Message ID | 1460044433-19282-9-git-send-email-sergey.fedorov@linaro.org (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Sergey Fedorov <sergey.fedorov@linaro.org> writes: > From: Sergey Fedorov <serge.fdrv@gmail.com> > > Ensure direct jump patching in AArch64 is atomic by using > atomic_read()/atomic_set() for code patching. > > Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> > Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> > --- > tcg/aarch64/tcg-target.inc.c | 14 +++++++++++++- > 1 file changed, 13 insertions(+), 1 deletion(-) > > diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c > index 0ed10a974121..15fdebec921f 100644 > --- a/tcg/aarch64/tcg-target.inc.c > +++ b/tcg/aarch64/tcg-target.inc.c > @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target) > *code_ptr = deposit32(*code_ptr, 0, 26, offset); > } > > +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, > + tcg_insn_unit *target) > +{ > + ptrdiff_t offset = target - code_ptr; > + tcg_insn_unit insn; > + assert(offset == sextract64(offset, 0, 26)); > + /* read instruction, mask away previous PC_REL26 parameter contents, > + set the proper offset, then write back the instruction. */ This comment could be moved from here and reloc_pc26 and made common for the two following functions. > + insn = atomic_read(code_ptr); > + atomic_set(code_ptr, deposit32(insn, 0, 26, offset)); > +} > + > static inline void reloc_pc19(tcg_insn_unit *code_ptr, tcg_insn_unit *target) > { > ptrdiff_t offset = target - code_ptr; > @@ -835,7 +847,7 @@ void aarch64_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr) > tcg_insn_unit *code_ptr = (tcg_insn_unit *)jmp_addr; > tcg_insn_unit *target = (tcg_insn_unit *)addr; > > - reloc_pc26(code_ptr, target); > + reloc_pc26_atomic(code_ptr, target); > flush_icache_range(jmp_addr, jmp_addr + 4); > } Otherwise: Reviewed-by: Alex Bennée <alex.bennee@linaro.org> -- Alex Bennée
On 04/20/2016 07:01 AM, Alex Bennée wrote: > > Sergey Fedorov <sergey.fedorov@linaro.org> writes: > >> From: Sergey Fedorov <serge.fdrv@gmail.com> >> >> Ensure direct jump patching in AArch64 is atomic by using >> atomic_read()/atomic_set() for code patching. >> >> Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> >> Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> >> --- >> tcg/aarch64/tcg-target.inc.c | 14 +++++++++++++- >> 1 file changed, 13 insertions(+), 1 deletion(-) >> >> diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c >> index 0ed10a974121..15fdebec921f 100644 >> --- a/tcg/aarch64/tcg-target.inc.c >> +++ b/tcg/aarch64/tcg-target.inc.c >> @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target) >> *code_ptr = deposit32(*code_ptr, 0, 26, offset); >> } >> >> +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, >> + tcg_insn_unit *target) >> +{ >> + ptrdiff_t offset = target - code_ptr; >> + tcg_insn_unit insn; >> + assert(offset == sextract64(offset, 0, 26)); >> + /* read instruction, mask away previous PC_REL26 parameter contents, >> + set the proper offset, then write back the instruction. */ > > This comment could be moved from here and reloc_pc26 and made common for > the two following functions. There's a significant amount of cleanup that ought to happen here, now that we're not re-translating TBs. I don't know if Sergey should be gated on that. r~
Richard Henderson <rth@twiddle.net> writes: > On 04/20/2016 07:01 AM, Alex Bennée wrote: >> >> Sergey Fedorov <sergey.fedorov@linaro.org> writes: >> >>> From: Sergey Fedorov <serge.fdrv@gmail.com> >>> >>> Ensure direct jump patching in AArch64 is atomic by using >>> atomic_read()/atomic_set() for code patching. >>> >>> Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> >>> Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> >>> --- >>> tcg/aarch64/tcg-target.inc.c | 14 +++++++++++++- >>> 1 file changed, 13 insertions(+), 1 deletion(-) >>> >>> diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c >>> index 0ed10a974121..15fdebec921f 100644 >>> --- a/tcg/aarch64/tcg-target.inc.c >>> +++ b/tcg/aarch64/tcg-target.inc.c >>> @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target) >>> *code_ptr = deposit32(*code_ptr, 0, 26, offset); >>> } >>> >>> +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, >>> + tcg_insn_unit *target) >>> +{ >>> + ptrdiff_t offset = target - code_ptr; >>> + tcg_insn_unit insn; >>> + assert(offset == sextract64(offset, 0, 26)); >>> + /* read instruction, mask away previous PC_REL26 parameter contents, >>> + set the proper offset, then write back the instruction. */ >> >> This comment could be moved from here and reloc_pc26 and made common for >> the two following functions. > > There's a significant amount of cleanup that ought to happen here, now that > we're not re-translating TBs. I don't know if Sergey should be gated > on that. Is this stuff already in the works? Otherwise we are trying to get pre-cursors to MTTCG into the code (once the tree re-opens) to keep the main diff down. This also is beneficial for linux-user stuff. > > > r~ -- Alex Bennée
On 20/04/16 18:08, Richard Henderson wrote: > On 04/20/2016 07:01 AM, Alex Bennée wrote: >> >> Sergey Fedorov <sergey.fedorov@linaro.org> writes: >> >>> From: Sergey Fedorov <serge.fdrv@gmail.com> >>> >>> Ensure direct jump patching in AArch64 is atomic by using >>> atomic_read()/atomic_set() for code patching. >>> >>> Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> >>> Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> >>> --- >>> tcg/aarch64/tcg-target.inc.c | 14 +++++++++++++- >>> 1 file changed, 13 insertions(+), 1 deletion(-) >>> >>> diff --git a/tcg/aarch64/tcg-target.inc.c >>> b/tcg/aarch64/tcg-target.inc.c >>> index 0ed10a974121..15fdebec921f 100644 >>> --- a/tcg/aarch64/tcg-target.inc.c >>> +++ b/tcg/aarch64/tcg-target.inc.c >>> @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit >>> *code_ptr, tcg_insn_unit *target) >>> *code_ptr = deposit32(*code_ptr, 0, 26, offset); >>> } >>> >>> +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, >>> + tcg_insn_unit *target) >>> +{ >>> + ptrdiff_t offset = target - code_ptr; >>> + tcg_insn_unit insn; >>> + assert(offset == sextract64(offset, 0, 26)); >>> + /* read instruction, mask away previous PC_REL26 parameter >>> contents, >>> + set the proper offset, then write back the instruction. */ >> >> This comment could be moved from here and reloc_pc26 and made common for >> the two following functions. > > There's a significant amount of cleanup that ought to happen here, now > that we're not re-translating TBs. I don't know if Sergey should be > gated on that. Do you mean I'd better avoid using stuff like reloc_pc26()? Kind regards, Sergey
On 04/20/2016 11:22 AM, Alex Bennée wrote: > > Richard Henderson <rth@twiddle.net> writes: > >> On 04/20/2016 07:01 AM, Alex Bennée wrote: >>> >>> Sergey Fedorov <sergey.fedorov@linaro.org> writes: >>> >>>> From: Sergey Fedorov <serge.fdrv@gmail.com> >>>> >>>> Ensure direct jump patching in AArch64 is atomic by using >>>> atomic_read()/atomic_set() for code patching. >>>> >>>> Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> >>>> Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> >>>> --- >>>> tcg/aarch64/tcg-target.inc.c | 14 +++++++++++++- >>>> 1 file changed, 13 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c >>>> index 0ed10a974121..15fdebec921f 100644 >>>> --- a/tcg/aarch64/tcg-target.inc.c >>>> +++ b/tcg/aarch64/tcg-target.inc.c >>>> @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target) >>>> *code_ptr = deposit32(*code_ptr, 0, 26, offset); >>>> } >>>> >>>> +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, >>>> + tcg_insn_unit *target) >>>> +{ >>>> + ptrdiff_t offset = target - code_ptr; >>>> + tcg_insn_unit insn; >>>> + assert(offset == sextract64(offset, 0, 26)); >>>> + /* read instruction, mask away previous PC_REL26 parameter contents, >>>> + set the proper offset, then write back the instruction. */ >>> >>> This comment could be moved from here and reloc_pc26 and made common for >>> the two following functions. >> >> There's a significant amount of cleanup that ought to happen here, now that >> we're not re-translating TBs. I don't know if Sergey should be gated >> on that. > > Is this stuff already in the works? Otherwise we are trying to get > pre-cursors to MTTCG into the code (once the tree re-opens) to keep the > main diff down. This also is beneficial for linux-user stuff. We're talking past one another. No, it's not yet in the works. I'm saying that this patch set should not wait for it. Thus I don't care if what he adds here is a little messy; we'll clean it all up at once later. Thus don't bother refactoring reloc_pc26 and reloc_pc26_atomic. r~
Richard Henderson <rth@twiddle.net> writes: > On 04/20/2016 11:22 AM, Alex Bennée wrote: >> >> Richard Henderson <rth@twiddle.net> writes: >> >>> On 04/20/2016 07:01 AM, Alex Bennée wrote: >>>> >>>> Sergey Fedorov <sergey.fedorov@linaro.org> writes: >>>> >>>>> From: Sergey Fedorov <serge.fdrv@gmail.com> >>>>> >>>>> Ensure direct jump patching in AArch64 is atomic by using >>>>> atomic_read()/atomic_set() for code patching. >>>>> >>>>> Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> >>>>> Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> >>>>> --- >>>>> tcg/aarch64/tcg-target.inc.c | 14 +++++++++++++- >>>>> 1 file changed, 13 insertions(+), 1 deletion(-) >>>>> >>>>> diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c >>>>> index 0ed10a974121..15fdebec921f 100644 >>>>> --- a/tcg/aarch64/tcg-target.inc.c >>>>> +++ b/tcg/aarch64/tcg-target.inc.c >>>>> @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target) >>>>> *code_ptr = deposit32(*code_ptr, 0, 26, offset); >>>>> } >>>>> >>>>> +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, >>>>> + tcg_insn_unit *target) >>>>> +{ >>>>> + ptrdiff_t offset = target - code_ptr; >>>>> + tcg_insn_unit insn; >>>>> + assert(offset == sextract64(offset, 0, 26)); >>>>> + /* read instruction, mask away previous PC_REL26 parameter contents, >>>>> + set the proper offset, then write back the instruction. */ >>>> >>>> This comment could be moved from here and reloc_pc26 and made common for >>>> the two following functions. >>> >>> There's a significant amount of cleanup that ought to happen here, now that >>> we're not re-translating TBs. I don't know if Sergey should be gated >>> on that. >> >> Is this stuff already in the works? Otherwise we are trying to get >> pre-cursors to MTTCG into the code (once the tree re-opens) to keep the >> main diff down. This also is beneficial for linux-user stuff. > > We're talking past one another. > > No, it's not yet in the works. I'm saying that this patch set should not wait > for it. Thus I don't care if what he adds here is a little messy; we'll clean > it all up at once later. Thus don't bother refactoring reloc_pc26 and > reloc_pc26_atomic. Ahh OK, cool ;-) > > > r~ -- Alex Bennée
On 07/04/16 18:53, Sergey Fedorov wrote: > From: Sergey Fedorov <serge.fdrv@gmail.com> > > Ensure direct jump patching in AArch64 is atomic by using > atomic_read()/atomic_set() for code patching. > > Signed-off-by: Sergey Fedorov <serge.fdrv@gmail.com> > Signed-off-by: Sergey Fedorov <sergey.fedorov@linaro.org> > --- > tcg/aarch64/tcg-target.inc.c | 14 +++++++++++++- > 1 file changed, 13 insertions(+), 1 deletion(-) > > diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c > index 0ed10a974121..15fdebec921f 100644 > --- a/tcg/aarch64/tcg-target.inc.c > +++ b/tcg/aarch64/tcg-target.inc.c > @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target) > *code_ptr = deposit32(*code_ptr, 0, 26, offset); > } > > +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, > + tcg_insn_unit *target) > +{ > + ptrdiff_t offset = target - code_ptr; > + tcg_insn_unit insn; > + assert(offset == sextract64(offset, 0, 26)); I'd better use tcg_debug_assert() here as in this patch: http://patchwork.ozlabs.org/patch/613020/ Kind regards, Sergey > + /* read instruction, mask away previous PC_REL26 parameter contents, > + set the proper offset, then write back the instruction. */ > + insn = atomic_read(code_ptr); > + atomic_set(code_ptr, deposit32(insn, 0, 26, offset)); > +} > + > static inline void reloc_pc19(tcg_insn_unit *code_ptr, tcg_insn_unit *target) > { > ptrdiff_t offset = target - code_ptr; > @@ -835,7 +847,7 @@ void aarch64_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr) > tcg_insn_unit *code_ptr = (tcg_insn_unit *)jmp_addr; > tcg_insn_unit *target = (tcg_insn_unit *)addr; > > - reloc_pc26(code_ptr, target); > + reloc_pc26_atomic(code_ptr, target); > flush_icache_range(jmp_addr, jmp_addr + 4); > } >
diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c index 0ed10a974121..15fdebec921f 100644 --- a/tcg/aarch64/tcg-target.inc.c +++ b/tcg/aarch64/tcg-target.inc.c @@ -73,6 +73,18 @@ static inline void reloc_pc26(tcg_insn_unit *code_ptr, tcg_insn_unit *target) *code_ptr = deposit32(*code_ptr, 0, 26, offset); } +static inline void reloc_pc26_atomic(tcg_insn_unit *code_ptr, + tcg_insn_unit *target) +{ + ptrdiff_t offset = target - code_ptr; + tcg_insn_unit insn; + assert(offset == sextract64(offset, 0, 26)); + /* read instruction, mask away previous PC_REL26 parameter contents, + set the proper offset, then write back the instruction. */ + insn = atomic_read(code_ptr); + atomic_set(code_ptr, deposit32(insn, 0, 26, offset)); +} + static inline void reloc_pc19(tcg_insn_unit *code_ptr, tcg_insn_unit *target) { ptrdiff_t offset = target - code_ptr; @@ -835,7 +847,7 @@ void aarch64_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr) tcg_insn_unit *code_ptr = (tcg_insn_unit *)jmp_addr; tcg_insn_unit *target = (tcg_insn_unit *)addr; - reloc_pc26(code_ptr, target); + reloc_pc26_atomic(code_ptr, target); flush_icache_range(jmp_addr, jmp_addr + 4); }