From: Jun Sun <jsun@mvista.com>
To: unlisted-recipients:; (no To-header on input)
Cc: linux-mips@fnet.fr, linux@cthulhu.engr.sgi.com
Subject: Re: Turning off cache ...
Date: Fri, 28 Jul 2000 19:43:33 -0700 [thread overview]
Message-ID: <398244D5.F86DE02C@mvista.com> (raw)
In-Reply-To: 20000728030836.A1906@bacchus.dhis.org
[-- Attachment #1: Type: text/plain, Size: 1274 bytes --]
Ralf Baechle wrote:
>
> On Thu, Jul 27, 2000 at 05:52:42PM -0700, Jun Sun wrote:
>
> > Is there is easy way to turn off caching entirely? I understand I need
> > to set k0 bits in config register. What about those C bits in TLB
> > entries? My CPU only has primary cache.
>
> The C bits are per page, the k0 bits are for KSEG0. If you want to
> turn of caching, then you need to:
>
> - change the k0 bits to uncached on startup, then flush the caches or a
> writeback might corrupt your data.
> - change the caching mode of the usermode pages by modifying the
> definitions for PAGE_NONE etc. in pgtable.h.
> - comment out the cache Create_Dirty_Exclusive instructions in r4xx0.c,
> using them on uncached pages would corrupt data.
>
> Which will make the kernel crawl awfully ...
>
> Ralf
I have successfully turned of cache. And of course my cache code is not
the source for the problem :-) - which means I need to probe further.
:-(
For those who are interested, here is the patch that gives how I did
it. Specifically, ld_mmu_r4k() has a line of code to turn on cache,
which needs to be commented out. I explicitly turned off cache in
ld_mmu_r4k(), right after a flush_all_cache() call, instead of at the
beginnig of kernel_entry.
Jun
[-- Attachment #2: linux-mips-v2.4.0-test2-r4k-cache-off.patch --]
[-- Type: text/plain, Size: 4068 bytes --]
--- include/asm/pgtable.h.orig Fri Jul 28 15:22:09 2000
+++ include/asm/pgtable.h Fri Jul 28 15:22:54 2000
@@ -132,7 +132,7 @@
#define _CACHE_CACHABLE_NO_WA (0<<9) /* R4600 only */
#define _CACHE_CACHABLE_WA (1<<9) /* R4600 only */
#define _CACHE_UNCACHED (2<<9) /* R4[0246]00 */
-#define _CACHE_CACHABLE_NONCOHERENT (3<<9) /* R4[0246]00 */
+#define _CACHE_CACHABLE_NONCOHERENT (2<<9) /* R4[0246]00 */
#define _CACHE_CACHABLE_CE (4<<9) /* R4[04]00 only */
#define _CACHE_CACHABLE_COW (5<<9) /* R4[04]00 only */
#define _CACHE_CACHABLE_CUW (6<<9) /* R4[04]00 only */
--- arch/mips/mm/r4xx0.c.cache-off Fri Jul 28 14:49:23 2000
+++ arch/mips/mm/r4xx0.c Fri Jul 28 19:38:17 2000
@@ -105,7 +105,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
}
@@ -134,7 +134,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
}
@@ -143,8 +143,8 @@
* This flavour of r4k_clear_page is for the R4600 V1.x. Cite from the
* IDT R4600 V1.7 errata:
*
- * 18. The CACHE instructions Hit_Writeback_Invalidate_D, Hit_Writeback_D,
- * Hit_Invalidate_D and Create_Dirty_Excl_D should only be
+ * 18. The CACHE instructions Hit_Writeback_Inv_D, Hit_Writeback_D,
+ * Hit_Invalidate_D and Hit_Writeback_Inv_D should only be
* executed if there is no other dcache activity. If the dcache is
* accessed for another instruction immeidately preceding when these
* cache instructions are executing, it is possible that the dcache
@@ -157,14 +157,14 @@
* nop
* nop
* nop
- * cache Hit_Writeback_Invalidate_D
+ * cache Hit_Writeback_Inv_D
*
* This is allowed: lw
* nop
* nop
* nop
* nop
- * cache Hit_Writeback_Invalidate_D
+ * cache Hit_Writeback_Inv_D
*/
static void r4k_clear_page_r4600_v1(void * page)
{
@@ -198,7 +198,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
}
@@ -234,7 +234,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
restore_flags(flags);
}
@@ -434,7 +434,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
}
static void r4k_copy_page_d32(void * to, void * from)
@@ -491,7 +491,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
}
/*
@@ -559,7 +559,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
}
static void r4k_copy_page_r4600_v2(void * to, void * from)
@@ -626,7 +626,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
restore_flags(flags);
}
@@ -2783,7 +2783,9 @@
printk("CPU revision is: %08x\n", read_32bit_cp0_register(CP0_PRID));
+ /* [jsun]
set_cp0_config(CONF_CM_CMASK, CONF_CM_CACHABLE_NONCOHERENT);
+ */
probe_icache(config);
probe_dcache(config);
@@ -2803,6 +2805,7 @@
}
flush_cache_all();
+ set_cp0_config(CONF_CM_CMASK, CONF_CM_UNCACHED);
write_32bit_cp0_register(CP0_WIRED, 0);
/*
WARNING: multiple messages have this Message-ID (diff)
From: Jun Sun <jsun@mvista.com>
Cc: linux-mips@fnet.fr, linux@cthulhu.engr.sgi.com
Subject: Re: Turning off cache ...
Date: Fri, 28 Jul 2000 19:43:33 -0700 [thread overview]
Message-ID: <398244D5.F86DE02C@mvista.com> (raw)
Message-ID: <20000729024333.AQKSKSLS3fQ4CydOrQt7hMYrhXSNoE2t79fYkc27mRc@z> (raw)
In-Reply-To: 20000728030836.A1906@bacchus.dhis.org
[-- Attachment #1: Type: text/plain, Size: 1274 bytes --]
Ralf Baechle wrote:
>
> On Thu, Jul 27, 2000 at 05:52:42PM -0700, Jun Sun wrote:
>
> > Is there is easy way to turn off caching entirely? I understand I need
> > to set k0 bits in config register. What about those C bits in TLB
> > entries? My CPU only has primary cache.
>
> The C bits are per page, the k0 bits are for KSEG0. If you want to
> turn of caching, then you need to:
>
> - change the k0 bits to uncached on startup, then flush the caches or a
> writeback might corrupt your data.
> - change the caching mode of the usermode pages by modifying the
> definitions for PAGE_NONE etc. in pgtable.h.
> - comment out the cache Create_Dirty_Exclusive instructions in r4xx0.c,
> using them on uncached pages would corrupt data.
>
> Which will make the kernel crawl awfully ...
>
> Ralf
I have successfully turned of cache. And of course my cache code is not
the source for the problem :-) - which means I need to probe further.
:-(
For those who are interested, here is the patch that gives how I did
it. Specifically, ld_mmu_r4k() has a line of code to turn on cache,
which needs to be commented out. I explicitly turned off cache in
ld_mmu_r4k(), right after a flush_all_cache() call, instead of at the
beginnig of kernel_entry.
Jun
[-- Attachment #2: linux-mips-v2.4.0-test2-r4k-cache-off.patch --]
[-- Type: text/plain, Size: 4068 bytes --]
--- include/asm/pgtable.h.orig Fri Jul 28 15:22:09 2000
+++ include/asm/pgtable.h Fri Jul 28 15:22:54 2000
@@ -132,7 +132,7 @@
#define _CACHE_CACHABLE_NO_WA (0<<9) /* R4600 only */
#define _CACHE_CACHABLE_WA (1<<9) /* R4600 only */
#define _CACHE_UNCACHED (2<<9) /* R4[0246]00 */
-#define _CACHE_CACHABLE_NONCOHERENT (3<<9) /* R4[0246]00 */
+#define _CACHE_CACHABLE_NONCOHERENT (2<<9) /* R4[0246]00 */
#define _CACHE_CACHABLE_CE (4<<9) /* R4[04]00 only */
#define _CACHE_CACHABLE_COW (5<<9) /* R4[04]00 only */
#define _CACHE_CACHABLE_CUW (6<<9) /* R4[04]00 only */
--- arch/mips/mm/r4xx0.c.cache-off Fri Jul 28 14:49:23 2000
+++ arch/mips/mm/r4xx0.c Fri Jul 28 19:38:17 2000
@@ -105,7 +105,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
}
@@ -134,7 +134,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
}
@@ -143,8 +143,8 @@
* This flavour of r4k_clear_page is for the R4600 V1.x. Cite from the
* IDT R4600 V1.7 errata:
*
- * 18. The CACHE instructions Hit_Writeback_Invalidate_D, Hit_Writeback_D,
- * Hit_Invalidate_D and Create_Dirty_Excl_D should only be
+ * 18. The CACHE instructions Hit_Writeback_Inv_D, Hit_Writeback_D,
+ * Hit_Invalidate_D and Hit_Writeback_Inv_D should only be
* executed if there is no other dcache activity. If the dcache is
* accessed for another instruction immeidately preceding when these
* cache instructions are executing, it is possible that the dcache
@@ -157,14 +157,14 @@
* nop
* nop
* nop
- * cache Hit_Writeback_Invalidate_D
+ * cache Hit_Writeback_Inv_D
*
* This is allowed: lw
* nop
* nop
* nop
* nop
- * cache Hit_Writeback_Invalidate_D
+ * cache Hit_Writeback_Inv_D
*/
static void r4k_clear_page_r4600_v1(void * page)
{
@@ -198,7 +198,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
}
@@ -234,7 +234,7 @@
:"=r" (page)
:"0" (page),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D)
+ "i" (Hit_Writeback_Inv_D)
:"$1","memory");
restore_flags(flags);
}
@@ -434,7 +434,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
}
static void r4k_copy_page_d32(void * to, void * from)
@@ -491,7 +491,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
}
/*
@@ -559,7 +559,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
}
static void r4k_copy_page_r4600_v2(void * to, void * from)
@@ -626,7 +626,7 @@
"=&r" (reg1), "=&r" (reg2), "=&r" (reg3), "=&r" (reg4)
:"0" (to), "1" (from),
"I" (PAGE_SIZE),
- "i" (Create_Dirty_Excl_D));
+ "i" (Hit_Writeback_Inv_D));
restore_flags(flags);
}
@@ -2783,7 +2783,9 @@
printk("CPU revision is: %08x\n", read_32bit_cp0_register(CP0_PRID));
+ /* [jsun]
set_cp0_config(CONF_CM_CMASK, CONF_CM_CACHABLE_NONCOHERENT);
+ */
probe_icache(config);
probe_dcache(config);
@@ -2803,6 +2805,7 @@
}
flush_cache_all();
+ set_cp0_config(CONF_CM_CMASK, CONF_CM_UNCACHED);
write_32bit_cp0_register(CP0_WIRED, 0);
/*
next prev parent reply other threads:[~2000-07-29 2:46 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2000-07-28 0:52 Turning off cache Jun Sun
2000-07-28 1:08 ` Ralf Baechle
2000-07-29 2:43 ` Jun Sun [this message]
2000-07-29 2:43 ` Jun Sun
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=398244D5.F86DE02C@mvista.com \
--to=jsun@mvista.com \
--cc=linux-mips@fnet.fr \
--cc=linux@cthulhu.engr.sgi.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox