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 ALLOC_PT_GPREGS_ON_STACK
104 .macro SAVE_C_REGS_HELPER offset
=0 rax
=1 rcx
=1 r8910
=1 r11
=1
106 movq
%r11
, 6*8+\
offset(%rsp
)
109 movq
%r10
, 7*8+\
offset(%rsp
)
110 movq
%r9
, 8*8+\
offset(%rsp
)
111 movq
%r8
, 9*8+\
offset(%rsp
)
114 movq
%rax
, 10*8+\
offset(%rsp
)
117 movq
%rcx
, 11*8+\
offset(%rsp
)
119 movq
%rdx
, 12*8+\
offset(%rsp
)
120 movq
%rsi
, 13*8+\
offset(%rsp
)
121 movq
%rdi
, 14*8+\
offset(%rsp
)
122 UNWIND_HINT_REGS offset
=\offset extra
=0
124 .macro SAVE_C_REGS offset
=0
125 SAVE_C_REGS_HELPER \offset
, 1, 1, 1, 1
127 .macro SAVE_C_REGS_EXCEPT_RAX_RCX offset
=0
128 SAVE_C_REGS_HELPER \offset
, 0, 0, 1, 1
130 .macro SAVE_C_REGS_EXCEPT_R891011
131 SAVE_C_REGS_HELPER
0, 1, 1, 0, 0
133 .macro SAVE_C_REGS_EXCEPT_RCX_R891011
134 SAVE_C_REGS_HELPER
0, 1, 0, 0, 0
136 .macro SAVE_C_REGS_EXCEPT_RAX_RCX_R11
137 SAVE_C_REGS_HELPER
0, 0, 0, 1, 0
140 .macro SAVE_EXTRA_REGS offset
=0
141 movq
%r15
, 0*8+\
offset(%rsp
)
142 movq
%r14
, 1*8+\
offset(%rsp
)
143 movq
%r13
, 2*8+\
offset(%rsp
)
144 movq
%r12
, 3*8+\
offset(%rsp
)
145 movq
%rbp
, 4*8+\
offset(%rsp
)
146 movq
%rbx
, 5*8+\
offset(%rsp
)
147 UNWIND_HINT_REGS offset
=\offset
150 .macro POP_EXTRA_REGS
176 * This is a sneaky trick to help the unwinder find pt_regs on the stack. The
177 * frame pointer is replaced with an encoded pointer to pt_regs. The encoding
178 * is just setting the LSB, which makes it an invalid stack address and is also
179 * a signal to the unwinder that it's a pt_regs pointer in disguise.
181 * NOTE: This macro must be used *after* SAVE_EXTRA_REGS because it corrupts
184 .macro ENCODE_FRAME_POINTER ptregs_offset
=0
185 #ifdef CONFIG_FRAME_POINTER
187 leaq \
ptregs_offset(%rsp
), %rbp
195 #ifdef CONFIG_PAGE_TABLE_ISOLATION
198 * PAGE_TABLE_ISOLATION PGDs are 8k. Flip bit 12 to switch between the two
201 #define PTI_USER_PGTABLE_BIT PAGE_SHIFT
202 #define PTI_USER_PGTABLE_MASK (1 << PTI_USER_PGTABLE_BIT)
203 #define PTI_USER_PCID_BIT X86_CR3_PTI_PCID_USER_BIT
204 #define PTI_USER_PCID_MASK (1 << PTI_USER_PCID_BIT)
205 #define PTI_USER_PGTABLE_AND_PCID_MASK (PTI_USER_PCID_MASK | PTI_USER_PGTABLE_MASK)
207 .macro SET_NOFLUSH_BIT reg
:req
208 bts $X86_CR3_PCID_NOFLUSH_BIT
, \reg
211 .macro ADJUST_KERNEL_CR3 reg
:req
212 ALTERNATIVE
"", "SET_NOFLUSH_BIT \reg", X86_FEATURE_PCID
213 /* Clear PCID and "PAGE_TABLE_ISOLATION bit", point CR3 at kernel pagetables: */
214 andq $
(~PTI_USER_PGTABLE_AND_PCID_MASK
), \reg
217 .macro SWITCH_TO_KERNEL_CR3 scratch_reg
:req
218 ALTERNATIVE
"jmp .Lend_\@", "", X86_FEATURE_PTI
219 mov
%cr3
, \scratch_reg
220 ADJUST_KERNEL_CR3 \scratch_reg
221 mov \scratch_reg
, %cr3
225 #define THIS_CPU_user_pcid_flush_mask \
226 PER_CPU_VAR(cpu_tlbstate) + TLB_STATE_user_pcid_flush_mask
228 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg
:req scratch_reg2
:req
229 ALTERNATIVE
"jmp .Lend_\@", "", X86_FEATURE_PTI
230 mov
%cr3
, \scratch_reg
232 ALTERNATIVE
"jmp .Lwrcr3_\@", "", X86_FEATURE_PCID
235 * Test if the ASID needs a flush.
237 movq \scratch_reg
, \scratch_reg2
238 andq $
(0x7FF), \scratch_reg
/* mask ASID */
239 bt \scratch_reg
, THIS_CPU_user_pcid_flush_mask
242 /* Flush needed, clear the bit */
243 btr \scratch_reg
, THIS_CPU_user_pcid_flush_mask
244 movq \scratch_reg2
, \scratch_reg
248 movq \scratch_reg2
, \scratch_reg
249 SET_NOFLUSH_BIT \scratch_reg
252 /* Flip the ASID to the user version */
253 orq $
(PTI_USER_PCID_MASK
), \scratch_reg
256 /* Flip the PGD to the user version */
257 orq $
(PTI_USER_PGTABLE_MASK
), \scratch_reg
258 mov \scratch_reg
, %cr3
262 .macro SWITCH_TO_USER_CR3_STACK scratch_reg
:req
264 SWITCH_TO_USER_CR3_NOSTACK scratch_reg
=\scratch_reg scratch_reg2
=%rax
268 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg
:req save_reg
:req
269 ALTERNATIVE
"jmp .Ldone_\@", "", X86_FEATURE_PTI
270 movq
%cr3
, \scratch_reg
271 movq \scratch_reg
, \save_reg
273 * Test the user pagetable bit. If set, then the user page tables
274 * are active. If clear CR3 already has the kernel page table
277 bt $PTI_USER_PGTABLE_BIT
, \scratch_reg
280 ADJUST_KERNEL_CR3 \scratch_reg
281 movq \scratch_reg
, %cr3
286 .macro RESTORE_CR3 scratch_reg
:req save_reg
:req
287 ALTERNATIVE
"jmp .Lend_\@", "", X86_FEATURE_PTI
289 ALTERNATIVE
"jmp .Lwrcr3_\@", "", X86_FEATURE_PCID
292 * KERNEL pages can always resume with NOFLUSH as we do
295 bt $PTI_USER_PGTABLE_BIT
, \save_reg
299 * Check if there's a pending flush for the user ASID we're
302 movq \save_reg
, \scratch_reg
303 andq $
(0x7FF), \scratch_reg
304 bt \scratch_reg
, THIS_CPU_user_pcid_flush_mask
307 btr \scratch_reg
, THIS_CPU_user_pcid_flush_mask
311 SET_NOFLUSH_BIT \save_reg
315 * The CR3 write could be avoided when not changing its value,
316 * but would require a CR3 read *and* a scratch register.
322 #else /* CONFIG_PAGE_TABLE_ISOLATION=n: */
324 .macro SWITCH_TO_KERNEL_CR3 scratch_reg
:req
326 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg
:req scratch_reg2
:req
328 .macro SWITCH_TO_USER_CR3_STACK scratch_reg
:req
330 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg
:req save_reg
:req
332 .macro RESTORE_CR3 scratch_reg
:req save_reg
:req
337 #endif /* CONFIG_X86_64 */
340 * This does 'call enter_from_user_mode' unless we can avoid it based on
341 * kernel config or using the static jump infrastructure.
343 .macro CALL_enter_from_user_mode
344 #ifdef CONFIG_CONTEXT_TRACKING
345 #ifdef HAVE_JUMP_LABEL
346 STATIC_JUMP_IF_FALSE
.Lafter_call_\@
, context_tracking_enabled
, def
=0
348 call enter_from_user_mode