Re: [PATCH] MIPS: fix boot with DT passed via UHI

From: David Daney
Date: Wed Jun 07 2017 - 18:47:21 EST


On 06/07/2017 06:16 AM, Daniel Schwierzeck wrote:


Am 06.06.2017 um 21:16 schrieb Andrea Merello:
commit 15f37e158892 ("MIPS: store the appended dtb address in a variable")
seems to have introduced code that relies on delay slots after branch,
however it seems that, since no directive ".set noreorder" is present, the
AS already fills delay slots with NOPs.

This caused failure in assigning proper DT blob address to fw_passed_dtb
variable, causing failure when booting passing DT via UHI; this has been
seen on a Lantiq VR9 SoC (Fritzbox 3370) and u-boot as bootloader.

[ 0.000000] Linux version 4.12.0-fritz+ (andrea@horizon) (gcc version 4.9.0 (GCC) ) #29 SMP Tue Jun 6 20:49:59 CEST 2017
[ 0.000000] SoC: xRX200 rev 1.2
[ 0.000000] bootconsole [early0] enabled
[ 0.000000] CPU0 revision is: 00019556 (MIPS 34Kc)
[ 0.000000] Determined physical RAM map:
[ 0.000000] memory: 00696000 @ 00002000 (usable)
[ 0.000000] memory: 00038000 @ 00698000 (usable after init)
[ 0.000000] Wasting 64 bytes for tracking 2 unused pages
[ 0.000000] Kernel panic - not syncing: No memory area to place a bootmap bitmap
[ 0.000000] Rebooting in 1 seconds..
[ 0.000000] Reboot failed -- System halted

This patch moves the instruction meant to be placed in the delay slot
before the preceding BEQ instruction, while the delay slot will be
filled with a NOP by the AS.

After this patch the kernel fetches the DR correctly

[ 0.000000] Linux version 4.12.0-fritz+ (andrea@horizon) (gcc version 4.9.0 (GCC) ) #30 SMP
Tue Jun 6 20:52:40 CEST 2017
[ 0.000000] SoC: xRX200 rev 1.2
[ 0.000000] bootconsole [early0] enabled
[ 0.000000] CPU0 revision is: 00019556 (MIPS 34Kc)
[ 0.000000] MIPS: machine is FRITZ3370 - Fritz!Box WLAN 3370
[ 0.000000] Determined physical RAM map:
[ 0.000000] memory: 08000000 @ 00000000 (usable)
[ 0.000000] Detected 1 available secondary CPU(s)
[ 0.000000] Primary instruction cache 32kB, VIPT, 4-way, linesize 32 bytes.
[ 0.000000] Primary data cache 32kB, 4-way, VIPT, cache aliases, linesize 32 bytes
[ 0.000000] Zone ranges:
[ 0.000000] Normal [mem 0x0000000000000000-0x0000000007ffffff]
[ 0.000000] Movable zone start for each node
[ 0.000000] Early memory node ranges
[ 0.000000] node 0: [mem 0x0000000000000000-0x0000000007ffffff]
[ 0.000000] Initmem setup node 0 [mem 0x0000000000000000-0x0000000007ffffff]
[ 0.000000] percpu: Embedded 15 pages/cpu @8110c000 s30176 r8192 d23072 u61440
[ 0.000000] Built 1 zonelists in Zone order, mobility grouping on. Total pages: 32512
[ 0.000000] Kernel command line: rootwait root=/dev/sda1 console=ttyLTQ0
...

Cc: linux-kernel@xxxxxxxxxxxxxxx
Cc: Jonas Gorski <jogo@xxxxxxxxxxx>
Cc: Daniel Schwierzeck <daniel.schwierzeck@xxxxxxxxx>
Signed-off-by: Andrea Merello <andrea.merello@xxxxxxxxx>

diff --git a/arch/mips/kernel/head.S b/arch/mips/kernel/head.S
index cf05220..d1bb506 100644
--- a/arch/mips/kernel/head.S
+++ b/arch/mips/kernel/head.S
@@ -106,8 +106,8 @@ NESTED(kernel_entry, 16, sp) # kernel entry point
beq t0, t1, dtb_found
#endif
li t1, -2
- beq a0, t1, dtb_found
move t2, a1
+ beq a0, t1, dtb_found
li t2, 0
dtb_found:


The fix looks correct. Without ".set noreorder" one should not

s/should/must/

manually
put instructions in the delay slot. This should be left to the AS as an
option for optimization.

By definition, it is what the assembler does. When ".set noreorder" is not in effect, the source code *must* be written as if branch delay slots do not exist. There is no option here.


Acked-by: Daniel Schwierzeck <daniel.schwierzeck@xxxxxxxxx>


Acked-by: David Daney <david.daney@xxxxxxxxxx>