[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <11e30cd0.47dc.16f118a143d.Coremail.sunguoyun@loongson.cn>
Date: Tue, 17 Dec 2019 09:47:39 +0800 (GMT+08:00)
From: 孙国云 <sunguoyun@...ngson.cn>
To: "Paul Burton" <paulburton@...nel.org>
Cc: "Ralf Baechle" <ralf@...ux-mips.org>,
"James Hogan" <jhogan@...nel.org>,
"Peter Zijlstra" <peterz@...radead.org>,
"Christian Brauner" <christian.brauner@...ntu.com>,
"Arnd Bergmann" <arnd@...db.de>,
"Heiko Carstens" <heiko.carstens@...ibm.com>,
"David Howells" <dhowells@...hat.com>,
"Geert Uytterhoeven" <geert@...ux-m68k.org>,
"Catalin Marinas" <catalin.marinas@....com>,
linux-mips@...r.kernel.org, linux-kernel@...r.kernel.org,
"Tiezhu Yang" <yangtiezhu@...ngson.cn>,
"Xuefeng Li" <lixuefeng@...ngson.cn>
Subject: Re: Re: [PATCH V3] MIPS: Add get_thread_area syscall
Dear Paul,
Thank you very much.
I didn`t know this way asm("rdhwr %0, $29" : "=r"(*ptls)) before.Now I have a test for it and it realy can work normally.
I just find get_thread_area syscall at path arch/x86.so I have a implement on MIPS platform. And I think set_thread_area and get_thread_area is matched in pairs.
Now it seems that the get_thread_area syscall is not necessary for criu.
Thanks,
Guoyun Sun
> -----原始邮件-----
> 发件人: "Paul Burton" <paulburton@...nel.org>
> 发送时间: 2019-12-17 03:33:33 (星期二)
> 收件人: "Guoyun Sun" <sunguoyun@...ngson.cn>
> 抄送: "Ralf Baechle" <ralf@...ux-mips.org>, "James Hogan" <jhogan@...nel.org>, "Peter Zijlstra" <peterz@...radead.org>, "Christian Brauner" <christian.brauner@...ntu.com>, "Arnd Bergmann" <arnd@...db.de>, "Heiko Carstens" <heiko.carstens@...ibm.com>, "David Howells" <dhowells@...hat.com>, "Geert Uytterhoeven" <geert@...ux-m68k.org>, "Catalin Marinas" <catalin.marinas@....com>, linux-mips@...r.kernel.org, linux-kernel@...r.kernel.org, "Tiezhu Yang" <yangtiezhu@...ngson.cn>, "Xuefeng Li" <lixuefeng@...ngson.cn>
> 主题: Re: [PATCH V3] MIPS: Add get_thread_area syscall
>
> Hi Guoyun,
>
> On Mon, Dec 16, 2019 at 10:17:51AM +0800, Guoyun Sun wrote:
> > CRIU(https://criu.org/) tools will dump TLS(Thread Local Storage) by
> > get_thread_area during checkpoint and restore TLS by set_thread_area during
> > restore. without this syscall, criu restore will fail on MIPS platform
> > because a variable with GCC __thread will be invalid.
> >
> > The following function will be called when criu restore
> >
> > static inline void restore_tls(tls_t *ptls) {
> > asm volatile(
> > "move $4, %0 \n"
> > "li $2, "__stringify(__NR_set_thread_area)" \n"
> > "syscall \n"
> > :
> > : "r"(*ptls)
> > : "$4","$2","memory");
> >
> > the *ptls can be obtained by get_thread_area on MIPS platform when criu
> > checkpoint. just like this:
> >
> > static inline void arch_get_tls(tls_t *ptls)
> > {
> > asm volatile(
> > "move $4, %0 \n"
> > "li $2, "__stringify(__NR_get_thread_area)" \n"
> > "syscall \n"
> > :
> > : "r"(ptls)
> > : "$4","$2","memory");
> >
> > }
>
> Why would you need a syscall for this?
>
> Why not retrieve the value using the rdhwr instruction just like every
> other bit of code using thread-local storage does?
>
> ie. something like:
>
> static inline void arch_get_tls(tls_t *ptls)
> {
> asm("rdhwr %0, $29" : "=r"(*ptls));
> }
>
> Thanks,
> Paul
</lixuefeng@...ngson.cn></yangtiezhu@...ngson.cn></catalin.marinas@....com></geert@...ux-m68k.org></dhowells@...hat.com></heiko.carstens@...ibm.com></arnd@...db.de></christian.brauner@...ntu.com></peterz@...radead.org></jhogan@...nel.org></ralf@...ux-mips.org></sunguoyun@...ngson.cn></paulburton@...nel.org>
Powered by blists - more mailing lists