1 /* SPDX-License-Identifier: GPL-2.0 */
2 #include <linux/jump_label.h>
3 #include <asm/unwind_hints.h>
4 #include <asm/cpufeatures.h>
5 #include <asm/page_types.h>
6 #include <asm/percpu.h>
7 #include <asm/asm-offsets.h>
8 #include <asm/processor-flags.h>
12 x86 function call convention, 64-bit:
13 -------------------------------------
14 arguments | callee-saved | extra caller-saved | return
15 [callee-clobbered] | | [callee-clobbered] |
16 ---------------------------------------------------------------------------
17 rdi rsi rdx rcx r8-9 | rbx rbp [*] r12-15 | r10-11 | rax, rdx [**]
19 ( rsp is obviously invariant across normal function calls. (gcc can 'merge'
20 functions when it sees tail-call optimization possibilities) rflags is
21 clobbered. Leftover arguments are passed over the stack frame.)
23 [*] In the frame-pointers case rbp is fixed to the stack frame.
25 [**] for struct return values wider than 64 bits the return convention is a
26 bit more complex: up to 128 bits width we return small structures
27 straight in rax, rdx. For structures larger than that (3 words or
28 larger) the caller puts a pointer to an on-stack return struct
29 [allocated in the caller's stack frame] into the first argument - i.e.
30 into rdi. All other arguments shift up by one in this case.
31 Fortunately this case is rare in the kernel.
33 For 32-bit we have the following conventions - kernel is built with
34 -mregparm=3 and -freg-struct-return:
36 x86 function calling convention, 32-bit:
37 ----------------------------------------
38 arguments | callee-saved | extra caller-saved | return
39 [callee-clobbered] | | [callee-clobbered] |
40 -------------------------------------------------------------------------
41 eax edx ecx | ebx edi esi ebp [*] | <none> | eax, edx [**]
43 ( here too esp is obviously invariant across normal function calls. eflags
44 is clobbered. Leftover arguments are passed over the stack frame. )
46 [*] In the frame-pointers case ebp is fixed to the stack frame.
48 [**] We build with -freg-struct-return, which on 32-bit means similar
49 semantics as on 64-bit: edx can be used for a second return value
50 (i.e. covering integer and structure sizes up to 64 bits) - after that
51 it gets more complex and more expensive: 3-word or larger struct returns
52 get done in the caller's frame and the pointer to the return struct goes
53 into regparm0, i.e. eax - the other arguments shift up and the
54 function's register parameters degenerate to regparm=2 in essence.
61 * 64-bit system call stack frame layout defines and helpers,
65 /* The layout forms the "struct pt_regs" on the stack: */
67 * C ABI says these regs are callee-preserved. They aren't saved on kernel entry
68 * unless syscall needs a complete, fully filled "struct pt_regs".
76 /* These regs are callee-clobbered. Always saved on kernel entry. */
87 * On syscall entry, this is syscall#. On CPU exception, this is error code.
88 * On hw interrupt, it's IRQ number:
91 /* Return frame for iretq */
98 #define SIZEOF_PTREGS 21*8
100 .macro PUSH_AND_CLEAR_REGS rdx
=%rdx rax
=%rax
102 * Push registers and sanitize registers of values that a
103 * speculation attack might otherwise want to exploit. The
104 * lower registers are likely clobbered well before they
105 * could be put to use in a speculative execution gadget.
106 * Interleave XOR with PUSH for better uop scheduling:
108 pushq
%rdi
/* pt_regs->di */
109 pushq
%rsi
/* pt_regs->si */
110 pushq
\rdx
/* pt_regs->dx */
111 pushq
%rcx
/* pt_regs->cx */
112 pushq
\rax
/* pt_regs->ax */
113 pushq
%r8
/* pt_regs->r8 */
114 xorq
%r8
, %r8
/* nospec r8 */
115 pushq
%r9
/* pt_regs->r9 */
116 xorq
%r9
, %r9
/* nospec r9 */
117 pushq
%r10
/* pt_regs->r10 */
118 xorq
%r10
, %r10
/* nospec r10 */
119 pushq
%r11
/* pt_regs->r11 */
120 xorq
%r11
, %r11
/* nospec r11*/
121 pushq
%rbx
/* pt_regs->rbx */
122 xorl
%ebx
, %ebx
/* nospec rbx*/
123 pushq
%rbp
/* pt_regs->rbp */
124 xorl
%ebp
, %ebp
/* nospec rbp*/
125 pushq
%r12
/* pt_regs->r12 */
126 xorq
%r12
, %r12
/* nospec r12*/
127 pushq
%r13
/* pt_regs->r13 */
128 xorq
%r13
, %r13
/* nospec r13*/
129 pushq
%r14
/* pt_regs->r14 */
130 xorq
%r14
, %r14
/* nospec r14*/
131 pushq
%r15
/* pt_regs->r15 */
132 xorq
%r15
, %r15
/* nospec r15*/
136 .macro POP_REGS pop_rdi
=1 skip_r11rcx
=0
165 * This is a sneaky trick to help the unwinder find pt_regs on the stack. The
166 * frame pointer is replaced with an encoded pointer to pt_regs. The encoding
167 * is just setting the LSB, which makes it an invalid stack address and is also
168 * a signal to the unwinder that it's a pt_regs pointer in disguise.
170 * NOTE: This macro must be used *after* PUSH_AND_CLEAR_REGS because it corrupts
173 .macro ENCODE_FRAME_POINTER ptregs_offset
=0
174 #ifdef CONFIG_FRAME_POINTER
176 leaq \
ptregs_offset(%rsp
), %rbp
184 #ifdef CONFIG_PAGE_TABLE_ISOLATION
187 * PAGE_TABLE_ISOLATION PGDs are 8k. Flip bit 12 to switch between the two
190 #define PTI_USER_PGTABLE_BIT PAGE_SHIFT
191 #define PTI_USER_PGTABLE_MASK (1 << PTI_USER_PGTABLE_BIT)
192 #define PTI_USER_PCID_BIT X86_CR3_PTI_PCID_USER_BIT
193 #define PTI_USER_PCID_MASK (1 << PTI_USER_PCID_BIT)
194 #define PTI_USER_PGTABLE_AND_PCID_MASK (PTI_USER_PCID_MASK | PTI_USER_PGTABLE_MASK)
196 .macro SET_NOFLUSH_BIT reg
:req
197 bts $X86_CR3_PCID_NOFLUSH_BIT
, \reg
200 .macro ADJUST_KERNEL_CR3 reg
:req
201 ALTERNATIVE
"", "SET_NOFLUSH_BIT \reg", X86_FEATURE_PCID
202 /* Clear PCID and "PAGE_TABLE_ISOLATION bit", point CR3 at kernel pagetables: */
203 andq $
(~PTI_USER_PGTABLE_AND_PCID_MASK
), \reg
206 .macro SWITCH_TO_KERNEL_CR3 scratch_reg
:req
207 ALTERNATIVE
"jmp .Lend_\@", "", X86_FEATURE_PTI
208 mov
%cr3
, \scratch_reg
209 ADJUST_KERNEL_CR3 \scratch_reg
210 mov \scratch_reg
, %cr3
214 #define THIS_CPU_user_pcid_flush_mask \
215 PER_CPU_VAR(cpu_tlbstate) + TLB_STATE_user_pcid_flush_mask
217 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg
:req scratch_reg2
:req
218 ALTERNATIVE
"jmp .Lend_\@", "", X86_FEATURE_PTI
219 mov
%cr3
, \scratch_reg
221 ALTERNATIVE
"jmp .Lwrcr3_\@", "", X86_FEATURE_PCID
224 * Test if the ASID needs a flush.
226 movq \scratch_reg
, \scratch_reg2
227 andq $
(0x7FF), \scratch_reg
/* mask ASID */
228 bt \scratch_reg
, THIS_CPU_user_pcid_flush_mask
231 /* Flush needed, clear the bit */
232 btr \scratch_reg
, THIS_CPU_user_pcid_flush_mask
233 movq \scratch_reg2
, \scratch_reg
237 movq \scratch_reg2
, \scratch_reg
238 SET_NOFLUSH_BIT \scratch_reg
241 /* Flip the ASID to the user version */
242 orq $
(PTI_USER_PCID_MASK
), \scratch_reg
245 /* Flip the PGD to the user version */
246 orq $
(PTI_USER_PGTABLE_MASK
), \scratch_reg
247 mov \scratch_reg
, %cr3
251 .macro SWITCH_TO_USER_CR3_STACK scratch_reg
:req
253 SWITCH_TO_USER_CR3_NOSTACK scratch_reg
=\scratch_reg scratch_reg2
=%rax
257 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg
:req save_reg
:req
258 ALTERNATIVE
"jmp .Ldone_\@", "", X86_FEATURE_PTI
259 movq
%cr3
, \scratch_reg
260 movq \scratch_reg
, \save_reg
262 * Test the user pagetable bit. If set, then the user page tables
263 * are active. If clear CR3 already has the kernel page table
266 bt $PTI_USER_PGTABLE_BIT
, \scratch_reg
269 ADJUST_KERNEL_CR3 \scratch_reg
270 movq \scratch_reg
, %cr3
275 .macro RESTORE_CR3 scratch_reg
:req save_reg
:req
276 ALTERNATIVE
"jmp .Lend_\@", "", X86_FEATURE_PTI
278 ALTERNATIVE
"jmp .Lwrcr3_\@", "", X86_FEATURE_PCID
281 * KERNEL pages can always resume with NOFLUSH as we do
284 bt $PTI_USER_PGTABLE_BIT
, \save_reg
288 * Check if there's a pending flush for the user ASID we're
291 movq \save_reg
, \scratch_reg
292 andq $
(0x7FF), \scratch_reg
293 bt \scratch_reg
, THIS_CPU_user_pcid_flush_mask
296 btr \scratch_reg
, THIS_CPU_user_pcid_flush_mask
300 SET_NOFLUSH_BIT \save_reg
304 * The CR3 write could be avoided when not changing its value,
305 * but would require a CR3 read *and* a scratch register.
311 #else /* CONFIG_PAGE_TABLE_ISOLATION=n: */
313 .macro SWITCH_TO_KERNEL_CR3 scratch_reg
:req
315 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg
:req scratch_reg2
:req
317 .macro SWITCH_TO_USER_CR3_STACK scratch_reg
:req
319 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg
:req save_reg
:req
321 .macro RESTORE_CR3 scratch_reg
:req save_reg
:req
326 #endif /* CONFIG_X86_64 */
329 * This does 'call enter_from_user_mode' unless we can avoid it based on
330 * kernel config or using the static jump infrastructure.
332 .macro CALL_enter_from_user_mode
333 #ifdef CONFIG_CONTEXT_TRACKING
334 #ifdef HAVE_JUMP_LABEL
335 STATIC_JUMP_IF_FALSE
.Lafter_call_\@
, context_tracking_enabled
, def
=0
337 call enter_from_user_mode