[PATCH] tile: support very large shared objects

[PATCH] tile: support very large shared objects

[Chris Metcalf]

Changelog:
2012-11-02  Chris Metcalf  <cmetcalf@tilera.com>

	* config.make.in (config-cflags-mcmodel-large): Define.
	* configure.in: Add libc_cv_cc_mcmodel_large.
	* configure: Regenerate.
	* elf/elf.h (R_TILEGX_IMM16_X[01]_HW[012]_(|LAST_)PLT_PCREL): Add.

ports/Changelog.tile:
2012-11-02  Chris Metcalf  <cmetcalf@tilera.com>

	* sysdeps/tile/Makefile [$(config-cflags-mcmodel-large) = yes]:
	Build elf-init.c with -mcmodel=large.
	* sysdeps/tile/crti.S: Support large memory model.
	* sysdeps/tile/start.S: Likewise.
	* sysdeps/tile/tilegx/tilegx64/configure.in: Test -mcmodel=large.
	* sysdeps/tile/tilegx/tilegx64/configure: Regenerate.

With gcc 4.8 tilegx has support for -mcmodel=large, to tolerate very
large shared objects.  This option changes the compiler output to
not include direct jump instructions, which have a range of only
2^30, i.e +/- 512MB.  Instead the compiler marshalls the target PCs
into registers and then uses jump- or call-to-register instructions.

For glibc, the upshot is that we need to add some additional
relocations (pc-relative PLT) and we need to arrange for a few
functions to tolerate the possibility of a large range between the
PC and the target.  In particular, the crti.S and start.S code needs
to be able to reach from .init to the PLT, and the elf-init code
similarly needs to be built with -mcmodel=large.  Note that libc.so
et al don't need any changes; the only changes are for code that
is statically linked into user code built with -mcmodel=large.

For the assembly code, I just rewrote it so that it unconditionally
uses the large model.  To be able to pass -mcmodel=large to
csu/elf-init.c, I need to check to see if the compiler supports
that flag, since gcc 4.7 doesn't; I added the support pretty
mechanically in the same way as libc_cv_cc_nofma, etc., are supported.
---
 config.make.in                                  |    1 +
 configure                                       |    2 ++
 configure.in                                    |    1 +
 elf/elf.h                                       |   17 +++++++++++++---
 ports/sysdeps/tile/Makefile                     |    8 ++++++++
 ports/sysdeps/tile/crti.S                       |   24 ++++++++++++++++++-----
 ports/sysdeps/tile/start.S                      |   23 +++++++++++++++-------
 ports/sysdeps/tile/tilegx/tilegx64/configure    |   21 ++++++++++++++++++++
 ports/sysdeps/tile/tilegx/tilegx64/configure.in |    7 +++++++
 11 files changed, 103 insertions(+), 15 deletions(-)
 create mode 100644 ports/sysdeps/tile/tilegx/tilegx64/configure
 create mode 100644 ports/sysdeps/tile/tilegx/tilegx64/configure.in

diff --git a/config.make.in b/config.make.in
index c22acf4..3b28f33 100644
--- a/config.make.in
+++ b/config.make.in
@@ -41,6 +41,7 @@ config-cflags-avx = @libc_cv_cc_avx@
 config-cflags-sse2avx = @libc_cv_cc_sse2avx@
 config-cflags-novzeroupper = @libc_cv_cc_novzeroupper@
 config-cflags-nofma = @libc_cv_cc_nofma@
+config-cflags-mcmodel-large = @libc_cv_cc_mcmodel_large@
 config-asflags-i686 = @libc_cv_as_i686@
 
 defines = @DEFINES@
diff --git a/configure b/configure
index 35b06c1..457173a 100755
--- a/configure
+++ b/configure
@@ -583,6 +583,7 @@ ldd_rewrite_script
 use_ldconfig
 libc_cv_sparc_as_vis3
 libc_cv_as_i686
+libc_cv_cc_mcmodel_large
 libc_cv_cc_nofma
 libc_cv_cc_fma4
 libc_cv_cc_novzeroupper
@@ -7510,6 +7511,7 @@ fi
 
 
 
+
 if test x$use_ldconfig = xyes; then
   $as_echo "#define USE_LDCONFIG 1" >>confdefs.h
 
diff --git a/configure.in b/configure.in
index 928b7b6..ca56686 100644
--- a/configure.in
+++ b/configure.in
@@ -2169,6 +2169,7 @@ AC_SUBST(libc_cv_cc_sse2avx)
 AC_SUBST(libc_cv_cc_novzeroupper)
 AC_SUBST(libc_cv_cc_fma4)
 AC_SUBST(libc_cv_cc_nofma)
+AC_SUBST(libc_cv_cc_mcmodel_large)
 AC_SUBST(libc_cv_as_i686)
 AC_SUBST(libc_cv_sparc_as_vis3)
 
diff --git a/elf/elf.h b/elf/elf.h
index 41fc3bd..2524924 100644
--- a/elf/elf.h
+++ b/elf/elf.h
@@ -2986,12 +2986,18 @@ typedef Elf32_Addr Elf32_Conflict;
 #define R_TILEGX_IMM16_X1_HW2_LAST_PCREL 63 /* X1 pipe PC-rel last hword 2 */
 #define R_TILEGX_IMM16_X0_HW0_GOT 64	/* X0 pipe hword 0 GOT offset */
 #define R_TILEGX_IMM16_X1_HW0_GOT 65	/* X1 pipe hword 0 GOT offset */
-/* Relocs 66-71 are currently not defined.  */
+#define R_TILEGX_IMM16_X0_HW0_PLT_PCREL 66 /* X0 pipe PC-rel PLT hword 0 */
+#define R_TILEGX_IMM16_X1_HW0_PLT_PCREL 67 /* X1 pipe PC-rel PLT hword 0 */
+#define R_TILEGX_IMM16_X0_HW1_PLT_PCREL 68 /* X0 pipe PC-rel PLT hword 1 */
+#define R_TILEGX_IMM16_X1_HW1_PLT_PCREL 69 /* X1 pipe PC-rel PLT hword 1 */
+#define R_TILEGX_IMM16_X0_HW2_PLT_PCREL 70 /* X0 pipe PC-rel PLT hword 2 */
+#define R_TILEGX_IMM16_X1_HW2_PLT_PCREL 71 /* X1 pipe PC-rel PLT hword 2 */
 #define R_TILEGX_IMM16_X0_HW0_LAST_GOT 72 /* X0 pipe last hword 0 GOT offset */
 #define R_TILEGX_IMM16_X1_HW0_LAST_GOT 73 /* X1 pipe last hword 0 GOT offset */
 #define R_TILEGX_IMM16_X0_HW1_LAST_GOT 74 /* X0 pipe last hword 1 GOT offset */
 #define R_TILEGX_IMM16_X1_HW1_LAST_GOT 75 /* X1 pipe last hword 1 GOT offset */
-/* Relocs 76-77 are currently not defined.  */
+#define R_TILEGX_IMM16_X0_HW3_PLT_PCREL 76 /* X0 pipe PC-rel PLT hword 3 */
+#define R_TILEGX_IMM16_X1_HW3_PLT_PCREL 77 /* X1 pipe PC-rel PLT hword 3 */
 #define R_TILEGX_IMM16_X0_HW0_TLS_GD 78	/* X0 pipe hword 0 TLS GD offset */
 #define R_TILEGX_IMM16_X1_HW0_TLS_GD 79	/* X1 pipe hword 0 TLS GD offset */
 #define R_TILEGX_IMM16_X0_HW0_TLS_LE 80	/* X0 pipe hword 0 TLS LE offset */
@@ -3007,7 +3013,12 @@ typedef Elf32_Addr Elf32_Conflict;
 /* Relocs 90-91 are currently not defined.  */
 #define R_TILEGX_IMM16_X0_HW0_TLS_IE 92	/* X0 pipe hword 0 TLS IE offset */
 #define R_TILEGX_IMM16_X1_HW0_TLS_IE 93	/* X1 pipe hword 0 TLS IE offset */
-/* Relocs 94-99 are currently not defined.  */
+#define R_TILEGX_IMM16_X0_HW0_LAST_PLT_PCREL 94 /* X0 pipe PC-rel PLT last hword 0 */
+#define R_TILEGX_IMM16_X1_HW0_LAST_PLT_PCREL 95 /* X1 pipe PC-rel PLT last hword 0 */
+#define R_TILEGX_IMM16_X0_HW1_LAST_PLT_PCREL 96 /* X0 pipe PC-rel PLT last hword 1 */
+#define R_TILEGX_IMM16_X1_HW1_LAST_PLT_PCREL 97 /* X1 pipe PC-rel PLT last hword 1 */
+#define R_TILEGX_IMM16_X0_HW2_LAST_PLT_PCREL 98 /* X0 pipe PC-rel PLT last hword 2 */
+#define R_TILEGX_IMM16_X1_HW2_LAST_PLT_PCREL 99 /* X1 pipe PC-rel PLT last hword 2 */
 #define R_TILEGX_IMM16_X0_HW0_LAST_TLS_IE 100 /* X0 pipe last hword 0 IE off */
 #define R_TILEGX_IMM16_X1_HW0_LAST_TLS_IE 101 /* X1 pipe last hword 0 IE off */
 #define R_TILEGX_IMM16_X0_HW1_LAST_TLS_IE 102 /* X0 pipe last hword 1 IE off */
diff --git a/ports/sysdeps/tile/Makefile b/ports/sysdeps/tile/Makefile
index 3cd4d1e..eb04627 100644
--- a/ports/sysdeps/tile/Makefile
+++ b/ports/sysdeps/tile/Makefile
@@ -11,3 +11,11 @@ ifeq ($(subdir),elf)
 # Extra shared linker files to link only into dl-allobjs.so.
 sysdep-rtld-routines += dl-start __tls_get_addr
 endif
+
+ifeq ($(subdir),csu)
+ifeq (yes,$(config-cflags-mcmodel-large))
+# Support largest code model since the init code at the start of the
+# linked shared object has to reach the .plt at the end of the object.
+CFLAGS-elf-init.c += -mcmodel=large
+endif
+endif
diff --git a/ports/sysdeps/tile/crti.S b/ports/sysdeps/tile/crti.S
index ccb4464..de87f4c 100644
--- a/ports/sysdeps/tile/crti.S
+++ b/ports/sysdeps/tile/crti.S
@@ -40,6 +40,7 @@
 
 #include <libc-symbols.h>
 #include <sysdep.h>
+#include <bits/wordsize.h>
 
 #ifndef PREINIT_FUNCTION
 # define PREINIT_FUNCTION __gmon_start__
@@ -70,16 +71,17 @@ _init:
 #if PREINIT_FUNCTION_WEAK
 	lnk r2
 0:
-#ifdef __tilegx__
+# ifdef __tilegx__
+	moveli r1, hw2_last(_GLOBAL_OFFSET_TABLE_ - 0b)
 	{
-	 moveli r1, hw1_last(_GLOBAL_OFFSET_TABLE_ - 0b)
+	 shl16insli r1, r1, hw1(_GLOBAL_OFFSET_TABLE_ - 0b)
 	 moveli r0, hw1_last_got(PREINIT_FUNCTION)
 	}
 	{
 	 shl16insli r1, r1, hw0(_GLOBAL_OFFSET_TABLE_ - 0b)
 	 shl16insli r0, r0, hw0_got(PREINIT_FUNCTION)
 	}
-#else
+# else
 	{
 	 moveli r1, lo16(_GLOBAL_OFFSET_TABLE_ - 0b)
 	 moveli r0, got_lo16(PREINIT_FUNCTION)
@@ -88,13 +90,25 @@ _init:
 	 auli r1, r1, ha16(_GLOBAL_OFFSET_TABLE_ - 0b)
 	 auli r0, r0, got_ha16(PREINIT_FUNCTION)
 	}
-#endif
+# endif
 	ADD_PTR r0, r0, r1
 	ADD_PTR r0, r0, r2
 	LD_PTR r0, r0
 	BEQZ r0, .Lno_weak_fn
-#endif
+	jalr r0
+#elif __WORDSIZE == 64
+	/* Since we are calling from the start of the object to the PLT,
+	   use the -mcmodel=large call model.  */
+	lnk r2
+0:
+	moveli r0, hw2_last_plt(PREINIT_FUNCTION - 0b)
+	shl16insli r0, r0, hw1_plt(PREINIT_FUNCTION - 0b)
+	shl16insli r0, r0, hw0_plt(PREINIT_FUNCTION - 0b)
+	add r0, r0, r2
+	jalr r0
+#else
 	jal plt(PREINIT_FUNCTION)
+#endif
 .Lno_weak_fn:
 
 	.section .fini,"ax",@progbits
diff --git a/ports/sysdeps/tile/start.S b/ports/sysdeps/tile/start.S
index 999bb53..54f015f 100644
--- a/ports/sysdeps/tile/start.S
+++ b/ports/sysdeps/tile/start.S
@@ -126,27 +126,37 @@ _start:
 	 moveli r0, hw2_last(main - .Lmy_pc)
 	}
 	{
-	 moveli r3, hw2_last(__libc_csu_init - .Lmy_pc)
 	 shl16insli r0, r0, hw1(main - .Lmy_pc)
+	 moveli r3, hw2_last(__libc_csu_init - .Lmy_pc)
 	}
 	{
-	 shl16insli r3, r3, hw1(__libc_csu_init - .Lmy_pc)
 	 shl16insli r0, r0, hw0(main - .Lmy_pc)
+	 shl16insli r3, r3, hw1(__libc_csu_init - .Lmy_pc)
 	}
 	{
+	 ADD_PTR r0, r0, r13
 	 shl16insli r3, r3, hw0(__libc_csu_init - .Lmy_pc)
+	}
+	{
+	 moveli r12, hw2_last_plt(__libc_start_main - .Lmy_pc)
+	 ADD_PTR r3, r3, r13
+	}
+	{
+	 shl16insli r12, r12, hw1_plt(__libc_start_main - .Lmy_pc)
 	 moveli r4, hw2_last(__libc_csu_fini - .Lmy_pc)
 	}
 	{
-	 ADD_PTR r0, r0, r13
+	 shl16insli r12, r12, hw0_plt(__libc_start_main - .Lmy_pc)
 	 shl16insli r4, r4, hw1(__libc_csu_fini - .Lmy_pc)
 	}
 	{
-	 ADD_PTR r3, r3, r13
+	 ADD_PTR r12, r12, r13
 	 shl16insli r4, r4, hw0(__libc_csu_fini - .Lmy_pc)
 	}
 	{
 	 ADD_PTR r4, r4, r13
+	 jalr r12
+	}
 #else
 	 addli r0, r13, lo16(main - .Lmy_pc)
 	}
@@ -160,13 +170,12 @@ _start:
 	}
 	{
 	 auli r4, r4, ha16(__libc_csu_fini - .Lmy_pc)
-
-#endif
-
 	 /* Call the user's main function, and exit with its value.
 	    But let the libc call main. */
 	 j plt(__libc_start_main)
 	}
+#endif
+
 	{
 	 /* Tell backtracer to give up (_start has no caller). */
 	 info INFO_OP_CANNOT_BACKTRACE
diff --git a/ports/sysdeps/tile/tilegx/tilegx64/configure b/ports/sysdeps/tile/tilegx/tilegx64/configure
new file mode 100644
index 0000000..dbe0a87
--- /dev/null
+++ b/ports/sysdeps/tile/tilegx/tilegx64/configure
@@ -0,0 +1,21 @@
+# This file is generated from configure.in by Autoconf.  DO NOT EDIT!
+ # Local configure fragment for sysdeps/tile/tilegx/tilegx64.
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for large memory model support" >&5
+$as_echo_n "checking for large memory model support... " >&6; }
+if ${libc_cv_cc_mcmodel_large+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  if { ac_try='${CC-cc} -mcmodel=large -xc /dev/null -S -o /dev/null'
+  { { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; } >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+  test $ac_status = 0; }; }; then :
+  libc_cv_cc_mcmodel_large=yes
+else
+  libc_cv_cc_mcmodel_large=no
+fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libc_cv_cc_mcmodel_large" >&5
+$as_echo "$libc_cv_cc_mcmodel_large" >&6; }
diff --git a/ports/sysdeps/tile/tilegx/tilegx64/configure.in b/ports/sysdeps/tile/tilegx/tilegx64/configure.in
new file mode 100644
index 0000000..5a7e912
--- /dev/null
+++ b/ports/sysdeps/tile/tilegx/tilegx64/configure.in
@@ -0,0 +1,7 @@
+GLIBC_PROVIDES dnl See aclocal.m4 in the top level source directory.
+# Local configure fragment for sysdeps/tile/tilegx/tilegx64.
+
+dnl Check if -mcmodel=large exists (gcc 4.8 and later).
+AC_CACHE_CHECK(for large memory model support, libc_cv_cc_mcmodel_large, [dnl
+LIBC_TRY_CC_OPTION([-mcmodel=large], [libc_cv_cc_mcmodel_large=yes], dnl
+[libc_cv_cc_mcmodel_large=no])])
-- 
1.7.10.3

Re: [PATCH] tile: support very large shared objects

[Roland McGrath]

The elf.h changes are OK provided binutils trunk already has all those
exact names and numbers for those reloc codes.  That's always good as
an isolated commit.

I think we can avoid the changes to the generic configure/config.make for
something only one port uses in a couple of ways.

In sysdeps/tile/Makefile do "include $(common-objpfx)config-tile.make".
That file will contain 'config-cflags-mcmodel = ...' or whatever you need.
Now, you have a couple of ways to go about creating that file.

One is to create it in configure.  I think you can probably just put in
a sysdeps/.../configure.in:

	AC_CONFIG_COMMANDS([config-tile.make],
	[echo "config-cflags-mcmodel-large = $libc_cv_cc_mcmodel_large" > config.tile.make],
	[libc_cv_cc_mcmodel_large=$libc_cv_cc_mcmodel_large])

and have it actually do the right thing.  But I haven't played with using
AC_CONFIG_COMMANDS this way to be sure it really works.  (Which macros
really work usefully in a fragment is hard to predict.)

The other way is to create on the fly at build time.  Either way you want
to add a $(common-objpfx)config-tile.make target in sysdeps/tile/Makefile.
If you are creating it with configure, then I think it can be just:
	$(common-objpfx)config-tile.make: $(common-objpfx)config.make ;
to take advantage of the config.make target in Makeconfig for correct
rebuilding after configure has been run again or changed.

If you don't do it in configure, then you just do it right there in your
target rule for config-tile.make.  I'd still make it depend on config.make,
but give it nonempty commands that actually create the file by running the
compiler test:
	$(common-objpfx)config-tile.make: $(common-objpfx)config.make
		mcmodel=no; \
		$(CC) -S -o /dev/null -xc /dev/null -mcmodel=large || mcmodel=yes; \
		echo "config-cflags-mcmodel = $mcmodel" > $@
or whatever it is.  In this case, you might want to make it a more specific
name since it's really an entirely local thing in sysdeps/tile/Makefile and
you might as well just have one file per test, so just make it
config-mcmodel-large.mk or something.

I generally think it's nice to do less in configure when we can do it in
make reasonably cheaply.  But I don't have a strong preference for which
way you do it, just that it be done cleanly and entirely by changing only
sysdeps configure and/or Makefile files.

csu/ is not enough.  You also need to use the flag for all of *_nonshared.a
(.oS files).  That too is code that's always statically linked into
applications.  And elf-init is not all of csu/, there's also gmon-start and
so forth.  Not to mention libc.a--do you not want fully-static linking to
work with -mcmodel=large?  So I think you want to do:

	CFLAGS-.o += -mcmodel=large

(This will cover .oS files too, which is what you're after most.)  If it's
too costly to do this for all of libc.a and you think people only want
static linking without the possibility of using -mcmodel=large or dynamic
linking with the possibility, then you can instead do:

	ifeq ($(subdir),csu)
	CFLAGS-.o += -mcmodel=large
	else
	CFLAGS-.oS += -mcmodel=large
	endif


The tile assembly bits are of course up to you and I don't presume to make
any sense of them.


Thanks,
Roland

Re: [PATCH] tile: support very large shared objects

[Chris Metcalf]

On 11/3/2012 12:47 AM, Roland McGrath wrote:
> The elf.h changes are OK provided binutils trunk already has all those
> exact names and numbers for those reloc codes.  That's always good as
> an isolated commit.

OK, pushed.

> I think we can avoid the changes to the generic configure/config.make for
> something only one port uses in a couple of ways.
>
> In sysdeps/tile/Makefile do "include $(common-objpfx)config-tile.make".
> That file will contain 'config-cflags-mcmodel = ...' or whatever you need.
> Now, you have a couple of ways to go about creating that file.
>
> [...]
> The other way is to create on the fly at build time.  Either way you want
> to add a $(common-objpfx)config-tile.make target in sysdeps/tile/Makefile.

I took this approach.

> 	$(common-objpfx)config-tile.make: $(common-objpfx)config.make
> 		mcmodel=no; \
> 		$(CC) -S -o /dev/null -xc /dev/null -mcmodel=large || mcmodel=yes; \
> 		echo "config-cflags-mcmodel = $mcmodel" > $@

I used cflags-mcmodel-large as the variable name, and
cflags-mcmodel-large.mk as the fragment filename.

> csu/ is not enough.  You also need to use the flag for all of *_nonshared.a
> (.oS files).  That too is code that's always statically linked into
> applications.

I did think about this.  My conclusion was that since *_nonshared.a is
linked last, it will always be at the end of the generated object, so jumps
and calls will always be able to "reach" other things linked in from
*_nonshared.a as well as the PLT, which immediately follows the text
section.  Thus, no need for large model.  In practice this does appear to
be the case.  (And this large model code is noticeably slower than regular
mode, and we've verified that it works at the customer with the actual
large object this way, for what that's worth.)

> And elf-init is not all of csu/, there's also gmon-start and
> so forth.

Ah, I didn't think about gmon-start.  That ends up linked into the object
statically for profiling, now that I look at it.  Is it true that the rest
are linked statically into shared objects?  I thought many of them were
just there for purely static build support.

> Not to mention libc.a--do you not want fully-static linking to
> work with -mcmodel=large?

I think the performance penalty is enough that the answer is "no" by
default.  (A one-instruction call or jump becomes a four-instruction "load
a register and jump" sequence.)  If some customer needs a ginormous static
binary, it's actually reasonable to just say "build a libc.a yourself and
link against it".  We just want to make sure the shared-object
infrastructure "just works" for the case of large objects (as you suggest
further down in your email).

>   So I think you want to do:
>
> 	CFLAGS-.o += -mcmodel=large
>
> (This will cover .oS files too, which is what you're after most.)  If it's
> too costly to do this for all of libc.a and you think people only want
> static linking without the possibility of using -mcmodel=large or dynamic
> linking with the possibility, then you can instead do:
>
> 	ifeq ($(subdir),csu)
> 	CFLAGS-.o += -mcmodel=large
> 	else
> 	CFLAGS-.oS += -mcmodel=large
> 	endif

Thanks for your feedback.  I'll post a modified patch for review shortly.

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com

Re: [PATCH] tile: support very large shared objects

[Chris Metcalf]

On 11/3/2012 9:28 AM, Chris Metcalf wrote:
> Ah, I didn't think about gmon-start.  That ends up linked into the object
> statically for profiling, now that I look at it.  Is it true that the rest
> are linked statically into shared objects?  I thought many of them were
> just there for purely static build support.

I convinced myself that only gmon-start.c needed to be passed
-mcmodel=large in addition to elf-init.c.

Roland, I cc'ed you on the patch v2, but didn't cc the rest of libc-alpha
since it is now completely platform-specific (thus went only to libc-ports).

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com

Re: [PATCH] tile: support very large shared objects

[Roland McGrath]

> > csu/ is not enough.  You also need to use the flag for all of *_nonshared.a
> > (.oS files).  That too is code that's always statically linked into
> > applications.
> 
> I did think about this.  My conclusion was that since *_nonshared.a is
> linked last, it will always be at the end of the generated object, so jumps
> and calls will always be able to "reach" other things linked in from
> *_nonshared.a as well as the PLT, which immediately follows the text
> section.  Thus, no need for large model.  In practice this does appear to
> be the case.  (And this large model code is noticeably slower than regular
> mode, and we've verified that it works at the customer with the actual
> large object this way, for what that's worth.)

There are many ways to arrive at a situation where these assumptions fall
down.  In general, we try to keep things fully general-purpose and not bake
in presumptions of this nature about what users do in the range of valid
inputs.  But since this is for a machine-specific, rarely-used option we'll
leave it up to you to decide about those tradeoffs.