selftests/bpf: Reject graph kptr use after RCU unlockgrafted

Add a sleepable verifier test that loads a graph-node local kptr in an
explicit RCU read-side critical section, then passes its node to
bpf_rbtree_remove() after the section ends.

Before the verifier fix, the stale NON_OWN_REF flag makes the node look like
a live borrowed reference and the program is accepted. After the fix, the
pointer is demoted without NON_OWN_REF and the graph kfunc argument is
rejected.

Also exercise a graph kptr loaded while a spin lock provides implicit RCU
protection. The pointer must be invalidated when the lock is released, which
guards the required ordering between non-owning-reference invalidation and
RCU demotion.

Update the existing fault-protected load test state description. The
post-unlock pointer no longer carries NON_OWN_REF, but remains readable
because the load is rewritten to use BPF_PROBE_MEM.

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Link: https://lore.kernel.org/r/20260904084325.52250-7-memxor@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

1 files changed, 73 insertions, 0 deletions

diff --git a/arch/arc/mm/highmem.c b/arch/arc/mm/highmem.c
new file mode 100644
index 000000000..c79912a6b
--- /dev/null
+++ b/arch/arc/mm/highmem.c
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2015 Synopsys, Inc. (www.synopsys.com)
+ */
+
+#include <linux/memblock.h>
+#include <linux/export.h>
+#include <linux/highmem.h>
+#include <linux/pgtable.h>
+#include <asm/processor.h>
+#include <asm/pgalloc.h>
+#include <asm/tlbflush.h>
+
+/*
+ * HIGHMEM API:
+ *
+ * kmap() API provides sleep semantics hence referred to as "permanent maps"
+ * It allows mapping LAST_PKMAP pages, using @last_pkmap_nr as the cursor
+ * for book-keeping
+ *
+ * kmap_atomic() can't sleep (calls pagefault_disable()), thus it provides
+ * shortlived ala "temporary mappings" which historically were implemented as
+ * fixmaps (compile time addr etc). Their book-keeping is done per cpu.
+ *
+ *	Both these facts combined (preemption disabled and per-cpu allocation)
+ *	means the total number of concurrent fixmaps will be limited to max
+ *	such allocations in a single control path. Thus KM_TYPE_NR (another
+ *	historic relic) is a small'ish number which caps max percpu fixmaps
+ *
+ * ARC HIGHMEM Details
+ *
+ * - the kernel vaddr space from 0x7z to 0x8z (currently used by vmalloc/module)
+ *   is now shared between vmalloc and kmap (non overlapping though)
+ *
+ * - Both fixmap/pkmap use a dedicated page table each, hooked up to swapper PGD
+ *   This means each only has 1 PGDIR_SIZE worth of kvaddr mappings, which means
+ *   2M of kvaddr space for typical config (8K page and 11:8:13 traversal split)
+ *
+ * - The fixed KMAP slots for kmap_local/atomic() require KM_MAX_IDX slots per
+ *   CPU. So the number of CPUs sharing a single PTE page is limited.
+ *
+ * - pkmap being preemptible, in theory could do with more than 256 concurrent
+ *   mappings. However, generic pkmap code: map_new_virtual(), doesn't traverse
+ *   the PGD and only works with a single page table @pkmap_page_table, hence
+ *   sets the limit
+ */
+
+extern pte_t * pkmap_page_table;
+
+static noinline pte_t * __init alloc_kmap_pgtable(unsigned long kvaddr)
+{
+	pmd_t *pmd_k = pmd_off_k(kvaddr);
+	pte_t *pte_k;
+
+	pte_k = (pte_t *)memblock_alloc_low(PAGE_SIZE, PAGE_SIZE);
+	if (!pte_k)
+		panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
+		      __func__, PAGE_SIZE, PAGE_SIZE);
+
+	pmd_populate_kernel(&init_mm, pmd_k, pte_k);
+	return pte_k;
+}
+
+void __init kmap_init(void)
+{
+	/* Due to recursive include hell, we can't do this in processor.h */
+	BUILD_BUG_ON(PAGE_OFFSET < (VMALLOC_END + FIXMAP_SIZE + PKMAP_SIZE));
+	BUILD_BUG_ON(LAST_PKMAP > PTRS_PER_PTE);
+	BUILD_BUG_ON(FIX_KMAP_SLOTS > PTRS_PER_PTE);
+
+	pkmap_page_table = alloc_kmap_pgtable(PKMAP_BASE);
+	alloc_kmap_pgtable(FIXMAP_BASE);
+}