- Oct 14, 2021
-
-
Daniel Lustig authored
-
- Oct 06, 2021
-
-
Andrew Waterman authored
Partially introduced by 7f45777b and 94ab414c4b470edae8bde4e7759163fe7ddf8ab4 The spec says that the platform will define which exceptions must populate mtval or stval with a nonzero value and which are permitted to zero it. But then later statements seem not to permit the possibility of zeroing the register, which is a contradiction. Resolve in a backwards-compatible manner by permitting zeros unless the platform says otherwise (which was the original intent, anyway). This permits e.g. zeroing mtval on an EBREAK instruction, where mepc provides sufficient information, as was the intent of 94ab414c4b470edae8bde4e7759163fe7ddf8ab4.
-
- Oct 05, 2021
-
-
Andrew Waterman authored
-
- Oct 02, 2021
-
-
Andrew Waterman authored
-
- Sep 30, 2021
-
-
Andrew Waterman authored
The description of FENCE.TSO as "optional" was incorrect. It has always been a mandatory instruction (because of the requirement that unused patterns in the fm field be treated as though fm=0).
-
- Sep 22, 2021
-
-
Andrew Waterman authored
-
Andrew Waterman authored
-
- Sep 16, 2021
-
-
Andrew Waterman authored
-
Andrew Waterman authored
-
Andrew Waterman authored
-
John Hauser authored
-
John Hauser authored
-
- Sep 15, 2021
-
-
John Hauser authored
-
Scott Johnson authored
-
Andrew Waterman authored
Related to forthcoming Svpbmt extension.
-
Andrew Waterman authored
-
- Sep 11, 2021
-
-
Andrew Waterman authored
-
- Sep 10, 2021
-
-
Andrew Waterman authored
* Changes to SSIP to support forthcoming interrupt controllers * Re: SSIP, say more by saying less
-
Andrew Waterman authored
* Speculative implicit reads, v2 These constraints are better expressed as PMAs, and so we relax the unprivileged spec to always allow speculative implicit reads "unless otherwise specified", then move the constraints into the PMAs. This proposal piggybacks on the non-idempotence PMA, constraining implicit reads to within a power-of-2 region of a non-speculative implicit read. Furthermore, speculative instruction fetch can also access the subsequent power-of-2 region. * Add preface entry
-
Daniel Lustig authored
Thanks to @YenHaoChen. Closes #731
-
- Sep 09, 2021
-
-
Andrew Waterman authored
Add mseccfg and *envcfg CSRs
-
Andrew Waterman authored
-
- Sep 06, 2021
-
-
John Hauser authored
Raise a virtual instruction exception when V = 1 if an attempted instruction is HS-qualified but is explicitly blocked by either a hypervisor CSR or a supervisor CSR. Previously, the document was inconsistent whether an illegal instruction or virtual instruction exception is raised when an instruction is blocked from executing in VU mode solely by a supervisor CSR such as scounteren.
-
- Sep 03, 2021
-
-
Andrew Waterman authored
-
Andrew Waterman authored
Writing the low part does not disturb the high part and vice versa.
-
- Sep 02, 2021
-
-
Andrew Waterman authored
-
John Hauser authored
When different privilege modes have different values for XLEN, the terms _RV32_ and _RV64_ are not sufficient to indicate the width of a privileged CSR that is accessed from a more-privileged mode (e.g. when 32-bit vsatp is accessed from 64-bit HS mode). Where needed, replace "RV32" and "RV64" with more specific references to the correct XLEN.
-
Andrew Waterman authored
-
- Aug 31, 2021
-
-
Andrew Waterman authored
This reverts commit 8551bd20 at @kasanovic's request.
-
Andrew Waterman authored
-
- Aug 30, 2021
-
-
Andrew Waterman authored
-
Andrew Waterman authored
-
Andrew Waterman authored
-
Andrew Waterman authored
-
John Hauser authored
-
Andrew Waterman authored
-
Andrew Waterman authored
-
Andrew Waterman authored
-
- Aug 29, 2021
-
-
Andrew Waterman authored
* Add Smdisc extension (i.e. mconfigptr CSR) * mconfigptr may be hardwired to zero * Add changelog entry for mconfigptr * Fix mconfigptr preface entry "points to the address of" implies an extra level of indirection. Should be "points to" or "contains the address of"
-
John Hauser authored
Make the manual more correct and consistent by dropping the term _execution environment interface_ and its abbreviation _EEI_ and replacing them everywhere with just _execution environment_.
-