Skip to content

Clear ARM64 instruction cache on mprotect exec #344

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
doanbaotrung wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Clear ARM64 instruction cache on mprotect exec #344

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions src/syscall/mem.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4540,6 +4540,11 @@ int64_t sys_mprotect(guest_t *g, uint64_t addr, uint64_t length, int prot)
return -LINUX_ENOMEM;
}
guest_region_set_prot(g, addr, mprot_end, prot);
if (prot & LINUX_PROT_EXEC) {

@cubic-dev-ai cubic-dev-ai Bot Aug 31, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1: The icache flush is added only in the high-VA branch (addr >= g->guest_size), but ordinary guest memory (text/heap/mmap below guest_size, which is the 64 GiB to 1 TiB primary window where ART JIT buffers live) is handled by the second mprot_off branch below, which never calls __builtin___clear_cache. For any mprotect(PROT_EXEC) at a normal guest VA, sys_mprotect returns without invalidating the instruction cache, so the stale-instruction bug the PR targets remains unfixed in the common case.

Prompt for AI agents 
Check if this issue is valid — if so, understand the root cause and fix it. At src/syscall/mem.c, line 4543:

<comment>The icache flush is added only in the high-VA branch (`addr >= g->guest_size`), but ordinary guest memory (text/heap/mmap below `guest_size`, which is the 64 GiB to 1 TiB primary window where ART JIT buffers live) is handled by the second `mprot_off` branch below, which never calls `__builtin___clear_cache`. For any mprotect(PROT_EXEC) at a normal guest VA, `sys_mprotect` returns without invalidating the instruction cache, so the stale-instruction bug the PR targets remains unfixed in the common case.</comment>

<file context>
@@ -4540,6 +4540,11 @@ int64_t sys_mprotect(guest_t *g, uint64_t addr, uint64_t length, int prot)
                     return -LINUX_ENOMEM;
             }
             guest_region_set_prot(g, addr, mprot_end, prot);
+            if (prot & LINUX_PROT_EXEC) {
+                char *host_start = (char *) g->host_base + addr;
+                char *host_end = (char *) g->host_base + mprot_end;
</file context>

char *host_start = (char *) g->host_base + addr;

@cubic-dev-ai cubic-dev-ai Bot Aug 31, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2: In the high-VA branch the range passed to __builtin___clear_cache is computed as g->host_base + addr, but pages with addr >= guest_size are not backed at host_base + addr. Per the comment at mem.c:1439-1441 and host_ptr_for_gpa (mem.c:796), high-VA pages live in named-mapping/overflow-segment host buffers, so this flushes the wrong (primary-window) memory and leaves the actual instruction cache untouched. Resolve the host range with host_ptr_for_gpa(g, addr) / host_ptr_for_gpa(g, mprot_end) instead.

Prompt for AI agents 
Check if this issue is valid — if so, understand the root cause and fix it. At src/syscall/mem.c, line 4544:

<comment>In the high-VA branch the range passed to `__builtin___clear_cache` is computed as `g->host_base + addr`, but pages with `addr >= guest_size` are not backed at `host_base + addr`. Per the comment at mem.c:1439-1441 and `host_ptr_for_gpa` (mem.c:796), high-VA pages live in named-mapping/overflow-segment host buffers, so this flushes the wrong (primary-window) memory and leaves the actual instruction cache untouched. Resolve the host range with `host_ptr_for_gpa(g, addr)` / `host_ptr_for_gpa(g, mprot_end)` instead.</comment>

<file context>
@@ -4540,6 +4540,11 @@ int64_t sys_mprotect(guest_t *g, uint64_t addr, uint64_t length, int prot)
             }
             guest_region_set_prot(g, addr, mprot_end, prot);
+            if (prot & LINUX_PROT_EXEC) {
+                char *host_start = (char *) g->host_base + addr;
+                char *host_end = (char *) g->host_base + mprot_end;
+                __builtin___clear_cache(host_start, host_end);
</file context>

char *host_end = (char *) g->host_base + mprot_end;
__builtin___clear_cache(host_start, host_end);
}
return 0;
}
uint64_t mprot_off = addr - g->ipa_base;
Expand Down
Loading