Arm semihosting v2 implementation isn't complete
Tamar Christina
Tamar.Christina@arm.com
Wed Dec 19 12:21:00 GMT 2018
Hi Alex,
The 12/18/2018 21:08, Alexander Fedotov wrote:
> Disassembling rdimon-v2m libraries for armv8 I see following:
> svc 0x0000f000
>
> while that document states immediate 0xf000 must be used with HLT only.
>
Yes, that's what your patch fixes, the change I mentioned was because it only needs to
use hlt 0xf000 for the -v2m and not the non -v2m versions which are also version 2.
newlib no longer builds v1 by default so your original patch with just #defined (SEMIHOST_V2)
is always true and semihosting v2 requires support for hlt on earlier architectures than Armv8 as well.
However I'd like to put this on hold till January as there is something I need to check with
a colleague whom has already gone on leave for the holidays.
Kind Regards,
Tamar
> If this v2m implementation assumes to use SVC and BKPT instructions with new immed values - should I create "pure" rdimon-v2 version then with HLT only ?
>
> Alex
>
>
> On Tue, 18 Dec 2018 at 19:54, Tamar Christina <Tamar.Christina@arm.com<mailto:Tamar.Christina@arm.com>> wrote:
> Hi Alexander,
>
> #ifdef THUMB_VXM
> #define AngelSWIInsn "bkpt"
> #define AngelSWIAsm bkpt
> +#elif defined (SEMIHOST_V2) && (__ARM_ARCH == 8)
>
> You'll want to use the same condition as the reason number setting code above,
>
> #if defined (SEMIHOST_V2) && defined (SEMIHOST_V2_MIXED_MODE)
>
> As we only want to use hlt in the mixed mode and not always.
> The Armv8 check isn't needed due to the note in the specification
> "Note: This requirement includes supporting the HLT encodings on ARMv7 and earlier processors, even though HLT
> is only defined as an instruction in ARMv8."
>
> So we just generate all the multilibs and leave it up to the implementations to support it or not.
>
> Thanks,
> Tamar
>
> + #define AngelSWIInsn "hlt"
> + #define AngelSWIAsm hlt
> #else
> #define AngelSWIInsn "swi"
> #define AngelSWIAsm swi
> --
>
>
> The 12/18/2018 15:09, Alexander Fedotov wrote:
> > I suggest following patch:
> >
> > On Tue, Dec 18, 2018 at 5:12 PM Alexander Fedotov <alfedotov@gmail.com<mailto:alfedotov@gmail.com>> wrote:
> > >
> > > Hi Tamar
> > >
> > > I see HLT only in comments, not in defines:
> > >
> > > #ifdef THUMB_VXM
> > > #define AngelSWIInsn "bkpt"
> > > #define AngelSWIAsm bkpt
> > > #else
> > > #define AngelSWIInsn "swi"
> > > #define AngelSWIAsm swi
> > > #endif
> > >
> > >
> > > On Tue, Dec 18, 2018 at 4:50 PM Tamar Christina <Tamar.Christina@arm.com<mailto:Tamar.Christina@arm.com>> wrote:
> > > >
> > > > Hi Alex
> > > >
> > > > > -----Original Message-----
> > > > > From: newlib-owner@sourceware.org<mailto:newlib-owner@sourceware.org> <newlib-owner@sourceware.org<mailto:newlib-owner@sourceware.org>> On
> > > > > Behalf Of Alexander Fedotov
> > > > > Sent: Tuesday, December 18, 2018 13:08
> > > > > To: Newlib <newlib@sourceware.org<mailto:newlib@sourceware.org>>
> > > > > Subject: Arm semihosting v2 implementation isn't complete
> > > > >
> > > > > Hi all
> > > > >
> > > > > I see that number of commits were dedicated to changes for Arm
> > > > > Semihosting v2 specification that states:
> > > > > ARM encourages semihosting callers to implement support for trapping using
> > > > > HLT on A32 and T32 as a configurable option. ARM strongly discourages
> > > > > semihosting callers from mixing the HLT and SVC mechanisms within the
> > > > > same executable.
> > > > >
> > > > > At the moment there is no HLT instruction used in libgloss/arm.
> > > >
> > > > There is, see swi.h, we don't support mixing of HLT and SVC (as mentioned in the specification bit you quoted) so they are in different libraries.
> > > > HLT is used when you used the MIXED_MODE spec files. E.g. aprofile-validation-v2m.specs will use HLT and aprofile-validation.specs uses SVC.
> > > >
> > > > Regards,
> > > > Tamar
> > > >
> > > > >
> > > > > So the question is: is there any reason to not have HLT right now ?
> > > > >
> > > > > Alex
> > >
> > >
> > >
> > > --
> > > Best regards,
> > > AF
> >
> >
> >
> > --
> > Best regards,
> > AF
>
>
>
> --
> --
> Best regards,
> AF
--
More information about the Newlib
mailing list