Skip to content

Commit

Permalink
More work for guests on several cores...
Browse files Browse the repository at this point in the history
  • Loading branch information
@didriklundberg
didriklundberg committed Aug 6, 2015
1 parent 9422213 commit 1a0b0ba
Show file tree
Hide file tree
Showing 5 changed files with 82 additions and 131 deletions.
1 change: 0 additions & 1 deletion core/hw/cpu/arm/arm_common/arm_guest_blob.S
View file
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,6 @@ arm_move_guest_blob:

foreach_guest:
/* Get size & padr. make sure size is 4-aligned. */

ldr r4, [r1], #-4 @@ VADR
ldr r5, [r1], #-4 @@ PSIZE
ldr r6, [r1], #-4 @@ FWSIZE
Expand Down
77 changes: 31 additions & 46 deletions core/hw/cpu/arm/arm_common/arm_pt.S
View file
Original file line number Diff line number Diff line change
Expand Up @@ -59,59 +59,24 @@ arm_setup_initial_pt_slave:
add r0, r0, r1
ldr r4, r0

/* Place zero in r1 for later usage, and store the physical starting address
* of the secondary page table with an offset of 0x4000 in r2. */
//mov r1, #0
//add r2, r0, #0x4000

/* First, clean the memory used to make all pages invalid. */
/* TODO: No need to do this again? mov r1, #0 is be removable in that case
1: str r1, [r0], #4
str r1, [r0], #4
str r1, [r0], #4
str r1, [r0], #4
teq r0, r2
bne 1b */

/* Do a 1->1 temporary mapping for our current position.
TODO: Should already be done.
ldr r6, =(HAL_PHYS_START >> 20)
ldr r7, =(HAL_PHYS_START | 0x40E) @ CB on, Section, XN=f, AP =01 PRIV RW USER NO ACCESS
str r7, [r4, r6, lsl #2] @ 1:1 mapping for current position */

/* Then, map us from HAL_VIR_START to HAL_PHY_START in L1.
* TODO: Changes needed here? */
//add r0, r4, #(HAL_VIRT_START & 0xff000000) >> 18
//str r7, [r0, #(HAL_VIRT_START & 0x00f00000) >> 18]! @ Map phys to HAL_VIRT_START

/* Enable MMU
* TODO: Do everything again? */
/* Enable MMU */
mov r0, #0
//mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
mcr p15, 0, r0, c8, c7, 0 @ flush I, D, TLB
mcr p15, 0, r0, c8, c7, 0 @ Flush instruction and data caches plus TLB
mcr p15, 0, r0, c1, c0, 0 @ Read control register
orr r0, r0 , #1 @ Enable MMU bit
mov r1, #-1 @ Sets manager to all domains
mcr p15, 0, r1, c3, c0, 0 @ Load domain access control
mcr p15, 0, r4, c2, c0, 0 @ Load page table pointer
mcr p15, 0, r0, c1, c0, 0 @ Write control reg

mcr p15, 0, r0, c1, c0, 0 @ read control register
orr r0, r0 , #1 @ enable MMU bit
mov r1, #-1 @ sets manager to all domains
mcr p15, 0, r1, c3, c0, 0 @ load domain access control
mcr p15, 0, r4, c2, c0, 0 @ load page table pointer

mcr p15, 0, r0, c1, c0, 0 @ write control reg

/* Adjust stack and jump to our virtual address.
* TODO: This part should probably be identical... */
/* Adjust stack and jump to our virtual address. */
ldr r1, =HAL_OFFSET
sub sp, sp, r1
sub pc, pc, r1

nop @ For PC jump above
nop @ For PC jump above

//ldr r4, =__hyper_pt_start__
//ldr r1, =HAL_OFFSET
//add r4, r4, r1
//mcr p15, 0, r4, c2, c0, 0 @ load page table pointer

sub pc, lr, r1 @ correct lr to virt space

arm_reset_initial_pt:
Expand Down Expand Up @@ -176,9 +141,29 @@ arm_setup_initial_pt:

/* Now running in virt space, we can remove 1:1 mapping now - or NOT.
* TODO: Current version should not remove the temporary mapping.
* We should instead copy it to
* We should instead copy it to */

/* TODO: Copy multiple words at a time? Which registers are used?
LDMCopy:
push {r4-r10}
LDMloop:
ldmia r1!, {r3 - r10}
stmia r0!, {r3 - r10}
subs r2, r2, #32
bge LDMloop
*/

ldr r0, =__hyper_pt_start_slave__
ldr r0, =__hyper_pt_start__
ldr r2, r6
copy_loop:
ldr r3, [r1], #4
str r3, [r0], #4
subs r2, r2, #4
bge copy_loop

mov r0, #0
ldr r3, =(HAL_PHYS_START >> 20)
str r0, [r4, r3, lsl #2] */
str r0, [r4, r3, lsl #2]

sub pc, lr, r1 @ correct lr to virt space
6 changes: 1 addition & 5 deletions core/hypervisor/handlers.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,10 +7,7 @@ extern void hypercall_dyn_new_pgd(addr_t *pgd_va);
extern void hypercall_dyn_set_pmd(addr_t *pmd, uint32_t desc);
extern void hypercall_dyn_set_pte(addr_t *l2pt_linux_entry_va, uint32_t linux_pte, uint32_t phys_pte);
extern int dmmu_handler(uint32_t p03, uint32_t p1, uint32_t p2);

//TODO: Fix this...
extern int __hyper_stack_bottom__;

extern virtual_machine* vms[4];

//This function returns a pointer to the VM currently running on this core,
Expand All @@ -33,11 +30,10 @@ virtual_machine* get_curr_vm(){
//TODO: Note that 13 is hard-coded from the stack size. In the slightly
//more general case of the stack size being a power of two,
temp_stack_pointer = temp_stack_pointer >> 13;
//TODO: Where do we decide how many VMs we have? Until then:

//4. Return curr_vm, which is a pointer to the current machine on this
//particular core.
return vms[temp_stack_pointer]; //TODO: Entirely cricket?
return vms[temp_stack_pointer];
}

#define USE_DMMU
Expand Down
2 changes: 1 addition & 1 deletion core/hypervisor/init.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
//TODO: Note: Added these to avoid warnings.
extern void dmmu_init();
extern uint32_t dmmu_map_L1_section(addr_t va, addr_t sec_base_add, uint32_t attrs);
extern void debug_inf(void); //TODO. Remove after debugging
extern void debug_inf(void); //TODO: Remove after debugging

//#define DEBUG_PG_CONTENT
//#define DEBUG_L1_PG_TYPE
Expand Down
127 changes: 49 additions & 78 deletions core/hypervisor/init_slave.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -35,16 +35,14 @@ void guests_init_multicore(){
uint32_t i, guest = 0;

//Set the ID of the virtual machines.
//TODO: Hard-coded for 4...
//TODO: Hard-coded for 4 cores/VMs...
vm_0.id = 0;
vm_1.id = 1;
vm_2.id = 2;
vm_3.id = 3;

//The virtual machines are stored in some sort of linked list.
//TODO: As long as we have an hard-coded array this linked list seems
//pointless...
//TODO: Hard-coded for 4...
//TODO: Hard-coded for 4 cores/VMs...
vm_0.next = &vm_1;
vm_1.next = &vm_2;
vm_2.next = &vm_3;
Expand Down Expand Up @@ -75,9 +73,6 @@ void guests_init_multicore(){
guests_db.guests[i].fwsize);
}

/* We start with vm_0 as the current virtual machine. */
curr_vm = &vm_0;

//The configuration of the guest contains location of the always cacheable
//region inside the guest, among other things. This can be found inside the
//sub-directory guest_config.
Expand All @@ -95,13 +90,9 @@ void guests_init_multicore(){
vm_2.config->firmware = get_guest(1 + guest++);
vm_3.config->firmware = get_guest(1 + guest++);

//TODO: Why do we store these values in temporary variables???
//TODO: Note: later on, we can't use the same vstart and pstart for the other guests.
// This would be the ideal starting point for a loop, setting pstart and vstart to
// different values.
uint32_t guest_number;
//TODO: 4 is hard-coded for 4 guests.
for (guest_number = 0; guest_number < 4; guest_number++){
/* We start with vm_0 as the current virtual machine. */
curr_vm = &vm_0;
do {

addr_t guest_psize = curr_vm->config->firmware->psize;
addr_t guest_vstart = curr_vm->config->firmware->vstart;
Expand All @@ -120,9 +111,6 @@ void guests_init_multicore(){
* addresses are mapped by the virtual addresses 0x00000000 to 0x004FFFFF.
* TODO: This memory sub-space must be accessible only to the hypervisor. */

//TODO: Note: We want to do this for every guest. For each loop, we want
//va_offset to start from a number which is guest_psize higher than the
//previous.
uint32_t va_offset;
for (va_offset = 0;
//TODO: Mathematically speaking, there is no reason to add section_size
Expand All @@ -132,8 +120,7 @@ void guests_init_multicore(){
va_offset + SECTION_SIZE <= guest_psize;
va_offset += SECTION_SIZE){
uint32_t offset, pmd;
//TODO: Take a look at the line below...
uint32_t va = vms[guest_number].config->reserved_va_for_pt_access_start + va_offset;
uint32_t va = curr_vm->config->reserved_va_for_pt_access_start + va_offset;
uint32_t pa = guest_pstart + va_offset;
pt_create_section(flpt_va, va, pa, MLT_HYPER_RAM);

Expand All @@ -143,36 +130,38 @@ void guests_init_multicore(){
COP_WRITE(COP_SYSTEM, COP_DCACHE_INVALIDATE_MVA, pmd);
}

//Invalidates cache (TODO: Automatically of current core?).
//Invalidate cache.
memory_commit();

//printf("HV pagetable after guests initialization:\n"); //DEBUG
//dump_mmu(flpt_va); //DEBUG
curr_vm = curr_vm->next;
} while (curr_vm.id != 0);

//We pin the L2s that can be created in the 32 KiB area of slpt_va.
//TODO: Presumably we need to do this once for each guest. We might want i
//to start from different numbers depending on the index of the guest.
dmmu_entry_t * bft = (dmmu_entry_t *)DMMU_BFT_BASE_VA;
for (i=0; i*4096<0x8000; i++) {
bft[PA_TO_PH_BLOCK((uint32_t)GET_PHYS(slpt_va) + i*4096)].type = PAGE_INFO_TYPE_L2PT;
bft[PA_TO_PH_BLOCK((uint32_t)GET_PHYS(slpt_va) + i*4096)].refcnt = 1;
}
//printf("HV pagetable after guests initialization:\n"); //DEBUG
//dump_mmu(flpt_va); //DEBUG

/* At this point we are finished initializing the master page table, and can
* start initializing the first guest page table.
* The master page table now contains
* 1) The virtual mapping to the hypervisor code and data.
* 2) A fixed virtual mapping to the guest PT.
* 3) Some reserved mapping that we ignore for now, e.g. IO‌REGS.
* 4) A 1-1 mapping to the guest memory (as defined in board_mem.c) writable
* and readable by the user.
* TODO: THIS‌ SETUP ‌MUST ‌BE ‌FIXED, SINCE ‌THE ‌GUEST ‌IS ‌NOT ‌ALLOWED ‌TO ‌WRITE
* INTO ITS ‌WHOLE‌ MEMORY */

//Create a copy of the master page table for the guest in the physical
//address pa_initial_l1.
//TODO: This needs to be done once for each guest, with different addresses.
//We pin the L2s that can be created in the 32 KiB area of slpt_va.
//TODO: This should only be done once.
dmmu_entry_t * bft = (dmmu_entry_t *)DMMU_BFT_BASE_VA;
for (i=0; i*4096<0x8000; i++) {
bft[PA_TO_PH_BLOCK((uint32_t)GET_PHYS(slpt_va) + i*4096)].type = PAGE_INFO_TYPE_L2PT;
bft[PA_TO_PH_BLOCK((uint32_t)GET_PHYS(slpt_va) + i*4096)].refcnt = 1;
}

/* At this point we are finished initializing the master page table, and can
* start initializing the first guest page table.
* The master page table now contains
* 1) The virtual mapping to the hypervisor code and data.
* 2) A fixed virtual mapping to the guest PT.
* 3) Some reserved mapping that we ignore for now, e.g. IO‌REGS.
* 4) A 1-1 mapping to the guest memory (as defined in board_mem.c) writable
* and readable by the user.
* TODO: THIS‌ SETUP ‌MUST ‌BE ‌FIXED, SINCE ‌THE ‌GUEST ‌IS ‌NOT ‌ALLOWED ‌TO ‌WRITE
* INTO ITS ‌WHOLE‌ MEMORY */

do {
/* Create a copy of the master page table for the guest in the physical
* address pa_initial_l1. */
uint32_t *guest_pt_va;
addr_t guest_pt_pa;

Expand All @@ -185,15 +174,13 @@ void guests_init_multicore(){
//dump_mmu(guest_pt_va); //DEBUG

/* Activate the guest page table. */
//TODO: Also needs to be done once for each guest.
memory_commit();
COP_WRITE(COP_SYSTEM, COP_SYSTEM_TRANSLATION_TABLE0, guest_pt_pa); //Set TTB0
isb();
memory_commit();

//Calling the create_L1_pt API to check the correctness of the L1 content and
//to change the page table type to 1.
//TODO: Also needs to be done once for each guest.
/* Calling the create_L1_pt API to check the correctness of the L1
* content and to change the page table type to 1. */
uint32_t res = dmmu_create_L1_pt(guest_pt_pa);
if (res != SUCCESS_MMU){
printf("Error: Failed to create the initial PT with error code %d.\n",
Expand All @@ -211,18 +198,17 @@ void guests_init_multicore(){
}
#endif

//Initialize the datastructures with the type for the initial L1.
//Create the attribute that allow the guest to read/write/execute.
/* Initialize the datastructures with the type for the initial L1.
* Create the attribute that allow the guest to read/write/execute. */
uint32_t attrs;
attrs = 0x12; // 0b1--10
attrs |= MMU_AP_USER_RW << MMU_SECTION_AP_SHIFT;
attrs = (attrs & (~0x10)) | 0xC | (HC_DOM_KERNEL << MMU_L1_DOMAIN_SHIFT);

//As default the guest has a 1-to-1 mapping to all its memory
//TODO: This loop needs to be done for all the guests.
/* As default the guest has a 1-to-1 mapping to all its memory
* TODO: This loop needs to be done for all the guests. */
uint32_t offset;
for (offset = 0;
//TODO: This loop really does leave 1 MiB at end, unlike the loop above.
offset + SECTION_SIZE <= guest_psize;
offset += SECTION_SIZE){
printf(" Creating initial mapping of %x to %x...\n",
Expand All @@ -247,7 +233,7 @@ void guests_init_multicore(){
}
#endif

// switch back to the master-pagetable
/* Switch back to the master page table */
memory_commit();
COP_WRITE(COP_SYSTEM, COP_SYSTEM_TRANSLATION_TABLE0, (uint32_t)GET_PHYS(__hyper_pt_start__)); //Set TTB0
isb();
Expand All @@ -256,22 +242,12 @@ void guests_init_multicore(){

/* END GUANCIO CHANGES */
/* END KTH CHANGES */
}

//TODO: This should only be done once.
#ifdef TRUSTED
//Gets the second guest, then increments guest counter by one.
get_guest(guest++);
curr_vm->mode_states[HC_GM_TRUSTED].ctx.sp = curr_vm->config->rpc_handlers->sp;
curr_vm->mode_states[HC_GM_TRUSTED].ctx.psr= ARM_INTERRUPT_MASK | ARM_MODE_USER;
#endif
curr_vm = curr_vm->next;
} while (curr_vm.id != 0);

guest = 0;

//Initialize the context with the physical addresses.
//TODO: This should only be done once.
//TODO: This appears to be the only part of this function which is already
//written for several VMs.
do{
/* Initialize default values */
for(i = 0; i < HC_NGUESTMODES;i++){
Expand All @@ -296,18 +272,19 @@ void guests_init_multicore(){
//that this entire loops works for several VMs.
} while(curr_vm != &vm_0);

//TODO: This obviously needs to be done once for every VM.
memory_commit();


// defines the active vm for each core
//Define the active virtual machine for each core
vms[0] = &vm_0;
vms[1] = &vm_1;
vms[2] = &vm_2;
vms[3] = &vm_3;

// set-up for each core current_context to point to the context of the active virtual machine
cpu_context_initial_set(&curr_vm->mode_states[HC_GM_KERNEL].ctx);
//Set-up for each core current_context to point to the context of the active virtual machine
do{
cpu_context_initial_set(&curr_vm->mode_states[HC_GM_KERNEL].ctx);
} while(curr_vm != &vm_0);
}


Expand All @@ -316,12 +293,6 @@ void slave_start_(){
* core/hw/cpu/family/model/cpu_init.c. Since caches are separate among
* cores, this should be run once for each core. */
cpu_init();

/* Initialize hypervisor guest modes and data structures according to config
* file in guest - defined in init.c. */
//guests_init(); //TODO: This should be run by the main core, and initialize
// all secondary cores.

start_guest(); //TODO: Currently we have a shared curr_vm, which includes
// among other things the state of processor registers.
start_guest();
}

0 comments on commit 1a0b0ba

Please sign in to comment.