* Re: [Discontig-devel] [PATCH] another discontig patch
@ 2003-06-21 9:06 Christoph Hellwig
2003-06-21 14:48 ` Martin J. Bligh
` (13 more replies)
0 siblings, 14 replies; 15+ messages in thread
From: Christoph Hellwig @ 2003-06-21 9:06 UTC (permalink / raw)
To: linux-ia64
On Thu, Jun 19, 2003 at 03:28:29PM -0700, Jesse Barnes wrote:
> @@ -703,6 +702,8 @@
> * get_free_pages() cannot be used before cpu_init() done. BSP allocates
> * "NR_CPUS" pages for all CPUs to avoid that AP calls get_zeroed_page().
> */
> +#ifndef CONFIG_DISCONTIGMEM
> + /* for discontig machines, we do this in discontig.c */
> if (smp_processor_id() = 0) {
> cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * NR_CPUS, PERCPU_PAGE_SIZE,
> __pa(MAX_DMA_ADDRESS));
> @@ -712,15 +713,13 @@
> cpu_data += PERCPU_PAGE_SIZE;
> }
> }
> +#endif
> cpu_data = __per_cpu_start + __per_cpu_offset[smp_processor_id()];
> #else /* !CONFIG_SMP */
> cpu_data = __phys_per_cpu_start;
> #endif /* !CONFIG_SMP */
Maybe this whole code should better be abstracted out then? Also
the use of smp_processor_id() here is not preempt-safe but I 'm not sure
whether preemption is already enabled at that point.
> +static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
> +static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
> +static bootmem_data_t bdata[NR_NODES] __initdata;
> +static unsigned long boot_pernode[NR_NODES] __initdata;
> +static unsigned long boot_pernodesize[NR_NODES] __initdata;
Maybe it's time for special pernode data like the percpu data?
Okay, okay, not revelant for this patch yet...
> struct page *zero_page_memmap_ptr; /* map entry for zero page */
>
> +extern int filter_rsvd_memory(unsigned long start, unsigned long end, void *arg);
Move this to a header?
> +#if defined(CONFIG_VIRTUAL_MEM_MAP) && !defined(CONFIG_DISCONTIGMEM)
> + extern int ia64_pfn_valid (unsigned long pfn);
> +# define pfn_valid(pfn) (((pfn) < max_mapnr) && ia64_pfn_valid(pfn))
> +#elif defined(CONFIG_VIRTUAL_MEM_MAP) && defined(CONFIG_DISCONTIGMEM)
> + extern int ia64_pfn_valid (unsigned long pfn);
> + extern unsigned long max_low_pfn;
> +# define pfn_valid(pfn) (((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
> +#else
> +# define pfn_valid(pfn) ((pfn) < max_mapnr)
> +#endif
Hmm, this ifdef mess look ugly. Can't we use max_low_pfn for the
!CONFIG_DISCONTIGMEM case, too and simp;ify it to something like:
#if defined(CONFIG_VIRTUAL_MEM_MAP)
extern int ia64_pfn_valid(unsigned long pfn);
#else
# define ia64_pfn_valid(pfn) (1)
#endif
#define pfn_valid(pfn) (((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
> +
> +#if defined(CONFIG_DISCONTIGMEM) && defined(CONFIG_VIRTUAL_MEM_MAP)
> + extern struct page *vmem_map;
> +# define pfn_to_page(pfn) (vmem_map + (pfn))
> +# define page_to_pfn(page) ((unsigned long) (page - vmem_map))
> +#else
> +# define pfn_to_page(pfn) (mem_map + (pfn))
> +# define page_to_pfn(page) ((unsigned long) (page - mem_map))
> +#endif /* CONFIG_DISCONTIGMEM && CONFIG_VIRTUAL_MEM_MAP */
> +
Hmm, doesn't CONFIG_VIRTUAL_MEM_MAP && !CONFIG_DISCONTIGMEM need
vmem_map also. Again, the ifdef mess is horrible :)
What about:
#ifndef CONFIG_VIRTUAL_MEM_MAP
#define vmem_map mem_map
#endif
#define pfn_to_page(pfn) (vmem_map + (pfn))
#define page_to_pfn(page) ((unsigned long) (page - vmem_map))
In fact I wonder what's so special about mem_map that the symbol
can't be used for the vmalloc'ed version..
BTW, what about per-node memmaps for SN2 like it's done for NUMAQ?
> #ifdef CONFIG_NUMA
> struct ia64_node_data *node_data;
> + int nodeid;
> + struct cpuinfo_ia64 *cpu_data[NR_CPUS];
> #endif
Should the big array be before the nodeid field?
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
@ 2003-06-21 14:48 ` Martin J. Bligh
2003-06-22 5:53 ` Jesse Barnes
` (12 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Martin J. Bligh @ 2003-06-21 14:48 UTC (permalink / raw)
To: linux-ia64
>> +static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
>> +static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
>> +static bootmem_data_t bdata[NR_NODES] __initdata;
>> +static unsigned long boot_pernode[NR_NODES] __initdata;
>> +static unsigned long boot_pernodesize[NR_NODES] __initdata;
The only use that I can *see* (without looking very hard) for
pg_data_ptr is as pg_data_ptr[node]->bdata, for which you already have
bdata[node], don't you? The fact that you have both pg_data_ptr and
pg_data_ptrs, which seem to do the same thing, but one as initdata,
and one not, is also rather confusing ...
memcpy(boot_node_data[0]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
hmmm.
And aren't boot_pernode and boot_pernodesize really part of bdata?
Seems like there's a lot of complexity in order to save a couple of
pointer dereferences ... during boot ;-) But maybe I'm just misreading
it ;-)
> Maybe it's time for special pernode data like the percpu data?
> Okay, okay, not revelant for this patch yet...
># ifndef CONFIG_VIRTUAL_MEM_MAP
># define vmem_map mem_map
># endif
>
># define pfn_to_page(pfn) (vmem_map + (pfn))
># define page_to_pfn(page) ((unsigned long) (page - vmem_map))
>
> In fact I wonder what's so special about mem_map that the symbol
> can't be used for the vmalloc'ed version..
>
> BTW, what about per-node memmaps for SN2 like it's done for NUMAQ?
I really don't see how you *can't* do that, and have it still work.
If the mem_map is placed in node local memory, and not a size that
happens to be a complete number of pages, even if you pack it down
into a vmem_map, it's still not contiguous. So adding a pfn (physical
page frame number) to the base of vmem_map can NOT give you the correct
address.
Presumably this *is* working for you ... but I'm buggered if I know how ;-)
Maybe I just need more coffee ;-)
Thanks,
M.
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
2003-06-21 14:48 ` Martin J. Bligh
@ 2003-06-22 5:53 ` Jesse Barnes
2003-06-22 5:57 ` Jesse Barnes
` (11 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Jesse Barnes @ 2003-06-22 5:53 UTC (permalink / raw)
To: linux-ia64
Thanks for looking at this, Christoph. Comments below.
On Sat, Jun 21, 2003 at 10:06:46AM +0100, Christoph Hellwig wrote:
> On Thu, Jun 19, 2003 at 03:28:29PM -0700, Jesse Barnes wrote:
> > #else /* !CONFIG_SMP */
> > cpu_data = __phys_per_cpu_start;
> > #endif /* !CONFIG_SMP */
>
> Maybe this whole code should better be abstracted out then? Also
> the use of smp_processor_id() here is not preempt-safe but I 'm not sure
> whether preemption is already enabled at that point.
I think we're ok preempt wise here, but you're right that it could
probably be abstracted a bit. The issue is that we need to create
per-node data structures containing each node's info before the per-cpu
stuff is initialized, so we do it all before the bootmem allocator is
even setup while we discover memory (see
discontig.c:find_pernode_space() I think).
> > +static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
> > +static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
> > +static bootmem_data_t bdata[NR_NODES] __initdata;
> > +static unsigned long boot_pernode[NR_NODES] __initdata;
> > +static unsigned long boot_pernodesize[NR_NODES] __initdata;
>
> Maybe it's time for special pernode data like the percpu data?
> Okay, okay, not revelant for this patch yet...
It's already there. It's really necessary to get decent performance,
and you have to watch out for cache aliasing effects.
> > +extern int filter_rsvd_memory(unsigned long start, unsigned long end, void *arg);
>
> Move this to a header?
Yeah, maybe it should be in asm/page.h? Or maybe we need an
asm/discontig.h?
> > +# define pfn_valid(pfn) ((pfn) < max_mapnr)
> > +#endif
>
> Hmm, this ifdef mess look ugly. Can't we use max_low_pfn for the
> !CONFIG_DISCONTIGMEM case, too and simp;ify it to something like:
Agreed, no question. I think this could use some additional cleanup.
> #if defined(CONFIG_VIRTUAL_MEM_MAP)
> extern int ia64_pfn_valid(unsigned long pfn);
> #else
> # define ia64_pfn_valid(pfn) (1)
> #endif
>
> #define pfn_valid(pfn) (((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
>
> > +
> > +#if defined(CONFIG_DISCONTIGMEM) && defined(CONFIG_VIRTUAL_MEM_MAP)
> > + extern struct page *vmem_map;
> > +# define pfn_to_page(pfn) (vmem_map + (pfn))
> > +# define page_to_pfn(page) ((unsigned long) (page - vmem_map))
> > +#else
> > +# define pfn_to_page(pfn) (mem_map + (pfn))
> > +# define page_to_pfn(page) ((unsigned long) (page - mem_map))
> > +#endif /* CONFIG_DISCONTIGMEM && CONFIG_VIRTUAL_MEM_MAP */
> > +
Though I'm not sure if this is the right way to do it... If discontig
on ia64 is going to depend on virtual mem map, then making this one
#ifdef somehow is probably best.
> Hmm, doesn't CONFIG_VIRTUAL_MEM_MAP && !CONFIG_DISCONTIGMEM need
> vmem_map also. Again, the ifdef mess is horrible :)
>
> What about:
>
> #ifndef CONFIG_VIRTUAL_MEM_MAP
> #define vmem_map mem_map
> #endif
>
> #define pfn_to_page(pfn) (vmem_map + (pfn))
> #define page_to_pfn(page) ((unsigned long) (page - vmem_map))
>
> In fact I wonder what's so special about mem_map that the symbol
> can't be used for the vmalloc'ed version..
Don't know, I'll have to look some more.
> BTW, what about per-node memmaps for SN2 like it's done for NUMAQ?
AFAIK, the per-node memmaps index into the big virtual memmap, but I'd
have to look at it some more. If we made per-node memory maps, then we
probably wouldn't need to use the virtual mem map, except that it makes
dealing with holes a bit easier (since on sn2 even memory within a node
isn't contiguous, and there are big holes between nodes).
> > #ifdef CONFIG_NUMA
> > struct ia64_node_data *node_data;
> > + int nodeid;
> > + struct cpuinfo_ia64 *cpu_data[NR_CPUS];
> > #endif
>
> Should the big array be before the nodeid field?
Good catch, probably not. :)
Thanks,
Jesse
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
2003-06-21 14:48 ` Martin J. Bligh
2003-06-22 5:53 ` Jesse Barnes
@ 2003-06-22 5:57 ` Jesse Barnes
2003-06-22 15:25 ` Martin J. Bligh
` (10 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Jesse Barnes @ 2003-06-22 5:57 UTC (permalink / raw)
To: linux-ia64
On Sat, Jun 21, 2003 at 07:48:08AM -0700, Martin J. Bligh wrote:
> >> +static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
> >> +static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
> >> +static bootmem_data_t bdata[NR_NODES] __initdata;
> >> +static unsigned long boot_pernode[NR_NODES] __initdata;
> >> +static unsigned long boot_pernodesize[NR_NODES] __initdata;
>
> The only use that I can *see* (without looking very hard) for
> pg_data_ptr is as pg_data_ptr[node]->bdata, for which you already have
> bdata[node], don't you? The fact that you have both pg_data_ptr and
> pg_data_ptrs, which seem to do the same thing, but one as initdata,
> and one not, is also rather confusing ...
Yeah, that's one thing that I keep meaning to remove, but didn't in the
initial versions to keep the 2.4 and 2.5 patches as close as possible.
> memcpy(boot_node_data[0]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
>
> hmmm.
>
> And aren't boot_pernode and boot_pernodesize really part of bdata?
> Seems like there's a lot of complexity in order to save a couple of
> pointer dereferences ... during boot ;-) But maybe I'm just misreading
> it ;-)
Well, there's definitely some pointer magic going on here, but I think
we use that stuff once the system is up as well, though from a different
pointer.
>
> > Maybe it's time for special pernode data like the percpu data?
> > Okay, okay, not revelant for this patch yet...
>
> ># ifndef CONFIG_VIRTUAL_MEM_MAP
> ># define vmem_map mem_map
> ># endif
> >
> ># define pfn_to_page(pfn) (vmem_map + (pfn))
> ># define page_to_pfn(page) ((unsigned long) (page - vmem_map))
> >
> > In fact I wonder what's so special about mem_map that the symbol
> > can't be used for the vmalloc'ed version..
> >
> > BTW, what about per-node memmaps for SN2 like it's done for NUMAQ?
>
> I really don't see how you *can't* do that, and have it still work.
> If the mem_map is placed in node local memory, and not a size that
> happens to be a complete number of pages, even if you pack it down
> into a vmem_map, it's still not contiguous. So adding a pfn (physical
> page frame number) to the base of vmem_map can NOT give you the correct
> address.
>
> Presumably this *is* working for you ... but I'm buggered if I know how ;-)
> Maybe I just need more coffee ;-)
No, there are some mixed metaphors here for sure (at least, as far as I
understand it!) :). You can find stuff either using the global vmemmap
or the per-node maps I think, but honestly, I'm still coming up to speed
on this code myself, Jack and Kimi are the real experts.
Thanks,
Jesse
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (2 preceding siblings ...)
2003-06-22 5:57 ` Jesse Barnes
@ 2003-06-22 15:25 ` Martin J. Bligh
2003-06-23 17:20 ` William Lee Irwin III
` (9 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Martin J. Bligh @ 2003-06-22 15:25 UTC (permalink / raw)
To: linux-ia64
>> >> +static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
>> >> +static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
>> >> +static bootmem_data_t bdata[NR_NODES] __initdata;
>> >> +static unsigned long boot_pernode[NR_NODES] __initdata;
>> >> +static unsigned long boot_pernodesize[NR_NODES] __initdata;
>>
>> The only use that I can *see* (without looking very hard) for
>> pg_data_ptr is as pg_data_ptr[node]->bdata, for which you already have
>> bdata[node], don't you? The fact that you have both pg_data_ptr and
>> pg_data_ptrs, which seem to do the same thing, but one as initdata,
>> and one not, is also rather confusing ...
>
> Yeah, that's one thing that I keep meaning to remove, but didn't in the
> initial versions to keep the 2.4 and 2.5 patches as close as possible.
OK, I don't think it's something you're introducing with your patch,
so it's probably best done separately anyway - you seem to be making
the current situation better in general ;-)
>> > BTW, what about per-node memmaps for SN2 like it's done for NUMAQ?
>>
>> I really don't see how you *can't* do that, and have it still work.
>> If the mem_map is placed in node local memory, and not a size that
>> happens to be a complete number of pages, even if you pack it down
>> into a vmem_map, it's still not contiguous. So adding a pfn (physical
>> page frame number) to the base of vmem_map can NOT give you the correct
>> address.
>>
>> Presumably this *is* working for you ... but I'm buggered if I know how ;-)
>> Maybe I just need more coffee ;-)
>
> No, there are some mixed metaphors here for sure (at least, as far as I
> understand it!) :). You can find stuff either using the global vmemmap
> or the per-node maps I think, but honestly, I'm still coming up to speed
> on this code myself, Jack and Kimi are the real experts.
OK. Am I mistaken in thinking that the point of vmem_map is to skip over
physical holes in RAM? If not, then disregard the following ;-) Perhaps
it's just about node locality of the struct pages, but that seems odd?
Maybe if the start end of memory areas are always hitting boundaries aligned
on PAGE_SIZE number of pages, (ie 16MB for 4K pages), then your mem_map is
always page aligned because it's in (sizeof (struct page) * PAGE_SIZE) chunks.
If you (plural) are relying on that, then I think it needs a huge stinking
comment - isn't terribly obvious to innocent bystanders ;-)
Whilst that would enable you to make the array node_local, you still can't
just add a pfn to the base as an offset anyway, which is what it looks like
you were doing - that assumes it's all contig. eg this:
pfn_to_page(pfn) (vmem_map + (pfn))
Unless you somehow warped the definition of pfn into a contiguous version
of the space, in which case, I think people's heads may explode ... we've
always had pfn = phys_addr / PAGE_SIZE as an identity.
M.
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (3 preceding siblings ...)
2003-06-22 15:25 ` Martin J. Bligh
@ 2003-06-23 17:20 ` William Lee Irwin III
2003-07-16 19:29 ` Jesse Barnes
` (8 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: William Lee Irwin III @ 2003-06-23 17:20 UTC (permalink / raw)
To: linux-ia64
At some point in the past, someone wrote:
>># define pfn_to_page(pfn) (vmem_map + (pfn))
>># define page_to_pfn(page) ((unsigned long) (page - vmem_map))
On Sat, Jun 21, 2003 at 07:48:08AM -0700, Martin J. Bligh wrote:
> I really don't see how you *can't* do that, and have it still work.
> If the mem_map is placed in node local memory, and not a size that
> happens to be a complete number of pages, even if you pack it down
> into a vmem_map, it's still not contiguous. So adding a pfn (physical
> page frame number) to the base of vmem_map can NOT give you the correct
> address.
> Presumably this *is* working for you ... but I'm buggered if I know how ;-)
> Maybe I just need more coffee ;-)
It's a sparse array. The interstices between nodes are padded with
empty virtualspace in order to cheapen the indexing operation. i.e. the
nodes' local mem_map's are virtually positioned according to the range
of pfns they span. If 32-bit weren't as pressed for virtualspace it
could do likewise.
-- wli
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (4 preceding siblings ...)
2003-06-23 17:20 ` William Lee Irwin III
@ 2003-07-16 19:29 ` Jesse Barnes
2003-07-16 19:40 ` Matthew Wilcox
` (7 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Jesse Barnes @ 2003-07-16 19:29 UTC (permalink / raw)
To: linux-ia64
Sorry it took so long for me to reply to this, but I've been on vacation
for the past 3 weeks. Here's a new discontig patch that incorporates
some of the feeback from this thread. Comments and/or integration into
2.6 appreciated.
Thanks,
Jesse
# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
# ChangeSet 1.1396 -> 1.1397
# include/asm-ia64/page.h 1.19 -> 1.20
# arch/ia64/kernel/setup.c 1.53 -> 1.54
# include/asm-ia64/pgtable.h 1.28 -> 1.29
# mm/bootmem.c 1.18 -> 1.19
# include/asm-ia64/numa.h 1.5 -> 1.6
# include/asm-ia64/processor.h 1.48 -> 1.49
# arch/ia64/mm/init.c 1.46 -> 1.47
# include/asm-ia64/nodedata.h 1.3 -> 1.4
# arch/ia64/mm/discontig.c 1.4 -> 1.5
# arch/ia64/Kconfig 1.38 -> 1.39
# drivers/acpi/Kconfig 1.12 -> 1.13
# include/asm-ia64/mmzone.h 1.4 -> 1.5
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 03/07/16 jbarnes@tomahawk.engr.sgi.com 1.1397
# discontig update
# --------------------------------------------
#
diff -Nru a/arch/ia64/Kconfig b/arch/ia64/Kconfig
--- a/arch/ia64/Kconfig Wed Jul 16 12:28:28 2003
+++ b/arch/ia64/Kconfig Wed Jul 16 12:28:28 2003
@@ -210,8 +210,8 @@
system with an A0 or A1 stepping CPU.
config NUMA
- bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
+ bool
+ default y if IA64_SGI_SN2 || IA64_GENERIC
help
Say Y to compile the kernel to support NUMA (Non-Uniform Memory
Access). This option is for configuring high-end multiprocessor
@@ -235,8 +235,7 @@
config DISCONTIGMEM
bool
- depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) && NUMA
- default y
+ default y if IA64_SGI_SN2 || IA64_GENERIC
help
Say Y to support efficient handling of discontiguous physical memory,
for architectures which are either NUMA (Non-Uniform Memory Access)
@@ -245,8 +244,7 @@
config VIRTUAL_MEM_MAP
bool "Enable Virtual Mem Map"
- depends on !NUMA
- default y if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
+ default y if !IA64_HP_SIM
help
Say Y to compile the kernel with support for a virtual mem map.
This is an alternate method of supporting large holes in the
@@ -259,8 +257,8 @@
are unsure, say Y.
config IA64_MCA
- bool "Enable IA-64 Machine Check Abort" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
+ bool "Enable IA-64 Machine Check Abort"
+ default y if !IA64_HP_SIM
help
Say Y here to enable machine check support for IA-64. If you're
unsure, answer Y.
@@ -292,43 +290,12 @@
depends on IA64_GENERIC || IA64_DIG || IA64_HP_ZX1 || IA64_SGI_SN2
default y
-config IA64_SGI_SN_DEBUG
- bool "Enable extra debugging code"
- depends on IA64_SGI_SN2
- help
- Turns on extra debugging code in the SGI SN (Scalable NUMA) platform
- for IA-64. Unless you are debugging problems on an SGI SN IA-64 box,
- say N.
-
config IA64_SGI_SN_SIM
bool "Enable SGI Medusa Simulator Support"
depends on IA64_SGI_SN2
help
If you are compiling a kernel that will run under SGI's IA-64
simulator (Medusa) then say Y, otherwise say N.
-
-config IA64_SGI_AUTOTEST
- bool "Enable autotest (llsc). Option to run cache test instead of booting"
- depends on IA64_SGI_SN2
- help
- Build a kernel used for hardware validation. If you include the
- keyword "autotest" on the boot command line, the kernel does NOT boot.
- Instead, it starts all cpus and runs cache coherency tests instead.
-
- If unsure, say N.
-
-config SERIAL_SGI_L1_PROTOCOL
- bool "Enable protocol mode for the L1 console"
- depends on IA64_SGI_SN2
- help
- Uses protocol mode instead of raw mode for the level 1 console on the
- SGI SN (Scalable NUMA) platform for IA-64. If you are compiling for
- an SGI SN box then Y is the recommended value, otherwise say N.
-
-config PERCPU_IRQ
- bool
- depends on IA64_SGI_SN2
- default y
# On IA-64, we always want an ELF /proc/kcore.
config KCORE_ELF
diff -Nru a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c
--- a/arch/ia64/kernel/setup.c Wed Jul 16 12:28:28 2003
+++ b/arch/ia64/kernel/setup.c Wed Jul 16 12:28:28 2003
@@ -138,7 +138,7 @@
call_pernode_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long rs, re;
- void (*func)(unsigned long, unsigned long, int, int);
+ void (*func)(unsigned long, unsigned long, int);
int i;
start = PAGE_ALIGN(start);
@@ -149,22 +149,21 @@
func = arg;
if (!num_memblks) {
- /*
- * This machine doesn't have SRAT, so call func with
- * nid=0, bank=0.
- */
+ /* No SRAT table, to assume one node (node 0) */
if (start < end)
- (*func)(start, end - start, 0, 0);
+ (*func)(start, end, 0);
return;
}
for (i = 0; i < num_memblks; i++) {
- rs = max(start, node_memblk[i].start_paddr);
- re = min(end, node_memblk[i].start_paddr+node_memblk[i].size);
+ rs = max(__pa(start), node_memblk[i].start_paddr);
+ re = min(__pa(end), node_memblk[i].start_paddr+node_memblk[i].size);
if (rs < re)
- (*func)(rs, re-rs, node_memblk[i].nid,
- node_memblk[i].bank);
+ (*func)((unsigned long)__va(rs), (unsigned long)__va(re), node_memblk[i].nid);
+
+ if ((unsigned long)__va(re) = end)
+ break;
}
}
@@ -180,7 +179,7 @@
filter_rsvd_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long range_start, range_end, prev_start;
- void (*func)(unsigned long, unsigned long);
+ void (*func)(unsigned long, unsigned long, int);
int i;
#if IGNORE_PFN0
@@ -202,9 +201,9 @@
if (range_start < range_end)
#ifdef CONFIG_DISCONTIGMEM
- call_pernode_memory(__pa(range_start), __pa(range_end), func);
+ call_pernode_memory(range_start, range_end, func);
#else
- (*func)(__pa(range_start), range_end - range_start);
+ (*func)(range_start, range_end, 0);
#endif
/* nothing more available in this segment */
@@ -703,6 +702,8 @@
* get_free_pages() cannot be used before cpu_init() done. BSP allocates
* "NR_CPUS" pages for all CPUs to avoid that AP calls get_zeroed_page().
*/
+#ifndef CONFIG_DISCONTIGMEM
+ /* for discontig machines, we do this in discontig.c */
if (smp_processor_id() = 0) {
cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * NR_CPUS, PERCPU_PAGE_SIZE,
__pa(MAX_DMA_ADDRESS));
@@ -714,6 +715,7 @@
per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
}
}
+#endif
cpu_data = __per_cpu_start + __per_cpu_offset[smp_processor_id()];
#else /* !CONFIG_SMP */
cpu_data = __phys_per_cpu_start;
diff -Nru a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c
--- a/arch/ia64/mm/discontig.c Wed Jul 16 12:28:28 2003
+++ b/arch/ia64/mm/discontig.c Wed Jul 16 12:28:28 2003
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000, 2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2001 Intel Corp.
* Copyright (c) 2001 Tony Luck <tony.luck@intel.com>
* Copyright (c) 2002 NEC Corp.
@@ -16,74 +16,60 @@
#include <linux/mmzone.h>
#include <linux/acpi.h>
#include <linux/efi.h>
-
+#include <asm/pgalloc.h>
+#include <asm/tlb.h>
/*
- * Round an address upward to the next multiple of GRANULE size.
+ * Round an address upward or downward to the next multiple of IA64_GRANULE_SIZE.
*/
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
-static struct ia64_node_data *node_data[NR_NODES];
-static long boot_pg_data[8*NR_NODES+sizeof(pg_data_t)] __initdata;
-static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
-static bootmem_data_t bdata[NR_NODES][NR_BANKS_PER_NODE+1] __initdata;
-
-extern int filter_rsvd_memory (unsigned long start, unsigned long end, void *arg);
+/*
+ * Used to locate BOOT_DATA prior to initializing the node data area.
+ */
+#define BOOT_NODE_DATA(node) pg_data_ptr[node]
/*
- * Return the compact node number of this cpu. Used prior to
- * setting up the cpu_data area.
- * Note - not fast, intended for boot use only!!
+ * To prevent cache aliasing effects, align per-node structures so that they
+ * start at addresses that are strided by node number.
*/
-int
-boot_get_local_nodeid(void)
-{
- int i;
+#define NODEDATA_ALIGN(addr, node) ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + (node)*PERCPU_PAGE_SIZE)
- for (i = 0; i < NR_CPUS; i++)
- if (node_cpuid[i].phys_id = hard_smp_processor_id())
- return node_cpuid[i].nid;
- /* node info missing, so nid should be 0.. */
- return 0;
-}
+static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
+static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
+static bootmem_data_t bdata[NR_NODES] __initdata;
+static unsigned long boot_pernode[NR_NODES] __initdata;
+static unsigned long boot_pernodesize[NR_NODES] __initdata;
-/*
- * Return a pointer to the pg_data structure for a node.
- * This function is used ONLY in early boot before the cpu_data
- * structure is available.
- */
-pg_data_t* __init
-boot_get_pg_data_ptr(long node)
-{
- return pg_data_ptr[node];
-}
+extern char __per_cpu_start[], __per_cpu_end[];
-/*
- * Return a pointer to the node data for the current node.
- * (boottime initialization only)
- */
-struct ia64_node_data *
+struct ia64_node_data*
get_node_data_ptr(void)
{
- return node_data[boot_get_local_nodeid()];
+ return boot_node_data[cpu_to_node_map[smp_processor_id()]]; /* ZZZ */
}
/*
* We allocate one of the bootmem_data_t structs for each piece of memory
* that we wish to treat as a contiguous block. Each such block must start
- * on a BANKSIZE boundary. Multiple banks per node is not supported.
+ * on a GRANULE boundary. Multiple banks per node is not supported.
+ * (Note: on SN2, all memory on a node is trated as a single bank.
+ * Holes within the bank are supported. This works because memory
+ * from different banks is not interleaved. The bootmap bitmap
+ * for the node is somewhat large but not too large).
*/
static int __init
-build_maps(unsigned long pstart, unsigned long length, int node)
+build_maps(unsigned long start, unsigned long end, int node)
{
bootmem_data_t *bdp;
unsigned long cstart, epfn;
- bdp = pg_data_ptr[node]->bdata;
- epfn = GRANULEROUNDUP(pstart + length) >> PAGE_SHIFT;
- cstart = pstart & ~(BANKSIZE - 1);
+ bdp = &bdata[node];
+ epfn = GRANULEROUNDUP(__pa(end)) >> PAGE_SHIFT;
+ cstart = GRANULEROUNDDOWN(__pa(start));
if (!bdp->node_low_pfn) {
bdp->node_boot_start = cstart;
@@ -99,34 +85,96 @@
return 0;
}
+
+/*
+ * Count the number of cpus on the node
+ */
+static __inline__ int
+count_cpus(int node)
+{
+ int cpu, n=0;
+
+ for (cpu=0; cpu < NR_CPUS; cpu++)
+ if (node = node_cpuid[cpu].nid)
+ n++;
+ return n;
+}
+
+
/*
- * Find space on each node for the bootmem map.
+ * Find space on each node for the bootmem map & other per-node data structures.
*
* Called by efi_memmap_walk to find boot memory on each node. Note that
* only blocks that are free are passed to this routine (currently filtered by
* free_available_memory).
*/
static int __init
-find_bootmap_space(unsigned long pstart, unsigned long length, int node)
+find_pernode_space(unsigned long start, unsigned long end, int node)
{
- unsigned long mapsize, pages, epfn;
+ unsigned long mapsize, pages, epfn, map=0, cpu, cpus;
+ unsigned long pernodesize=0, pernode;
+ unsigned long cpu_data;
+ unsigned long pstart, length;
bootmem_data_t *bdp;
+ pstart = __pa(start);
+ length = end - start;
epfn = (pstart + length) >> PAGE_SHIFT;
- bdp = &pg_data_ptr[node]->bdata[0];
+ bdp = &bdata[node];
if (pstart < bdp->node_boot_start || epfn > bdp->node_low_pfn)
return 0;
- if (!bdp->node_bootmem_map) {
+ if (!boot_pernode[node]) {
+ cpus = count_cpus(node);
+ pernodesize += PERCPU_PAGE_SIZE * cpus;
+ pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
+ pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+ pernodesize = PAGE_ALIGN(pernodesize);
+ pernode = NODEDATA_ALIGN(pstart, node);
+
+ if (pstart + length > (pernode + pernodesize)) {
+ boot_pernode[node] = pernode;
+ boot_pernodesize[node] = pernodesize;
+ memset(__va(pernode), 0, pernodesize);
+
+ cpu_data = pernode;
+ pernode += PERCPU_PAGE_SIZE * cpus;
+
+ pg_data_ptr[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ boot_node_data[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+
+ pg_data_ptr[node]->bdata = &bdata[node];
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ if (node = node_cpuid[cpu].nid) {
+ extern char __per_cpu_start[], __phys_per_cpu_start[];
+ memcpy((void*)cpu_data, __phys_per_cpu_start, __per_cpu_end - __per_cpu_start);
+ __per_cpu_offset[cpu] = (char*)__va(cpu_data) - __per_cpu_start;
+ cpu_data += PERCPU_PAGE_SIZE;
+ }
+ }
+ }
+ }
+
+ pernode = boot_pernode[node];
+ pernodesize = boot_pernodesize[node];
+ if (pernode && !bdp->node_bootmem_map) {
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- if (length > mapsize) {
- init_bootmem_node(
- BOOT_NODE_DATA(node),
- pstart>>PAGE_SHIFT,
- bdp->node_boot_start>>PAGE_SHIFT,
- bdp->node_low_pfn);
+
+ if (pernode - pstart > mapsize)
+ map = pstart;
+ else if (pstart + length - pernode - pernodesize > mapsize)
+ map = pernode + pernodesize;
+
+ if (map) {
+ init_bootmem_node(BOOT_NODE_DATA(node), map>>PAGE_SHIFT,
+ bdp->node_boot_start>>PAGE_SHIFT, bdp->node_low_pfn);
}
}
@@ -143,9 +191,9 @@
*
*/
static int __init
-discontig_free_bootmem_node(unsigned long pstart, unsigned long length, int node)
+discontig_free_bootmem_node(unsigned long start, unsigned long end, int node)
{
- free_bootmem_node(BOOT_NODE_DATA(node), pstart, length);
+ free_bootmem_node(BOOT_NODE_DATA(node), __pa(start), end - start);
return 0;
}
@@ -158,53 +206,50 @@
discontig_reserve_bootmem(void)
{
int node;
- unsigned long mapbase, mapsize, pages;
+ unsigned long base, size, pages;
bootmem_data_t *bdp;
for (node = 0; node < numnodes; node++) {
bdp = BOOT_NODE_DATA(node)->bdata;
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
- mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- mapbase = __pa(bdp->node_bootmem_map);
- reserve_bootmem_node(BOOT_NODE_DATA(node), mapbase, mapsize);
+ size = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
+ base = __pa(bdp->node_bootmem_map);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
+
+ size = boot_pernodesize[node];
+ base = __pa(boot_pernode[node]);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
}
}
/*
- * Allocate per node tables.
- * - the pg_data structure is allocated on each node. This minimizes offnode
- * memory references
- * - the node data is allocated & initialized. Portions of this structure is read-only (after
- * boot) and contains node-local pointers to usefuls data structures located on
- * other nodes.
+ * Initialize per-node data
+ *
+ * Finish setting up the node data for this node, then copy it to the other nodes.
*
- * We also switch to using the "real" pg_data structures at this point. Earlier in boot, we
- * use a different structure. The only use for pg_data prior to the point in boot is to get
- * the pointer to the bdata for the node.
*/
static void __init
-allocate_pernode_structures(void)
+initialize_pernode_data(void)
{
- pg_data_t *pgdat=0, *new_pgdat_list=0;
- int node, mynode;
+ int cpu, node;
+
+ memcpy(boot_node_data[0]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
+ memcpy(boot_node_data[0]->node_data_ptrs, boot_node_data, sizeof(boot_node_data));
- mynode = boot_get_local_nodeid();
- for (node = numnodes - 1; node >= 0 ; node--) {
- node_data[node] = alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof (struct ia64_node_data));
- pgdat = __alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof(pg_data_t), SMP_CACHE_BYTES, 0);
- pgdat->bdata = &(bdata[node][0]);
- pg_data_ptr[node] = pgdat;
- pgdat->pgdat_next = new_pgdat_list;
- new_pgdat_list = pgdat;
+ for (node=1; node < numnodes; node++) {
+ memcpy(boot_node_data[node], boot_node_data[0], sizeof(struct ia64_node_data));
+ boot_node_data[node]->node = node;
}
-
- memcpy(node_data[mynode]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
- memcpy(node_data[mynode]->node_data_ptrs, node_data, sizeof(node_data));
- pgdat_list = new_pgdat_list;
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ node = node_cpuid[cpu].nid;
+ per_cpu(cpu_info, cpu).node_data = boot_node_data[node];
+ per_cpu(cpu_info, cpu).nodeid = node;
+ }
}
+
/*
* Called early in boot to setup the boot memory allocator, and to
* allocate the node-local pg_data & node-directory data structures..
@@ -212,96 +257,19 @@
void __init
discontig_mem_init(void)
{
- int node;
-
if (numnodes = 0) {
printk(KERN_ERR "node info missing!\n");
numnodes = 1;
}
- for (node = 0; node < numnodes; node++) {
- pg_data_ptr[node] = (pg_data_t*) &boot_pg_data[node];
- pg_data_ptr[node]->bdata = &bdata[node][0];
- }
-
min_low_pfn = -1;
max_low_pfn = 0;
efi_memmap_walk(filter_rsvd_memory, build_maps);
- efi_memmap_walk(filter_rsvd_memory, find_bootmap_space);
+ efi_memmap_walk(filter_rsvd_memory, find_pernode_space);
efi_memmap_walk(filter_rsvd_memory, discontig_free_bootmem_node);
- discontig_reserve_bootmem();
- allocate_pernode_structures();
-}
-
-/*
- * Initialize the paging system.
- * - determine sizes of each node
- * - initialize the paging system for the node
- * - build the nodedir for the node. This contains pointers to
- * the per-bank mem_map entries.
- * - fix the page struct "virtual" pointers. These are bank specific
- * values that the paging system doesn't understand.
- * - replicate the nodedir structure to other nodes
- */
-
-void __init
-discontig_paging_init(void)
-{
- int node, mynode;
- unsigned long max_dma, zones_size[MAX_NR_ZONES];
- unsigned long kaddr, ekaddr, bid;
- struct page *page;
- bootmem_data_t *bdp;
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
- mynode = boot_get_local_nodeid();
- for (node = 0; node < numnodes; node++) {
- long pfn, startpfn;
-
- memset(zones_size, 0, sizeof(zones_size));
-
- startpfn = -1;
- bdp = BOOT_NODE_DATA(node)->bdata;
- pfn = bdp->node_boot_start >> PAGE_SHIFT;
- if (startpfn = -1)
- startpfn = pfn;
- if (pfn > max_dma)
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - pfn);
- else if (bdp->node_low_pfn < max_dma)
- zones_size[ZONE_DMA] += (bdp->node_low_pfn - pfn);
- else {
- zones_size[ZONE_DMA] += (max_dma - pfn);
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - max_dma);
- }
-
- free_area_init_node(node, NODE_DATA(node), NULL, zones_size, startpfn, 0);
-
- page = NODE_DATA(node)->node_mem_map;
-
- bdp = BOOT_NODE_DATA(node)->bdata;
-
- kaddr = (unsigned long)__va(bdp->node_boot_start);
- ekaddr = (unsigned long)__va(bdp->node_low_pfn << PAGE_SHIFT);
- while (kaddr < ekaddr) {
- if (paddr_to_nid(__pa(kaddr)) = node) {
- bid = BANK_MEM_MAP_INDEX(kaddr);
- node_data[mynode]->node_id_map[bid] = node;
- node_data[mynode]->bank_mem_map_base[bid] = page;
- }
- kaddr += BANKSIZE;
- page += BANKSIZE/PAGE_SIZE;
- }
- }
-
- /*
- * Finish setting up the node data for this node, then copy it to the other nodes.
- */
- for (node=0; node < numnodes; node++)
- if (mynode != node) {
- memcpy(node_data[node], node_data[mynode], sizeof(struct ia64_node_data));
- node_data[node]->node = node;
- }
+ discontig_reserve_bootmem();
+ initialize_pernode_data();
}
diff -Nru a/arch/ia64/mm/init.c b/arch/ia64/mm/init.c
--- a/arch/ia64/mm/init.c Wed Jul 16 12:28:28 2003
+++ b/arch/ia64/mm/init.c Wed Jul 16 12:28:28 2003
@@ -44,7 +44,7 @@
#ifdef CONFIG_VIRTUAL_MEM_MAP
# define LARGE_GAP 0x40000000 /* Use virtual mem map if hole is > than this */
unsigned long vmalloc_end = VMALLOC_END_INIT;
- static struct page *vmem_map;
+ struct page *vmem_map;
static unsigned long num_dma_physpages;
#endif
@@ -240,7 +240,7 @@
else if (page_count(pgdat->node_mem_map + i))
shared += page_count(pgdat->node_mem_map + i) - 1;
}
- printk("\t%d pages of RAM\n", pgdat->node_spanned_pages);
+ printk("\t%ld pages of RAM\n", pgdat->node_spanned_pages);
printk("\t%d reserved pages\n", reserved);
printk("\t%d pages shared\n", shared);
printk("\t%d pages swap cached\n", cached);
@@ -397,6 +397,7 @@
{
unsigned long address, start_page, end_page;
struct page *map_start, *map_end;
+ int node;
pgd_t *pgd;
pmd_t *pmd;
pte_t *pte;
@@ -406,19 +407,20 @@
start_page = (unsigned long) map_start & PAGE_MASK;
end_page = PAGE_ALIGN((unsigned long) map_end);
+ node = paddr_to_nid(__pa(start));
for (address = start_page; address < end_page; address += PAGE_SIZE) {
pgd = pgd_offset_k(address);
if (pgd_none(*pgd))
- pgd_populate(&init_mm, pgd, alloc_bootmem_pages(PAGE_SIZE));
+ pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pmd = pmd_offset(pgd, address);
if (pmd_none(*pmd))
- pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages(PAGE_SIZE));
+ pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pte = pte_offset_kernel(pmd, address);
if (pte_none(*pte))
- set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages(PAGE_SIZE)) >> PAGE_SHIFT,
+ set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
PAGE_KERNEL));
}
return 0;
@@ -431,6 +433,14 @@
unsigned long zone;
};
+struct memmap_count_callback_data {
+ int node;
+ unsigned long num_physpages;
+ unsigned long num_dma_physpages;
+ unsigned long min_pfn;
+ unsigned long max_pfn;
+} cdata;
+
static int
virtual_memmap_init (u64 start, u64 end, void *arg)
{
@@ -489,16 +499,6 @@
}
static int
-count_dma_pages (u64 start, u64 end, void *arg)
-{
- unsigned long *count = arg;
-
- if (end <= MAX_DMA_ADDRESS)
- *count += (end - start) >> PAGE_SHIFT;
- return 0;
-}
-
-static int
find_largest_hole (u64 start, u64 end, void *arg)
{
u64 *max_gap = arg;
@@ -514,102 +514,101 @@
}
#endif /* CONFIG_VIRTUAL_MEM_MAP */
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
+#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
+#define ORDERROUNDDOWN(n) ((n) & ~((PAGE_SIZE<<MAX_ORDER)-1))
static int
-count_pages (u64 start, u64 end, void *arg)
+count_pages (unsigned long start, unsigned long end, int node)
{
- unsigned long *count = arg;
+ start = __pa(start);
+ end = __pa(end);
- *count += (end - start) >> PAGE_SHIFT;
+ if (node = cdata.node) {
+ cdata.num_physpages += (end - start) >> PAGE_SHIFT;
+ if (start <= __pa(MAX_DMA_ADDRESS))
+ cdata.num_dma_physpages += (min(end, __pa(MAX_DMA_ADDRESS)) - start) >> PAGE_SHIFT;
+ start = GRANULEROUNDDOWN(__pa(start));
+ start = ORDERROUNDDOWN(start);
+ end = GRANULEROUNDUP(__pa(end));
+ cdata.max_pfn = max(cdata.max_pfn, end >> PAGE_SHIFT);
+ cdata.min_pfn = min(cdata.min_pfn, start >> PAGE_SHIFT);
+ }
return 0;
}
/*
* Set up the page tables.
*/
-
-#ifdef CONFIG_DISCONTIGMEM
void
paging_init (void)
{
- extern void discontig_paging_init(void);
-
- discontig_paging_init();
- efi_memmap_walk(count_pages, &num_physpages);
- zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
-}
-#else /* !CONFIG_DISCONTIGMEM */
-void
-paging_init (void)
-{
- unsigned long max_dma;
+ unsigned long max_dma_pfn;
unsigned long zones_size[MAX_NR_ZONES];
# ifdef CONFIG_VIRTUAL_MEM_MAP
unsigned long zholes_size[MAX_NR_ZONES];
unsigned long max_gap;
# endif
+ int node;
- /* initialize mem_map[] */
-
- memset(zones_size, 0, sizeof(zones_size));
-
- num_physpages = 0;
- efi_memmap_walk(count_pages, &num_physpages);
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
-
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- memset(zholes_size, 0, sizeof(zholes_size));
-
- num_dma_physpages = 0;
- efi_memmap_walk(count_dma_pages, &num_dma_physpages);
-
- if (max_low_pfn < max_dma) {
- zones_size[ZONE_DMA] = max_low_pfn;
- zholes_size[ZONE_DMA] = max_low_pfn - num_dma_physpages;
- } else {
- zones_size[ZONE_DMA] = max_dma;
- zholes_size[ZONE_DMA] = max_dma - num_dma_physpages;
- if (num_physpages > num_dma_physpages) {
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
- zholes_size[ZONE_NORMAL] = ((max_low_pfn - max_dma)
- - (num_physpages - num_dma_physpages));
- }
- }
-
+ max_dma_pfn = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
max_gap = 0;
efi_memmap_walk(find_largest_hole, (u64 *)&max_gap);
- if (max_gap < LARGE_GAP) {
- vmem_map = (struct page *) 0;
- free_area_init_node(0, &contig_page_data, NULL, zones_size, 0, zholes_size);
- mem_map = contig_page_data.node_mem_map;
- }
- else {
- unsigned long map_size;
-
- /* allocate virtual_mem_map */
- map_size = PAGE_ALIGN(max_low_pfn * sizeof(struct page));
- vmalloc_end -= map_size;
- vmem_map = (struct page *) vmalloc_end;
- efi_memmap_walk(create_mem_map_page_table, 0);
-
- free_area_init_node(0, &contig_page_data, vmem_map, zones_size, 0, zholes_size);
+ for (node = 0; node < numnodes; node++) {
+ memset(zones_size, 0, sizeof(zones_size));
+ memset(zholes_size, 0, sizeof(zholes_size));
+ memset(&cdata, 0, sizeof(cdata));
+
+ cdata.node = node;
+ cdata.min_pfn = ~0;
+
+ efi_memmap_walk(filter_rsvd_memory, count_pages);
+ num_dma_physpages += cdata.num_dma_physpages;
+ num_physpages += cdata.num_physpages;
+
+ if (cdata.min_pfn >= max_dma_pfn) {
+ /* Above the DMA zone */
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn - cdata.num_physpages;
+ } else if (cdata.max_pfn < max_dma_pfn) {
+ /* This block is DMAable */
+ zones_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn - cdata.num_dma_physpages;
+ } else {
+ zones_size[ZONE_DMA] = max_dma_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - cdata.num_dma_physpages;
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - max_dma_pfn;
+ zholes_size[ZONE_NORMAL] = zones_size[ZONE_NORMAL] - (cdata.num_physpages - cdata.num_dma_physpages);
+ }
- mem_map = contig_page_data.node_mem_map;
- printk("Virtual mem_map starts at 0x%p\n", mem_map);
- }
-# else /* !CONFIG_VIRTUAL_MEM_MAP */
- if (max_low_pfn < max_dma)
- zones_size[ZONE_DMA] = max_low_pfn;
- else {
- zones_size[ZONE_DMA] = max_dma;
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
+ if (numnodes = 1 && max_gap < LARGE_GAP) {
+ /* Just one node with no big holes... */
+ vmem_map = (struct page *)0;
+ zones_size[ZONE_DMA] += cdata.min_pfn;
+ zholes_size[ZONE_DMA] += cdata.min_pfn;
+ free_area_init_node(0, NODE_DATA(node), NODE_DATA(node)->node_mem_map,
+ zones_size, 0, zholes_size);
+ }
+ else {
+ /* allocate virtual mem_map */
+ if (node = 0) {
+ unsigned long map_size;
+ map_size = PAGE_ALIGN(max_low_pfn*sizeof(struct page));
+ vmalloc_end -= map_size;
+ vmem_map = (struct page *) vmalloc_end;
+ efi_memmap_walk(create_mem_map_page_table, 0);
+ printk("Virtual mem_map starts at 0x%p\n", vmem_map);
+#ifndef CONFIG_DISCONTIGMEM
+ mem_map = vmem_map;
+#endif
+ }
+ free_area_init_node(node, NODE_DATA(node), vmem_map + cdata.min_pfn,
+ zones_size, cdata.min_pfn, zholes_size);
+ }
}
- free_area_init(zones_size);
-# endif /* !CONFIG_VIRTUAL_MEM_MAP */
+
zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
}
-#endif /* !CONFIG_DISCONTIGMEM */
static int
count_reserved_pages (u64 start, u64 end, void *arg)
diff -Nru a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
--- a/drivers/acpi/Kconfig Wed Jul 16 12:28:28 2003
+++ b/drivers/acpi/Kconfig Wed Jul 16 12:28:28 2003
@@ -133,7 +133,7 @@
config ACPI_NUMA
bool "NUMA support" if NUMA && (IA64 && !IA64_HP_SIM || X86 && ACPI && !ACPI_HT_ONLY && !X86_64)
- default y if IA64 && IA64_SGI_SN
+ default y if IA64_GENERIC || IA64_SGI_SN2
config ACPI_ASUS
tristate "ASUS/Medion Laptop Extras"
diff -Nru a/include/asm-ia64/mmzone.h b/include/asm-ia64/mmzone.h
--- a/include/asm-ia64/mmzone.h Wed Jul 16 12:28:28 2003
+++ b/include/asm-ia64/mmzone.h Wed Jul 16 12:28:28 2003
@@ -3,7 +3,7 @@
* License. See the file "COPYING" in the main directory of this archive
* for more details.
*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000,2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2002 NEC Corp.
* Copyright (c) 2002 Erich Focht <efocht@ess.nec.de>
* Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com>
@@ -14,150 +14,50 @@
#include <linux/config.h>
#include <linux/init.h>
-/*
- * Given a kaddr, find the base mem_map address for the start of the mem_map
- * entries for the bank containing the kaddr.
- */
-#define BANK_MEM_MAP_BASE(kaddr) local_node_data->bank_mem_map_base[BANK_MEM_MAP_INDEX(kaddr)]
-
-/*
- * Given a kaddr, this macro return the relative map number
- * within the bank.
- */
-#define BANK_MAP_NR(kaddr) (BANK_OFFSET(kaddr) >> PAGE_SHIFT)
-/*
- * Given a pte, this macro returns a pointer to the page struct for the pte.
- */
-#define pte_page(pte) virt_to_page(PAGE_OFFSET | (pte_val(pte)&_PFN_MASK))
+#ifdef CONFIG_NUMA
-/*
- * Determine if a kaddr is a valid memory address of memory that
- * actually exists.
- *
- * The check consists of 2 parts:
- * - verify that the address is a region 7 address & does not
- * contain any bits that preclude it from being a valid platform
- * memory address
- * - verify that the chunk actually exists.
- *
- * Note that IO addresses are NOT considered valid addresses.
- *
- * Note, many platforms can simply check if kaddr exceeds a specific size.
- * (However, this won't work on SGI platforms since IO space is embedded
- * within the range of valid memory addresses & nodes have holes in the
- * address range between banks).
- */
-#define kern_addr_valid(kaddr) ({long _kav=(long)(kaddr); \
- VALID_MEM_KADDR(_kav);})
-
-/*
- * Given a kaddr, return a pointer to the page struct for the page.
- * If the kaddr does not represent RAM memory that potentially exists, return
- * a pointer the page struct for max_mapnr. IO addresses will
- * return the page for max_nr. Addresses in unpopulated RAM banks may
- * return undefined results OR may panic the system.
- *
- */
-#define virt_to_page(kaddr) ({long _kvtp=(long)(kaddr); \
- (VALID_MEM_KADDR(_kvtp)) \
- ? BANK_MEM_MAP_BASE(_kvtp) + BANK_MAP_NR(_kvtp) \
- : NULL;})
+#ifdef CONFIG_IA64_DIG
/*
- * Given a page struct entry, return the physical address that the page struct represents.
- * Since IA64 has all memory in the DMA zone, the following works:
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define page_to_phys(page) __pa(page_address(page))
-
-#define node_mem_map(nid) (NODE_DATA(nid)->node_mem_map)
+#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
+#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES * 32)
-#define node_localnr(pfn, nid) ((pfn) - NODE_DATA(nid)->node_start_pfn)
-#define pfn_to_page(pfn) (struct page *)(node_mem_map(pfn_to_nid(pfn)) + node_localnr(pfn, pfn_to_nid(pfn)))
-#define pfn_to_nid(pfn) local_node_data->node_id_map[(pfn << PAGE_SHIFT) >> BANKSHIFT]
-
-#define page_to_pfn(page) (long)((page - page_zone(page)->zone_mem_map) + page_zone(page)->zone_start_pfn)
+#elif CONFIG_IA64_SGI_SN2
/*
- * pfn_valid should be made as fast as possible, and the current definition
- * is valid for machines that are NUMA, but still contiguous, which is what
- * is currently supported. A more generalised, but slower definition would
- * be something like this - mbligh:
- * ( pfn_to_pgdat(pfn) && (pfn < node_end_pfn(pfn_to_nid(pfn))) )
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define pfn_valid(pfn) (pfn < max_low_pfn)
-extern unsigned long max_low_pfn;
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of compute nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
+#elif CONFIG_IA64_GENERIC
-#ifdef CONFIG_IA64_DIG
/*
- * Platform definitions for DIG platform with contiguous memory.
+ * Platform definitions for GENERIC platform with contiguous or discontiguous memory.
*/
-#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
-#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
-#define MAX_PHYS_MEMORY (1UL << 40) /* 1 TB */
-/*
- * Bank definitions.
- * Configurable settings for DIG: 512MB/bank: 16GB/node,
- * 2048MB/bank: 64GB/node,
- * 8192MB/bank: 256GB/node.
- */
-#define NR_BANKS_PER_NODE 32
-#if defined(CONFIG_IA64_NODESIZE_16GB)
-# define BANKSHIFT 29
-#elif defined(CONFIG_IA64_NODESIZE_64GB)
-# define BANKSHIFT 31
-#elif defined(CONFIG_IA64_NODESIZE_256GB)
-# define BANKSHIFT 33
#else
-# error Unsupported bank and nodesize!
+#error unknown platform
#endif
-#define BANKSIZE (1UL << BANKSHIFT)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-/*
- * VALID_MEM_KADDR returns a boolean to indicate if a kaddr is
- * potentially a valid cacheable identity mapped RAM memory address.
- * Note that the RAM may or may not actually be present!!
- */
-#define VALID_MEM_KADDR(kaddr) 1
+extern void build_cpu_to_node_map(void);
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#else /* CONFIG_NUMA */
-#elif defined(CONFIG_IA64_SGI_SN2)
-/*
- * SGI SN2 discontig definitions
- */
-#define MAX_PHYSNODE_ID 2048 /* 2048 node ids (also called nasid) */
-#define NR_NODES 128 /* Maximum number of nodes in SSI */
-#define MAX_PHYS_MEMORY (1UL << 49)
-
-#define BANKSHIFT 38
-#define NR_BANKS_PER_NODE 4
-#define SN2_NODE_SIZE (64UL*1024*1024*1024) /* 64GB per node */
-#define BANKSIZE (SN2_NODE_SIZE/NR_BANKS_PER_NODE)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-#define VALID_MEM_KADDR(kaddr) 1
-
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#define NR_NODES 1
-#endif /* CONFIG_IA64_DIG */
+#endif /* CONFIG_NUMA */
#endif /* _ASM_IA64_MMZONE_H */
diff -Nru a/include/asm-ia64/nodedata.h b/include/asm-ia64/nodedata.h
--- a/include/asm-ia64/nodedata.h Wed Jul 16 12:28:28 2003
+++ b/include/asm-ia64/nodedata.h Wed Jul 16 12:28:28 2003
@@ -13,7 +13,7 @@
#ifndef _ASM_IA64_NODEDATA_H
#define _ASM_IA64_NODEDATA_H
-
+#include <asm/percpu.h>
#include <asm/mmzone.h>
/*
@@ -22,15 +22,17 @@
struct pglist_data;
struct ia64_node_data {
- short active_cpu_count;
short node;
+ short active_cpu_count;
+ /*
+ * The fields are read-only (after boot). They contain pointers to various structures
+ * located on other nodes. Ths data is replicated on each node in order to reduce
+ * off-node references.
+ */
struct pglist_data *pg_data_ptrs[NR_NODES];
- struct page *bank_mem_map_base[NR_BANKS];
struct ia64_node_data *node_data_ptrs[NR_NODES];
- short node_id_map[NR_BANKS];
};
-
/*
* Return a pointer to the node_data structure for the executing cpu.
*/
@@ -40,7 +42,8 @@
/*
* Return a pointer to the node_data structure for the specified node.
*/
-#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
/*
* Get a pointer to the node_id/node_data for the current cpu.
@@ -48,29 +51,5 @@
*/
extern int boot_get_local_nodeid(void);
extern struct ia64_node_data *get_node_data_ptr(void);
-
-/*
- * Given a node id, return a pointer to the pg_data_t for the node.
- * The following 2 macros are similar.
- *
- * NODE_DATA - should be used in all code not related to system
- * initialization. It uses pernode data structures to minimize
- * offnode memory references. However, these structure are not
- * present during boot. This macro can be used once cpu_init
- * completes.
- *
- * BOOT_NODE_DATA
- * - should be used during system initialization
- * prior to freeing __initdata. It does not depend on the percpu
- * area being present.
- *
- * NOTE: The names of these macros are misleading but are difficult to change
- * since they are used in generic linux & on other architecures.
- */
-#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
-#define BOOT_NODE_DATA(nid) boot_get_pg_data_ptr((long)(nid))
-
-struct pglist_data;
-extern struct pglist_data * __init boot_get_pg_data_ptr(long);
#endif /* _ASM_IA64_NODEDATA_H */
diff -Nru a/include/asm-ia64/numa.h b/include/asm-ia64/numa.h
--- a/include/asm-ia64/numa.h Wed Jul 16 12:28:28 2003
+++ b/include/asm-ia64/numa.h Wed Jul 16 12:28:28 2003
@@ -15,13 +15,21 @@
#ifdef CONFIG_DISCONTIGMEM
# include <asm/mmzone.h>
-# define NR_MEMBLKS (NR_BANKS)
#else
# define NR_NODES (8)
# define NR_MEMBLKS (NR_NODES * 8)
#endif
#include <linux/cache.h>
+#include <linux/threads.h>
+#include <linux/smp.h>
+
+#define NODEMASK_WORDCOUNT ((NR_NODES+(BITS_PER_LONG-1))/BITS_PER_LONG)
+
+#define NODE_MASK_NONE { [0 ... ((NR_NODES+BITS_PER_LONG-1)/BITS_PER_LONG)-1] = 0 }
+
+typedef unsigned long nodemask_t[NODEMASK_WORDCOUNT];
+
extern volatile char cpu_to_node_map[NR_CPUS] __cacheline_aligned;
extern volatile unsigned long node_to_cpu_mask[NR_NODES] __cacheline_aligned;
@@ -63,6 +71,12 @@
extern int paddr_to_nid(unsigned long paddr);
#define local_nodeid (cpu_to_node_map[smp_processor_id()])
+
+#else /* !CONFIG_NUMA */
+
+#define node_distance(from,to) 10
+#define paddr_to_nid(x) 0
+#define local_nodeid 0
#endif /* CONFIG_NUMA */
diff -Nru a/include/asm-ia64/page.h b/include/asm-ia64/page.h
--- a/include/asm-ia64/page.h Wed Jul 16 12:28:28 2003
+++ b/include/asm-ia64/page.h Wed Jul 16 12:28:28 2003
@@ -93,18 +93,26 @@
#define virt_addr_valid(kaddr) pfn_valid(__pa(kaddr) >> PAGE_SHIFT)
-#ifndef CONFIG_DISCONTIGMEM
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- extern int ia64_pfn_valid (unsigned long pfn);
-# define pfn_valid(pfn) (((pfn) < max_mapnr) && ia64_pfn_valid(pfn))
-# else
-# define pfn_valid(pfn) ((pfn) < max_mapnr)
-# endif
-#define virt_to_page(kaddr) pfn_to_page(__pa(kaddr) >> PAGE_SHIFT)
-#define page_to_pfn(page) ((unsigned long) (page - mem_map))
-#define pfn_to_page(pfn) (mem_map + (pfn))
-#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
+#ifdef CONFIG_VIRTUAL_MEM_MAP
+extern int ia64_pfn_valid(unsigned long pfn);
+#else
+#define ia64_pfn_valid(pfn) (1)
+#endif
+
+extern unsigned long max_low_pfn;
+#define pfn_valid(pfn) (((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
+
+#if defined(CONFIG_VIRTUAL_MEM_MAP) && !defined(CONFIG_DISCONTIGMEM)
+#define vmem_map mem_map
+#else
+extern struct page *vmem_map;
#endif
+
+#define pfn_to_page(pfn) (vmem_map + (pfn))
+#define page_to_pfn(page) ((unsigned long) (page - vmem_map))
+
+#define virt_to_page(kaddr) (pfn_to_page(__pa(kaddr) >> PAGE_SHIFT))
+#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
typedef union ia64_va {
struct {
diff -Nru a/include/asm-ia64/pgtable.h b/include/asm-ia64/pgtable.h
--- a/include/asm-ia64/pgtable.h Wed Jul 16 12:28:28 2003
+++ b/include/asm-ia64/pgtable.h Wed Jul 16 12:28:28 2003
@@ -174,7 +174,6 @@
return (addr & (local_cpu_data->unimpl_pa_mask)) = 0;
}
-#ifndef CONFIG_DISCONTIGMEM
/*
* kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
* memory. For the return value to be meaningful, ADDR must be >@@ -190,7 +189,6 @@
*/
#define kern_addr_valid(addr) (1)
-#endif
/*
* Now come the defines and routines to manage and access the three-level
@@ -241,10 +239,8 @@
#define pte_none(pte) (!pte_val(pte))
#define pte_present(pte) (pte_val(pte) & (_PAGE_P | _PAGE_PROTNONE))
#define pte_clear(pte) (pte_val(*(pte)) = 0UL)
-#ifndef CONFIG_DISCONTIGMEM
/* pte_page() returns the "struct page *" corresponding to the PTE: */
#define pte_page(pte) virt_to_page(((pte_val(pte) & _PFN_MASK) + PAGE_OFFSET))
-#endif
#define pmd_none(pmd) (!pmd_val(pmd))
#define pmd_bad(pmd) (!ia64_phys_addr_valid(pmd_val(pmd)))
@@ -416,6 +412,7 @@
extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
extern void paging_init (void);
+extern int filter_rsvd_memory(unsigned long start, unsigned long end, void *arg);
/*
* Note: The macros below rely on the fact that MAX_SWAPFILES_SHIFT <= number of
diff -Nru a/include/asm-ia64/processor.h b/include/asm-ia64/processor.h
--- a/include/asm-ia64/processor.h Wed Jul 16 12:28:28 2003
+++ b/include/asm-ia64/processor.h Wed Jul 16 12:28:28 2003
@@ -185,6 +185,8 @@
#endif
#ifdef CONFIG_NUMA
struct ia64_node_data *node_data;
+ struct cpuinfo_ia64 *cpu_data[NR_CPUS];
+ int nodeid;
#endif
};
diff -Nru a/mm/bootmem.c b/mm/bootmem.c
--- a/mm/bootmem.c Wed Jul 16 12:28:28 2003
+++ b/mm/bootmem.c Wed Jul 16 12:28:28 2003
@@ -48,8 +48,24 @@
bootmem_data_t *bdata = pgdat->bdata;
unsigned long mapsize = ((end - start)+7)/8;
- pgdat->pgdat_next = pgdat_list;
- pgdat_list = pgdat;
+
+ /*
+ * sort pgdat_list so that the lowest one comes first,
+ * which makes alloc_bootmem_low_pages work as desired.
+ */
+ if (!pgdat_list || pgdat_list->node_start_pfn > pgdat->node_start_pfn) {
+ pgdat->pgdat_next = pgdat_list;
+ pgdat_list = pgdat;
+ } else {
+ pg_data_t *tmp = pgdat_list;
+ while (tmp->pgdat_next) {
+ if (tmp->pgdat_next->node_start_pfn > pgdat->node_start_pfn)
+ break;
+ tmp = tmp->pgdat_next;
+ }
+ pgdat->pgdat_next = tmp->pgdat_next;
+ tmp->pgdat_next = pgdat;
+ }
mapsize = (mapsize + (sizeof(long) - 1UL)) & ~(sizeof(long) - 1UL);
bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
@@ -251,7 +267,7 @@
static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
{
- struct page *page = pgdat->node_mem_map;
+ struct page *page;
bootmem_data_t *bdata = pgdat->bdata;
unsigned long i, count, total = 0;
unsigned long idx;
@@ -260,23 +276,23 @@
if (!bdata->node_bootmem_map) BUG();
count = 0;
+ page = virt_to_page(phys_to_virt(bdata->node_boot_start));
idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
map = bdata->node_bootmem_map;
for (i = 0; i < idx; ) {
unsigned long v = ~map[i / BITS_PER_LONG];
if (v) {
unsigned long m;
- for (m = 1; m && i < idx; m<<=1, page++, i++) {
+ for (m = 1; m && i < idx; m<<=1, i++) {
if (v & m) {
count++;
- ClearPageReserved(page);
- set_page_count(page, 1);
- __free_page(page);
+ ClearPageReserved(page+i);
+ set_page_count(page+i, 1);
+ __free_page(page+i);
}
}
} else {
i+=BITS_PER_LONG;
- page += BITS_PER_LONG;
}
}
total += count;
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (5 preceding siblings ...)
2003-07-16 19:29 ` Jesse Barnes
@ 2003-07-16 19:40 ` Matthew Wilcox
2003-07-16 19:51 ` Jesse Barnes
` (6 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Matthew Wilcox @ 2003-07-16 19:40 UTC (permalink / raw)
To: linux-ia64
On Wed, Jul 16, 2003 at 12:29:52PM -0700, Jesse Barnes wrote:
> @@ -210,8 +210,8 @@
> system with an A0 or A1 stepping CPU.
>
> config NUMA
> - bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
> - default y if IA64_SGI_SN2
> + bool
> + default y if IA64_SGI_SN2 || IA64_GENERIC
> help
> Say Y to compile the kernel to support NUMA (Non-Uniform Memory
> Access). This option is for configuring high-end multiprocessor
If you're removing the question, you can remove the helptext too.
> @@ -235,8 +235,7 @@
>
> config DISCONTIGMEM
> bool
> - depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) && NUMA
> - default y
> + default y if IA64_SGI_SN2 || IA64_GENERIC
> help
> Say Y to support efficient handling of discontiguous physical memory,
> for architectures which are either NUMA (Non-Uniform Memory Access)
This one already shouldn't have helptext ;-)
> - * on a BANKSIZE boundary. Multiple banks per node is not supported.
> + * on a GRANULE boundary. Multiple banks per node is not supported.
Multiple banks *are* not supported ;-) The feature of multiple banks
*is* not supported.
> @@ -22,15 +22,17 @@
>
> struct pglist_data;
> struct ia64_node_data {
> - short active_cpu_count;
> short node;
> + short active_cpu_count;
> + /*
> + * The fields are read-only (after boot). They contain pointers to various structures
> + * located on other nodes. Ths data is replicated on each node in order to reduce
> + * off-node references.
> + */
Can you wrap comments at 80 columns? It makes them much easier to read.
--
"It's not Hollywood. War is real, war is primarily not about defeat or
victory, it is about death. I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (6 preceding siblings ...)
2003-07-16 19:40 ` Matthew Wilcox
@ 2003-07-16 19:51 ` Jesse Barnes
2003-07-16 19:56 ` Erich Focht
` (5 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Jesse Barnes @ 2003-07-16 19:51 UTC (permalink / raw)
To: linux-ia64
Thanks for looking at it. Here's another one that fixes that stuff up.
On Wed, Jul 16, 2003 at 08:40:51PM +0100, Matthew Wilcox wrote:
> > + default y if IA64_SGI_SN2 || IA64_GENERIC
> > help
> > Say Y to compile the kernel to support NUMA (Non-Uniform Memory
> > Access). This option is for configuring high-end multiprocessor
>
> If you're removing the question, you can remove the helptext too.
>
> > + default y if IA64_SGI_SN2 || IA64_GENERIC
> > help
> > Say Y to support efficient handling of discontiguous physical memory,
> > for architectures which are either NUMA (Non-Uniform Memory Access)
>
> This one already shouldn't have helptext ;-)
>
> > - * on a BANKSIZE boundary. Multiple banks per node is not supported.
> > + * on a GRANULE boundary. Multiple banks per node is not supported.
>
> Multiple banks *are* not supported ;-) The feature of multiple banks
> *is* not supported.
>
> > @@ -22,15 +22,17 @@
> >
> > struct pglist_data;
> > struct ia64_node_data {
> > - short active_cpu_count;
> > short node;
> > + short active_cpu_count;
> > + /*
> > + * The fields are read-only (after boot). They contain pointers to various structures
> > + * located on other nodes. Ths data is replicated on each node in order to reduce
> > + * off-node references.
> > + */
>
> Can you wrap comments at 80 columns? It makes them much easier to read.
I prefer it that way too, but David likes 100 columns, AFAICT. :) Fixed
anyway.
Thanks,
Jesse
# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
# ChangeSet 1.1396 -> 1.1398
# include/asm-ia64/page.h 1.19 -> 1.20
# arch/ia64/kernel/setup.c 1.53 -> 1.54
# include/asm-ia64/pgtable.h 1.28 -> 1.29
# mm/bootmem.c 1.18 -> 1.19
# include/asm-ia64/numa.h 1.5 -> 1.6
# include/asm-ia64/processor.h 1.48 -> 1.49
# arch/ia64/mm/init.c 1.46 -> 1.47
# include/asm-ia64/nodedata.h 1.3 -> 1.5
# arch/ia64/mm/discontig.c 1.4 -> 1.6
# arch/ia64/Kconfig 1.38 -> 1.40
# drivers/acpi/Kconfig 1.12 -> 1.13
# include/asm-ia64/mmzone.h 1.4 -> 1.5
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 03/07/16 jbarnes@tomahawk.engr.sgi.com 1.1397
# discontig update
# --------------------------------------------
# 03/07/16 jbarnes@tomahawk.engr.sgi.com 1.1398
# more discontig stuff
# --------------------------------------------
#
diff -Nru a/arch/ia64/Kconfig b/arch/ia64/Kconfig
--- a/arch/ia64/Kconfig Wed Jul 16 12:50:02 2003
+++ b/arch/ia64/Kconfig Wed Jul 16 12:50:02 2003
@@ -210,12 +210,8 @@
system with an A0 or A1 stepping CPU.
config NUMA
- bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
- help
- Say Y to compile the kernel to support NUMA (Non-Uniform Memory
- Access). This option is for configuring high-end multiprocessor
- server systems. If in doubt, say N.
+ bool
+ default y if IA64_SGI_SN2 || IA64_GENERIC
choice
prompt "Maximum Memory per NUMA Node" if NUMA && IA64_DIG
@@ -235,18 +231,11 @@
config DISCONTIGMEM
bool
- depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) && NUMA
- default y
- help
- Say Y to support efficient handling of discontiguous physical memory,
- for architectures which are either NUMA (Non-Uniform Memory Access)
- or have huge holes in the physical address space for other reasons.
- See <file:Documentation/vm/numa> for more.
+ default y if IA64_SGI_SN2 || IA64_GENERIC
config VIRTUAL_MEM_MAP
bool "Enable Virtual Mem Map"
- depends on !NUMA
- default y if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
+ default y if !IA64_HP_SIM
help
Say Y to compile the kernel with support for a virtual mem map.
This is an alternate method of supporting large holes in the
@@ -259,8 +248,8 @@
are unsure, say Y.
config IA64_MCA
- bool "Enable IA-64 Machine Check Abort" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
+ bool "Enable IA-64 Machine Check Abort"
+ default y if !IA64_HP_SIM
help
Say Y here to enable machine check support for IA-64. If you're
unsure, answer Y.
@@ -292,43 +281,12 @@
depends on IA64_GENERIC || IA64_DIG || IA64_HP_ZX1 || IA64_SGI_SN2
default y
-config IA64_SGI_SN_DEBUG
- bool "Enable extra debugging code"
- depends on IA64_SGI_SN2
- help
- Turns on extra debugging code in the SGI SN (Scalable NUMA) platform
- for IA-64. Unless you are debugging problems on an SGI SN IA-64 box,
- say N.
-
config IA64_SGI_SN_SIM
bool "Enable SGI Medusa Simulator Support"
depends on IA64_SGI_SN2
help
If you are compiling a kernel that will run under SGI's IA-64
simulator (Medusa) then say Y, otherwise say N.
-
-config IA64_SGI_AUTOTEST
- bool "Enable autotest (llsc). Option to run cache test instead of booting"
- depends on IA64_SGI_SN2
- help
- Build a kernel used for hardware validation. If you include the
- keyword "autotest" on the boot command line, the kernel does NOT boot.
- Instead, it starts all cpus and runs cache coherency tests instead.
-
- If unsure, say N.
-
-config SERIAL_SGI_L1_PROTOCOL
- bool "Enable protocol mode for the L1 console"
- depends on IA64_SGI_SN2
- help
- Uses protocol mode instead of raw mode for the level 1 console on the
- SGI SN (Scalable NUMA) platform for IA-64. If you are compiling for
- an SGI SN box then Y is the recommended value, otherwise say N.
-
-config PERCPU_IRQ
- bool
- depends on IA64_SGI_SN2
- default y
# On IA-64, we always want an ELF /proc/kcore.
config KCORE_ELF
diff -Nru a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c
--- a/arch/ia64/kernel/setup.c Wed Jul 16 12:50:02 2003
+++ b/arch/ia64/kernel/setup.c Wed Jul 16 12:50:02 2003
@@ -138,7 +138,7 @@
call_pernode_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long rs, re;
- void (*func)(unsigned long, unsigned long, int, int);
+ void (*func)(unsigned long, unsigned long, int);
int i;
start = PAGE_ALIGN(start);
@@ -149,22 +149,21 @@
func = arg;
if (!num_memblks) {
- /*
- * This machine doesn't have SRAT, so call func with
- * nid=0, bank=0.
- */
+ /* No SRAT table, to assume one node (node 0) */
if (start < end)
- (*func)(start, end - start, 0, 0);
+ (*func)(start, end, 0);
return;
}
for (i = 0; i < num_memblks; i++) {
- rs = max(start, node_memblk[i].start_paddr);
- re = min(end, node_memblk[i].start_paddr+node_memblk[i].size);
+ rs = max(__pa(start), node_memblk[i].start_paddr);
+ re = min(__pa(end), node_memblk[i].start_paddr+node_memblk[i].size);
if (rs < re)
- (*func)(rs, re-rs, node_memblk[i].nid,
- node_memblk[i].bank);
+ (*func)((unsigned long)__va(rs), (unsigned long)__va(re), node_memblk[i].nid);
+
+ if ((unsigned long)__va(re) = end)
+ break;
}
}
@@ -180,7 +179,7 @@
filter_rsvd_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long range_start, range_end, prev_start;
- void (*func)(unsigned long, unsigned long);
+ void (*func)(unsigned long, unsigned long, int);
int i;
#if IGNORE_PFN0
@@ -202,9 +201,9 @@
if (range_start < range_end)
#ifdef CONFIG_DISCONTIGMEM
- call_pernode_memory(__pa(range_start), __pa(range_end), func);
+ call_pernode_memory(range_start, range_end, func);
#else
- (*func)(__pa(range_start), range_end - range_start);
+ (*func)(range_start, range_end, 0);
#endif
/* nothing more available in this segment */
@@ -703,6 +702,8 @@
* get_free_pages() cannot be used before cpu_init() done. BSP allocates
* "NR_CPUS" pages for all CPUs to avoid that AP calls get_zeroed_page().
*/
+#ifndef CONFIG_DISCONTIGMEM
+ /* for discontig machines, we do this in discontig.c */
if (smp_processor_id() = 0) {
cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * NR_CPUS, PERCPU_PAGE_SIZE,
__pa(MAX_DMA_ADDRESS));
@@ -714,6 +715,7 @@
per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
}
}
+#endif
cpu_data = __per_cpu_start + __per_cpu_offset[smp_processor_id()];
#else /* !CONFIG_SMP */
cpu_data = __phys_per_cpu_start;
diff -Nru a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c
--- a/arch/ia64/mm/discontig.c Wed Jul 16 12:50:02 2003
+++ b/arch/ia64/mm/discontig.c Wed Jul 16 12:50:02 2003
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000, 2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2001 Intel Corp.
* Copyright (c) 2001 Tony Luck <tony.luck@intel.com>
* Copyright (c) 2002 NEC Corp.
@@ -16,74 +16,60 @@
#include <linux/mmzone.h>
#include <linux/acpi.h>
#include <linux/efi.h>
-
+#include <asm/pgalloc.h>
+#include <asm/tlb.h>
/*
- * Round an address upward to the next multiple of GRANULE size.
+ * Round an address upward or downward to the next multiple of IA64_GRANULE_SIZE.
*/
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
-static struct ia64_node_data *node_data[NR_NODES];
-static long boot_pg_data[8*NR_NODES+sizeof(pg_data_t)] __initdata;
-static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
-static bootmem_data_t bdata[NR_NODES][NR_BANKS_PER_NODE+1] __initdata;
-
-extern int filter_rsvd_memory (unsigned long start, unsigned long end, void *arg);
+/*
+ * Used to locate BOOT_DATA prior to initializing the node data area.
+ */
+#define BOOT_NODE_DATA(node) pg_data_ptr[node]
/*
- * Return the compact node number of this cpu. Used prior to
- * setting up the cpu_data area.
- * Note - not fast, intended for boot use only!!
+ * To prevent cache aliasing effects, align per-node structures so that they
+ * start at addresses that are strided by node number.
*/
-int
-boot_get_local_nodeid(void)
-{
- int i;
+#define NODEDATA_ALIGN(addr, node) ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + (node)*PERCPU_PAGE_SIZE)
- for (i = 0; i < NR_CPUS; i++)
- if (node_cpuid[i].phys_id = hard_smp_processor_id())
- return node_cpuid[i].nid;
- /* node info missing, so nid should be 0.. */
- return 0;
-}
+static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
+static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
+static bootmem_data_t bdata[NR_NODES] __initdata;
+static unsigned long boot_pernode[NR_NODES] __initdata;
+static unsigned long boot_pernodesize[NR_NODES] __initdata;
-/*
- * Return a pointer to the pg_data structure for a node.
- * This function is used ONLY in early boot before the cpu_data
- * structure is available.
- */
-pg_data_t* __init
-boot_get_pg_data_ptr(long node)
-{
- return pg_data_ptr[node];
-}
+extern char __per_cpu_start[], __per_cpu_end[];
-/*
- * Return a pointer to the node data for the current node.
- * (boottime initialization only)
- */
-struct ia64_node_data *
+struct ia64_node_data*
get_node_data_ptr(void)
{
- return node_data[boot_get_local_nodeid()];
+ return boot_node_data[cpu_to_node_map[smp_processor_id()]]; /* ZZZ */
}
/*
* We allocate one of the bootmem_data_t structs for each piece of memory
* that we wish to treat as a contiguous block. Each such block must start
- * on a BANKSIZE boundary. Multiple banks per node is not supported.
+ * on a GRANULE boundary. Multiple banks per node are not supported.
+ * (Note: on SN2, all memory on a node is trated as a single bank.
+ * Holes within the bank are supported. This works because memory
+ * from different banks is not interleaved. The bootmap bitmap
+ * for the node is somewhat large but not too large).
*/
static int __init
-build_maps(unsigned long pstart, unsigned long length, int node)
+build_maps(unsigned long start, unsigned long end, int node)
{
bootmem_data_t *bdp;
unsigned long cstart, epfn;
- bdp = pg_data_ptr[node]->bdata;
- epfn = GRANULEROUNDUP(pstart + length) >> PAGE_SHIFT;
- cstart = pstart & ~(BANKSIZE - 1);
+ bdp = &bdata[node];
+ epfn = GRANULEROUNDUP(__pa(end)) >> PAGE_SHIFT;
+ cstart = GRANULEROUNDDOWN(__pa(start));
if (!bdp->node_low_pfn) {
bdp->node_boot_start = cstart;
@@ -99,34 +85,96 @@
return 0;
}
+
+/*
+ * Count the number of cpus on the node
+ */
+static __inline__ int
+count_cpus(int node)
+{
+ int cpu, n=0;
+
+ for (cpu=0; cpu < NR_CPUS; cpu++)
+ if (node = node_cpuid[cpu].nid)
+ n++;
+ return n;
+}
+
+
/*
- * Find space on each node for the bootmem map.
+ * Find space on each node for the bootmem map & other per-node data structures.
*
* Called by efi_memmap_walk to find boot memory on each node. Note that
* only blocks that are free are passed to this routine (currently filtered by
* free_available_memory).
*/
static int __init
-find_bootmap_space(unsigned long pstart, unsigned long length, int node)
+find_pernode_space(unsigned long start, unsigned long end, int node)
{
- unsigned long mapsize, pages, epfn;
+ unsigned long mapsize, pages, epfn, map=0, cpu, cpus;
+ unsigned long pernodesize=0, pernode;
+ unsigned long cpu_data;
+ unsigned long pstart, length;
bootmem_data_t *bdp;
+ pstart = __pa(start);
+ length = end - start;
epfn = (pstart + length) >> PAGE_SHIFT;
- bdp = &pg_data_ptr[node]->bdata[0];
+ bdp = &bdata[node];
if (pstart < bdp->node_boot_start || epfn > bdp->node_low_pfn)
return 0;
- if (!bdp->node_bootmem_map) {
+ if (!boot_pernode[node]) {
+ cpus = count_cpus(node);
+ pernodesize += PERCPU_PAGE_SIZE * cpus;
+ pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
+ pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+ pernodesize = PAGE_ALIGN(pernodesize);
+ pernode = NODEDATA_ALIGN(pstart, node);
+
+ if (pstart + length > (pernode + pernodesize)) {
+ boot_pernode[node] = pernode;
+ boot_pernodesize[node] = pernodesize;
+ memset(__va(pernode), 0, pernodesize);
+
+ cpu_data = pernode;
+ pernode += PERCPU_PAGE_SIZE * cpus;
+
+ pg_data_ptr[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ boot_node_data[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+
+ pg_data_ptr[node]->bdata = &bdata[node];
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ if (node = node_cpuid[cpu].nid) {
+ extern char __per_cpu_start[], __phys_per_cpu_start[];
+ memcpy((void*)cpu_data, __phys_per_cpu_start, __per_cpu_end - __per_cpu_start);
+ __per_cpu_offset[cpu] = (char*)__va(cpu_data) - __per_cpu_start;
+ cpu_data += PERCPU_PAGE_SIZE;
+ }
+ }
+ }
+ }
+
+ pernode = boot_pernode[node];
+ pernodesize = boot_pernodesize[node];
+ if (pernode && !bdp->node_bootmem_map) {
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- if (length > mapsize) {
- init_bootmem_node(
- BOOT_NODE_DATA(node),
- pstart>>PAGE_SHIFT,
- bdp->node_boot_start>>PAGE_SHIFT,
- bdp->node_low_pfn);
+
+ if (pernode - pstart > mapsize)
+ map = pstart;
+ else if (pstart + length - pernode - pernodesize > mapsize)
+ map = pernode + pernodesize;
+
+ if (map) {
+ init_bootmem_node(BOOT_NODE_DATA(node), map>>PAGE_SHIFT,
+ bdp->node_boot_start>>PAGE_SHIFT, bdp->node_low_pfn);
}
}
@@ -143,9 +191,9 @@
*
*/
static int __init
-discontig_free_bootmem_node(unsigned long pstart, unsigned long length, int node)
+discontig_free_bootmem_node(unsigned long start, unsigned long end, int node)
{
- free_bootmem_node(BOOT_NODE_DATA(node), pstart, length);
+ free_bootmem_node(BOOT_NODE_DATA(node), __pa(start), end - start);
return 0;
}
@@ -158,53 +206,50 @@
discontig_reserve_bootmem(void)
{
int node;
- unsigned long mapbase, mapsize, pages;
+ unsigned long base, size, pages;
bootmem_data_t *bdp;
for (node = 0; node < numnodes; node++) {
bdp = BOOT_NODE_DATA(node)->bdata;
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
- mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- mapbase = __pa(bdp->node_bootmem_map);
- reserve_bootmem_node(BOOT_NODE_DATA(node), mapbase, mapsize);
+ size = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
+ base = __pa(bdp->node_bootmem_map);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
+
+ size = boot_pernodesize[node];
+ base = __pa(boot_pernode[node]);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
}
}
/*
- * Allocate per node tables.
- * - the pg_data structure is allocated on each node. This minimizes offnode
- * memory references
- * - the node data is allocated & initialized. Portions of this structure is read-only (after
- * boot) and contains node-local pointers to usefuls data structures located on
- * other nodes.
+ * Initialize per-node data
+ *
+ * Finish setting up the node data for this node, then copy it to the other nodes.
*
- * We also switch to using the "real" pg_data structures at this point. Earlier in boot, we
- * use a different structure. The only use for pg_data prior to the point in boot is to get
- * the pointer to the bdata for the node.
*/
static void __init
-allocate_pernode_structures(void)
+initialize_pernode_data(void)
{
- pg_data_t *pgdat=0, *new_pgdat_list=0;
- int node, mynode;
+ int cpu, node;
+
+ memcpy(boot_node_data[0]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
+ memcpy(boot_node_data[0]->node_data_ptrs, boot_node_data, sizeof(boot_node_data));
- mynode = boot_get_local_nodeid();
- for (node = numnodes - 1; node >= 0 ; node--) {
- node_data[node] = alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof (struct ia64_node_data));
- pgdat = __alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof(pg_data_t), SMP_CACHE_BYTES, 0);
- pgdat->bdata = &(bdata[node][0]);
- pg_data_ptr[node] = pgdat;
- pgdat->pgdat_next = new_pgdat_list;
- new_pgdat_list = pgdat;
+ for (node=1; node < numnodes; node++) {
+ memcpy(boot_node_data[node], boot_node_data[0], sizeof(struct ia64_node_data));
+ boot_node_data[node]->node = node;
}
-
- memcpy(node_data[mynode]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
- memcpy(node_data[mynode]->node_data_ptrs, node_data, sizeof(node_data));
- pgdat_list = new_pgdat_list;
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ node = node_cpuid[cpu].nid;
+ per_cpu(cpu_info, cpu).node_data = boot_node_data[node];
+ per_cpu(cpu_info, cpu).nodeid = node;
+ }
}
+
/*
* Called early in boot to setup the boot memory allocator, and to
* allocate the node-local pg_data & node-directory data structures..
@@ -212,96 +257,19 @@
void __init
discontig_mem_init(void)
{
- int node;
-
if (numnodes = 0) {
printk(KERN_ERR "node info missing!\n");
numnodes = 1;
}
- for (node = 0; node < numnodes; node++) {
- pg_data_ptr[node] = (pg_data_t*) &boot_pg_data[node];
- pg_data_ptr[node]->bdata = &bdata[node][0];
- }
-
min_low_pfn = -1;
max_low_pfn = 0;
efi_memmap_walk(filter_rsvd_memory, build_maps);
- efi_memmap_walk(filter_rsvd_memory, find_bootmap_space);
+ efi_memmap_walk(filter_rsvd_memory, find_pernode_space);
efi_memmap_walk(filter_rsvd_memory, discontig_free_bootmem_node);
- discontig_reserve_bootmem();
- allocate_pernode_structures();
-}
-
-/*
- * Initialize the paging system.
- * - determine sizes of each node
- * - initialize the paging system for the node
- * - build the nodedir for the node. This contains pointers to
- * the per-bank mem_map entries.
- * - fix the page struct "virtual" pointers. These are bank specific
- * values that the paging system doesn't understand.
- * - replicate the nodedir structure to other nodes
- */
-
-void __init
-discontig_paging_init(void)
-{
- int node, mynode;
- unsigned long max_dma, zones_size[MAX_NR_ZONES];
- unsigned long kaddr, ekaddr, bid;
- struct page *page;
- bootmem_data_t *bdp;
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
- mynode = boot_get_local_nodeid();
- for (node = 0; node < numnodes; node++) {
- long pfn, startpfn;
-
- memset(zones_size, 0, sizeof(zones_size));
-
- startpfn = -1;
- bdp = BOOT_NODE_DATA(node)->bdata;
- pfn = bdp->node_boot_start >> PAGE_SHIFT;
- if (startpfn = -1)
- startpfn = pfn;
- if (pfn > max_dma)
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - pfn);
- else if (bdp->node_low_pfn < max_dma)
- zones_size[ZONE_DMA] += (bdp->node_low_pfn - pfn);
- else {
- zones_size[ZONE_DMA] += (max_dma - pfn);
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - max_dma);
- }
-
- free_area_init_node(node, NODE_DATA(node), NULL, zones_size, startpfn, 0);
-
- page = NODE_DATA(node)->node_mem_map;
-
- bdp = BOOT_NODE_DATA(node)->bdata;
-
- kaddr = (unsigned long)__va(bdp->node_boot_start);
- ekaddr = (unsigned long)__va(bdp->node_low_pfn << PAGE_SHIFT);
- while (kaddr < ekaddr) {
- if (paddr_to_nid(__pa(kaddr)) = node) {
- bid = BANK_MEM_MAP_INDEX(kaddr);
- node_data[mynode]->node_id_map[bid] = node;
- node_data[mynode]->bank_mem_map_base[bid] = page;
- }
- kaddr += BANKSIZE;
- page += BANKSIZE/PAGE_SIZE;
- }
- }
-
- /*
- * Finish setting up the node data for this node, then copy it to the other nodes.
- */
- for (node=0; node < numnodes; node++)
- if (mynode != node) {
- memcpy(node_data[node], node_data[mynode], sizeof(struct ia64_node_data));
- node_data[node]->node = node;
- }
+ discontig_reserve_bootmem();
+ initialize_pernode_data();
}
diff -Nru a/arch/ia64/mm/init.c b/arch/ia64/mm/init.c
--- a/arch/ia64/mm/init.c Wed Jul 16 12:50:02 2003
+++ b/arch/ia64/mm/init.c Wed Jul 16 12:50:02 2003
@@ -44,7 +44,7 @@
#ifdef CONFIG_VIRTUAL_MEM_MAP
# define LARGE_GAP 0x40000000 /* Use virtual mem map if hole is > than this */
unsigned long vmalloc_end = VMALLOC_END_INIT;
- static struct page *vmem_map;
+ struct page *vmem_map;
static unsigned long num_dma_physpages;
#endif
@@ -240,7 +240,7 @@
else if (page_count(pgdat->node_mem_map + i))
shared += page_count(pgdat->node_mem_map + i) - 1;
}
- printk("\t%d pages of RAM\n", pgdat->node_spanned_pages);
+ printk("\t%ld pages of RAM\n", pgdat->node_spanned_pages);
printk("\t%d reserved pages\n", reserved);
printk("\t%d pages shared\n", shared);
printk("\t%d pages swap cached\n", cached);
@@ -397,6 +397,7 @@
{
unsigned long address, start_page, end_page;
struct page *map_start, *map_end;
+ int node;
pgd_t *pgd;
pmd_t *pmd;
pte_t *pte;
@@ -406,19 +407,20 @@
start_page = (unsigned long) map_start & PAGE_MASK;
end_page = PAGE_ALIGN((unsigned long) map_end);
+ node = paddr_to_nid(__pa(start));
for (address = start_page; address < end_page; address += PAGE_SIZE) {
pgd = pgd_offset_k(address);
if (pgd_none(*pgd))
- pgd_populate(&init_mm, pgd, alloc_bootmem_pages(PAGE_SIZE));
+ pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pmd = pmd_offset(pgd, address);
if (pmd_none(*pmd))
- pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages(PAGE_SIZE));
+ pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pte = pte_offset_kernel(pmd, address);
if (pte_none(*pte))
- set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages(PAGE_SIZE)) >> PAGE_SHIFT,
+ set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
PAGE_KERNEL));
}
return 0;
@@ -431,6 +433,14 @@
unsigned long zone;
};
+struct memmap_count_callback_data {
+ int node;
+ unsigned long num_physpages;
+ unsigned long num_dma_physpages;
+ unsigned long min_pfn;
+ unsigned long max_pfn;
+} cdata;
+
static int
virtual_memmap_init (u64 start, u64 end, void *arg)
{
@@ -489,16 +499,6 @@
}
static int
-count_dma_pages (u64 start, u64 end, void *arg)
-{
- unsigned long *count = arg;
-
- if (end <= MAX_DMA_ADDRESS)
- *count += (end - start) >> PAGE_SHIFT;
- return 0;
-}
-
-static int
find_largest_hole (u64 start, u64 end, void *arg)
{
u64 *max_gap = arg;
@@ -514,102 +514,101 @@
}
#endif /* CONFIG_VIRTUAL_MEM_MAP */
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
+#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
+#define ORDERROUNDDOWN(n) ((n) & ~((PAGE_SIZE<<MAX_ORDER)-1))
static int
-count_pages (u64 start, u64 end, void *arg)
+count_pages (unsigned long start, unsigned long end, int node)
{
- unsigned long *count = arg;
+ start = __pa(start);
+ end = __pa(end);
- *count += (end - start) >> PAGE_SHIFT;
+ if (node = cdata.node) {
+ cdata.num_physpages += (end - start) >> PAGE_SHIFT;
+ if (start <= __pa(MAX_DMA_ADDRESS))
+ cdata.num_dma_physpages += (min(end, __pa(MAX_DMA_ADDRESS)) - start) >> PAGE_SHIFT;
+ start = GRANULEROUNDDOWN(__pa(start));
+ start = ORDERROUNDDOWN(start);
+ end = GRANULEROUNDUP(__pa(end));
+ cdata.max_pfn = max(cdata.max_pfn, end >> PAGE_SHIFT);
+ cdata.min_pfn = min(cdata.min_pfn, start >> PAGE_SHIFT);
+ }
return 0;
}
/*
* Set up the page tables.
*/
-
-#ifdef CONFIG_DISCONTIGMEM
void
paging_init (void)
{
- extern void discontig_paging_init(void);
-
- discontig_paging_init();
- efi_memmap_walk(count_pages, &num_physpages);
- zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
-}
-#else /* !CONFIG_DISCONTIGMEM */
-void
-paging_init (void)
-{
- unsigned long max_dma;
+ unsigned long max_dma_pfn;
unsigned long zones_size[MAX_NR_ZONES];
# ifdef CONFIG_VIRTUAL_MEM_MAP
unsigned long zholes_size[MAX_NR_ZONES];
unsigned long max_gap;
# endif
+ int node;
- /* initialize mem_map[] */
-
- memset(zones_size, 0, sizeof(zones_size));
-
- num_physpages = 0;
- efi_memmap_walk(count_pages, &num_physpages);
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
-
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- memset(zholes_size, 0, sizeof(zholes_size));
-
- num_dma_physpages = 0;
- efi_memmap_walk(count_dma_pages, &num_dma_physpages);
-
- if (max_low_pfn < max_dma) {
- zones_size[ZONE_DMA] = max_low_pfn;
- zholes_size[ZONE_DMA] = max_low_pfn - num_dma_physpages;
- } else {
- zones_size[ZONE_DMA] = max_dma;
- zholes_size[ZONE_DMA] = max_dma - num_dma_physpages;
- if (num_physpages > num_dma_physpages) {
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
- zholes_size[ZONE_NORMAL] = ((max_low_pfn - max_dma)
- - (num_physpages - num_dma_physpages));
- }
- }
-
+ max_dma_pfn = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
max_gap = 0;
efi_memmap_walk(find_largest_hole, (u64 *)&max_gap);
- if (max_gap < LARGE_GAP) {
- vmem_map = (struct page *) 0;
- free_area_init_node(0, &contig_page_data, NULL, zones_size, 0, zholes_size);
- mem_map = contig_page_data.node_mem_map;
- }
- else {
- unsigned long map_size;
-
- /* allocate virtual_mem_map */
- map_size = PAGE_ALIGN(max_low_pfn * sizeof(struct page));
- vmalloc_end -= map_size;
- vmem_map = (struct page *) vmalloc_end;
- efi_memmap_walk(create_mem_map_page_table, 0);
-
- free_area_init_node(0, &contig_page_data, vmem_map, zones_size, 0, zholes_size);
+ for (node = 0; node < numnodes; node++) {
+ memset(zones_size, 0, sizeof(zones_size));
+ memset(zholes_size, 0, sizeof(zholes_size));
+ memset(&cdata, 0, sizeof(cdata));
+
+ cdata.node = node;
+ cdata.min_pfn = ~0;
+
+ efi_memmap_walk(filter_rsvd_memory, count_pages);
+ num_dma_physpages += cdata.num_dma_physpages;
+ num_physpages += cdata.num_physpages;
+
+ if (cdata.min_pfn >= max_dma_pfn) {
+ /* Above the DMA zone */
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn - cdata.num_physpages;
+ } else if (cdata.max_pfn < max_dma_pfn) {
+ /* This block is DMAable */
+ zones_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn - cdata.num_dma_physpages;
+ } else {
+ zones_size[ZONE_DMA] = max_dma_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - cdata.num_dma_physpages;
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - max_dma_pfn;
+ zholes_size[ZONE_NORMAL] = zones_size[ZONE_NORMAL] - (cdata.num_physpages - cdata.num_dma_physpages);
+ }
- mem_map = contig_page_data.node_mem_map;
- printk("Virtual mem_map starts at 0x%p\n", mem_map);
- }
-# else /* !CONFIG_VIRTUAL_MEM_MAP */
- if (max_low_pfn < max_dma)
- zones_size[ZONE_DMA] = max_low_pfn;
- else {
- zones_size[ZONE_DMA] = max_dma;
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
+ if (numnodes = 1 && max_gap < LARGE_GAP) {
+ /* Just one node with no big holes... */
+ vmem_map = (struct page *)0;
+ zones_size[ZONE_DMA] += cdata.min_pfn;
+ zholes_size[ZONE_DMA] += cdata.min_pfn;
+ free_area_init_node(0, NODE_DATA(node), NODE_DATA(node)->node_mem_map,
+ zones_size, 0, zholes_size);
+ }
+ else {
+ /* allocate virtual mem_map */
+ if (node = 0) {
+ unsigned long map_size;
+ map_size = PAGE_ALIGN(max_low_pfn*sizeof(struct page));
+ vmalloc_end -= map_size;
+ vmem_map = (struct page *) vmalloc_end;
+ efi_memmap_walk(create_mem_map_page_table, 0);
+ printk("Virtual mem_map starts at 0x%p\n", vmem_map);
+#ifndef CONFIG_DISCONTIGMEM
+ mem_map = vmem_map;
+#endif
+ }
+ free_area_init_node(node, NODE_DATA(node), vmem_map + cdata.min_pfn,
+ zones_size, cdata.min_pfn, zholes_size);
+ }
}
- free_area_init(zones_size);
-# endif /* !CONFIG_VIRTUAL_MEM_MAP */
+
zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
}
-#endif /* !CONFIG_DISCONTIGMEM */
static int
count_reserved_pages (u64 start, u64 end, void *arg)
diff -Nru a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
--- a/drivers/acpi/Kconfig Wed Jul 16 12:50:02 2003
+++ b/drivers/acpi/Kconfig Wed Jul 16 12:50:02 2003
@@ -133,7 +133,7 @@
config ACPI_NUMA
bool "NUMA support" if NUMA && (IA64 && !IA64_HP_SIM || X86 && ACPI && !ACPI_HT_ONLY && !X86_64)
- default y if IA64 && IA64_SGI_SN
+ default y if IA64_GENERIC || IA64_SGI_SN2
config ACPI_ASUS
tristate "ASUS/Medion Laptop Extras"
diff -Nru a/include/asm-ia64/mmzone.h b/include/asm-ia64/mmzone.h
--- a/include/asm-ia64/mmzone.h Wed Jul 16 12:50:02 2003
+++ b/include/asm-ia64/mmzone.h Wed Jul 16 12:50:02 2003
@@ -3,7 +3,7 @@
* License. See the file "COPYING" in the main directory of this archive
* for more details.
*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000,2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2002 NEC Corp.
* Copyright (c) 2002 Erich Focht <efocht@ess.nec.de>
* Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com>
@@ -14,150 +14,50 @@
#include <linux/config.h>
#include <linux/init.h>
-/*
- * Given a kaddr, find the base mem_map address for the start of the mem_map
- * entries for the bank containing the kaddr.
- */
-#define BANK_MEM_MAP_BASE(kaddr) local_node_data->bank_mem_map_base[BANK_MEM_MAP_INDEX(kaddr)]
-
-/*
- * Given a kaddr, this macro return the relative map number
- * within the bank.
- */
-#define BANK_MAP_NR(kaddr) (BANK_OFFSET(kaddr) >> PAGE_SHIFT)
-/*
- * Given a pte, this macro returns a pointer to the page struct for the pte.
- */
-#define pte_page(pte) virt_to_page(PAGE_OFFSET | (pte_val(pte)&_PFN_MASK))
+#ifdef CONFIG_NUMA
-/*
- * Determine if a kaddr is a valid memory address of memory that
- * actually exists.
- *
- * The check consists of 2 parts:
- * - verify that the address is a region 7 address & does not
- * contain any bits that preclude it from being a valid platform
- * memory address
- * - verify that the chunk actually exists.
- *
- * Note that IO addresses are NOT considered valid addresses.
- *
- * Note, many platforms can simply check if kaddr exceeds a specific size.
- * (However, this won't work on SGI platforms since IO space is embedded
- * within the range of valid memory addresses & nodes have holes in the
- * address range between banks).
- */
-#define kern_addr_valid(kaddr) ({long _kav=(long)(kaddr); \
- VALID_MEM_KADDR(_kav);})
-
-/*
- * Given a kaddr, return a pointer to the page struct for the page.
- * If the kaddr does not represent RAM memory that potentially exists, return
- * a pointer the page struct for max_mapnr. IO addresses will
- * return the page for max_nr. Addresses in unpopulated RAM banks may
- * return undefined results OR may panic the system.
- *
- */
-#define virt_to_page(kaddr) ({long _kvtp=(long)(kaddr); \
- (VALID_MEM_KADDR(_kvtp)) \
- ? BANK_MEM_MAP_BASE(_kvtp) + BANK_MAP_NR(_kvtp) \
- : NULL;})
+#ifdef CONFIG_IA64_DIG
/*
- * Given a page struct entry, return the physical address that the page struct represents.
- * Since IA64 has all memory in the DMA zone, the following works:
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define page_to_phys(page) __pa(page_address(page))
-
-#define node_mem_map(nid) (NODE_DATA(nid)->node_mem_map)
+#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
+#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES * 32)
-#define node_localnr(pfn, nid) ((pfn) - NODE_DATA(nid)->node_start_pfn)
-#define pfn_to_page(pfn) (struct page *)(node_mem_map(pfn_to_nid(pfn)) + node_localnr(pfn, pfn_to_nid(pfn)))
-#define pfn_to_nid(pfn) local_node_data->node_id_map[(pfn << PAGE_SHIFT) >> BANKSHIFT]
-
-#define page_to_pfn(page) (long)((page - page_zone(page)->zone_mem_map) + page_zone(page)->zone_start_pfn)
+#elif CONFIG_IA64_SGI_SN2
/*
- * pfn_valid should be made as fast as possible, and the current definition
- * is valid for machines that are NUMA, but still contiguous, which is what
- * is currently supported. A more generalised, but slower definition would
- * be something like this - mbligh:
- * ( pfn_to_pgdat(pfn) && (pfn < node_end_pfn(pfn_to_nid(pfn))) )
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define pfn_valid(pfn) (pfn < max_low_pfn)
-extern unsigned long max_low_pfn;
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of compute nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
+#elif CONFIG_IA64_GENERIC
-#ifdef CONFIG_IA64_DIG
/*
- * Platform definitions for DIG platform with contiguous memory.
+ * Platform definitions for GENERIC platform with contiguous or discontiguous memory.
*/
-#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
-#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
-#define MAX_PHYS_MEMORY (1UL << 40) /* 1 TB */
-/*
- * Bank definitions.
- * Configurable settings for DIG: 512MB/bank: 16GB/node,
- * 2048MB/bank: 64GB/node,
- * 8192MB/bank: 256GB/node.
- */
-#define NR_BANKS_PER_NODE 32
-#if defined(CONFIG_IA64_NODESIZE_16GB)
-# define BANKSHIFT 29
-#elif defined(CONFIG_IA64_NODESIZE_64GB)
-# define BANKSHIFT 31
-#elif defined(CONFIG_IA64_NODESIZE_256GB)
-# define BANKSHIFT 33
#else
-# error Unsupported bank and nodesize!
+#error unknown platform
#endif
-#define BANKSIZE (1UL << BANKSHIFT)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-/*
- * VALID_MEM_KADDR returns a boolean to indicate if a kaddr is
- * potentially a valid cacheable identity mapped RAM memory address.
- * Note that the RAM may or may not actually be present!!
- */
-#define VALID_MEM_KADDR(kaddr) 1
+extern void build_cpu_to_node_map(void);
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#else /* CONFIG_NUMA */
-#elif defined(CONFIG_IA64_SGI_SN2)
-/*
- * SGI SN2 discontig definitions
- */
-#define MAX_PHYSNODE_ID 2048 /* 2048 node ids (also called nasid) */
-#define NR_NODES 128 /* Maximum number of nodes in SSI */
-#define MAX_PHYS_MEMORY (1UL << 49)
-
-#define BANKSHIFT 38
-#define NR_BANKS_PER_NODE 4
-#define SN2_NODE_SIZE (64UL*1024*1024*1024) /* 64GB per node */
-#define BANKSIZE (SN2_NODE_SIZE/NR_BANKS_PER_NODE)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-#define VALID_MEM_KADDR(kaddr) 1
-
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#define NR_NODES 1
-#endif /* CONFIG_IA64_DIG */
+#endif /* CONFIG_NUMA */
#endif /* _ASM_IA64_MMZONE_H */
diff -Nru a/include/asm-ia64/nodedata.h b/include/asm-ia64/nodedata.h
--- a/include/asm-ia64/nodedata.h Wed Jul 16 12:50:02 2003
+++ b/include/asm-ia64/nodedata.h Wed Jul 16 12:50:02 2003
@@ -13,7 +13,7 @@
#ifndef _ASM_IA64_NODEDATA_H
#define _ASM_IA64_NODEDATA_H
-
+#include <asm/percpu.h>
#include <asm/mmzone.h>
/*
@@ -22,15 +22,17 @@
struct pglist_data;
struct ia64_node_data {
- short active_cpu_count;
short node;
+ short active_cpu_count;
+ /*
+ * The fields are read-only (after boot). They contain pointers
+ * to various structures located on other nodes. Ths data is
+ * replicated on each node in order to reduce off-node references.
+ */
struct pglist_data *pg_data_ptrs[NR_NODES];
- struct page *bank_mem_map_base[NR_BANKS];
struct ia64_node_data *node_data_ptrs[NR_NODES];
- short node_id_map[NR_BANKS];
};
-
/*
* Return a pointer to the node_data structure for the executing cpu.
*/
@@ -40,7 +42,8 @@
/*
* Return a pointer to the node_data structure for the specified node.
*/
-#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
/*
* Get a pointer to the node_id/node_data for the current cpu.
@@ -48,29 +51,5 @@
*/
extern int boot_get_local_nodeid(void);
extern struct ia64_node_data *get_node_data_ptr(void);
-
-/*
- * Given a node id, return a pointer to the pg_data_t for the node.
- * The following 2 macros are similar.
- *
- * NODE_DATA - should be used in all code not related to system
- * initialization. It uses pernode data structures to minimize
- * offnode memory references. However, these structure are not
- * present during boot. This macro can be used once cpu_init
- * completes.
- *
- * BOOT_NODE_DATA
- * - should be used during system initialization
- * prior to freeing __initdata. It does not depend on the percpu
- * area being present.
- *
- * NOTE: The names of these macros are misleading but are difficult to change
- * since they are used in generic linux & on other architecures.
- */
-#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
-#define BOOT_NODE_DATA(nid) boot_get_pg_data_ptr((long)(nid))
-
-struct pglist_data;
-extern struct pglist_data * __init boot_get_pg_data_ptr(long);
#endif /* _ASM_IA64_NODEDATA_H */
diff -Nru a/include/asm-ia64/numa.h b/include/asm-ia64/numa.h
--- a/include/asm-ia64/numa.h Wed Jul 16 12:50:02 2003
+++ b/include/asm-ia64/numa.h Wed Jul 16 12:50:02 2003
@@ -15,13 +15,21 @@
#ifdef CONFIG_DISCONTIGMEM
# include <asm/mmzone.h>
-# define NR_MEMBLKS (NR_BANKS)
#else
# define NR_NODES (8)
# define NR_MEMBLKS (NR_NODES * 8)
#endif
#include <linux/cache.h>
+#include <linux/threads.h>
+#include <linux/smp.h>
+
+#define NODEMASK_WORDCOUNT ((NR_NODES+(BITS_PER_LONG-1))/BITS_PER_LONG)
+
+#define NODE_MASK_NONE { [0 ... ((NR_NODES+BITS_PER_LONG-1)/BITS_PER_LONG)-1] = 0 }
+
+typedef unsigned long nodemask_t[NODEMASK_WORDCOUNT];
+
extern volatile char cpu_to_node_map[NR_CPUS] __cacheline_aligned;
extern volatile unsigned long node_to_cpu_mask[NR_NODES] __cacheline_aligned;
@@ -63,6 +71,12 @@
extern int paddr_to_nid(unsigned long paddr);
#define local_nodeid (cpu_to_node_map[smp_processor_id()])
+
+#else /* !CONFIG_NUMA */
+
+#define node_distance(from,to) 10
+#define paddr_to_nid(x) 0
+#define local_nodeid 0
#endif /* CONFIG_NUMA */
diff -Nru a/include/asm-ia64/page.h b/include/asm-ia64/page.h
--- a/include/asm-ia64/page.h Wed Jul 16 12:50:02 2003
+++ b/include/asm-ia64/page.h Wed Jul 16 12:50:02 2003
@@ -93,18 +93,26 @@
#define virt_addr_valid(kaddr) pfn_valid(__pa(kaddr) >> PAGE_SHIFT)
-#ifndef CONFIG_DISCONTIGMEM
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- extern int ia64_pfn_valid (unsigned long pfn);
-# define pfn_valid(pfn) (((pfn) < max_mapnr) && ia64_pfn_valid(pfn))
-# else
-# define pfn_valid(pfn) ((pfn) < max_mapnr)
-# endif
-#define virt_to_page(kaddr) pfn_to_page(__pa(kaddr) >> PAGE_SHIFT)
-#define page_to_pfn(page) ((unsigned long) (page - mem_map))
-#define pfn_to_page(pfn) (mem_map + (pfn))
-#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
+#ifdef CONFIG_VIRTUAL_MEM_MAP
+extern int ia64_pfn_valid(unsigned long pfn);
+#else
+#define ia64_pfn_valid(pfn) (1)
+#endif
+
+extern unsigned long max_low_pfn;
+#define pfn_valid(pfn) (((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
+
+#if defined(CONFIG_VIRTUAL_MEM_MAP) && !defined(CONFIG_DISCONTIGMEM)
+#define vmem_map mem_map
+#else
+extern struct page *vmem_map;
#endif
+
+#define pfn_to_page(pfn) (vmem_map + (pfn))
+#define page_to_pfn(page) ((unsigned long) (page - vmem_map))
+
+#define virt_to_page(kaddr) (pfn_to_page(__pa(kaddr) >> PAGE_SHIFT))
+#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
typedef union ia64_va {
struct {
diff -Nru a/include/asm-ia64/pgtable.h b/include/asm-ia64/pgtable.h
--- a/include/asm-ia64/pgtable.h Wed Jul 16 12:50:02 2003
+++ b/include/asm-ia64/pgtable.h Wed Jul 16 12:50:02 2003
@@ -174,7 +174,6 @@
return (addr & (local_cpu_data->unimpl_pa_mask)) = 0;
}
-#ifndef CONFIG_DISCONTIGMEM
/*
* kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
* memory. For the return value to be meaningful, ADDR must be >@@ -190,7 +189,6 @@
*/
#define kern_addr_valid(addr) (1)
-#endif
/*
* Now come the defines and routines to manage and access the three-level
@@ -241,10 +239,8 @@
#define pte_none(pte) (!pte_val(pte))
#define pte_present(pte) (pte_val(pte) & (_PAGE_P | _PAGE_PROTNONE))
#define pte_clear(pte) (pte_val(*(pte)) = 0UL)
-#ifndef CONFIG_DISCONTIGMEM
/* pte_page() returns the "struct page *" corresponding to the PTE: */
#define pte_page(pte) virt_to_page(((pte_val(pte) & _PFN_MASK) + PAGE_OFFSET))
-#endif
#define pmd_none(pmd) (!pmd_val(pmd))
#define pmd_bad(pmd) (!ia64_phys_addr_valid(pmd_val(pmd)))
@@ -416,6 +412,7 @@
extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
extern void paging_init (void);
+extern int filter_rsvd_memory(unsigned long start, unsigned long end, void *arg);
/*
* Note: The macros below rely on the fact that MAX_SWAPFILES_SHIFT <= number of
diff -Nru a/include/asm-ia64/processor.h b/include/asm-ia64/processor.h
--- a/include/asm-ia64/processor.h Wed Jul 16 12:50:02 2003
+++ b/include/asm-ia64/processor.h Wed Jul 16 12:50:02 2003
@@ -185,6 +185,8 @@
#endif
#ifdef CONFIG_NUMA
struct ia64_node_data *node_data;
+ struct cpuinfo_ia64 *cpu_data[NR_CPUS];
+ int nodeid;
#endif
};
diff -Nru a/mm/bootmem.c b/mm/bootmem.c
--- a/mm/bootmem.c Wed Jul 16 12:50:02 2003
+++ b/mm/bootmem.c Wed Jul 16 12:50:02 2003
@@ -48,8 +48,24 @@
bootmem_data_t *bdata = pgdat->bdata;
unsigned long mapsize = ((end - start)+7)/8;
- pgdat->pgdat_next = pgdat_list;
- pgdat_list = pgdat;
+
+ /*
+ * sort pgdat_list so that the lowest one comes first,
+ * which makes alloc_bootmem_low_pages work as desired.
+ */
+ if (!pgdat_list || pgdat_list->node_start_pfn > pgdat->node_start_pfn) {
+ pgdat->pgdat_next = pgdat_list;
+ pgdat_list = pgdat;
+ } else {
+ pg_data_t *tmp = pgdat_list;
+ while (tmp->pgdat_next) {
+ if (tmp->pgdat_next->node_start_pfn > pgdat->node_start_pfn)
+ break;
+ tmp = tmp->pgdat_next;
+ }
+ pgdat->pgdat_next = tmp->pgdat_next;
+ tmp->pgdat_next = pgdat;
+ }
mapsize = (mapsize + (sizeof(long) - 1UL)) & ~(sizeof(long) - 1UL);
bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
@@ -251,7 +267,7 @@
static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
{
- struct page *page = pgdat->node_mem_map;
+ struct page *page;
bootmem_data_t *bdata = pgdat->bdata;
unsigned long i, count, total = 0;
unsigned long idx;
@@ -260,23 +276,23 @@
if (!bdata->node_bootmem_map) BUG();
count = 0;
+ page = virt_to_page(phys_to_virt(bdata->node_boot_start));
idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
map = bdata->node_bootmem_map;
for (i = 0; i < idx; ) {
unsigned long v = ~map[i / BITS_PER_LONG];
if (v) {
unsigned long m;
- for (m = 1; m && i < idx; m<<=1, page++, i++) {
+ for (m = 1; m && i < idx; m<<=1, i++) {
if (v & m) {
count++;
- ClearPageReserved(page);
- set_page_count(page, 1);
- __free_page(page);
+ ClearPageReserved(page+i);
+ set_page_count(page+i, 1);
+ __free_page(page+i);
}
}
} else {
i+=BITS_PER_LONG;
- page += BITS_PER_LONG;
}
}
total += count;
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (7 preceding siblings ...)
2003-07-16 19:51 ` Jesse Barnes
@ 2003-07-16 19:56 ` Erich Focht
2003-07-16 22:37 ` Jesse Barnes
` (4 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Erich Focht @ 2003-07-16 19:56 UTC (permalink / raw)
To: linux-ia64
On Wednesday 16 July 2003 21:40, Matthew Wilcox wrote:
> On Wed, Jul 16, 2003 at 12:29:52PM -0700, Jesse Barnes wrote:
> > @@ -210,8 +210,8 @@
> > system with an A0 or A1 stepping CPU.
> >
> > config NUMA
> > - bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
> > - default y if IA64_SGI_SN2
> > + bool
> > + default y if IA64_SGI_SN2 || IA64_GENERIC
> > help
> > Say Y to compile the kernel to support NUMA (Non-Uniform Memory
> > Access). This option is for configuring high-end multiprocessor
There are other NUMA machines around which are not from SGI and are
IA64_DIG. You excluded them on purpose? I hope not...
> If you're removing the question, you can remove the helptext too.
>
> > @@ -235,8 +235,7 @@
> >
> > config DISCONTIGMEM
> > bool
> > - depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) &&
> > NUMA - default y
> > + default y if IA64_SGI_SN2 || IA64_GENERIC
> > help
> > Say Y to support efficient handling of discontiguous physical memory,
> > for architectures which are either NUMA (Non-Uniform Memory Access)
Same comment as above. NEC TX7 is an IA64_DIG platform.
Erich
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (8 preceding siblings ...)
2003-07-16 19:56 ` Erich Focht
@ 2003-07-16 22:37 ` Jesse Barnes
2003-07-17 8:23 ` Erich Focht
` (3 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Jesse Barnes @ 2003-07-16 22:37 UTC (permalink / raw)
To: linux-ia64
On Wed, Jul 16, 2003 at 09:56:32PM +0200, Erich Focht wrote:
> On Wednesday 16 July 2003 21:40, Matthew Wilcox wrote:
> > On Wed, Jul 16, 2003 at 12:29:52PM -0700, Jesse Barnes wrote:
> > > @@ -210,8 +210,8 @@
> > > system with an A0 or A1 stepping CPU.
> > >
> > > config NUMA
> > > - bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
> > > - default y if IA64_SGI_SN2
> > > + bool
> > > + default y if IA64_SGI_SN2 || IA64_GENERIC
> > > help
> > > Say Y to compile the kernel to support NUMA (Non-Uniform Memory
> > > Access). This option is for configuring high-end multiprocessor
>
> There are other NUMA machines around which are not from SGI and are
> IA64_DIG. You excluded them on purpose? I hope not...
>
> > If you're removing the question, you can remove the helptext too.
> >
> > > @@ -235,8 +235,7 @@
> > >
> > > config DISCONTIGMEM
> > > bool
> > > - depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) &&
> > > NUMA - default y
> > > + default y if IA64_SGI_SN2 || IA64_GENERIC
> > > help
> > > Say Y to support efficient handling of discontiguous physical memory,
> > > for architectures which are either NUMA (Non-Uniform Memory Access)
>
> Same comment as above. NEC TX7 is an IA64_DIG platform.
Oops, sorry. Here's another one. How does it look?
Thanks,
Jesse
# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
# ChangeSet 1.1397 -> 1.1398
# include/asm-ia64/page.h 1.19 -> 1.20
# arch/ia64/kernel/setup.c 1.53 -> 1.54
# include/asm-ia64/pgtable.h 1.28 -> 1.29
# mm/bootmem.c 1.18 -> 1.19
# include/asm-ia64/numa.h 1.5 -> 1.6
# include/asm-ia64/processor.h 1.48 -> 1.49
# arch/ia64/mm/init.c 1.46 -> 1.47
# include/asm-ia64/nodedata.h 1.3 -> 1.4
# arch/ia64/mm/discontig.c 1.4 -> 1.5
# arch/ia64/Kconfig 1.38 -> 1.39
# drivers/acpi/Kconfig 1.12 -> 1.13
# include/asm-ia64/mmzone.h 1.4 -> 1.5
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 03/07/16 jbarnes@tomahawk.engr.sgi.com 1.1398
# latest discontig
# --------------------------------------------
#
diff -Nru a/arch/ia64/Kconfig b/arch/ia64/Kconfig
--- a/arch/ia64/Kconfig Wed Jul 16 15:36:33 2003
+++ b/arch/ia64/Kconfig Wed Jul 16 15:36:33 2003
@@ -211,7 +211,7 @@
config NUMA
bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
+ default y if IA64_SGI_SN2 || IA64_GENERIC
help
Say Y to compile the kernel to support NUMA (Non-Uniform Memory
Access). This option is for configuring high-end multiprocessor
@@ -234,9 +234,8 @@
endchoice
config DISCONTIGMEM
- bool
- depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) && NUMA
- default y
+ bool "Discontiguous memory support" if IA64_DIG
+ default y if IA64_SGI_SN2 || IA64_GENERIC
help
Say Y to support efficient handling of discontiguous physical memory,
for architectures which are either NUMA (Non-Uniform Memory Access)
@@ -245,8 +244,7 @@
config VIRTUAL_MEM_MAP
bool "Enable Virtual Mem Map"
- depends on !NUMA
- default y if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
+ default y if !IA64_HP_SIM
help
Say Y to compile the kernel with support for a virtual mem map.
This is an alternate method of supporting large holes in the
@@ -259,8 +257,8 @@
are unsure, say Y.
config IA64_MCA
- bool "Enable IA-64 Machine Check Abort" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
+ bool "Enable IA-64 Machine Check Abort"
+ default y if !IA64_HP_SIM
help
Say Y here to enable machine check support for IA-64. If you're
unsure, answer Y.
@@ -292,43 +290,12 @@
depends on IA64_GENERIC || IA64_DIG || IA64_HP_ZX1 || IA64_SGI_SN2
default y
-config IA64_SGI_SN_DEBUG
- bool "Enable extra debugging code"
- depends on IA64_SGI_SN2
- help
- Turns on extra debugging code in the SGI SN (Scalable NUMA) platform
- for IA-64. Unless you are debugging problems on an SGI SN IA-64 box,
- say N.
-
config IA64_SGI_SN_SIM
bool "Enable SGI Medusa Simulator Support"
depends on IA64_SGI_SN2
help
If you are compiling a kernel that will run under SGI's IA-64
simulator (Medusa) then say Y, otherwise say N.
-
-config IA64_SGI_AUTOTEST
- bool "Enable autotest (llsc). Option to run cache test instead of booting"
- depends on IA64_SGI_SN2
- help
- Build a kernel used for hardware validation. If you include the
- keyword "autotest" on the boot command line, the kernel does NOT boot.
- Instead, it starts all cpus and runs cache coherency tests instead.
-
- If unsure, say N.
-
-config SERIAL_SGI_L1_PROTOCOL
- bool "Enable protocol mode for the L1 console"
- depends on IA64_SGI_SN2
- help
- Uses protocol mode instead of raw mode for the level 1 console on the
- SGI SN (Scalable NUMA) platform for IA-64. If you are compiling for
- an SGI SN box then Y is the recommended value, otherwise say N.
-
-config PERCPU_IRQ
- bool
- depends on IA64_SGI_SN2
- default y
# On IA-64, we always want an ELF /proc/kcore.
config KCORE_ELF
diff -Nru a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c
--- a/arch/ia64/kernel/setup.c Wed Jul 16 15:36:33 2003
+++ b/arch/ia64/kernel/setup.c Wed Jul 16 15:36:33 2003
@@ -138,7 +138,7 @@
call_pernode_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long rs, re;
- void (*func)(unsigned long, unsigned long, int, int);
+ void (*func)(unsigned long, unsigned long, int);
int i;
start = PAGE_ALIGN(start);
@@ -149,22 +149,21 @@
func = arg;
if (!num_memblks) {
- /*
- * This machine doesn't have SRAT, so call func with
- * nid=0, bank=0.
- */
+ /* No SRAT table, to assume one node (node 0) */
if (start < end)
- (*func)(start, end - start, 0, 0);
+ (*func)(start, end, 0);
return;
}
for (i = 0; i < num_memblks; i++) {
- rs = max(start, node_memblk[i].start_paddr);
- re = min(end, node_memblk[i].start_paddr+node_memblk[i].size);
+ rs = max(__pa(start), node_memblk[i].start_paddr);
+ re = min(__pa(end), node_memblk[i].start_paddr+node_memblk[i].size);
if (rs < re)
- (*func)(rs, re-rs, node_memblk[i].nid,
- node_memblk[i].bank);
+ (*func)((unsigned long)__va(rs), (unsigned long)__va(re), node_memblk[i].nid);
+
+ if ((unsigned long)__va(re) = end)
+ break;
}
}
@@ -180,7 +179,7 @@
filter_rsvd_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long range_start, range_end, prev_start;
- void (*func)(unsigned long, unsigned long);
+ void (*func)(unsigned long, unsigned long, int);
int i;
#if IGNORE_PFN0
@@ -202,9 +201,9 @@
if (range_start < range_end)
#ifdef CONFIG_DISCONTIGMEM
- call_pernode_memory(__pa(range_start), __pa(range_end), func);
+ call_pernode_memory(range_start, range_end, func);
#else
- (*func)(__pa(range_start), range_end - range_start);
+ (*func)(range_start, range_end, 0);
#endif
/* nothing more available in this segment */
@@ -703,6 +702,8 @@
* get_free_pages() cannot be used before cpu_init() done. BSP allocates
* "NR_CPUS" pages for all CPUs to avoid that AP calls get_zeroed_page().
*/
+#ifndef CONFIG_DISCONTIGMEM
+ /* for discontig machines, we do this in discontig.c */
if (smp_processor_id() = 0) {
cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * NR_CPUS, PERCPU_PAGE_SIZE,
__pa(MAX_DMA_ADDRESS));
@@ -714,6 +715,7 @@
per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
}
}
+#endif
cpu_data = __per_cpu_start + __per_cpu_offset[smp_processor_id()];
#else /* !CONFIG_SMP */
cpu_data = __phys_per_cpu_start;
diff -Nru a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c
--- a/arch/ia64/mm/discontig.c Wed Jul 16 15:36:33 2003
+++ b/arch/ia64/mm/discontig.c Wed Jul 16 15:36:33 2003
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000, 2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2001 Intel Corp.
* Copyright (c) 2001 Tony Luck <tony.luck@intel.com>
* Copyright (c) 2002 NEC Corp.
@@ -16,74 +16,60 @@
#include <linux/mmzone.h>
#include <linux/acpi.h>
#include <linux/efi.h>
-
+#include <asm/pgalloc.h>
+#include <asm/tlb.h>
/*
- * Round an address upward to the next multiple of GRANULE size.
+ * Round an address upward or downward to the next multiple of IA64_GRANULE_SIZE.
*/
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
-static struct ia64_node_data *node_data[NR_NODES];
-static long boot_pg_data[8*NR_NODES+sizeof(pg_data_t)] __initdata;
-static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
-static bootmem_data_t bdata[NR_NODES][NR_BANKS_PER_NODE+1] __initdata;
-
-extern int filter_rsvd_memory (unsigned long start, unsigned long end, void *arg);
+/*
+ * Used to locate BOOT_DATA prior to initializing the node data area.
+ */
+#define BOOT_NODE_DATA(node) pg_data_ptr[node]
/*
- * Return the compact node number of this cpu. Used prior to
- * setting up the cpu_data area.
- * Note - not fast, intended for boot use only!!
+ * To prevent cache aliasing effects, align per-node structures so that they
+ * start at addresses that are strided by node number.
*/
-int
-boot_get_local_nodeid(void)
-{
- int i;
+#define NODEDATA_ALIGN(addr, node) ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + (node)*PERCPU_PAGE_SIZE)
- for (i = 0; i < NR_CPUS; i++)
- if (node_cpuid[i].phys_id = hard_smp_processor_id())
- return node_cpuid[i].nid;
- /* node info missing, so nid should be 0.. */
- return 0;
-}
+static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
+static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
+static bootmem_data_t bdata[NR_NODES] __initdata;
+static unsigned long boot_pernode[NR_NODES] __initdata;
+static unsigned long boot_pernodesize[NR_NODES] __initdata;
-/*
- * Return a pointer to the pg_data structure for a node.
- * This function is used ONLY in early boot before the cpu_data
- * structure is available.
- */
-pg_data_t* __init
-boot_get_pg_data_ptr(long node)
-{
- return pg_data_ptr[node];
-}
+extern char __per_cpu_start[], __per_cpu_end[];
-/*
- * Return a pointer to the node data for the current node.
- * (boottime initialization only)
- */
-struct ia64_node_data *
+struct ia64_node_data*
get_node_data_ptr(void)
{
- return node_data[boot_get_local_nodeid()];
+ return boot_node_data[cpu_to_node_map[smp_processor_id()]]; /* ZZZ */
}
/*
* We allocate one of the bootmem_data_t structs for each piece of memory
* that we wish to treat as a contiguous block. Each such block must start
- * on a BANKSIZE boundary. Multiple banks per node is not supported.
+ * on a GRANULE boundary. Multiple banks per node are not supported.
+ * (Note: on SN2, all memory on a node is trated as a single bank.
+ * Holes within the bank are supported. This works because memory
+ * from different banks is not interleaved. The bootmap bitmap
+ * for the node is somewhat large but not too large).
*/
static int __init
-build_maps(unsigned long pstart, unsigned long length, int node)
+build_maps(unsigned long start, unsigned long end, int node)
{
bootmem_data_t *bdp;
unsigned long cstart, epfn;
- bdp = pg_data_ptr[node]->bdata;
- epfn = GRANULEROUNDUP(pstart + length) >> PAGE_SHIFT;
- cstart = pstart & ~(BANKSIZE - 1);
+ bdp = &bdata[node];
+ epfn = GRANULEROUNDUP(__pa(end)) >> PAGE_SHIFT;
+ cstart = GRANULEROUNDDOWN(__pa(start));
if (!bdp->node_low_pfn) {
bdp->node_boot_start = cstart;
@@ -99,34 +85,96 @@
return 0;
}
+
+/*
+ * Count the number of cpus on the node
+ */
+static __inline__ int
+count_cpus(int node)
+{
+ int cpu, n=0;
+
+ for (cpu=0; cpu < NR_CPUS; cpu++)
+ if (node = node_cpuid[cpu].nid)
+ n++;
+ return n;
+}
+
+
/*
- * Find space on each node for the bootmem map.
+ * Find space on each node for the bootmem map & other per-node data structures.
*
* Called by efi_memmap_walk to find boot memory on each node. Note that
* only blocks that are free are passed to this routine (currently filtered by
* free_available_memory).
*/
static int __init
-find_bootmap_space(unsigned long pstart, unsigned long length, int node)
+find_pernode_space(unsigned long start, unsigned long end, int node)
{
- unsigned long mapsize, pages, epfn;
+ unsigned long mapsize, pages, epfn, map=0, cpu, cpus;
+ unsigned long pernodesize=0, pernode;
+ unsigned long cpu_data;
+ unsigned long pstart, length;
bootmem_data_t *bdp;
+ pstart = __pa(start);
+ length = end - start;
epfn = (pstart + length) >> PAGE_SHIFT;
- bdp = &pg_data_ptr[node]->bdata[0];
+ bdp = &bdata[node];
if (pstart < bdp->node_boot_start || epfn > bdp->node_low_pfn)
return 0;
- if (!bdp->node_bootmem_map) {
+ if (!boot_pernode[node]) {
+ cpus = count_cpus(node);
+ pernodesize += PERCPU_PAGE_SIZE * cpus;
+ pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
+ pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+ pernodesize = PAGE_ALIGN(pernodesize);
+ pernode = NODEDATA_ALIGN(pstart, node);
+
+ if (pstart + length > (pernode + pernodesize)) {
+ boot_pernode[node] = pernode;
+ boot_pernodesize[node] = pernodesize;
+ memset(__va(pernode), 0, pernodesize);
+
+ cpu_data = pernode;
+ pernode += PERCPU_PAGE_SIZE * cpus;
+
+ pg_data_ptr[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ boot_node_data[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+
+ pg_data_ptr[node]->bdata = &bdata[node];
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ if (node = node_cpuid[cpu].nid) {
+ extern char __per_cpu_start[], __phys_per_cpu_start[];
+ memcpy((void*)cpu_data, __phys_per_cpu_start, __per_cpu_end - __per_cpu_start);
+ __per_cpu_offset[cpu] = (char*)__va(cpu_data) - __per_cpu_start;
+ cpu_data += PERCPU_PAGE_SIZE;
+ }
+ }
+ }
+ }
+
+ pernode = boot_pernode[node];
+ pernodesize = boot_pernodesize[node];
+ if (pernode && !bdp->node_bootmem_map) {
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- if (length > mapsize) {
- init_bootmem_node(
- BOOT_NODE_DATA(node),
- pstart>>PAGE_SHIFT,
- bdp->node_boot_start>>PAGE_SHIFT,
- bdp->node_low_pfn);
+
+ if (pernode - pstart > mapsize)
+ map = pstart;
+ else if (pstart + length - pernode - pernodesize > mapsize)
+ map = pernode + pernodesize;
+
+ if (map) {
+ init_bootmem_node(BOOT_NODE_DATA(node), map>>PAGE_SHIFT,
+ bdp->node_boot_start>>PAGE_SHIFT, bdp->node_low_pfn);
}
}
@@ -143,9 +191,9 @@
*
*/
static int __init
-discontig_free_bootmem_node(unsigned long pstart, unsigned long length, int node)
+discontig_free_bootmem_node(unsigned long start, unsigned long end, int node)
{
- free_bootmem_node(BOOT_NODE_DATA(node), pstart, length);
+ free_bootmem_node(BOOT_NODE_DATA(node), __pa(start), end - start);
return 0;
}
@@ -158,53 +206,50 @@
discontig_reserve_bootmem(void)
{
int node;
- unsigned long mapbase, mapsize, pages;
+ unsigned long base, size, pages;
bootmem_data_t *bdp;
for (node = 0; node < numnodes; node++) {
bdp = BOOT_NODE_DATA(node)->bdata;
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
- mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- mapbase = __pa(bdp->node_bootmem_map);
- reserve_bootmem_node(BOOT_NODE_DATA(node), mapbase, mapsize);
+ size = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
+ base = __pa(bdp->node_bootmem_map);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
+
+ size = boot_pernodesize[node];
+ base = __pa(boot_pernode[node]);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
}
}
/*
- * Allocate per node tables.
- * - the pg_data structure is allocated on each node. This minimizes offnode
- * memory references
- * - the node data is allocated & initialized. Portions of this structure is read-only (after
- * boot) and contains node-local pointers to usefuls data structures located on
- * other nodes.
+ * Initialize per-node data
+ *
+ * Finish setting up the node data for this node, then copy it to the other nodes.
*
- * We also switch to using the "real" pg_data structures at this point. Earlier in boot, we
- * use a different structure. The only use for pg_data prior to the point in boot is to get
- * the pointer to the bdata for the node.
*/
static void __init
-allocate_pernode_structures(void)
+initialize_pernode_data(void)
{
- pg_data_t *pgdat=0, *new_pgdat_list=0;
- int node, mynode;
+ int cpu, node;
+
+ memcpy(boot_node_data[0]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
+ memcpy(boot_node_data[0]->node_data_ptrs, boot_node_data, sizeof(boot_node_data));
- mynode = boot_get_local_nodeid();
- for (node = numnodes - 1; node >= 0 ; node--) {
- node_data[node] = alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof (struct ia64_node_data));
- pgdat = __alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof(pg_data_t), SMP_CACHE_BYTES, 0);
- pgdat->bdata = &(bdata[node][0]);
- pg_data_ptr[node] = pgdat;
- pgdat->pgdat_next = new_pgdat_list;
- new_pgdat_list = pgdat;
+ for (node=1; node < numnodes; node++) {
+ memcpy(boot_node_data[node], boot_node_data[0], sizeof(struct ia64_node_data));
+ boot_node_data[node]->node = node;
}
-
- memcpy(node_data[mynode]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
- memcpy(node_data[mynode]->node_data_ptrs, node_data, sizeof(node_data));
- pgdat_list = new_pgdat_list;
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ node = node_cpuid[cpu].nid;
+ per_cpu(cpu_info, cpu).node_data = boot_node_data[node];
+ per_cpu(cpu_info, cpu).nodeid = node;
+ }
}
+
/*
* Called early in boot to setup the boot memory allocator, and to
* allocate the node-local pg_data & node-directory data structures..
@@ -212,96 +257,19 @@
void __init
discontig_mem_init(void)
{
- int node;
-
if (numnodes = 0) {
printk(KERN_ERR "node info missing!\n");
numnodes = 1;
}
- for (node = 0; node < numnodes; node++) {
- pg_data_ptr[node] = (pg_data_t*) &boot_pg_data[node];
- pg_data_ptr[node]->bdata = &bdata[node][0];
- }
-
min_low_pfn = -1;
max_low_pfn = 0;
efi_memmap_walk(filter_rsvd_memory, build_maps);
- efi_memmap_walk(filter_rsvd_memory, find_bootmap_space);
+ efi_memmap_walk(filter_rsvd_memory, find_pernode_space);
efi_memmap_walk(filter_rsvd_memory, discontig_free_bootmem_node);
- discontig_reserve_bootmem();
- allocate_pernode_structures();
-}
-
-/*
- * Initialize the paging system.
- * - determine sizes of each node
- * - initialize the paging system for the node
- * - build the nodedir for the node. This contains pointers to
- * the per-bank mem_map entries.
- * - fix the page struct "virtual" pointers. These are bank specific
- * values that the paging system doesn't understand.
- * - replicate the nodedir structure to other nodes
- */
-
-void __init
-discontig_paging_init(void)
-{
- int node, mynode;
- unsigned long max_dma, zones_size[MAX_NR_ZONES];
- unsigned long kaddr, ekaddr, bid;
- struct page *page;
- bootmem_data_t *bdp;
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
- mynode = boot_get_local_nodeid();
- for (node = 0; node < numnodes; node++) {
- long pfn, startpfn;
-
- memset(zones_size, 0, sizeof(zones_size));
-
- startpfn = -1;
- bdp = BOOT_NODE_DATA(node)->bdata;
- pfn = bdp->node_boot_start >> PAGE_SHIFT;
- if (startpfn = -1)
- startpfn = pfn;
- if (pfn > max_dma)
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - pfn);
- else if (bdp->node_low_pfn < max_dma)
- zones_size[ZONE_DMA] += (bdp->node_low_pfn - pfn);
- else {
- zones_size[ZONE_DMA] += (max_dma - pfn);
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - max_dma);
- }
-
- free_area_init_node(node, NODE_DATA(node), NULL, zones_size, startpfn, 0);
-
- page = NODE_DATA(node)->node_mem_map;
-
- bdp = BOOT_NODE_DATA(node)->bdata;
-
- kaddr = (unsigned long)__va(bdp->node_boot_start);
- ekaddr = (unsigned long)__va(bdp->node_low_pfn << PAGE_SHIFT);
- while (kaddr < ekaddr) {
- if (paddr_to_nid(__pa(kaddr)) = node) {
- bid = BANK_MEM_MAP_INDEX(kaddr);
- node_data[mynode]->node_id_map[bid] = node;
- node_data[mynode]->bank_mem_map_base[bid] = page;
- }
- kaddr += BANKSIZE;
- page += BANKSIZE/PAGE_SIZE;
- }
- }
-
- /*
- * Finish setting up the node data for this node, then copy it to the other nodes.
- */
- for (node=0; node < numnodes; node++)
- if (mynode != node) {
- memcpy(node_data[node], node_data[mynode], sizeof(struct ia64_node_data));
- node_data[node]->node = node;
- }
+ discontig_reserve_bootmem();
+ initialize_pernode_data();
}
diff -Nru a/arch/ia64/mm/init.c b/arch/ia64/mm/init.c
--- a/arch/ia64/mm/init.c Wed Jul 16 15:36:33 2003
+++ b/arch/ia64/mm/init.c Wed Jul 16 15:36:33 2003
@@ -44,7 +44,7 @@
#ifdef CONFIG_VIRTUAL_MEM_MAP
# define LARGE_GAP 0x40000000 /* Use virtual mem map if hole is > than this */
unsigned long vmalloc_end = VMALLOC_END_INIT;
- static struct page *vmem_map;
+ struct page *vmem_map;
static unsigned long num_dma_physpages;
#endif
@@ -240,7 +240,7 @@
else if (page_count(pgdat->node_mem_map + i))
shared += page_count(pgdat->node_mem_map + i) - 1;
}
- printk("\t%d pages of RAM\n", pgdat->node_spanned_pages);
+ printk("\t%ld pages of RAM\n", pgdat->node_spanned_pages);
printk("\t%d reserved pages\n", reserved);
printk("\t%d pages shared\n", shared);
printk("\t%d pages swap cached\n", cached);
@@ -397,6 +397,7 @@
{
unsigned long address, start_page, end_page;
struct page *map_start, *map_end;
+ int node;
pgd_t *pgd;
pmd_t *pmd;
pte_t *pte;
@@ -406,19 +407,20 @@
start_page = (unsigned long) map_start & PAGE_MASK;
end_page = PAGE_ALIGN((unsigned long) map_end);
+ node = paddr_to_nid(__pa(start));
for (address = start_page; address < end_page; address += PAGE_SIZE) {
pgd = pgd_offset_k(address);
if (pgd_none(*pgd))
- pgd_populate(&init_mm, pgd, alloc_bootmem_pages(PAGE_SIZE));
+ pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pmd = pmd_offset(pgd, address);
if (pmd_none(*pmd))
- pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages(PAGE_SIZE));
+ pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pte = pte_offset_kernel(pmd, address);
if (pte_none(*pte))
- set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages(PAGE_SIZE)) >> PAGE_SHIFT,
+ set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
PAGE_KERNEL));
}
return 0;
@@ -431,6 +433,14 @@
unsigned long zone;
};
+struct memmap_count_callback_data {
+ int node;
+ unsigned long num_physpages;
+ unsigned long num_dma_physpages;
+ unsigned long min_pfn;
+ unsigned long max_pfn;
+} cdata;
+
static int
virtual_memmap_init (u64 start, u64 end, void *arg)
{
@@ -489,16 +499,6 @@
}
static int
-count_dma_pages (u64 start, u64 end, void *arg)
-{
- unsigned long *count = arg;
-
- if (end <= MAX_DMA_ADDRESS)
- *count += (end - start) >> PAGE_SHIFT;
- return 0;
-}
-
-static int
find_largest_hole (u64 start, u64 end, void *arg)
{
u64 *max_gap = arg;
@@ -514,102 +514,101 @@
}
#endif /* CONFIG_VIRTUAL_MEM_MAP */
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
+#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
+#define ORDERROUNDDOWN(n) ((n) & ~((PAGE_SIZE<<MAX_ORDER)-1))
static int
-count_pages (u64 start, u64 end, void *arg)
+count_pages (unsigned long start, unsigned long end, int node)
{
- unsigned long *count = arg;
+ start = __pa(start);
+ end = __pa(end);
- *count += (end - start) >> PAGE_SHIFT;
+ if (node = cdata.node) {
+ cdata.num_physpages += (end - start) >> PAGE_SHIFT;
+ if (start <= __pa(MAX_DMA_ADDRESS))
+ cdata.num_dma_physpages += (min(end, __pa(MAX_DMA_ADDRESS)) - start) >> PAGE_SHIFT;
+ start = GRANULEROUNDDOWN(__pa(start));
+ start = ORDERROUNDDOWN(start);
+ end = GRANULEROUNDUP(__pa(end));
+ cdata.max_pfn = max(cdata.max_pfn, end >> PAGE_SHIFT);
+ cdata.min_pfn = min(cdata.min_pfn, start >> PAGE_SHIFT);
+ }
return 0;
}
/*
* Set up the page tables.
*/
-
-#ifdef CONFIG_DISCONTIGMEM
void
paging_init (void)
{
- extern void discontig_paging_init(void);
-
- discontig_paging_init();
- efi_memmap_walk(count_pages, &num_physpages);
- zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
-}
-#else /* !CONFIG_DISCONTIGMEM */
-void
-paging_init (void)
-{
- unsigned long max_dma;
+ unsigned long max_dma_pfn;
unsigned long zones_size[MAX_NR_ZONES];
# ifdef CONFIG_VIRTUAL_MEM_MAP
unsigned long zholes_size[MAX_NR_ZONES];
unsigned long max_gap;
# endif
+ int node;
- /* initialize mem_map[] */
-
- memset(zones_size, 0, sizeof(zones_size));
-
- num_physpages = 0;
- efi_memmap_walk(count_pages, &num_physpages);
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
-
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- memset(zholes_size, 0, sizeof(zholes_size));
-
- num_dma_physpages = 0;
- efi_memmap_walk(count_dma_pages, &num_dma_physpages);
-
- if (max_low_pfn < max_dma) {
- zones_size[ZONE_DMA] = max_low_pfn;
- zholes_size[ZONE_DMA] = max_low_pfn - num_dma_physpages;
- } else {
- zones_size[ZONE_DMA] = max_dma;
- zholes_size[ZONE_DMA] = max_dma - num_dma_physpages;
- if (num_physpages > num_dma_physpages) {
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
- zholes_size[ZONE_NORMAL] = ((max_low_pfn - max_dma)
- - (num_physpages - num_dma_physpages));
- }
- }
-
+ max_dma_pfn = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
max_gap = 0;
efi_memmap_walk(find_largest_hole, (u64 *)&max_gap);
- if (max_gap < LARGE_GAP) {
- vmem_map = (struct page *) 0;
- free_area_init_node(0, &contig_page_data, NULL, zones_size, 0, zholes_size);
- mem_map = contig_page_data.node_mem_map;
- }
- else {
- unsigned long map_size;
-
- /* allocate virtual_mem_map */
- map_size = PAGE_ALIGN(max_low_pfn * sizeof(struct page));
- vmalloc_end -= map_size;
- vmem_map = (struct page *) vmalloc_end;
- efi_memmap_walk(create_mem_map_page_table, 0);
-
- free_area_init_node(0, &contig_page_data, vmem_map, zones_size, 0, zholes_size);
+ for (node = 0; node < numnodes; node++) {
+ memset(zones_size, 0, sizeof(zones_size));
+ memset(zholes_size, 0, sizeof(zholes_size));
+ memset(&cdata, 0, sizeof(cdata));
+
+ cdata.node = node;
+ cdata.min_pfn = ~0;
+
+ efi_memmap_walk(filter_rsvd_memory, count_pages);
+ num_dma_physpages += cdata.num_dma_physpages;
+ num_physpages += cdata.num_physpages;
+
+ if (cdata.min_pfn >= max_dma_pfn) {
+ /* Above the DMA zone */
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn - cdata.num_physpages;
+ } else if (cdata.max_pfn < max_dma_pfn) {
+ /* This block is DMAable */
+ zones_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn - cdata.num_dma_physpages;
+ } else {
+ zones_size[ZONE_DMA] = max_dma_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - cdata.num_dma_physpages;
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - max_dma_pfn;
+ zholes_size[ZONE_NORMAL] = zones_size[ZONE_NORMAL] - (cdata.num_physpages - cdata.num_dma_physpages);
+ }
- mem_map = contig_page_data.node_mem_map;
- printk("Virtual mem_map starts at 0x%p\n", mem_map);
- }
-# else /* !CONFIG_VIRTUAL_MEM_MAP */
- if (max_low_pfn < max_dma)
- zones_size[ZONE_DMA] = max_low_pfn;
- else {
- zones_size[ZONE_DMA] = max_dma;
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
+ if (numnodes = 1 && max_gap < LARGE_GAP) {
+ /* Just one node with no big holes... */
+ vmem_map = (struct page *)0;
+ zones_size[ZONE_DMA] += cdata.min_pfn;
+ zholes_size[ZONE_DMA] += cdata.min_pfn;
+ free_area_init_node(0, NODE_DATA(node), NODE_DATA(node)->node_mem_map,
+ zones_size, 0, zholes_size);
+ }
+ else {
+ /* allocate virtual mem_map */
+ if (node = 0) {
+ unsigned long map_size;
+ map_size = PAGE_ALIGN(max_low_pfn*sizeof(struct page));
+ vmalloc_end -= map_size;
+ vmem_map = (struct page *) vmalloc_end;
+ efi_memmap_walk(create_mem_map_page_table, 0);
+ printk("Virtual mem_map starts at 0x%p\n", vmem_map);
+#ifndef CONFIG_DISCONTIGMEM
+ mem_map = vmem_map;
+#endif
+ }
+ free_area_init_node(node, NODE_DATA(node), vmem_map + cdata.min_pfn,
+ zones_size, cdata.min_pfn, zholes_size);
+ }
}
- free_area_init(zones_size);
-# endif /* !CONFIG_VIRTUAL_MEM_MAP */
+
zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
}
-#endif /* !CONFIG_DISCONTIGMEM */
static int
count_reserved_pages (u64 start, u64 end, void *arg)
diff -Nru a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
--- a/drivers/acpi/Kconfig Wed Jul 16 15:36:33 2003
+++ b/drivers/acpi/Kconfig Wed Jul 16 15:36:33 2003
@@ -133,7 +133,7 @@
config ACPI_NUMA
bool "NUMA support" if NUMA && (IA64 && !IA64_HP_SIM || X86 && ACPI && !ACPI_HT_ONLY && !X86_64)
- default y if IA64 && IA64_SGI_SN
+ default y if IA64_GENERIC || IA64_SGI_SN2
config ACPI_ASUS
tristate "ASUS/Medion Laptop Extras"
diff -Nru a/include/asm-ia64/mmzone.h b/include/asm-ia64/mmzone.h
--- a/include/asm-ia64/mmzone.h Wed Jul 16 15:36:33 2003
+++ b/include/asm-ia64/mmzone.h Wed Jul 16 15:36:33 2003
@@ -3,7 +3,7 @@
* License. See the file "COPYING" in the main directory of this archive
* for more details.
*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000,2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2002 NEC Corp.
* Copyright (c) 2002 Erich Focht <efocht@ess.nec.de>
* Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com>
@@ -14,150 +14,50 @@
#include <linux/config.h>
#include <linux/init.h>
-/*
- * Given a kaddr, find the base mem_map address for the start of the mem_map
- * entries for the bank containing the kaddr.
- */
-#define BANK_MEM_MAP_BASE(kaddr) local_node_data->bank_mem_map_base[BANK_MEM_MAP_INDEX(kaddr)]
-
-/*
- * Given a kaddr, this macro return the relative map number
- * within the bank.
- */
-#define BANK_MAP_NR(kaddr) (BANK_OFFSET(kaddr) >> PAGE_SHIFT)
-/*
- * Given a pte, this macro returns a pointer to the page struct for the pte.
- */
-#define pte_page(pte) virt_to_page(PAGE_OFFSET | (pte_val(pte)&_PFN_MASK))
+#ifdef CONFIG_NUMA
-/*
- * Determine if a kaddr is a valid memory address of memory that
- * actually exists.
- *
- * The check consists of 2 parts:
- * - verify that the address is a region 7 address & does not
- * contain any bits that preclude it from being a valid platform
- * memory address
- * - verify that the chunk actually exists.
- *
- * Note that IO addresses are NOT considered valid addresses.
- *
- * Note, many platforms can simply check if kaddr exceeds a specific size.
- * (However, this won't work on SGI platforms since IO space is embedded
- * within the range of valid memory addresses & nodes have holes in the
- * address range between banks).
- */
-#define kern_addr_valid(kaddr) ({long _kav=(long)(kaddr); \
- VALID_MEM_KADDR(_kav);})
-
-/*
- * Given a kaddr, return a pointer to the page struct for the page.
- * If the kaddr does not represent RAM memory that potentially exists, return
- * a pointer the page struct for max_mapnr. IO addresses will
- * return the page for max_nr. Addresses in unpopulated RAM banks may
- * return undefined results OR may panic the system.
- *
- */
-#define virt_to_page(kaddr) ({long _kvtp=(long)(kaddr); \
- (VALID_MEM_KADDR(_kvtp)) \
- ? BANK_MEM_MAP_BASE(_kvtp) + BANK_MAP_NR(_kvtp) \
- : NULL;})
+#ifdef CONFIG_IA64_DIG
/*
- * Given a page struct entry, return the physical address that the page struct represents.
- * Since IA64 has all memory in the DMA zone, the following works:
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define page_to_phys(page) __pa(page_address(page))
-
-#define node_mem_map(nid) (NODE_DATA(nid)->node_mem_map)
+#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
+#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES * 32)
-#define node_localnr(pfn, nid) ((pfn) - NODE_DATA(nid)->node_start_pfn)
-#define pfn_to_page(pfn) (struct page *)(node_mem_map(pfn_to_nid(pfn)) + node_localnr(pfn, pfn_to_nid(pfn)))
-#define pfn_to_nid(pfn) local_node_data->node_id_map[(pfn << PAGE_SHIFT) >> BANKSHIFT]
-
-#define page_to_pfn(page) (long)((page - page_zone(page)->zone_mem_map) + page_zone(page)->zone_start_pfn)
+#elif CONFIG_IA64_SGI_SN2
/*
- * pfn_valid should be made as fast as possible, and the current definition
- * is valid for machines that are NUMA, but still contiguous, which is what
- * is currently supported. A more generalised, but slower definition would
- * be something like this - mbligh:
- * ( pfn_to_pgdat(pfn) && (pfn < node_end_pfn(pfn_to_nid(pfn))) )
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define pfn_valid(pfn) (pfn < max_low_pfn)
-extern unsigned long max_low_pfn;
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of compute nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
+#elif CONFIG_IA64_GENERIC
-#ifdef CONFIG_IA64_DIG
/*
- * Platform definitions for DIG platform with contiguous memory.
+ * Platform definitions for GENERIC platform with contiguous or discontiguous memory.
*/
-#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
-#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
-#define MAX_PHYS_MEMORY (1UL << 40) /* 1 TB */
-/*
- * Bank definitions.
- * Configurable settings for DIG: 512MB/bank: 16GB/node,
- * 2048MB/bank: 64GB/node,
- * 8192MB/bank: 256GB/node.
- */
-#define NR_BANKS_PER_NODE 32
-#if defined(CONFIG_IA64_NODESIZE_16GB)
-# define BANKSHIFT 29
-#elif defined(CONFIG_IA64_NODESIZE_64GB)
-# define BANKSHIFT 31
-#elif defined(CONFIG_IA64_NODESIZE_256GB)
-# define BANKSHIFT 33
#else
-# error Unsupported bank and nodesize!
+#error unknown platform
#endif
-#define BANKSIZE (1UL << BANKSHIFT)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-/*
- * VALID_MEM_KADDR returns a boolean to indicate if a kaddr is
- * potentially a valid cacheable identity mapped RAM memory address.
- * Note that the RAM may or may not actually be present!!
- */
-#define VALID_MEM_KADDR(kaddr) 1
+extern void build_cpu_to_node_map(void);
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#else /* CONFIG_NUMA */
-#elif defined(CONFIG_IA64_SGI_SN2)
-/*
- * SGI SN2 discontig definitions
- */
-#define MAX_PHYSNODE_ID 2048 /* 2048 node ids (also called nasid) */
-#define NR_NODES 128 /* Maximum number of nodes in SSI */
-#define MAX_PHYS_MEMORY (1UL << 49)
-
-#define BANKSHIFT 38
-#define NR_BANKS_PER_NODE 4
-#define SN2_NODE_SIZE (64UL*1024*1024*1024) /* 64GB per node */
-#define BANKSIZE (SN2_NODE_SIZE/NR_BANKS_PER_NODE)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-#define VALID_MEM_KADDR(kaddr) 1
-
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#define NR_NODES 1
-#endif /* CONFIG_IA64_DIG */
+#endif /* CONFIG_NUMA */
#endif /* _ASM_IA64_MMZONE_H */
diff -Nru a/include/asm-ia64/nodedata.h b/include/asm-ia64/nodedata.h
--- a/include/asm-ia64/nodedata.h Wed Jul 16 15:36:33 2003
+++ b/include/asm-ia64/nodedata.h Wed Jul 16 15:36:33 2003
@@ -13,7 +13,7 @@
#ifndef _ASM_IA64_NODEDATA_H
#define _ASM_IA64_NODEDATA_H
-
+#include <asm/percpu.h>
#include <asm/mmzone.h>
/*
@@ -22,15 +22,17 @@
struct pglist_data;
struct ia64_node_data {
- short active_cpu_count;
short node;
+ short active_cpu_count;
+ /*
+ * The fields are read-only (after boot). They contain pointers
+ * to various structures located on other nodes. Ths data is
+ * replicated on each node in order to reduce off-node references.
+ */
struct pglist_data *pg_data_ptrs[NR_NODES];
- struct page *bank_mem_map_base[NR_BANKS];
struct ia64_node_data *node_data_ptrs[NR_NODES];
- short node_id_map[NR_BANKS];
};
-
/*
* Return a pointer to the node_data structure for the executing cpu.
*/
@@ -40,7 +42,8 @@
/*
* Return a pointer to the node_data structure for the specified node.
*/
-#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
/*
* Get a pointer to the node_id/node_data for the current cpu.
@@ -48,29 +51,5 @@
*/
extern int boot_get_local_nodeid(void);
extern struct ia64_node_data *get_node_data_ptr(void);
-
-/*
- * Given a node id, return a pointer to the pg_data_t for the node.
- * The following 2 macros are similar.
- *
- * NODE_DATA - should be used in all code not related to system
- * initialization. It uses pernode data structures to minimize
- * offnode memory references. However, these structure are not
- * present during boot. This macro can be used once cpu_init
- * completes.
- *
- * BOOT_NODE_DATA
- * - should be used during system initialization
- * prior to freeing __initdata. It does not depend on the percpu
- * area being present.
- *
- * NOTE: The names of these macros are misleading but are difficult to change
- * since they are used in generic linux & on other architecures.
- */
-#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
-#define BOOT_NODE_DATA(nid) boot_get_pg_data_ptr((long)(nid))
-
-struct pglist_data;
-extern struct pglist_data * __init boot_get_pg_data_ptr(long);
#endif /* _ASM_IA64_NODEDATA_H */
diff -Nru a/include/asm-ia64/numa.h b/include/asm-ia64/numa.h
--- a/include/asm-ia64/numa.h Wed Jul 16 15:36:33 2003
+++ b/include/asm-ia64/numa.h Wed Jul 16 15:36:33 2003
@@ -15,13 +15,21 @@
#ifdef CONFIG_DISCONTIGMEM
# include <asm/mmzone.h>
-# define NR_MEMBLKS (NR_BANKS)
#else
# define NR_NODES (8)
# define NR_MEMBLKS (NR_NODES * 8)
#endif
#include <linux/cache.h>
+#include <linux/threads.h>
+#include <linux/smp.h>
+
+#define NODEMASK_WORDCOUNT ((NR_NODES+(BITS_PER_LONG-1))/BITS_PER_LONG)
+
+#define NODE_MASK_NONE { [0 ... ((NR_NODES+BITS_PER_LONG-1)/BITS_PER_LONG)-1] = 0 }
+
+typedef unsigned long nodemask_t[NODEMASK_WORDCOUNT];
+
extern volatile char cpu_to_node_map[NR_CPUS] __cacheline_aligned;
extern volatile unsigned long node_to_cpu_mask[NR_NODES] __cacheline_aligned;
@@ -63,6 +71,12 @@
extern int paddr_to_nid(unsigned long paddr);
#define local_nodeid (cpu_to_node_map[smp_processor_id()])
+
+#else /* !CONFIG_NUMA */
+
+#define node_distance(from,to) 10
+#define paddr_to_nid(x) 0
+#define local_nodeid 0
#endif /* CONFIG_NUMA */
diff -Nru a/include/asm-ia64/page.h b/include/asm-ia64/page.h
--- a/include/asm-ia64/page.h Wed Jul 16 15:36:33 2003
+++ b/include/asm-ia64/page.h Wed Jul 16 15:36:33 2003
@@ -93,18 +93,26 @@
#define virt_addr_valid(kaddr) pfn_valid(__pa(kaddr) >> PAGE_SHIFT)
-#ifndef CONFIG_DISCONTIGMEM
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- extern int ia64_pfn_valid (unsigned long pfn);
-# define pfn_valid(pfn) (((pfn) < max_mapnr) && ia64_pfn_valid(pfn))
-# else
-# define pfn_valid(pfn) ((pfn) < max_mapnr)
-# endif
-#define virt_to_page(kaddr) pfn_to_page(__pa(kaddr) >> PAGE_SHIFT)
-#define page_to_pfn(page) ((unsigned long) (page - mem_map))
-#define pfn_to_page(pfn) (mem_map + (pfn))
-#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
+#ifdef CONFIG_VIRTUAL_MEM_MAP
+extern int ia64_pfn_valid(unsigned long pfn);
+#else
+#define ia64_pfn_valid(pfn) (1)
+#endif
+
+extern unsigned long max_low_pfn;
+#define pfn_valid(pfn) (((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
+
+#if defined(CONFIG_VIRTUAL_MEM_MAP) && !defined(CONFIG_DISCONTIGMEM)
+#define vmem_map mem_map
+#else
+extern struct page *vmem_map;
#endif
+
+#define pfn_to_page(pfn) (vmem_map + (pfn))
+#define page_to_pfn(page) ((unsigned long) (page - vmem_map))
+
+#define virt_to_page(kaddr) (pfn_to_page(__pa(kaddr) >> PAGE_SHIFT))
+#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
typedef union ia64_va {
struct {
diff -Nru a/include/asm-ia64/pgtable.h b/include/asm-ia64/pgtable.h
--- a/include/asm-ia64/pgtable.h Wed Jul 16 15:36:33 2003
+++ b/include/asm-ia64/pgtable.h Wed Jul 16 15:36:33 2003
@@ -174,7 +174,6 @@
return (addr & (local_cpu_data->unimpl_pa_mask)) = 0;
}
-#ifndef CONFIG_DISCONTIGMEM
/*
* kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
* memory. For the return value to be meaningful, ADDR must be >@@ -190,7 +189,6 @@
*/
#define kern_addr_valid(addr) (1)
-#endif
/*
* Now come the defines and routines to manage and access the three-level
@@ -241,10 +239,8 @@
#define pte_none(pte) (!pte_val(pte))
#define pte_present(pte) (pte_val(pte) & (_PAGE_P | _PAGE_PROTNONE))
#define pte_clear(pte) (pte_val(*(pte)) = 0UL)
-#ifndef CONFIG_DISCONTIGMEM
/* pte_page() returns the "struct page *" corresponding to the PTE: */
#define pte_page(pte) virt_to_page(((pte_val(pte) & _PFN_MASK) + PAGE_OFFSET))
-#endif
#define pmd_none(pmd) (!pmd_val(pmd))
#define pmd_bad(pmd) (!ia64_phys_addr_valid(pmd_val(pmd)))
@@ -416,6 +412,7 @@
extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
extern void paging_init (void);
+extern int filter_rsvd_memory(unsigned long start, unsigned long end, void *arg);
/*
* Note: The macros below rely on the fact that MAX_SWAPFILES_SHIFT <= number of
diff -Nru a/include/asm-ia64/processor.h b/include/asm-ia64/processor.h
--- a/include/asm-ia64/processor.h Wed Jul 16 15:36:33 2003
+++ b/include/asm-ia64/processor.h Wed Jul 16 15:36:33 2003
@@ -185,6 +185,8 @@
#endif
#ifdef CONFIG_NUMA
struct ia64_node_data *node_data;
+ struct cpuinfo_ia64 *cpu_data[NR_CPUS];
+ int nodeid;
#endif
};
diff -Nru a/mm/bootmem.c b/mm/bootmem.c
--- a/mm/bootmem.c Wed Jul 16 15:36:33 2003
+++ b/mm/bootmem.c Wed Jul 16 15:36:33 2003
@@ -48,8 +48,24 @@
bootmem_data_t *bdata = pgdat->bdata;
unsigned long mapsize = ((end - start)+7)/8;
- pgdat->pgdat_next = pgdat_list;
- pgdat_list = pgdat;
+
+ /*
+ * sort pgdat_list so that the lowest one comes first,
+ * which makes alloc_bootmem_low_pages work as desired.
+ */
+ if (!pgdat_list || pgdat_list->node_start_pfn > pgdat->node_start_pfn) {
+ pgdat->pgdat_next = pgdat_list;
+ pgdat_list = pgdat;
+ } else {
+ pg_data_t *tmp = pgdat_list;
+ while (tmp->pgdat_next) {
+ if (tmp->pgdat_next->node_start_pfn > pgdat->node_start_pfn)
+ break;
+ tmp = tmp->pgdat_next;
+ }
+ pgdat->pgdat_next = tmp->pgdat_next;
+ tmp->pgdat_next = pgdat;
+ }
mapsize = (mapsize + (sizeof(long) - 1UL)) & ~(sizeof(long) - 1UL);
bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
@@ -251,7 +267,7 @@
static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
{
- struct page *page = pgdat->node_mem_map;
+ struct page *page;
bootmem_data_t *bdata = pgdat->bdata;
unsigned long i, count, total = 0;
unsigned long idx;
@@ -260,23 +276,23 @@
if (!bdata->node_bootmem_map) BUG();
count = 0;
+ page = virt_to_page(phys_to_virt(bdata->node_boot_start));
idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
map = bdata->node_bootmem_map;
for (i = 0; i < idx; ) {
unsigned long v = ~map[i / BITS_PER_LONG];
if (v) {
unsigned long m;
- for (m = 1; m && i < idx; m<<=1, page++, i++) {
+ for (m = 1; m && i < idx; m<<=1, i++) {
if (v & m) {
count++;
- ClearPageReserved(page);
- set_page_count(page, 1);
- __free_page(page);
+ ClearPageReserved(page+i);
+ set_page_count(page+i, 1);
+ __free_page(page+i);
}
}
} else {
i+=BITS_PER_LONG;
- page += BITS_PER_LONG;
}
}
total += count;
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (9 preceding siblings ...)
2003-07-16 22:37 ` Jesse Barnes
@ 2003-07-17 8:23 ` Erich Focht
2003-07-18 0:16 ` Jesse Barnes
` (2 subsequent siblings)
13 siblings, 0 replies; 15+ messages in thread
From: Erich Focht @ 2003-07-17 8:23 UTC (permalink / raw)
To: linux-ia64
> > Same comment as above. NEC TX7 is an IA64_DIG platform.
>
> Oops, sorry. Here's another one. How does it look?
...
> --- a/arch/ia64/Kconfig Wed Jul 16 15:36:33 2003
> +++ b/arch/ia64/Kconfig Wed Jul 16 15:36:33 2003
> @@ -211,7 +211,7 @@
>
> config NUMA
> bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
> - default y if IA64_SGI_SN2
> + default y if IA64_SGI_SN2 || IA64_GENERIC
> help
> Say Y to compile the kernel to support NUMA (Non-Uniform Memory
> Access). This option is for configuring high-end multiprocessor
> @@ -234,9 +234,8 @@
> endchoice
OK. Thanks.
>
> config DISCONTIGMEM
> - bool
> - depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) &&
> NUMA - default y
> + bool "Discontiguous memory support" if IA64_DIG
> + default y if IA64_SGI_SN2 || IA64_GENERIC
> help
> Say Y to support efficient handling of discontiguous physical memory,
> for architectures which are either NUMA (Non-Uniform Memory Access)
> @@ -245,8 +244,7 @@
Maybe that should better be
bool "Discontiguous memory support" if (IA64_DIG && NUMA)
Discontigmem alone doesn't make sense, yet, I think...
Regards,
Erich
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (10 preceding siblings ...)
2003-07-17 8:23 ` Erich Focht
@ 2003-07-18 0:16 ` Jesse Barnes
2003-07-18 16:15 ` Erich Focht
2003-07-21 18:46 ` Takayoshi Kochi
13 siblings, 0 replies; 15+ messages in thread
From: Jesse Barnes @ 2003-07-18 0:16 UTC (permalink / raw)
To: linux-ia64
On Thu, Jul 17, 2003 at 10:23:37AM +0200, Erich Focht wrote:
> Maybe that should better be
> bool "Discontiguous memory support" if (IA64_DIG && NUMA)
>
> Discontigmem alone doesn't make sense, yet, I think...
Not for ia64 I guess, unless we make zx1 use discontig as well. Patch
appended.
Thanks,
Jesse
# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
# ChangeSet 1.1397 -> 1.1399
# include/asm-ia64/page.h 1.19 -> 1.20
# arch/ia64/kernel/setup.c 1.53 -> 1.54
# include/asm-ia64/pgtable.h 1.28 -> 1.29
# mm/bootmem.c 1.18 -> 1.19
# include/asm-ia64/numa.h 1.5 -> 1.6
# include/asm-ia64/processor.h 1.48 -> 1.49
# arch/ia64/mm/init.c 1.46 -> 1.47
# include/asm-ia64/nodedata.h 1.3 -> 1.4
# arch/ia64/mm/discontig.c 1.4 -> 1.5
# arch/ia64/Kconfig 1.38 -> 1.40
# drivers/acpi/Kconfig 1.12 -> 1.13
# include/asm-ia64/mmzone.h 1.4 -> 1.5
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 03/07/16 jbarnes@tomahawk.engr.sgi.com 1.1398
# latest discontig
# --------------------------------------------
# 03/07/17 jbarnes@tomahawk.engr.sgi.com 1.1399
# another discontig patch
# --------------------------------------------
#
diff -Nru a/arch/ia64/Kconfig b/arch/ia64/Kconfig
--- a/arch/ia64/Kconfig Thu Jul 17 16:59:05 2003
+++ b/arch/ia64/Kconfig Thu Jul 17 16:59:05 2003
@@ -211,7 +211,7 @@
config NUMA
bool "Enable NUMA support" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
+ default y if IA64_SGI_SN2 || IA64_GENERIC
help
Say Y to compile the kernel to support NUMA (Non-Uniform Memory
Access). This option is for configuring high-end multiprocessor
@@ -234,9 +234,8 @@
endchoice
config DISCONTIGMEM
- bool
- depends on IA64_SGI_SN2 || (IA64_GENERIC || IA64_DIG || IA64_HP_ZX1) && NUMA
- default y
+ bool "Discontiguous memory support" if (IA64_DIG && NUMA)
+ default y if IA64_SGI_SN2 || IA64_GENERIC
help
Say Y to support efficient handling of discontiguous physical memory,
for architectures which are either NUMA (Non-Uniform Memory Access)
@@ -245,8 +244,7 @@
config VIRTUAL_MEM_MAP
bool "Enable Virtual Mem Map"
- depends on !NUMA
- default y if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
+ default y if !IA64_HP_SIM
help
Say Y to compile the kernel with support for a virtual mem map.
This is an alternate method of supporting large holes in the
@@ -259,8 +257,8 @@
are unsure, say Y.
config IA64_MCA
- bool "Enable IA-64 Machine Check Abort" if IA64_GENERIC || IA64_DIG || IA64_HP_ZX1
- default y if IA64_SGI_SN2
+ bool "Enable IA-64 Machine Check Abort"
+ default y if !IA64_HP_SIM
help
Say Y here to enable machine check support for IA-64. If you're
unsure, answer Y.
@@ -292,43 +290,12 @@
depends on IA64_GENERIC || IA64_DIG || IA64_HP_ZX1 || IA64_SGI_SN2
default y
-config IA64_SGI_SN_DEBUG
- bool "Enable extra debugging code"
- depends on IA64_SGI_SN2
- help
- Turns on extra debugging code in the SGI SN (Scalable NUMA) platform
- for IA-64. Unless you are debugging problems on an SGI SN IA-64 box,
- say N.
-
config IA64_SGI_SN_SIM
bool "Enable SGI Medusa Simulator Support"
depends on IA64_SGI_SN2
help
If you are compiling a kernel that will run under SGI's IA-64
simulator (Medusa) then say Y, otherwise say N.
-
-config IA64_SGI_AUTOTEST
- bool "Enable autotest (llsc). Option to run cache test instead of booting"
- depends on IA64_SGI_SN2
- help
- Build a kernel used for hardware validation. If you include the
- keyword "autotest" on the boot command line, the kernel does NOT boot.
- Instead, it starts all cpus and runs cache coherency tests instead.
-
- If unsure, say N.
-
-config SERIAL_SGI_L1_PROTOCOL
- bool "Enable protocol mode for the L1 console"
- depends on IA64_SGI_SN2
- help
- Uses protocol mode instead of raw mode for the level 1 console on the
- SGI SN (Scalable NUMA) platform for IA-64. If you are compiling for
- an SGI SN box then Y is the recommended value, otherwise say N.
-
-config PERCPU_IRQ
- bool
- depends on IA64_SGI_SN2
- default y
# On IA-64, we always want an ELF /proc/kcore.
config KCORE_ELF
diff -Nru a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c
--- a/arch/ia64/kernel/setup.c Thu Jul 17 16:59:04 2003
+++ b/arch/ia64/kernel/setup.c Thu Jul 17 16:59:04 2003
@@ -138,7 +138,7 @@
call_pernode_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long rs, re;
- void (*func)(unsigned long, unsigned long, int, int);
+ void (*func)(unsigned long, unsigned long, int);
int i;
start = PAGE_ALIGN(start);
@@ -149,22 +149,21 @@
func = arg;
if (!num_memblks) {
- /*
- * This machine doesn't have SRAT, so call func with
- * nid=0, bank=0.
- */
+ /* No SRAT table, to assume one node (node 0) */
if (start < end)
- (*func)(start, end - start, 0, 0);
+ (*func)(start, end, 0);
return;
}
for (i = 0; i < num_memblks; i++) {
- rs = max(start, node_memblk[i].start_paddr);
- re = min(end, node_memblk[i].start_paddr+node_memblk[i].size);
+ rs = max(__pa(start), node_memblk[i].start_paddr);
+ re = min(__pa(end), node_memblk[i].start_paddr+node_memblk[i].size);
if (rs < re)
- (*func)(rs, re-rs, node_memblk[i].nid,
- node_memblk[i].bank);
+ (*func)((unsigned long)__va(rs), (unsigned long)__va(re), node_memblk[i].nid);
+
+ if ((unsigned long)__va(re) = end)
+ break;
}
}
@@ -180,7 +179,7 @@
filter_rsvd_memory (unsigned long start, unsigned long end, void *arg)
{
unsigned long range_start, range_end, prev_start;
- void (*func)(unsigned long, unsigned long);
+ void (*func)(unsigned long, unsigned long, int);
int i;
#if IGNORE_PFN0
@@ -202,9 +201,9 @@
if (range_start < range_end)
#ifdef CONFIG_DISCONTIGMEM
- call_pernode_memory(__pa(range_start), __pa(range_end), func);
+ call_pernode_memory(range_start, range_end, func);
#else
- (*func)(__pa(range_start), range_end - range_start);
+ (*func)(range_start, range_end, 0);
#endif
/* nothing more available in this segment */
@@ -703,6 +702,8 @@
* get_free_pages() cannot be used before cpu_init() done. BSP allocates
* "NR_CPUS" pages for all CPUs to avoid that AP calls get_zeroed_page().
*/
+#ifndef CONFIG_DISCONTIGMEM
+ /* for discontig machines, we do this in discontig.c */
if (smp_processor_id() = 0) {
cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * NR_CPUS, PERCPU_PAGE_SIZE,
__pa(MAX_DMA_ADDRESS));
@@ -714,6 +715,7 @@
per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
}
}
+#endif
cpu_data = __per_cpu_start + __per_cpu_offset[smp_processor_id()];
#else /* !CONFIG_SMP */
cpu_data = __phys_per_cpu_start;
diff -Nru a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c
--- a/arch/ia64/mm/discontig.c Thu Jul 17 16:59:05 2003
+++ b/arch/ia64/mm/discontig.c Thu Jul 17 16:59:05 2003
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000, 2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2001 Intel Corp.
* Copyright (c) 2001 Tony Luck <tony.luck@intel.com>
* Copyright (c) 2002 NEC Corp.
@@ -16,74 +16,60 @@
#include <linux/mmzone.h>
#include <linux/acpi.h>
#include <linux/efi.h>
-
+#include <asm/pgalloc.h>
+#include <asm/tlb.h>
/*
- * Round an address upward to the next multiple of GRANULE size.
+ * Round an address upward or downward to the next multiple of IA64_GRANULE_SIZE.
*/
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
-static struct ia64_node_data *node_data[NR_NODES];
-static long boot_pg_data[8*NR_NODES+sizeof(pg_data_t)] __initdata;
-static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
-static bootmem_data_t bdata[NR_NODES][NR_BANKS_PER_NODE+1] __initdata;
-
-extern int filter_rsvd_memory (unsigned long start, unsigned long end, void *arg);
+/*
+ * Used to locate BOOT_DATA prior to initializing the node data area.
+ */
+#define BOOT_NODE_DATA(node) pg_data_ptr[node]
/*
- * Return the compact node number of this cpu. Used prior to
- * setting up the cpu_data area.
- * Note - not fast, intended for boot use only!!
+ * To prevent cache aliasing effects, align per-node structures so that they
+ * start at addresses that are strided by node number.
*/
-int
-boot_get_local_nodeid(void)
-{
- int i;
+#define NODEDATA_ALIGN(addr, node) ((((addr) + 1024*1024-1) & ~(1024*1024-1)) + (node)*PERCPU_PAGE_SIZE)
- for (i = 0; i < NR_CPUS; i++)
- if (node_cpuid[i].phys_id = hard_smp_processor_id())
- return node_cpuid[i].nid;
- /* node info missing, so nid should be 0.. */
- return 0;
-}
+static struct ia64_node_data *boot_node_data[NR_NODES] __initdata;
+static pg_data_t *pg_data_ptr[NR_NODES] __initdata;
+static bootmem_data_t bdata[NR_NODES] __initdata;
+static unsigned long boot_pernode[NR_NODES] __initdata;
+static unsigned long boot_pernodesize[NR_NODES] __initdata;
-/*
- * Return a pointer to the pg_data structure for a node.
- * This function is used ONLY in early boot before the cpu_data
- * structure is available.
- */
-pg_data_t* __init
-boot_get_pg_data_ptr(long node)
-{
- return pg_data_ptr[node];
-}
+extern char __per_cpu_start[], __per_cpu_end[];
-/*
- * Return a pointer to the node data for the current node.
- * (boottime initialization only)
- */
-struct ia64_node_data *
+struct ia64_node_data*
get_node_data_ptr(void)
{
- return node_data[boot_get_local_nodeid()];
+ return boot_node_data[cpu_to_node_map[smp_processor_id()]]; /* ZZZ */
}
/*
* We allocate one of the bootmem_data_t structs for each piece of memory
* that we wish to treat as a contiguous block. Each such block must start
- * on a BANKSIZE boundary. Multiple banks per node is not supported.
+ * on a GRANULE boundary. Multiple banks per node are not supported.
+ * (Note: on SN2, all memory on a node is trated as a single bank.
+ * Holes within the bank are supported. This works because memory
+ * from different banks is not interleaved. The bootmap bitmap
+ * for the node is somewhat large but not too large).
*/
static int __init
-build_maps(unsigned long pstart, unsigned long length, int node)
+build_maps(unsigned long start, unsigned long end, int node)
{
bootmem_data_t *bdp;
unsigned long cstart, epfn;
- bdp = pg_data_ptr[node]->bdata;
- epfn = GRANULEROUNDUP(pstart + length) >> PAGE_SHIFT;
- cstart = pstart & ~(BANKSIZE - 1);
+ bdp = &bdata[node];
+ epfn = GRANULEROUNDUP(__pa(end)) >> PAGE_SHIFT;
+ cstart = GRANULEROUNDDOWN(__pa(start));
if (!bdp->node_low_pfn) {
bdp->node_boot_start = cstart;
@@ -99,34 +85,96 @@
return 0;
}
+
+/*
+ * Count the number of cpus on the node
+ */
+static __inline__ int
+count_cpus(int node)
+{
+ int cpu, n=0;
+
+ for (cpu=0; cpu < NR_CPUS; cpu++)
+ if (node = node_cpuid[cpu].nid)
+ n++;
+ return n;
+}
+
+
/*
- * Find space on each node for the bootmem map.
+ * Find space on each node for the bootmem map & other per-node data structures.
*
* Called by efi_memmap_walk to find boot memory on each node. Note that
* only blocks that are free are passed to this routine (currently filtered by
* free_available_memory).
*/
static int __init
-find_bootmap_space(unsigned long pstart, unsigned long length, int node)
+find_pernode_space(unsigned long start, unsigned long end, int node)
{
- unsigned long mapsize, pages, epfn;
+ unsigned long mapsize, pages, epfn, map=0, cpu, cpus;
+ unsigned long pernodesize=0, pernode;
+ unsigned long cpu_data;
+ unsigned long pstart, length;
bootmem_data_t *bdp;
+ pstart = __pa(start);
+ length = end - start;
epfn = (pstart + length) >> PAGE_SHIFT;
- bdp = &pg_data_ptr[node]->bdata[0];
+ bdp = &bdata[node];
if (pstart < bdp->node_boot_start || epfn > bdp->node_low_pfn)
return 0;
- if (!bdp->node_bootmem_map) {
+ if (!boot_pernode[node]) {
+ cpus = count_cpus(node);
+ pernodesize += PERCPU_PAGE_SIZE * cpus;
+ pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
+ pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+ pernodesize = PAGE_ALIGN(pernodesize);
+ pernode = NODEDATA_ALIGN(pstart, node);
+
+ if (pstart + length > (pernode + pernodesize)) {
+ boot_pernode[node] = pernode;
+ boot_pernodesize[node] = pernodesize;
+ memset(__va(pernode), 0, pernodesize);
+
+ cpu_data = pernode;
+ pernode += PERCPU_PAGE_SIZE * cpus;
+
+ pg_data_ptr[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ boot_node_data[node] = __va(pernode);
+ pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
+
+ pg_data_ptr[node]->bdata = &bdata[node];
+ pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
+
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ if (node = node_cpuid[cpu].nid) {
+ extern char __per_cpu_start[], __phys_per_cpu_start[];
+ memcpy((void*)cpu_data, __phys_per_cpu_start, __per_cpu_end - __per_cpu_start);
+ __per_cpu_offset[cpu] = (char*)__va(cpu_data) - __per_cpu_start;
+ cpu_data += PERCPU_PAGE_SIZE;
+ }
+ }
+ }
+ }
+
+ pernode = boot_pernode[node];
+ pernodesize = boot_pernodesize[node];
+ if (pernode && !bdp->node_bootmem_map) {
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- if (length > mapsize) {
- init_bootmem_node(
- BOOT_NODE_DATA(node),
- pstart>>PAGE_SHIFT,
- bdp->node_boot_start>>PAGE_SHIFT,
- bdp->node_low_pfn);
+
+ if (pernode - pstart > mapsize)
+ map = pstart;
+ else if (pstart + length - pernode - pernodesize > mapsize)
+ map = pernode + pernodesize;
+
+ if (map) {
+ init_bootmem_node(BOOT_NODE_DATA(node), map>>PAGE_SHIFT,
+ bdp->node_boot_start>>PAGE_SHIFT, bdp->node_low_pfn);
}
}
@@ -143,9 +191,9 @@
*
*/
static int __init
-discontig_free_bootmem_node(unsigned long pstart, unsigned long length, int node)
+discontig_free_bootmem_node(unsigned long start, unsigned long end, int node)
{
- free_bootmem_node(BOOT_NODE_DATA(node), pstart, length);
+ free_bootmem_node(BOOT_NODE_DATA(node), __pa(start), end - start);
return 0;
}
@@ -158,53 +206,50 @@
discontig_reserve_bootmem(void)
{
int node;
- unsigned long mapbase, mapsize, pages;
+ unsigned long base, size, pages;
bootmem_data_t *bdp;
for (node = 0; node < numnodes; node++) {
bdp = BOOT_NODE_DATA(node)->bdata;
pages = bdp->node_low_pfn - (bdp->node_boot_start>>PAGE_SHIFT);
- mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
- mapbase = __pa(bdp->node_bootmem_map);
- reserve_bootmem_node(BOOT_NODE_DATA(node), mapbase, mapsize);
+ size = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
+ base = __pa(bdp->node_bootmem_map);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
+
+ size = boot_pernodesize[node];
+ base = __pa(boot_pernode[node]);
+ reserve_bootmem_node(BOOT_NODE_DATA(node), base, size);
}
}
/*
- * Allocate per node tables.
- * - the pg_data structure is allocated on each node. This minimizes offnode
- * memory references
- * - the node data is allocated & initialized. Portions of this structure is read-only (after
- * boot) and contains node-local pointers to usefuls data structures located on
- * other nodes.
+ * Initialize per-node data
+ *
+ * Finish setting up the node data for this node, then copy it to the other nodes.
*
- * We also switch to using the "real" pg_data structures at this point. Earlier in boot, we
- * use a different structure. The only use for pg_data prior to the point in boot is to get
- * the pointer to the bdata for the node.
*/
static void __init
-allocate_pernode_structures(void)
+initialize_pernode_data(void)
{
- pg_data_t *pgdat=0, *new_pgdat_list=0;
- int node, mynode;
+ int cpu, node;
+
+ memcpy(boot_node_data[0]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
+ memcpy(boot_node_data[0]->node_data_ptrs, boot_node_data, sizeof(boot_node_data));
- mynode = boot_get_local_nodeid();
- for (node = numnodes - 1; node >= 0 ; node--) {
- node_data[node] = alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof (struct ia64_node_data));
- pgdat = __alloc_bootmem_node(BOOT_NODE_DATA(node), sizeof(pg_data_t), SMP_CACHE_BYTES, 0);
- pgdat->bdata = &(bdata[node][0]);
- pg_data_ptr[node] = pgdat;
- pgdat->pgdat_next = new_pgdat_list;
- new_pgdat_list = pgdat;
+ for (node=1; node < numnodes; node++) {
+ memcpy(boot_node_data[node], boot_node_data[0], sizeof(struct ia64_node_data));
+ boot_node_data[node]->node = node;
}
-
- memcpy(node_data[mynode]->pg_data_ptrs, pg_data_ptr, sizeof(pg_data_ptr));
- memcpy(node_data[mynode]->node_data_ptrs, node_data, sizeof(node_data));
- pgdat_list = new_pgdat_list;
+ for (cpu=0; cpu < NR_CPUS; cpu++) {
+ node = node_cpuid[cpu].nid;
+ per_cpu(cpu_info, cpu).node_data = boot_node_data[node];
+ per_cpu(cpu_info, cpu).nodeid = node;
+ }
}
+
/*
* Called early in boot to setup the boot memory allocator, and to
* allocate the node-local pg_data & node-directory data structures..
@@ -212,96 +257,19 @@
void __init
discontig_mem_init(void)
{
- int node;
-
if (numnodes = 0) {
printk(KERN_ERR "node info missing!\n");
numnodes = 1;
}
- for (node = 0; node < numnodes; node++) {
- pg_data_ptr[node] = (pg_data_t*) &boot_pg_data[node];
- pg_data_ptr[node]->bdata = &bdata[node][0];
- }
-
min_low_pfn = -1;
max_low_pfn = 0;
efi_memmap_walk(filter_rsvd_memory, build_maps);
- efi_memmap_walk(filter_rsvd_memory, find_bootmap_space);
+ efi_memmap_walk(filter_rsvd_memory, find_pernode_space);
efi_memmap_walk(filter_rsvd_memory, discontig_free_bootmem_node);
- discontig_reserve_bootmem();
- allocate_pernode_structures();
-}
-
-/*
- * Initialize the paging system.
- * - determine sizes of each node
- * - initialize the paging system for the node
- * - build the nodedir for the node. This contains pointers to
- * the per-bank mem_map entries.
- * - fix the page struct "virtual" pointers. These are bank specific
- * values that the paging system doesn't understand.
- * - replicate the nodedir structure to other nodes
- */
-
-void __init
-discontig_paging_init(void)
-{
- int node, mynode;
- unsigned long max_dma, zones_size[MAX_NR_ZONES];
- unsigned long kaddr, ekaddr, bid;
- struct page *page;
- bootmem_data_t *bdp;
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
- mynode = boot_get_local_nodeid();
- for (node = 0; node < numnodes; node++) {
- long pfn, startpfn;
-
- memset(zones_size, 0, sizeof(zones_size));
-
- startpfn = -1;
- bdp = BOOT_NODE_DATA(node)->bdata;
- pfn = bdp->node_boot_start >> PAGE_SHIFT;
- if (startpfn = -1)
- startpfn = pfn;
- if (pfn > max_dma)
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - pfn);
- else if (bdp->node_low_pfn < max_dma)
- zones_size[ZONE_DMA] += (bdp->node_low_pfn - pfn);
- else {
- zones_size[ZONE_DMA] += (max_dma - pfn);
- zones_size[ZONE_NORMAL] += (bdp->node_low_pfn - max_dma);
- }
-
- free_area_init_node(node, NODE_DATA(node), NULL, zones_size, startpfn, 0);
-
- page = NODE_DATA(node)->node_mem_map;
-
- bdp = BOOT_NODE_DATA(node)->bdata;
-
- kaddr = (unsigned long)__va(bdp->node_boot_start);
- ekaddr = (unsigned long)__va(bdp->node_low_pfn << PAGE_SHIFT);
- while (kaddr < ekaddr) {
- if (paddr_to_nid(__pa(kaddr)) = node) {
- bid = BANK_MEM_MAP_INDEX(kaddr);
- node_data[mynode]->node_id_map[bid] = node;
- node_data[mynode]->bank_mem_map_base[bid] = page;
- }
- kaddr += BANKSIZE;
- page += BANKSIZE/PAGE_SIZE;
- }
- }
-
- /*
- * Finish setting up the node data for this node, then copy it to the other nodes.
- */
- for (node=0; node < numnodes; node++)
- if (mynode != node) {
- memcpy(node_data[node], node_data[mynode], sizeof(struct ia64_node_data));
- node_data[node]->node = node;
- }
+ discontig_reserve_bootmem();
+ initialize_pernode_data();
}
diff -Nru a/arch/ia64/mm/init.c b/arch/ia64/mm/init.c
--- a/arch/ia64/mm/init.c Thu Jul 17 16:59:05 2003
+++ b/arch/ia64/mm/init.c Thu Jul 17 16:59:05 2003
@@ -44,7 +44,7 @@
#ifdef CONFIG_VIRTUAL_MEM_MAP
# define LARGE_GAP 0x40000000 /* Use virtual mem map if hole is > than this */
unsigned long vmalloc_end = VMALLOC_END_INIT;
- static struct page *vmem_map;
+ struct page *vmem_map;
static unsigned long num_dma_physpages;
#endif
@@ -240,7 +240,7 @@
else if (page_count(pgdat->node_mem_map + i))
shared += page_count(pgdat->node_mem_map + i) - 1;
}
- printk("\t%d pages of RAM\n", pgdat->node_spanned_pages);
+ printk("\t%ld pages of RAM\n", pgdat->node_spanned_pages);
printk("\t%d reserved pages\n", reserved);
printk("\t%d pages shared\n", shared);
printk("\t%d pages swap cached\n", cached);
@@ -397,6 +397,7 @@
{
unsigned long address, start_page, end_page;
struct page *map_start, *map_end;
+ int node;
pgd_t *pgd;
pmd_t *pmd;
pte_t *pte;
@@ -406,19 +407,20 @@
start_page = (unsigned long) map_start & PAGE_MASK;
end_page = PAGE_ALIGN((unsigned long) map_end);
+ node = paddr_to_nid(__pa(start));
for (address = start_page; address < end_page; address += PAGE_SIZE) {
pgd = pgd_offset_k(address);
if (pgd_none(*pgd))
- pgd_populate(&init_mm, pgd, alloc_bootmem_pages(PAGE_SIZE));
+ pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pmd = pmd_offset(pgd, address);
if (pmd_none(*pmd))
- pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages(PAGE_SIZE));
+ pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
pte = pte_offset_kernel(pmd, address);
if (pte_none(*pte))
- set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages(PAGE_SIZE)) >> PAGE_SHIFT,
+ set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
PAGE_KERNEL));
}
return 0;
@@ -431,6 +433,14 @@
unsigned long zone;
};
+struct memmap_count_callback_data {
+ int node;
+ unsigned long num_physpages;
+ unsigned long num_dma_physpages;
+ unsigned long min_pfn;
+ unsigned long max_pfn;
+} cdata;
+
static int
virtual_memmap_init (u64 start, u64 end, void *arg)
{
@@ -489,16 +499,6 @@
}
static int
-count_dma_pages (u64 start, u64 end, void *arg)
-{
- unsigned long *count = arg;
-
- if (end <= MAX_DMA_ADDRESS)
- *count += (end - start) >> PAGE_SHIFT;
- return 0;
-}
-
-static int
find_largest_hole (u64 start, u64 end, void *arg)
{
u64 *max_gap = arg;
@@ -514,102 +514,101 @@
}
#endif /* CONFIG_VIRTUAL_MEM_MAP */
+#define GRANULEROUNDDOWN(n) ((n) & ~(IA64_GRANULE_SIZE-1))
+#define GRANULEROUNDUP(n) (((n)+IA64_GRANULE_SIZE-1) & ~(IA64_GRANULE_SIZE-1))
+#define ORDERROUNDDOWN(n) ((n) & ~((PAGE_SIZE<<MAX_ORDER)-1))
static int
-count_pages (u64 start, u64 end, void *arg)
+count_pages (unsigned long start, unsigned long end, int node)
{
- unsigned long *count = arg;
+ start = __pa(start);
+ end = __pa(end);
- *count += (end - start) >> PAGE_SHIFT;
+ if (node = cdata.node) {
+ cdata.num_physpages += (end - start) >> PAGE_SHIFT;
+ if (start <= __pa(MAX_DMA_ADDRESS))
+ cdata.num_dma_physpages += (min(end, __pa(MAX_DMA_ADDRESS)) - start) >> PAGE_SHIFT;
+ start = GRANULEROUNDDOWN(__pa(start));
+ start = ORDERROUNDDOWN(start);
+ end = GRANULEROUNDUP(__pa(end));
+ cdata.max_pfn = max(cdata.max_pfn, end >> PAGE_SHIFT);
+ cdata.min_pfn = min(cdata.min_pfn, start >> PAGE_SHIFT);
+ }
return 0;
}
/*
* Set up the page tables.
*/
-
-#ifdef CONFIG_DISCONTIGMEM
void
paging_init (void)
{
- extern void discontig_paging_init(void);
-
- discontig_paging_init();
- efi_memmap_walk(count_pages, &num_physpages);
- zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
-}
-#else /* !CONFIG_DISCONTIGMEM */
-void
-paging_init (void)
-{
- unsigned long max_dma;
+ unsigned long max_dma_pfn;
unsigned long zones_size[MAX_NR_ZONES];
# ifdef CONFIG_VIRTUAL_MEM_MAP
unsigned long zholes_size[MAX_NR_ZONES];
unsigned long max_gap;
# endif
+ int node;
- /* initialize mem_map[] */
-
- memset(zones_size, 0, sizeof(zones_size));
-
- num_physpages = 0;
- efi_memmap_walk(count_pages, &num_physpages);
-
- max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
-
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- memset(zholes_size, 0, sizeof(zholes_size));
-
- num_dma_physpages = 0;
- efi_memmap_walk(count_dma_pages, &num_dma_physpages);
-
- if (max_low_pfn < max_dma) {
- zones_size[ZONE_DMA] = max_low_pfn;
- zholes_size[ZONE_DMA] = max_low_pfn - num_dma_physpages;
- } else {
- zones_size[ZONE_DMA] = max_dma;
- zholes_size[ZONE_DMA] = max_dma - num_dma_physpages;
- if (num_physpages > num_dma_physpages) {
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
- zholes_size[ZONE_NORMAL] = ((max_low_pfn - max_dma)
- - (num_physpages - num_dma_physpages));
- }
- }
-
+ max_dma_pfn = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
max_gap = 0;
efi_memmap_walk(find_largest_hole, (u64 *)&max_gap);
- if (max_gap < LARGE_GAP) {
- vmem_map = (struct page *) 0;
- free_area_init_node(0, &contig_page_data, NULL, zones_size, 0, zholes_size);
- mem_map = contig_page_data.node_mem_map;
- }
- else {
- unsigned long map_size;
-
- /* allocate virtual_mem_map */
- map_size = PAGE_ALIGN(max_low_pfn * sizeof(struct page));
- vmalloc_end -= map_size;
- vmem_map = (struct page *) vmalloc_end;
- efi_memmap_walk(create_mem_map_page_table, 0);
-
- free_area_init_node(0, &contig_page_data, vmem_map, zones_size, 0, zholes_size);
+ for (node = 0; node < numnodes; node++) {
+ memset(zones_size, 0, sizeof(zones_size));
+ memset(zholes_size, 0, sizeof(zholes_size));
+ memset(&cdata, 0, sizeof(cdata));
+
+ cdata.node = node;
+ cdata.min_pfn = ~0;
+
+ efi_memmap_walk(filter_rsvd_memory, count_pages);
+ num_dma_physpages += cdata.num_dma_physpages;
+ num_physpages += cdata.num_physpages;
+
+ if (cdata.min_pfn >= max_dma_pfn) {
+ /* Above the DMA zone */
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_NORMAL] = cdata.max_pfn - cdata.min_pfn - cdata.num_physpages;
+ } else if (cdata.max_pfn < max_dma_pfn) {
+ /* This block is DMAable */
+ zones_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = cdata.max_pfn - cdata.min_pfn - cdata.num_dma_physpages;
+ } else {
+ zones_size[ZONE_DMA] = max_dma_pfn - cdata.min_pfn;
+ zholes_size[ZONE_DMA] = zones_size[ZONE_DMA] - cdata.num_dma_physpages;
+ zones_size[ZONE_NORMAL] = cdata.max_pfn - max_dma_pfn;
+ zholes_size[ZONE_NORMAL] = zones_size[ZONE_NORMAL] - (cdata.num_physpages - cdata.num_dma_physpages);
+ }
- mem_map = contig_page_data.node_mem_map;
- printk("Virtual mem_map starts at 0x%p\n", mem_map);
- }
-# else /* !CONFIG_VIRTUAL_MEM_MAP */
- if (max_low_pfn < max_dma)
- zones_size[ZONE_DMA] = max_low_pfn;
- else {
- zones_size[ZONE_DMA] = max_dma;
- zones_size[ZONE_NORMAL] = max_low_pfn - max_dma;
+ if (numnodes = 1 && max_gap < LARGE_GAP) {
+ /* Just one node with no big holes... */
+ vmem_map = (struct page *)0;
+ zones_size[ZONE_DMA] += cdata.min_pfn;
+ zholes_size[ZONE_DMA] += cdata.min_pfn;
+ free_area_init_node(0, NODE_DATA(node), NODE_DATA(node)->node_mem_map,
+ zones_size, 0, zholes_size);
+ }
+ else {
+ /* allocate virtual mem_map */
+ if (node = 0) {
+ unsigned long map_size;
+ map_size = PAGE_ALIGN(max_low_pfn*sizeof(struct page));
+ vmalloc_end -= map_size;
+ vmem_map = (struct page *) vmalloc_end;
+ efi_memmap_walk(create_mem_map_page_table, 0);
+ printk("Virtual mem_map starts at 0x%p\n", vmem_map);
+#ifndef CONFIG_DISCONTIGMEM
+ mem_map = vmem_map;
+#endif
+ }
+ free_area_init_node(node, NODE_DATA(node), vmem_map + cdata.min_pfn,
+ zones_size, cdata.min_pfn, zholes_size);
+ }
}
- free_area_init(zones_size);
-# endif /* !CONFIG_VIRTUAL_MEM_MAP */
+
zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
}
-#endif /* !CONFIG_DISCONTIGMEM */
static int
count_reserved_pages (u64 start, u64 end, void *arg)
diff -Nru a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
--- a/drivers/acpi/Kconfig Thu Jul 17 16:59:05 2003
+++ b/drivers/acpi/Kconfig Thu Jul 17 16:59:05 2003
@@ -133,7 +133,7 @@
config ACPI_NUMA
bool "NUMA support" if NUMA && (IA64 && !IA64_HP_SIM || X86 && ACPI && !ACPI_HT_ONLY && !X86_64)
- default y if IA64 && IA64_SGI_SN
+ default y if IA64_GENERIC || IA64_SGI_SN2
config ACPI_ASUS
tristate "ASUS/Medion Laptop Extras"
diff -Nru a/include/asm-ia64/mmzone.h b/include/asm-ia64/mmzone.h
--- a/include/asm-ia64/mmzone.h Thu Jul 17 16:59:05 2003
+++ b/include/asm-ia64/mmzone.h Thu Jul 17 16:59:05 2003
@@ -3,7 +3,7 @@
* License. See the file "COPYING" in the main directory of this archive
* for more details.
*
- * Copyright (c) 2000 Silicon Graphics, Inc. All rights reserved.
+ * Copyright (c) 2000,2003 Silicon Graphics, Inc. All rights reserved.
* Copyright (c) 2002 NEC Corp.
* Copyright (c) 2002 Erich Focht <efocht@ess.nec.de>
* Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com>
@@ -14,150 +14,50 @@
#include <linux/config.h>
#include <linux/init.h>
-/*
- * Given a kaddr, find the base mem_map address for the start of the mem_map
- * entries for the bank containing the kaddr.
- */
-#define BANK_MEM_MAP_BASE(kaddr) local_node_data->bank_mem_map_base[BANK_MEM_MAP_INDEX(kaddr)]
-
-/*
- * Given a kaddr, this macro return the relative map number
- * within the bank.
- */
-#define BANK_MAP_NR(kaddr) (BANK_OFFSET(kaddr) >> PAGE_SHIFT)
-/*
- * Given a pte, this macro returns a pointer to the page struct for the pte.
- */
-#define pte_page(pte) virt_to_page(PAGE_OFFSET | (pte_val(pte)&_PFN_MASK))
+#ifdef CONFIG_NUMA
-/*
- * Determine if a kaddr is a valid memory address of memory that
- * actually exists.
- *
- * The check consists of 2 parts:
- * - verify that the address is a region 7 address & does not
- * contain any bits that preclude it from being a valid platform
- * memory address
- * - verify that the chunk actually exists.
- *
- * Note that IO addresses are NOT considered valid addresses.
- *
- * Note, many platforms can simply check if kaddr exceeds a specific size.
- * (However, this won't work on SGI platforms since IO space is embedded
- * within the range of valid memory addresses & nodes have holes in the
- * address range between banks).
- */
-#define kern_addr_valid(kaddr) ({long _kav=(long)(kaddr); \
- VALID_MEM_KADDR(_kav);})
-
-/*
- * Given a kaddr, return a pointer to the page struct for the page.
- * If the kaddr does not represent RAM memory that potentially exists, return
- * a pointer the page struct for max_mapnr. IO addresses will
- * return the page for max_nr. Addresses in unpopulated RAM banks may
- * return undefined results OR may panic the system.
- *
- */
-#define virt_to_page(kaddr) ({long _kvtp=(long)(kaddr); \
- (VALID_MEM_KADDR(_kvtp)) \
- ? BANK_MEM_MAP_BASE(_kvtp) + BANK_MAP_NR(_kvtp) \
- : NULL;})
+#ifdef CONFIG_IA64_DIG
/*
- * Given a page struct entry, return the physical address that the page struct represents.
- * Since IA64 has all memory in the DMA zone, the following works:
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define page_to_phys(page) __pa(page_address(page))
-
-#define node_mem_map(nid) (NODE_DATA(nid)->node_mem_map)
+#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
+#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES * 32)
-#define node_localnr(pfn, nid) ((pfn) - NODE_DATA(nid)->node_start_pfn)
-#define pfn_to_page(pfn) (struct page *)(node_mem_map(pfn_to_nid(pfn)) + node_localnr(pfn, pfn_to_nid(pfn)))
-#define pfn_to_nid(pfn) local_node_data->node_id_map[(pfn << PAGE_SHIFT) >> BANKSHIFT]
-
-#define page_to_pfn(page) (long)((page - page_zone(page)->zone_mem_map) + page_zone(page)->zone_start_pfn)
+#elif CONFIG_IA64_SGI_SN2
/*
- * pfn_valid should be made as fast as possible, and the current definition
- * is valid for machines that are NUMA, but still contiguous, which is what
- * is currently supported. A more generalised, but slower definition would
- * be something like this - mbligh:
- * ( pfn_to_pgdat(pfn) && (pfn < node_end_pfn(pfn_to_nid(pfn))) )
+ * Platform definitions for DIG platform with contiguous memory.
*/
-#define pfn_valid(pfn) (pfn < max_low_pfn)
-extern unsigned long max_low_pfn;
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of compute nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
+#elif CONFIG_IA64_GENERIC
-#ifdef CONFIG_IA64_DIG
/*
- * Platform definitions for DIG platform with contiguous memory.
+ * Platform definitions for GENERIC platform with contiguous or discontiguous memory.
*/
-#define MAX_PHYSNODE_ID 8 /* Maximum node number +1 */
-#define NR_NODES 8 /* Maximum number of nodes in SSI */
+#define MAX_PHYSNODE_ID 2048 /* Maximum node number +1 */
+#define NR_NODES 256 /* Maximum number of nodes in SSI */
+#define NR_MEMBLKS (NR_NODES)
-#define MAX_PHYS_MEMORY (1UL << 40) /* 1 TB */
-/*
- * Bank definitions.
- * Configurable settings for DIG: 512MB/bank: 16GB/node,
- * 2048MB/bank: 64GB/node,
- * 8192MB/bank: 256GB/node.
- */
-#define NR_BANKS_PER_NODE 32
-#if defined(CONFIG_IA64_NODESIZE_16GB)
-# define BANKSHIFT 29
-#elif defined(CONFIG_IA64_NODESIZE_64GB)
-# define BANKSHIFT 31
-#elif defined(CONFIG_IA64_NODESIZE_256GB)
-# define BANKSHIFT 33
#else
-# error Unsupported bank and nodesize!
+#error unknown platform
#endif
-#define BANKSIZE (1UL << BANKSHIFT)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-/*
- * VALID_MEM_KADDR returns a boolean to indicate if a kaddr is
- * potentially a valid cacheable identity mapped RAM memory address.
- * Note that the RAM may or may not actually be present!!
- */
-#define VALID_MEM_KADDR(kaddr) 1
+extern void build_cpu_to_node_map(void);
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#else /* CONFIG_NUMA */
-#elif defined(CONFIG_IA64_SGI_SN2)
-/*
- * SGI SN2 discontig definitions
- */
-#define MAX_PHYSNODE_ID 2048 /* 2048 node ids (also called nasid) */
-#define NR_NODES 128 /* Maximum number of nodes in SSI */
-#define MAX_PHYS_MEMORY (1UL << 49)
-
-#define BANKSHIFT 38
-#define NR_BANKS_PER_NODE 4
-#define SN2_NODE_SIZE (64UL*1024*1024*1024) /* 64GB per node */
-#define BANKSIZE (SN2_NODE_SIZE/NR_BANKS_PER_NODE)
-#define BANK_OFFSET(addr) ((unsigned long)(addr) & (BANKSIZE-1))
-#define NR_BANKS (NR_BANKS_PER_NODE * NR_NODES)
-#define VALID_MEM_KADDR(kaddr) 1
-
-/*
- * Given a nodeid & a bank number, find the address of the mem_map
- * entry for the first page of the bank.
- */
-#define BANK_MEM_MAP_INDEX(kaddr) \
- (((unsigned long)(kaddr) & (MAX_PHYS_MEMORY-1)) >> BANKSHIFT)
+#define NR_NODES 1
-#endif /* CONFIG_IA64_DIG */
+#endif /* CONFIG_NUMA */
#endif /* _ASM_IA64_MMZONE_H */
diff -Nru a/include/asm-ia64/nodedata.h b/include/asm-ia64/nodedata.h
--- a/include/asm-ia64/nodedata.h Thu Jul 17 16:59:05 2003
+++ b/include/asm-ia64/nodedata.h Thu Jul 17 16:59:05 2003
@@ -13,7 +13,7 @@
#ifndef _ASM_IA64_NODEDATA_H
#define _ASM_IA64_NODEDATA_H
-
+#include <asm/percpu.h>
#include <asm/mmzone.h>
/*
@@ -22,15 +22,17 @@
struct pglist_data;
struct ia64_node_data {
- short active_cpu_count;
short node;
+ short active_cpu_count;
+ /*
+ * The fields are read-only (after boot). They contain pointers
+ * to various structures located on other nodes. Ths data is
+ * replicated on each node in order to reduce off-node references.
+ */
struct pglist_data *pg_data_ptrs[NR_NODES];
- struct page *bank_mem_map_base[NR_BANKS];
struct ia64_node_data *node_data_ptrs[NR_NODES];
- short node_id_map[NR_BANKS];
};
-
/*
* Return a pointer to the node_data structure for the executing cpu.
*/
@@ -40,7 +42,8 @@
/*
* Return a pointer to the node_data structure for the specified node.
*/
-#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define node_data(node) (local_node_data->node_data_ptrs[node])
+#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
/*
* Get a pointer to the node_id/node_data for the current cpu.
@@ -48,29 +51,5 @@
*/
extern int boot_get_local_nodeid(void);
extern struct ia64_node_data *get_node_data_ptr(void);
-
-/*
- * Given a node id, return a pointer to the pg_data_t for the node.
- * The following 2 macros are similar.
- *
- * NODE_DATA - should be used in all code not related to system
- * initialization. It uses pernode data structures to minimize
- * offnode memory references. However, these structure are not
- * present during boot. This macro can be used once cpu_init
- * completes.
- *
- * BOOT_NODE_DATA
- * - should be used during system initialization
- * prior to freeing __initdata. It does not depend on the percpu
- * area being present.
- *
- * NOTE: The names of these macros are misleading but are difficult to change
- * since they are used in generic linux & on other architecures.
- */
-#define NODE_DATA(nid) (local_node_data->pg_data_ptrs[nid])
-#define BOOT_NODE_DATA(nid) boot_get_pg_data_ptr((long)(nid))
-
-struct pglist_data;
-extern struct pglist_data * __init boot_get_pg_data_ptr(long);
#endif /* _ASM_IA64_NODEDATA_H */
diff -Nru a/include/asm-ia64/numa.h b/include/asm-ia64/numa.h
--- a/include/asm-ia64/numa.h Thu Jul 17 16:59:05 2003
+++ b/include/asm-ia64/numa.h Thu Jul 17 16:59:05 2003
@@ -15,13 +15,21 @@
#ifdef CONFIG_DISCONTIGMEM
# include <asm/mmzone.h>
-# define NR_MEMBLKS (NR_BANKS)
#else
# define NR_NODES (8)
# define NR_MEMBLKS (NR_NODES * 8)
#endif
#include <linux/cache.h>
+#include <linux/threads.h>
+#include <linux/smp.h>
+
+#define NODEMASK_WORDCOUNT ((NR_NODES+(BITS_PER_LONG-1))/BITS_PER_LONG)
+
+#define NODE_MASK_NONE { [0 ... ((NR_NODES+BITS_PER_LONG-1)/BITS_PER_LONG)-1] = 0 }
+
+typedef unsigned long nodemask_t[NODEMASK_WORDCOUNT];
+
extern volatile char cpu_to_node_map[NR_CPUS] __cacheline_aligned;
extern volatile unsigned long node_to_cpu_mask[NR_NODES] __cacheline_aligned;
@@ -63,6 +71,12 @@
extern int paddr_to_nid(unsigned long paddr);
#define local_nodeid (cpu_to_node_map[smp_processor_id()])
+
+#else /* !CONFIG_NUMA */
+
+#define node_distance(from,to) 10
+#define paddr_to_nid(x) 0
+#define local_nodeid 0
#endif /* CONFIG_NUMA */
diff -Nru a/include/asm-ia64/page.h b/include/asm-ia64/page.h
--- a/include/asm-ia64/page.h Thu Jul 17 16:59:04 2003
+++ b/include/asm-ia64/page.h Thu Jul 17 16:59:04 2003
@@ -93,18 +93,26 @@
#define virt_addr_valid(kaddr) pfn_valid(__pa(kaddr) >> PAGE_SHIFT)
-#ifndef CONFIG_DISCONTIGMEM
-# ifdef CONFIG_VIRTUAL_MEM_MAP
- extern int ia64_pfn_valid (unsigned long pfn);
-# define pfn_valid(pfn) (((pfn) < max_mapnr) && ia64_pfn_valid(pfn))
-# else
-# define pfn_valid(pfn) ((pfn) < max_mapnr)
-# endif
-#define virt_to_page(kaddr) pfn_to_page(__pa(kaddr) >> PAGE_SHIFT)
-#define page_to_pfn(page) ((unsigned long) (page - mem_map))
-#define pfn_to_page(pfn) (mem_map + (pfn))
-#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
+#ifdef CONFIG_VIRTUAL_MEM_MAP
+extern int ia64_pfn_valid(unsigned long pfn);
+#else
+#define ia64_pfn_valid(pfn) (1)
+#endif
+
+extern unsigned long max_low_pfn;
+#define pfn_valid(pfn) (((pfn) < max_low_pfn) && ia64_pfn_valid(pfn))
+
+#if defined(CONFIG_VIRTUAL_MEM_MAP) && !defined(CONFIG_DISCONTIGMEM)
+#define vmem_map mem_map
+#else
+extern struct page *vmem_map;
#endif
+
+#define pfn_to_page(pfn) (vmem_map + (pfn))
+#define page_to_pfn(page) ((unsigned long) (page - vmem_map))
+
+#define virt_to_page(kaddr) (pfn_to_page(__pa(kaddr) >> PAGE_SHIFT))
+#define page_to_phys(page) (page_to_pfn(page) << PAGE_SHIFT)
typedef union ia64_va {
struct {
diff -Nru a/include/asm-ia64/pgtable.h b/include/asm-ia64/pgtable.h
--- a/include/asm-ia64/pgtable.h Thu Jul 17 16:59:04 2003
+++ b/include/asm-ia64/pgtable.h Thu Jul 17 16:59:05 2003
@@ -174,7 +174,6 @@
return (addr & (local_cpu_data->unimpl_pa_mask)) = 0;
}
-#ifndef CONFIG_DISCONTIGMEM
/*
* kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
* memory. For the return value to be meaningful, ADDR must be >@@ -190,7 +189,6 @@
*/
#define kern_addr_valid(addr) (1)
-#endif
/*
* Now come the defines and routines to manage and access the three-level
@@ -241,10 +239,8 @@
#define pte_none(pte) (!pte_val(pte))
#define pte_present(pte) (pte_val(pte) & (_PAGE_P | _PAGE_PROTNONE))
#define pte_clear(pte) (pte_val(*(pte)) = 0UL)
-#ifndef CONFIG_DISCONTIGMEM
/* pte_page() returns the "struct page *" corresponding to the PTE: */
#define pte_page(pte) virt_to_page(((pte_val(pte) & _PFN_MASK) + PAGE_OFFSET))
-#endif
#define pmd_none(pmd) (!pmd_val(pmd))
#define pmd_bad(pmd) (!ia64_phys_addr_valid(pmd_val(pmd)))
@@ -416,6 +412,7 @@
extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
extern void paging_init (void);
+extern int filter_rsvd_memory(unsigned long start, unsigned long end, void *arg);
/*
* Note: The macros below rely on the fact that MAX_SWAPFILES_SHIFT <= number of
diff -Nru a/include/asm-ia64/processor.h b/include/asm-ia64/processor.h
--- a/include/asm-ia64/processor.h Thu Jul 17 16:59:05 2003
+++ b/include/asm-ia64/processor.h Thu Jul 17 16:59:05 2003
@@ -185,6 +185,8 @@
#endif
#ifdef CONFIG_NUMA
struct ia64_node_data *node_data;
+ struct cpuinfo_ia64 *cpu_data[NR_CPUS];
+ int nodeid;
#endif
};
diff -Nru a/mm/bootmem.c b/mm/bootmem.c
--- a/mm/bootmem.c Thu Jul 17 16:59:05 2003
+++ b/mm/bootmem.c Thu Jul 17 16:59:05 2003
@@ -48,8 +48,24 @@
bootmem_data_t *bdata = pgdat->bdata;
unsigned long mapsize = ((end - start)+7)/8;
- pgdat->pgdat_next = pgdat_list;
- pgdat_list = pgdat;
+
+ /*
+ * sort pgdat_list so that the lowest one comes first,
+ * which makes alloc_bootmem_low_pages work as desired.
+ */
+ if (!pgdat_list || pgdat_list->node_start_pfn > pgdat->node_start_pfn) {
+ pgdat->pgdat_next = pgdat_list;
+ pgdat_list = pgdat;
+ } else {
+ pg_data_t *tmp = pgdat_list;
+ while (tmp->pgdat_next) {
+ if (tmp->pgdat_next->node_start_pfn > pgdat->node_start_pfn)
+ break;
+ tmp = tmp->pgdat_next;
+ }
+ pgdat->pgdat_next = tmp->pgdat_next;
+ tmp->pgdat_next = pgdat;
+ }
mapsize = (mapsize + (sizeof(long) - 1UL)) & ~(sizeof(long) - 1UL);
bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
@@ -251,7 +267,7 @@
static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
{
- struct page *page = pgdat->node_mem_map;
+ struct page *page;
bootmem_data_t *bdata = pgdat->bdata;
unsigned long i, count, total = 0;
unsigned long idx;
@@ -260,23 +276,23 @@
if (!bdata->node_bootmem_map) BUG();
count = 0;
+ page = virt_to_page(phys_to_virt(bdata->node_boot_start));
idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
map = bdata->node_bootmem_map;
for (i = 0; i < idx; ) {
unsigned long v = ~map[i / BITS_PER_LONG];
if (v) {
unsigned long m;
- for (m = 1; m && i < idx; m<<=1, page++, i++) {
+ for (m = 1; m && i < idx; m<<=1, i++) {
if (v & m) {
count++;
- ClearPageReserved(page);
- set_page_count(page, 1);
- __free_page(page);
+ ClearPageReserved(page+i);
+ set_page_count(page+i, 1);
+ __free_page(page+i);
}
}
} else {
i+=BITS_PER_LONG;
- page += BITS_PER_LONG;
}
}
total += count;
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (11 preceding siblings ...)
2003-07-18 0:16 ` Jesse Barnes
@ 2003-07-18 16:15 ` Erich Focht
2003-07-21 18:46 ` Takayoshi Kochi
13 siblings, 0 replies; 15+ messages in thread
From: Erich Focht @ 2003-07-18 16:15 UTC (permalink / raw)
To: linux-ia64
Looks good, applies cleanly on top of 2.6.0-test1 + ia64 but doesn't
boot on TX7 :-( Dies right after the elilo output lines... Maybe Tak
(CC'd) finds some time to have a look?
Regards,
Erich
On Friday 18 July 2003 02:16, Jesse Barnes wrote:
> On Thu, Jul 17, 2003 at 10:23:37AM +0200, Erich Focht wrote:
> > Maybe that should better be
> > bool "Discontiguous memory support" if (IA64_DIG && NUMA)
> >
> > Discontigmem alone doesn't make sense, yet, I think...
>
> Not for ia64 I guess, unless we make zx1 use discontig as well. Patch
> appended.
>
> Thanks,
> Jesse
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [Discontig-devel] [PATCH] another discontig patch
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
` (12 preceding siblings ...)
2003-07-18 16:15 ` Erich Focht
@ 2003-07-21 18:46 ` Takayoshi Kochi
13 siblings, 0 replies; 15+ messages in thread
From: Takayoshi Kochi @ 2003-07-21 18:46 UTC (permalink / raw)
To: linux-ia64
From: Erich Focht <efocht@hpce.nec.com>
Subject: Re: [Discontig-devel] [PATCH] another discontig patch
Date: Fri, 18 Jul 2003 18:15:44 +0200
Message-ID: <200307181815.44593.efocht@hpce.nec.com>
> Looks good, applies cleanly on top of 2.6.0-test1 + ia64 but doesn't
> boot on TX7 :-( Dies right after the elilo output lines... Maybe Tak
> (CC'd) finds some time to have a look?
I'll attend OLS this week so I think I can't have time to test it.
I'll look into it next week.
---
1st Computer Software Division, NEC Corporation
Takayoshi Kochi <kochi@hpc.bs1.fc.nec.co.jp/t-kochi@bq.jp.nec.com>
^ permalink raw reply [flat|nested] 15+ messages in thread
end of thread, other threads:[~2003-07-21 18:46 UTC | newest]
Thread overview: 15+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2003-06-21 9:06 [Discontig-devel] [PATCH] another discontig patch Christoph Hellwig
2003-06-21 14:48 ` Martin J. Bligh
2003-06-22 5:53 ` Jesse Barnes
2003-06-22 5:57 ` Jesse Barnes
2003-06-22 15:25 ` Martin J. Bligh
2003-06-23 17:20 ` William Lee Irwin III
2003-07-16 19:29 ` Jesse Barnes
2003-07-16 19:40 ` Matthew Wilcox
2003-07-16 19:51 ` Jesse Barnes
2003-07-16 19:56 ` Erich Focht
2003-07-16 22:37 ` Jesse Barnes
2003-07-17 8:23 ` Erich Focht
2003-07-18 0:16 ` Jesse Barnes
2003-07-18 16:15 ` Erich Focht
2003-07-21 18:46 ` Takayoshi Kochi
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.