m1n1/proxyclient/sprr_test_permissions.py
Sven Peter e58c264a07 gxf: add SPRR/GXF proxyclient experiments
Signed-off-by: Sven Peter <sven@svenpeter.dev>
2021-05-11 15:48:40 +09:00

363 lines
11 KiB
Python

#!/usr/bin/env python3
from setup import *
import asm
from contextlib import contextmanager
p.smp_start_secondaries()
class ARMPageTable:
PAGESIZE = 0x4000
def __init__(self, memalign, free):
self.memalign = memalign
self.free = free
self.l0 = self.memalign(self.PAGESIZE, self.PAGESIZE)
self.l1 = [self.memalign(self.PAGESIZE, self.PAGESIZE), self.memalign(
self.PAGESIZE, self.PAGESIZE)]
self.l2 = {}
p.write64(self.l0, self.make_table_pte(self.l1[0]))
p.write64(self.l0+8, self.make_table_pte(self.l1[1]))
def make_table_pte(self, addr):
# table mapping, access bit set
return addr | 0b11 | (1 << 10)
def map_page(self, vaddr, paddr, access_bits):
ap = (access_bits & 0b1100) >> 2
pxn = (access_bits & 0b0010) >> 1
uxn = (access_bits & 0b0001)
# block mapping, access bit set
pte = paddr | 0b01 | (1 << 10)
# move access bits in place
pte |= ap << 6
pte |= pxn << 54
pte |= uxn << 53
l0_idx = (vaddr >> (25+11+11)) & 1
l1_idx = (vaddr >> (25+11)) & 0x7ff
l2_idx = (vaddr >> 25) & 0x7ff
tbl = self.l2.get((l0_idx, l1_idx), None)
if not tbl:
tbl = self.memalign(self.PAGESIZE, self.PAGESIZE)
self.l2[(l0_idx, l1_idx)] = tbl
p.write64(self.l1[l0_idx] + 8*l1_idx, self.make_table_pte(tbl))
p.write64(tbl + 8*l2_idx, pte)
def map(self, vaddr, paddr, sz, access_bits):
assert sz % 0x2000000 == 0
assert vaddr % 0x2000000 == 0
assert paddr % 0x2000000 == 0
assert access_bits <= 0b1111
while sz > 0:
self.map_page(vaddr, paddr, access_bits)
sz -= 0x2000000
vaddr += 0x2000000
paddr += 0x2000000
def build_and_write_code(heap, code):
page = heap.memalign(0x4000, 0x4000)
compiled = asm.ARMAsm(code, page).data
iface.writemem(page, compiled)
p.dc_cvau(page, len(compiled))
p.ic_ivau(page, len(compiled))
return page
def setup_exception_vectors(heap, gxf=False):
if gxf:
elr = "S3_6_C15_C10_6"
eret = ".long 0x201400"
indicator = 0xf2
else:
elr = "ELR_EL1"
eret = "eret"
indicator = 0xf0
return build_and_write_code(heap, """
.rept 16
b 1f
.align 7
.endr
1:
// store that we failed
mov x10, 0x{indicator:x}
// move PC two instruction further and clear 0xf0 0000 0000 to
// make sure we end up in the r-x mapping either way and don't
// repeat the instruction that just faulted
// we skip the second instruction since that one is used to
// indicate success
ldr x11, =0x0fffffffff
mrs x12, {elr}
add x12, x12, #8
and x12, x12, x11
msr {elr}, x12
isb
{eret}
""".format(eret=eret, elr=elr, indicator=indicator))
print("Setting up..")
pagetable = ARMPageTable(u.memalign, u.free)
pagetable.map(0x800000000, 0x800000000, 0xc00000000, 0)
pagetable.map(0xf800000000, 0x800000000, 0xc00000000, 1)
el2_vectors = setup_exception_vectors(u.heap, gxf=False)
gl2_vectors = setup_exception_vectors(u.heap, gxf=True)
probe_page = build_and_write_code(u.heap, "mov x10, 0x80\nret\nret\nret\n")
probe_page_vaddr = probe_page | 0xf000000000
code_page = build_and_write_code(u.heap, """
#define SPRR_PERM_EL0 S3_6_C15_C1_5
#define SPRR_PERM_EL1 S3_6_C15_C1_6
#define SPRR_PERM_EL2 S3_6_C15_C1_7
#define GXF_CONFIG_EL2 s3_6_c15_c1_2
#define GXF_ENTER_EL2 s3_6_c15_c8_1
#define GXF_ABORT_EL2 s3_6_c15_c8_2
#define MPIDR_GL2 S3_6_C15_C10_1
#define VBAR_GL2 S3_6_C15_C10_2
#define SPSR_GL2 S3_6_C15_C10_3
#define ELR_GL2 S3_6_C15_C10_6
#define FAR_GL2 S3_6_C15_C10_7
#define genter .long 0x00201420
#define gexit .long 0x201400
// just store everything since i'm too lazy to think about
// register assignments
str x30, [sp, #-16]!
stp x28, x29, [sp, #-16]!
stp x26, x27, [sp, #-16]!
stp x24, x25, [sp, #-16]!
stp x22, x23, [sp, #-16]!
stp x20, x21, [sp, #-16]!
stp x18, x19, [sp, #-16]!
stp x16, x17, [sp, #-16]!
stp x14, x15, [sp, #-16]!
stp x12, x13, [sp, #-16]!
stp x10, x11, [sp, #-16]!
stp x8, x9, [sp, #-16]!
stp x6, x7, [sp, #-16]!
stp x4, x5, [sp, #-16]!
stp x2, x3, [sp, #-16]!
stp x0, x1, [sp, #-16]!
mov x20, x0 // store SPRR value for later
mov x21, 0 // clear result
// setup exception vectors
ldr x0, =0x{vectors:x}
msr VBAR_EL2, x0
isb
// prepare MMU
ldr x0, =0x0400ff
msr MAIR_EL1, x0
ldr x0, =0x27510b510
msr TCR_EL1, x0
ldr x0, =0x{ttbr:x}
msr TTBR0_EL2, x0
// enable SPPR
mov x0, 1
msr s3_6_c15_c1_0, x0
msr s3_6_c15_c1_3, xzr
isb
// clear all SPPR registers
// (note that reads from/writes to EL1 will be redirected to EL2 anyway)
ldr x0, =0
msr SPRR_PERM_EL0, x0
msr SPRR_PERM_EL1, x0
msr SPRR_PERM_EL2, x0
msr s3_6_c15_c1_3, x0
// setup SPPR_EL2
msr SPRR_PERM_EL2, x20
isb
dsb ishst
tlbi vmalle1is
dsb ish
isb
msr s3_6_c15_c1_3, xzr
isb
// enable MMU
ldr x1, =0x1005
mrs x0, SCTLR_EL1
mov x3, x0
orr x0, x0, x1
msr SCTLR_EL1, x0
isb
// configure and enable GXF
mov x0, 1
msr GXF_CONFIG_EL2, x0
isb
ldr x0, =gxf_entry
msr GXF_ENTER_EL2, x0
ldr x0, =gxf_abort
msr GXF_ABORT_EL2, x0
isb
// test GXF access
genter
// test execute access
ldr x1, =0x{probe_page:x}
mov x10, 0
blr x1
lsl x21, x21, #8
orr x21, x21, x10
// test read access
ldr x1, =0x{probe_page:x}
mov x10, 0
ldr x1, [x1]
mov x10, 0x80
lsl x21, x21, #8
orr x21, x21, x10
// test write access
ldr x1, =0x{probe_page:x}
add x1, x1, 0x20
mov x10, 0
str x1, [x1]
mov x10, 0x80
lsl x21, x21, #8
orr x21, x21, x10
// disable MMU again
dsb ish
isb
msr SCTLR_EL1, x3
isb
mov x0, 0
msr GXF_CONFIG_EL2, x0
msr s3_6_c15_c1_0, x0
mov x0, x21
// restore everything except for x0
add sp, sp, #8
ldr x1, [sp], #8
ldp x2, x3, [sp], #16
ldp x4, x5, [sp], #16
ldp x6, x7, [sp], #16
ldp x8, x9, [sp], #16
ldp x10, x11, [sp], #16
ldp x12, x13, [sp], #16
ldp x14, x15, [sp], #16
ldp x16, x17, [sp], #16
ldp x18, x19, [sp], #16
ldp x20, x21, [sp], #16
ldp x22, x23, [sp], #16
ldp x24, x25, [sp], #16
ldp x26, x27, [sp], #16
ldp x28, x29, [sp], #16
ldr x30, [sp], #16
ret
gxf_entry:
// setup GL exception vectors
ldr x0, =0x{gxf_vectors:x}
msr VBAR_GL2, x0
isb
// we might double fault -> store state here
mrs x14, S3_6_C15_C10_3
mrs x15, S3_6_C15_C10_4
mrs x16, S3_6_C15_C10_5
mrs x17, ELR_GL2
mrs x18, FAR_GL2
// test execute access
ldr x1, =0x{probe_page:x}
mov x10, 0
blr x1
lsl x21, x21, #8
orr x21, x21, x10
// test read access
ldr x1, =0x{probe_page:x}
mov x10, 0
ldr x1, [x1]
mov x10, 0x80
lsl x21, x21, #8
orr x21, x21, x10
// test write access
ldr x1, =0x{probe_page:x}
add x1, x1, #0x20
mov x10, 0
str x1, [x1]
mov x10, 0x80
lsl x21, x21, #8
orr x21, x21, x10
// restore state in case we faulted in here
msr S3_6_C15_C10_3, x14
msr S3_6_C15_C10_4, x15
msr S3_6_C15_C10_5, x16
msr ELR_GL2, x17
msr FAR_GL2, x18
isb
gexit
gxf_abort:
// store that we failed
mov x10, 0xf1
// move PC two instruction further and clear 0xf0 0000 0000 to
// make sure we end up in the r-x mapping either way and don't
// repeat the instruction that just faulted
// we skip the second instruction since that one is used to
// indicate success
ldr x11, =0x0fffffffff
mrs x12, ELR_GL2
add x12, x12, #8
and x12, x12, x11
msr ELR_GL2, x12
isb
gexit
""".format(ttbr=pagetable.l0, vectors=el2_vectors, probe_page=probe_page_vaddr, gxf_vectors=gl2_vectors))
print("Running code now...")
for i in range(0x10):
sprr_val = 0x5 | ((i & 0xf) << 4)
ret = p.smp_call_sync(1, code_page, sprr_val)
glret = ret >> 24
glx = 'x' if (glret >> 16) & 0xff == 0x80 else '-'
glr = 'r' if (glret >> 8) & 0xff == 0x80 else '-'
glw = 'w' if glret & 0xff == 0x80 else '-'
x = 'x' if (ret >> 16) & 0xff == 0x80 else '-'
r = 'r' if (ret >> 8) & 0xff == 0x80 else '-'
w = 'w' if ret & 0xff == 0x80 else '-'
print("SPRR: {0:04b} result: {1:x} GL: {2}{3}{4} EL: {5}{6}{7}".format(
i, ret, glr, glw, glx, r, w, x))