linux-mips
[Top] [All Lists]

Re: [PATCH 4/9] mips: deal with larger physical offsets

To: Florian Fainelli <florian@openwrt.org>, Ralf Baechle <ralf@linux-mips.org>
Subject: Re: [PATCH 4/9] mips: deal with larger physical offsets
From: David Daney <ddaney@caviumnetworks.com>
Date: Mon, 01 Jun 2009 11:13:29 -0700
Cc: linux-mips@linux-mips.org
In-reply-to: <200906011359.47216.florian@openwrt.org>
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <200906011359.47216.florian@openwrt.org>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Thunderbird 2.0.0.21 (X11/20090320)
Florian Fainelli wrote:
TI AR7 has larger physical offsets than other MIPS-based
systems. A long jump to the target is required instead of
a short jump like on other systems.

Signed-off-by: Eugene Konev <ejka@imfi.kspu.ru>
Signed-off-by: Florian Fainelli <florian@openwrt.org>
---
diff --git a/arch/mips/kernel/traps.c b/arch/mips/kernel/traps.c
index e83da17..44f7141 100644
--- a/arch/mips/kernel/traps.c
+++ b/arch/mips/kernel/traps.c
@@ -1256,9 +1256,22 @@ void *set_except_vector(int n, void *addr)
exception_handlers[n] = handler;
        if (n == 0 && cpu_has_divec) {
-               *(u32 *)(ebase + 0x200) = 0x08000000 |
-                                         (0x03ffffff & (handler >> 2));
-               local_flush_icache_range(ebase + 0x200, ebase + 0x204);
+               if ((handler ^ (ebase + 4)) & 0xfc000000) {
+                       /* lui k0, 0x0000 */
+                       *(u32 *)(ebase + 0x200) = 0x3c1a0000 | (handler >> 16);
+                       /* ori k0, 0x0000 */
+                       *(u32 *)(ebase + 0x204) =
+                                       0x375a0000 | (handler & 0xffff);
+                       /* jr k0 */
+                       *(u32 *)(ebase + 0x208) = 0x03400008;
+                       /* nop */
+                       *(u32 *)(ebase + 0x20C) = 0x00000000;
+                       local_flush_icache_range(ebase + 0x200, ebase + 0x210);
+               } else {
+                       *(u32 *)(ebase + 0x200) =
+                               0x08000000 | (0x03ffffff & (handler >> 2));
+                       local_flush_icache_range(ebase + 0x200, ebase + 0x204);
+               }
        }
        return (void *)old_handler;
 }


I had to do a similar change for my mapped kernel patch. However in mine, I used uasm. I wonder if the extra memory occupied by uasm is too steep a penalty to pay for the ability to use it outside of __init code. If not, we should move uasm out of __init and use it instead of this manual assembly.

David Daney

<Prev in Thread] Current Thread [Next in Thread>