Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221213174234.688534-1-dave.hansen@linux.intel.com>
 <CAHk-=wi=TY3Kte5Z1_nvfcsEh+rcz86pYnzeASw=pbG9QtpJEQ@mail.gmail.com>
 <20221215123007.cagd7qiidehqd77k@box.shutemov.name> <CAHk-=whC_ixb3FDdMhW_wiKw7-bB700kvUyqN+_cPUNp=1hNsQ@mail.gmail.com>
In-Reply-To: <CAHk-=whC_ixb3FDdMhW_wiKw7-bB700kvUyqN+_cPUNp=1hNsQ@mail.gmail.com>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Thu, 15 Dec 2022 09:26:19 -0800
Message-ID: <CAHk-=wjyy3iKS0B=A-yAPqjE3xiUU_5AiXApCasuYKTNu842dQ@mail.gmail.com>
Subject: Re: [GIT PULL] x86/mm for 6.2
To:     kirill.shutemov@linux.intel.com
Cc:     Dave Hansen <dave.hansen@linux.intel.com>,
        linux-kernel@vger.kernel.org, x86@kernel.org, jejb@linux.ibm.com,
        martin.petersen@oracle.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Dec 15, 2022 at 9:17 AM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> Here's *one* suggested solution:

Note again: this is not a "you need to do it this way" suggestion.
This is just a "at least this way doesn't have the issues I object
to". There are bound to be other ways to do it.

But if you feel like all threads have to share the same LAM state, it
does seem a lot simpler if you just say "you need to set that state
before you start any threads". No?

> And yes, I would actually suggest that _any_ thread creation locks it,
> so that you never *EVER* have any issues with "oh, now I need to
> synchronize with other threads". A process can set its LAM state at
> startup, not in the middle of running!

Note that this "no serialization needed" is just about the SW side.

The *hardware* side may still need the IPI just to make sure that it
forces a TLB flush - even if we are single-threaded, that single
thread may have run on other CPU's before.

But I think at that point it's just a regular TLB flush, and doesn't
need that LAM-specific IPI.

But maybe there's some bigger HW serialization that is needed for the
LAM bit, I have not looked at that enough to know.

              Linus