| Message ID | 20200918144327.7qgh64f57myg6njs@arm.com |
|---|---|
| State | New |
| Headers | show |
| Series |
|
| Related | show |
> -----Original Message----- > From: Newlib <newlib-bounces@sourceware.org> On Behalf Of Alex Coplan > Sent: 18 September 2020 15:43 > To: newlib@sourceware.org > Cc: Richard Earnshaw <Richard.Earnshaw@arm.com> > Subject: [PATCH] libgloss/aarch64: Add support for Armv8-R AArch64 > > Hello, > > This patch adds support for Armv8-R AArch64. We update the AArch64 CPU > boot code to work on Armv8-R if __ARM_ARCH_PROFILE is set to 'R'. > Armv8-R AArch64 has no EL3, so we don't set vbar_el3, and adjust the > code to set up the MPU for Armv8-R. > > We also add a specs file for use with Armv8-R AArch64 models. > > Testing: > * Ran AArch64 GCC testsuite using --with-arch=armv8-r together with the > new specs file and boot code in an Armv8-R AArch64 model. > * Ran newlib testsuite, no regressions. > > OK for master? > FWIW this looks good to me but I'm not a maintainer. Thanks, Kyrill > Thanks, > Alex > > --- > > 2020-09-18 Alex Coplan <alex.coplan@arm.com> > > libgloss/ChangeLog: > > * aarch64/Makefile.in: Install new specs file. > * aarch64/cpu-init/Makefile.in: Also build boot code for Armv8-R. > * aarch64/cpu-init/rdimon-aem-el3.S: Add support for Armv8-R. > * aarch64/elf-aem-v8-r.specs: New.
On 18/09/2020 15:43, Alex Coplan wrote: > Hello, > > This patch adds support for Armv8-R AArch64. We update the AArch64 CPU > boot code to work on Armv8-R if __ARM_ARCH_PROFILE is set to 'R'. > Armv8-R AArch64 has no EL3, so we don't set vbar_el3, and adjust the > code to set up the MPU for Armv8-R. > > We also add a specs file for use with Armv8-R AArch64 models. > > Testing: > * Ran AArch64 GCC testsuite using --with-arch=armv8-r together with the > new specs file and boot code in an Armv8-R AArch64 model. > * Ran newlib testsuite, no regressions. > > OK for master? > > Thanks, > Alex > > --- > > 2020-09-18 Alex Coplan <alex.coplan@arm.com> > > libgloss/ChangeLog: > > * aarch64/Makefile.in: Install new specs file. > * aarch64/cpu-init/Makefile.in: Also build boot code for Armv8-R. > * aarch64/cpu-init/rdimon-aem-el3.S: Add support for Armv8-R. > * aarch64/elf-aem-v8-r.specs: New. > The main problem I see with this is that it assumes that the compiler being used is *not* configured for a default CPU architecture of ARMv8-r. If it is, then the normal rdimon-aem-el3.o file will in fact become the same as the v8-r version, which is not really what we want. Perhaps, rather than hacking around the __ARM_ARCH_PROFILE setting you should just set things up to build the file twice, but with a different conventional pre-define set each time. R.
Hi Richard, On 29/09/2020 15:06, Richard Earnshaw (lists) wrote: > On 18/09/2020 15:43, Alex Coplan wrote: > > Hello, > > > > This patch adds support for Armv8-R AArch64. We update the AArch64 CPU > > boot code to work on Armv8-R if __ARM_ARCH_PROFILE is set to 'R'. > > Armv8-R AArch64 has no EL3, so we don't set vbar_el3, and adjust the > > code to set up the MPU for Armv8-R. > > > > We also add a specs file for use with Armv8-R AArch64 models. > > > > Testing: > > * Ran AArch64 GCC testsuite using --with-arch=armv8-r together with the > > new specs file and boot code in an Armv8-R AArch64 model. > > * Ran newlib testsuite, no regressions. > > > > OK for master? > > > > Thanks, > > Alex > > > > --- > > > > 2020-09-18 Alex Coplan <alex.coplan@arm.com> > > > > libgloss/ChangeLog: > > > > * aarch64/Makefile.in: Install new specs file. > > * aarch64/cpu-init/Makefile.in: Also build boot code for Armv8-R. > > * aarch64/cpu-init/rdimon-aem-el3.S: Add support for Armv8-R. > > * aarch64/elf-aem-v8-r.specs: New. > > > > The main problem I see with this is that it assumes that the compiler > being used is *not* configured for a default CPU architecture of > ARMv8-r. If it is, then the normal rdimon-aem-el3.o file will in fact > become the same as the v8-r version, which is not really what we want. Good catch. Thanks for the review. > > Perhaps, rather than hacking around the __ARM_ARCH_PROFILE setting you > should just set things up to build the file twice, but with a different > conventional pre-define set each time. > > R. Sounds sensible. Is the revised patch (attached) OK? Thanks, Alex diff --git a/libgloss/aarch64/Makefile.in b/libgloss/aarch64/Makefile.in index 546557e3c..b8c2ca43c 100644 --- a/libgloss/aarch64/Makefile.in +++ b/libgloss/aarch64/Makefile.in @@ -85,7 +85,8 @@ RDIMON_BSP = librdimon${${MULTILIBNAME}}.a RDIMON_OBJS = $(patsubst %,rdimon-%,$(OBJS)) RDIMON_SCRIPTS = rdimon${${MULTILIBNAME}}.specs \ aem-ve${${MULTILIBNAME}}.specs \ - aem-validation${${MULTILIBNAME}}.specs + aem-validation${${MULTILIBNAME}}.specs \ + aem-v8-r${${MULTILIBNAME}}.specs RDIMON_INSTALL = install-rdimon CFLAGS = -g diff --git a/libgloss/aarch64/cpu-init/Makefile.in b/libgloss/aarch64/cpu-init/Makefile.in index 08aa1760f..c96f4ee0e 100644 --- a/libgloss/aarch64/cpu-init/Makefile.in +++ b/libgloss/aarch64/cpu-init/Makefile.in @@ -62,7 +62,7 @@ AR = @AR@ LD = @LD@ RANLIB = @RANLIB@ -CPU_INIT_OBJS = rdimon-aem-el3.o +CPU_INIT_OBJS = rdimon-aem-el3.o rdimon-aem-v8-r.o CPU_INIT_INSTALL = install-cpu-init CFLAGS = -g @@ -79,10 +79,12 @@ all: ${CPU_INIT_OBJS} .PHONY: test test: -# Static pattern rule for assembling cpu init files to object files. -${CPU_INIT_OBJS}: %.o: %.S +rdimon-aem-el3.o : rdimon-aem-el3.S $(CC) $(CFLAGS_FOR_TARGET) $(CFLAGS) $(INCLUDES) -DARM_RDI_MONITOR -o $@ -c $< +rdimon-aem-v8-r.o : rdimon-aem-el3.S + $(CC) $(CFLAGS_FOR_TARGET) $(CFLAGS) $(INCLUDES) -DARM_RDI_MONITOR -DBUILD_FOR_R_PROFILE -o $@ -c $< + clean mostlyclean: rm -f a.out core *.i *.o *-test *.srec *.dis *.x diff --git a/libgloss/aarch64/cpu-init/rdimon-aem-el3.S b/libgloss/aarch64/cpu-init/rdimon-aem-el3.S index e00f0b2c8..0296a8054 100644 --- a/libgloss/aarch64/cpu-init/rdimon-aem-el3.S +++ b/libgloss/aarch64/cpu-init/rdimon-aem-el3.S @@ -43,7 +43,9 @@ _init_vectors: /* Installs a table of exception vectors to catch and handle all exceptions by terminating the process with a diagnostic. */ adr x0, vectors +#ifndef BUILD_FOR_R_PROFILE msr vbar_el3, x0 +#endif msr vbar_el2, x0 msr vbar_el1, x0 ret @@ -110,6 +112,14 @@ vectors: .text .align 2 _flat_map: +#ifdef BUILD_FOR_R_PROFILE + mrs x0, sctlr_el2 + orr x0, x0, #1 // SCTLR_EL2.M (enable MPU) + orr x0, x0, #(1 << 17) // SCTLR_EL2.BR (background regions) + msr sctlr_el2, x0 + isb + ret +#else /* Page table setup (identity mapping). */ adrp x0, ttb add x0, x0, :lo12:ttb @@ -173,6 +183,7 @@ _flat_map: isb .Lnosve: ret +#endif .data .align 12 diff --git a/libgloss/aarch64/elf-aem-v8-r.specs b/libgloss/aarch64/elf-aem-v8-r.specs new file mode 100644 index 000000000..4daeffd11 --- /dev/null +++ b/libgloss/aarch64/elf-aem-v8-r.specs @@ -0,0 +1,20 @@ +# elf-aem-v8-r.specs +# +# Spec file for AArch64 baremetal newlib, libgloss on Armv8-R with version 2 of +# AngelAPI semi-hosting. + +%rename link old_link + +*link: +-Ttext-segment 0x10000 %(old_link) + +%rename lib libc + +*libgloss: +-lrdimon + +*lib: +cpu-init/rdimon-aem-v8-r.o%s --start-group %(libc) %(libgloss) --end-group + +*startfile: +crti%O%s crtbegin%O%s %{!pg:rdimon-crt0%O%s} %{pg:rdimon-crt0%O%s}
On 29/09/2020 17:27, Alex Coplan via Newlib wrote: > Hi Richard, > > On 29/09/2020 15:06, Richard Earnshaw (lists) wrote: >> On 18/09/2020 15:43, Alex Coplan wrote: >>> Hello, >>> >>> This patch adds support for Armv8-R AArch64. We update the AArch64 CPU >>> boot code to work on Armv8-R if __ARM_ARCH_PROFILE is set to 'R'. >>> Armv8-R AArch64 has no EL3, so we don't set vbar_el3, and adjust the >>> code to set up the MPU for Armv8-R. >>> >>> We also add a specs file for use with Armv8-R AArch64 models. >>> >>> Testing: >>> * Ran AArch64 GCC testsuite using --with-arch=armv8-r together with the >>> new specs file and boot code in an Armv8-R AArch64 model. >>> * Ran newlib testsuite, no regressions. >>> >>> OK for master? >>> >>> Thanks, >>> Alex >>> >>> --- >>> >>> 2020-09-18 Alex Coplan <alex.coplan@arm.com> >>> >>> libgloss/ChangeLog: >>> >>> * aarch64/Makefile.in: Install new specs file. >>> * aarch64/cpu-init/Makefile.in: Also build boot code for Armv8-R. >>> * aarch64/cpu-init/rdimon-aem-el3.S: Add support for Armv8-R. >>> * aarch64/elf-aem-v8-r.specs: New. >>> >> >> The main problem I see with this is that it assumes that the compiler >> being used is *not* configured for a default CPU architecture of >> ARMv8-r. If it is, then the normal rdimon-aem-el3.o file will in fact >> become the same as the v8-r version, which is not really what we want. > > Good catch. Thanks for the review. > >> >> Perhaps, rather than hacking around the __ARM_ARCH_PROFILE setting you >> should just set things up to build the file twice, but with a different >> conventional pre-define set each time. >> >> R. > > Sounds sensible. Is the revised patch (attached) OK? > > Thanks, > Alex > Pushed. I've used the following as the commit message (note that newlib doesn't use ChangeLog files anymore). libgloss: aarch64: Add support for Armv8-R AArch64 This patch adds support for Armv8-R AArch64. Armv8-R AArch64 has no EL3, so we don't set vbar_el3, and adjust the code to set up the MPU for Armv8-R. So build a different flavour of the startup code to support that. We also add a specs file that uses this alternative startup code which can be used with Armv8-R AArch64 models. R.
Brian Hawley Luminex Software, Inc. 1-951-781-4100 x112 www.luminex.com Original Message From: newlib@sourceware.org Sent: September 30, 2020 3:46 AM To: alex.coplan@arm.com Reply-to: Richard.Earnshaw@arm.com Cc: newlib@sourceware.org Subject: Re: [PATCH] libgloss/aarch64: Add support for Armv8-R AArch64 On 29/09/2020 17:27, Alex Coplan via Newlib wrote: > Hi Richard, > > On 29/09/2020 15:06, Richard Earnshaw (lists) wrote: >> On 18/09/2020 15:43, Alex Coplan wrote: >>> Hello, >>> >>> This patch adds support for Armv8-R AArch64. We update the AArch64 CPU >>> boot code to work on Armv8-R if __ARM_ARCH_PROFILE is set to 'R'. >>> Armv8-R AArch64 has no EL3, so we don't set vbar_el3, and adjust the >>> code to set up the MPU for Armv8-R. >>> >>> We also add a specs file for use with Armv8-R AArch64 models. >>> >>> Testing: >>> * Ran AArch64 GCC testsuite using --with-arch=armv8-r together with the >>> new specs file and boot code in an Armv8-R AArch64 model. >>> * Ran newlib testsuite, no regressions. >>> >>> OK for master? >>> >>> Thanks, >>> Alex >>> >>> --- >>> >>> 2020-09-18 Alex Coplan <alex.coplan@arm.com> >>> >>> libgloss/ChangeLog: >>> >>> * aarch64/Makefile.in: Install new specs file. >>> * aarch64/cpu-init/Makefile.in: Also build boot code for Armv8-R. >>> * aarch64/cpu-init/rdimon-aem-el3.S: Add support for Armv8-R. >>> * aarch64/elf-aem-v8-r.specs: New. >>> >> >> The main problem I see with this is that it assumes that the compiler >> being used is *not* configured for a default CPU architecture of >> ARMv8-r. If it is, then the normal rdimon-aem-el3.o file will in fact >> become the same as the v8-r version, which is not really what we want. > > Good catch. Thanks for the review. > >> >> Perhaps, rather than hacking around the __ARM_ARCH_PROFILE setting you >> should just set things up to build the file twice, but with a different >> conventional pre-define set each time. >> >> R. > > Sounds sensible. Is the revised patch (attached) OK? > > Thanks, > Alex > Pushed. I've used the following as the commit message (note that newlib doesn't use ChangeLog files anymore). libgloss: aarch64: Add support for Armv8-R AArch64 This patch adds support for Armv8-R AArch64. Armv8-R AArch64 has no EL3, so we don't set vbar_el3, and adjust the code to set up the MPU for Armv8-R. So build a different flavour of the startup code to support that. We also add a specs file that uses this alternative startup code which can be used with Armv8-R AArch64 models. R.
diff --git a/libgloss/aarch64/Makefile.in b/libgloss/aarch64/Makefile.in index 546557e3c..b8c2ca43c 100644 --- a/libgloss/aarch64/Makefile.in +++ b/libgloss/aarch64/Makefile.in @@ -85,7 +85,8 @@ RDIMON_BSP = librdimon${${MULTILIBNAME}}.a RDIMON_OBJS = $(patsubst %,rdimon-%,$(OBJS)) RDIMON_SCRIPTS = rdimon${${MULTILIBNAME}}.specs \ aem-ve${${MULTILIBNAME}}.specs \ - aem-validation${${MULTILIBNAME}}.specs + aem-validation${${MULTILIBNAME}}.specs \ + aem-v8-r${${MULTILIBNAME}}.specs RDIMON_INSTALL = install-rdimon CFLAGS = -g diff --git a/libgloss/aarch64/cpu-init/Makefile.in b/libgloss/aarch64/cpu-init/Makefile.in index 08aa1760f..3ada50d64 100644 --- a/libgloss/aarch64/cpu-init/Makefile.in +++ b/libgloss/aarch64/cpu-init/Makefile.in @@ -62,7 +62,7 @@ AR = @AR@ LD = @LD@ RANLIB = @RANLIB@ -CPU_INIT_OBJS = rdimon-aem-el3.o +CPU_INIT_OBJS = rdimon-aem-el3.o rdimon-aem-v8-r.o CPU_INIT_INSTALL = install-cpu-init CFLAGS = -g @@ -79,10 +79,15 @@ all: ${CPU_INIT_OBJS} .PHONY: test test: -# Static pattern rule for assembling cpu init files to object files. -${CPU_INIT_OBJS}: %.o: %.S +rdimon-aem-el3.o : rdimon-aem-el3.S $(CC) $(CFLAGS_FOR_TARGET) $(CFLAGS) $(INCLUDES) -DARM_RDI_MONITOR -o $@ -c $< +rdimon-aem-v8-r.o : rdimon-aem-el3.S + # Force the ACLE macro __ARM_ARCH_PROFILE='R'. This should be defined by + # the compiler for compilers supporting -march=armv8-r on AArch64 but we + # want to be able to build this code with older compilers. + $(CC) $(CFLAGS_FOR_TARGET) $(CFLAGS) $(INCLUDES) -DARM_RDI_MONITOR -D__ARM_ARCH_PROFILE=82 -o $@ -c $< + clean mostlyclean: rm -f a.out core *.i *.o *-test *.srec *.dis *.x diff --git a/libgloss/aarch64/cpu-init/rdimon-aem-el3.S b/libgloss/aarch64/cpu-init/rdimon-aem-el3.S index e00f0b2c8..93dcdb353 100644 --- a/libgloss/aarch64/cpu-init/rdimon-aem-el3.S +++ b/libgloss/aarch64/cpu-init/rdimon-aem-el3.S @@ -43,7 +43,9 @@ _init_vectors: /* Installs a table of exception vectors to catch and handle all exceptions by terminating the process with a diagnostic. */ adr x0, vectors +#if __ARM_ARCH_PROFILE != 'R' msr vbar_el3, x0 +#endif msr vbar_el2, x0 msr vbar_el1, x0 ret @@ -110,6 +112,14 @@ vectors: .text .align 2 _flat_map: +#if __ARM_ARCH_PROFILE == 'R' + mrs x0, sctlr_el2 + orr x0, x0, #1 // SCTLR_EL2.M (enable MPU) + orr x0, x0, #(1 << 17) // SCTLR_EL2.BR (background regions) + msr sctlr_el2, x0 + isb + ret +#else /* Page table setup (identity mapping). */ adrp x0, ttb add x0, x0, :lo12:ttb @@ -173,6 +183,7 @@ _flat_map: isb .Lnosve: ret +#endif .data .align 12 diff --git a/libgloss/aarch64/elf-aem-v8-r.specs b/libgloss/aarch64/elf-aem-v8-r.specs new file mode 100644 index 000000000..4daeffd11 --- /dev/null +++ b/libgloss/aarch64/elf-aem-v8-r.specs @@ -0,0 +1,20 @@ +# elf-aem-v8-r.specs +# +# Spec file for AArch64 baremetal newlib, libgloss on Armv8-R with version 2 of +# AngelAPI semi-hosting. + +%rename link old_link + +*link: +-Ttext-segment 0x10000 %(old_link) + +%rename lib libc + +*libgloss: +-lrdimon + +*lib: +cpu-init/rdimon-aem-v8-r.o%s --start-group %(libc) %(libgloss) --end-group + +*startfile: +crti%O%s crtbegin%O%s %{!pg:rdimon-crt0%O%s} %{pg:rdimon-crt0%O%s}