early ioremap not working with 2.4.19-pre1-aa1 ?

early ioremap not working with 2.4.19-pre1-aa1 ?

[Martin J. Bligh]

I have code for the NUMA-Q systems that does an ioremap
as the first thing in smp_boot_cpus (ia32 tree). This seems to 
work fine until I install the aa patches ... then it hangs in the 
ioremap.

Has anyone got any idea why this might be? I'd really like to
test out the -aa vm patches on this box ... I can debug it some
more - just looking for an easy answer ;-)

Thanks,

Martin.

Re: early ioremap not working with 2.4.19-pre1-aa1 ?

[Andrea Arcangeli]

On Fri, Mar 01, 2002 at 03:46:14PM -0800, Martin J. Bligh wrote:
> I have code for the NUMA-Q systems that does an ioremap
> as the first thing in smp_boot_cpus (ia32 tree). This seems to 
> work fine until I install the aa patches ... then it hangs in the 
> ioremap.

this sounds like the same problem of the MXT patch. In short pte_alloc
and in turn ioremap was usable only after the initcalls.

Does this incremental patch fix it?  (untested)

--- 2.4.19pre1aa1/include/linux/highmem.h.~1~	Fri Mar  1 20:19:05 2002
+++ 2.4.19pre1aa1/include/linux/highmem.h	Sat Mar  2 03:43:42 2002
@@ -12,6 +12,7 @@
 
 /* declarations for linux/mm/highmem.c */
 unsigned int nr_free_highpages(void);
+extern void init_kmap(void);
 
 extern struct buffer_head *create_bounce(int rw, struct buffer_head * bh_orig);
 
@@ -64,6 +65,7 @@
 #else /* CONFIG_HIGHMEM */
 
 static inline unsigned int nr_free_highpages(void) { return 0; }
+#define init_kmap() do { } while(0)
 
 static inline void *kmap(struct page *page) { return page_address(page); }
 
--- 2.4.19pre1aa1/init/main.c.~1~	Wed Feb 27 12:46:19 2002
+++ 2.4.19pre1aa1/init/main.c	Sat Mar  2 03:42:52 2002
@@ -599,6 +599,7 @@
 	mem_init();
 	kmem_cache_sizes_init();
 	pgtable_cache_init();
+	init_kmap();
 
 #ifdef CONFIG_PERFMON
 	perfmon_init();
--- 2.4.19pre1aa1/mm/highmem.c.~1~	Wed Feb 27 12:46:13 2002
+++ 2.4.19pre1aa1/mm/highmem.c	Sat Mar  2 03:43:50 2002
@@ -45,7 +45,7 @@
 
 static wait_queue_head_t pkmap_map_wait[KM_NR_SERIES];
 
-static __init int init_kmap(void)
+void __init init_kmap(void)
 {
 	int i;
 
@@ -56,7 +56,6 @@
 #endif
 	return 0;
 }
-__initcall(init_kmap);
 
 static void flush_all_zero_pkmaps(void)
 {
> 
> Has anyone got any idea why this might be? I'd really like to
> test out the -aa vm patches on this box ... I can debug it some
> more - just looking for an easy answer ;-)
> 
> Thanks,
> 
> Martin.


Andrea

Re: early ioremap not working with 2.4.19-pre1-aa1 ?

[Martin J. Bligh]

>> I have code for the NUMA-Q systems that does an ioremap
>> as the first thing in smp_boot_cpus (ia32 tree). This seems to 
>> work fine until I install the aa patches ... then it hangs in the 
>> ioremap.
> 
> this sounds like the same problem of the MXT patch. In short pte_alloc
> and in turn ioremap was usable only after the initcalls.
> 
> Does this incremental patch fix it?  (untested)

Sorry for the slow test cycle - this works just great ... will
this make it back to your main tree?

Thanks very much for the patch,

Martin.

Re: early ioremap not working with 2.4.19-pre1-aa1 ?

[Mikael Pettersson]

On Fri, 01 Mar 2002 15:46:14 -0800, Martin J. Bligh wrote:
>I have code for the NUMA-Q systems that does an ioremap
>as the first thing in smp_boot_cpus (ia32 tree). This seems to 
>work fine until I install the aa patches ... then it hangs in the 
>ioremap.

You may want to use the new boot-time ioremap, which is based
on the fixmap mechanism. It's included in 2.5.6-pre1; a separate
patch for 2.4.18 is available at
<http://www.csd.uu.se/~mikpe/linux/patches/2.4/>.

/Mikael