riscv: Ensure gp is NULL or points to valid data

This ensures constructs like `if (gd & gd->...) { ... }` work when
accessing the global data pointer. Without this change, it was possible for
a very early trap to cause _exit_trap to directly or indirectly (through
printf) to read arbitrary memory. This could cause a second trap,
preventing show_regs from being printed.

printf (and specifically puts) uses gd to determine what function to print
with. These functions in turn use gd to find the serial device, etc.
However, before accessing gd, puts first checks to see if it is non-NULL.
This indicates an existing (perhaps undocumented) assumption that either gd
is NULL or it is completely valid.

Before this patch, gd either points to unexpected data (because it retains
the value it did from the prior-stage) or points to uninitialized data
(because it has not yet been initialized by board_init_f_init_reserve)
until the hart has acquired available_harts_lock. This can cause two
problems, depending on the value of gd->flags. If GD_FLG_SERIAL_READY is
unset, then some garbage data will be printed to stdout, but there will not
be a second trap. However, if GD_FLG_SERIAL_READY is set, then puts will
try to print with serial_puts, which will likely cause a second trap.

After this patch, gd is zero up until either a hart has set it in
wait_for_gd_init, or until it is set by arch_init_gd. This prevents its
usage before its data is initialized because both handle_trap and puts
ensure that gd is nonzero before using it. After gd has been set, it is OK
to access it because its data has been cleared (and so flags is valid).

XIP cannot use locks because flash is not writable. This leaves it
vulnerable to the same class of bugs regarding already-pending IPIs as
before this series. Fixing that would require finding another method of
synchronization, which is outside the scope of this series.

Fixes: 7c6ca03eae ("riscv: additional crash information")
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>

1 files changed, 25 insertions, 3 deletions

diff --git a/arch/riscv/cpu/start.S b/arch/riscv/cpu/start.S
index 66ca1c70204..eb852538cab 100644
--- a/arch/riscv/cpu/start.S
+++ b/arch/riscv/cpu/start.S
@@ -47,6 +47,13 @@ _start:
 	mv	tp, a0
 	mv	s1, a1
 
+	/*
+	 * Set the global data pointer to a known value in case we get a very
+	 * early trap. The global data pointer will be set its actual value only
+	 * after it has been initialized.
+	 */
+	mv	gp, zero
+
 	la	t0, trap_entry
 	csrw	MODE_PREFIX(tvec), t0
 
@@ -85,10 +92,10 @@ call_board_init_f_0:
 	jal	board_init_f_alloc_reserve
 
 	/*
-	 * Set global data pointer here for all harts, uninitialized at this
-	 * point.
+	 * Save global data pointer for later. We don't set it here because it
+	 * is not initialized yet.
 	 */
-	mv	gp, a0
+	mv	s0, a0
 
 	/* setup stack */
 #if CONFIG_IS_ENABLED(SMP)
@@ -109,6 +116,14 @@ call_board_init_f_0:
 	amoswap.w s2, t1, 0(t0)
 	bnez	s2, wait_for_gd_init
 #else
+	/*
+	 * FIXME: gp is set before it is initialized. If an XIP U-Boot ever
+	 * encounters a pending IPI on boot it is liable to jump to whatever
+	 * memory happens to be in ipi_data.addr on boot. It may also run into
+	 * problems if it encounters an exception too early (because printf/puts
+	 * accesses gd).
+	 */
+	mv	gp, s0
 	bnez	tp, secondary_hart_loop
 #endif
 
@@ -133,6 +148,13 @@ wait_for_gd_init:
 1:	amoswap.w.aq t1, t1, 0(t0)
 	bnez	t1, 1b
 
+	/*
+	 * Set the global data pointer only when gd_t has been initialized.
+	 * This was already set by arch_setup_gd on the boot hart, but all other
+	 * harts' global data pointers gets set here.
+	 */
+	mv	gp, s0
+
 	/* register available harts in the available_harts mask */
 	li	t1, 1
 	sll	t1, t1, tp