ia64_mca_rendez_int_handler use of hard_smp_processor_id
arch/ia64/kernel/mca.c:ia64_mca_rendez_int_handler has
#ifdef CONFIG_SMP cpu = cpu_logical_id(hard_smp_processor_id()); #endif ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_DONE;
All the other code that runs imi_rendez_checkin does so using logical cpu numbers. Why does ia64_mca_rendez_int_handler use that convoluted expression instead of the simpler
cpu = smp_processor_id();
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Keith Owens
Keith> arch/ia64/kernel/mca.c:ia64_mca_rendez_int_handler has Keith> #ifdef CONFIG_SMP Keith> cpu = cpu_logical_id(hard_smp_processor_id()); Keith> #endif Keith> ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_DONE;
Keith> All the other code that runs imi_rendez_checkin does so using logical Keith> cpu numbers. Why does ia64_mca_rendez_int_handler use that convoluted Keith> expression instead of the simpler
Keith> cpu = smp_processor_id();
I don't know either. Perhaps the original author remembers (Jenna or Tony, perhaps?).
--david
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Keith Owens
Keith> arch/ia64/kernel/mca.c:ia64_mca_rendez_int_handler has Keith> #ifdef CONFIG_SMP Keith> cpu = cpu_logical_id(hard_smp_processor_id()); Keith> #endif Keith> ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_DONE;
Keith> All the other code that runs imi_rendez_checkin does so using logical Keith> cpu numbers. Why does ia64_mca_rendez_int_handler use that convoluted Keith> expression instead of the simpler
Keith> cpu = smp_processor_id();
David> I don't know either. Perhaps the original author remembers (Jenna or David> Tony, perhaps?).
This code predates me. I checked with Jenna and she says that it's before her time too.
<speculation>
That convoluted expression avoids use of the per-cpu mapping, but I can't see why we'd be scared to use that here, and not be paranoid elsewhere. How was smp_processor_id() implemented far back in days of 2.4.0 and before? </speculation>
-Tony
Quoted 16 lines · Tony
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Keith Owens
Keith> arch/ia64/kernel/mca.c:ia64_mca_rendez_int_handler has Keith> #ifdef
CONFIG_SMP
Keith> cpu = cpu_logical_id(hard_smp_processor_id());
Keith> #endif
Keith> ia64_mc_info.imi_rendez_checkin[cpu] = >IA64_MCA_RENDEZ_CHECKIN_DONE;Keith> All the other code that runs
imi_rendez_checkindoes so using >logical Keith> cpu numbers. Why doesia64_mca_rendez_int_handleruse that >convoluted Keith> expression instead of the simpler
Keith> cpu =smp_processor_id();David> I don't know either. Perhaps the original author remembers >(Jenna or David> Tony, perhaps?).
Quoted 8 lines
This code predates me. I checked with Jenna and she says that it's >before her time too.
<speculation>
That convoluted expression avoids use of the per-cpu mapping, but I >can't see why we'd be scared to use that here, and not be paranoid elsewhere. How >wassmp_processor_id()implemented far back in days of 2.4.0 and before? </speculation>
At one time there was confusion between which arrays used the logical cpu number in smp_processor_id() and which arrays used the hardware cpu number. This is probably a holdover from that period.
Index: 20.5/arch/ia64/kernel/mca.c
arch/ia64/kernel/mca.c Wed, 26 Mar 2003 08:14:29 +1100 kaos (linux-2.4/s/c/5_mca.c 1.1.3.2.3.1.1.1.1.3 644) renamed from arch/ia64/kernel/mca.c Wed, 11 Dec 2002 20:58:53 +1100 kaos (linux-2.4/s/c/5_mca.c 1.1.3.2.3.1.1.1.1.3 644) +1 -4
@@ -640,13 +640,10 @@ ia64_mca_wakeup_all(void) void ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *ptregs) { - int flags, cpu = 0; + int flags, cpu = smp_processor_id(); /* Mask all interrupts */ save_and_cli(flags); -#ifdef CONFIG_SMP - cpu = cpu_logical_id(hard_smp_processor_id()); -#endif ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_DONE; /* Register with the SAL monarch that the slave has * reached SAL
Index: 20.5/arch/ia64/kernel/mca.c
arch/ia64/kernel/mca.c Wed, 26 Mar 2003 08:14:29 +1100 kaos (linux-2.4/s/c/5_mca.c 1.1.3.2.3.1.1.1.1.3 644) renamed from arch/ia64/kernel/mca.c Wed, 11 Dec 2002 20:58:53 +1100 kaos (linux-2.4/s/c/5_mca.c 1.1.3.2.3.1.1.1.1.3 644) +1 -1
@@ -640,13 +640,10 @@ ia64_mca_wakeup_all(void) void ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *ptregs) { - int flags, cpu = 0; + int flags, cpu = smp_processor_id(); /* Mask all interrupts */ save_and_cli(flags);
If this is called from a preemptible context, then move the cpu cmp_processor_id() within the save_and_cli() section, or it won't be preempt safe. (thread could be migrated between getting the processor
ID and stopping interrupts.)
Peter C
I applied this for 2.4.
Quoted 18 lines · Keith Owens
Index:
20.5/arch/ia64/kernel/mca.carch/ia64/kernel/mca.c Wed, 26 Mar 2003 08:14:29 +1100 kaos (linux-2.4/s/c/5_mca.c 1.1.3.2.3.1.1.1.1.3 644) renamed from arch/ia64/kernel/mca.c Wed, 11 Dec 2002 20:58:53 +1100 kaos (linux-2.4/s/c/5_mca.c 1.1.3.2.3.1.1.1.1.3 644) +1
-4@@ -640,13 +640,10 @@ ia64_mca_wakeup_all(void) void ia64_mca_rendez_int_handler(int rendez_irq, void *arg, struct pt_regs *ptregs) { - int flags, cpu = 0; + int flags, cpu = smp_processor_id(); /* Mask all interrupts */ save_and_cli(flags); -#ifdef CONFIG_SMP - cpu = cpu_logical_id(hard_smp_processor_id()); -#endif ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_DONE; /* Register with the SAL monarch that the slave has * reached SAL