Albert Lee [Wed, 16 Nov 2005 09:06:18 +0000 (17:06 +0800)]
[PATCH] libata: honor the transfer cycle time speficied by the EIDE device
The following code segment is not functional because the transfer cycle time speficied by
the EIDE device is later overwritten by ata_timing_quantize():
/*
* If the drive is an EIDE drive, it can tell us it needs extended
* PIO/MW_DMA cycle timing.
*/
if (adev->id[ATA_ID_FIELD_VALID] & 2) { /* EIDE drive */
memset(&p, 0, sizeof(p));
(snip)
ata_timing_merge(&p, t, t, ATA_TIMING_CYCLE | ATA_TIMING_CYC8B);
<== uninitialized "t" is used here
}
/*
* Convert the timing to bus clock counts.
*/
ata_timing_quantize(s, t, T, UT); <== t is overwritten by quantized s
The patch has been submitted for ide-timing.h before:
http://marc.theaimsgroup.com/?l=linux-ide&m=110820013425454&w=2
Resubmitted for libata.
Changes:
- Minor fix to honor the following transfer cycle time speficied by the device
- id[65]: Minimum Multiword DMA transfer cycle time per word
- id[67]: Minimum PIO transfer cycle time without flow control
- id[68]: Minimum PIO transfer cycle time with IORDY
Signed-off-by: Albert Lee <albertcc@tw.ibm.com>
======= Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
[PATCH] powerpc: Make the vDSO functions set error code (#2)
The vDSO functions should have the same calling convention as a syscall.
Unfortunately, they currently don't set the cr0.so bit which is used to
indicate an error. This patch makes them clear this bit unconditionally
since all functions currently succeed. The syscall fallback done by some
of them will eventually override this if the syscall fails.
This also changes the symbol version of all vdso exports to make sure
glibc can differenciate between old and fixed calls for existing ones
like __kernel_gettimeofday.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
Paul Mackerras [Wed, 16 Nov 2005 02:38:21 +0000 (13:38 +1100)]
powerpc: Move ppc64 boot wrapper code over to arch/powerpc
This also extends the code to handle 32-bit ELF vmlinux files as well
as 64-bit ones. This is sufficient for booting on new-world 32-bit
powermacs (i.e. all recent machines).
The merge of machine types broke boot with yaboot & ARCH=ppc due to the
old code still retreiving the old-syle machine type passed in by yaboot.
This patch fixes it by translating those old numbers. Since that whole
mecanism is deprecated, this is a temporary fix until ARCH=ppc uses the
new prom_init that the merged architecture now uses for both ppc32 and
ppc64 (after 2.6.15)
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
Guido Guenther [Tue, 15 Nov 2005 06:28:05 +0000 (17:28 +1100)]
[PATCH] PowerBook 6,1: headphone not detected after suspend
ever since suspend to disk works I had the problem that headphone
(un)plugging doesn't get detected properly anymore after the first
resume.
Reloading the module worked around this ever since, however the real
cause of the problem was that after a resume the driver only got
interrupts on "unplug" not on "plug". Reactivating the headphone status
interrupt in tumbler_resume fixes this. This shouldn't cause
any trouble with software suspend, but it would be nice if somebody
could confirm this:
Signed-off-by: Guido Guenther <agx@sigxcpu.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
I discovered that in some cases (PowerMac for example) we wouldn't
properly map the PCI IO space on recent kernels. In addition, the code
for initializing PCI host bridges was scattered all over the place with
some duplication between platforms.
This patch fixes the problem and does a small cleanup by creating a
pcibios_alloc_controller() in pci_64.c that is similar to the one in
pci_32.c (just takes an additional device node argument) that takes care
of all the grunt allocation and initialisation work. It should work for
both boot time and dynamically allocated PHBs.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
My patch moving ppc64 RTC to genrtc was supposed to update all
defconfigs, but for some reason, the patch actually posted only had the
pseries one... ouch. This patch properly updates all defconfigs.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
Currently 8xx fails to boot due to endless pagefaults.
Seems the bug is exposed by the change which avoids flushing the
TLB when not necessary (in case the pte has not changed), introduced
recently:
__handle_mm_fault():
entry = pte_mkyoung(entry);
if (!pte_same(old_entry, entry)) {
ptep_set_access_flags(vma, address, pte, entry, write_access);
update_mmu_cache(vma, address, entry);
lazy_mmu_prot_update(entry);
} else {
/*
* This is needed only for protection faults but the arch code
* is not yet telling us if this is a protection fault or not.
* This still avoids useless tlb flushes for .text page faults
* with threads.
*/
if (write_access)
flush_tlb_page(vma, address);
}
The "update_mmu_cache()" call was unconditional before, which caused the TLB
to be flushed by:
if (pfn_valid(pfn)) {
struct page *page = pfn_to_page(pfn);
if (!PageReserved(page)
&& !test_bit(PG_arch_1, &page->flags)) {
if (vma->vm_mm == current->active_mm) {
#ifdef CONFIG_8xx
/* On 8xx, cache control instructions (particularly
* "dcbst" from flush_dcache_icache) fault as write
* operation if there is an unpopulated TLB entry
* for the address in question. To workaround that,
* we invalidate the TLB here, thus avoiding dcbst
* misbehaviour.
*/
_tlbie(address);
#endif
__flush_dcache_icache((void *) address);
} else
flush_dcache_icache_page(page);
set_bit(PG_arch_1, &page->flags);
}
Which worked to due to pure luck: PG_arch_1 was always unset before, but
now it isnt.
The root of the problem are the changes against the 8xx TLB handlers introduced
during v2.6. What happens is the TLBMiss handlers load the zeroed pte into
the TLB, causing the TLBError handler to be invoked (thats two TLB faults per
pagefault), which then jumps to the generic MM code to setup the pte.
The bug is that the zeroed TLB is not invalidated (the same reason
for the "dcbst" misbehaviour), resulting in infinite TLBError faults.
The "two exception" approach requires a TLB flush (to nuke the zeroed TLB)
at each PTE update for correct behaviour:
Signed-off-by: Marcelo Tosatti <marcelo.tosatti@cyclades.com> Signed-off-by: Paul Mackerras <paulus@samba.org>
This patch should fix the crashes we have been seeing on 64-bit
powerpc systems with a memory hole when sparsemem is enabled.
I'd appreciate it if people who know more about NUMA and sparsemem
than me could look over it.
There were two bugs. The first was that if NUMA was enabled but there
was no NUMA information for the machine, the setup_nonnuma() function
was adding a single region, assuming memory was contiguous. The
second was that the loops in mem_init() and show_mem() assumed that
all pages within the span of a pgdat were valid (had a valid struct
page).
I also fixed the incorrect setting of num_physpages that Mike Kravetz
pointed out.
Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
KOVACS Krisztian [Wed, 16 Nov 2005 00:47:34 +0000 (16:47 -0800)]
[NETFILTER] Free layer-3 specific protocol tables at cleanup
Although the comment around the allocation code tells us that
the layer-3 specific protocol tables will be freed when cleaning up,
they aren't. And this makes nfsim complain loudly...
Signed-off-by: KOVACS Krisztian <hidden@balabit.hu> Signed-off-by: Harald Welte <laforge@netfilter.org> Signed-off-by: David S. Miller <davem@davemloft.net>
KOVACS Krisztian [Wed, 16 Nov 2005 00:47:09 +0000 (16:47 -0800)]
[NETFILTER] Remove nf_conntrack stat proc file when cleaning up
Fix nf_conntrack statistics proc file removal. Looks like the old bug
was forward-ported from ip_conntrack. :-]
Signed-off-by: KOVACS Krisztian <hidden@balabit.hu> Signed-off-by: Harald Welte <laforge@netfilter.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Chen, Kenneth W [Tue, 15 Nov 2005 22:37:05 +0000 (14:37 -0800)]
[PATCH] ia64: cpu_idle performance bug fix
Our performance validation on 2.6.15-rc1 caught a disastrous performance
regression on ia64 with netperf (-98%) and volanomark (-58%) compares to
previous kernel version 2.6.14-git7. See the following chart (result
group 1 & 2).
This changeset broke the ia64 task resched notification. In
sched.c:resched_task(), a reschedule IPI is conditioned upon
TIF_POLLING_NRFLAG. However, the above changeset unconditionally set
the polling thread flag for idle tasks regardless whether pal_halt_light
is in use or not. As a result, resched IPI is not sent from
resched_task(). And since the default behavior on ia64 is to use
pal_halt_light, we end up delaying the rescheduling task until next
timer tick, and thus cause the performance regression.
This fixes the performance bug. I'm glad our performance suite is
turning up bad performance bug like this in time.
Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
- Fix a regression in command completion, which prevented
the restart of the DMA engine after the device throws
an error.
- Pack more hardware info into the port-reset error message.
- Promote "welcome to our timeout" message from debug msg
to normal printk.
Dave Jones [Tue, 15 Nov 2005 08:09:24 +0000 (00:09 -0800)]
[PATCH] oops-tracing: mention extended VGA
Something I've found handy countless times when users do this..
Signed-off-by: Dave Jones <davej@redhat.com> Acked-by: Randy Dunlap <rdunlap@xenotime.net> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Grant Coady [Tue, 15 Nov 2005 08:09:20 +0000 (00:09 -0800)]
[PATCH] cciss_scsi warning fix
drivers/block/cciss_scsi.c:264: warning: `print_bytes' defined but not used
drivers/block/cciss_scsi.c:298: warning: `print_cmd' defined but not used
Signed-off-by: Grant Coady <gcoady@gmail.com> Acked-by: Mike Miller <mike.miller@hp.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
NeilBrown [Tue, 15 Nov 2005 08:09:12 +0000 (00:09 -0800)]
[PATCH] md: make md threads interruptible again
Despite the fact that md threads don't need to be signalled, and won't
respond to signals anyway, we need to have an 'interruptible' wait, else
they stay in 'D' state and add to the load average.
(akpm: the signal_pending() test is unneeded - we'll fix that up in the next
round. For now, leave it there because that's how the code used to be).
Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
NeilBrown [Tue, 15 Nov 2005 08:09:11 +0000 (00:09 -0800)]
[PATCH] md: mark START_ARRAY deprecated with a date
This was marked deprecated "after 2.6" back in the 2.5 days. But now it
seems there isn't going to be any "after 2.6", and we deprecate by date
now. So set a date.
Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
NeilBrown [Tue, 15 Nov 2005 08:09:10 +0000 (00:09 -0800)]
[PATCH] knfsd: make sure nfsd doesn't hog a cpu forever
Being kernel-threads, nfsd servers don't get pre-empted (depending on
CONFIG). If there is a steady stream of NFS requests that can be served
from cache, an nfsd thread may hold on to a cpu indefinitely, which isn't
very friendly.
So it is good to have a cond_resched in there (just before looking for a
new request to serve), to make sure we play nice.
Signed-off-by: Neil Brown <neilb@suse.de> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Corey Minyard [Tue, 15 Nov 2005 08:09:07 +0000 (00:09 -0800)]
[PATCH] ipmi: bump-driver-version
Lots of good changes to the driver lately that userspace will care about
the version of the driver. Bump the version from 36.0 to 38.0 to be higher
than 37 that the 2.4 driver came out with a few weeks ago which doesn't
have all the same changes.
Signed-off-by: Matt Domsch <Matt_Domsch@dell.com> Signed-off-by: Corey Minyard <minyard@acm.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Along with above patch, another patch was posted and has been merged
(3d1675b41b02d64bd1185903ea0d25a8c0bb6dea). That patch was dependent on
the above patch and now it should also be reverted.
Cc: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Jochen Friedrich [Tue, 15 Nov 2005 05:58:18 +0000 (21:58 -0800)]
[LLC]: Fix typo
Signed-off-by: Jochen Friedrich <jochen@scram.de> Acked-by: Arnaldo Carvalho de Melo <acme@mandriva.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jochen Friedrich [Tue, 15 Nov 2005 05:57:46 +0000 (21:57 -0800)]
[LLC]: Make core block on remote busy.
Signed-off-by: Jochen Friedrich <jochen@scram.de> Acked-by: Arnaldo Carvalho de Melo <acme@mandriva.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Jochen Friedrich [Tue, 15 Nov 2005 05:57:15 +0000 (21:57 -0800)]
[LLC]: Fix TX window scaling
Signed-off-by: Jochen Friedrich <jochen@scram.de> Acked-by: Arnaldo Carvalho de Melo <acme@mandriva.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Andi Kleen [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: Remove CONFIG_CHECKING and add command line option for pagefault tracing
CONFIG_CHECKING covered some debugging code used in the early times
of the port. But it wasn't even SMP safe for quite some time
and the bugs it checked for seem to be gone.
This patch removes all the code to verify GS at kernel entry. There
haven't been any new bugs in this area for a long time.
Previously it also covered the sysctl for the page fault tracing.
That didn't make much sense because that code was unconditionally
compiled in. I made that a boot option now because it is typically
only useful at boot.
Magnus Damm [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: Make node boundaries consistent
The current x86_64 NUMA memory code is inconsequent when it comes to node
memory ranges. The exact behaviour varies depending on which config option
that is used.
setup_node_bootmem() has start and end as arguments and these are used to
calculate the size of the node like this: (end - start). This is all fine
if end is pointing to the first non-available byte. The problem is that the
current x86_64 code sometimes treats it as the last present byte and sometimes
as the first non-available byte. The result is that some configurations might
lose a page at the end of the range.
This patch tries to fix CONFIG_ACPI_NUMA, CONFIG_K8_NUMA and CONFIG_NUMA_EMU
so they all treat the end variable as the first non-available byte. This is
the same way as the single node code.
The patch is boot tested on dual x86_64 hardware with the above configurations,
but maybe the removed code is needed as some workaround?
Andi Kleen [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: Log machine checks from boot on Intel systems
The logging for boot errors was turned off because it was broken
on some AMD systems. But give Intel EM64T systems a chance because they are
supposed to be correct there.
The advantage is that there is a chance to actually log uncorrected
machine checks after the reset.
[PATCH] x86_64: Make ACPI NUMA and NUMA emulation peers of K8_NUMA in Kconfig
On x86_64 arches, there is no way to choose ACPI_NUMA without having to choose
K8_NUMA. CONFIG_K8_NUMA is not needed for Intel EM64T NUMA boxes. It also
looks odd if you have to select ACPI_NUMA from the power management menu.
This patch fixes those oddities. Patch does the following:
1. Makes NUMA a config option like other arches
2. Makes topology detection options like K8_NUMA dependent on NUMA
3. Choosing ACPI NUMA detection can be done from the standard
"Processor type and features" menu
AK: I fixed up the dependencies and changed the help texts a bit
on top of Kiran's patch.
Keeping this function does not makes sense because it's a copied (and
buggy) copy of sys_time. The only difference is that now.tv_sec (which is
a time_t, i.e. a 64-bit long) is copied (and truncated) into a int
(32-bit).
The prototype is the same (they both take a long __user *), so let's drop
this and redirect it to sys_time (and make sure it exists by defining
__ARCH_WANT_SYS_TIME).
Only disadvantage is that the sys_stime definition is also compiled (may be
fixed if needed by adding a separate __ARCH_WANT_SYS_STIME macro, and
defining it for all arch's defining __ARCH_WANT_SYS_TIME except x86_64).
[PATCH] x86_64: Set ____cacheline_maxaligned_in_smp alignment to 128 bytes
The current value was correct before the introduction of Intel EM64T support -
but now L1_CACHE_SHIFT_MAX can be less than L1_CACHE_SHIFT, which _is_ funny!
Between the few users of ____cacheline_maxaligned_in_smp, we also have (for
example) rcu_ctrlblk, and struct zone, with zone->{lru_,}lock. I.e. we have
a lot of excess cacheline bouncing on them.
No correctness issues, obviously. So this could even be merged for 2.6.14
(I'm not a fan of this idea, though).
Andi Kleen [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: Remove optimization for B stepping AMD K8
B stepping were the first shipping Opterons. memcpy/memset/copy_page/
clear_page had special optimized version for them. These are really
old and in the minority now and the difference to the generic versions
(using rep microcode) is not that big anyways. So just remove them.
TODO: figure out optimized versions for Intel Netburst based EM64T
[PATCH] x86-64/i386: Intel HT, Multi core detection fixes
Fields obtained through cpuid vector 0x1(ebx[16:23]) and
vector 0x4(eax[14:25], eax[26:31]) indicate the maximum values and might not
always be the same as what is available and what OS sees. So make sure
"siblings" and "cpu cores" values in /proc/cpuinfo reflect the values as seen
by OS instead of what cpuid instruction says. This will also fix the buggy BIOS
cases (for example where cpuid on a single core cpu says there are "2" siblings,
even when HT is disabled in the BIOS.
http://bugzilla.kernel.org/show_bug.cgi?id=4359)
Andi Kleen [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: Don't enable interrupt unconditionally in reboot path
When they were disabled before (e.g. after a panic) it's better
to keep them off, otherwise followon panics can happen from timer
interrupt handlers etc.
Drawback is that pageup in the console won't work anymore though.
Shaohua Li [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: Force correct address space size for MTRR on some 64bit Intel Xeons
They report 40bit, but only have 36bits of physical address space.
This caused problems with setting up the correct masks for MTRR.
CPUID workaround for steppings 0F33h(supporting x86) and 0F34h(supporting x86
and EM64T). Detail info can be found at:
http://download.intel.com/design/Xeon/specupdt/30240216.pdf
http://download.intel.com/design/Pentium4/specupdt/30235221.pdf
Eric Dumazet [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: Optimize NUMA node hash function
Compute the highest possible value for memnode_shift, in order to reduce
footprint of memnodemap[] to the minimum, thus making all users
(phys_to_nid(), kfree()), more cache friendly.
Andi Kleen [Sat, 5 Nov 2005 16:25:54 +0000 (17:25 +0100)]
[PATCH] x86_64: New heuristics to find out hotpluggable CPUs.
With a NR_CPUS==128 kernel with CPU hotplug enabled we would waste 4MB
on per CPU data of all possible CPUs. The reason was that HOTPLUG
always set up possible map to NR_CPUS cpus and then we need to allocate
that much (each per CPU data is roughly ~32k now)
The underlying problem is that ACPI didn't tell us how many hotplug CPUs
the platform supports. So the old code just assumed all, which would
lead to this memory wastage.
This implements some new heuristics:
- If the BIOS specified disabled CPUs in the ACPI/mptables assume they
can be enabled later (this is bending the ACPI specification a bit,
but seems like a obvious extension)
- The user can overwrite it with a new additionals_cpus=NUM option
- Otherwise use half of the available CPUs or 2, whatever is more.
Andi Kleen [Sat, 5 Nov 2005 16:25:53 +0000 (17:25 +0100)]
[PATCH] x86_64: Only use asm/sections.h to declare section symbols
Adding __initdata_* to asm-generic/sections.h
Replaces a lot of open coded externs in arch/x86_64/*
I had to change __bss_end to __bss_stop to match the other architectures.
We should zap the low mappings, as soon as possible, so that we can catch
kernel bugs more effectively. Previously early boot had NULL mapped
and didn't trap on NULL references.
This patch introduces boot_level4_pgt, which will always have low identity
addresses mapped. Druing boot, all the processors will use this as their
level4 pgt. On BP, we will switch to init_level4_pgt as soon as we enter C
code and zap the low mappings as soon as we are done with the usage of
identity low mapped addresses. On AP's we will zap the low mappings as
soon as we jump to C code.
Andi Kleen [Sat, 5 Nov 2005 16:25:53 +0000 (17:25 +0100)]
[PATCH] x86_64: Fix gcc 4 warning in aperture.c
Fix
arch/x86_64/kernel/aperture.c: In function #iommu_hole_init#:
arch/x86_64/kernel/aperture.c:199: warning: #aper_order# may be used uninitialized in this function
James Cleverdon [Sat, 5 Nov 2005 16:25:53 +0000 (17:25 +0100)]
[PATCH] i386/x86-64: Share interrupt vectors when there is a large number of interrupt sources
Here's a patch that builds on Natalie Protasevich's IRQ compression
patch and tries to work for MPS boots as well as ACPI. It is meant for
a 4-node IBM x460 NUMA box, which was dying because it had interrupt
pins with GSI numbers > NR_IRQS and thus overflowed irq_desc.
The problem is that this system has 270 GSIs (which are 1:1 mapped with
I/O APIC RTEs) and an 8-node box would have 540. This is much bigger
than NR_IRQS (224 for both i386 and x86_64). Also, there aren't enough
vectors to go around. There are about 190 usable vectors, not counting
the reserved ones and the unused vectors at 0x20 to 0x2F. So, my patch
attempts to compress the GSI range and share vectors by sharing IRQs.
Jacob Shin [Sat, 5 Nov 2005 16:25:53 +0000 (17:25 +0100)]
[PATCH] x86_64: Support for AMD specific MCE Threshold.
MC4_MISC - DRAM Errors Threshold Register realized under AMD K8 Rev F.
This register is used to count correctable and uncorrectable ECC errors that occur during DRAM read operations.
The user may interface through sysfs files in order to change the threshold configuration.
bank%d/error_count - reads current error count, write to clear.
bank%d/interrupt_enable - set/clear interrupt enable.
bank%d/threshold_limit - read/write the threshold limit.
APIC vector 0xF9 in hw_irq.h.
5 software defined bank ids in mce.h.
new apic.c function to setup threshold apic lvt.
defaults to interrupt off, count enabled, and threshold limit max.
sysfs interface created on /sys/devices/system/threshold.
Andi Kleen [Sat, 5 Nov 2005 16:25:53 +0000 (17:25 +0100)]
[PATCH] x86_64: Account mem_map in VM holes accounting
The VM needs to know about lost memory in zones to accurately
balance dirty pages. This patch accounts mem_map in there too,
which fixes a constant errror of a few percent. Also some
other misc mappings and the kernel text itself are accounted
too.
Andi Kleen [Sat, 5 Nov 2005 16:25:53 +0000 (17:25 +0100)]
[PATCH] x86_64: Add 4GB DMA32 zone
Add a new 4GB GFP_DMA32 zone between the GFP_DMA and GFP_NORMAL zones.
As a bit of historical background: when the x86-64 port
was originally designed we had some discussion if we should
use a 16MB DMA zone like i386 or a 4GB DMA zone like IA64 or
both. Both was ruled out at this point because it was in early
2.4 when VM is still quite shakey and had bad troubles even
dealing with one DMA zone. We settled on the 16MB DMA zone mainly
because we worried about older soundcards and the floppy.
But this has always caused problems since then because
device drivers had trouble getting enough DMA able memory. These days
the VM works much better and the wide use of NUMA has proven
it can deal with many zones successfully.
So this patch adds both zones.
This helps drivers who need a lot of memory below 4GB because
their hardware is not accessing more (graphic drivers - proprietary
and free ones, video frame buffer drivers, sound drivers etc.).
Previously they could only use IOMMU+16MB GFP_DMA, which
was not enough memory.
Another common problem is that hardware who has full memory
addressing for >4GB misses it for some control structures in memory
(like transmit rings or other metadata). They tended to allocate memory
in the 16MB GFP_DMA or the IOMMU/swiotlb then using pci_alloc_consistent,
but that can tie up a lot of precious 16MB GFPDMA/IOMMU/swiotlb memory
(even on AMD systems the IOMMU tends to be quite small) especially if you have
many devices. With the new zone pci_alloc_consistent can just put
this stuff into memory below 4GB which works better.
One argument was still if the zone should be 4GB or 2GB. The main
motivation for 2GB would be an unnamed not so unpopular hardware
raid controller (mostly found in older machines from a particular four letter
company) who has a strange 2GB restriction in firmware. But
that one works ok with swiotlb/IOMMU anyways, so it doesn't really
need GFP_DMA32. I chose 4GB to be compatible with IA64 and because
it seems to be the most common restriction.
The new zone is so far added only for x86-64.
For other architectures who don't set up this
new zone nothing changes. Architectures can set a compatibility
define in Kconfig CONFIG_DMA_IS_DMA32 that will define GFP_DMA32
as GFP_DMA. Otherwise it's a nop because on 32bit architectures
it's normally not needed because GFP_NORMAL (=0) is DMA able
enough.
One problem is still that GFP_DMA means different things on different
architectures. e.g. some drivers used to have #ifdef ia64 use GFP_DMA
(trusting it to be 4GB) #elif __x86_64__ (use other hacks like
the swiotlb because 16MB is not enough) ... . This was quite
ugly and is now obsolete.
These should be now converted to use GFP_DMA32 unconditionally. I haven't done
this yet. Or best only use pci_alloc_consistent/dma_alloc_coherent
which will use GFP_DMA32 transparently.