The spec says "The PMP access control bits shall be the same for all
physical addresses in the cache block [... else] the behavior of a CBO
instruction is UNSPECIFIED."
Thus, we only need to check the byte rs1 points to
(instead of the entire cache block).
This change was made ages ago in the spec.
I did not actually test that the new privilege checks in ebreak and
c.ebreak are correct, but all the existing debug tests still pass.
The etrigger match on exceptions doesn't work properly in cases like
the following:
1) M-mode delegates ECALLs to S-mode
2) A CPU hardware point mechanism is used to place a breakpoint on the
Umode instruction that executes the ECALL from Umode to Smode. In
effect, this creates a breakpoint etrigger based on Umode.
In the above, the expectation is that #2 will first cause an exit to
the Smode handler (stvec), and the hardware breakpoint exception will
be triggered following an entry into the handler.
However, since etrigger currently checks the current privilege mode, we
will never get a match on conditions like #2.
The patch attempts to address the issue by using the stashed version of
the previous privilege mode for the etrigger match.
cc: YenHaoChen <howard25336284@gmail.com>
Signed-off-by: Atul Khare <atulkhare@rivosinc.com>
The current version of mode_match() is based on the current privilege
level. This adds an explicit privilege and virtual mode parameters in
anticipation of an upcoming patch for matching trap triggers.
This adds the prev_v field to track the previous virtual mode state. We
also assign it unconditionally to handle cases for trigger matching
like the following (pointed out by Scott Johnson):
1) SRET from HS to VU: prev_v is set to 0
2) Trap from VU to VS: state.v/prev_v won't be assigned because of
unchanged v, and remain 0.
3) An etrigger that's set to break on a VU-mode trap won't match
properly because prev_v is incorrect
This be used in a forthcoming patch for trigger matching.
This commit lets the mstatus.MPP be a valid value if unsupporting
U-mode. Without this commit, the mret may result in a corrupted state
without properly setting the MPP to M-mode (if unsupporting U-mode).
This commit adds the *.pc files for Spike's simulation library,
enabling dynamic and static linking without the need to directly
reference Spike sources. Using Spike as a stand-alone library
provides an interesting option for developing tools
and applications based on Spike.
Andrew Waterman
YenHaoChen
demin.han
Weiwei Li
Tim Newsome
Jerry Zhao
Gianluca Guida
Scott Johnson
Atul Khare
Wojciech Bartczak
rbuchner