Received-SPF: pass (google.com: domain of linux-kernel+bounces-159072-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <CAJZ5v0iUUmMUo86vBzYJjL4NjoFzpDwD1+c292aP+T++PLv6vQ@mail.gmail.com>
 <CAHk-=wj8J78-12QfAoKaLdRi2g1=_U7sv02POShjotcJ6t4nzw@mail.gmail.com>
In-Reply-To: <CAHk-=wj8J78-12QfAoKaLdRi2g1=_U7sv02POShjotcJ6t4nzw@mail.gmail.com>
From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Thu, 25 Apr 2024 12:18:17 -0700
Message-ID: <CAHk-=wj52PUZ0xtoLs79B9uar6h7FVaKC0gbD-a_wZxDjH2ViQ@mail.gmail.com>
Subject: Re: [GIT PULL] ACPI fixes for v6.9-rc6
To: "Rafael J. Wysocki" <rafael@kernel.org>, Jarred White <jarredwhite@linux.microsoft.com>
Cc: ACPI Devel Maling List <linux-acpi@vger.kernel.org>, Linux PM <linux-pm@vger.kernel.org>, 
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"

On Thu, 25 Apr 2024 at 11:58, Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> And maybe this time, it's not a buggy mess?

Actually, even with MASK_VAL() fixed, I think it's *STILL* a buggy mess.

Why? Beuse the *uses* of MASK_VAL() seem entirely bogus.

In particular, we have this in cpc_write():

        if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
                val = MASK_VAL(reg, val);

        switch (size) {
        case 8:
                writeb_relaxed(val, vaddr);
                break;
        case 16:
                writew_relaxed(val, vaddr);
                break;
        ...

and I strongly suspect that it needs to update the 'vaddr' too. Something like

        if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
                val = MASK_VAL(reg, val);
  #ifdef __LITTLE_ENDIAN
                vaddr += reg->bit_offset >> 3;
                if (reg->bit_offset & 7)
                        return -EFAULT;
  #else
                /* Fixme if we ever care */
                if (reg->bit_offset)
                        return -EFAULT;
  #endif
        }

*might* be changing this in the right direction, but it's unclear and
I neither know that CPC rules, nor did I think _that_ much about it.

Anyway, the take-away should be that all this code is entirely broken
and somebody didn't think enough about it.

It's possible that that whole cpc_write() ACPI_ADR_SPACE_SYSTEM_MEMORY
case should be done as a 64-bit "read-mask-write" sequence.

Possibly with "reg->bit_offset == 0" and the 8/16/32/64-bit cases as a
special case for "just do the write".

Or, maybe writes with a non-zero bit offset shouldn't be allowed at
all, and there are CPC rules that aren't checked. I don't know. I only
know that the current code is seriously broken.

                   Linus