summaryrefslogtreecommitdiff
path: root/src/arch/x86/include/mode_switch.h
diff options
context:
space:
mode:
authorJeremy Compostella <jeremy.compostella@intel.com>2023-08-30 15:42:09 -0700
committerJulius Werner <jwerner@chromium.org>2023-09-14 21:02:07 +0000
commitb7832de0260b042c25bf8f53abcb32e20a29ae9c (patch)
treedea12cf2118d848e1681254d52fa26a15fc008b4 /src/arch/x86/include/mode_switch.h
parent79f2e1fc8b6192e96f99c05f71baeb77d4633d40 (diff)
x86: Add .data section support for pre-memory stages
x86 pre-memory stages do not support the `.data` section and as a result developers are required to include runtime initialization code instead of relying on C global variable definition. To illustrate the impact of this lack of `.data` section support, here are two limitations I personally ran into: 1. The inclusion of libgfxinit in romstage for Raptor Lake has required some changes in libgfxinit to ensure data is initialized at runtime. In addition, we had to manually map some `.data` symbols in the `_bss` region. 2. CBFS cache is currently not supported in pre-memory stages and enabling it would require to add an initialization function and find a generic spot to call it. Other platforms do not have that limitation. Hence, resolving it would help to align code and reduce compilation based restriction (cf. the use of `ENV_HAS_DATA_SECTION` compilation flag in various places of coreboot code). We identified three cases to consider: 1. eXecute-In-Place pre-memory stages - code is in SPINOR - data is also stored in SPINOR but must be linked in Cache-As-RAM and copied there at runtime 2. `bootblock` stage is a bit different as it uses Cache-As-Ram but the memory mapping and its entry code different 3. pre-memory stages loaded in and executed from Cache-As-RAM (cf. `CONFIG_NO_XIP_EARLY_STAGES`). eXecute-In-Place pre-memory stages (#1) require the creation of a new ELF segment as the code segment Virtual Memory Address and Load Memory Address are identical but the data needs to be linked in cache-As-RAM (VMA) but to be stored right after the code (LMA). Here is the output `readelf --segments` on a `romstage.debug` ELF binary. Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align LOAD 0x000080 0x02000000 0x02000000 0x21960 0x21960 R E 0x20 LOAD 0x0219e0 0xfefb1640 0x02021960 0x00018 0x00018 RW 0x4 Section to Segment mapping: Segment Sections... 00 .text 01 .data Segment 0 `VirtAddr` and `PhysAddr` are at the same address while they are totally different for the Segment 1 holding the `.data` section. Since we need the data section `VirtAddr` to be in the Cache-As-Ram and its `PhysAddr` right after the `.text` section, the use of a new segment is mandatory. `bootblock` (#2) also uses this new segment to store the data right after the code and load it to Cache-As-RAM at runtime. However, the code involved is different. Not eXecute-In-Place pre-memory stages (#3) do not really need any special work other than enabling a data section as the code and data VMA / LMA translation vector is the same. TEST=#1 and #2 verified on rex and qemu 32 and 64 bits: - The `bootblock.debug`, `romstage.debug` and `verstage.debug` all have data stored at the end of the `.text` section and code to copy the data content to the Cache-As-RAM. - The CBFS stages included in the final image has not improperly relocated any of the `.data` section symbol. - Test purposes global data symbols we added in bootblock, romstage and verstage are properly accessible at runtime #3: for "Intel Apollolake DDR3 RVP1" board, we verified that the generated romstage ELF includes a .data section similarly to a regular memory enabled stage. Change-Id: I030407fcc72776e59def476daa5b86ad0495debe Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com> Reviewed-on: https://review.coreboot.org/c/coreboot/+/77289 Tested-by: build bot (Jenkins) <no-reply@coreboot.org> Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Diffstat (limited to 'src/arch/x86/include/mode_switch.h')
0 files changed, 0 insertions, 0 deletions