Hi,
I needed to move the edit from c1_LIRGenerator_<cpu>.cpp to
the shared file after "8201543: Modularize C1 GC barriers"
New webrev:
http://cr.openjdk.java.net/~goetz/wr18/8201593-lenInAIOOB/06/
@Stuart
Thanks for testing!
> so as to accommodate the array pointer you are pushing onto the stack?
Yes, what you are pointing out seems to be wrong, I changed it to '2'.
Best regards,
Goetz.
> -----Original Message-----
> From: Stuart Monteith [mailto:stuart.monteith at linaro.org]
> Sent: Freitag, 27. April 2018 16:37
> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
> Cc: hotspot-compiler-dev at openjdk.java.net; aarch64-port-
> dev at openjdk.java.net; hotspot-runtime-dev at openjdk.java.net; aarch32-
> port-dev at openjdk.java.net
> Subject: Re: RFR(M): 8201593: Print array length in
> ArrayIndexOutOfBoundsException.
>
> Hi,
> JTregs hasn't flagged any issues, so it should be ok.
>
> Regarding the 32-bit arm code, in "void
> RangeCheckStub::emit_code(LIR_Assembler* ce)" should:
> ce->verify_reserved_argument_area_size(1);
> be
> ce->verify_reserved_argument_area_size(2);
>
> so as to accommodate the array pointer you are pushing onto the stack?
>
> I've not tested 32-bit arm.
>
>
> BR,
> Stuart
>
> On 26 April 2018 at 15:31, Stuart Monteith <stuart.monteith at linaro.org>
> wrote:
> > Thanks, I'm happy with that.
> >
> > The registers have a clean path to call_RT - r22 and r23 aren't used
> > inbetween. They are an arbitrary choice - c_rarg0 and c_rarg1 were
> > always going to cause problems. If _array->as_pointer_register()
> > and/or _index->as_register() or _index->as_jint() were the registers
> > we were using as parameters there would be trouble. However, with
> > pd_last_allocatable_cpu_reg = 16, that shouldn't happen with r22/23,
> > or indeed anything else in the range r17 to r28.
> >
> > I'm going to run all of JTRegs and seem what that produces now.
> >
> > BR,
> > Stuart
> >
> >
> >
> >
> > On 26 April 2018 at 15:14, Lindenmaier, Goetz
> <goetz.lindenmaier at sap.com> wrote:
> >> Hi Stuart,
> >>
> >> thanks for fixing this! Webrev with your changes:
> >> http://cr.openjdk.java.net/~goetz/wr18/8201593-lenInAIOOB/05/
> >>
> >>> There is the possibility of overwriting live values though, aren't
> >>> there? The registers are saved by call_RT. Should I be concerned about
> >>> deopt and debugging going wrong? Furthermore, won't there be issues
> in
> >>> exception handlers?
> >> As I understand, this just has to survive the far_call.
> >> The call_RT in c1_Runtime then moves it into the
> >> proper argument registers. This is just the handling of an
> >> exception, and in these few instructions no java code is
> >> executed, no safepoint is passed, so this should be fine.
> >>
> >> incremental diff:
> >> iff -r 874f2b999ff6 src/hotspot/cpu/aarch64/c1_CodeStubs_aarch64.cpp
> >> --- a/src/hotspot/cpu/aarch64/c1_CodeStubs_aarch64.cpp Mon Apr 16
> 15:17:20 2018 +0200
> >> +++ b/src/hotspot/cpu/aarch64/c1_CodeStubs_aarch64.cpp Thu Apr 26
> 15:55:18 2018 +0200
> >> @@ -75,16 +75,16 @@
> >> }
> >>
> >> if (_index->is_cpu_register()) {
> >> - __ mov(rscratch1, _index->as_register());
> >> + __ mov(r22, _index->as_register());
> >> } else {
> >> - __ mov(rscratch1, _index->as_jint());
> >> + __ mov(r22, _index->as_jint());
> >> }
> >> Runtime1::StubID stub_id;
> >> if (_throw_index_out_of_bounds_exception) {
> >> stub_id = Runtime1::throw_index_exception_id;
> >> } else {
> >> assert(_array != NULL, "sanity");
> >> - __ mov(rscratch2, _array->as_pointer_register());
> >> + __ mov(r23, _array->as_pointer_register());
> >> stub_id = Runtime1::throw_range_check_failed_id;
> >> }
> >> __ far_call(RuntimeAddress(Runtime1::entry_for(stub_id)), NULL,
> rscratch2);
> >> diff -r 874f2b999ff6 src/hotspot/cpu/aarch64/c1_Runtime1_aarch64.cpp
> >> --- a/src/hotspot/cpu/aarch64/c1_Runtime1_aarch64.cpp Mon Apr 16
> 15:17:20 2018 +0200
> >> +++ b/src/hotspot/cpu/aarch64/c1_Runtime1_aarch64.cpp Thu Apr 26
> 15:55:18 2018 +0200
> >> @@ -327,7 +327,7 @@
> >>
> >>
> >> // target: the entry point of the method that creates and posts the
> exception oop
> >> -// has_argument: true if the exception needs an argument (passed in
> rscratch1)
> >> +// has_argument: true if the exception needs arguments (passed in r22
> and r23)
> >>
> >> OopMapSet* Runtime1::generate_exception_throw(StubAssembler*
> sasm, address target, bool has_argument) {
> >> // make a frame and preserve the caller's caller-save registers
> >> @@ -336,7 +336,7 @@
> >> if (!has_argument) {
> >> call_offset = __ call_RT(noreg, noreg, target);
> >> } else {
> >> - call_offset = __ call_RT(noreg, noreg, target, rscratch1, rscratch2);
> >> + call_offset = __ call_RT(noreg, noreg, target, r22, r23);
> >> }
> >> OopMapSet* oop_maps = new OopMapSet();
> >> oop_maps->add_gc_map(call_offset, oop_map);
> >>
> >> Best regards,
> >> Goetz.
> >>
> >>
> >>> -----Original Message-----
> >>> From: Stuart Monteith [mailto:stuart.monteith at linaro.org]
> >>> Sent: Donnerstag, 26. April 2018 12:52
> >>> To: Andrew Haley <aph at redhat.com>
> >>> Cc: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; hotspot-
> compiler-
> >>> dev at openjdk.java.net; aarch64-port-dev at openjdk.java.net; hotspot-
> >>> runtime-dev at openjdk.java.net; aarch32-port-dev at openjdk.java.net
> >>> Subject: Re: RFR(M): 8201593: Print array length in
> >>> ArrayIndexOutOfBoundsException.
> >>>
> >>> Hi,
> >>> Using c_rarg1 and c_rarg2 instead of rscratch1 and overwriting
> >>> rscratch2 causes a SIGSEGV.
> >>> Using r22 and r23 instead, the test ran successfully.
> >>>
> >>> In c1_CodeStubs_aarch64.cpp
> >>> :
> >>> 77 if (_index->is_cpu_register()) {
> >>> 78 __ mov(r22, _index->as_register());
> >>> 79 } else {
> >>> 80 __ mov(r22, _index->as_jint());
> >>> 81 }
> >>> 82 Runtime1::StubID stub_id;
> >>> 83 if (_throw_index_out_of_bounds_exception) {
> >>> 84 stub_id = Runtime1::throw_index_exception_id;
> >>> 85 } else {
> >>> 86 assert(_array != NULL, "sanity");
> >>> 87 __ mov(r23, _array->as_pointer_register());
> >>> 88 stub_id = Runtime1::throw_range_check_failed_id;
> >>> 89 }
> >>>
> >>> in c1_Runtime_aarch64.cpp:
> >>>
> >>> 336 if (!has_argument) {
> >>> 337 call_offset = __ call_RT(noreg, noreg, target);
> >>> 338 } else {
> >>> 339 call_offset = __ call_RT(noreg, noreg, target, r22, r23);
> >>> 340 }
> >>>
> >>> There is the possibility of overwriting live values though, aren't
> >>> there? The registers are saved by call_RT. Should I be concerned about
> >>> deopt and debugging going wrong? Furthermore, won't there be issues
> in
> >>> exception handlers?
> >>>
> >>> BR,
> >>> Stuart
> >>>
> >>>
> >>> On 25 April 2018 at 16:49, Stuart Monteith <stuart.monteith at linaro.org>
> >>> wrote:
> >>> > Indeed - and that is what I am seeing. Usually no parameters are being
> >>> > called with this pattern, or rscratch1, with the temporary variable
> >>> > being changed to use rscratch2 in such circumstances.
> >>> > I'll try c_rarg1 and c_rarg2 - they should pass straight through,if I
> >>> > interpret the code correcting.
> >>> >
> >>> > On 25 April 2018 at 16:26, Andrew Haley <aph at redhat.com> wrote:
> >>> >> On 04/25/2018 04:00 PM, Stuart Monteith wrote:
> >>> >>> I'm not quite sure to solve this yet - we'll need to use the stack in
> >>> >>> some safe way.
> >>> >>
> >>> >> It's not a great idea to pass arguments in rscratch1 or rscratch2. These
> >>> >> registers are for use in macros and should be treated as volatile.
> Given
> >>> >> that you're throwing an exception, registers will be clobbered
> anyway.
> >>> >>
> >>> >> --
> >>> >> Andrew Haley
> >>> >> Java Platform Lead Engineer
> >>> >> Red Hat UK Ltd. <https://www.redhat.com>
> >>> >> EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671