[Barrelfish-users] Two problems in arm appcore booting code

[Wang Nan]

Hi,

I found 2 problems when booting armv7 appcores:

1:

After appcore bootup, monitor unmap cpu_buf_memory AFTER the cap destroied, which will cause cap-not-found error.

I suggest following patch, which unmap the buffer BEFORE destroy the cap.

diff --git a/usr/monitor/arch/armv7/boot.c b/usr/monitor/arch/armv7/boot.c
index 876b475..885c1ee 100644
--- a/usr/monitor/arch/armv7/boot.c
+++ b/usr/monitor/arch/armv7/boot.c
@@ -392,6 +392,12 @@ errval_t spawn_xcore_monitor(coreid_t coreid, int hwid, enum cpu_type cpu_type,


     /* Clean up */ // XXX: Should not delete the remote cap
+
+    err = vspace_unmap(cpu_buf_memory);
+    if (err_is_fail(err)) {
+        USER_PANIC_ERR(err, "vspace unmap CPU driver memory failed");
+    }
+
     err = cap_destroy(cpu_memory_cap);
     if (err_is_fail(err)) {
         USER_PANIC_ERR(err, "cap_destroy failed");
@@ -400,10 +406,6 @@ errval_t spawn_xcore_monitor(coreid_t coreid, int hwid, enum cpu_type cpu_type,
     if (err_is_fail(err)) {
         USER_PANIC_ERR(err, "cap_destroy failed");
     }
-    err = vspace_unmap(cpu_buf_memory);
-    if (err_is_fail(err)) {
-        USER_PANIC_ERR(err, "vspace unmap CPU driver memory failed");
-    }

     return SYS_ERR_OK;
 }


2.

monitor use ram_set_affinity before allocing cpu_memory_cap, which I believe is to make cpu driver to reside in directly mapped physical memory. But spawn_memory_cap is freely allocated, and not
guarantee to reside in directly mapped memory.

I suggest to use ram_set_affinity before following line:

err = frame_alloc(&spawn_memory_cap, ARM_CORE_DATA_PAGES * arch_page_size,
                      NULL);