Backport following patches:
 - RISC-V: Add shorten_memrefs pass.
 - RISC-V: Make unique SECCAT_SRODATA names start with .srodata (not .sdata2)
 - RISC-V: Update march parser
 - RISC-V: Handle implied extension for -march parser
 - Fix alignment for local variable [PR90811]
 - RISC-V: Optimize si to di zero-extend followed by left shift.
 - RISC-V: Make __divdi3 handle div by zero same as hardware.
 - RISC-V: Describe correct USEs for gpr_save pattern [PR95252]
 - RISC-V: Unify the output asm pattern between gpr_save and gpr_restore pattern.

Bump gcc10

 - With few bugfix backport:
   - riscv: Fix up riscv_atomic_assign_expand_fenv [PR94950]
   - testsuite: Require gnu-tm support for pr94856.C

This reverts commit b6561b1a.

qemu 5.0 using new kernel struct define for rv32, which require latest
glibc, however it's not upstream yet, so I decide delay the update
until rv32 glibc upstream.

Handle underline in arch string for target board option

Bump qemu to 5.0.0

Bump gdb to 9.1 with gdb sim.

Use proper kernel headers

Update whitelist for newlib nano

Bump binutils to 2.34

Tweak makefile rule for check binutils gdb 

This testcase is used to check whether the linker is work normally when both
the --wrap=SYMBOL and -flto are set.  It seems that the current fix only can
work when the wrapped symbol `SYMBOL` is defined in the other files and will be
linked.  See the pr-24406-1 test, the `read` symbol is defined in the glibc’s
libc.so.6, and we will also link the libc.so.6 by the linux toolchain, so it
works.  But the `read` symbol isn’t defined in the newlib, or other linked
objects or archives.  Therefore, the pr-24406-1 testsuite is always failed
for riscv elf toolchain, but seems to work for riscv linux toolchain.

On the other hand, if the wrapped symbol is `read_p`, it isn't defined in the
other linked files, so we also get the error link for the linux toolchain.
The fix in upstream seems not complete.  I suppose that we shoud just skip the
pr-24406-1 testcase until upstream fix or remove it.

 - dash won't expand it as expect

Add a newlib nano copy of libm.

[configure] fix unexpected operator line 3295

equality of two string is `=` not `==` see `test(1)`

The gamma functions use the reentrant structure which changes size in newlib
nano, so we need a nano specific copy of libm.a to go with libc.a.

Added python3 to Ubuntu prerequisites.

Bump newlib to 3.2.0

Fix non-portable operator in shell conditional

Use = to compare strings with the test builtin command instead of ==,
which is non-standard and sometimes unsupported by stricter POSIX shells
(e.g., dash as /bin/sh).

Add a pr69728 optimization success to the whitelist

I'm not sure what's going on here, but as far as I can tell we're optimizing
this while the test suite thinks we shouldn't be able to.

Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>

Don't enable gcc checking by default.

Also, support all gcc checking options instead of just yes and release.
Also, fix copy-and-paste error in cmodel support refering to gcc checking.