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;
From:   Thomas Gleixner <tglx@linutronix.de>
To:     Nikolay Borisov <nik.borisov@suse.com>, x86@kernel.org
Cc:     linux-kernel@vger.kernel.org, mhocko@suse.com, jslaby@suse.cz
Subject: Re: [PATCH v3 5/5] x86/entry: Make IA32 syscalls' availability
 depend on ia32_enabled()
In-Reply-To: <cf24db95-9c91-5d30-fe83-3a1e2cbf0cc1@suse.com>
References: <20230616125730.1164989-1-nik.borisov@suse.com>
 <20230616125730.1164989-6-nik.borisov@suse.com> <87mt0wqxkt.ffs@tglx>
 <cf24db95-9c91-5d30-fe83-3a1e2cbf0cc1@suse.com>
Date:   Mon, 19 Jun 2023 10:40:46 +0200
Message-ID: <87a5wvrgj5.ffs@tglx>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Mon, Jun 19 2023 at 09:28, Nikolay Borisov wrote:
> On 19.06.23 =D0=B3. 0:17 =D1=87., Thomas Gleixner wrote:
>> On Fri, Jun 16 2023 at 15:57, Nikolay Borisov wrote:
>>> Another major aspect of supporting running of 32bit processes is the
>>> ability to access 32bit syscalls. Such syscalls are invoked either by
>>> using the legacy int 0x80 call gate interface or via the newer sysenter
>>> instruction.
>>>
>>> Ensure that if ia32 emulation is disabled (either at compile time or
>>> runtime) then those 2 syscall mechanisms are also disabled.
>>=20
>> AFAICT there are _three_ mechanisms for 32bit syscalls, no?
>
> int 0x80 and sysenter make it 2? Which one is the 3rd one - the "native=20
> 64bit syscall" used in for X32 ABI ? This patch specifically deals with=20
> the first 2?

int 80, sysenter, syscall =3D 3

They obviously depend on the vendor preference when the CPU has enabled
long mode:

                      AMD    Intel
compat_int 80         y      y
compat_sysenter       #UD    y
compat_syscall        y      #UD

On Intel SYSENTER is trivial to disable by setting MSR_IA32_SYSENTER_CS
to 0 which makes sysenter raise #GP.

The nasty one is SYSCALL on AMD. If MSR_EFER.SCE=3D1 then MSR_CSTAR must
contain a valid kernel text address because otherwise compat SYSCALL
faults with CPL0 and user GSBASE. That's the whole reason for the stub
function which just sets EAX to -ENOSYS and returns via SYSRET.

And your patch deals with all _three_:

                    compat=3DON                    compat=3DOFF
compat_int 80:      Set system interrupt gate    ---

compat_sysenter:    Set up SYSENTER MSRs for     Invalidate SYSENTER
                    entry_SYSENTER_compat()      MSRs

compat_syscall:     Set MSR_CSTAR to             Set MSR_CSTAR to
                    entry_SYSCALL_compat()       stub function
                    (AMD only)                   (AMD only)

No?

Changelogs have to be precise. Otherwise they are useless and in the
worst case actively misleading.

Thanks,

        tglx