summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorArthur Heymans <arthur@aheymans.xyz>2019-11-20 23:44:25 +0100
committerPatrick Georgi <pgeorgi@google.com>2020-04-14 09:55:47 +0000
commit15161d92842e95f4b8ab10e4b1be118756fde437 (patch)
treedd83d5a15dbec78b68e0a813ac21eec8dbf6a346
parent26ea43a5c2fb40e8d1235f8106a99533057f79eb (diff)
4.12 release notes: Add some explanation behind deprecations
Some features are made mandatory, meaning that some platforms have been dropped from master. This also explains that further development on these popular platforms can happen on the 4.11 branch. TODO is this really the right place or is it too technical for release notes? Change-Id: I95e01c301e7db6f81ef88a89d709ebab35c9ccfb Signed-off-by: Arthur Heymans <arthur@aheymans.xyz> Reviewed-on: https://review.coreboot.org/c/coreboot/+/37064 Tested-by: build bot (Jenkins) <no-reply@coreboot.org> Reviewed-by: Angel Pons <th3fanbus@gmail.com> Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net> Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com> Reviewed-by: David Hendricks <david.hendricks@gmail.com>
-rw-r--r--Documentation/releases/coreboot-4.12-relnotes.md63
1 files changed, 63 insertions, 0 deletions
diff --git a/Documentation/releases/coreboot-4.12-relnotes.md b/Documentation/releases/coreboot-4.12-relnotes.md
index 7943aa7161..b172c4a92e 100644
--- a/Documentation/releases/coreboot-4.12-relnotes.md
+++ b/Documentation/releases/coreboot-4.12-relnotes.md
@@ -10,6 +10,69 @@ notes.
* The chip and board additions and removals will be updated right
before the release, so those do not need to be added.
+Deprecations
+------------
+
+For the 4.12 release a few features on x86 became mandatory. These are
+relocatable ramstage, postcar stage and C_ENVIRONMENT_BOOTBLOCK.
+
+### Relocatable ramstage
+
+Relocatable stages are a feature implemented only on x86, where stages
+can be relocated at runtime. This is used to place ramstage in a better
+location that does not collide with memory the OS or the payload tends
+to use. The rationale behind making this mandatory is that you always
+want cbmem to be cached so it's a good location to run ramstage from.
+It avoids using lower memory altogether so the OS can make use of it
+and no backing up needs to happen on S3 resume.
+
+### Postcar stage
+
+With Postcar stage tearing down Cache-as-Ram is done in a separate
+stage. This means that romstage has a clean program boundary and
+that all variables in romstage can be accessed via their linked
+addresses without runtime resolution. There is no need to link
+global and static variables via the CAR\_GLOBAL macro and no need
+to access them with car\_set/get\_var/ptr functions.
+
+### C\_ENVIRONMENT\_BOOTBLOCK
+
+Historically the bootblock on x86 platforms has been compiled with
+romcc. This means that the generated code only uses CPU registers
+and therefore no stack. This 20K+ LOC compiler is limited and hard
+to maintain and so is the code that one has to write in that
+environment. A different solution is to set up Cache-as-Ram in the
+bootblock and run GCC compiled code in the bootblock. The advantages
+are increased flexibility and consistency with other architectures as
+well as other stages: e.g. printing to console is possible and
+VBOOT can run before romstage, making romstage updatable via RW FMAP
+regions.
+
+### Platforms dropped from master
+
+The following platforms did not implement those feature are dropped
+from master to allow the master branch to move on:
+- AMDFAM10
+- all FSP1.0 platforms: BROADWELL_DE, FSP_BAYTRAIL, RANGELEY
+- VIA VX900
+- TODO (AMD?)
+
+In particular on FSP1.0 it is impossible to implement POSTCAR stage.
+The reason is that FSP1.0 relocates the CAR region to the HOB before
+returning to coreboot. This means that after FSP returns to coreboot
+accessing variables via their original address is not possible. One
+way of obtaining that behavior would be to set up Cache-as-Ram again
+(but with open source code) and copy the relocated data from the HOB
+there. This solution is deemed too hacky. Maybe a lesson can be
+learned from this: blobs should not interfere with the execution
+environment, as this makes proper integration much harder.
+
+### 4.11_branch
+
+Given that some platforms supported by FSP1.0 are being produced and
+popular, the 4.11 release was made into a branch in which further
+development can happen.
+
Significant changes
-------------------