xref: /openbmc/linux/arch/x86/entry/calling.h (revision 708078f6)
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>
9 #include <asm/inst.h>
10 
11 /*
12 
13  x86 function call convention, 64-bit:
14  -------------------------------------
15   arguments           |  callee-saved      | extra caller-saved | return
16  [callee-clobbered]   |                    | [callee-clobbered] |
17  ---------------------------------------------------------------------------
18  rdi rsi rdx rcx r8-9 | rbx rbp [*] r12-15 | r10-11             | rax, rdx [**]
19 
20  ( rsp is obviously invariant across normal function calls. (gcc can 'merge'
21    functions when it sees tail-call optimization possibilities) rflags is
22    clobbered. Leftover arguments are passed over the stack frame.)
23 
24  [*]  In the frame-pointers case rbp is fixed to the stack frame.
25 
26  [**] for struct return values wider than 64 bits the return convention is a
27       bit more complex: up to 128 bits width we return small structures
28       straight in rax, rdx. For structures larger than that (3 words or
29       larger) the caller puts a pointer to an on-stack return struct
30       [allocated in the caller's stack frame] into the first argument - i.e.
31       into rdi. All other arguments shift up by one in this case.
32       Fortunately this case is rare in the kernel.
33 
34 For 32-bit we have the following conventions - kernel is built with
35 -mregparm=3 and -freg-struct-return:
36 
37  x86 function calling convention, 32-bit:
38  ----------------------------------------
39   arguments         | callee-saved        | extra caller-saved | return
40  [callee-clobbered] |                     | [callee-clobbered] |
41  -------------------------------------------------------------------------
42  eax edx ecx        | ebx edi esi ebp [*] | <none>             | eax, edx [**]
43 
44  ( here too esp is obviously invariant across normal function calls. eflags
45    is clobbered. Leftover arguments are passed over the stack frame. )
46 
47  [*]  In the frame-pointers case ebp is fixed to the stack frame.
48 
49  [**] We build with -freg-struct-return, which on 32-bit means similar
50       semantics as on 64-bit: edx can be used for a second return value
51       (i.e. covering integer and structure sizes up to 64 bits) - after that
52       it gets more complex and more expensive: 3-word or larger struct returns
53       get done in the caller's frame and the pointer to the return struct goes
54       into regparm0, i.e. eax - the other arguments shift up and the
55       function's register parameters degenerate to regparm=2 in essence.
56 
57 */
58 
59 #ifdef CONFIG_X86_64
60 
61 /*
62  * 64-bit system call stack frame layout defines and helpers,
63  * for assembly code:
64  */
65 
66 /* The layout forms the "struct pt_regs" on the stack: */
67 /*
68  * C ABI says these regs are callee-preserved. They aren't saved on kernel entry
69  * unless syscall needs a complete, fully filled "struct pt_regs".
70  */
71 #define R15		0*8
72 #define R14		1*8
73 #define R13		2*8
74 #define R12		3*8
75 #define RBP		4*8
76 #define RBX		5*8
77 /* These regs are callee-clobbered. Always saved on kernel entry. */
78 #define R11		6*8
79 #define R10		7*8
80 #define R9		8*8
81 #define R8		9*8
82 #define RAX		10*8
83 #define RCX		11*8
84 #define RDX		12*8
85 #define RSI		13*8
86 #define RDI		14*8
87 /*
88  * On syscall entry, this is syscall#. On CPU exception, this is error code.
89  * On hw interrupt, it's IRQ number:
90  */
91 #define ORIG_RAX	15*8
92 /* Return frame for iretq */
93 #define RIP		16*8
94 #define CS		17*8
95 #define EFLAGS		18*8
96 #define RSP		19*8
97 #define SS		20*8
98 
99 #define SIZEOF_PTREGS	21*8
100 
101 .macro PUSH_AND_CLEAR_REGS rdx=%rdx rax=%rax save_ret=0
102 	/*
103 	 * Push registers and sanitize registers of values that a
104 	 * speculation attack might otherwise want to exploit. The
105 	 * lower registers are likely clobbered well before they
106 	 * could be put to use in a speculative execution gadget.
107 	 * Interleave XOR with PUSH for better uop scheduling:
108 	 */
109 	.if \save_ret
110 	pushq	%rsi		/* pt_regs->si */
111 	movq	8(%rsp), %rsi	/* temporarily store the return address in %rsi */
112 	movq	%rdi, 8(%rsp)	/* pt_regs->di (overwriting original return address) */
113 	.else
114 	pushq   %rdi		/* pt_regs->di */
115 	pushq   %rsi		/* pt_regs->si */
116 	.endif
117 	pushq	\rdx		/* pt_regs->dx */
118 	xorl	%edx, %edx	/* nospec   dx */
119 	pushq   %rcx		/* pt_regs->cx */
120 	xorl	%ecx, %ecx	/* nospec   cx */
121 	pushq   \rax		/* pt_regs->ax */
122 	pushq   %r8		/* pt_regs->r8 */
123 	xorl	%r8d, %r8d	/* nospec   r8 */
124 	pushq   %r9		/* pt_regs->r9 */
125 	xorl	%r9d, %r9d	/* nospec   r9 */
126 	pushq   %r10		/* pt_regs->r10 */
127 	xorl	%r10d, %r10d	/* nospec   r10 */
128 	pushq   %r11		/* pt_regs->r11 */
129 	xorl	%r11d, %r11d	/* nospec   r11*/
130 	pushq	%rbx		/* pt_regs->rbx */
131 	xorl    %ebx, %ebx	/* nospec   rbx*/
132 	pushq	%rbp		/* pt_regs->rbp */
133 	xorl    %ebp, %ebp	/* nospec   rbp*/
134 	pushq	%r12		/* pt_regs->r12 */
135 	xorl	%r12d, %r12d	/* nospec   r12*/
136 	pushq	%r13		/* pt_regs->r13 */
137 	xorl	%r13d, %r13d	/* nospec   r13*/
138 	pushq	%r14		/* pt_regs->r14 */
139 	xorl	%r14d, %r14d	/* nospec   r14*/
140 	pushq	%r15		/* pt_regs->r15 */
141 	xorl	%r15d, %r15d	/* nospec   r15*/
142 	UNWIND_HINT_REGS
143 	.if \save_ret
144 	pushq	%rsi		/* return address on top of stack */
145 	.endif
146 .endm
147 
148 .macro POP_REGS pop_rdi=1 skip_r11rcx=0
149 	popq %r15
150 	popq %r14
151 	popq %r13
152 	popq %r12
153 	popq %rbp
154 	popq %rbx
155 	.if \skip_r11rcx
156 	popq %rsi
157 	.else
158 	popq %r11
159 	.endif
160 	popq %r10
161 	popq %r9
162 	popq %r8
163 	popq %rax
164 	.if \skip_r11rcx
165 	popq %rsi
166 	.else
167 	popq %rcx
168 	.endif
169 	popq %rdx
170 	popq %rsi
171 	.if \pop_rdi
172 	popq %rdi
173 	.endif
174 .endm
175 
176 /*
177  * This is a sneaky trick to help the unwinder find pt_regs on the stack.  The
178  * frame pointer is replaced with an encoded pointer to pt_regs.  The encoding
179  * is just setting the LSB, which makes it an invalid stack address and is also
180  * a signal to the unwinder that it's a pt_regs pointer in disguise.
181  *
182  * NOTE: This macro must be used *after* PUSH_AND_CLEAR_REGS because it corrupts
183  * the original rbp.
184  */
185 .macro ENCODE_FRAME_POINTER ptregs_offset=0
186 #ifdef CONFIG_FRAME_POINTER
187 	leaq 1+\ptregs_offset(%rsp), %rbp
188 #endif
189 .endm
190 
191 #ifdef CONFIG_PAGE_TABLE_ISOLATION
192 
193 /*
194  * PAGE_TABLE_ISOLATION PGDs are 8k.  Flip bit 12 to switch between the two
195  * halves:
196  */
197 #define PTI_USER_PGTABLE_BIT		PAGE_SHIFT
198 #define PTI_USER_PGTABLE_MASK		(1 << PTI_USER_PGTABLE_BIT)
199 #define PTI_USER_PCID_BIT		X86_CR3_PTI_PCID_USER_BIT
200 #define PTI_USER_PCID_MASK		(1 << PTI_USER_PCID_BIT)
201 #define PTI_USER_PGTABLE_AND_PCID_MASK  (PTI_USER_PCID_MASK | PTI_USER_PGTABLE_MASK)
202 
203 .macro SET_NOFLUSH_BIT	reg:req
204 	bts	$X86_CR3_PCID_NOFLUSH_BIT, \reg
205 .endm
206 
207 .macro ADJUST_KERNEL_CR3 reg:req
208 	ALTERNATIVE "", "SET_NOFLUSH_BIT \reg", X86_FEATURE_PCID
209 	/* Clear PCID and "PAGE_TABLE_ISOLATION bit", point CR3 at kernel pagetables: */
210 	andq    $(~PTI_USER_PGTABLE_AND_PCID_MASK), \reg
211 .endm
212 
213 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
214 	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
215 	mov	%cr3, \scratch_reg
216 	ADJUST_KERNEL_CR3 \scratch_reg
217 	mov	\scratch_reg, %cr3
218 .Lend_\@:
219 .endm
220 
221 #define THIS_CPU_user_pcid_flush_mask   \
222 	PER_CPU_VAR(cpu_tlbstate) + TLB_STATE_user_pcid_flush_mask
223 
224 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req
225 	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
226 	mov	%cr3, \scratch_reg
227 
228 	ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID
229 
230 	/*
231 	 * Test if the ASID needs a flush.
232 	 */
233 	movq	\scratch_reg, \scratch_reg2
234 	andq	$(0x7FF), \scratch_reg		/* mask ASID */
235 	bt	\scratch_reg, THIS_CPU_user_pcid_flush_mask
236 	jnc	.Lnoflush_\@
237 
238 	/* Flush needed, clear the bit */
239 	btr	\scratch_reg, THIS_CPU_user_pcid_flush_mask
240 	movq	\scratch_reg2, \scratch_reg
241 	jmp	.Lwrcr3_pcid_\@
242 
243 .Lnoflush_\@:
244 	movq	\scratch_reg2, \scratch_reg
245 	SET_NOFLUSH_BIT \scratch_reg
246 
247 .Lwrcr3_pcid_\@:
248 	/* Flip the ASID to the user version */
249 	orq	$(PTI_USER_PCID_MASK), \scratch_reg
250 
251 .Lwrcr3_\@:
252 	/* Flip the PGD to the user version */
253 	orq     $(PTI_USER_PGTABLE_MASK), \scratch_reg
254 	mov	\scratch_reg, %cr3
255 .Lend_\@:
256 .endm
257 
258 .macro SWITCH_TO_USER_CR3_STACK	scratch_reg:req
259 	pushq	%rax
260 	SWITCH_TO_USER_CR3_NOSTACK scratch_reg=\scratch_reg scratch_reg2=%rax
261 	popq	%rax
262 .endm
263 
264 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req
265 	ALTERNATIVE "jmp .Ldone_\@", "", X86_FEATURE_PTI
266 	movq	%cr3, \scratch_reg
267 	movq	\scratch_reg, \save_reg
268 	/*
269 	 * Test the user pagetable bit. If set, then the user page tables
270 	 * are active. If clear CR3 already has the kernel page table
271 	 * active.
272 	 */
273 	bt	$PTI_USER_PGTABLE_BIT, \scratch_reg
274 	jnc	.Ldone_\@
275 
276 	ADJUST_KERNEL_CR3 \scratch_reg
277 	movq	\scratch_reg, %cr3
278 
279 .Ldone_\@:
280 .endm
281 
282 .macro RESTORE_CR3 scratch_reg:req save_reg:req
283 	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
284 
285 	ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID
286 
287 	/*
288 	 * KERNEL pages can always resume with NOFLUSH as we do
289 	 * explicit flushes.
290 	 */
291 	bt	$PTI_USER_PGTABLE_BIT, \save_reg
292 	jnc	.Lnoflush_\@
293 
294 	/*
295 	 * Check if there's a pending flush for the user ASID we're
296 	 * about to set.
297 	 */
298 	movq	\save_reg, \scratch_reg
299 	andq	$(0x7FF), \scratch_reg
300 	bt	\scratch_reg, THIS_CPU_user_pcid_flush_mask
301 	jnc	.Lnoflush_\@
302 
303 	btr	\scratch_reg, THIS_CPU_user_pcid_flush_mask
304 	jmp	.Lwrcr3_\@
305 
306 .Lnoflush_\@:
307 	SET_NOFLUSH_BIT \save_reg
308 
309 .Lwrcr3_\@:
310 	/*
311 	 * The CR3 write could be avoided when not changing its value,
312 	 * but would require a CR3 read *and* a scratch register.
313 	 */
314 	movq	\save_reg, %cr3
315 .Lend_\@:
316 .endm
317 
318 #else /* CONFIG_PAGE_TABLE_ISOLATION=n: */
319 
320 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
321 .endm
322 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req
323 .endm
324 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req
325 .endm
326 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req
327 .endm
328 .macro RESTORE_CR3 scratch_reg:req save_reg:req
329 .endm
330 
331 #endif
332 
333 .macro STACKLEAK_ERASE_NOCLOBBER
334 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
335 	PUSH_AND_CLEAR_REGS
336 	call stackleak_erase
337 	POP_REGS
338 #endif
339 .endm
340 
341 .macro SAVE_AND_SET_GSBASE scratch_reg:req save_reg:req
342 	rdgsbase \save_reg
343 	GET_PERCPU_BASE \scratch_reg
344 	wrgsbase \scratch_reg
345 .endm
346 
347 #endif /* CONFIG_X86_64 */
348 
349 .macro STACKLEAK_ERASE
350 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
351 	call stackleak_erase
352 #endif
353 .endm
354 
355 #ifdef CONFIG_SMP
356 
357 /*
358  * CPU/node NR is loaded from the limit (size) field of a special segment
359  * descriptor entry in GDT.
360  */
361 .macro LOAD_CPU_AND_NODE_SEG_LIMIT reg:req
362 	movq	$__CPUNODE_SEG, \reg
363 	lsl	\reg, \reg
364 .endm
365 
366 /*
367  * Fetch the per-CPU GSBASE value for this processor and put it in @reg.
368  * We normally use %gs for accessing per-CPU data, but we are setting up
369  * %gs here and obviously can not use %gs itself to access per-CPU data.
370  */
371 .macro GET_PERCPU_BASE reg:req
372 	ALTERNATIVE \
373 		"LOAD_CPU_AND_NODE_SEG_LIMIT \reg", \
374 		"RDPID	\reg", \
375 		X86_FEATURE_RDPID
376 	andq	$VDSO_CPUNODE_MASK, \reg
377 	movq	__per_cpu_offset(, \reg, 8), \reg
378 .endm
379 
380 #else
381 
382 .macro GET_PERCPU_BASE reg:req
383 	movq	pcpu_unit_offsets(%rip), \reg
384 .endm
385 
386 #endif /* CONFIG_SMP */
387 
388 /*
389  * This does 'call enter_from_user_mode' unless we can avoid it based on
390  * kernel config or using the static jump infrastructure.
391  */
392 .macro CALL_enter_from_user_mode
393 #ifdef CONFIG_CONTEXT_TRACKING
394 #ifdef CONFIG_JUMP_LABEL
395 	STATIC_JUMP_IF_FALSE .Lafter_call_\@, context_tracking_enabled, def=0
396 #endif
397 	call enter_from_user_mode
398 .Lafter_call_\@:
399 #endif
400 .endm
401