Confused... Does this replace the previous series (which at first glance looks pretty much the same) ?
I already applied the previous one in my local tree, but didn't ask to pull yet. On 09/04, Denys Vlasenko wrote: > > After adding these, it's clear we have some awkward choices there. > Some valid instructions are prohibited from uprobing while > several invalid ones are allowed. > > Hopefully future edits to the good-opcode tables will fix wrong bits > or explain why those bits are not wrong. > > No actual code changes. > > Signed-off-by: Denys Vlasenko <[email protected]> > CC: Jim Keniston <[email protected]> > CC: Masami Hiramatsu <[email protected]> > CC: Srikar Dronamraju <[email protected]> > CC: Ingo Molnar <[email protected]> > CC: Oleg Nesterov <[email protected]> > --- > arch/x86/kernel/uprobes.c | 153 > ++++++++++++++++++++++++++++++++++++++++------ > 1 file changed, 134 insertions(+), 19 deletions(-) > > diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c > index 5d1cbfe..9ba2fb2 100644 > --- a/arch/x86/kernel/uprobes.c > +++ b/arch/x86/kernel/uprobes.c > @@ -66,6 +66,49 @@ > * Good-instruction tables for 32-bit apps. This is non-const and volatile > * to keep gcc from statically optimizing it out, as variable_test_bit makes > * some versions of gcc to think only *(unsigned long*) is used. > + * > + * Prefixes. Most marked as "bad", but it doesn't matter, since insn decoder > + * won't report *prefixes* as OPCODE1(insn). > + * 0f - 2-byte opcode prefix > + * 26,2e,36,3e - es:/cs:/ss:/ds: > + * 64 - fs: (marked as "good", why?) > + * 65 - gs: (marked as "good", why?) > + * 66 - operand-size prefix > + * 67 - address-size prefix > + * f0 - lock prefix > + * f2 - repnz (marked as "good", why?) > + * f3 - rep/repz (marked as "good", why?) > + * > + * Opcodes we'll probably never support: > + * 6c-6f - ins,outs. SEGVs if used in userspace > + * e4-e7 - in,out imm. SEGVs if used in userspace > + * ec-ef - in,out acc. SEGVs if used in userspace > + * cc - int3. SIGTRAP if used in userspace > + * ce - into. Not used in userspace - no kernel support to make it useful. > SEGVs > + * (why we support bound (62) then? it's similar, and similarly unused...) > + * f1 - int1. SIGTRAP if used in userspace > + * f4 - hlt. SEGVs if used in userspace > + * fa - cli. SEGVs if used in userspace > + * fb - sti. SEGVs if used in userspace > + * > + * Opcodes which need some work to be supported: > + * 07,17,1f - pop es/ss/ds > + * Normally not used in userspace, but would execute if used. > + * Can cause GP or stack exception if tries to load wrong segment > descriptor. > + * We hesitate to run them under single step since kernel's handling > + * of userspace single-stepping (TF flag) is fragile. > + * We can easily refuse to support push es/cs/ss/ds (06/0e/16/1e) > + * on the same grounds that they are never used. > + * cd - int N. > + * Used by userspace for "int 80" syscall entry. (Other "int N" > + * cause GP -> SEGV since their IDT gates don't allow calls from CPL 3). > + * Not supported since kernel's handling of userspace single-stepping > + * (TF flag) is fragile. > + * cf - iret. Normally not used in userspace. Doesn't SEGV unless arguments > are bad > + * > + * Opcodes which can be enabled right away: > + * 63 - arpl. This insn has no unusual exceptions (it's basically an arith > op). > + * d6 - salc. Undocumented "sign-extend carry flag to AL" insn > */ > #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) > static volatile u32 good_insns_32[256 / 32] = { > @@ -94,7 +137,55 @@ static volatile u32 good_insns_32[256 / 32] = { > #define good_insns_32 NULL > #endif > > -/* Good-instruction tables for 64-bit apps */ > +/* Good-instruction tables for 64-bit apps. > + * > + * Prefixes. Most marked as "bad", but it doesn't matter, since insn decoder > + * won't report *prefixes* as OPCODE1(insn). > + * 0f - 2-byte opcode prefix > + * 26,2e,36,3e - es:/cs:/ss:/ds: > + * 40-4f - rex prefixes > + * 64 - fs: (marked as "good", why?) > + * 65 - gs: (marked as "good", why?) > + * 66 - operand-size prefix > + * 67 - address-size prefix > + * f0 - lock prefix > + * f2 - repnz (marked as "good", why?) > + * f3 - rep/repz (marked as "good", why?) > + * > + * Genuinely invalid opcodes: > + * 06,07 - formerly push/pop es > + * 0e - formerly push cs > + * 16,17 - formerly push/pop ss > + * 1e,1f - formerly push/pop ds > + * 27,2f,37,3f - formerly daa/das/aaa/aas > + * 60,61 - formerly pusha/popa > + * 62 - formerly bound. EVEX prefix for AVX512 > + * 82 - formerly redundant encoding of Group1 > + * 9a - formerly call seg:ofs (marked as "supported"???) > + * c4,c5 - formerly les/lds. VEX prefixes for AVX > + * ce - formerly into > + * d4,d5 - formerly aam/aad > + * d6 - formerly undocumented salc > + * ea - formerly jmp seg:ofs (marked as "supported"???) > + * > + * Opcodes we'll probably never support: > + * 6c-6f - ins,outs. SEGVs if used in userspace > + * e4-e7 - in,out imm. SEGVs if used in userspace > + * ec-ef - in,out acc. SEGVs if used in userspace > + * cc - int3. SIGTRAP if used in userspace > + * f1 - int1. SIGTRAP if used in userspace > + * f4 - hlt. SEGVs if used in userspace > + * fa - cli. SEGVs if used in userspace > + * fb - sti. SEGVs if used in userspace > + * > + * Opcodes which need some work to be supported: > + * cd - int N. > + * Used by userspace for "int 80" syscall entry. (Other "int N" > + * cause GP -> SEGV since their IDT gates don't allow calls from CPL 3). > + * Not supported since kernel's handling of userspace single-stepping > + * (TF flag) is fragile. > + * cf - iret. Normally not used in userspace. Doesn't SEGV unless arguments > are bad > + */ > #if defined(CONFIG_X86_64) > static volatile u32 good_insns_64[256 / 32] = { > /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ > @@ -122,7 +213,48 @@ static volatile u32 good_insns_64[256 / 32] = { > #define good_insns_64 NULL > #endif > > -/* Using this for both 64-bit and 32-bit apps */ > +/* Using this for both 64-bit and 32-bit apps. > + * Opcodes we don't support: > + * 0f 00 - SLDT/STR/LLDT/LTR/VERR/VERW/-/- group. System insns > + * 0f 01 - SGDT/SIDT/LGDT/LIDT/SMSW/-/LMSW/INVLPG group. > + * Also encodes tons of other system insns if mod=11. > + * Some are in fact non-system: xend, xtest, rdtscp, maybe more > + * 0f 02 - lar (why? should be safe, it throws no exceptipons) > + * 0f 03 - lsl (why? should be safe, it throws no exceptipons) > + * 0f 04 - undefined > + * 0f 05 - syscall > + * 0f 06 - clts (CPL0 insn) > + * 0f 07 - sysret > + * 0f 08 - invd (CPL0 insn) > + * 0f 09 - wbinvd (CPL0 insn) > + * 0f 0a - undefined > + * 0f 0b - ud2 > + * 0f 0c - undefined > + * 0f 0d - prefetchFOO (amd prefetch insns) > + * 0f 18 - prefetchBAR (intel prefetch insns) > + * 0f 24 - mov from test regs (perhaps entire 20-27 area can be disabled > (special reg ops)) > + * 0f 25 - undefined > + * 0f 26 - mov to test regs > + * 0f 27 - undefined > + * 0f 30 - wrmsr (CPL0 insn) > + * 0f 34 - sysenter > + * 0f 35 - sysexit > + * 0f 36 - undefined > + * 0f 37 - getsec > + * 0f 38-3f - 3-byte opcodes (why?? all look safe) > + * 0f 78 - vmread > + * 0f 79 - vmwrite > + * 0f 7a - undefined > + * 0f 7b - undefined > + * 0f 7c - undefined > + * 0f 7d - undefined > + * 0f a6 - undefined > + * 0f a7 - undefined > + * 0f b8 - popcnt (why?? it's an ordinary ALU op) > + * 0f d0 - undefined > + * 0f f0 - lddqu (why?? it's an ordinary vector load op) > + * 0f ff - undefined > + */ > static volatile u32 good_2byte_insns[256 / 32] = { > /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ > /* ---------------------------------------------- */ > @@ -148,23 +280,6 @@ static volatile u32 good_2byte_insns[256 / 32] = { > #undef W > > /* > - * opcodes we'll probably never support: > - * > - * 6c-6d, e4-e5, ec-ed - in > - * 6e-6f, e6-e7, ee-ef - out > - * cc, cd - int3, int > - * cf - iret > - * d6 - illegal instruction > - * f1 - int1/icebp > - * f4 - hlt > - * fa, fb - cli, sti > - * 0f - lar, lsl, syscall, clts, sysret, sysenter, sysexit, invd, wbinvd, > ud2 > - * > - * invalid opcodes in 64-bit mode: > - * > - * 06, 0e, 16, 1e, 27, 2f, 37, 3f, 60-62, 82, c4-c5, d4-d5 > - * 63 - we support this opcode in x86_64 but not in i386. > - * > * opcodes we may need to refine support for: > * > * 0f - 2-byte instructions: For many of these instructions, the validity > -- > 1.8.1.4 > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
