musl - Re: Porting to RISC-V

Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-id: <BF67221A-08E1-4578-9268-B46D4F4289CF@mac.com>
Date: Thu, 03 May 2018 08:37:13 +1200
From: Michael Clark <michaeljclark@....com>
To: musl@...ts.openwall.com, Dean Michael Ancajas <dbancajas@...il.com>,
 Palmer Dabbelt <palmer@...belt.com>, Rich Felker <dalias@...c.org>
Subject: Re: Porting to RISC-V



> On 3/05/2018, at 7:45 AM, Rich Felker <dalias@...c.org> wrote:
> 
> On Wed, May 02, 2018 at 03:40:04PM -0400, Dean Michael Ancajas wrote:
>> Can you send a link of the wiki?
> 
> https://wiki.musl-libc.org/
> https://wiki.musl-libc.org/porting.html

Here’s a pointer to my fork of Aric Belsito’s tree, including several fixes to get threads and atomics passing libc-tests.

	- https://github.com/michaeljclark/musl-riscv

This is the list of contributors as far as I know, but I might have to do a deeper inspection of the git history:

	Aric Belsito <lluixhi@...il.com>
	Alex Suykov <alex.suykov@...il.com>
	Michael Clark <michaeljclark@....com>

I’ve talked to Palmer Dabbelt about moving the port to the riscv github organisation retaining all of the contributor history. Typically riscv repos are prefixed with riscv- versus suffixed however that is a minor detail. We’ll need to squash the port into some more logical commits as there is quite a bit of churn in the history, however we’ll tag the repo in its current state to keep the contributor history.

Threads and mutexes are working. I need to sync with latest musl and run libc-tests again and we need to run the tests in RISC-V Linux versus RISC-V QEMU linux-user. Running against linux-kernel will give more accurate results compared to QEMU’s linux-user emulation which may not be 100% accurate. This is easier to do now as there are several glibc based full Linux distros that can be run in QEMU RISC-V and on real hardware with networking and block storage. i.e. we can rsync binaries in over ssh in the QEMU virt machine. Indeed folk have been running self-hosted GCC bootstraps in the Fedora RISC-V port which has toolchain packages. Now there is a Debian port, and iirc there may even be a SUSE port (Palmer?)

Here is some recent info on QEMU for RISC-V, which might help with the porting effort:

	- https://github.com/riscv/riscv-qemu/wiki
	- https://www.sifive.com/blog/2018/04/25/risc-v-qemu-part-2-the-risc-v-qemu-port-is-upstream/

Here is the last update I sent regarding the RISC-V musl port…

Issues fixed since picking up GSoC musl-riscv branch:

	• gcc patch to set the musl dynamic linker name (ELF interp) is upstream
		• /lib/ld-musl-riscv32.so.1 (-mabi=ilp32d, default, hard float)
		• /lib/ld-musl-riscv64.so.1 (-mabi=lp64d, default, hard float)
		• /lib/ld-musl-riscv32-sf.so.1 (-mabi=ilp32, soft float)
		• /lib/ld-musl-riscv64-sf.so.1 (-mabi=lp64, soft float)
		• /lib/ld-musl-riscv32-sp.so.1 (-mabi=ilp32f, single precision)
		• /lib/ld-musl-riscv64-sp.so.1 (-mabi=lp64f, single precision)
	• fixed failing pthread tests.
		• a_cas was deadlocking (updated a_cas in atomic_a.h, fixed missing inline asm constraint)
		• defined the minimal set of atomics required by the musl library
	• fixed failing sigaltstack tests (update sigaltstack and ucontext in signal.h)
	• fixed failing ipc_sem tests (added struct semid_ds in sem.h)
	• fixed failing stat tests (defined blksize_t and nlink_t in alltypes.h.in)
	• rename sigcontext __regs to gregs so that gcc would compile
	• rename _gp to __global_pointer$ in the crt to work with current binutils
	• change definition of long double to quadruple precision
	• update syscalls.h.in to use asm-generic syscall definitions
	• update stat.h to use asm-generic stat definition

Remaining issues:

	• rebase to current musl-libc
	• audit arch/riscv32 and arch/riscv64 headers to make sure they match linux-4.16
	• check results of tests that are expected to fail (compare with other architectures)
	• ELF thread local variables are not being initialised
		• tls_init test is failing

Note: riscv32 glibc is not yet upstream so the 32-bit ABI is not yet frozen.

Rich, BTW It seems the TLS offset is directly above the thread pointer (tp).

	$ cat foo.c
	__thread int i = 42;

	void foo()
	{
		i++;
	}

	0000000000010226 <foo>:
	   10226:	00022703          	lw	a4,0(tp) # 0 <i>
	   1022a:	2705                	addiw	a4,a4,1
	   1022c:	00e22023          	sw	a4,0(tp) # 0 <i>
	   10230:	8082                	ret

	0000000000010a7a <__set_thread_area>:
	   10a7a:	822a                	mv	tp,a0
	   10a7c:	4501                	li	a0,0
	   10a7e:	8082                	ret

Michael.
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.