meta-virtualization/recipes-extended/xen/files/xen-arm64-implement-atomic-fetch-add.patch
Christopher Clark 5bf7fae581 xen: add patch to fix build on ARM64 with gcc 10.1.0
The newer gcc toolchain won't link due to a missing builtin:
__sync_fetch_and_add. Xen is built with -fno-builtin, so add a
patch to implement the one required function.

Signed-off-by: Christopher Clark <christopher.w.clark@gmail.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
2020-07-06 16:25:56 -04:00

51 lines
1.9 KiB
Diff

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Christopher Clark <christopher.w.clark@gmail.com>
Date: Fri, 26 June 2020 16:34:00 -0800
Subject: [PATCH] xen: implement atomic op to fix arm64 compilation
Xen's ARM implementation of arch_fetch_and_add since f9cc3cd9
uses a builtin, despite the build being performed with -fno-builtin.
With gcc 10.1.0, this now breaks prelinking spinlock.c, so
implement the one atomic operation that is required with logic
derived from Linux's atomic_ll_sc.h: ATOMIC_FETCH_OP and comparison with
the binary produced with and without the patch with gcc 9.2.0.
Signed-off-by: Christopher Clark <christopher.w.clark@gmail.com>
diff --git a/xen/include/asm-arm/system.h b/xen/include/asm-arm/system.h
index e5d062667d..c46dd3ac71 100644
--- a/xen/include/asm-arm/system.h
+++ b/xen/include/asm-arm/system.h
@@ -55,7 +55,32 @@ static inline int local_abort_is_enabled(void)
return !(flags & PSR_ABT_MASK);
}
+#ifdef CONFIG_ARM_64
+
+/* see atomic_ll_sc.h: ATOMIC_FETCH_OP(name, mb, acq, rel, cl, op, asm_op, constraint) */
+static inline int arch_fetch_and_add(unsigned int *ptr, unsigned long i)
+{
+ int register lptr asm("x0");
+ int register result asm("w1");
+ int register newval asm("w2");
+ int register status asm("w3");
+
+ asm volatile(
+ " mov %[lptr], %[ptr]\n"
+ "1: ldxr %w[result], [%[lptr]]\n"
+ " add %w[newval], %w[result], %w[i]\n"
+ " stlxr %w[status], %w[newval], [%[lptr]]\n"
+ " cbnz %w[status], 1b\n"
+ " dmb ish\n"
+ : [result] "=&r" (result), [lptr] "=&r" (lptr), [newval] "=&r" (newval), [status] "=&r" (status), [i] "+r" (i), "+Q" (*ptr)
+ : [ptr] "r" (ptr), "r" (i)
+ : "memory");
+
+ return result;
+}
+#else
#define arch_fetch_and_add(x, v) __sync_fetch_and_add(x, v)
+#endif
extern struct vcpu *__context_switch(struct vcpu *prev, struct vcpu *next);