Related: #411, #362, #248
Similar to how seL4_BootInfo_t works, we should pass a bootinfo structure in the $a0 register through to init(), which can contain physical and virtual addresses of various memory regions, populated by the capDL initialiser; instead of doing this at build time with ELF patching. This helps especially with x86, where paddr are often unknown at runtime.
void init(microkit_bootinfo_t *bi) {
// etc
}This structure could look something like:
typedef struct _microkit_bootinfo {
paddr_t paddrs[...];
vaddr_t vaddrs[...];
void *preloaded_data_regions[...];
} microkit_bootinfo_t;
and have everything in the system description file have a numeric ID; or alternatively be a bit more wasteful of memory and pass e.g. an array of memory regions which contains the string-name, and id, the vaddr, possibly the paddr, etc. (essentially, serialising the SDF file).
This would also make it a bit easier for non-microkit users to re-use microkit PDs (e.g. Djawula), as this requires no runtime ELF-patching.
Related: #411, #362, #248
Similar to how
seL4_BootInfo_tworks, we should pass a bootinfo structure in the$a0register through to init(), which can contain physical and virtual addresses of various memory regions, populated by the capDL initialiser; instead of doing this at build time with ELF patching. This helps especially with x86, where paddr are often unknown at runtime.This structure could look something like:
and have everything in the system description file have a numeric ID; or alternatively be a bit more wasteful of memory and pass e.g. an array of memory regions which contains the string-name, and id, the vaddr, possibly the paddr, etc. (essentially, serialising the SDF file).
This would also make it a bit easier for non-microkit users to re-use microkit PDs (e.g. Djawula), as this requires no runtime ELF-patching.