pagetables: allow user and kernel maps in same pgd

KTF allows user mode to allocate high addresses, which normally are only
available to kernel. However, to handle this somewhat gracefully, we need
to check if already-existent page table protection bits conflicts with
those of a new pgentry.

This means that we may fixup NX, RW or USER flags if the new mapping is
more permissive than current ones. This is a bit sketchy, yet useful for
certains tests. Therefore, print a big warning if this ever happens.

Signed-off-by: Johannes Wikner <[email protected]>

arch/x86/pagetables.c

@@ -128,6 +128,23 @@ static mfn_t get_cr3_mfn(cr3_t *cr3_entry) {
     return cr3_entry->mfn;
 }
 
+static inline void pgentry_fixup_flags(pgentry_t *entry, unsigned long flags) {
+    /* Our new flags will take precedence over previous one.  */
+    unsigned long entry_new = *entry;
+    entry_new |= flags & _PAGE_USER;
+    entry_new |= flags & _PAGE_RW;
+    entry_new &= ~(flags & _PAGE_NX);
+    if (unlikely(*entry != entry_new)) {
+        char flags_str_old[16];
+        char flags_str_new[16];
+        printk("WARNING: Already-present page table protection conflicts with our.\n"
+               "         Updating present page table protection: %s -> %s\n",
+               dump_pte_flags(flags_str_old, 16, (pte_t) *entry),
+               dump_pte_flags(flags_str_new, 16, (pte_t) entry_new));
+        *entry = entry_new;
+    }
+}
+
 static mfn_t get_pgentry_mfn(mfn_t tab_mfn, pt_index_t index, unsigned long flags) {
     pgentry_t *tab, *entry;
     mfn_t mfn;
@@ -147,6 +164,10 @@ static mfn_t get_pgentry_mfn(mfn_t tab_mfn, pt_index_t index, unsigned long flag
         tab = init_map_mfn(mfn);
         memset(tab, 0, PAGE_SIZE);
     }
+    else {
+        /* Page table already exists but its flags may conflict with our. Maybe fixup */
+        pgentry_fixup_flags(entry, flags);
+    }
 
     return mfn;
 }

tests/unittests.c

@@ -33,6 +33,7 @@
 #include <test.h>
 #include <usermode.h>
 
+#include <mm/pmm.h>
 #include <smp/smp.h>
 
 static char opt_string[4];
@@ -97,6 +98,12 @@ static unsigned long __user_text test_user_task_func3(void *arg) {
     return -16;
 }
 
+#define HIGH_USER_PTR _ptr(0xffffffff80222000)
+static unsigned long __user_text test_user_task_func4(void *arg) {
+    printf(USTR("access: %lx\n"), _ul(HIGH_USER_PTR));
+    return *(unsigned long *) HIGH_USER_PTR;
+}
+
 int unit_tests(void *_unused) {
     printk("\nLet the UNITTESTs begin\n");
     printk("Commandline parsing: %s\n", kernel_cmdline);
@@ -165,20 +172,24 @@ int unit_tests(void *_unused) {
     cpu_freq_expect("Prototyp Amazing Foo One @ 1GHz", 1000000000);
     cpu_freq_expect("Prototyp Amazing Foo Two @ 1.00GHz", 1000000000);
 
-    task_t *task1, *task2, *task_user1, *task_user2, *task_user3;
+    task_t *task1, *task2, *task_user1, *task_user2, *task_user3, *task_user4;
 
     task1 = new_kernel_task("test1", test_kernel_task_func, _ptr(98));
     task2 = new_kernel_task("test2", test_kernel_task_func, _ptr(-99));
     task_user1 = new_user_task("test1 user", test_user_task_func1, NULL);
     task_user2 = new_user_task("test2 user", test_user_task_func2, NULL);
     task_user3 = new_user_task("test3 user", test_user_task_func3, NULL);
+    task_user4 = new_user_task("test4 user", test_user_task_func4, NULL);
+
+    vmap_user_4k(HIGH_USER_PTR, get_free_frame()->mfn, L1_PROT_USER);
 
     set_task_repeat(task1, 10);
     schedule_task(task1, get_bsp_cpu());
     schedule_task(task2, get_cpu(1));
     schedule_task(task_user1, get_bsp_cpu());
     schedule_task(task_user2, get_cpu(1));
     schedule_task(task_user3, get_bsp_cpu());
+    schedule_task(task_user4, get_cpu(1));
 
     printk("Long mode to real mode transition:\n");
     long_to_real();