androm3da added a comment.
This patch spans several areas, so I would be happy to decompose it into
smaller parts if it made it easier for reviewers.
================
Comment at: compiler-rt/lib/xray/xray_hexagon.cpp:76-84
+static void WriteInstFlushCache(void *Addr, uint32_t NewInstruction) {
+ asm volatile("icinva(%[inst_addr])\n\t"
+ "isync\n\t"
+ "memw(%[inst_addr]) = %[new_inst]\n\t"
+ "dccleaninva(%[inst_addr])\n\t"
+ "syncht\n\t"
+ :
----------------
If we need to support cases where the entire sled spans cache lines, then we
must iterate over the icinva() instead.
================
Comment at: compiler-rt/lib/xray/xray_hexagon.cpp:76-84
+static void WriteInstFlushCache(void *Addr, uint32_t NewInstruction) {
+ asm volatile("icinva(%[inst_addr])\n\t"
+ "isync\n\t"
+ "memw(%[inst_addr]) = %[new_inst]\n\t"
+ "dccleaninva(%[inst_addr])\n\t"
+ "syncht\n\t"
+ :
----------------
androm3da wrote:
> If we need to support cases where the entire sled spans cache lines, then we
> must iterate over the icinva() instead.
>
Other arches do atomic stores. I think we will need to change this to: LL
first, then icinva/isync, then SC.
If the SC fails then either we re-run all those steps because we want to
clobber it or we do nothing because we assume some other thread applied the
same patch first?
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D113638/new/
https://reviews.llvm.org/D113638
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
