From a40a6e16ed76e5e26a0f60226b64c311d4a62c9f Mon Sep 17 00:00:00 2001
From: Mathias Kresin <dev@kresin.me>
Date: Sun, 31 Oct 2021 23:04:54 +0100
Subject: [PATCH] lzma: force 8bit reads

At least since gcc 7.3.0 (OpenWrt 18.06) lwr/lwl are used in the
assembly of LzmaProps_Decode. While the decission made by the compiler
looks perfect fine, it triggers some obscure hang on lantiq danube-s
v1.5 with MX29LV640EB NOR flash chips.

Only if the offset 1 is used, the hang can be observed. Using any other
offset works fine:

  lwl s0,0(a1) - s0 == 0x6d000080
  lwl s0,1(a1) - hangs
  lwl s0,2(a1) - s0 == 0x0080xxxx
  lwl s0,3(a1) - s0 == 0x80xxxxxx

It isn't clear whether it is a limitation of the flash chip, the EBU or
something else.

Force 8bit reads to prevent gcc optimizing the read with lwr/lwl
instructions.

Signed-off-by: Mathias Kresin <dev@kresin.me>
---
 lib/lzma/LzmaDec.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

--- a/lib/lzma/LzmaDec.c
+++ b/lib/lzma/LzmaDec.c
@@ -7,6 +7,7 @@
 #include "LzmaDec.h"
 
 #include <linux/string.h>
+#include <asm/io.h>
 
 #define kNumTopBits 24
 #define kTopValue ((UInt32)1 << kNumTopBits)
@@ -703,7 +704,7 @@ static ELzmaDummy LzmaDec_TryDummy(const
 
 static void LzmaDec_InitRc(CLzmaDec *p, const Byte *data)
 {
-  p->code = ((UInt32)data[1] << 24) | ((UInt32)data[2] << 16) | ((UInt32)data[3] << 8) | ((UInt32)data[4]);
+  p->code = ((UInt32)readb(&data[1]) << 24) | ((UInt32)readb(&data[2]) << 16) | ((UInt32)readb(&data[3]) << 8) | ((UInt32)readb(&data[4]));
   p->range = 0xFFFFFFFF;
   p->needFlush = 0;
 }
@@ -929,7 +930,7 @@ SRes LzmaProps_Decode(CLzmaProps *p, con
   if (size < LZMA_PROPS_SIZE)
     return SZ_ERROR_UNSUPPORTED;
   else
-    dicSize = data[1] | ((UInt32)data[2] << 8) | ((UInt32)data[3] << 16) | ((UInt32)data[4] << 24);
+    dicSize = readb(&data[1]) | ((UInt32)readb(&data[2]) << 8) | ((UInt32)readb(&data[3]) << 16) | ((UInt32)readb(&data[4]) << 24);
 
   if (dicSize < LZMA_DIC_MIN)
     dicSize = LZMA_DIC_MIN;