voyager: fix up for moving the generic smp call func to pointers
PATCH 1/1 voyager: fix up for moving the generic smp call func to pointers
commit e7986739a76cde5079da08809d8bbc6878387ae0
Author: Mike Travis <[email protected]> Date: Tue Dec 16 17:33:52 2008 -0800
x86 smp: modify send_IPI_mask interface to accept cpumask_t pointers
Changed all the IPI ops to take pointers, but didn't update voyager, which is now broken. Make the changes in voyager as well.
--- arch/x86/mach-voyager/voyager_smp.c | 12 ++++++------ 1 files changed, 6 insertions(+), 6 deletions(-)
arch/x86/mach-voyager/voyager_smp.c +6 -6
@@ -81,7 +81,7 @@ static void enable_local_vic_irq(unsigned int irq);static void disable_local_vic_irq (unsigned int irq);static void before_handle_vic_irq (unsigned int irq);static void after_handle_vic_irq (unsigned int irq); -static void set_vic_irq_affinity (unsigned int irq,cpumask_t mask); +static void set_vic_irq_affinity (unsigned int irq,const struct cpumask *mask);static void ack_vic_irq (unsigned int irq);static void vic_enable_cpi (void );static void do_boot_cpu (__u8 cpuid);
@@ -1597,16 +1597,16 @@ static void after_handle_vic_irq(unsigned int irq)* change the mask and then do an interrupt enable CPI to re-enable on * the selected processors */ -void set_vic_irq_affinity (unsigned int irq,cpumask_t mask) +void set_vic_irq_affinity (unsigned int irq,const struct cpumask *mask) {/* Only extended processors handle interrupts */ unsigned long real_mask;unsigned long irq_mask =1 << irq;int cpu; - real_mask =cpus_addr (mask)[0 ] & voyager_extended_vic_processors; + real_mask =cpus_addr (*mask)[0 ] & voyager_extended_vic_processors; -if (cpus_addr (mask)[0 ] ==0 ) +if (cpus_addr (*mask)[0 ] ==0 )/* can't have no CPUs to accept the interrupt -- extremely * bad things will happen */ return ;
@@ -1782,9 +1782,9 @@ void __init smp_setup_processor_id(void)x86_write_percpu (cpu_number,hard_smp_processor_id ()); } -static void voyager_send_call_func (cpumask_t callmask) +static void voyager_send_call_func (const struct cpumask *callmask) { -__u32 mask =cpus_addr (callmask)[0 ] & ~(1 <<smp_processor_id ()); +__u32 mask =cpus_addr (*callmask)[0 ] & ~(1 <<smp_processor_id ());send_CPI (mask,VIC_CALL_FUNCTION_CPI ); }
Signature · 1 line
1.5.6.6
James,
Quoted 8 lines · James Bottomley
commit
e7986739a76cde5079da08809d8bbc6878387ae0Author: Mike Travis <[email protected]> Date: Tue Dec 16 17:33:52 2008 -0800x86 smp: modify
send_IPI_maskinterface to acceptcpumask_tpointersChanged all the IPI ops to take pointers, but didn't update voyager, which is now broken. Make the changes in voyager as well.
Note: earlier this week i've rearchitectured the x86 subarchitecture and APIC code - so there's a whole lot more to be done on the Voyager side than just the cpumask conversion.
The x86/Voyager subarch is not fully ported yet (i cleaned up its Kconfig impact already) and hence it is disabled for the time being. It ought to be relatively straightforward to port it to the new code - if you are interested in doing that.
You can find the latest code in tip/master at:
http://people.redhat.com/mingo/tip.git/README
it's tentatively aimed for v2.6.30.
If you have any questions/suggestions about this code, feel free ...
Ingo
Quoted 8 lines · Ingo Molnar
James,
Quoted 4 lines · James Bottomley
commit
e7986739a76cde5079da08809d8bbc6878387ae0Author: Mike Travis <[email protected]> Date: Tue Dec 16 17:33:52 2008 -0800
Quoted 8 lines
Quoted 4 lines
x86 smp: modify
send_IPI_maskinterface to acceptcpumask_tpointersChanged all the IPI ops to take pointers, but didn't update voyager, which is now broken. Make the changes in voyager as well.
Note: earlier this week i've rearchitectured the x86 subarchitecture and APIC code - so there's a whole lot more to be done on the Voyager side than just the cpumask conversion.
Not for 2.6.29 there isn't.
Quoted 10 lines
The x86/Voyager subarch is not fully ported yet (i cleaned up its
Kconfigimpact already) and hence it is disabled for the time being. It ought to be relatively straightforward to port it to the new code - if you are interested in doing that.You can find the latest code in tip/master at:
http://people.redhat.com/mingo/tip.git/README
it's tentatively aimed for v2.6.30.
Right, and I'll fix the rest for 2.6.30. However, this patch is needed now to get 2.6.29-rc3 to boot.
James
Quoted 25 lines · James Bottomley
Quoted 8 lines · Ingo Molnar
James,
Quoted 4 lines · James Bottomley
commit
e7986739a76cde5079da08809d8bbc6878387ae0Author: Mike Travis <[email protected]> Date: Tue Dec 16 17:33:52 2008 -0800Quoted 8 lines
Quoted 4 lines
x86 smp: modify
send_IPI_maskinterface to acceptcpumask_tpointersChanged all the IPI ops to take pointers, but didn't update voyager, which is now broken. Make the changes in voyager as well.
Note: earlier this week i've rearchitectured the x86 subarchitecture and APIC code - so there's a whole lot more to be done on the Voyager side than just the cpumask conversion.
Not for 2.6.29 there isn't.
Quoted 5 lines
The x86/Voyager subarch is not fully ported yet (i cleaned up its
Kconfigimpact already) and hence it is disabled for the time being. It ought to be relatively straightforward to port it to the new code - if you are interested in doing that.
Quoted 7 lines
Quoted 5 lines
You can find the latest code in tip/master at:
http://people.redhat.com/mingo/tip.git/README
it's tentatively aimed for v2.6.30.
Right, and I'll fix the rest for 2.6.30. [...]
ok.
Quoted 1 line
[...] However, this patch is needed now to get 2.6.29-rc3 to boot.
and there's also the other ones. We can put them in now that we've got the new subarch code. Could you send a pull request with all the voyager fixes collected into one tree?
Ingo
Quoted 35 lines · Ingo Molnar
Quoted 25 lines · James Bottomley
Quoted 8 lines · Ingo Molnar
James,
Quoted 4 lines · James Bottomley
commit
e7986739a76cde5079da08809d8bbc6878387ae0Author: Mike Travis <[email protected]> Date: Tue Dec 16 17:33:52 2008 -0800Quoted 8 lines
Quoted 4 lines
x86 smp: modify
send_IPI_maskinterface to acceptcpumask_tpointersChanged all the IPI ops to take pointers, but didn't update voyager, which is now broken. Make the changes in voyager as well.
Note: earlier this week i've rearchitectured the x86 subarchitecture and APIC code - so there's a whole lot more to be done on the Voyager side than just the cpumask conversion.
Not for 2.6.29 there isn't.
Quoted 5 lines
The x86/Voyager subarch is not fully ported yet (i cleaned up its
Kconfigimpact already) and hence it is disabled for the time being. It ought to be relatively straightforward to port it to the new code - if you are interested in doing that.Quoted 7 lines
Quoted 5 lines
You can find the latest code in tip/master at:
http://people.redhat.com/mingo/tip.git/README
it's tentatively aimed for v2.6.30.
Right, and I'll fix the rest for 2.6.30. [...]
Quoted 7 lines
ok.
Quoted 1 line
[...] However, this patch is needed now to get 2.6.29-rc3 to boot.
and there's also the other ones. We can put them in now that we've got the new subarch code. Could you send a pull request with all the voyager fixes collected into one tree?
Sure, they're currently all here:
master.kernel.org:/pub/scm/linux/kernel/git/jejb/voyager-2.6.git
James
PATCH 1/1 x86/Voyager: make it build and boot
Quoted 9 lines · James Bottomley
Quoted 5 lines
Quoted 1 line
[...] However, this patch is needed now to get 2.6.29-rc3 to boot.
and there's also the other ones. We can put them in now that we've got the new subarch code. Could you send a pull request with all the voyager fixes collected into one tree?
Sure, they're currently all here:
master.kernel.org:/pub/scm/linux/kernel/git/jejb/voyager-2.6.git
I couldnt pull this tree as it had nearly as many problems as commits:
1) The tree was rebased freshly, losing history.
2) Commit febf1fe ("[VOYAGER] x86: fix voyager QIC interrupt controller") is bogus as it adds only comment but claims to be an ioremap_cache() fix. Probably a rebasing artifact.
4) Commit 569a588 is bogus as it introduces a duplicate Kconfig entry. Probably a rebasing/merging artifact.
3) The standard x86 commit log format is not:
[VOYAGER] fix cpu bootmaps
But something like:
x86/Voyager: fix cpu bootmaps
You even mixed them inconsistently:
[VOYAGER] fix cpu bootmaps
[VOYAGER] x86: fix voyager QIC interrupt controller
5) Commit 78543b1 ("[VOYAGER] x86: add ability to test for boot CPU") is ugly and possibly breaks the standard build:
+#ifdef CONFIG_ACPI
#include <asm/acpi.h>
+#endif If then such conditionals should be added to the acpi.h header itself and any knock-on build breakages addressed. The construct you used are magnets for standard kernel build breakages.
6) Neither did i like the commit logs berating x86 developers for breaking voyager - while in reality there's only a single Linux developer on the planet who _can_ test Voyager and its assumptions (and that is you).
Unless you take part in the x86 development process fully (which you havent for a long time - you always came one stuff hit upstream and sometimes in late rcs jumping months of x86 development), you cannot expect special treatment nor can you expect people to even be aware of the quirks of Voyager. Development is complex already without a constant distraction from a subarch that is not used by any developer but you.
Anyway, to move forward i consolidated the acceptable bits into a single low-impact commit and queued it up in x86/urgent - see it below. The acpi.h bit you probably still need to build Voyager but it needs reworking. Please submit it separately.
Thanks,
Ingo
-------------------->
Quoted 1 line
From
92ab78315c638515d0e81b0c70b2082f713582d9Mon Sep 17 00:00:00 2001
From: James Bottomley <[email protected]>
Date: Sat, 31 Jan 2009 17:24:43 +0100
Subject: [PATCH] x86/Voyager: make it build and boot
[
[email protected]: these fixes are a subset of changes cherry-picked from:
git://git.kernel.org:/pub/scm/linux/kernel/git/jejb/voyager-2.6.git
They fix various problems that recent x86 changes caused in the Voyager
subarchitecture: both APIC changes and cpumask changes and certain cleanups caused subarch assumptions to break.
Most of these changes are obsolete as the subarch code has been removed from the x86 development tree - but we merge them upstream to make Voyager build and boot.
]
--- arch/x86/kernel/irqinit_32.c | 12 ------------ arch/x86/mach-default/setup.c | 12 ++++++++++++ arch/x86/mach-voyager/setup.c | 12 +++++++++++- arch/x86/mach-voyager/voyager_smp.c | 25 ++++++++++++------------- 4 files changed, 35 insertions(+), 26 deletions(-)
arch/x86/kernel/irqinit_32.c +0 -12
@@ -78,15 +78,6 @@ void __init init_ISA_irqs(void) } } -/* -* IRQ2 is cascade interrupt to second interrupt controller -*/ -static struct irqaction irq2 = { - .handler = no_action, - .mask =CPU_MASK_NONE , - .name ="cascade" , -}; - DEFINE_PER_CPU(vector_irq_t, vector_irq) = { [0 ... IRQ0_VECTOR - 1] = -1, [IRQ0_VECTOR] = 0,
@@ -178,9 +169,6 @@ void __init native_init_IRQ(void)alloc_intr_gate (THERMAL_APIC_VECTOR , thermal_interrupt);#endif -if (!acpi_ioapic) -setup_irq (2 , &irq2); -/* setup after call gates are initialised (usually add in * the architecture specific gates) */
arch/x86/mach-default/setup.c +12 -0
@@ -38,6 +38,15 @@ void __init pre_intr_init_hook(void)init_ISA_irqs (); } +/* +* IRQ2 is cascade interrupt to second interrupt controller +*/ +static struct irqaction irq2 = { + .handler = no_action, + .mask =CPU_MASK_NONE , + .name ="cascade" , +}; +/** * intr_init_hook - post gate setup interrupt initialisation *
@@ -53,6 +62,9 @@ void __init intr_init_hook(void)if (x86_quirks->arch_intr_init ())return ; } +if (!acpi_ioapic) +setup_irq (2 , &irq2); + }/**
arch/x86/mach-voyager/setup.c +11 -1
@@ -33,13 +33,23 @@ void __init intr_init_hook(void)setup_irq (2 , &irq2); } -void __init pre_setup_arch_hook (void ) +static void voyager_disable_tsc (void ) {/* Voyagers run their CPUs from independent clocks, so disable * the TSC code because we can't sync them */ setup_clear_cpu_cap (X86_FEATURE_TSC ); } +void __init pre_setup_arch_hook (void ) +{ +voyager_disable_tsc (); +} + +void __init pre_time_init_hook (void ) +{ +voyager_disable_tsc (); +} +void __init trap_init_hook (void ) { }
arch/x86/mach-voyager/voyager_smp.c +12 -13
@@ -81,7 +81,7 @@ static void enable_local_vic_irq(unsigned int irq);static void disable_local_vic_irq (unsigned int irq);static void before_handle_vic_irq (unsigned int irq);static void after_handle_vic_irq (unsigned int irq); -static void set_vic_irq_affinity (unsigned int irq,cpumask_t mask); +static void set_vic_irq_affinity (unsigned int irq,const struct cpumask *mask);static void ack_vic_irq (unsigned int irq);static void vic_enable_cpi (void );static void do_boot_cpu (__u8 cpuid);
@@ -211,8 +211,6 @@ static __u32 cpu_booted_map;static cpumask_t smp_commenced_mask =CPU_MASK_NONE ;/* This is for the new dynamic CPU boot code */ -cpumask_t cpu_callin_map =CPU_MASK_NONE ; -cpumask_t cpu_callout_map =CPU_MASK_NONE ;/* The per processor IRQ masks (these are usually kept in sync) */ static __u16 vic_irq_mask[NR_CPUS] __cacheline_aligned;
@@ -378,7 +376,7 @@ void __init find_smp_config(void)cpus_addr (phys_cpu_present_map)[0 ] |=voyager_extended_cmos_read (VOYAGER_PROCESSOR_PRESENT_MASK +3 ) <<24 ; - cpu_possible_map = phys_cpu_present_map; +init_cpu_possible (&phys_cpu_present_map);printk ("VOYAGER SMP: phys_cpu_present_map = 0x%lx\n" ,cpus_addr (phys_cpu_present_map)[0 ]);/* Here we set up the VIC to enable SMP */
@@ -1599,16 +1597,16 @@ static void after_handle_vic_irq(unsigned int irq)* change the mask and then do an interrupt enable CPI to re-enable on * the selected processors */ -void set_vic_irq_affinity (unsigned int irq,cpumask_t mask) +void set_vic_irq_affinity (unsigned int irq,const struct cpumask *mask) {/* Only extended processors handle interrupts */ unsigned long real_mask;unsigned long irq_mask =1 << irq;int cpu; - real_mask =cpus_addr (mask)[0 ] & voyager_extended_vic_processors; + real_mask =cpus_addr (*mask)[0 ] & voyager_extended_vic_processors; -if (cpus_addr (mask)[0 ] ==0 ) +if (cpus_addr (*mask)[0 ] ==0 )/* can't have no CPUs to accept the interrupt -- extremely * bad things will happen */ return ;
@@ -1750,10 +1748,11 @@ static void __cpuinit voyager_smp_prepare_boot_cpu(void)init_gdt (smp_processor_id ());switch_to_new_gdt (); -cpu_set (smp_processor_id (), cpu_online_map); -cpu_set (smp_processor_id (), cpu_callout_map); -cpu_set (smp_processor_id (), cpu_possible_map); -cpu_set (smp_processor_id (), cpu_present_map); + cpu_online_map =cpumask_of_cpu (smp_processor_id ()); + cpu_callout_map =cpumask_of_cpu (smp_processor_id ()); + cpu_callin_map =CPU_MASK_NONE ; + cpu_present_map =cpumask_of_cpu (smp_processor_id ()); + }static int __cpuinit voyager_cpu_up (unsigned int cpu)
@@ -1783,9 +1782,9 @@ void __init smp_setup_processor_id(void)x86_write_percpu (cpu_number,hard_smp_processor_id ()); } -static void voyager_send_call_func (cpumask_t callmask) +static void voyager_send_call_func (const struct cpumask *callmask) { -__u32 mask =cpus_addr (callmask)[0 ] & ~(1 <<smp_processor_id ()); +__u32 mask =cpus_addr (*callmask)[0 ] & ~(1 <<smp_processor_id ());send_CPI (mask,VIC_CALL_FUNCTION_CPI ); }
Re: [PATCH] x86/Voyager: make it build and boot
Quoted 8 lines · Ingo Molnar
Quoted 6 lines · James Bottomley
Quoted 5 lines
Quoted 1 line
[...] However, this patch is needed now to get 2.6.29-rc3 to boot.
and there's also the other ones. We can put them in now that we've got the new subarch code. Could you send a pull request with all the voyager fixes collected into one tree?
Quoted 7 lines
Quoted 3 lines
Sure, they're currently all here:
master.kernel.org:/pub/scm/linux/kernel/git/jejb/voyager-2.6.git
I couldnt pull this tree as it had nearly as many problems as commits:
1) The tree was rebased freshly, losing history.
? Rebasing doesn't lose any history ... just read the manual page. It also tends to be required operating procedure constructing a patch set against a moving target
Quoted 3 lines
2) Commit febf1fe ("[VOYAGER] x86: fix voyager QIC interrupt controller") is bogus as it adds only comment but claims to be an
ioremap_cache()fix. Probably a rebasing artifact.
No it's still required. That was the problem that occurred when ioremap was switched from cached to uncached. You were just trigger happy when I explained the problem and how to fix it. The ioremap_cached needs a comment explaining why it has to be cached.
Quoted 2 lines
4) Commit 569a588 is bogus as it introduces a duplicate
Kconfigentry. Probably a rebasing/merging artifact.
Yes, good catch ... when they go in in different places rebasing doesn't always catch the updates.
Quoted 7 lines
3) The standard x86 commit log format is not:
[VOYAGER] fix cpu bootmaps
But something like:
x86/Voyager: fix cpu bootmaps
Well, you actually have to say what you want. A simple git log will tell you that [VOYAGER] has been the standard form for years.
Quoted 4 lines
You even mixed them inconsistently:
[VOYAGER] fix cpu bootmaps
[VOYAGER] x86: fix voyager QIC interrupt controller
So if I understand, you want x86 and voyager?
Quoted 6 lines
5) Commit 78543b1 ("[VOYAGER] x86: add ability to test for boot CPU") is ugly and possibly breaks the standard build:
+#ifdef CONFIG_ACPI #include <asm/acpi.h> +#endif
Really nothing should be including asm/acpi.h ... it should be including linux/acpi.h, which would resolve the issue ... but I didn't understand why it wanted the asm version in the first place, so I was reluctant to try fixing it properly in case there was large breakage in configurations I don't check. If fixmap.h wants to continue using the asm version, the ifdef hedging is about the best way: it duplicates what linux/acpi.h does, if you look (asm/acpi is gated by the same ifdef).
Quoted 3 lines
If then such conditionals should be added to the
acpi.hheader itself and any knock-on build breakages addressed. The construct you used are magnets for standard kernel build breakages.
Because you're planning to put non-ACPI symbols in asm/acpi.h? Since linux/acpi.h works in the same way, you'll break a lot of other things if you do that.
Quoted 3 lines
6) Neither did i like the commit logs berating x86 developers for breaking voyager - while in reality there's only a single Linux developer on the planet who _can_ test Voyager and its assumptions (and that is you).
They're pretty neutral to my point of view. They identify the commit causing the problem and describe the fix. This is standard changelog behaviour. If you identify problem language, I can fix it, but I think just saying this commit broke and it was fixed by doing that is about as neutral as you can get.
As I keep explaining, there are two other voyager users, one in Italy and one in Argentina. Both institutions with virtually no IT budget that happened to be stuck with an inventory of the systems. It might not be massive, but I bet it's bigger than some other architectures user bases.
Quoted 7 lines
Unless you take part in the x86 development process fully (which you havent for a long time - you always came one stuff hit upstream and sometimes in late rcs jumping months of x86 development), you cannot expect special treatment nor can you expect people to even be aware of the quirks of Voyager. Development is complex already without a constant distraction from a subarch that is not used by any developer but you.
OK, How?
For example: If you want to be involved in SCSI development, you follow linux-scsi: every patch that ever makes it into any tree goes over that
list for comment (this includes every patch I do). If I look at your x86/apic branch for instance, most of the patches appear to have gone straight from your fancy into your git tree without ever traversing a mailing list (or at least not one indexed by google).
If you develop by banging stuff into a git tree as soon as you think of it, it's a bit hard to participate in that process (I gave up telepathy in favour of contact sports a while ago). The best method is, in fact, to wait until your stuff gets into mainline and then fix it all up, which, by co-incidence is the way the voyager tree currently operates.
Quoted 4 lines
Anyway, to move forward i consolidated the acceptable bits into a single low-impact commit and queued it up in x86/urgent - see it below. The
acpi.hbit you probably still need to build Voyager but it needs reworking. Please submit it separately.
Thanks.
I really would recommend the fix I proposed: all it does, as I said, is precisely what linux/acpi.h does. The problem with trying to include linux/acpi.h in asm/fixmap.h is that fixmap is an early include and gets us into all sorts of nasty circular tangles that will be very invasive to sort out.
James
Re: [PATCH] x86/Voyager: make it build and boot
Quoted 13 lines · James Bottomley
Quoted 12 lines · Ingo Molnar
Quoted 9 lines · James Bottomley
Quoted 5 lines
Quoted 1 line
[...] However, this patch is needed now to get 2.6.29-rc3 to boot.
and there's also the other ones. We can put them in now that we've got the new subarch code. Could you send a pull request with all the voyager fixes collected into one tree?
Sure, they're currently all here:
master.kernel.org:/pub/scm/linux/kernel/git/jejb/voyager-2.6.git
Quoted 7 lines
Quoted 3 lines
I couldnt pull this tree as it had nearly as many problems as commits:
1) The tree was rebased freshly, losing history.
? Rebasing doesn't lose any history ... just read the manual page. It also tends to be required operating procedure constructing a patch set against a moving target
That is nonsense - rebasing loses a lot of history: it loses the precise place of the commit and re-anchors the commit to another place. It hides the cases where you did conflict resolution in a rebase, and it also hides the cases where Git auto-merged something. It makes it hard to trust the dates in the tree because literally, according to the timestamps, the tree you sent me was just a few hours hold.
We dont ever rebase maintainer Git trees - only in the most exceptional circumstances. Rebasing is fine for certain situations during development, but it is considered evil as a maintenance model.
Quoted 8 lines
Quoted 3 lines
2) Commit febf1fe ("[VOYAGER] x86: fix voyager QIC interrupt controller") is bogus as it adds only comment but claims to be an
ioremap_cache()fix. Probably a rebasing artifact.No it's still required. That was the problem that occurred when ioremap was switched from cached to uncached. You were just trigger happy when I explained the problem and how to fix it. The
ioremap_cachedneeds a comment explaining why it has to be cached.
Have you considered to _take a look_ at the commit i pointed out to you? The commit does this:
| From febf1fee10022544674aa01f518a7603b2231191 Mon Sep 17 00:00:00 2001 | From: James Bottomley <[email protected]> | Date: Fri, 25 Apr 2008 09:02:51 -0500 | | | [...] | + /* The QIC address is really special. QIC does cross | + * CPU interrupts via cache line states, so if there's | + * no caching on these addresses the interrupts stop | + * working. If ever the QIC stops booting but VIC is | + * fine, suspect that the ioremap here is no longer | + * producing cached memory */ | qic_addr = (unsigned long)ioremap_cache(qic_addr, 0x400);
This commit _only adds a comment_. We fixed the ioremap bug upstream eons ago (when you pointed it out), and you probably did a conflict resolution in a rebase and then perhaps forgot about this.
Quoted 5 lines
Quoted 2 lines
4) Commit 569a588 is bogus as it introduces a duplicate
Kconfigentry. Probably a rebasing/merging artifact.Yes, good catch ... when they go in in different places rebasing doesn't always catch the updates.
Yes, this is one of the typical dangers of rebasing. Besides losing true history, you also risk mishaps like that.
That is why the kernel maintainer 101 is: "never rebase". I'm surprised you never even heard about that... Does this mean that you frequently rebase the SCSI tree?
Quoted 11 lines
Quoted 7 lines
3) The standard x86 commit log format is not:
[VOYAGER] fix cpu bootmaps
But something like:
x86/Voyager: fix cpu bootmaps
Well, you actually have to say what you want. A simple git log will tell you that [VOYAGER] has been the standard form for years.
Quoted 6 lines
Quoted 4 lines
You even mixed them inconsistently:
[VOYAGER] fix cpu bootmaps
[VOYAGER] x86: fix voyager QIC interrupt controllerSo if I understand, you want x86 and voyager?
The format i suggested above would be fine.
Quoted 33 lines
Quoted 6 lines
5) Commit 78543b1 ("[VOYAGER] x86: add ability to test for boot CPU") is ugly and possibly breaks the standard build:
+#ifdef CONFIG_ACPI #include <asm/acpi.h> +#endifReally nothing should be including
asm/acpi.h... it should be includinglinux/acpi.h, which would resolve the issue ... but I didn't understand why it wanted the asm version in the first place, so I was reluctant to try fixing it properly in case there was large breakage in configurations I don't check. Iffixmap.hwants to continue using the asm version, the ifdef hedging is about the best way: it duplicates whatlinux/acpi.hdoes, if you look (asm/acpi is gated by the same ifdef).Quoted 3 lines
If then such conditionals should be added to the
acpi.hheader itself and any knock-on build breakages addressed. The construct you used are magnets for standard kernel build breakages.Because you're planning to put non-ACPI symbols in
asm/acpi.h? Sincelinux/acpi.hworks in the same way, you'll break a lot of other things if you do that.Quoted 3 lines
6) Neither did i like the commit logs berating x86 developers for breaking voyager - while in reality there's only a single Linux developer on the planet who _can_ test Voyager and its assumptions (and that is you).
They're pretty neutral to my point of view. They identify the commit causing the problem and describe the fix. This is standard changelog behaviour. If you identify problem language, I can fix it, but I think just saying this commit broke and it was fixed by doing that is about as neutral as you can get.
Quoted 4 lines
As I keep explaining, there are two other voyager users, one in Italy and one in Argentina. Both institutions with virtually no IT budget that happened to be stuck with an inventory of the systems. It might not be massive, but I bet it's bigger than some other architectures user bases.
What development kernels are they using? I never heard any bugreport from them.
Quoted 8 lines
Quoted 7 lines
Unless you take part in the x86 development process fully (which you havent for a long time - you always came one stuff hit upstream and sometimes in late rcs jumping months of x86 development), you cannot expect special treatment nor can you expect people to even be aware of the quirks of Voyager. Development is complex already without a constant distraction from a subarch that is not used by any developer but you.
Quoted 8 lines
OK, How?
For example: If you want to be involved in SCSI development, you follow
linux-scsi: every patch that ever makes it into any tree goes over that list for comment (this includes every patch I do). If I look at your x86/apic branch for instance, most of the patches appear to have gone straight from your fancy into your git tree without ever traversing a mailing list (or at least not one indexed by google).
Regarding the changes in the x86/apic branch, see this thread in your lkml folder:
x86: unify genapic code, unify subarchitectures, remove old subarchitecture code
http://lwn.net/Articles/317050/
http://thread.gmane.org/gmane.linux.kernel/786890
With 114 patches submitted to lkml. (there's ~10 more in the works that i'll submit to lkml once i'm happy with them)
All patches to the x86 tree are submitted to lkml (including my patches and that of other maintainers) and all development is done there.
So i dont get it how you can claim it credibly that you cannot take part in x86 development.
Quoted 5 lines
If you develop by banging stuff into a git tree as soon as you think of it, it's a bit hard to participate in that process (I gave up telepathy in favour of contact sports a while ago). The best method is, in fact, to wait until your stuff gets into mainline and then fix it all up, which, by co-incidence is the way the voyager tree currently operates.
Nice rant which ignores the fact that i submitted those patches to lkml.
Quoted 5 lines
Quoted 4 lines
Anyway, to move forward i consolidated the acceptable bits into a single low-impact commit and queued it up in x86/urgent - see it below. The
acpi.hbit you probably still need to build Voyager but it needs reworking. Please submit it separately.
Quoted 7 lines
Thanks.
I really would recommend the fix I proposed: all it does, as I said, is precisely what
linux/acpi.hdoes. The problem with trying to includelinux/acpi.hinasm/fixmap.his that fixmap is an early include and gets us into all sorts of nasty circular tangles that will be very invasive to sort out.
Well, the #ifdef hack you did is out of question. The clean approach would be to remove the acpi.h include from asm/fixmap.h and fix the dependency fallout. The hack you did deepens the spaghetti and makes it config-conditional as well - which is even worse than plain header dependency mess and which is a step backwards.
Ingo