x86, objtool: several fixes/improvements
v2 cover x86, objtool: several fixes/improvements
v2:
- Use ud2 instead of calling kvm_spurious_fault() in vmx_vmenter() [Paolo]
- Fix ____kvm_handle_fault_on_reboot() comment [Paolo]
- Tweak "switch jump table" comment [Peter]
- Add review tags
Patches also available at:git://git.kernel.org/pub/scm/linux/kernel/git/jpoimboe/linux.git objtool-many-fixes-v2
v1 was posted here:
https://lkml.kernel.org/r/[email protected]
-------------------
There have been a lot of objtool bug reports lately, mainly related to Clang and BPF. As part of fixing those bugs, I added some improvements to objtool which uncovered yet more bugs (some kernel, some objtool).
I've given these patches a lot of testing with both GCC and Clang. More compile testing of objtool would be appreciated, as the kbuild test robot doesn't seem to be paying much attention to my branches lately.
There are still at least three outstanding issues:
1) With clang I see:
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.o: warning: objtool: .altinstr_replacement+0x88: redundant UACCESS disable
I haven't dug into it yet.
2) There's also an issue in clang where a large switch table had a bunch of unused (bad) entries. It's not a code correctness issue, but hopefully it can get fixed in clang anyway. See patch 20/22 for more details.
3) CONFIG_LIVEPATCH is causing some objtool "unreachable instruction" warnings due to the new -flive-patching flag. I have some fixes pending, but this patch set is already long enough.
Jann Horn (1):
objtool: Support repeated uses of the same C jump table
Josh Poimboeuf (21):
x86/paravirt: Fix callee-saved function ELF sizes
x86/kvm: Fix fastop function ELF metadata
x86/kvm: Replace vmx_vmenter()'s call to kvm_spurious_fault() with UD2 x86/kvm: Don't call kvm_spurious_fault() from .fixup x86/entry: Fix thunk function ELF sizes
x86/head/64: Annotate start_cpu0() as non-callable x86/uaccess: Remove ELF function annotation from
copy_user_handle_tail()
x86/uaccess: Don't leak AC flag into fentry from mcsafe_handle_tail() x86/uaccess: Remove redundant CLACs in getuser/putuser error paths
bpf: Disable GCC -fgcse optimization for ___bpf_prog_run()
objtool: Add mcsafe_handle_tail() to the uaccess safe list
objtool: Track original function across branches
objtool: Refactor function alias logic
objtool: Warn on zero-length functions
objtool: Change dead_end_function() to return boolean
objtool: Do frame pointer check before dead end check
objtool: Refactor sibling call detection logic
objtool: Refactor jump table code
objtool: Fix seg fault on bad switch table entry
objtool: convert insn type to enum
objtool: Support conditional retpolines
arch/x86/entry/thunk_64.S | 5 +- arch/x86/include/asm/kvm_host.h | 34 ++-- arch/x86/include/asm/paravirt.h | 1 + arch/x86/kernel/head_64.S | 4 +- arch/x86/kernel/kvm.c | 1 + arch/x86/kvm/emulate.c | 44 +++-- arch/x86/kvm/vmx/vmenter.S | 6 +- arch/x86/lib/copy_user_64.S | 2 +- arch/x86/lib/getuser.S | 20 +-- arch/x86/lib/putuser.S | 29 +-- arch/x86/lib/usercopy_64.c | 2 +- include/linux/compiler-gcc.h | 2 + include/linux/compiler_types.h | 4 + kernel/bpf/core.c | 2 +- tools/objtool/arch.h | 36 ++-- tools/objtool/arch/x86/decode.c | 2 +- tools/objtool/check.c | 308 ++++++++++++++++---------------- tools/objtool/check.h | 3 +- tools/objtool/elf.c | 4 +- tools/objtool/elf.h | 3 +- 20 files changed, 278 insertions(+), 234 deletions(-)
Signature · 1 line
2.20.1
PATCH v2 1/22 x86/paravirt: Fix callee-saved function ELF sizes
The __raw_callee_save_*() functions have an ELF symbol size of zero, which confuses objtool and other tools.
Fixes a bunch of warnings like the following:
arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_pte_val() is missing an ELF size annotation arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_pgd_val() is missing an ELF size annotation arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_make_pte() is missing an ELF size annotation arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_make_pgd() is missing an ELF size annotation
---
Cc: Juergen Gross <[email protected]>
--- arch/x86/include/asm/paravirt.h | 1 + arch/x86/kernel/kvm.c | 1 + 2 files changed, 2 insertions(+)
arch/x86/include/asm/paravirt.h +1 -0
@@ -746,6 +746,7 @@ bool __raw_callee_save___native_vcpu_is_preempted(long cpu);PV_RESTORE_ALL_CALLER_REGS \ FRAME_END \"ret;" \ + ".size " PV_THUNK_NAME(func) ", .-" PV_THUNK_NAME(func) ";" \ ".popsection")/* Get a reference to a callee-save function */
arch/x86/kernel/kvm.c +1 -0
@@ -838,6 +838,7 @@ asm( "cmpb $0, " __stringify(KVM_STEAL_TIME_preempted) "+steal_time(%rax);""setne %al;" "ret;" +".size __raw_callee_save___kvm_vcpu_is_preempted, .-__raw_callee_save___kvm_vcpu_is_preempted;" ".popsection");#endif
Signature · 1 line
2.20.1
PATCH v2 2/22 x86/kvm: Fix fastop function ELF metadata
In reply to Josh PoimboeufSome of the fastop functions, e.g. em_setcc(), are actually just used as global labels which point to blocks of functions. The global labels are incorrectly annotated as functions. Also the functions themselves don't have size annotations.
Fixes a bunch of warnings like the following:
arch/x86/kvm/emulate.o: warning: objtool: seto() is missing an ELF size annotation arch/x86/kvm/emulate.o: warning: objtool: em_setcc() is missing an ELF size annotation arch/x86/kvm/emulate.o: warning: objtool: setno() is missing an ELF size annotation arch/x86/kvm/emulate.o: warning: objtool: setc() is missing an ELF size annotation
---
Cc: Paolo Bonzini <[email protected]>
Cc: Radim Krčmář <[email protected]>
--- arch/x86/kvm/emulate.c | 44 +++++++++++++++++++++++++++++------------- 1 file changed, 31 insertions(+), 13 deletions(-)
arch/x86/kvm/emulate.c +31 -13
@@ -312,29 +312,42 @@ static void invalidate_registers(struct x86_emulate_ctxt *ctxt)static int fastop (struct x86_emulate_ctxt *ctxt,void (*fop)(struct fastop *)); -#define FOP_FUNC (name) \ +#define __FOP_FUNC (name) \ ".align " __stringify(FASTOP_SIZE) " \n\t" \ ".type " name ", @function \n\t" \ name ":\n\t" -#define FOP_RET "ret \n\t" +#define FOP_FUNC (name) \ + __FOP_FUNC(#name) + +#define __FOP_RET (name) \ + "ret \n\t" \ + ".size " name ", .-" name "\n\t" + +#define FOP_RET (name) \ + __FOP_RET(#name)#define FOP_START (op) \ extern void em_##op(struct fastop *fake); \ asm(".pushsection .text, \"ax\" \n\t" \ ".global em_" #op " \n\t" \ - FOP_FUNC("em_" #op) + ".align " __stringify(FASTOP_SIZE) " \n\t" \ + "em_" #op ":\n\t"#define FOP_END \ ".popsection") +#define __FOPNOP (name) \ + __FOP_FUNC(name) \ + __FOP_RET(name) +#define FOPNOP () \ - FOP_FUNC(__stringify(__UNIQUE_ID(nop))) \ - FOP_RET + __FOPNOP(__stringify(__UNIQUE_ID(nop)))#define FOP1E (op, dst) \ - FOP_FUNC(#op "_" #dst) \ - "10: " #op " %" #dst " \n\t" FOP_RET + __FOP_FUNC(#op "_" #dst) \ + "10: " #op " %" #dst " \n\t" \ + __FOP_RET(#op "_" #dst)#define FOP1EEX (op, dst) \ FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception)
@@ -366,8 +379,9 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *)); FOP_END#define FOP2E (op, dst, src) \ - FOP_FUNC(#op "_" #dst "_" #src) \ - #op " %" #src ", %" #dst " \n\t" FOP_RET + __FOP_FUNC(#op "_" #dst "_" #src) \ + #op " %" #src ", %" #dst " \n\t" \ + __FOP_RET(#op "_" #dst "_" #src)#define FASTOP2 (op) \ FOP_START(op) \
@@ -405,8 +419,9 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *)); FOP_END#define FOP3E (op, dst, src, src2) \ - FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \ - #op " %" #src2 ", %" #src ", %" #dst " \n\t" FOP_RET + __FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \ + #op " %" #src2 ", %" #src ", %" #dst " \n\t"\ + __FOP_RET(#op "_" #dst "_" #src "_" #src2)/* 3-operand, word-only, src2=cl */ #define FASTOP3WCL (op) \
@@ -423,7 +438,7 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *)); ".type " #op ", @function \n\t" \ #op ": \n\t" \ #op " %al \n\t" \ - FOP_RET + __FOP_RET(#op) asm(".pushsection .fixup, \"ax\"\n" ".global kvm_fastop_exception \n"
@@ -449,7 +464,10 @@ FOP_SETCC(setle)FOP_SETCC (setnle)FOP_END ; -FOP_START(salc) "pushf; sbb %al, %al; popf \n\t" FOP_RET +FOP_START(salc) +FOP_FUNC(salc) +"pushf; sbb %al, %al; popf \n\t" +FOP_RET (salc) FOP_END;/*
Signature · 1 line
2.20.1
PATCH v2 4/22 x86/kvm: Don't call kvm_spurious_fault() from .fixup
In reply to Josh PoimboeufAfter making a change to improve objtool's sibling call detection, it started showing the following warning:
arch/x86/kvm/vmx/nested.o: warning: objtool: .fixup+0x15: sibling call from callable instruction with modified stack frame
The problem is the ____kvm_handle_fault_on_reboot() macro. It does a fake call by pushing a fake RIP and doing a jump. That tricks the unwinder into printing the function which triggered the exception, rather than the .fixup code.
Instead of the hack to make it look like the original function made the call, just change the macro so that the original function actually does make the call. This allows removal of the hack, and also makes objtool happy.
I triggered a vmx instruction exception and verified that the stack trace is still sane:
kernel BUG at arch/x86/kvm/x86.c:358!
invalid opcode: 0000 [#1] SMP PTI
CPU: 28 PID: 4096 Comm: qemu-kvm Not tainted 5.2.0+ #16 Hardware name: Lenovo THINKSYSTEM SD530 -[7X2106Z000]-/-[7X2106Z000]-, BIOS -[TEE113Z-1.00]- 07/17/2017 RIP: 0010:kvm_spurious_fault+0x5/0x10 Code: 00 00 00 00 00 8b 44 24 10 89 d2 45 89 c9 48 89 44 24 10 8b 44 24 08 48 89 44 24 08 e9 d4 40 22 00 0f 1f 40 00 0f 1f 44 00 00 <0f> 0b 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 55 49 89 fd 41 RSP: 0018:ffffbf91c683bd00 EFLAGS: 00010246 RAX: 000061f040000000 RBX: ffff9e159c77bba0 RCX: ffff9e15a5c87000 RDX: 0000000665c87000 RSI: ffff9e15a5c87000 RDI: ffff9e159c77bba0 RBP: 0000000000000000 R08: 0000000000000000 R09: ffff9e15a5c87000 R10: 0000000000000000 R11: fffff8f2d99721c0 R12: ffff9e159c77bba0 R13: ffffbf91c671d960 R14: ffff9e159c778000 R15: 0000000000000000 FS: 00007fa341cbe700(0000) GS:ffff9e15b7400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fdd38356804 CR3: 00000006759de003 CR4: 00000000007606e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 PKRU: 55555554 Call Trace: loaded_vmcs_init+0x4f/0xe0 alloc_loaded_vmcs+0x38/0xd0 vmx_create_vcpu+0xf7/0x600 kvm_vm_ioctl+0x5e9/0x980 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? free_one_page+0x13f/0x4e0 do_vfs_ioctl+0xa4/0x630 ksys_ioctl+0x60/0x90 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x55/0x1c0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fa349b1ee5b
---
v2: Fix ____kvm_handle_fault_on_reboot() comment [Paolo]
--- arch/x86/include/asm/kvm_host.h | 34 ++++++++++++++++++--------------- 1 file changed, 19 insertions(+), 15 deletions(-)
arch/x86/include/asm/kvm_host.h +19 -15
@@ -1496,25 +1496,29 @@ enum {#define kvm_arch_vcpu_memslots_id (vcpu) ((vcpu)->arch.hflags & HF_SMM_MASK ? 1 : 0)#define kvm_memslots_for_spte_role (kvm, role) __kvm_memslots(kvm, (role).smm) +asmlinkage void __noreturn kvm_spurious_fault(void); +/* * Hardware virtualization extension instructions may fault if a * reboot turns off virtualization while processes are running. -* Trap the fault and ignore the instruction if that happens. +* Usually after catching the fault we just panic; during reboot +* instead the instruction is ignored. */ -asmlinkage void kvm_spurious_fault(void); - -#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ - "666: " insn "\n\t" \ - "668: \n\t" \ - ".pushsection .fixup, \"ax\" \n" \ - "667: \n\t" \ - cleanup_insn "\n\t" \ - "cmpb $0, kvm_rebooting \n\t" \ - "jne 668b \n\t" \ - __ASM_SIZE(push) " $666b \n\t" \ - "jmp kvm_spurious_fault \n\t" \ - ".popsection \n\t" \ - _ASM_EXTABLE(666b, 667b) +#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ + "666: \n\t" \ + insn "\n\t" \ + "jmp 668f \n\t" \ + "667: \n\t" \ + "call kvm_spurious_fault \n\t" \ + "668: \n\t" \ + ".pushsection .fixup, \"ax\" \n\t" \ + "700: \n\t" \ + cleanup_insn "\n\t" \ + "cmpb $0, kvm_rebooting\n\t" \ + "je 667b \n\t" \ + "jmp 668b \n\t" \ + ".popsection \n\t" \ + _ASM_EXTABLE(666b, 700b)#define __kvm_handle_fault_on_reboot (insn) \ ____kvm_handle_fault_on_reboot(insn, "")
Signature · 1 line
2.20.1
PATCH v2 3/22 x86/kvm: Replace vmx_vmenter()'s call to kvm_spurious_fault() with UD2
In reply to Josh PoimboeufObjtool reports the following:
arch/x86/kvm/vmx/vmenter.o: warning: objtool: vmx_vmenter()+0x14: call without frame pointer save/setup
But frame pointers are necessarily broken anyway, because __vmx_vcpu_run() clobbers RBP with the guest's value before calling vmx_vmenter(). So calling without a frame pointer doesn't make things any worse.
Make objtool happy by changing the call to a UD2.
---
v2: ud2 instead of kvm_spurious_fault() [Paolo]
--- arch/x86/kvm/vmx/vmenter.S | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
arch/x86/kvm/vmx/vmenter.S +3 -3
@@ -54,9 +54,9 @@ ENTRY(vmx_vmenter) ret 3: cmpb $0, kvm_rebooting - jne 4f - call kvm_spurious_fault -4: ret + je 4f + ret +4: ud2 .pushsection .fixup, "ax" 5: jmp 3b
Signature · 1 line
2.20.1
PATCH v2 6/22 x86/head/64: Annotate start_cpu0() as non-callable
In reply to Josh PoimboeufAfter an objtool improvement, it complains about the fact that start_cpu0() jumps to code which has an LRET instruction.
arch/x86/kernel/head_64.o: warning: objtool: .head.text+0xe4: unsupported instruction in callable function
Technically, start_cpu0() is callable, but it acts nothing like a callable function. Prevent objtool from treating it like one by removing its ELF function annotation.
--- arch/x86/kernel/head_64.S | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/head_64.S +2 -2
@@ -253,10 +253,10 @@ END(secondary_startup_64) * start_secondary() via .Ljump_to_C_code. */ ENTRY(start_cpu0) - movq initial_stack(%rip), %rsp UNWIND_HINT_EMPTY + movq initial_stack(%rip), %rsp jmp .Ljump_to_C_code -ENDPROC(start_cpu0) +END(start_cpu0) #endif /* Both SMP bootup and ACPI suspend change these variables */
Signature · 1 line
2.20.1
PATCH v2 9/22 x86/uaccess: Remove redundant CLACs in getuser/putuser error paths
In reply to Josh PoimboeufThe same getuser/putuser error paths are used regardless of whether AC is set. In non-exception failure cases, this results in an unnecessary CLAC.
Fixes the following warnings:
arch/x86/lib/getuser.o: warning: objtool: .altinstr_replacement+0x18: redundant UACCESS disable arch/x86/lib/putuser.o: warning: objtool: .altinstr_replacement+0x18: redundant UACCESS disable
--- arch/x86/lib/getuser.S | 20 ++++++++++---------- arch/x86/lib/putuser.S | 29 ++++++++++++++++------------- 2 files changed, 26 insertions(+), 23 deletions(-)
arch/x86/lib/getuser.S +10 -10
@@ -115,29 +115,29 @@ ENDPROC(__get_user_8) EXPORT_SYMBOL(__get_user_8) +bad_get_user_clac: + ASM_CLAC bad_get_user: xor %edx,%edx mov $(-EFAULT),%_ASM_AX - ASM_CLAC ret -END(bad_get_user) #ifdef CONFIG_X86_32 +bad_get_user_8_clac: + ASM_CLAC bad_get_user_8: xor %edx,%edx xor %ecx,%ecx mov $(-EFAULT),%_ASM_AX - ASM_CLAC ret -END(bad_get_user_8) #endif - _ASM_EXTABLE_UA(1b, bad_get_user) - _ASM_EXTABLE_UA(2b, bad_get_user) - _ASM_EXTABLE_UA(3b, bad_get_user) + _ASM_EXTABLE_UA(1b, bad_get_user_clac) + _ASM_EXTABLE_UA(2b, bad_get_user_clac) + _ASM_EXTABLE_UA(3b, bad_get_user_clac) #ifdef CONFIG_X86_64 - _ASM_EXTABLE_UA(4b, bad_get_user) + _ASM_EXTABLE_UA(4b, bad_get_user_clac) #else - _ASM_EXTABLE_UA(4b, bad_get_user_8) - _ASM_EXTABLE_UA(5b, bad_get_user_8) + _ASM_EXTABLE_UA(4b, bad_get_user_8_clac) + _ASM_EXTABLE_UA(5b, bad_get_user_8_clac) #endif
arch/x86/lib/putuser.S +16 -13
@@ -32,8 +32,6 @@ */ #define ENTER mov PER_CPU_VAR(current_task), %_ASM_BX -#define EXIT ASM_CLAC ; \ - ret .text ENTRY(__put_user_1)
@@ -43,7 +41,8 @@ ENTRY(__put_user_1) ASM_STAC 1: movb %al,(%_ASM_CX) xor %eax,%eax - EXIT + ASM_CLAC + ret ENDPROC(__put_user_1) EXPORT_SYMBOL(__put_user_1)
@@ -56,7 +55,8 @@ ENTRY(__put_user_2) ASM_STAC 2: movw %ax,(%_ASM_CX) xor %eax,%eax - EXIT + ASM_CLAC + ret ENDPROC(__put_user_2) EXPORT_SYMBOL(__put_user_2)
@@ -69,7 +69,8 @@ ENTRY(__put_user_4) ASM_STAC 3: movl %eax,(%_ASM_CX) xor %eax,%eax - EXIT + ASM_CLAC + ret ENDPROC(__put_user_4) EXPORT_SYMBOL(__put_user_4)
@@ -85,19 +86,21 @@ ENTRY(__put_user_8) 5: movl %edx,4(%_ASM_CX) #endif xor %eax,%eax - EXIT + ASM_CLAC + RET ENDPROC(__put_user_8) EXPORT_SYMBOL(__put_user_8) +bad_put_user_clac: + ASM_CLAC bad_put_user: movl $-EFAULT,%eax - EXIT -END(bad_put_user) + RET - _ASM_EXTABLE_UA(1b, bad_put_user) - _ASM_EXTABLE_UA(2b, bad_put_user) - _ASM_EXTABLE_UA(3b, bad_put_user) - _ASM_EXTABLE_UA(4b, bad_put_user) + _ASM_EXTABLE_UA(1b, bad_put_user_clac) + _ASM_EXTABLE_UA(2b, bad_put_user_clac) + _ASM_EXTABLE_UA(3b, bad_put_user_clac) + _ASM_EXTABLE_UA(4b, bad_put_user_clac) #ifdef CONFIG_X86_32 - _ASM_EXTABLE_UA(5b, bad_put_user) + _ASM_EXTABLE_UA(5b, bad_put_user_clac) #endif
Signature · 1 line
2.20.1
PATCH v2 18/22 objtool: Refactor jump table code
In reply to Josh PoimboeufNow that C jump tables are supported, call them "jump tables" instead of "switch tables". Also rename some other variables, add comments, and simplify the code flow a bit.
---
v2:
- comment tweak: "switch jump table" -> "jump table" [PeterZ]
--- tools/objtool/check.c | 82 +++++++++++++++++++++++-------------------- tools/objtool/elf.c | 2 +- tools/objtool/elf.h | 2 +- 3 files changed, 46 insertions(+), 40 deletions(-)
tools/objtool/check.c +44 -38
@@ -627,7 +627,7 @@ static int add_jump_destinations(struct objtool_file *file)* However this code can't completely replace the * read_symbols() code because this doesn't detect the * case where the parent function's only reference to a -* subfunction is through a switch table. +* subfunction is through a jump table. */ if (!strstr (insn->func->name,".cold." ) &&strstr (insn->jump_dest->func->name,".cold." )) {
@@ -899,20 +899,24 @@ static int add_special_section_alts(struct objtool_file *file)return ret; } -static int add_switch_table (struct objtool_file *file,struct instruction *insn, +static int add_jump_table (struct objtool_file *file,struct instruction *insn,struct rela *table,struct rela *next_table) {struct rela *rela = table; -struct instruction *alt_insn; +struct instruction *dest_insn;struct alternative *alt;struct symbol *pfunc = insn->func->pfunc;unsigned int prev_offset =0 ; -list_for_each_entry_from (rela, &table->rela_sec->rela_list, list) { +/* +* Each @rela is a switch table relocation which points to the target +* instruction. +*/ +list_for_each_entry_from (rela, &table->sec->rela_list, list) {if (rela == next_table)break ; -/* Make sure the switch table entries are consecutive: */ +/* Make sure the table entries are consecutive: */ if (prev_offset && rela->offset != prev_offset +8 )break ;
@@ -921,12 +925,12 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn, rela->addend == pfunc->offset)break ; - alt_insn =find_insn (file, rela->sym->sec, rela->addend); -if (!alt_insn) + dest_insn =find_insn (file, rela->sym->sec, rela->addend); +if (!dest_insn)break ; -/* Make sure the jmp dest is in the function or subfunction: */ -if (alt_insn->func->pfunc != pfunc) +/* Make sure the destination is in the same function: */ +if (dest_insn->func->pfunc != pfunc)break ; alt =malloc (sizeof (*alt));
@@ -935,7 +939,7 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn,return -1 ; } - alt->insn = alt_insn; + alt->insn = dest_insn;list_add_tail (&alt->list, &insn->alts); prev_offset = rela->offset; }
@@ -950,7 +954,7 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn, }/* -* find_switch_table() - Given a dynamic jump, find the switch jump table in +* find_jump_table() - Given a dynamic jump, find the switch jump table in * .rodata associated with it. * * There are 3 basic patterns:
@@ -992,13 +996,13 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn,* * NOTE: RETPOLINE made it harder still to decode dynamic jumps. */ -static struct rela *find_switch_table (struct objtool_file *file, +static struct rela *find_jump_table (struct objtool_file *file,struct symbol *func,struct instruction *insn) { -struct rela *text_rela, *rodata_rela; +struct rela *text_rela, *table_rela;struct instruction *orig_insn = insn; -struct section *rodata_sec; +struct section *table_sec;unsigned long table_offset;/*
@@ -1031,7 +1035,7 @@ static struct rela *find_switch_table(struct objtool_file *file,continue ; table_offset = text_rela->addend; - rodata_sec = text_rela->sym->sec; + table_sec = text_rela->sym->sec;if (text_rela->type ==R_X86_64_PC32 ) table_offset +=4 ;
@@ -1045,29 +1049,31 @@ static struct rela *find_switch_table(struct objtool_file *file,* need to be placed in the C_JUMP_TABLE_SECTION section. They * have symbols associated with them. */ -if (find_symbol_containing (rodata_sec, table_offset) && -strcmp (rodata_sec->name,C_JUMP_TABLE_SECTION )) +if (find_symbol_containing (table_sec, table_offset) && +strcmp (table_sec->name,C_JUMP_TABLE_SECTION ))continue ; - rodata_rela =find_rela_by_dest (rodata_sec, table_offset); -if (rodata_rela) { -/* -* Use of RIP-relative switch jumps is quite rare, and -* indicates a rare GCC quirk/bug which can leave dead -* code behind. -*/ -if (text_rela->type ==R_X86_64_PC32 ) - file->ignore_unreachables = true; +/* Each table entry has a rela associated with it. */ + table_rela =find_rela_by_dest (table_sec, table_offset); +if (!table_rela) +continue ; -return rodata_rela; - } +/* +* Use of RIP-relative switch jumps is quite rare, and +* indicates a rare GCC quirk/bug which can leave dead code +* behind. +*/ +if (text_rela->type ==R_X86_64_PC32 ) + file->ignore_unreachables = true; + +return table_rela; }return NULL ; } -static int add_func_switch_tables (struct objtool_file *file, +static int add_func_jump_tables (struct objtool_file *file,struct symbol *func) {struct instruction *insn, *last =NULL , *prev_jump =NULL ;
@@ -1080,7 +1086,7 @@ static int add_func_switch_tables(struct objtool_file *file,/* * Store back-pointers for unconditional forward jumps such -* that find_switch_table() can back-track using those and +* that find_jump_table() can back-track using those and * avoid some potentially confusing code. */ if (insn->type == INSN_JUMP_UNCONDITIONAL && insn->jump_dest &&
@@ -1095,17 +1101,17 @@ static int add_func_switch_tables(struct objtool_file *file,if (insn->type !=INSN_JUMP_DYNAMIC )continue ; - rela =find_switch_table (file, func, insn); + rela =find_jump_table (file, func, insn);if (!rela)continue ;/* -* We found a switch table, but we don't know yet how big it +* We found a jump table, but we don't know yet how big it * is. Don't add it until we reach the end of the function or -* the beginning of another switch table in the same function. +* the beginning of another jump table in the same function. */ if (prev_jump) { - ret =add_switch_table (file, prev_jump, prev_rela, rela); + ret =add_jump_table (file, prev_jump, prev_rela, rela);if (ret)return ret; }
@@ -1115,7 +1121,7 @@ static int add_func_switch_tables(struct objtool_file *file, }if (prev_jump) { - ret =add_switch_table (file, prev_jump, prev_rela,NULL ); + ret =add_jump_table (file, prev_jump, prev_rela,NULL );if (ret)return ret; }
@@ -1128,7 +1134,7 @@ static int add_func_switch_tables(struct objtool_file *file,* section which contains a list of addresses within the function to jump to. * This finds these jump tables and adds them to the insn->alts lists. */ -static int add_switch_table_alts (struct objtool_file *file) +static int add_jump_table_alts (struct objtool_file *file) {struct section *sec;struct symbol *func;
@@ -1142,7 +1148,7 @@ static int add_switch_table_alts(struct objtool_file *file)if (func->type !=STT_FUNC )continue ; - ret =add_func_switch_tables (file, func); + ret =add_func_jump_tables (file, func);if (ret)return ret; }
@@ -1339,7 +1345,7 @@ static int decode_sections(struct objtool_file *file)if (ret)return ret; - ret =add_switch_table_alts (file); + ret =add_jump_table_alts (file);if (ret)return ret;
tools/objtool/elf.c +1 -1
@@ -385,7 +385,7 @@ static int read_relas(struct elf *elf) rela->offset = rela->rela.r_offset; symndx =GELF_R_SYM (rela->rela.r_info); rela->sym =find_symbol_by_index (elf, symndx); - rela->rela_sec = sec; + rela->sec = sec;if (!rela->sym) {WARN ("can't find rela entry symbol %d for %s" , symndx, sec->name);
tools/objtool/elf.h +1 -1
@@ -57,7 +57,7 @@ struct rela {struct list_head list;struct hlist_node hash;GElf_Rela rela; -struct section *rela_sec; +struct section *sec;struct symbol *sym;unsigned int type;unsigned long offset;
Signature · 1 line
2.20.1
PATCH v2 20/22 objtool: Fix seg fault on bad switch table entry
In reply to Josh PoimboeufIn one rare case, Clang generated the following code:
5ca: 83 e0 21 and $0x21,%eax
5cd: b9 04 00 00 00 mov $0x4,%ecx
5d2: ff 24 c5 00 00 00 00 jmpq *0x0(,%rax,8)
5d5: R_X86_64_32S .rodata+0x38which uses the corresponding jump table relocations:
000000000038 000200000001 R_X86_64_64 0000000000000000 .text + 834 000000000040 000200000001 R_X86_64_64 0000000000000000 .text + 5d9 000000000048 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000050 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000058 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000060 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000068 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000070 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000078 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000080 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000088 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000090 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000098 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000a0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000a8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000b0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000b8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000c0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000c8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000d0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000d8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000e0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000e8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000f0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000f8 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000100 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000108 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000110 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000118 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000120 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000128 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000130 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000138 000200000001 R_X86_64_64 0000000000000000 .text + 82f 000000000140 000200000001 R_X86_64_64 0000000000000000 .text + 828
Since %eax was masked with 0x21, only the first two and the last two entries are possible.
Objtool doesn't actually emulate all the code, so it isn't smart enough to know that all the middle entries aren't reachable. They point to the NOP padding area after the end of the function, so objtool seg faulted when it tried to dereference a NULL insn->func.
After this fix, objtool still gives an "unreachable" error because it stops reading the jump table when it encounters the bad addresses:
/home/jpoimboe/objtool-tests/adm1275.o: warning: objtool: adm1275_probe()+0x828: unreachable instruction
While the above code is technically correct, it's very wasteful of memory -- it uses 34 jump table entries when only 4 are needed. It's also not possible for objtool to validate this type of switch table because the unused entries point outside the function and objtool has no way of determining if that's intentional. Hopefully the Clang folks can fix it.
--- tools/objtool/check.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/objtool/check.c +1 -1
@@ -932,7 +932,7 @@ static int add_jump_table(struct objtool_file *file, struct instruction *insn,break ;/* Make sure the destination is in the same function: */ -if (dest_insn->func->pfunc != pfunc) +if (!dest_insn->func || dest_insn->func->pfunc != pfunc)break ; alt =malloc (sizeof (*alt));
Signature · 1 line
2.20.1
PATCH v2 21/22 objtool: convert insn type to enum
In reply to Josh PoimboeufThis makes it easier to add new instruction types. Also it's hopefully more robust since the compiler should warn about out-of-range enums.
--- tools/objtool/arch.h | 35 +++++++++++++++++---------------- tools/objtool/arch/x86/decode.c | 2 +- tools/objtool/check.c | 7 ------- tools/objtool/check.h | 2 +- 4 files changed, 20 insertions(+), 26 deletions(-)
tools/objtool/arch.h +18 -17
@@ -11,22 +11,23 @@#include "elf.h" #include "cfi.h" -#define INSN_JUMP_CONDITIONAL 1 -#define INSN_JUMP_UNCONDITIONAL 2 -#define INSN_JUMP_DYNAMIC 3 -#define INSN_CALL 4 -#define INSN_CALL_DYNAMIC 5 -#define INSN_RETURN 6 -#define INSN_CONTEXT_SWITCH 7 -#define INSN_STACK 8 -#define INSN_BUG 9 -#define INSN_NOP 10 -#define INSN_STAC 11 -#define INSN_CLAC 12 -#define INSN_STD 13 -#define INSN_CLD 14 -#define INSN_OTHER 15 -#define INSN_LAST INSN_OTHER +enum insn_type { +INSN_JUMP_CONDITIONAL , +INSN_JUMP_UNCONDITIONAL , +INSN_JUMP_DYNAMIC , +INSN_CALL , +INSN_CALL_DYNAMIC , +INSN_RETURN , +INSN_CONTEXT_SWITCH , +INSN_STACK , +INSN_BUG , +INSN_NOP , +INSN_STAC , +INSN_CLAC , +INSN_STD , +INSN_CLD , +INSN_OTHER , +};enum op_dest_type {OP_DEST_REG ,
@@ -68,7 +69,7 @@ void arch_initial_func_cfi_state(struct cfi_state *state);int arch_decode_instruction (struct elf *elf,struct section *sec,unsigned long offset,unsigned int maxlen, -unsigned int *len,unsigned char *type, +unsigned int *len,enum insn_type *type,unsigned long *immediate,struct stack_op *op);bool arch_callee_saved_reg (unsigned char reg);
tools/objtool/arch/x86/decode.c +1 -1
@@ -68,7 +68,7 @@ bool arch_callee_saved_reg(unsigned char reg)int arch_decode_instruction (struct elf *elf,struct section *sec,unsigned long offset,unsigned int maxlen, -unsigned int *len,unsigned char *type, +unsigned int *len,enum insn_type *type,unsigned long *immediate,struct stack_op *op) {struct insn insn;
tools/objtool/check.c +0 -7
@@ -267,13 +267,6 @@ static int decode_instructions(struct objtool_file *file)if (ret) goto err; -if (!insn->type || insn->type >INSN_LAST ) { -WARN_FUNC ("invalid instruction type %d" , - insn->sec, insn->offset, insn->type); - ret =-1 ; - goto err; - } -hash_add (file->insn_hash, &insn->hash, insn->offset);list_add_tail (&insn->list, &file->insn_list); }
tools/objtool/check.h +1 -1
@@ -31,7 +31,7 @@ struct instruction {struct section *sec;unsigned long offset;unsigned int len; -unsigned char type; +enum insn_type type;unsigned long immediate;bool alt_group, visited, dead_end, ignore, hint, save, restore, ignore_alts;bool retpoline_safe;
Signature · 1 line
2.20.1
PATCH v2 19/22 objtool: Support repeated uses of the same C jump table
In reply to Josh PoimboeufFrom: Jann Horn <[email protected]>
This fixes objtool for both a GCC issue and a Clang issue:
1) GCC issue:
kernel/bpf/core.o: warning: objtool: ___bpf_prog_run()+0x8d5: sibling call from callable instruction with modified stack frame
With CONFIG_RETPOLINE=n, GCC is doing the following optimization in ___bpf_prog_run().
Before:
select_insn:
jmp *jumptable(,%rax,8)
...
ALU64_ADD_X:
...
jmp select_insn
ALU_ADD_X:
...
jmp select_insnAfter:
select_insn:
jmp *jumptable(, %rax, 8)
...
ALU64_ADD_X:
...
jmp *jumptable(, %rax, 8)
ALU_ADD_X:
...
jmp *jumptable(, %rax, 8)This confuses objtool. It has never seen multiple indirect jump sites which use the same jump table.
For GCC switch tables, the only way of detecting the size of a table is by continuing to scan for more tables. The size of the previous table can only be determined after another switch table is found, or when the scan reaches the end of the function.
That logic was reused for C jump tables, and was based on the assumption that each jump table only has a single jump site. The above optimization breaks that assumption.
2) Clang issue:
drivers/usb/misc/sisusbvga/sisusb.o: warning: objtool: sisusb_write_mem_bulk()+0x588: can't find switch jump table
With clang 9, code can be generated where a function contains two indirect jump instructions which use the same switch table.
The fix is the same for both issues: split the jump table parsing into two passes.
In the first pass, locate the heads of all switch tables for the function and mark their locations.
In the second pass, parse the switch tables and add them.
--- tools/objtool/check.c | 53 +++++++++++++++++++++++-------------------- tools/objtool/check.h | 1 + tools/objtool/elf.h | 1 + 3 files changed, 30 insertions(+), 25 deletions(-)
tools/objtool/check.c +28 -25
@@ -900,7 +900,7 @@ static int add_special_section_alts(struct objtool_file *file) }static int add_jump_table (struct objtool_file *file,struct instruction *insn, -struct rela *table,struct rela *next_table) +struct rela *table) {struct rela *rela = table;struct instruction *dest_insn;
@@ -913,7 +913,9 @@ static int add_jump_table(struct objtool_file *file, struct instruction *insn,* instruction. */ list_for_each_entry_from (rela, &table->sec->rela_list, list) { -if (rela == next_table) + +/* Check for the end of the table: */ +if (rela != table && rela->jump_table_start)break ;/* Make sure the table entries are consecutive: */
@@ -1072,13 +1074,15 @@ static struct rela *find_jump_table(struct objtool_file *file,return NULL ; } - -static int add_func_jump_tables (struct objtool_file *file, -struct symbol *func) +/* +* First pass: Mark the head of each jump table so that in the next pass, +* we know when a given jump table ends and the next one starts. +*/ +static void mark_func_jump_tables (struct objtool_file *file, +struct symbol *func) { -struct instruction *insn, *last =NULL , *prev_jump =NULL ; -struct rela *rela, *prev_rela =NULL ; -int ret; +struct instruction *insn, *last =NULL ; +struct rela *rela;func_for_each_insn_all (file, func, insn) {if (!last)
@@ -1102,26 +1106,24 @@ static int add_func_jump_tables(struct objtool_file *file,continue ; rela =find_jump_table (file, func, insn); -if (!rela) -continue ; - -/* -* We found a jump table, but we don't know yet how big it -* is. Don't add it until we reach the end of the function or -* the beginning of another jump table in the same function. -*/ -if (prev_jump) { - ret =add_jump_table (file, prev_jump, prev_rela, rela); -if (ret) -return ret; +if (rela) { + rela->jump_table_start = true; + insn->jump_table = rela; } - - prev_jump = insn; - prev_rela = rela; } +} + +static int add_func_jump_tables (struct objtool_file *file, +struct symbol *func) +{ +struct instruction *insn; +int ret; + +func_for_each_insn_all (file, func, insn) { +if (!insn->jump_table) +continue ; -if (prev_jump) { - ret =add_jump_table (file, prev_jump, prev_rela,NULL ); + ret =add_jump_table (file, insn, insn->jump_table);if (ret)return ret; }
@@ -1148,6 +1150,7 @@ static int add_jump_table_alts(struct objtool_file *file)if (func->type !=STT_FUNC )continue ; +mark_func_jump_tables (file, func); ret =add_func_jump_tables (file, func);if (ret)return ret;
tools/objtool/check.h +1 -0
@@ -38,6 +38,7 @@ struct instruction {struct symbol *call_dest;struct instruction *jump_dest;struct instruction *first_jump_src; +struct rela *jump_table;struct list_head alts;struct symbol *func;struct stack_op stack_op;
tools/objtool/elf.h +1 -0
@@ -62,6 +62,7 @@ struct rela {unsigned int type;unsigned long offset;int addend; +bool jump_table_start; };struct elf {
Signature · 1 line
2.20.1
PATCH v2 22/22 objtool: Support conditional retpolines
In reply to Josh PoimboeufA Clang-built kernel is showing the following warning:
arch/x86/kernel/platform-quirks.o: warning: objtool: x86_early_init_platform_quirks()+0x84: unreachable instruction
That corresponds to this code:
7e: 0f 85 00 00 00 00 jne 84 <x86_early_init_platform_quirks+0x84>
80: R_X86_64_PC32 __x86_indirect_thunk_r11-0x4
84: c3 retq
This is a conditional retpoline sibling call, which is now possible thanks to retpolines. Objtool hasn't seen that before. It's incorrectly interpreting the conditional jump as an unconditional dynamic jump.
--- tools/objtool/arch.h | 1 + tools/objtool/check.c | 12 ++++++++++-- 2 files changed, 11 insertions(+), 2 deletions(-)
tools/objtool/arch.h +1 -0
@@ -15,6 +15,7 @@ enum insn_type {INSN_JUMP_CONDITIONAL ,INSN_JUMP_UNCONDITIONAL ,INSN_JUMP_DYNAMIC , +INSN_JUMP_DYNAMIC_CONDITIONAL ,INSN_CALL ,INSN_CALL_DYNAMIC , INSN_RETURN,
tools/objtool/check.c +10 -2
@@ -575,7 +575,11 @@ static int add_jump_destinations(struct objtool_file *file)* Retpoline jumps are really dynamic jumps in * disguise, so convert them accordingly. */ - insn->type =INSN_JUMP_DYNAMIC ; +if (insn->type ==INSN_JUMP_UNCONDITIONAL ) + insn->type =INSN_JUMP_DYNAMIC ; +else + insn->type =INSN_JUMP_DYNAMIC_CONDITIONAL ; + insn->retpoline_safe = true;continue ; }else {
@@ -2114,13 +2118,17 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,break ;case INSN_JUMP_DYNAMIC : +case INSN_JUMP_DYNAMIC_CONDITIONAL :if (func &&is_sibling_call (insn)) { ret =validate_sibling_call (insn, &state);if (ret)return ret; } -return 0 ; +if (insn->type ==INSN_JUMP_DYNAMIC ) +return 0 ; + +break ;case INSN_CONTEXT_SWITCH :if (func && (!next_insn || !next_insn->hint)) {
Signature · 1 line
2.20.1
PATCH v2 15/22 objtool: Change dead_end_function() to return boolean
In reply to Josh Poimboeufdead_end_function() can no longer return an error. Simplify its interface by making it return boolean.
--- tools/objtool/check.c | 36 ++++++++++++++---------------------- 1 file changed, 14 insertions(+), 22 deletions(-)
tools/objtool/check.c +14 -22
@@ -105,14 +105,9 @@ static struct instruction *next_insn_same_func(struct objtool_file *file,* * For local functions, we have to detect them manually by simply looking for * the lack of a return instruction. -* -* Returns: -* -1: error -* 0: no dead end -* 1: dead end */ -static int __dead_end_function (struct objtool_file *file,struct symbol *func, -int recursion) +static bool __dead_end_function (struct objtool_file *file,struct symbol *func, +int recursion) {int i;struct instruction *insn;
@@ -139,29 +134,29 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func, };if (func->bind ==STB_WEAK ) -return 0 ; +return false;if (func->bind ==STB_GLOBAL )for (i =0 ; i <ARRAY_SIZE (global_noreturns); i++)if (!strcmp (func->name, global_noreturns[i])) -return 1 ; +return true;if (!func->len) -return 0 ; +return false; insn =find_insn (file, func->sec, func->offset);if (!insn->func) -return 0 ; +return false;func_for_each_insn_all (file, func, insn) { empty = false;if (insn->type ==INSN_RETURN ) -return 0 ; +return false; }if (empty) -return 0 ; +return false;/* * A function can have a sibling call instead of a return. In that
@@ -174,7 +169,7 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func,if (!dest)/* sibling call to another file */ -return 0 ; +return false;if (dest->func && dest->func->pfunc != insn->func->pfunc) {
@@ -186,7 +181,7 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func,* This is a very rare case. It means * they aren't dead ends. */ -return 0 ; +return false; } return __dead_end_function(file, dest->func,
@@ -196,13 +191,13 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func,if (insn->type ==INSN_JUMP_DYNAMIC &&list_empty (&insn->alts))/* sibling call */ -return 0 ; +return false; } -return 1 ; +return true; } -static int dead_end_function (struct objtool_file *file,struct symbol *func) +static bool dead_end_function (struct objtool_file *file,struct symbol *func) {return __dead_end_function (file, func,0 ); }
@@ -2080,11 +2075,8 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,if (is_fentry_call (insn))break ; - ret =dead_end_function (file, insn->call_dest); -if (ret ==1 ) +if (dead_end_function (file, insn->call_dest))return 0 ; -if (ret ==-1 ) -return 1 ; }if (!no_fp && func && !has_valid_stack_frame (&state)) {
Signature · 1 line
2.20.1
PATCH v2 14/22 objtool: Warn on zero-length functions
In reply to Josh PoimboeufAll callable functions should have an ELF size.
--- tools/objtool/check.c | 6 ++++++ 1 file changed, 6 insertions(+)
tools/objtool/check.c +6 -0
@@ -2357,6 +2357,12 @@ static int validate_functions(struct objtool_file *file)if (func->type !=STT_FUNC )continue ; +if (!func->len) { +WARN ("%s() is missing an ELF size annotation" , + func->name); + warnings++; + } +if (func->pfunc != func || func->alias != func)continue ;
Signature · 1 line
2.20.1
PATCH v2 12/22 objtool: Track original function across branches
In reply to Josh PoimboeufIf 'insn->func' is NULL, objtool skips some important checks, including sibling call validation. So if some .fixup code does an invalid sibling call, objtool ignores it.
Treat all code branches (including alts) as part of the original function by keeping track of the original func value from validate_functions().
This improves the usefulness of some clang function fallthrough warnings, and exposes some additional kernel bugs in the process.
--- tools/objtool/check.c | 28 ++++++++++++---------------- 1 file changed, 12 insertions(+), 16 deletions(-)
tools/objtool/check.c +12 -16
@@ -1934,13 +1934,12 @@ static int validate_sibling_call(struct instruction *insn, struct insn_state *st* each instruction and validate all the rules described in * tools/objtool/Documentation/stack-validation.txt. */ -static int validate_branch (struct objtool_file *file,struct instruction *first, -struct insn_state state) +static int validate_branch (struct objtool_file *file,struct symbol *func, +struct instruction *first,struct insn_state state) {struct alternative *alt;struct instruction *insn, *next_insn;struct section *sec; -struct symbol *func =NULL ;int ret; insn = first;
@@ -1961,9 +1960,6 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,return 1 ; } -if (insn->func) - func = insn->func->pfunc; -if (func && insn->ignore) {WARN_FUNC ("BUG: why am I validating an ignored function?" , sec, insn->offset);
@@ -1985,7 +1981,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first, i = insn; save_insn =NULL ; -func_for_each_insn_continue_reverse (file, insn->func, i) { +func_for_each_insn_continue_reverse (file, func, i) {if (i->save) { save_insn = i;break ;
@@ -2031,7 +2027,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,if (alt->skip_orig) skip_orig = true; - ret =validate_branch (file, alt->insn, state); + ret =validate_branch (file, func, alt->insn, state);if (ret) {if (backtrace)BT_FUNC ("(alt)" , insn);
@@ -2069,7 +2065,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,if (state.bp_scratch) {WARN ("%s uses BP as a scratch register" , - insn->func->name); + func->name);return 1 ; }
@@ -2109,8 +2105,8 @@ static int validate_branch(struct objtool_file *file, struct instruction *first, }else if (insn->jump_dest && (!func || !insn->jump_dest->func || insn->jump_dest->func->pfunc == func)) { - ret =validate_branch (file, insn->jump_dest, - state); + ret =validate_branch (file, func, + insn->jump_dest, state);if (ret) {if (backtrace)BT_FUNC ("(branch)" , insn);
@@ -2176,7 +2172,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,break ;case INSN_CLAC : -if (!state.uaccess && insn->func) { +if (!state.uaccess && func) {WARN_FUNC ("redundant UACCESS disable" , sec, insn->offset);return 1 ; }
@@ -2197,7 +2193,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,break ;case INSN_CLD : -if (!state.df && insn->func) +if (!state.df && func)WARN_FUNC ("redundant CLD" , sec, insn->offset); state.df = false;
@@ -2236,7 +2232,7 @@ static int validate_unwind_hints(struct objtool_file *file)for_each_insn (file, insn) {if (insn->hint && !insn->visited) { - ret =validate_branch (file, insn, state); + ret =validate_branch (file, insn->func, insn, state);if (ret && backtrace)BT_FUNC ("<=== (hint)" , insn); warnings += ret;
@@ -2363,12 +2359,12 @@ static int validate_functions(struct objtool_file *file)continue ; insn =find_insn (file, sec, func->offset); -if (!insn || insn->ignore) +if (!insn || insn->ignore || insn->visited)continue ; state.uaccess = func->alias->uaccess_safe; - ret =validate_branch (file, insn, state); + ret =validate_branch (file, func, insn, state);if (ret && backtrace)BT_FUNC ("<=== (func)" , insn); warnings += ret;
Signature · 1 line
2.20.1
PATCH v2 17/22 objtool: Refactor sibling call detection logic
In reply to Josh PoimboeufSimplify the sibling call detection logic a bit.
--- tools/objtool/check.c | 65 ++++++++++++++++++++++--------------------- 1 file changed, 33 insertions(+), 32 deletions(-)
tools/objtool/check.c +33 -32
@@ -97,6 +97,20 @@ static struct instruction *next_insn_same_func(struct objtool_file *file,for (insn =next_insn_same_sec (file, insn); insn; \ insn =next_insn_same_sec (file, insn)) +static bool is_sibling_call(struct instruction *insn) +{ +/* An indirect jump is either a sibling call or a jump to a table. */ +if (insn->type ==INSN_JUMP_DYNAMIC ) +return list_empty (&insn->alts); + +if (insn->type !=INSN_JUMP_CONDITIONAL && + insn->type !=INSN_JUMP_UNCONDITIONAL ) +return false; + +/* add_jump_destinations() sets insn->call_dest for sibling calls. */ +return !!insn->call_dest; +} +/* * This checks to see if the given function is a "noreturn" function. *
@@ -167,34 +181,25 @@ static bool __dead_end_function(struct objtool_file *file, struct symbol *func,* of the sibling call returns. */ func_for_each_insn_all (file, func, insn) { -if (insn->type ==INSN_JUMP_UNCONDITIONAL ) { +if (is_sibling_call (insn)) {struct instruction *dest = insn->jump_dest;if (!dest)/* sibling call to another file */ return false; -if (dest->func && dest->func->pfunc != insn->func->pfunc) { - -/* local sibling call */ -if (recursion ==5 ) { -/* -* Infinite recursion: two functions -* have sibling calls to each other. -* This is a very rare case. It means -* they aren't dead ends. -*/ -return false; - } - -return __dead_end_function (file, dest->func, - recursion +1 ); +/* local sibling call */ +if (recursion ==5 ) { +/* +* Infinite recursion: two functions have +* sibling calls to each other. This is a very +* rare case. It means they aren't dead ends. +*/ +return false; } - } -if (insn->type ==INSN_JUMP_DYNAMIC &&list_empty (&insn->alts)) -/* sibling call */ -return false; +return __dead_end_function (file, dest->func, recursion+1 ); + } }return true;
@@ -581,9 +586,8 @@ static int add_jump_destinations(struct objtool_file *file) insn->retpoline_safe = true;continue ; }else { -/* sibling call */ +/* external sibling call */ insn->call_dest = rela->sym; - insn->jump_dest =NULL ;continue ; }
@@ -633,9 +637,8 @@ static int add_jump_destinations(struct objtool_file *file) }else if (insn->jump_dest->func->pfunc != insn->func->pfunc && insn->jump_dest->offset == insn->jump_dest->func->offset) { -/* sibling class */ +/* internal sibling call */ insn->call_dest = insn->jump_dest->func; - insn->jump_dest =NULL ; } } }
@@ -1889,7 +1892,7 @@ static inline bool func_uaccess_safe(struct symbol *func)return false; } -static inline const char *insn_dest_name (struct instruction *insn) +static inline const char *call_dest_name (struct instruction *insn) {if (insn->call_dest)return insn->call_dest->name;
@@ -1901,13 +1904,13 @@ static int validate_call(struct instruction *insn, struct insn_state *state) {if (state->uaccess && !func_uaccess_safe (insn->call_dest)) {WARN_FUNC ("call to %s() with UACCESS enabled" , - insn->sec, insn->offset,insn_dest_name (insn)); + insn->sec, insn->offset,call_dest_name (insn));return 1 ; }if (state->df) {WARN_FUNC ("call to %s() with DF set" , - insn->sec, insn->offset,insn_dest_name (insn)); + insn->sec, insn->offset,call_dest_name (insn));return 1 ; }
@@ -2088,14 +2091,12 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,case INSN_JUMP_CONDITIONAL :case INSN_JUMP_UNCONDITIONAL : -if (func && !insn->jump_dest) { +if (func &&is_sibling_call (insn)) { ret =validate_sibling_call (insn, &state);if (ret)return ret; - }else if (insn->jump_dest && - (!func || !insn->jump_dest->func || - insn->jump_dest->func->pfunc == func)) { + }else if (insn->jump_dest) { ret =validate_branch (file, func, insn->jump_dest, state);if (ret) {
@@ -2111,7 +2112,7 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,break ;case INSN_JUMP_DYNAMIC : -if (func &&list_empty (&insn->alts)) { +if (func &&is_sibling_call (insn)) { ret =validate_sibling_call (insn, &state);if (ret)return ret;
Signature · 1 line
2.20.1
PATCH v2 10/22 bpf: Disable GCC -fgcse optimization for ___bpf_prog_run()
In reply to Josh PoimboeufOn x86-64, with CONFIG_RETPOLINE=n, GCC's "global common subexpression elimination" optimization results in ___bpf_prog_run()'s jumptable code changing from this:
select_insn: jmp *jumptable(, %rax, 8) ... ALU64_ADD_X: ... jmp *jumptable(, %rax, 8) ALU_ADD_X: ... jmp *jumptable(, %rax, 8)
to this:
select_insn: mov jumptable, %r12 jmp *(%r12, %rax, 8) ... ALU64_ADD_X: ... jmp *(%r12, %rax, 8) ALU_ADD_X: ... jmp *(%r12, %rax, 8)
The jumptable address is placed in a register once, at the beginning of the function. The function execution can then go through multiple indirect jumps which rely on that same register value. This has a few issues:
1) Objtool isn't smart enough to be able to track such a register value across multiple recursive indirect jumps through the jump table.
2) With CONFIG_RETPOLINE enabled, this optimization actually results in a small slowdown. I measured a ~4.7% slowdown in the test_bpf "tcpdump port 22" selftest.
This slowdown is actually predicted by the GCC manual:
Note: When compiling a program using computed gotos, a GCC extension, you may get better run-time performance if you disable the global common subexpression elimination pass by adding -fno-gcse to the command line.
So just disable the optimization for this function.
---
Cc: Alexei Starovoitov <[email protected]>
Cc: Daniel Borkmann <[email protected]>
--- include/linux/compiler-gcc.h | 2 ++ include/linux/compiler_types.h | 4 ++++ kernel/bpf/core.c | 2 +- 3 files changed, 7 insertions(+), 1 deletion(-)
include/linux/compiler-gcc.h +2 -0
@@ -170,3 +170,5 @@#else #define __diag_GCC_8 (s)#endif + +#define __no_fgcse __attribute__((optimize("-fno-gcse")))
include/linux/compiler_types.h +4 -0
@@ -189,6 +189,10 @@ struct ftrace_likely_data { #define asm_volatile_goto(x...) asm goto(x)#endif +#ifndef __no_fgcse +# define __no_fgcse +#endif +/* Are two types/vars the same type (ignoring qualifiers)? */ #define __same_type (a, b) __builtin_types_compatible_p(typeof(a), typeof(b))
kernel/bpf/core.c +1 -1
@@ -1295,7 +1295,7 @@ bool bpf_opcode_in_insntable(u8 code)* * Decode and execute eBPF instructions. */ -static u64 ___bpf_prog_run (u64 *regs,const struct bpf_insn *insn,u64 *stack) +static u64 __no_fgcse ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack) {#define BPF_INSN_2_LBL (x, y) [BPF_##x | BPF_##y] = &&x##_##y#define BPF_INSN_3_LBL (x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
Signature · 1 line
2.20.1
PATCH v2 16/22 objtool: Do frame pointer check before dead end check
In reply to Josh PoimboeufEven calls to __noreturn functions need the frame pointer setup first. Such functions often dump the stack.
--- tools/objtool/check.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-)
tools/objtool/check.c +9 -9
@@ -133,6 +133,9 @@ static bool __dead_end_function(struct objtool_file *file, struct symbol *func, "rewind_stack_do_exit", }; +if (!func) +return false; +if (func->bind ==STB_WEAK )return false;
@@ -2071,19 +2074,16 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,if (ret)return ret; -if (insn->type ==INSN_CALL ) { -if (is_fentry_call (insn)) -break ; - -if (dead_end_function (file, insn->call_dest)) -return 0 ; - } - -if (!no_fp && func && !has_valid_stack_frame (&state)) { +if (!no_fp && func && !is_fentry_call (insn) && + !has_valid_stack_frame (&state)) {WARN_FUNC ("call without frame pointer save/setup" , sec, insn->offset);return 1 ; } + +if (dead_end_function (file, insn->call_dest)) +return 0 ; +break ;case INSN_JUMP_CONDITIONAL :
Signature · 1 line
2.20.1
PATCH v2 13/22 objtool: Refactor function alias logic
In reply to Josh Poimboeuf- Add an alias check in validate_functions(). With this change, aliases no longer need uaccess_safe set.
- Add an alias check in decode_instructions(). With this change, the "if (!insn->func)" check is no longer needed.
- Don't create aliases for zero-length functions, as it can have unexpected results. The next patch will spit out a warning for zero-length functions anyway.
--- tools/objtool/check.c | 16 +++++++++------- tools/objtool/elf.c | 2 +- 2 files changed, 10 insertions(+), 8 deletions(-)
tools/objtool/check.c +9 -7
@@ -276,7 +276,7 @@ static int decode_instructions(struct objtool_file *file) }list_for_each_entry (func, &sec->symbol_list, list) { -if (func->type !=STT_FUNC ) +if (func->type !=STT_FUNC || func->alias != func)continue ;if (!find_insn (file, sec, func->offset)) {
@@ -286,8 +286,7 @@ static int decode_instructions(struct objtool_file *file) }func_for_each_insn (file, func, insn) -if (!insn->func) - insn->func = func; + insn->func = func; } }
@@ -508,7 +507,7 @@ static void add_uaccess_safe(struct objtool_file *file)if (!func)continue ; - func->alias->uaccess_safe = true; + func->uaccess_safe = true; } }
@@ -1887,7 +1886,7 @@ static bool insn_state_match(struct instruction *insn, struct insn_state *state)static inline bool func_uaccess_safe (struct symbol *func) {if (func) -return func->alias->uaccess_safe; +return func->uaccess_safe;return false; }
@@ -2355,14 +2354,17 @@ static int validate_functions(struct objtool_file *file)for_each_sec (file, sec) {list_for_each_entry (func, &sec->symbol_list, list) { -if (func->type !=STT_FUNC || func->pfunc != func) +if (func->type !=STT_FUNC ) +continue ; + +if (func->pfunc != func || func->alias != func)continue ; insn =find_insn (file, sec, func->offset);if (!insn || insn->ignore || insn->visited)continue ; - state.uaccess = func->alias->uaccess_safe; + state.uaccess = func->uaccess_safe; ret =validate_branch (file, func, insn, state);if (ret && backtrace)
tools/objtool/elf.c +1 -1
@@ -278,7 +278,7 @@ static int read_symbols(struct elf *elf) }if (sym->offset == s->offset) { -if (sym->len == s->len && alias == sym) +if (sym->len && sym->len == s->len && alias == sym) alias = s;if (sym->len >= s->len) {
Signature · 1 line
2.20.1
PATCH v2 8/22 x86/uaccess: Don't leak AC flag into fentry from mcsafe_handle_tail()
In reply to Josh PoimboeufAfter adding mcsafe_handle_tail() to the objtool uaccess safe list, objtool reports:
arch/x86/lib/usercopy_64.o: warning: objtool: mcsafe_handle_tail()+0x0: call to __fentry__() with UACCESS enabled
With SMAP, this function is called with AC=1, so it needs to be careful about which functions it calls. Disable the ftrace entry hook, which can potentially pull in a lot of extra code.
--- arch/x86/lib/usercopy_64.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/lib/usercopy_64.c +1 -1
@@ -60,7 +60,7 @@ EXPORT_SYMBOL(clear_user);* but reuse __memcpy_mcsafe in case a new read error is encountered. * clac() is handled in _copy_to_iter_mcsafe(). */ -__visible unsigned long +__visible notrace unsigned longmcsafe_handle_tail (char *to,char *from,unsigned len) {for (; len; --len, to++, from++) {
Signature · 1 line
2.20.1
PATCH v2 11/22 objtool: Add mcsafe_handle_tail() to the uaccess safe list
In reply to Josh PoimboeufAfter an objtool improvement, it's reporting that __memcpy_mcsafe() is calling mcsafe_handle_tail() with AC=1:
arch/x86/lib/memcpy_64.o: warning: objtool: .fixup+0x13: call to mcsafe_handle_tail() with UACCESS enabled
arch/x86/lib/memcpy_64.o: warning: objtool: __memcpy_mcsafe()+0x34: (alt) arch/x86/lib/memcpy_64.o: warning: objtool: __memcpy_mcsafe()+0xb: (branch) arch/x86/lib/memcpy_64.o: warning: objtool: __memcpy_mcsafe()+0x0: <=== (func)
mcsafe_handle_tail() is basically an extension of __memcpy_mcsafe(), so AC=1 is supposed to be set. Add mcsafe_handle_tail() to the uaccess safe list.
--- tools/objtool/check.c | 1 + 1 file changed, 1 insertion(+)
tools/objtool/check.c +1 -0
@@ -490,6 +490,7 @@ static const char *uaccess_safe_builtin[] = {/* misc */ "csum_partial_copy_generic" ,"__memcpy_mcsafe" , +"mcsafe_handle_tail" ,"ftrace_likely_update" ,/* CONFIG_TRACE_BRANCH_PROFILING */ NULL };
Signature · 1 line
2.20.1
PATCH v2 7/22 x86/uaccess: Remove ELF function annotation from copy_user_handle_tail()
In reply to Josh PoimboeufAfter an objtool improvement, it's complaining about the CLAC in copy_user_handle_tail():
arch/x86/lib/copy_user_64.o: warning: objtool: .altinstr_replacement+0x12: redundant UACCESS disable
arch/x86/lib/copy_user_64.o: warning: objtool: copy_user_handle_tail()+0x6: (alt) arch/x86/lib/copy_user_64.o: warning: objtool: copy_user_handle_tail()+0x2: (alt) arch/x86/lib/copy_user_64.o: warning: objtool: copy_user_handle_tail()+0x0: <=== (func)
copy_user_handle_tail() is incorrectly marked as a callable function, so objtool is rightfully concerned about the CLAC with no corresponding STAC.
Remove the ELF function annotation. The copy_user_handle_tail() code path is already verified by objtool because it's jumped to by other callable asm code (which does the corresponding STAC).
--- arch/x86/lib/copy_user_64.S | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/lib/copy_user_64.S +1 -1
@@ -239,7 +239,7 @@ copy_user_handle_tail: ret _ASM_EXTABLE_UA(1b, 2b) -ENDPROC(copy_user_handle_tail) +END(copy_user_handle_tail) /* * copy_user_nocache - Uncached memory copy with exception handling
Signature · 1 line
2.20.1
PATCH v2 5/22 x86/entry: Fix thunk function ELF sizes
In reply to Josh PoimboeufFix the following warnings:
arch/x86/entry/thunk_64.o: warning: objtool: trace_hardirqs_on_thunk() is missing an ELF size annotation arch/x86/entry/thunk_64.o: warning: objtool: trace_hardirqs_off_thunk() is missing an ELF size annotation arch/x86/entry/thunk_64.o: warning: objtool: lockdep_sys_exit_thunk() is missing an ELF size annotation
--- arch/x86/entry/thunk_64.S | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)
arch/x86/entry/thunk_64.S +2 -3
@@ -12,9 +12,7 @@ /* rdi: arg1 ... normal C conventions. rax is saved/restored. */ .macro THUNK name, func, put_ret_addr_in_rdi=0 - .globl \name - .type \name, @function -\name: + ENTRY(\name) pushq %rbp movq %rsp, %rbp
@@ -35,6 +33,7 @@ call \func jmp .L_restore + ENDPROC(\name) _ASM_NOKPROBE(\name) .endm
Signature · 1 line
2.20.1
Re: [PATCH v2 03/22] x86/kvm: Replace vmx_vmenter()'s call to kvm_spurious_fault() with UD2
In reply to Josh PoimboeufQuoted 31 lines · Josh Poimboeuf
Objtool reports the following:
arch/x86/kvm/vmx/vmenter.o: warning: objtool: vmx_vmenter()+0x14: call without frame pointer save/setupBut frame pointers are necessarily broken anyway, because
__vmx_vcpu_run()clobbers RBP with the guest's value before callingvmx_vmenter(). So calling without a frame pointer doesn't make things any worse.Make objtool happy by changing the call to a UD2.
Suggested-by Paolo Bonzini <[email protected]>Signed-off-by Josh Poimboeuf <[email protected]>---
v2: ud2 instead ofkvm_spurious_fault()[Paolo]Cc Paolo Bonzini <[email protected]>Cc Radim Krčmář <[email protected]>--- arch/x86/kvm/vmx/vmenter.S | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)arch/x86/kvm/vmx/vmenter.S +0
-0@@ -54,9 +54,9 @@ ENTRY(vmx_vmenter) ret
arch/x86/kvm/vmx/vmenter.S ENTRY(vmx_vmenter) · line 62
Quoted 10 lines
arch/x86/kvm/vmx/vmenter.S +3
-33: cmpb $0, kvm_rebooting - jne 4f - call kvm_spurious_fault -4: ret + je 4f + ret +4: ud2 .pushsection .fixup, "ax" 5: jmp 3b
Re: [PATCH v2 04/22] x86/kvm: Don't call kvm_spurious_fault() from .fixup
In reply to Josh PoimboeufQuoted 105 lines · Josh Poimboeuf
After making a change to improve objtool's sibling call detection, it started showing the following warning:
arch/x86/kvm/vmx/nested.o: warning: objtool: .fixup+0x15: sibling call from callable instruction with modified stack frameThe problem is the
____kvm_handle_fault_on_reboot()macro. It does a fake call by pushing a fake RIP and doing a jump. That tricks the unwinder into printing the function which triggered the exception, rather than the .fixup code.Instead of the hack to make it look like the original function made the call, just change the macro so that the original function actually does make the call. This allows removal of the hack, and also makes objtool happy.
I triggered a vmx instruction exception and verified that the stack trace is still sane:
kernel BUG at
arch/x86/kvm/x86.c:358!
invalid opcode: 0000 [#1] SMP PTICPU: 28 PID: 4096 Comm: qemu-kvm Not tainted 5.2.0+ #16 Hardware name: Lenovo THINKSYSTEM SD530 -[7X2106Z000]-/-[7X2106Z000]-, BIOS -[TEE113Z-1.00]- 07/17/2017 RIP: 0010:kvm_spurious_fault+0x5/0x10 Code: 00 00 00 00 00 8b 44 24 10 89 d2 45 89 c9 48 89 44 24 10 8b 44 24 08 48 89 44 24 08 e9 d4 40 22 00 0f 1f 40 00 0f 1f 44 00 00 <0f> 0b 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 55 49 89 fd 41 RSP: 0018:ffffbf91c683bd00 EFLAGS: 00010246 RAX: 000061f040000000 RBX: ffff9e159c77bba0 RCX: ffff9e15a5c87000 RDX: 0000000665c87000 RSI: ffff9e15a5c87000 RDI: ffff9e159c77bba0 RBP: 0000000000000000 R08: 0000000000000000 R09: ffff9e15a5c87000 R10: 0000000000000000 R11: fffff8f2d99721c0 R12: ffff9e159c77bba0 R13: ffffbf91c671d960 R14: ffff9e159c778000 R15: 0000000000000000 FS: 00007fa341cbe700(0000) GS:ffff9e15b7400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fdd38356804 CR3: 00000006759de003 CR4: 00000000007606e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 PKRU: 55555554 Call Trace: loaded_vmcs_init+0x4f/0xe0 alloc_loaded_vmcs+0x38/0xd0 vmx_create_vcpu+0xf7/0x600 kvm_vm_ioctl+0x5e9/0x980 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? free_one_page+0x13f/0x4e0 do_vfs_ioctl+0xa4/0x630 ksys_ioctl+0x60/0x90 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x55/0x1c0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fa349b1ee5bSigned-off-by Josh Poimboeuf <[email protected]>Acked-by Paolo Bonzini <[email protected]>---
v2: Fix____kvm_handle_fault_on_reboot()comment [Paolo]Cc Paolo Bonzini <[email protected]>Cc Radim Krčmář <[email protected]>--- arch/x86/include/asm/kvm_host.h | 34 ++++++++++++++++++--------------- 1 file changed, 19 insertions(+), 15 deletions(-)arch/x86/include/asm/kvm_host.h +12
-15@@ -1496,25 +1496,29 @@ enum {#define kvm_arch_vcpu_memslots_id (vcpu) ((vcpu)->arch.hflags & HF_SMM_MASK ? 1 : 0)#define kvm_memslots_for_spte_role (kvm, role) __kvm_memslots(kvm, (role).smm) +asmlinkage void __noreturn kvm_spurious_fault(void); +/* * Hardware virtualization extension instructions may fault if a * reboot turns off virtualization while processes are running. -* Trap the fault and ignore the instruction if that happens. +* Usually after catching the fault we just panic; during reboot +* instead the instruction is ignored. */ -asmlinkage void kvm_spurious_fault(void); - -#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ - "666: " insn "\n\t" \ - "668: \n\t" \ - ".pushsection .fixup, \"ax\" \n" \ - "667: \n\t" \ - cleanup_insn "\n\t" \ - "cmpb $0, kvm_rebooting \n\t" \ - "jne 668b \n\t" \ - __ASM_SIZE(push) " $666b \n\t" \ - "jmp kvm_spurious_fault \n\t" \ - ".popsection \n\t" \ - _ASM_EXTABLE(666b, 667b) +#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ + "666: \n\t" \ + insn "\n\t" \ + "jmp 668f \n\t" \ + "667: \n\t" \ + "call kvm_spurious_fault \n\t" \ + "668: \n\t" \ + ".pushsection .fixup, \"ax\" \n\t" \
arch/x86/include/asm/kvm_host.h enum { · line 1524
Quoted 10 lines
arch/x86/include/asm/kvm_host.h +7
-0+"700: \n\t" \ + cleanup_insn"\n\t" \ +"cmpb $0, kvm_rebooting\n\t" \ +"je 667b \n\t" \ +"jmp 668b \n\t" \ +".popsection \n\t" \ + _ASM_EXTABLE(666b, 700b)#define __kvm_handle_fault_on_reboot (insn) \ ____kvm_handle_fault_on_reboot(insn, "")
This has a side effect of adding a jump in a generally hot path, but let's hope that the speculation gods for once help us.
Paolo
Re: [PATCH v2 04/22] x86/kvm: Don't call kvm_spurious_fault() from .fixup
Quoted 74 lines · Paolo Bonzini
Quoted 73 lines · Josh Poimboeuf
After making a change to improve objtool's sibling call detection, it started showing the following warning:
arch/x86/kvm/vmx/nested.o: warning: objtool: .fixup+0x15: sibling call from callable instruction with modified stack frameThe problem is the
____kvm_handle_fault_on_reboot()macro. It does a fake call by pushing a fake RIP and doing a jump. That tricks the unwinder into printing the function which triggered the exception, rather than the .fixup code.Instead of the hack to make it look like the original function made the call, just change the macro so that the original function actually does make the call. This allows removal of the hack, and also makes objtool happy.
I triggered a vmx instruction exception and verified that the stack trace is still sane:
kernel BUG at
arch/x86/kvm/x86.c:358!
invalid opcode: 0000 [#1] SMP PTICPU: 28 PID: 4096 Comm: qemu-kvm Not tainted 5.2.0+ #16 Hardware name: Lenovo THINKSYSTEM SD530 -[7X2106Z000]-/-[7X2106Z000]-, BIOS -[TEE113Z-1.00]- 07/17/2017 RIP: 0010:kvm_spurious_fault+0x5/0x10 Code: 00 00 00 00 00 8b 44 24 10 89 d2 45 89 c9 48 89 44 24 10 8b 44 24 08 48 89 44 24 08 e9 d4 40 22 00 0f 1f 40 00 0f 1f 44 00 00 <0f> 0b 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 55 49 89 fd 41 RSP: 0018:ffffbf91c683bd00 EFLAGS: 00010246 RAX: 000061f040000000 RBX: ffff9e159c77bba0 RCX: ffff9e15a5c87000 RDX: 0000000665c87000 RSI: ffff9e15a5c87000 RDI: ffff9e159c77bba0 RBP: 0000000000000000 R08: 0000000000000000 R09: ffff9e15a5c87000 R10: 0000000000000000 R11: fffff8f2d99721c0 R12: ffff9e159c77bba0 R13: ffffbf91c671d960 R14: ffff9e159c778000 R15: 0000000000000000 FS: 00007fa341cbe700(0000) GS:ffff9e15b7400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fdd38356804 CR3: 00000006759de003 CR4: 00000000007606e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 PKRU: 55555554 Call Trace: loaded_vmcs_init+0x4f/0xe0 alloc_loaded_vmcs+0x38/0xd0 vmx_create_vcpu+0xf7/0x600 kvm_vm_ioctl+0x5e9/0x980 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? free_one_page+0x13f/0x4e0 do_vfs_ioctl+0xa4/0x630 ksys_ioctl+0x60/0x90 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x55/0x1c0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fa349b1ee5bSigned-off-by Josh Poimboeuf <[email protected]>Acked-by Paolo Bonzini <[email protected]>---
v2: Fix____kvm_handle_fault_on_reboot()comment [Paolo]Cc Paolo Bonzini <[email protected]>Cc Radim Krčmář <[email protected]>--- arch/x86/include/asm/kvm_host.h | 34 ++++++++++++++++++--------------- 1 file changed, 19 insertions(+), 15 deletions(-)arch/x86/include/asm/kvm_host.h +0
-0@@ -1496,25 +1496,29 @@ enum {#define kvm_arch_vcpu_memslots_id (vcpu) ((vcpu)->arch.hflags & HF_SMM_MASK ? 1 : 0)#define kvm_memslots_for_spte_role (kvm, role) __kvm_memslots(kvm, (role).smm)
Quoted 1 line
arch/x86/include/asm/kvm_host.h enum { · line 1499
Quoted 1 line
arch/x86/include/asm/kvm_host.h +1
-0+asmlinkage void __noreturn kvm_spurious_fault(void);
With __noreturn added, can the entry in __dead_end_function() in tools/objtool/check.c be removed?
Quoted 38 lines
Quoted 38 lines
arch/x86/include/asm/kvm_host.h +18
-15+/* * Hardware virtualization extension instructions may fault if a * reboot turns off virtualization while processes are running. -* Trap the fault and ignore the instruction if that happens. +* Usually after catching the fault we just panic; during reboot +* instead the instruction is ignored. */ -asmlinkage void kvm_spurious_fault(void); - -#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ - "666: " insn "\n\t" \ - "668: \n\t" \ - ".pushsection .fixup, \"ax\" \n" \ - "667: \n\t" \ - cleanup_insn "\n\t" \ - "cmpb $0, kvm_rebooting \n\t" \ - "jne 668b \n\t" \ - __ASM_SIZE(push) " $666b \n\t" \ - "jmp kvm_spurious_fault \n\t" \ - ".popsection \n\t" \ - _ASM_EXTABLE(666b, 667b) +#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ + "666: \n\t" \ + insn "\n\t" \ + "jmp 668f \n\t" \ + "667: \n\t" \ + "call kvm_spurious_fault \n\t" \ + "668: \n\t" \ + ".pushsection .fixup, \"ax\" \n\t" \ + "700: \n\t" \ + cleanup_insn "\n\t" \ + "cmpb $0, kvm_rebooting\n\t" \ + "je 667b \n\t" \ + "jmp 668b \n\t" \ + ".popsection \n\t" \ + _ASM_EXTABLE(666b, 700b)
Quoted 8 lines
arch/x86/include/asm/kvm_host.h enum { · line 1524
Quoted 3 lines
arch/x86/include/asm/kvm_host.h +0
-0#define __kvm_handle_fault_on_reboot (insn) \ ____kvm_handle_fault_on_reboot(insn, "")Acked-by Paolo Bonzini <[email protected]>This has a side effect of adding a jump in a generally hot path, but let's hope that the speculation gods for once help us.
Any reason not to take the same approach as vmx_vmenter() and ud2 directly from fixup? I've never found kvm_spurious_fault() to be all that helpful, IMO it's a win win. :-)
Paolo Bonzini
maintainerRe: [PATCH v2 04/22] x86/kvm: Don't call kvm_spurious_fault() from .fixup
Quoted 7 lines · Sean Christopherson
Quoted 4 lines
Acked-by Paolo Bonzini <[email protected]>This has a side effect of adding a jump in a generally hot path, but let's hope that the speculation gods for once help us.
Any reason not to take the same approach as
vmx_vmenter()and ud2 directly from fixup? I've never foundkvm_spurious_fault()to be all that helpful, IMO it's a win win. :-)
Honestly I've never seen a backtrace from here but I would rather not regret this when a customer encounters it...
Paolo
Re: [PATCH v2 04/22] x86/kvm: Don't call kvm_spurious_fault() from .fixup
In reply to Sean ChristophersonQuoted 8 lines · Sean Christopherson
Quoted 8 lines
Quoted 8 lines
arch/x86/include/asm/kvm_host.h +0
-0@@ -1496,25 +1496,29 @@ enum {#define kvm_arch_vcpu_memslots_id (vcpu) ((vcpu)->arch.hflags & HF_SMM_MASK ? 1 : 0)#define kvm_memslots_for_spte_role (kvm, role) __kvm_memslots(kvm, (role).smm)
Quoted 4 lines
Quoted 1 line
arch/x86/include/asm/kvm_host.h enum { · line 1499
Quoted 1 line
arch/x86/include/asm/kvm_host.h +1
-0+asmlinkage void __noreturn kvm_spurious_fault(void);With
__noreturnadded, can the entry in__dead_end_function()intools/objtool/check.cbe removed?
No, that's actually still needed because objtool can't see the __noreturn annotation. So it still needs to know that the "call kvm_spurious_fault" doesn't return.
Signature · 1 line
Josh
Re: [PATCH v2 04/22] x86/kvm: Don't call kvm_spurious_fault() from .fixup
In reply to Paolo BonziniQuoted 11 lines · Paolo Bonzini
Quoted 7 lines · Sean Christopherson
Quoted 4 lines
Acked-by Paolo Bonzini <[email protected]>This has a side effect of adding a jump in a generally hot path, but let's hope that the speculation gods for once help us.
Any reason not to take the same approach as
vmx_vmenter()and ud2 directly from fixup? I've never foundkvm_spurious_fault()to be all that helpful, IMO it's a win win. :-)Honestly I've never seen a backtrace from here but I would rather not regret this when a customer encounters it...
In theory, changing the "call kvm_spurious_fault" to ud2 should be fine. It should be tested, of course.
I would defer to Sean to make the patch on top of mine :-)
Signature · 1 line
Josh
Paolo Bonzini
maintainerRe: [PATCH v2 04/22] x86/kvm: Don't call kvm_spurious_fault() from .fixup
Quoted 10 lines · Josh Poimboeuf
Quoted 9 lines · Paolo Bonzini
Quoted 7 lines · Sean Christopherson
Quoted 4 lines
Acked-by Paolo Bonzini <[email protected]>This has a side effect of adding a jump in a generally hot path, but let's hope that the speculation gods for once help us.
Any reason not to take the same approach as
vmx_vmenter()and ud2 directly from fixup? I've never foundkvm_spurious_fault()to be all that helpful, IMO it's a win win. :-)
Quoted 8 lines
Quoted 2 lines
Honestly I've never seen a backtrace from here but I would rather not regret this when a customer encounters it...
In theory, changing the "call
kvm_spurious_fault" to ud2 should be fine. It should be tested, of course.I would defer to Sean to make the patch on top of mine :-)
Yes, this can be done easily on top.
Paolo
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/paravirt: Fix callee-saved function ELF sizes
In reply to Josh Poimboeufx86/paravirt: Fix callee-saved function ELF sizes
Commit-ID: 083db6764821996526970e42d09c1ab2f4155dd4 Gitweb: https://git.kernel.org/tip/083db6764821996526970e42d09c1ab2f4155dd4 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:36 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:03 +0200
x86/paravirt: Fix callee-saved function ELF sizes
The __raw_callee_save_*() functions have an ELF symbol size of zero, which confuses objtool and other tools.
Fixes a bunch of warnings like the following:
arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_pte_val() is missing an ELF size annotation arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_pgd_val() is missing an ELF size annotation arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_make_pte() is missing an ELF size annotation arch/x86/xen/mmu_pv.o: warning: objtool: __raw_callee_save_xen_make_pgd() is missing an ELF size annotation
--- arch/x86/include/asm/paravirt.h | 1 + arch/x86/kernel/kvm.c | 1 + 2 files changed, 2 insertions(+)
arch/x86/include/asm/paravirt.h +1 -0
@@ -746,6 +746,7 @@ bool __raw_callee_save___native_vcpu_is_preempted(long cpu);PV_RESTORE_ALL_CALLER_REGS \ FRAME_END \"ret;" \ + ".size " PV_THUNK_NAME(func) ", .-" PV_THUNK_NAME(func) ";" \ ".popsection")/* Get a reference to a callee-save function */
arch/x86/kernel/kvm.c +1 -0
@@ -838,6 +838,7 @@ asm( "cmpb $0, " __stringify(KVM_STEAL_TIME_preempted) "+steal_time(%rax);""setne %al;" "ret;" +".size __raw_callee_save___kvm_vcpu_is_preempted, .-__raw_callee_save___kvm_vcpu_is_preempted;" ".popsection");#endif
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/kvm: Fix fastop function ELF metadata
In reply to Josh Poimboeufx86/kvm: Fix fastop function ELF metadata
Commit-ID: d99a6ce70ec6ed990b74bd4e34232fd830d20d27 Gitweb: https://git.kernel.org/tip/d99a6ce70ec6ed990b74bd4e34232fd830d20d27 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:37 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:03 +0200
x86/kvm: Fix fastop function ELF metadata
Some of the fastop functions, e.g. em_setcc(), are actually just used as global labels which point to blocks of functions. The global labels are incorrectly annotated as functions. Also the functions themselves don't have size annotations.
Fixes a bunch of warnings like the following:
arch/x86/kvm/emulate.o: warning: objtool: seto() is missing an ELF size annotation arch/x86/kvm/emulate.o: warning: objtool: em_setcc() is missing an ELF size annotation arch/x86/kvm/emulate.o: warning: objtool: setno() is missing an ELF size annotation arch/x86/kvm/emulate.o: warning: objtool: setc() is missing an ELF size annotation
--- arch/x86/kvm/emulate.c | 44 +++++++++++++++++++++++++++++++------------- 1 file changed, 31 insertions(+), 13 deletions(-)
arch/x86/kvm/emulate.c +31 -13
@@ -312,29 +312,42 @@ static void invalidate_registers(struct x86_emulate_ctxt *ctxt)static int fastop (struct x86_emulate_ctxt *ctxt,void (*fop)(struct fastop *)); -#define FOP_FUNC (name) \ +#define __FOP_FUNC (name) \ ".align " __stringify(FASTOP_SIZE) " \n\t" \ ".type " name ", @function \n\t" \ name ":\n\t" -#define FOP_RET "ret \n\t" +#define FOP_FUNC (name) \ + __FOP_FUNC(#name) + +#define __FOP_RET (name) \ + "ret \n\t" \ + ".size " name ", .-" name "\n\t" + +#define FOP_RET (name) \ + __FOP_RET(#name)#define FOP_START (op) \ extern void em_##op(struct fastop *fake); \ asm(".pushsection .text, \"ax\" \n\t" \ ".global em_" #op " \n\t" \ - FOP_FUNC("em_" #op) + ".align " __stringify(FASTOP_SIZE) " \n\t" \ + "em_" #op ":\n\t"#define FOP_END \ ".popsection") +#define __FOPNOP (name) \ + __FOP_FUNC(name) \ + __FOP_RET(name) +#define FOPNOP () \ - FOP_FUNC(__stringify(__UNIQUE_ID(nop))) \ - FOP_RET + __FOPNOP(__stringify(__UNIQUE_ID(nop)))#define FOP1E (op, dst) \ - FOP_FUNC(#op "_" #dst) \ - "10: " #op " %" #dst " \n\t" FOP_RET + __FOP_FUNC(#op "_" #dst) \ + "10: " #op " %" #dst " \n\t" \ + __FOP_RET(#op "_" #dst)#define FOP1EEX (op, dst) \ FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception)
@@ -366,8 +379,9 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *)); FOP_END#define FOP2E (op, dst, src) \ - FOP_FUNC(#op "_" #dst "_" #src) \ - #op " %" #src ", %" #dst " \n\t" FOP_RET + __FOP_FUNC(#op "_" #dst "_" #src) \ + #op " %" #src ", %" #dst " \n\t" \ + __FOP_RET(#op "_" #dst "_" #src)#define FASTOP2 (op) \ FOP_START(op) \
@@ -405,8 +419,9 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *)); FOP_END#define FOP3E (op, dst, src, src2) \ - FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \ - #op " %" #src2 ", %" #src ", %" #dst " \n\t" FOP_RET + __FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \ + #op " %" #src2 ", %" #src ", %" #dst " \n\t"\ + __FOP_RET(#op "_" #dst "_" #src "_" #src2)/* 3-operand, word-only, src2=cl */ #define FASTOP3WCL (op) \
@@ -423,7 +438,7 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *)); ".type " #op ", @function \n\t" \ #op ": \n\t" \ #op " %al \n\t" \ - FOP_RET + __FOP_RET(#op) asm(".pushsection .fixup, \"ax\"\n" ".global kvm_fastop_exception \n"
@@ -449,7 +464,10 @@ FOP_SETCC(setle)FOP_SETCC (setnle)FOP_END ; -FOP_START(salc) "pushf; sbb %al, %al; popf \n\t" FOP_RET +FOP_START(salc) +FOP_FUNC(salc) +"pushf; sbb %al, %al; popf \n\t" +FOP_RET (salc) FOP_END;/*
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/kvm: Replace vmx_vmenter()'s call to kvm_spurious_fault() with UD2
In reply to Josh Poimboeufx86/kvm: Replace vmx_vmenter()'s call to kvm_spurious_fault() with UD2
Commit-ID: 19f2d8fa98644c7b78845b1d66abeae4e3d9dfa8 Gitweb: https://git.kernel.org/tip/19f2d8fa98644c7b78845b1d66abeae4e3d9dfa8 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:38 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:03 +0200
x86/kvm: Replace vmx_vmenter()'s call to kvm_spurious_fault() with UD2
Objtool reports the following:
arch/x86/kvm/vmx/vmenter.o: warning: objtool: vmx_vmenter()+0x14: call without frame pointer save/setup
But frame pointers are necessarily broken anyway, because __vmx_vcpu_run() clobbers RBP with the guest's value before calling vmx_vmenter(). So calling without a frame pointer doesn't make things any worse.
Make objtool happy by changing the call to a UD2.
--- arch/x86/kvm/vmx/vmenter.S | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
arch/x86/kvm/vmx/vmenter.S +3 -3
@@ -54,9 +54,9 @@ ENTRY(vmx_vmenter) ret 3: cmpb $0, kvm_rebooting - jne 4f - call kvm_spurious_fault -4: ret + je 4f + ret +4: ud2 .pushsection .fixup, "ax" 5: jmp 3b
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/kvm: Don't call kvm_spurious_fault() from .fixup
In reply to Josh Poimboeufx86/kvm: Don't call kvm_spurious_fault() from .fixup
Commit-ID: 3901336ed9887b075531bffaeef7742ba614058b Gitweb: https://git.kernel.org/tip/3901336ed9887b075531bffaeef7742ba614058b Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:39 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:04 +0200
x86/kvm: Don't call kvm_spurious_fault() from .fixup
After making a change to improve objtool's sibling call detection, it started showing the following warning:
arch/x86/kvm/vmx/nested.o: warning: objtool: .fixup+0x15: sibling call from callable instruction with modified stack frame
The problem is the ____kvm_handle_fault_on_reboot() macro. It does a fake call by pushing a fake RIP and doing a jump. That tricks the unwinder into printing the function which triggered the exception, rather than the .fixup code.
Instead of the hack to make it look like the original function made the call, just change the macro so that the original function actually does make the call. This allows removal of the hack, and also makes objtool happy.
I triggered a vmx instruction exception and verified that the stack trace is still sane:
kernel BUG at arch/x86/kvm/x86.c:358!
invalid opcode: 0000 [#1] SMP PTI
CPU: 28 PID: 4096 Comm: qemu-kvm Not tainted 5.2.0+ #16 Hardware name: Lenovo THINKSYSTEM SD530 -[7X2106Z000]-/-[7X2106Z000]-, BIOS -[TEE113Z-1.00]- 07/17/2017 RIP: 0010:kvm_spurious_fault+0x5/0x10 Code: 00 00 00 00 00 8b 44 24 10 89 d2 45 89 c9 48 89 44 24 10 8b 44 24 08 48 89 44 24 08 e9 d4 40 22 00 0f 1f 40 00 0f 1f 44 00 00 <0f> 0b 66 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 55 49 89 fd 41 RSP: 0018:ffffbf91c683bd00 EFLAGS: 00010246 RAX: 000061f040000000 RBX: ffff9e159c77bba0 RCX: ffff9e15a5c87000 RDX: 0000000665c87000 RSI: ffff9e15a5c87000 RDI: ffff9e159c77bba0 RBP: 0000000000000000 R08: 0000000000000000 R09: ffff9e15a5c87000 R10: 0000000000000000 R11: fffff8f2d99721c0 R12: ffff9e159c77bba0 R13: ffffbf91c671d960 R14: ffff9e159c778000 R15: 0000000000000000 FS: 00007fa341cbe700(0000) GS:ffff9e15b7400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fdd38356804 CR3: 00000006759de003 CR4: 00000000007606e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 PKRU: 55555554 Call Trace: loaded_vmcs_init+0x4f/0xe0 alloc_loaded_vmcs+0x38/0xd0 vmx_create_vcpu+0xf7/0x600 kvm_vm_ioctl+0x5e9/0x980 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? __switch_to_asm+0x40/0x70 ? __switch_to_asm+0x34/0x70 ? free_one_page+0x13f/0x4e0 do_vfs_ioctl+0xa4/0x630 ksys_ioctl+0x60/0x90 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x55/0x1c0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fa349b1ee5b
--- arch/x86/include/asm/kvm_host.h | 34 +++++++++++++++++++--------------- 1 file changed, 19 insertions(+), 15 deletions(-)
arch/x86/include/asm/kvm_host.h +19 -15
@@ -1496,25 +1496,29 @@ enum {#define kvm_arch_vcpu_memslots_id (vcpu) ((vcpu)->arch.hflags & HF_SMM_MASK ? 1 : 0)#define kvm_memslots_for_spte_role (kvm, role) __kvm_memslots(kvm, (role).smm) +asmlinkage void __noreturn kvm_spurious_fault(void); +/* * Hardware virtualization extension instructions may fault if a * reboot turns off virtualization while processes are running. -* Trap the fault and ignore the instruction if that happens. +* Usually after catching the fault we just panic; during reboot +* instead the instruction is ignored. */ -asmlinkage void kvm_spurious_fault(void); - -#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ - "666: " insn "\n\t" \ - "668: \n\t" \ - ".pushsection .fixup, \"ax\" \n" \ - "667: \n\t" \ - cleanup_insn "\n\t" \ - "cmpb $0, kvm_rebooting \n\t" \ - "jne 668b \n\t" \ - __ASM_SIZE(push) " $666b \n\t" \ - "jmp kvm_spurious_fault \n\t" \ - ".popsection \n\t" \ - _ASM_EXTABLE(666b, 667b) +#define ____kvm_handle_fault_on_reboot (insn, cleanup_insn) \ + "666: \n\t" \ + insn "\n\t" \ + "jmp 668f \n\t" \ + "667: \n\t" \ + "call kvm_spurious_fault \n\t" \ + "668: \n\t" \ + ".pushsection .fixup, \"ax\" \n\t" \ + "700: \n\t" \ + cleanup_insn "\n\t" \ + "cmpb $0, kvm_rebooting\n\t" \ + "je 667b \n\t" \ + "jmp 668b \n\t" \ + ".popsection \n\t" \ + _ASM_EXTABLE(666b, 700b)#define __kvm_handle_fault_on_reboot (insn) \ ____kvm_handle_fault_on_reboot(insn, "")
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/entry: Fix thunk function ELF sizes
In reply to Josh Poimboeufx86/entry: Fix thunk function ELF sizes
Commit-ID: e6dd47394493061c605285a868fc72eae2e9c866 Gitweb: https://git.kernel.org/tip/e6dd47394493061c605285a868fc72eae2e9c866 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:40 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:04 +0200
x86/entry: Fix thunk function ELF sizes
Fix the following warnings:
arch/x86/entry/thunk_64.o: warning: objtool: trace_hardirqs_on_thunk() is missing an ELF size annotation arch/x86/entry/thunk_64.o: warning: objtool: trace_hardirqs_off_thunk() is missing an ELF size annotation arch/x86/entry/thunk_64.o: warning: objtool: lockdep_sys_exit_thunk() is missing an ELF size annotation
--- arch/x86/entry/thunk_64.S | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)
arch/x86/entry/thunk_64.S +2 -3
@@ -12,9 +12,7 @@ /* rdi: arg1 ... normal C conventions. rax is saved/restored. */ .macro THUNK name, func, put_ret_addr_in_rdi=0 - .globl \name - .type \name, @function -\name: + ENTRY(\name) pushq %rbp movq %rsp, %rbp
@@ -35,6 +33,7 @@ call \func jmp .L_restore + ENDPROC(\name) _ASM_NOKPROBE(\name) .endm
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/head/64: Annotate start_cpu0() as non-callable
In reply to Josh Poimboeufx86/head/64: Annotate start_cpu0() as non-callable
Commit-ID: 61a73f5cd1a5794626d216cc56e20a1b195c5d0c Gitweb: https://git.kernel.org/tip/61a73f5cd1a5794626d216cc56e20a1b195c5d0c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:41 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:04 +0200
x86/head/64: Annotate start_cpu0() as non-callable
After an objtool improvement, it complains about the fact that start_cpu0() jumps to code which has an LRET instruction.
arch/x86/kernel/head_64.o: warning: objtool: .head.text+0xe4: unsupported instruction in callable function
Technically, start_cpu0() is callable, but it acts nothing like a callable function. Prevent objtool from treating it like one by removing its ELF function annotation.
--- arch/x86/kernel/head_64.S | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/head_64.S +2 -2
@@ -253,10 +253,10 @@ END(secondary_startup_64) * start_secondary() via .Ljump_to_C_code. */ ENTRY(start_cpu0) - movq initial_stack(%rip), %rsp UNWIND_HINT_EMPTY + movq initial_stack(%rip), %rsp jmp .Ljump_to_C_code -ENDPROC(start_cpu0) +END(start_cpu0) #endif /* Both SMP bootup and ACPI suspend change these variables */
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/uaccess: Remove ELF function annotation from copy_user_handle_tail()
In reply to Josh Poimboeufx86/uaccess: Remove ELF function annotation from copy_user_handle_tail()
Commit-ID: 3a6ab4bcc52263dd5b1d2fd2e4ce95a38c798b4d Gitweb: https://git.kernel.org/tip/3a6ab4bcc52263dd5b1d2fd2e4ce95a38c798b4d Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:42 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:05 +0200
x86/uaccess: Remove ELF function annotation from copy_user_handle_tail()
After an objtool improvement, it's complaining about the CLAC in copy_user_handle_tail():
arch/x86/lib/copy_user_64.o: warning: objtool: .altinstr_replacement+0x12: redundant UACCESS disable
arch/x86/lib/copy_user_64.o: warning: objtool: copy_user_handle_tail()+0x6: (alt) arch/x86/lib/copy_user_64.o: warning: objtool: copy_user_handle_tail()+0x2: (alt) arch/x86/lib/copy_user_64.o: warning: objtool: copy_user_handle_tail()+0x0: <=== (func)
copy_user_handle_tail() is incorrectly marked as a callable function, so objtool is rightfully concerned about the CLAC with no corresponding STAC.
Remove the ELF function annotation. The copy_user_handle_tail() code path is already verified by objtool because it's jumped to by other callable asm code (which does the corresponding STAC).
--- arch/x86/lib/copy_user_64.S | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/lib/copy_user_64.S +1 -1
@@ -239,7 +239,7 @@ copy_user_handle_tail: ret _ASM_EXTABLE_UA(1b, 2b) -ENDPROC(copy_user_handle_tail) +END(copy_user_handle_tail) /* * copy_user_nocache - Uncached memory copy with exception handling
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/uaccess: Don't leak AC flag into fentry from mcsafe_handle_tail()
In reply to Josh Poimboeufx86/uaccess: Don't leak AC flag into fentry from mcsafe_handle_tail()
Commit-ID: 5e307a6bc7b600999742675dd182bcb8a6fe308e Gitweb: https://git.kernel.org/tip/5e307a6bc7b600999742675dd182bcb8a6fe308e Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:43 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:05 +0200
x86/uaccess: Don't leak AC flag into fentry from mcsafe_handle_tail()
After adding mcsafe_handle_tail() to the objtool uaccess safe list, objtool reports:
arch/x86/lib/usercopy_64.o: warning: objtool: mcsafe_handle_tail()+0x0: call to __fentry__() with UACCESS enabled
With SMAP, this function is called with AC=1, so it needs to be careful about which functions it calls. Disable the ftrace entry hook, which can potentially pull in a lot of extra code.
--- arch/x86/lib/usercopy_64.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/lib/usercopy_64.c +1 -1
@@ -60,7 +60,7 @@ EXPORT_SYMBOL(clear_user);* but reuse __memcpy_mcsafe in case a new read error is encountered. * clac() is handled in _copy_to_iter_mcsafe(). */ -__visible unsigned long +__visible notrace unsigned longmcsafe_handle_tail (char *to,char *from,unsigned len) {for (; len; --len, to++, from++) {
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] x86/uaccess: Remove redundant CLACs in getuser/putuser error paths
In reply to Josh Poimboeufx86/uaccess: Remove redundant CLACs in getuser/putuser error paths
Commit-ID: 82e844a6536d1a3c12a73e44712f4021d90a4b53 Gitweb: https://git.kernel.org/tip/82e844a6536d1a3c12a73e44712f4021d90a4b53 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:44 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200
x86/uaccess: Remove redundant CLACs in getuser/putuser error paths
The same getuser/putuser error paths are used regardless of whether AC is set. In non-exception failure cases, this results in an unnecessary CLAC.
Fixes the following warnings:
arch/x86/lib/getuser.o: warning: objtool: .altinstr_replacement+0x18: redundant UACCESS disable arch/x86/lib/putuser.o: warning: objtool: .altinstr_replacement+0x18: redundant UACCESS disable
--- arch/x86/lib/getuser.S | 20 ++++++++++---------- arch/x86/lib/putuser.S | 29 ++++++++++++++++------------- 2 files changed, 26 insertions(+), 23 deletions(-)
arch/x86/lib/getuser.S +10 -10
@@ -115,29 +115,29 @@ ENDPROC(__get_user_8) EXPORT_SYMBOL(__get_user_8) +bad_get_user_clac: + ASM_CLAC bad_get_user: xor %edx,%edx mov $(-EFAULT),%_ASM_AX - ASM_CLAC ret -END(bad_get_user) #ifdef CONFIG_X86_32 +bad_get_user_8_clac: + ASM_CLAC bad_get_user_8: xor %edx,%edx xor %ecx,%ecx mov $(-EFAULT),%_ASM_AX - ASM_CLAC ret -END(bad_get_user_8) #endif - _ASM_EXTABLE_UA(1b, bad_get_user) - _ASM_EXTABLE_UA(2b, bad_get_user) - _ASM_EXTABLE_UA(3b, bad_get_user) + _ASM_EXTABLE_UA(1b, bad_get_user_clac) + _ASM_EXTABLE_UA(2b, bad_get_user_clac) + _ASM_EXTABLE_UA(3b, bad_get_user_clac) #ifdef CONFIG_X86_64 - _ASM_EXTABLE_UA(4b, bad_get_user) + _ASM_EXTABLE_UA(4b, bad_get_user_clac) #else - _ASM_EXTABLE_UA(4b, bad_get_user_8) - _ASM_EXTABLE_UA(5b, bad_get_user_8) + _ASM_EXTABLE_UA(4b, bad_get_user_8_clac) + _ASM_EXTABLE_UA(5b, bad_get_user_8_clac) #endif
arch/x86/lib/putuser.S +16 -13
@@ -32,8 +32,6 @@ */ #define ENTER mov PER_CPU_VAR(current_task), %_ASM_BX -#define EXIT ASM_CLAC ; \ - ret .text ENTRY(__put_user_1)
@@ -43,7 +41,8 @@ ENTRY(__put_user_1) ASM_STAC 1: movb %al,(%_ASM_CX) xor %eax,%eax - EXIT + ASM_CLAC + ret ENDPROC(__put_user_1) EXPORT_SYMBOL(__put_user_1)
@@ -56,7 +55,8 @@ ENTRY(__put_user_2) ASM_STAC 2: movw %ax,(%_ASM_CX) xor %eax,%eax - EXIT + ASM_CLAC + ret ENDPROC(__put_user_2) EXPORT_SYMBOL(__put_user_2)
@@ -69,7 +69,8 @@ ENTRY(__put_user_4) ASM_STAC 3: movl %eax,(%_ASM_CX) xor %eax,%eax - EXIT + ASM_CLAC + ret ENDPROC(__put_user_4) EXPORT_SYMBOL(__put_user_4)
@@ -85,19 +86,21 @@ ENTRY(__put_user_8) 5: movl %edx,4(%_ASM_CX) #endif xor %eax,%eax - EXIT + ASM_CLAC + RET ENDPROC(__put_user_8) EXPORT_SYMBOL(__put_user_8) +bad_put_user_clac: + ASM_CLAC bad_put_user: movl $-EFAULT,%eax - EXIT -END(bad_put_user) + RET - _ASM_EXTABLE_UA(1b, bad_put_user) - _ASM_EXTABLE_UA(2b, bad_put_user) - _ASM_EXTABLE_UA(3b, bad_put_user) - _ASM_EXTABLE_UA(4b, bad_put_user) + _ASM_EXTABLE_UA(1b, bad_put_user_clac) + _ASM_EXTABLE_UA(2b, bad_put_user_clac) + _ASM_EXTABLE_UA(3b, bad_put_user_clac) + _ASM_EXTABLE_UA(4b, bad_put_user_clac) #ifdef CONFIG_X86_32 - _ASM_EXTABLE_UA(5b, bad_put_user) + _ASM_EXTABLE_UA(5b, bad_put_user_clac) #endif
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] bpf: Disable GCC -fgcse optimization for ___bpf_prog_run()
In reply to Josh Poimboeufbpf: Disable GCC -fgcse optimization for ___bpf_prog_run()
Commit-ID: 3193c0836f203a91bef96d88c64cccf0be090d9c Gitweb: https://git.kernel.org/tip/3193c0836f203a91bef96d88c64cccf0be090d9c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:45 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200
bpf: Disable GCC -fgcse optimization for ___bpf_prog_run()
On x86-64, with CONFIG_RETPOLINE=n, GCC's "global common subexpression elimination" optimization results in ___bpf_prog_run()'s jumptable code changing from this:
select_insn: jmp *jumptable(, %rax, 8) ... ALU64_ADD_X: ... jmp *jumptable(, %rax, 8) ALU_ADD_X: ... jmp *jumptable(, %rax, 8)
to this:
select_insn: mov jumptable, %r12 jmp *(%r12, %rax, 8) ... ALU64_ADD_X: ... jmp *(%r12, %rax, 8) ALU_ADD_X: ... jmp *(%r12, %rax, 8)
The jumptable address is placed in a register once, at the beginning of the function. The function execution can then go through multiple indirect jumps which rely on that same register value. This has a few issues:
1) Objtool isn't smart enough to be able to track such a register value across multiple recursive indirect jumps through the jump table.
2) With CONFIG_RETPOLINE enabled, this optimization actually results in a small slowdown. I measured a ~4.7% slowdown in the test_bpf "tcpdump port 22" selftest.
This slowdown is actually predicted by the GCC manual:
Note: When compiling a program using computed gotos, a GCC extension, you may get better run-time performance if you disable the global common subexpression elimination pass by adding -fno-gcse to the command line.
So just disable the optimization for this function.
--- include/linux/compiler-gcc.h | 2 ++ include/linux/compiler_types.h | 4 ++++ kernel/bpf/core.c | 2 +- 3 files changed, 7 insertions(+), 1 deletion(-)
include/linux/compiler-gcc.h +2 -0
@@ -170,3 +170,5 @@#else #define __diag_GCC_8 (s)#endif + +#define __no_fgcse __attribute__((optimize("-fno-gcse")))
include/linux/compiler_types.h +4 -0
@@ -189,6 +189,10 @@ struct ftrace_likely_data { #define asm_volatile_goto(x...) asm goto(x)#endif +#ifndef __no_fgcse +# define __no_fgcse +#endif +/* Are two types/vars the same type (ignoring qualifiers)? */ #define __same_type (a, b) __builtin_types_compatible_p(typeof(a), typeof(b))
kernel/bpf/core.c +1 -1
@@ -1295,7 +1295,7 @@ bool bpf_opcode_in_insntable(u8 code)* * Decode and execute eBPF instructions. */ -static u64 ___bpf_prog_run (u64 *regs,const struct bpf_insn *insn,u64 *stack) +static u64 __no_fgcse ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn, u64 *stack) {#define BPF_INSN_2_LBL (x, y) [BPF_##x | BPF_##y] = &&x##_##y#define BPF_INSN_3_LBL (x, y, z) [BPF_##x | BPF_##y | BPF_##z] = &&x##_##y##_##z
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Add mcsafe_handle_tail() to the uaccess safe list
In reply to Josh Poimboeufobjtool: Add mcsafe_handle_tail() to the uaccess safe list
Commit-ID: a7e47f26039c26312a4144c3001b4e9fa886bd45 Gitweb: https://git.kernel.org/tip/a7e47f26039c26312a4144c3001b4e9fa886bd45 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:46 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200
objtool: Add mcsafe_handle_tail() to the uaccess safe list
After an objtool improvement, it's reporting that __memcpy_mcsafe() is calling mcsafe_handle_tail() with AC=1:
arch/x86/lib/memcpy_64.o: warning: objtool: .fixup+0x13: call to mcsafe_handle_tail() with UACCESS enabled
arch/x86/lib/memcpy_64.o: warning: objtool: __memcpy_mcsafe()+0x34: (alt) arch/x86/lib/memcpy_64.o: warning: objtool: __memcpy_mcsafe()+0xb: (branch) arch/x86/lib/memcpy_64.o: warning: objtool: __memcpy_mcsafe()+0x0: <=== (func)
mcsafe_handle_tail() is basically an extension of __memcpy_mcsafe(), so AC=1 is supposed to be set. Add mcsafe_handle_tail() to the uaccess safe list.
--- tools/objtool/check.c | 1 + 1 file changed, 1 insertion(+)
tools/objtool/check.c +1 -0
@@ -490,6 +490,7 @@ static const char *uaccess_safe_builtin[] = {/* misc */ "csum_partial_copy_generic" ,"__memcpy_mcsafe" , +"mcsafe_handle_tail" ,"ftrace_likely_update" ,/* CONFIG_TRACE_BRANCH_PROFILING */ NULL };
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Track original function across branches
In reply to Josh Poimboeufobjtool: Track original function across branches
Commit-ID: c705cecc8431951b4f34178e6b1db51b4a504c43 Gitweb: https://git.kernel.org/tip/c705cecc8431951b4f34178e6b1db51b4a504c43 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:47 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:07 +0200
objtool: Track original function across branches
If 'insn->func' is NULL, objtool skips some important checks, including sibling call validation. So if some .fixup code does an invalid sibling call, objtool ignores it.
Treat all code branches (including alts) as part of the original function by keeping track of the original func value from validate_functions().
This improves the usefulness of some clang function fallthrough warnings, and exposes some additional kernel bugs in the process.
--- tools/objtool/check.c | 28 ++++++++++++---------------- 1 file changed, 12 insertions(+), 16 deletions(-)
tools/objtool/check.c +12 -16
@@ -1934,13 +1934,12 @@ static int validate_sibling_call(struct instruction *insn, struct insn_state *st* each instruction and validate all the rules described in * tools/objtool/Documentation/stack-validation.txt. */ -static int validate_branch (struct objtool_file *file,struct instruction *first, -struct insn_state state) +static int validate_branch (struct objtool_file *file,struct symbol *func, +struct instruction *first,struct insn_state state) {struct alternative *alt;struct instruction *insn, *next_insn;struct section *sec; -struct symbol *func =NULL ;int ret; insn = first;
@@ -1961,9 +1960,6 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,return 1 ; } -if (insn->func) - func = insn->func->pfunc; -if (func && insn->ignore) {WARN_FUNC ("BUG: why am I validating an ignored function?" , sec, insn->offset);
@@ -1985,7 +1981,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first, i = insn; save_insn =NULL ; -func_for_each_insn_continue_reverse (file, insn->func, i) { +func_for_each_insn_continue_reverse (file, func, i) {if (i->save) { save_insn = i;break ;
@@ -2031,7 +2027,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,if (alt->skip_orig) skip_orig = true; - ret =validate_branch (file, alt->insn, state); + ret =validate_branch (file, func, alt->insn, state);if (ret) {if (backtrace)BT_FUNC ("(alt)" , insn);
@@ -2069,7 +2065,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,if (state.bp_scratch) {WARN ("%s uses BP as a scratch register" , - insn->func->name); + func->name);return 1 ; }
@@ -2109,8 +2105,8 @@ static int validate_branch(struct objtool_file *file, struct instruction *first, }else if (insn->jump_dest && (!func || !insn->jump_dest->func || insn->jump_dest->func->pfunc == func)) { - ret =validate_branch (file, insn->jump_dest, - state); + ret =validate_branch (file, func, + insn->jump_dest, state);if (ret) {if (backtrace)BT_FUNC ("(branch)" , insn);
@@ -2176,7 +2172,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,break ;case INSN_CLAC : -if (!state.uaccess && insn->func) { +if (!state.uaccess && func) {WARN_FUNC ("redundant UACCESS disable" , sec, insn->offset);return 1 ; }
@@ -2197,7 +2193,7 @@ static int validate_branch(struct objtool_file *file, struct instruction *first,break ;case INSN_CLD : -if (!state.df && insn->func) +if (!state.df && func)WARN_FUNC ("redundant CLD" , sec, insn->offset); state.df = false;
@@ -2236,7 +2232,7 @@ static int validate_unwind_hints(struct objtool_file *file)for_each_insn (file, insn) {if (insn->hint && !insn->visited) { - ret =validate_branch (file, insn, state); + ret =validate_branch (file, insn->func, insn, state);if (ret && backtrace)BT_FUNC ("<=== (hint)" , insn); warnings += ret;
@@ -2363,12 +2359,12 @@ static int validate_functions(struct objtool_file *file)continue ; insn =find_insn (file, sec, func->offset); -if (!insn || insn->ignore) +if (!insn || insn->ignore || insn->visited)continue ; state.uaccess = func->alias->uaccess_safe; - ret =validate_branch (file, insn, state); + ret =validate_branch (file, func, insn, state);if (ret && backtrace)BT_FUNC ("<=== (func)" , insn); warnings += ret;
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Refactor function alias logic
In reply to Josh Poimboeufobjtool: Refactor function alias logic
Commit-ID: e10cd8fe8ddfd28a172d2be57ae0e90c7f752e6a Gitweb: https://git.kernel.org/tip/e10cd8fe8ddfd28a172d2be57ae0e90c7f752e6a Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:48 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:07 +0200
objtool: Refactor function alias logic
- Add an alias check in validate_functions(). With this change, aliases no longer need uaccess_safe set.
- Add an alias check in decode_instructions(). With this change, the "if (!insn->func)" check is no longer needed.
- Don't create aliases for zero-length functions, as it can have unexpected results. The next patch will spit out a warning for zero-length functions anyway.
--- tools/objtool/check.c | 16 +++++++++------- tools/objtool/elf.c | 2 +- 2 files changed, 10 insertions(+), 8 deletions(-)
tools/objtool/check.c +9 -7
@@ -276,7 +276,7 @@ static int decode_instructions(struct objtool_file *file) }list_for_each_entry (func, &sec->symbol_list, list) { -if (func->type !=STT_FUNC ) +if (func->type !=STT_FUNC || func->alias != func)continue ;if (!find_insn (file, sec, func->offset)) {
@@ -286,8 +286,7 @@ static int decode_instructions(struct objtool_file *file) }func_for_each_insn (file, func, insn) -if (!insn->func) - insn->func = func; + insn->func = func; } }
@@ -508,7 +507,7 @@ static void add_uaccess_safe(struct objtool_file *file)if (!func)continue ; - func->alias->uaccess_safe = true; + func->uaccess_safe = true; } }
@@ -1887,7 +1886,7 @@ static bool insn_state_match(struct instruction *insn, struct insn_state *state)static inline bool func_uaccess_safe (struct symbol *func) {if (func) -return func->alias->uaccess_safe; +return func->uaccess_safe;return false; }
@@ -2355,14 +2354,17 @@ static int validate_functions(struct objtool_file *file)for_each_sec (file, sec) {list_for_each_entry (func, &sec->symbol_list, list) { -if (func->type !=STT_FUNC || func->pfunc != func) +if (func->type !=STT_FUNC ) +continue ; + +if (func->pfunc != func || func->alias != func)continue ; insn =find_insn (file, sec, func->offset);if (!insn || insn->ignore || insn->visited)continue ; - state.uaccess = func->alias->uaccess_safe; + state.uaccess = func->uaccess_safe; ret =validate_branch (file, func, insn, state);if (ret && backtrace)
tools/objtool/elf.c +1 -1
@@ -278,7 +278,7 @@ static int read_symbols(struct elf *elf) }if (sym->offset == s->offset) { -if (sym->len == s->len && alias == sym) +if (sym->len && sym->len == s->len && alias == sym) alias = s;if (sym->len >= s->len) {
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Warn on zero-length functions
In reply to Josh Poimboeufobjtool: Warn on zero-length functions
Commit-ID: 61e9b75a0ccf1fecacc28a2d77ea4a19aa404e39 Gitweb: https://git.kernel.org/tip/61e9b75a0ccf1fecacc28a2d77ea4a19aa404e39 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:49 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:07 +0200
objtool: Warn on zero-length functions
All callable functions should have an ELF size.
--- tools/objtool/check.c | 6 ++++++ 1 file changed, 6 insertions(+)
tools/objtool/check.c +6 -0
@@ -2357,6 +2357,12 @@ static int validate_functions(struct objtool_file *file)if (func->type !=STT_FUNC )continue ; +if (!func->len) { +WARN ("%s() is missing an ELF size annotation" , + func->name); + warnings++; + } +if (func->pfunc != func || func->alias != func)continue ;
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Change dead_end_function() to return boolean
In reply to Josh Poimboeufobjtool: Change dead_end_function() to return boolean
Commit-ID: 8e25c9f8b482ea8d8b6fb4f6f5c09bcc5ee18663 Gitweb: https://git.kernel.org/tip/8e25c9f8b482ea8d8b6fb4f6f5c09bcc5ee18663 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:50 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:08 +0200
objtool: Change dead_end_function() to return boolean
dead_end_function() can no longer return an error. Simplify its interface by making it return boolean.
--- tools/objtool/check.c | 36 ++++++++++++++---------------------- 1 file changed, 14 insertions(+), 22 deletions(-)
tools/objtool/check.c +14 -22
@@ -105,14 +105,9 @@ static struct instruction *next_insn_same_func(struct objtool_file *file,* * For local functions, we have to detect them manually by simply looking for * the lack of a return instruction. -* -* Returns: -* -1: error -* 0: no dead end -* 1: dead end */ -static int __dead_end_function (struct objtool_file *file,struct symbol *func, -int recursion) +static bool __dead_end_function (struct objtool_file *file,struct symbol *func, +int recursion) {int i;struct instruction *insn;
@@ -139,29 +134,29 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func, };if (func->bind ==STB_WEAK ) -return 0 ; +return false;if (func->bind ==STB_GLOBAL )for (i =0 ; i <ARRAY_SIZE (global_noreturns); i++)if (!strcmp (func->name, global_noreturns[i])) -return 1 ; +return true;if (!func->len) -return 0 ; +return false; insn =find_insn (file, func->sec, func->offset);if (!insn->func) -return 0 ; +return false;func_for_each_insn_all (file, func, insn) { empty = false;if (insn->type ==INSN_RETURN ) -return 0 ; +return false; }if (empty) -return 0 ; +return false;/* * A function can have a sibling call instead of a return. In that
@@ -174,7 +169,7 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func,if (!dest)/* sibling call to another file */ -return 0 ; +return false;if (dest->func && dest->func->pfunc != insn->func->pfunc) {
@@ -186,7 +181,7 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func,* This is a very rare case. It means * they aren't dead ends. */ -return 0 ; +return false; } return __dead_end_function(file, dest->func,
@@ -196,13 +191,13 @@ static int __dead_end_function(struct objtool_file *file, struct symbol *func,if (insn->type ==INSN_JUMP_DYNAMIC &&list_empty (&insn->alts))/* sibling call */ -return 0 ; +return false; } -return 1 ; +return true; } -static int dead_end_function (struct objtool_file *file,struct symbol *func) +static bool dead_end_function (struct objtool_file *file,struct symbol *func) {return __dead_end_function (file, func,0 ); }
@@ -2080,11 +2075,8 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,if (is_fentry_call (insn))break ; - ret =dead_end_function (file, insn->call_dest); -if (ret ==1 ) +if (dead_end_function (file, insn->call_dest))return 0 ; -if (ret ==-1 ) -return 1 ; }if (!no_fp && func && !has_valid_stack_frame (&state)) {
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Do frame pointer check before dead end check
In reply to Josh Poimboeufobjtool: Do frame pointer check before dead end check
Commit-ID: c9bab22bc449ad2496a6bbbf68acc711d9c5301c Gitweb: https://git.kernel.org/tip/c9bab22bc449ad2496a6bbbf68acc711d9c5301c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:51 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:08 +0200
objtool: Do frame pointer check before dead end check
Even calls to __noreturn functions need the frame pointer setup first. Such functions often dump the stack.
--- tools/objtool/check.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-)
tools/objtool/check.c +9 -9
@@ -133,6 +133,9 @@ static bool __dead_end_function(struct objtool_file *file, struct symbol *func, "rewind_stack_do_exit", }; +if (!func) +return false; +if (func->bind ==STB_WEAK )return false;
@@ -2071,19 +2074,16 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,if (ret)return ret; -if (insn->type ==INSN_CALL ) { -if (is_fentry_call (insn)) -break ; - -if (dead_end_function (file, insn->call_dest)) -return 0 ; - } - -if (!no_fp && func && !has_valid_stack_frame (&state)) { +if (!no_fp && func && !is_fentry_call (insn) && + !has_valid_stack_frame (&state)) {WARN_FUNC ("call without frame pointer save/setup" , sec, insn->offset);return 1 ; } + +if (dead_end_function (file, insn->call_dest)) +return 0 ; +break ;case INSN_JUMP_CONDITIONAL :
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Refactor sibling call detection logic
In reply to Josh Poimboeufobjtool: Refactor sibling call detection logic
Commit-ID: 0c1ddd33177530feb3685a800bba1ac4cc58cc4b Gitweb: https://git.kernel.org/tip/0c1ddd33177530feb3685a800bba1ac4cc58cc4b Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:52 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:08 +0200
objtool: Refactor sibling call detection logic
Simplify the sibling call detection logic a bit.
--- tools/objtool/check.c | 65 ++++++++++++++++++++++++++------------------------- 1 file changed, 33 insertions(+), 32 deletions(-)
tools/objtool/check.c +33 -32
@@ -97,6 +97,20 @@ static struct instruction *next_insn_same_func(struct objtool_file *file,for (insn =next_insn_same_sec (file, insn); insn; \ insn =next_insn_same_sec (file, insn)) +static bool is_sibling_call(struct instruction *insn) +{ +/* An indirect jump is either a sibling call or a jump to a table. */ +if (insn->type ==INSN_JUMP_DYNAMIC ) +return list_empty (&insn->alts); + +if (insn->type !=INSN_JUMP_CONDITIONAL && + insn->type !=INSN_JUMP_UNCONDITIONAL ) +return false; + +/* add_jump_destinations() sets insn->call_dest for sibling calls. */ +return !!insn->call_dest; +} +/* * This checks to see if the given function is a "noreturn" function. *
@@ -167,34 +181,25 @@ static bool __dead_end_function(struct objtool_file *file, struct symbol *func,* of the sibling call returns. */ func_for_each_insn_all (file, func, insn) { -if (insn->type ==INSN_JUMP_UNCONDITIONAL ) { +if (is_sibling_call (insn)) {struct instruction *dest = insn->jump_dest;if (!dest)/* sibling call to another file */ return false; -if (dest->func && dest->func->pfunc != insn->func->pfunc) { - -/* local sibling call */ -if (recursion ==5 ) { -/* -* Infinite recursion: two functions -* have sibling calls to each other. -* This is a very rare case. It means -* they aren't dead ends. -*/ -return false; - } - -return __dead_end_function (file, dest->func, - recursion +1 ); +/* local sibling call */ +if (recursion ==5 ) { +/* +* Infinite recursion: two functions have +* sibling calls to each other. This is a very +* rare case. It means they aren't dead ends. +*/ +return false; } - } -if (insn->type ==INSN_JUMP_DYNAMIC &&list_empty (&insn->alts)) -/* sibling call */ -return false; +return __dead_end_function (file, dest->func, recursion+1 ); + } }return true;
@@ -581,9 +586,8 @@ static int add_jump_destinations(struct objtool_file *file) insn->retpoline_safe = true;continue ; }else { -/* sibling call */ +/* external sibling call */ insn->call_dest = rela->sym; - insn->jump_dest =NULL ;continue ; }
@@ -633,9 +637,8 @@ static int add_jump_destinations(struct objtool_file *file) }else if (insn->jump_dest->func->pfunc != insn->func->pfunc && insn->jump_dest->offset == insn->jump_dest->func->offset) { -/* sibling class */ +/* internal sibling call */ insn->call_dest = insn->jump_dest->func; - insn->jump_dest =NULL ; } } }
@@ -1889,7 +1892,7 @@ static inline bool func_uaccess_safe(struct symbol *func)return false; } -static inline const char *insn_dest_name (struct instruction *insn) +static inline const char *call_dest_name (struct instruction *insn) {if (insn->call_dest)return insn->call_dest->name;
@@ -1901,13 +1904,13 @@ static int validate_call(struct instruction *insn, struct insn_state *state) {if (state->uaccess && !func_uaccess_safe (insn->call_dest)) {WARN_FUNC ("call to %s() with UACCESS enabled" , - insn->sec, insn->offset,insn_dest_name (insn)); + insn->sec, insn->offset,call_dest_name (insn));return 1 ; }if (state->df) {WARN_FUNC ("call to %s() with DF set" , - insn->sec, insn->offset,insn_dest_name (insn)); + insn->sec, insn->offset,call_dest_name (insn));return 1 ; }
@@ -2088,14 +2091,12 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,case INSN_JUMP_CONDITIONAL :case INSN_JUMP_UNCONDITIONAL : -if (func && !insn->jump_dest) { +if (func &&is_sibling_call (insn)) { ret =validate_sibling_call (insn, &state);if (ret)return ret; - }else if (insn->jump_dest && - (!func || !insn->jump_dest->func || - insn->jump_dest->func->pfunc == func)) { + }else if (insn->jump_dest) { ret =validate_branch (file, func, insn->jump_dest, state);if (ret) {
@@ -2111,7 +2112,7 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,break ;case INSN_JUMP_DYNAMIC : -if (func &&list_empty (&insn->alts)) { +if (func &&is_sibling_call (insn)) { ret =validate_sibling_call (insn, &state);if (ret)return ret;
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Refactor jump table code
In reply to Josh Poimboeufobjtool: Refactor jump table code
Commit-ID: e7c2bc37bfae120bce3e7cc8c8abf9d110af0757 Gitweb: https://git.kernel.org/tip/e7c2bc37bfae120bce3e7cc8c8abf9d110af0757 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:53 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:09 +0200
objtool: Refactor jump table code
Now that C jump tables are supported, call them "jump tables" instead of "switch tables". Also rename some other variables, add comments, and simplify the code flow a bit.
--- tools/objtool/check.c | 82 +++++++++++++++++++++++++++------------------------ tools/objtool/elf.c | 2 +- tools/objtool/elf.h | 2 +- 3 files changed, 46 insertions(+), 40 deletions(-)
tools/objtool/check.c +44 -38
@@ -627,7 +627,7 @@ static int add_jump_destinations(struct objtool_file *file)* However this code can't completely replace the * read_symbols() code because this doesn't detect the * case where the parent function's only reference to a -* subfunction is through a switch table. +* subfunction is through a jump table. */ if (!strstr (insn->func->name,".cold." ) &&strstr (insn->jump_dest->func->name,".cold." )) {
@@ -899,20 +899,24 @@ out:return ret; } -static int add_switch_table (struct objtool_file *file,struct instruction *insn, +static int add_jump_table (struct objtool_file *file,struct instruction *insn,struct rela *table,struct rela *next_table) {struct rela *rela = table; -struct instruction *alt_insn; +struct instruction *dest_insn;struct alternative *alt;struct symbol *pfunc = insn->func->pfunc;unsigned int prev_offset =0 ; -list_for_each_entry_from (rela, &table->rela_sec->rela_list, list) { +/* +* Each @rela is a switch table relocation which points to the target +* instruction. +*/ +list_for_each_entry_from (rela, &table->sec->rela_list, list) {if (rela == next_table)break ; -/* Make sure the switch table entries are consecutive: */ +/* Make sure the table entries are consecutive: */ if (prev_offset && rela->offset != prev_offset +8 )break ;
@@ -921,12 +925,12 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn, rela->addend == pfunc->offset)break ; - alt_insn =find_insn (file, rela->sym->sec, rela->addend); -if (!alt_insn) + dest_insn =find_insn (file, rela->sym->sec, rela->addend); +if (!dest_insn)break ; -/* Make sure the jmp dest is in the function or subfunction: */ -if (alt_insn->func->pfunc != pfunc) +/* Make sure the destination is in the same function: */ +if (dest_insn->func->pfunc != pfunc)break ; alt =malloc (sizeof (*alt));
@@ -935,7 +939,7 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn,return -1 ; } - alt->insn = alt_insn; + alt->insn = dest_insn;list_add_tail (&alt->list, &insn->alts); prev_offset = rela->offset; }
@@ -950,7 +954,7 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn, }/* -* find_switch_table() - Given a dynamic jump, find the switch jump table in +* find_jump_table() - Given a dynamic jump, find the switch jump table in * .rodata associated with it. * * There are 3 basic patterns:
@@ -992,13 +996,13 @@ static int add_switch_table(struct objtool_file *file, struct instruction *insn,* * NOTE: RETPOLINE made it harder still to decode dynamic jumps. */ -static struct rela *find_switch_table (struct objtool_file *file, +static struct rela *find_jump_table (struct objtool_file *file,struct symbol *func,struct instruction *insn) { -struct rela *text_rela, *rodata_rela; +struct rela *text_rela, *table_rela;struct instruction *orig_insn = insn; -struct section *rodata_sec; +struct section *table_sec;unsigned long table_offset;/*
@@ -1031,7 +1035,7 @@ static struct rela *find_switch_table(struct objtool_file *file,continue ; table_offset = text_rela->addend; - rodata_sec = text_rela->sym->sec; + table_sec = text_rela->sym->sec;if (text_rela->type ==R_X86_64_PC32 ) table_offset +=4 ;
@@ -1045,29 +1049,31 @@ static struct rela *find_switch_table(struct objtool_file *file,* need to be placed in the C_JUMP_TABLE_SECTION section. They * have symbols associated with them. */ -if (find_symbol_containing (rodata_sec, table_offset) && -strcmp (rodata_sec->name,C_JUMP_TABLE_SECTION )) +if (find_symbol_containing (table_sec, table_offset) && +strcmp (table_sec->name,C_JUMP_TABLE_SECTION ))continue ; - rodata_rela =find_rela_by_dest (rodata_sec, table_offset); -if (rodata_rela) { -/* -* Use of RIP-relative switch jumps is quite rare, and -* indicates a rare GCC quirk/bug which can leave dead -* code behind. -*/ -if (text_rela->type ==R_X86_64_PC32 ) - file->ignore_unreachables = true; +/* Each table entry has a rela associated with it. */ + table_rela =find_rela_by_dest (table_sec, table_offset); +if (!table_rela) +continue ; -return rodata_rela; - } +/* +* Use of RIP-relative switch jumps is quite rare, and +* indicates a rare GCC quirk/bug which can leave dead code +* behind. +*/ +if (text_rela->type ==R_X86_64_PC32 ) + file->ignore_unreachables = true; + +return table_rela; }return NULL ; } -static int add_func_switch_tables (struct objtool_file *file, +static int add_func_jump_tables (struct objtool_file *file,struct symbol *func) {struct instruction *insn, *last =NULL , *prev_jump =NULL ;
@@ -1080,7 +1086,7 @@ static int add_func_switch_tables(struct objtool_file *file,/* * Store back-pointers for unconditional forward jumps such -* that find_switch_table() can back-track using those and +* that find_jump_table() can back-track using those and * avoid some potentially confusing code. */ if (insn->type == INSN_JUMP_UNCONDITIONAL && insn->jump_dest &&
@@ -1095,17 +1101,17 @@ static int add_func_switch_tables(struct objtool_file *file,if (insn->type !=INSN_JUMP_DYNAMIC )continue ; - rela =find_switch_table (file, func, insn); + rela =find_jump_table (file, func, insn);if (!rela)continue ;/* -* We found a switch table, but we don't know yet how big it +* We found a jump table, but we don't know yet how big it * is. Don't add it until we reach the end of the function or -* the beginning of another switch table in the same function. +* the beginning of another jump table in the same function. */ if (prev_jump) { - ret =add_switch_table (file, prev_jump, prev_rela, rela); + ret =add_jump_table (file, prev_jump, prev_rela, rela);if (ret)return ret; }
@@ -1115,7 +1121,7 @@ static int add_func_switch_tables(struct objtool_file *file, }if (prev_jump) { - ret =add_switch_table (file, prev_jump, prev_rela,NULL ); + ret =add_jump_table (file, prev_jump, prev_rela,NULL );if (ret)return ret; }
@@ -1128,7 +1134,7 @@ static int add_func_switch_tables(struct objtool_file *file,* section which contains a list of addresses within the function to jump to. * This finds these jump tables and adds them to the insn->alts lists. */ -static int add_switch_table_alts (struct objtool_file *file) +static int add_jump_table_alts (struct objtool_file *file) {struct section *sec;struct symbol *func;
@@ -1142,7 +1148,7 @@ static int add_switch_table_alts(struct objtool_file *file)if (func->type !=STT_FUNC )continue ; - ret =add_func_switch_tables (file, func); + ret =add_func_jump_tables (file, func);if (ret)return ret; }
@@ -1339,7 +1345,7 @@ static int decode_sections(struct objtool_file *file)if (ret)return ret; - ret =add_switch_table_alts (file); + ret =add_jump_table_alts (file);if (ret)return ret;
tools/objtool/elf.c +1 -1
@@ -385,7 +385,7 @@ static int read_relas(struct elf *elf) rela->offset = rela->rela.r_offset; symndx =GELF_R_SYM (rela->rela.r_info); rela->sym =find_symbol_by_index (elf, symndx); - rela->rela_sec = sec; + rela->sec = sec;if (!rela->sym) {WARN ("can't find rela entry symbol %d for %s" , symndx, sec->name);
tools/objtool/elf.h +1 -1
@@ -57,7 +57,7 @@ struct rela {struct list_head list;struct hlist_node hash;GElf_Rela rela; -struct section *rela_sec; +struct section *sec;struct symbol *sym;unsigned int type;unsigned long offset;
tip-bot for Jann Horn
merge bot[tip:core/urgent] objtool: Support repeated uses of the same C jump table
In reply to Josh Poimboeufobjtool: Support repeated uses of the same C jump table
Commit-ID: bd98c81346468fc2f86aeeb44d4d0d6f763a62b7 Gitweb: https://git.kernel.org/tip/bd98c81346468fc2f86aeeb44d4d0d6f763a62b7 Author: Jann Horn <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:54 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:09 +0200
objtool: Support repeated uses of the same C jump table
This fixes objtool for both a GCC issue and a Clang issue:
1) GCC issue:
kernel/bpf/core.o: warning: objtool: ___bpf_prog_run()+0x8d5: sibling call from callable instruction with modified stack frame
With CONFIG_RETPOLINE=n, GCC is doing the following optimization in ___bpf_prog_run().
Before:
select_insn:
jmp *jumptable(,%rax,8)
...
ALU64_ADD_X:
...
jmp select_insn
ALU_ADD_X:
...
jmp select_insnAfter:
select_insn:
jmp *jumptable(, %rax, 8)
...
ALU64_ADD_X:
...
jmp *jumptable(, %rax, 8)
ALU_ADD_X:
...
jmp *jumptable(, %rax, 8)This confuses objtool. It has never seen multiple indirect jump sites which use the same jump table.
For GCC switch tables, the only way of detecting the size of a table is by continuing to scan for more tables. The size of the previous table can only be determined after another switch table is found, or when the scan reaches the end of the function.
That logic was reused for C jump tables, and was based on the assumption that each jump table only has a single jump site. The above optimization breaks that assumption.
2) Clang issue:
drivers/usb/misc/sisusbvga/sisusb.o: warning: objtool: sisusb_write_mem_bulk()+0x588: can't find switch jump table
With clang 9, code can be generated where a function contains two indirect jump instructions which use the same switch table.
The fix is the same for both issues: split the jump table parsing into two passes.
In the first pass, locate the heads of all switch tables for the function and mark their locations.
In the second pass, parse the switch tables and add them.
--- tools/objtool/check.c | 53 +++++++++++++++++++++++++++------------------------ tools/objtool/check.h | 1 + tools/objtool/elf.h | 1 + 3 files changed, 30 insertions(+), 25 deletions(-)
tools/objtool/check.c +28 -25
@@ -900,7 +900,7 @@ out: }static int add_jump_table (struct objtool_file *file,struct instruction *insn, -struct rela *table,struct rela *next_table) +struct rela *table) {struct rela *rela = table;struct instruction *dest_insn;
@@ -913,7 +913,9 @@ static int add_jump_table(struct objtool_file *file, struct instruction *insn,* instruction. */ list_for_each_entry_from (rela, &table->sec->rela_list, list) { -if (rela == next_table) + +/* Check for the end of the table: */ +if (rela != table && rela->jump_table_start)break ;/* Make sure the table entries are consecutive: */
@@ -1072,13 +1074,15 @@ static struct rela *find_jump_table(struct objtool_file *file,return NULL ; } - -static int add_func_jump_tables (struct objtool_file *file, -struct symbol *func) +/* +* First pass: Mark the head of each jump table so that in the next pass, +* we know when a given jump table ends and the next one starts. +*/ +static void mark_func_jump_tables (struct objtool_file *file, +struct symbol *func) { -struct instruction *insn, *last =NULL , *prev_jump =NULL ; -struct rela *rela, *prev_rela =NULL ; -int ret; +struct instruction *insn, *last =NULL ; +struct rela *rela;func_for_each_insn_all (file, func, insn) {if (!last)
@@ -1102,26 +1106,24 @@ static int add_func_jump_tables(struct objtool_file *file,continue ; rela =find_jump_table (file, func, insn); -if (!rela) -continue ; - -/* -* We found a jump table, but we don't know yet how big it -* is. Don't add it until we reach the end of the function or -* the beginning of another jump table in the same function. -*/ -if (prev_jump) { - ret =add_jump_table (file, prev_jump, prev_rela, rela); -if (ret) -return ret; +if (rela) { + rela->jump_table_start = true; + insn->jump_table = rela; } - - prev_jump = insn; - prev_rela = rela; } +} + +static int add_func_jump_tables (struct objtool_file *file, +struct symbol *func) +{ +struct instruction *insn; +int ret; + +func_for_each_insn_all (file, func, insn) { +if (!insn->jump_table) +continue ; -if (prev_jump) { - ret =add_jump_table (file, prev_jump, prev_rela,NULL ); + ret =add_jump_table (file, insn, insn->jump_table);if (ret)return ret; }
@@ -1148,6 +1150,7 @@ static int add_jump_table_alts(struct objtool_file *file)if (func->type !=STT_FUNC )continue ; +mark_func_jump_tables (file, func); ret =add_func_jump_tables (file, func);if (ret)return ret;
tools/objtool/check.h +1 -0
@@ -38,6 +38,7 @@ struct instruction {struct symbol *call_dest;struct instruction *jump_dest;struct instruction *first_jump_src; +struct rela *jump_table;struct list_head alts;struct symbol *func;struct stack_op stack_op;
tools/objtool/elf.h +1 -0
@@ -62,6 +62,7 @@ struct rela {unsigned int type;unsigned long offset;int addend; +bool jump_table_start; };struct elf {
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Fix seg fault on bad switch table entry
In reply to Josh Poimboeufobjtool: Fix seg fault on bad switch table entry
Commit-ID: e65050b94d8c518fdbee572ea4ca6d352e1fda37 Gitweb: https://git.kernel.org/tip/e65050b94d8c518fdbee572ea4ca6d352e1fda37 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:55 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:10 +0200
objtool: Fix seg fault on bad switch table entry
In one rare case, Clang generated the following code:
5ca: 83 e0 21 and $0x21,%eax
5cd: b9 04 00 00 00 mov $0x4,%ecx
5d2: ff 24 c5 00 00 00 00 jmpq *0x0(,%rax,8)
5d5: R_X86_64_32S .rodata+0x38which uses the corresponding jump table relocations:
000000000038 000200000001 R_X86_64_64 0000000000000000 .text + 834 000000000040 000200000001 R_X86_64_64 0000000000000000 .text + 5d9 000000000048 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000050 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000058 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000060 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000068 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000070 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000078 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000080 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000088 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000090 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000098 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000a0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000a8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000b0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000b8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000c0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000c8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000d0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000d8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000e0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000e8 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000f0 000200000001 R_X86_64_64 0000000000000000 .text + b96 0000000000f8 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000100 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000108 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000110 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000118 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000120 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000128 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000130 000200000001 R_X86_64_64 0000000000000000 .text + b96 000000000138 000200000001 R_X86_64_64 0000000000000000 .text + 82f 000000000140 000200000001 R_X86_64_64 0000000000000000 .text + 828
Since %eax was masked with 0x21, only the first two and the last two entries are possible.
Objtool doesn't actually emulate all the code, so it isn't smart enough to know that all the middle entries aren't reachable. They point to the NOP padding area after the end of the function, so objtool seg faulted when it tried to dereference a NULL insn->func.
After this fix, objtool still gives an "unreachable" error because it stops reading the jump table when it encounters the bad addresses:
/home/jpoimboe/objtool-tests/adm1275.o: warning: objtool: adm1275_probe()+0x828: unreachable instruction
While the above code is technically correct, it's very wasteful of memory -- it uses 34 jump table entries when only 4 are needed. It's also not possible for objtool to validate this type of switch table because the unused entries point outside the function and objtool has no way of determining if that's intentional. Hopefully the Clang folks can fix it.
--- tools/objtool/check.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/objtool/check.c +1 -1
@@ -932,7 +932,7 @@ static int add_jump_table(struct objtool_file *file, struct instruction *insn,break ;/* Make sure the destination is in the same function: */ -if (dest_insn->func->pfunc != pfunc) +if (!dest_insn->func || dest_insn->func->pfunc != pfunc)break ; alt =malloc (sizeof (*alt));
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Convert insn type to enum
In reply to Josh Poimboeufobjtool: Convert insn type to enum
Commit-ID: 9fe7b7642fe2c5158904d06fe31b740ca0695a01 Gitweb: https://git.kernel.org/tip/9fe7b7642fe2c5158904d06fe31b740ca0695a01 Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:56 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:10 +0200
objtool: Convert insn type to enum
This makes it easier to add new instruction types. Also it's hopefully more robust since the compiler should warn about out-of-range enums.
--- tools/objtool/arch.h | 35 ++++++++++++++++++----------------- tools/objtool/arch/x86/decode.c | 2 +- tools/objtool/check.c | 7 ------- tools/objtool/check.h | 2 +- 4 files changed, 20 insertions(+), 26 deletions(-)
tools/objtool/arch.h +18 -17
@@ -11,22 +11,23 @@#include "elf.h" #include "cfi.h" -#define INSN_JUMP_CONDITIONAL 1 -#define INSN_JUMP_UNCONDITIONAL 2 -#define INSN_JUMP_DYNAMIC 3 -#define INSN_CALL 4 -#define INSN_CALL_DYNAMIC 5 -#define INSN_RETURN 6 -#define INSN_CONTEXT_SWITCH 7 -#define INSN_STACK 8 -#define INSN_BUG 9 -#define INSN_NOP 10 -#define INSN_STAC 11 -#define INSN_CLAC 12 -#define INSN_STD 13 -#define INSN_CLD 14 -#define INSN_OTHER 15 -#define INSN_LAST INSN_OTHER +enum insn_type { +INSN_JUMP_CONDITIONAL , +INSN_JUMP_UNCONDITIONAL , +INSN_JUMP_DYNAMIC , +INSN_CALL , +INSN_CALL_DYNAMIC , +INSN_RETURN , +INSN_CONTEXT_SWITCH , +INSN_STACK , +INSN_BUG , +INSN_NOP , +INSN_STAC , +INSN_CLAC , +INSN_STD , +INSN_CLD , +INSN_OTHER , +};enum op_dest_type {OP_DEST_REG ,
@@ -68,7 +69,7 @@ void arch_initial_func_cfi_state(struct cfi_state *state);int arch_decode_instruction (struct elf *elf,struct section *sec,unsigned long offset,unsigned int maxlen, -unsigned int *len,unsigned char *type, +unsigned int *len,enum insn_type *type,unsigned long *immediate,struct stack_op *op);bool arch_callee_saved_reg (unsigned char reg);
tools/objtool/arch/x86/decode.c +1 -1
@@ -68,7 +68,7 @@ bool arch_callee_saved_reg(unsigned char reg)int arch_decode_instruction (struct elf *elf,struct section *sec,unsigned long offset,unsigned int maxlen, -unsigned int *len,unsigned char *type, +unsigned int *len,enum insn_type *type,unsigned long *immediate,struct stack_op *op) {struct insn insn;
tools/objtool/check.c +0 -7
@@ -267,13 +267,6 @@ static int decode_instructions(struct objtool_file *file)if (ret) goto err; -if (!insn->type || insn->type >INSN_LAST ) { -WARN_FUNC ("invalid instruction type %d" , - insn->sec, insn->offset, insn->type); - ret =-1 ; - goto err; - } -hash_add (file->insn_hash, &insn->hash, insn->offset);list_add_tail (&insn->list, &file->insn_list); }
tools/objtool/check.h +1 -1
@@ -31,7 +31,7 @@ struct instruction {struct section *sec;unsigned long offset;unsigned int len; -unsigned char type; +enum insn_type type;unsigned long immediate;bool alt_group, visited, dead_end, ignore, hint, save, restore, ignore_alts;bool retpoline_safe;
tip-bot for Josh Poimboeuf
merge bot[tip:core/urgent] objtool: Support conditional retpolines
In reply to Josh Poimboeufobjtool: Support conditional retpolines
Commit-ID: b68b9907069a8d3a65bc16a35360bf8f8603c8fa Gitweb: https://git.kernel.org/tip/b68b9907069a8d3a65bc16a35360bf8f8603c8fa Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:57 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:10 +0200
objtool: Support conditional retpolines
A Clang-built kernel is showing the following warning:
arch/x86/kernel/platform-quirks.o: warning: objtool: x86_early_init_platform_quirks()+0x84: unreachable instruction
That corresponds to this code:
7e: 0f 85 00 00 00 00 jne 84 <x86_early_init_platform_quirks+0x84>
80: R_X86_64_PC32 __x86_indirect_thunk_r11-0x4
84: c3 retq
This is a conditional retpoline sibling call, which is now possible thanks to retpolines. Objtool hasn't seen that before. It's incorrectly interpreting the conditional jump as an unconditional dynamic jump.
--- tools/objtool/arch.h | 1 + tools/objtool/check.c | 12 ++++++++++-- 2 files changed, 11 insertions(+), 2 deletions(-)
tools/objtool/arch.h +1 -0
@@ -15,6 +15,7 @@ enum insn_type {INSN_JUMP_CONDITIONAL ,INSN_JUMP_UNCONDITIONAL ,INSN_JUMP_DYNAMIC , +INSN_JUMP_DYNAMIC_CONDITIONAL ,INSN_CALL ,INSN_CALL_DYNAMIC , INSN_RETURN,
tools/objtool/check.c +10 -2
@@ -575,7 +575,11 @@ static int add_jump_destinations(struct objtool_file *file)* Retpoline jumps are really dynamic jumps in * disguise, so convert them accordingly. */ - insn->type =INSN_JUMP_DYNAMIC ; +if (insn->type ==INSN_JUMP_UNCONDITIONAL ) + insn->type =INSN_JUMP_DYNAMIC ; +else + insn->type =INSN_JUMP_DYNAMIC_CONDITIONAL ; + insn->retpoline_safe = true;continue ; }else {
@@ -2114,13 +2118,17 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,break ;case INSN_JUMP_DYNAMIC : +case INSN_JUMP_DYNAMIC_CONDITIONAL :if (func &&is_sibling_call (insn)) { ret =validate_sibling_call (insn, &state);if (ret)return ret; } -return 0 ; +if (insn->type ==INSN_JUMP_DYNAMIC ) +return 0 ; + +break ;case INSN_CONTEXT_SWITCH :if (func && (!next_insn || !next_insn->hint)) {
BPF vs objtool again
In reply to tip-bot for Josh PoimboeufQuoted 8 lines · tip-bot for Josh Poimboeuf
Commit-ID: 3193c0836f203a91bef96d88c64cccf0be090d9c Gitweb: https://git.kernel.org/tip/3193c0836f203a91bef96d88c64cccf0be090d9c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:45 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200bpf: Disable GCC -fgcse optimization for
___bpf_prog_run()
For some reason, this
__attribute__((optimize("-fno-gcse")))
is disabling frame pointers in ___bpf_prog_run(). If you compile with CONFIG_FRAME_POINTER it'll show something like:
kernel/bpf/core.o: warning: objtool: ___bpf_prog_run.cold()+0x7: call without frame pointer save/setup
Also, since GCC 9.1, the GCC docs say "The optimize attribute should be used for debugging purposes only. It is not suitable in production code." That doesn't sound too promising.
So it seems like this commit should be reverted. But then we're back to objtool being broken again in the RETPOLINE=n case, which means no ORC coverage in this function. (See above commit for the details)
Some ideas:
- Skip objtool checking of that func/file (at least for RETPOLINE=n) -- but then it won't have ORC coverage.
- Get rid of the "double goto" in ___bpf_prog_run(), which simplifies it enough for objtool to understand -- but then the text explodes for RETPOLINE=y.
- Add -fno-gfcse to the Makefile for kernel/bpf/core.c -- but then that affects the optimization of other functions in the file. However I don't think the impact is significant.
- Move ___bpf_prog_run() to its own file with the -fno-gfcse flag. I'm thinking this could be the least bad option. Alexei?
Signature · 1 line
Josh
Re: BPF vs objtool again
Quoted 14 lines · Josh Poimboeuf
Quoted 8 lines · tip-bot for Josh Poimboeuf
Commit-ID: 3193c0836f203a91bef96d88c64cccf0be090d9c Gitweb: https://git.kernel.org/tip/3193c0836f203a91bef96d88c64cccf0be090d9c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:45 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200bpf: Disable GCC -fgcse optimization for
___bpf_prog_run()For some reason, this
__attribute__((optimize("-fno-gcse")))
Quoted 9 lines
is disabling frame pointers in
___bpf_prog_run(). If you compile withCONFIG_FRAME_POINTERit'll show something like:
kernel/bpf/core.o: warning: objtool: ___bpf_prog_run.cold()+0x7: call without frame pointer save/setupAlso, since GCC 9.1, the GCC docs say "The optimize attribute should be used for debugging purposes only. It is not suitable in production code." That doesn't sound too promising.
It turns out that the optimize attribute doesn't append options to the command-line arguments, it starts from the defaults and only adds whatever you specify in the attribute. So it's not very useful for production code.
See this for eg where the same thing came up in a different context.
https://lore.kernel.org/lkml/[email protected]/
Re: BPF vs objtool again
In reply to Josh PoimboeufQuoted 10 lines · Josh Poimboeuf
Quoted 8 lines · tip-bot for Josh Poimboeuf
Commit-ID: 3193c0836f203a91bef96d88c64cccf0be090d9c Gitweb: https://git.kernel.org/tip/3193c0836f203a91bef96d88c64cccf0be090d9c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:45 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200bpf: Disable GCC -fgcse optimization for
___bpf_prog_run()
Quoted 8 lines
For some reason, this
__attribute__((optimize("-fno-gcse")))
is disabling frame pointers in
___bpf_prog_run(). If you compile withCONFIG_FRAME_POINTERit'll show something like:
kernel/bpf/core.o: warning: objtool: ___bpf_prog_run.cold()+0x7: call without frame pointer save/setup
you mean it started to disable frame pointers from some version of gcc? It wasn't doing this before, since objtool wasn't complaining, right? Sounds like gcc bug?
Quoted 8 lines
Also, since GCC 9.1, the GCC docs say "The optimize attribute should be used for debugging purposes only. It is not suitable in production code." That doesn't sound too promising.
So it seems like this commit should be reverted. But then we're back to objtool being broken again in the RETPOLINE=n case, which means no ORC coverage in this function. (See above commit for the details)
Quoted 8 lines
Some ideas:
- Skip objtool checking of that func/file (at least for RETPOLINE=n) -- but then it won't have ORC coverage.
- Get rid of the "double goto" in
___bpf_prog_run(), which simplifies it enough for objtool to understand -- but then the text explodes for RETPOLINE=y.
How that will look like?
That could be the best option.
Quoted 6 lines
- Add -fno-gfcse to the
Makefileforkernel/bpf/core.c-- but then that affects the optimization of other functions in the file. However I don't think the impact is significant.- Move
___bpf_prog_run()to its own file with the -fno-gfcse flag. I'm thinking this could be the least bad option. Alexei?
I think it would be easier to move some of the hot path functions out of core.c instead.
Like *ksym*, BPF_CALL*, bpf_jit*, bpf_prog*.
I think resulting churn will be less.
imo it's more important to keep git blame history for interpreter than for the other funcs.
Sounds like it's a fix that needs to be sent for the next RC ? Please send a patch for bpf tree then.
Daniel, thoughts?
Re: BPF vs objtool again
Quoted 15 lines · Alexei Starovoitov
Quoted 14 lines · Josh Poimboeuf
Quoted 8 lines · tip-bot for Josh Poimboeuf
Commit-ID: 3193c0836f203a91bef96d88c64cccf0be090d9c Gitweb: https://git.kernel.org/tip/3193c0836f203a91bef96d88c64cccf0be090d9c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:45 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200bpf: Disable GCC -fgcse optimization for
___bpf_prog_run()For some reason, this
__attribute__((optimize("-fno-gcse")))
Quoted 8 lines
Quoted 4 lines
is disabling frame pointers in
___bpf_prog_run(). If you compile withCONFIG_FRAME_POINTERit'll show something like:
kernel/bpf/core.o: warning: objtool: ___bpf_prog_run.cold()+0x7: call without frame pointer save/setupyou mean it started to disable frame pointers from some version of gcc? It wasn't doing this before, since objtool wasn't complaining, right? Sounds like gcc bug?
I actually think this warning has been around for a while. I just only recently looked at it. I don't think anything changed in GCC, it's just that almost nobody uses CONFIG_FRAME_POINTER, so it wasn't really noticed.
Quoted 13 lines
Quoted 13 lines
Also, since GCC 9.1, the GCC docs say "The optimize attribute should be used for debugging purposes only. It is not suitable in production code." That doesn't sound too promising.
So it seems like this commit should be reverted. But then we're back to objtool being broken again in the RETPOLINE=n case, which means no ORC coverage in this function. (See above commit for the details)
Some ideas:
- Skip objtool checking of that func/file (at least for RETPOLINE=n) -- but then it won't have ORC coverage.
Quoted 6 lines
Quoted 3 lines
- Get rid of the "double goto" in
___bpf_prog_run(), which simplifies it enough for objtool to understand -- but then the text explodes for RETPOLINE=y.How that will look like?
That could be the best option.
For example:
#define GOTO ({ goto *jumptable[insn->code]; })
and then replace all 'goto select_insn' with 'GOTO;'
The problem is that with RETPOLINE=y, the function text size grows from 5k to 7k, because for each of the 160+ retpoline JMPs, GCC (stupidly) reloads the jump table register into a scratch register.
Quoted 7 lines
Quoted 6 lines
- Add -fno-gfcse to the
Makefileforkernel/bpf/core.c-- but then that affects the optimization of other functions in the file. However I don't think the impact is significant.- Move
___bpf_prog_run()to its own file with the -fno-gfcse flag. I'm thinking this could be the least bad option. Alexei?
Quoted 8 lines
I think it would be easier to move some of the hot path functions out of
core.cinstead.
Like *ksym*, BPF_CALL*, bpf_jit*, bpf_prog*.
I think resulting churn will be less.
imo it's more important to keep git blame history for interpreter than for the other funcs.
Sounds like it's a fix that needs to be sent for the next RC ? Please send a patch for bpf tree then.
I can make a patch, what file would you recommend moving those hot path functions to?
Signature · 1 line
Josh
Re: BPF vs objtool again
Quoted 52 lines · Josh Poimboeuf
Quoted 15 lines · Alexei Starovoitov
Quoted 14 lines · Josh Poimboeuf
Quoted 8 lines · tip-bot for Josh Poimboeuf
Commit-ID: 3193c0836f203a91bef96d88c64cccf0be090d9c Gitweb: https://git.kernel.org/tip/3193c0836f203a91bef96d88c64cccf0be090d9c Author: Josh Poimboeuf <[email protected]> AuthorDate: Wed, 17 Jul 2019 20:36:45 -0500 Committer: Thomas Gleixner <[email protected]> CommitDate: Thu, 18 Jul 2019 21:01:06 +0200bpf: Disable GCC -fgcse optimization for
___bpf_prog_run()For some reason, this
__attribute__((optimize("-fno-gcse")))
Quoted 8 lines
Quoted 4 lines
is disabling frame pointers in
___bpf_prog_run(). If you compile withCONFIG_FRAME_POINTERit'll show something like:
kernel/bpf/core.o: warning: objtool: ___bpf_prog_run.cold()+0x7: call without frame pointer save/setupyou mean it started to disable frame pointers from some version of gcc? It wasn't doing this before, since objtool wasn't complaining, right? Sounds like gcc bug?
I actually think this warning has been around for a while. I just only recently looked at it. I don't think anything changed in GCC, it's just that almost nobody uses
CONFIG_FRAME_POINTER, so it wasn't really noticed.Quoted 13 lines
Quoted 13 lines
Also, since GCC 9.1, the GCC docs say "The optimize attribute should be used for debugging purposes only. It is not suitable in production code." That doesn't sound too promising.
So it seems like this commit should be reverted. But then we're back to objtool being broken again in the RETPOLINE=n case, which means no ORC coverage in this function. (See above commit for the details)
Some ideas:
- Skip objtool checking of that func/file (at least for RETPOLINE=n) -- but then it won't have ORC coverage.
Quoted 6 lines
Quoted 3 lines
- Get rid of the "double goto" in
___bpf_prog_run(), which simplifies it enough for objtool to understand -- but then the text explodes for RETPOLINE=y.How that will look like?
That could be the best option.For example:
Quoted 7 lines
#define GOTO ({ goto *jumptable[insn->code]; })
and then replace all 'goto
select_insn' with 'GOTO;'The problem is that with RETPOLINE=y, the function text size grows from 5k to 7k, because for each of the 160+ retpoline JMPs, GCC (stupidly) reloads the jump table register into a scratch register.
that would be a tiny change, right?
I'd rather go with that and gate it with 'ifdef CONFIG_FRAME_POINTER' Like:
#ifndef CONFIG_FRAME_POINTER
#define CONT ({ insn++; goto select_insn; })
#define CONT_JMP ({ insn++; goto select_insn; })
#else
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#define CONT_JMP ({ insn++; goto *jumptable[insn->code]; }) #endif
The reason this CONT and CONT_JMP macros are there because a combination of different gcc versions together with different cpus make branch predictor behave 'unpredictably'.
I've played with CONT and CONT_JMP either both doing direct goto or indirect goto and observed quite different performance characteristics from the interpreter.
What you see right now was the best tune I could get from a set of cpus I had to play with and compilers. If I did the same tuning today the outcome could have been different.
So I think it's totally fine to use above code. I think some cpus may actually see performance gains with certain versions of gcc.
The retpoline text increase is unfortunate but when retpoline is on other security knobs should be on too. In particular CONFIG_BPF_JIT_ALWAYS_ON should be on as well. Which will remove interpreter from .text completely.
Re: BPF vs objtool again
Quoted 25 lines · Alexei Starovoitov
Quoted 9 lines
For example:
#define GOTO ({ goto *jumptable[insn->code]; })
and then replace all 'goto
select_insn' with 'GOTO;'The problem is that with RETPOLINE=y, the function text size grows from 5k to 7k, because for each of the 160+ retpoline JMPs, GCC (stupidly) reloads the jump table register into a scratch register.
that would be a tiny change, right?
I'd rather go with that and gate it with 'ifdefCONFIG_FRAME_POINTER' Like:
#ifndefCONFIG_FRAME_POINTER
#define CONT ({ insn++; gotoselect_insn; })
#defineCONT_JMP({ insn++; gotoselect_insn; })
#else
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#defineCONT_JMP({ insn++; goto *jumptable[insn->code]; }) #endifThe reason this CONT and
CONT_JMPmacros are there because a combination of different gcc versions together with different cpus make branch predictor behave 'unpredictably'.
Quoted 11 lines
I've played with CONT and
CONT_JMPeither both doing direct goto or indirect goto and observed quite different performance characteristics from the interpreter.
What you see right now was the best tune I could get from a set of cpus I had to play with and compilers. If I did the same tuning today the outcome could have been different.
So I think it's totally fine to use above code. I think some cpus may actually see performance gains with certain versions of gcc.
The retpoline text increase is unfortunate but when retpoline is on other security knobs should be on too. In particularCONFIG_BPF_JIT_ALWAYS_ONshould be on as well. Which will remove interpreter from .text completely.
This would actually be contingent on RETPOLINE, not FRAME_POINTER.
(FRAME_POINTER was the other issue with the "optimize" attribute, which we're reverting so it'll no longer be a problem.)
So if you're not concerned about the retpoline text growth, it could be as simple as:
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#define CONT_JMP ({ insn++; goto *jumptable[insn->code]; })
Or, if you wanted to avoid the text growth, it could be:
#ifdef CONFIG_RETPOLINE
/*
* Avoid a 40% increase in function text size by getting GCC to generate a
* single retpoline jump instead of 160+.
*/
#define CONT ({ insn++; goto select_insn; })
#define CONT_JMP ({ insn++; goto select_insn; })
select_insn:
goto *jumptable[insn->code];
#else /* !CONFIG_RETPOLINE */
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#define CONT_JMP ({ insn++; goto *jumptable[insn->code]; }) #endif /* CONFIG_RETPOLINE */
But since this is legacy path, I think the first one is much nicer.
Also, JMP_TAIL_CALL has a "goto select_insn", is it ok to convert that to CONT?
Signature · 1 line
Josh
Re: BPF vs objtool again
Quoted 46 lines · Josh Poimboeuf
Quoted 25 lines · Alexei Starovoitov
Quoted 9 lines
For example:
#define GOTO ({ goto *jumptable[insn->code]; })
and then replace all 'goto
select_insn' with 'GOTO;'The problem is that with RETPOLINE=y, the function text size grows from 5k to 7k, because for each of the 160+ retpoline JMPs, GCC (stupidly) reloads the jump table register into a scratch register.
that would be a tiny change, right?
I'd rather go with that and gate it with 'ifdefCONFIG_FRAME_POINTER' Like:
#ifndefCONFIG_FRAME_POINTER
#define CONT ({ insn++; gotoselect_insn; })
#defineCONT_JMP({ insn++; gotoselect_insn; })
#else
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#defineCONT_JMP({ insn++; goto *jumptable[insn->code]; }) #endifThe reason this CONT and
CONT_JMPmacros are there because a combination of different gcc versions together with different cpus make branch predictor behave 'unpredictably'.Quoted 11 lines
I've played with CONT and
CONT_JMPeither both doing direct goto or indirect goto and observed quite different performance characteristics from the interpreter.
What you see right now was the best tune I could get from a set of cpus I had to play with and compilers. If I did the same tuning today the outcome could have been different.
So I think it's totally fine to use above code. I think some cpus may actually see performance gains with certain versions of gcc.
The retpoline text increase is unfortunate but when retpoline is on other security knobs should be on too. In particularCONFIG_BPF_JIT_ALWAYS_ONshould be on as well. Which will remove interpreter from .text completely.This would actually be contingent on RETPOLINE, not
FRAME_POINTER.(
FRAME_POINTERwas the other issue with the "optimize" attribute, which we're reverting so it'll no longer be a problem.)So if you're not concerned about the retpoline text growth, it could be as simple as:
Quoted 7 lines
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#defineCONT_JMP({ insn++; goto *jumptable[insn->code]; })Or, if you wanted to avoid the text growth, it could be:
#ifdef
CONFIG_RETPOLINE
I'm a bit lost. So objtool is fine with the asm when retpoline is on? Then pls do:
#if defined(CONFIG_RETPOLINE) || !defined(CONFIG_X86)
since there is no need to mess with other archs.
Quoted 16 lines
/*
* Avoid a 40% increase in function text size by getting GCC to generate a
* single retpoline jump instead of 160+.
*/
#define CONT ({ insn++; gotoselect_insn; })
#defineCONT_JMP({ insn++; gotoselect_insn; })
select_insn:goto *jumptable[insn->code];#else /* !CONFIG_RETPOLINE */
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#defineCONT_JMP({ insn++; goto *jumptable[insn->code]; }) #endif /*CONFIG_RETPOLINE*/
Quoted 5 lines
But since this is legacy path, I think the first one is much nicer.
Also,
JMP_TAIL_CALLhas a "gotoselect_insn", is it ok to convert that to CONT?
yep
Re: BPF vs objtool again
Quoted 12 lines · Alexei Starovoitov
Quoted 12 lines
This would actually be contingent on RETPOLINE, not
FRAME_POINTER.(
FRAME_POINTERwas the other issue with the "optimize" attribute, which we're reverting so it'll no longer be a problem.)So if you're not concerned about the retpoline text growth, it could be as simple as:
#define CONT ({ insn++; goto *jumptable[insn->code]; })
#defineCONT_JMP({ insn++; goto *jumptable[insn->code]; })
Quoted 5 lines
Quoted 3 lines
Or, if you wanted to avoid the text growth, it could be:
#ifdef
CONFIG_RETPOLINEI'm a bit lost. So objtool is fine with the asm when retpoline is on?
Yeah, it's confusing... this has been quite an adventure with GCC.
Objtool is fine with the RETPOLINE double goto. It's only the !RETPOLINE double goto which is the problem, because that triggers more GCC weirdness (see 3193c0836f20).
Quoted 4 lines
Then pls do:
#ifdefined(CONFIG_RETPOLINE)|| !defined(CONFIG_X86)since there is no need to mess with other archs.
Getting rid of select_insn altogether would make the code a lot simpler, but it's your call. I'll make a patch soon.
Signature · 1 line
Josh