Unify MMU translation APIs #533
Conversation
ChinYikMing
force-pushed
the
unify-mmu-trans
branch
from
ff5d4d7 to
88f83c8
Compare
src/riscv.h
Outdated
| * ifetch do not leverage this translation because basic block | ||
| * might be retranslated and the corresponding PTE is NULL. | ||
| */ | ||
| typedef riscv_word_t (*riscv_mem_trans)(riscv_t *rv, |
There was a problem hiding this comment.
Specify _t suffix for typedef.
There was a problem hiding this comment.
I am not sure if riscv_mem_trans is meaningful enough. How about mmu_translate_t or mem_translate_t?
There was a problem hiding this comment.
Specify
_tsuffix for typedef.
For only this one? What about the others?
There was a problem hiding this comment.
I am not sure if
riscv_mem_transis meaningful enough. How aboutmmu_translate_tormem_translate_t?
Use riscv prefix for consistency with others. I would go with riscv_mem_translate_t, is it OK?
There was a problem hiding this comment.
Use
riscvprefix for consistency with others. I would go withriscv_mem_translate_t, is it OK?
It looks good.
The reason not to use "trans" is that it is an existing adjective which might not be intended to mean "translate." See https://dictionary.cambridge.org/dictionary/english/trans
To enable guest OS to run SDL applications using bidirectional queues,
the VMM or system emulator must access the guest OS's virtual memory,
as the queues are defined there. This requires MMU translation. In
addition, the queues event data size likely larger than a word (maybe a
page), thus the existing rv->io.mem_{read,write}_w do not sufficient for
such memory manipulation.
Export rv->io.mem_translate() to perform gVA2gPA translation. Unifying
MMU translation logic by using rv->io.mem_translate(), eliminating
redundant flows:
mmu_walk -> check_pg_fault -> check_signal -> get_ppn_and_offset.
Also, this interface allowing src/syscall_sdl.c to handle virtual
memory by first translating the gVA2gPA, followed by memory_{read/write}
the chunk of data at once.
TODO: dTLB can be introduced in rv->io.mem_trans() to cache the gVA2gPA
translation.
See: sysprog21#310, sysprog21#510
ChinYikMing
force-pushed
the
unify-mmu-trans
branch
from
88f83c8 to
dfab12e
Compare
|
Thank @ChinYikMing for contributing! |
To enable guest OS to run SDL applications using bidirectional queues, the VMM or system emulator must access the guest OS's virtual memory, as the queues are defined there. This requires MMU translation. In addition, the queues event data size likely larger than a word (maybe a page), thus the existing rv->io.mem_{read,write}_w do not sufficient for such memory manipulation.
Export rv->io.mem_translate() to perform gVA2gPA translation. Unifying MMU translation logic by using
rv->io.mem_translate(), eliminating redundant flows: mmu_walk -> check_pg_fault -> check_signal -> get_ppn_and_offset.
Also, this interface allowing src/syscall_sdl.c to handle virtual memory by first translating the gVA2gPA, followed by memory_{read/write} the chunk of data at once.
TODO: dTLB can be introduced in rv->io.mem_trans() to cache the gVA2gPA translation.
See: #310, #510