without an assigned ASID");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_page_directory_cap:
if (cap_page_directory_cap_get_capPDIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a PD cap without an assigned ASID");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_page_table_cap:
if (cap_page_table_cap_get_capPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a PT cap without an assigned ASID");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_frame_cap:
ret.cap = cap_frame_cap_set_capFMappedASID(cap, asidInvalid);
ret.status = EXCEPTION_NONE;
return ret;
case cap_asid_control_cap:
case cap_asid_pool_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
#ifdef CONFIG_ARM_HYPERVISOR_SUPPORT
case cap_vcpu_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
#endif
default:
/* This assert has no equivalent in haskell,
* as the options are restricted by type */
fail("Invalid arch cap");
}
}
deriveCap_ret_t Arch_deriveCap(cte_t *slot, cap_t cap)
{
deriveCap_ret_t ret;
switch (cap_get_capType(cap)) {
case cap_page_table_cap:
if (cap_page_table_cap_get_capPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving an unmapped PT cap");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_page_directory_cap:
if (cap_page_directory_cap_get_capPDIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a PD cap without an assigned ASID");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_asid_control_cap:
case cap_asid_pool_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
case cap_io_port_control_cap:
ret.status = EXCEPTION_NONE;
ret.cap = cap_null_cap_new();
return ret;
case cap_io_port_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
#ifdef CONFIG_IOMMU
case cap_io_space_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
case cap_io_page_table_cap:
if (cap_io_page_table_cap_get_capIOPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
#endif
#ifdef CONFIG_VTX
case cap_vcpu_cap:
ret.cap = cap;
ret.status = EXCEPTION_NONE;
return ret;
case cap_ept_pml4_cap:
if (cap_ept_pml4_cap_get_capPML4IsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving a EPT PML4 cap without an assigned ASID.");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_ept_pdpt_cap:
if (cap_ept_pdpt_cap_get_capPDPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving an unmapped EPT PDPT cap.");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_ept_pd_cap:
if (cap_ept_pd_cap_get_capPDIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving an unmapped EPT PD cap.");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
case cap_ept_pt_cap:
if (cap_ept_pt_cap_get_capPTIsMapped(cap)) {
ret.cap = cap;
ret.status = EXCEPTION_NONE;
} else {
userError("Deriving an unmapped EPT PT cap.");
current_syscall_error.type = seL4_IllegalOperation;
ret.cap = cap_null_cap_new();
ret.status = EXCEPTION_SYSCALL_ERROR;
}
return ret;
#endif
default:
return Mode_deriveCap(slot, cap);
}
}
void
fastpath_reply_recv(word_t cptr, word_t msgInfo)
{
seL4_MessageInfo_t info;
cap_t ep_cap;
endpoint_t *ep_ptr;
word_t length;
cte_t *callerSlot;
cap_t callerCap;
tcb_t *caller;
word_t badge;
tcb_t *endpointTail;
word_t fault_type;
cap_t newVTable;
pde_t *cap_pd;
pde_t stored_hw_asid;
/* Get message info and length */
info = messageInfoFromWord_raw(msgInfo);
length = seL4_MessageInfo_get_length(info);
fault_type = fault_get_faultType(ksCurThread->tcbFault);
#ifdef CONFIG_BENCHMARK_TRACK_KERNEL_ENTRIES
ksKernelEntry.path = Entry_Syscall;
ksKernelEntry.syscall_no = SysReplyRecv;
ksKernelEntry.cap_type = cap_endpoint_cap;
ksKernelEntry.invocation_tag = seL4_MessageInfo_get_label(info);
ksKernelEntry.is_fastpath = true;
benchmark_track_start();
#endif
#ifdef CONFIG_BENCHMARK_TRACK_UTILISATION
benchmark_utilisation_kentry_stamp();
#endif /* CONFIG_BENCHMARK_TRACK_UTILISATION */
/* Check there's no extra caps, the length is ok and there's no
* saved fault. */
if (unlikely(fastpath_mi_check(msgInfo) ||
fault_type != fault_null_fault)) {
slowpath(SysReplyRecv);
}
/* Lookup the cap */
ep_cap = lookup_fp(TCB_PTR_CTE_PTR(ksCurThread, tcbCTable)->cap,
cptr);
/* Check it's an endpoint */
if (unlikely(!cap_capType_equals(ep_cap, cap_endpoint_cap) ||
!cap_endpoint_cap_get_capCanReceive(ep_cap))) {
slowpath(SysReplyRecv);
}
/* Check there is nothing waiting on the notification */
if (ksCurThread->tcbBoundNotification &&
notification_ptr_get_state(ksCurThread->tcbBoundNotification) == NtfnState_Active) {
slowpath(SysReplyRecv);
}
/* Get the endpoint address */
ep_ptr = EP_PTR(cap_endpoint_cap_get_capEPPtr(ep_cap));
/* Check that there's not a thread waiting to send */
if (unlikely(endpoint_ptr_get_state(ep_ptr) == EPState_Send)) {
slowpath(SysReplyRecv);
}
/* Only reply if the reply cap is valid. */
callerSlot = TCB_PTR_CTE_PTR(ksCurThread, tcbCaller);
callerCap = callerSlot->cap;
if (unlikely(!fastpath_reply_cap_check(callerCap))) {
slowpath(SysReplyRecv);
}
/* Determine who the caller is. */
caller = TCB_PTR(cap_reply_cap_get_capTCBPtr(callerCap));
/* Check that the caller has not faulted, in which case a fault
reply is generated instead. */
fault_type = fault_get_faultType(caller->tcbFault);
if (unlikely(fault_type != fault_null_fault)) {
slowpath(SysReplyRecv);
}
/* Get destination thread.*/
newVTable = TCB_PTR_CTE_PTR(caller, tcbVTable)->cap;
/* Get vspace root. */
#if defined(ARCH_ARM) || !defined(CONFIG_PAE_PAGING)
cap_pd = PDE_PTR(cap_page_directory_cap_get_capPDBasePtr(newVTable));
#else
cap_pd = PDE_PTR(cap_pdpt_cap_get_capPDPTBasePtr(newVTable));
#endif
/* Ensure that the destination has a valid MMU. */
if (unlikely(! isValidVTableRoot_fp (newVTable))) {
slowpath(SysReplyRecv);
}
#ifdef ARCH_ARM
/* Get HWASID. */
stored_hw_asid = cap_pd[PD_ASID_SLOT];
#endif
/* Ensure the original caller can be scheduled directly. */
if (unlikely(caller->tcbPriority < ksCurThread->tcbPriority)) {
slowpath(SysReplyRecv);
}
#ifdef ARCH_ARM
/* Ensure the HWASID is valid. */
if (unlikely(!pde_pde_invalid_get_stored_asid_valid(stored_hw_asid))) {
slowpath(SysReplyRecv);
}
#endif
/* Ensure the original caller is in the current domain and can be scheduled directly. */
if (unlikely(caller->tcbDomain != ksCurDomain && maxDom)) {
slowpath(SysReplyRecv);
}
/*
* --- POINT OF NO RETURN ---
*
* At this stage, we have committed to performing the IPC.
*/
#ifdef ARCH_X86
/* Need to update NextIP in the calling thread */
setRegister(ksCurThread, NextIP, getRegister(ksCurThread, NextIP) + 2);
#endif
/* Set thread state to BlockedOnReceive */
thread_state_ptr_mset_blockingObject_tsType(
&ksCurThread->tcbState, (word_t)ep_ptr, ThreadState_BlockedOnReceive);
/* Place the thread in the endpoint queue */
endpointTail = TCB_PTR(endpoint_ptr_get_epQueue_tail(ep_ptr));
if (likely(!endpointTail)) {
ksCurThread->tcbEPPrev = NULL;
ksCurThread->tcbEPNext = NULL;
/* Set head/tail of queue and endpoint state. */
endpoint_ptr_set_epQueue_head_np(ep_ptr, TCB_REF(ksCurThread));
endpoint_ptr_mset_epQueue_tail_state(ep_ptr, TCB_REF(ksCurThread),
EPState_Recv);
} else {
/* Append current thread onto the queue. */
endpointTail->tcbEPNext = ksCurThread;
ksCurThread->tcbEPPrev = endpointTail;
ksCurThread->tcbEPNext = NULL;
/* Update tail of queue. */
endpoint_ptr_mset_epQueue_tail_state(ep_ptr, TCB_REF(ksCurThread),
EPState_Recv);
}
/* Delete the reply cap. */
mdb_node_ptr_mset_mdbNext_mdbRevocable_mdbFirstBadged(
&CTE_PTR(mdb_node_get_mdbPrev(callerSlot->cteMDBNode))->cteMDBNode,
0, 1, 1);
callerSlot->cap = cap_null_cap_new();
callerSlot->cteMDBNode = nullMDBNode;
/* I know there's no fault, so straight to the transfer. */
/* Replies don't have a badge. */
badge = 0;
fastpath_copy_mrs (length, ksCurThread, caller);
/* Dest thread is set Running, but not queued. */
thread_state_ptr_set_tsType_np(&caller->tcbState,
ThreadState_Running);
switchToThread_fp(caller, cap_pd, stored_hw_asid);
msgInfo = wordFromMessageInfo(seL4_MessageInfo_set_capsUnwrapped(info, 0));
#ifdef CONFIG_BENCHMARK_TRACK_KERNEL_ENTRIES
benchmark_track_exit();
#endif
fastpath_restore(badge, msgInfo, ksCurThread);
}
