diff mbox series

Re: [review] WP documentation for flashrom users

Message ID Y1pypjwlPtAIcLaS@MjU3Nj
State New
Headers show
Series Re: [review] WP documentation for flashrom users | expand

Commit Message

Sergii Dmytruk Oct. 27, 2022, 11:59 a.m. UTC
Thanks for taking a look.  Diff of changes and new version is below.

================================================================================

 bootblock-protection.md | 6 +++---
 heads-and-wp.md         | 9 +++++----
 2 files changed, 8 insertions(+), 7 deletions(-)


================================================================================

# Firmware updates vs. SPI write-protection

Enabling write-protection of any kind is meant to obstruct changing data, but it
also limits what you can do to the part of firmware that's still writable. This
document is meant to cover some of the origins of such limitations and
situations which might arise after part of a flash chip has been protected.

## Firmware updates after locking bootblock

This section is primarily concerned with `coreboot`, but similar problems can
happen for any kind of firmware.

### Risks of partial updates

Partial updates can produce an unbootable image if an old bootblock doesn't work
with a more recent version of `coreboot`. This can be manifested in various
ways ranging from an old bootblock not being able to find new romstage to system
booting successfully but data in `coreboot` tables being mangled or incomplete.

The incompatibilities might happen when switching version of firmware or even
when using the same version with a slightly different configuration.

Another thing that can potentially cause trouble is CBFS layout. When bootblock
is part of CBFS, it doesn't necessarily have a fixed address, moreover it can
change location as well if it depends on file size (when bootblock's last byte
must be the last byte of the image, which is the case on x86). If newer
bootblock is smaller such that an old WP range now covers bootblock and some
other file, this file won't be fully updated due to write-protection,
potentially resulting in a corrupt image. Luckily, when bootblock is the last
file it's normally preceded by a significant amount of empty space, which won't
let this situation to occur.

On top of that, last 4 bytes of the image contain offset to the master header of
CBFS. Depending on the coreboot version this offset might be crucial for the
loading of romstage, in which case moving CBFS within the image without updating
the offset (when it's locked by WP) can also prevent the system from booting.

### Recovering from a broken state

Since broken flash won't let the system to boot, the way to fix it is to flash
the chip externally by connecting it to a different device. A possible
alternative could be to have a backup flash created beforehand and swapping it
for the broken one (mainly applicable if swapping doesn't require soldering).

## Flashing whole firmware image

The function of the hardware protection mechanism (`W#` or `W/` pin of flash
chips) is to lock state of software protection thus preventing it from being
disabled. After the chip is physically unlocked by changing the state of the
pin, the state of the write protection doesn't change. However, in this state
the protection can be easily turned off programmatically, which is what
`flashrom` tries to do before performing an operation on a chip.

In other words, changing state of the WP pin might be enough to be able to
flash the chip in full. If `flashrom` errors or you don't want to rely on the
automatic behaviour, you can try to explicitly disable the protection by running
`flashrom` like this:

```
flashrom --wp-disable
```

If you need to pass extra parameters to flash your chip (e.g., programmer or
chip name), add them to the above command (order of such parameters shouldn't
matter).

Mind that as of now (26 October 2022) `flashrom` uses different code for
temporarily disabling protection automatically before an operation (the state
is restored afterwards) and for handling `--wp-disable`. This means that effects
on write protection can be different depending on which code changes it, which
is why you should know about existence of both of these code paths.

================================================================================

# Example of partial write-protection

This document provides demonstration of how one can protect part of a flash
chip from writing using `flashrom` and its support for manipulating SPI write
protection (WP). This kind of protection requires changing connection of WP
pin of the chip to prevent any attempt of disabling the protection by software
alone.

**Not to be confused** with protection by flash controller of your
motherboard (PCH protection).

## `flashrom` version

At the time of writing (26 October 2022) there hasn't been a `flashrom` release
that includes WP manipulation facilities. You might have to build one from
scratch (assuming you've already installed build dependencies):

```
git clone --depth 1 https://github.com/flashrom/flashrom
cd flashrom
# the simplest case of building using GNU make
make
# flashrom executable will appear in current directory
```

## Programmer support of WP

Not all programmers support manipulating WP configuration. A suitable
programmer must either provide a dedicated API for working with WP or give
sufficiently comprehensive access to the interface of the flash chip.

In particular, on Intel platforms *internal* programmer might allow only limited
access to WP feature of chips or effectively deny it. Read "Intel chipsets"
section of `flashrom`'s manpage for details on how you can try choosing
sequencing type to possibly make WP work for you.

In some cases external flashing might be the only option and you need to
unscrew your device, find the chip, connect it to another device through a
suitable adapter and finally be able to configure it as you wish.

## Chip support in `flashrom`

There is a great variety of chips with some not supporting write protection at
all and others doing it in their own peculiar way of which `flashrom` has no
idea. So the first thing to do is to make sure that `flashrom` knows how WP
works for your chip and chipset doesn't get in the way. Run a command like
(adjust this and similar commands below if you're not using *internal*
programmer or need to specify other options):

```
flashrom --programmer internal --wp-status
```

Seeing this output line would mean that `flashrom` doesn't know how to use WP
feature of the chip you have:

```
Failed to get WP status: WP operations are not implemented for this chip
```

Otherwise the output might contain something similar to this:

```
Protection range: start=0x00000000 length=0x00000000 (none)
Protection mode: disabled
```

If so, you can continue with the rest of the instructions.

## Collecting information about the range

You need to know where the area you want to protect starts and ends. The example
below assumes you're trying to protect bootblock stored in CBFS at the end of
some `coreboot` firmware. In other cases it might be a separate file which is
put at the beginning of a chip. You need to have an idea of what you're doing
here or have some reliable instructions to follow.

In this case `cbfstool` can be used to list information about bootblock like
this:

```
$ cbfstool rom print | sed -n '2p; /bootblock/p'
Name                           Offset     Type           Size   Comp
bootblock                      0x3ef100   bootblock       36544 none
```

However, the offset is relative to the start of CBFS region, so we also need to
find out offset of CBFS:

```
$ cbfstool rom layout | grep CBFS
'COREBOOT' (CBFS, size 4161536, offset 12615680)
```

Now we can calculate:

 * start offset (CBFS offset + 64 + bootblock offset): \
   `12615680 + 64 + 0x3ef100 = 0xff7140` \
   (`printf "%#x\n" $(( 12615680 + 64 + 0x3ef100 ))`)
 * end offset (start offset + bootblock size - 1): \
   `0xff7140 + 36544 - 1 = 0xffffff` \
   (`printf "%#x\n" $(( 0xff7140 + 36544 - 1 ))`)

Thus we need to write-protect the smallest area that covers the range from
`0xff7140` to `0xffffff` (both bounds are inclusive).

"64" in the computation of start offset is offset of booblock data.
Unfortunately, current tooling doesn't provide a reliable way of determining
actual offset, but 64 is the typical "extra offset" one needs to add to account
for file metadata of CBFS (otherwise it can be its multiple 128 or bigger).
Bootblock should normally end at the last byte of ROM on x86 systems, giving you
a way to test the result of computations.

## Finding a matching range

In most chips the list of supported ranges is fixed and you can't specify an
arbitrary one. Some others allow more fine-grained control, but that feature is
not supported by `flashrom` as of now.

Obtain list of supported ranges from which we'll pick the best match:

```
$ flashrom --programmer internal --wp-list
...
Available protection ranges:
        start=0x00000000 length=0x00000000 (none)
        start=0x00000000 length=0x00001000 (lower 1/4096)
        start=0x00fff000 length=0x00001000 (upper 1/4096)
        start=0x00000000 length=0x00002000 (lower 1/2048)
        start=0x00ffe000 length=0x00002000 (upper 1/2048)
        start=0x00000000 length=0x00004000 (lower 1/1024)
        start=0x00ffc000 length=0x00004000 (upper 1/1024)
        start=0x00000000 length=0x00008000 (lower 1/512)
        start=0x00ff8000 length=0x00008000 (upper 1/512)
        start=0x00000000 length=0x00040000 (lower 1/64)
        start=0x00fc0000 length=0x00040000 (upper 1/64)
        start=0x00000000 length=0x00080000 (lower 1/32)
        start=0x00f80000 length=0x00080000 (upper 1/32)
        start=0x00000000 length=0x00100000 (lower 1/16)
        start=0x00f00000 length=0x00100000 (upper 1/16)
        start=0x00000000 length=0x00200000 (lower 1/8)
        start=0x00e00000 length=0x00200000 (upper 1/8)
        start=0x00000000 length=0x00400000 (lower 1/4)
        start=0x00c00000 length=0x00400000 (upper 1/4)
        start=0x00000000 length=0x00800000 (lower 1/2)
        start=0x00800000 length=0x00800000 (upper 1/2)
        start=0x00000000 length=0x00c00000 (lower 3/4)
        start=0x00400000 length=0x00c00000 (upper 3/4)
        start=0x00000000 length=0x00e00000 (lower 7/8)
        start=0x00200000 length=0x00e00000 (upper 7/8)
        start=0x00000000 length=0x00f00000 (lower 15/16)
        start=0x00100000 length=0x00f00000 (upper 15/16)
        start=0x00000000 length=0x00f80000 (lower 31/32)
        start=0x00080000 length=0x00f80000 (upper 31/32)
        start=0x00000000 length=0x00fc0000 (lower 63/64)
        start=0x00040000 length=0x00fc0000 (upper 63/64)
        start=0x00000000 length=0x00ff8000 (lower 511/512)
        start=0x00008000 length=0x00ff8000 (upper 511/512)
        start=0x00000000 length=0x00ffc000 (lower 1023/1024)
        start=0x00004000 length=0x00ffc000 (upper 1023/1024)
        start=0x00000000 length=0x00ffe000 (lower 2047/2048)
        start=0x00002000 length=0x00ffe000 (upper 2047/2048)
        start=0x00000000 length=0x00fff000 (lower 4095/4096)
        start=0x00001000 length=0x00fff000 (upper 4095/4096)
        start=0x00000000 length=0x01000000 (all)
```

Pick a range by scanning the list in the top down order (because the smaller
ranges come first):

 - if bootblock is at the start of a chip, look for the first lower range whose
   length is greater than the end offset
 - if bootblock is at the end of a chip, look for the first upper range which
   starts before or at the start offset
 - mind that you're unlikely to find an ideal match and will probably protect
   more than you need; this is fine if that's just an empty space, but can
   cause troubles with future updates if that's some data or metadata which
   changes with every release

This is the first upper range starting before `0xff7140`:

```
        start=0x00fc0000 length=0x00040000 (upper 1/64)
```

It covers `0x00fc0000 -- 0x00ffffff` which includes our bootblock. This area
takes up 256 KiB, about 7 times bigger than our bootblock, but there is no better
choice in this case and output of `cbfstool rom layout` shows that we
additionally include a part of 876 KiB empty space which will hopefully remain
there in future firmware versions (it's a good idea to check before a firmware
update).

## Protection setup

The following command sets the range and enables WP at the same time, the values
are taken from the chosen range above:

```
flashrom --programmer internal --wp-range=0x00fc0000,0x00040000 --wp-enable
```

You can set the range and change WP status independently as well if needed
(just specify one `--wp-*` option at a time). Make sure that hardware
protection is off (state of `W#`/`W/` pin of the chip) or you won't be able to
change WP configuration.

On success, the output of the above command will include such lines:

```
Enabled hardware protection
Activated protection range: start=0x00fc0000 length=0x00040000 (upper 1/64)
```

**Caveat**: `flashrom` automatically tries to disable WP before any operation
on a chip (read, write, erase, verify), so double-check status of WP before
changing state of WP pin on your chip!

## Verifying hardware protection

Once you're happy with the configuration and changed state of WP pin, you can
try disabling WP using `flashrom --dp-disable` to make sure that it fails now.

Comments

Anastasia Klimchuk Nov. 4, 2022, 1:22 a.m. UTC | #1
I only noticed now that the mailing list got lost from the last two
messages. I apologise for that, it was completely unintentional!
Re-sending again, and now including the mailing list. Text below.

On Wed, Nov 2, 2022 at 12:17 AM Sergii Dmytruk <sergii.dmytruk@3mdeb.com> wrote:
>
> On Tue, Nov 01, 2022 at 09:32:27AM +1100, Anastasia Klimchuk wrote:
> > Sergii, you mentioned in the earlier thread, you can convert the text
> > to mediawiki format? That would be great!
> > Also, would you be able to update the wiki page by yourself? We can
> > create an account for you.
> > Then you can preview/correct the text , until the result looks good to you.
> >
> > As we agreed in the earlier thread, this should go to
> > https://www.flashrom.org/Documentation#Using_flashrom section.
> >
> > Thank you so much for your contributions!
>
> Yes, I can deal with MediaWiki and handle formatting issues myself if
> I'll get an account.
>
> Just for information, it's trivial to convert to Wiki automatically:
>
>     pandoc -w mediawiki bootblock-protection.md -o bootblock-protection.wiki
>
> The important thing is to have empty lines before lists in Markdown,
> pandoc doesn't recognize lists otherwise.  There can be other minor
> differences.
Sergii Dmytruk Nov. 4, 2022, 3:59 p.m. UTC | #2
Thanks, updated https://www.flashrom.org/Documentation#Using_flashrom
with links to new pages:

 * https://www.flashrom.org/Firmware_updates_vs._SPI_write-protection
 * https://www.flashrom.org/Example_of_partial_write-protection

Also fixed link to man page source there (although it would be better to
have it rendered, groff isn't very readable).

On Fri, Nov 04, 2022 at 12:22:20PM +1100, Anastasia Klimchuk wrote:
> I only noticed now that the mailing list got lost from the last two
> messages. I apologise for that, it was completely unintentional!
> Re-sending again, and now including the mailing list. Text below.
>
> On Wed, Nov 2, 2022 at 12:17 AM Sergii Dmytruk <sergii.dmytruk@3mdeb.com> wrote:
> >
> > On Tue, Nov 01, 2022 at 09:32:27AM +1100, Anastasia Klimchuk wrote:
> > > Sergii, you mentioned in the earlier thread, you can convert the text
> > > to mediawiki format? That would be great!
> > > Also, would you be able to update the wiki page by yourself? We can
> > > create an account for you.
> > > Then you can preview/correct the text , until the result looks good to you.
> > >
> > > As we agreed in the earlier thread, this should go to
> > > https://www.flashrom.org/Documentation#Using_flashrom section.
> > >
> > > Thank you so much for your contributions!
> >
> > Yes, I can deal with MediaWiki and handle formatting issues myself if
> > I'll get an account.
> >
> > Just for information, it's trivial to convert to Wiki automatically:
> >
> >     pandoc -w mediawiki bootblock-protection.md -o bootblock-protection.wiki
> >
> > The important thing is to have empty lines before lists in Markdown,
> > pandoc doesn't recognize lists otherwise.  There can be other minor
> > differences.
diff mbox series

Patch

diff --git a/bootblock-protection.md b/bootblock-protection.md
index fafb32e..66a6bb2 100644
--- a/bootblock-protection.md
+++ b/bootblock-protection.md
@@ -11,7 +11,7 @@  motherboard (PCH protection).
 
 ## `flashrom` version
 
-At the time of writing (11 October 2022) there hasn't been a `flashrom` release
+At the time of writing (26 October 2022) there hasn't been a `flashrom` release
 that includes WP manipulation facilities. You might have to build one from
 scratch (assuming you've already installed build dependencies):
 
@@ -217,5 +217,5 @@  changing state of WP pin on your chip!
 
 ## Verifying hardware protection
 
-Once you've happy with the configuration and changed state of WP pin, you can
-try disabling WP using `flashrom` to make sure that it fails now.
+Once you're happy with the configuration and changed state of WP pin, you can
+try disabling WP using `flashrom --dp-disable` to make sure that it fails now.
diff --git a/heads-and-wp.md b/heads-and-wp.md
index 70a80fd..0f5485c 100644
--- a/heads-and-wp.md
+++ b/heads-and-wp.md
@@ -64,7 +64,8 @@  If you need to pass extra parameters to flash your chip (e.g., programmer or
 chip name), add them to the above command (order of such parameters shouldn't
 matter).
 
-Mind that in `flashrom` the code for disabling protection automatically is
-different from the code which handles `--wp-disable`. This means that they can
-sometimes produce different results, this is why you should be aware of both
-code paths that affect WP.
+Mind that as of now (26 October 2022) `flashrom` uses different code for
+temporarily disabling protection automatically before an operation (the state
+is restored afterwards) and for handling `--wp-disable`. This means that effects
+on write protection can be different depending on which code changes it, which
+is why you should know about existence of both of these code paths.