Skip to contentSkip to conversation

Wrong BogoMIPS on alpha

DISCUSSION Bob McElrath 3 participants 3 messages Sep 4, 2001 – Sep 14, 2001 Alpha port

lore
Share & tools
lore
#0 lore raw

Recently the bogomips measurement has gone all haywire. Every once in a while when I boot up the bogomips measurement will be absurly high (i.e. 5 Terahertz), with obvious associated problems. This happens in recent kernels with both gcc-2.96-81 (redhat) and egcs-2.91.66. It does not happen with early 2.4 kernels or 2.2 kernels. It does not happen all the time. It never happens from a cold boot, and (almost) always happens in a warm reboot. In particular, MILO (which is a stripped-down linux 2.0 kernel that loads the real kernel) always measures the bogomips correctly immediately before loading a new kernel that measures it incorrectly.

The system is:
Alpha LX164 (21164 chip) at 600MHz
Kernels 2.4.7-2.4.9.

Cheers,
-- Bob

Bob McElrath ([email protected]) Univ. of Wisconsin at Madison, Department of Physics

#1 lore raw
Quoted 3 lines · Bob McElrath

Recently the bogomips measurement has gone all haywire. Every once in a while when I boot up the bogomips measurement will be absurly high (i.e. 5 Terahertz), with obvious associated problems.

[...]

Quoted 2 lines

The system is:
Alpha LX164 (21164 chip) at 600MHz

Your clock chip has gone bad. This happens regularly with the shitty chip they put in LX and SX systems.

r~

#2 lore raw
In reply to Bob McElrath
Quoted 10 lines · Richard Henderson
Quoted 3 lines · Bob McElrath

Recently the bogomips measurement has gone all haywire. Every once in a while when I boot up the bogomips measurement will be absurly high (i.e. 5 Terahertz), with obvious associated problems.

[...]

Quoted 2 lines

The system is:
Alpha LX164 (21164 chip) at 600MHz

Your clock chip has gone bad. This happens regularly with the shitty chip they put in LX and SX systems.

Actually it was a problem with another chip - integrated 8254 timer on intel SIO bridge. It was triggered on lx164 and reportedly on pc164 by the new clock calibration code. I've seen that myself on lx164 two or three times before my vacation, but can't reproduce it since then... Anyway, better sanity check should fix that. Also:
- 8254 naming changed to PIT (programmable interval timer) for consistency;
- reference to HZ removed for better accuracy;
- few unrelated compile and warning fixes for 2.4.10-pre9.

Ivan.

include/asm-alpha/processor.h +2 -0
@@ -7,6 +7,8 @@
 #ifndef __ASM_ALPHA_PROCESSOR_H
 #define __ASM_ALPHA_PROCESSOR_H
 
+#include <linux/personality.h>
+
 /*
  * Returns current instruction pointer ("program counter").
  */
include/asm-alpha/hw_irq.h +1 -1
@@ -5,7 +5,7 @@
 
 static inline void hw_resend_irq(struct hw_interrupt_type *h, unsigned int i) {}
 
-extern volatile unsigned long irq_err_count;
+extern unsigned long irq_err_count;
 
 #ifdef CONFIG_ALPHA_GENERIC
 #define ACTUAL_NR_IRQS	alpha_mv.nr_irqs
arch/alpha/mm/init.c +1 -1
@@ -267,7 +267,7 @@ callback_init(void * kernel_end)
 
 		/* Let vmalloc know that we've allocated some space.  */
 		console_remap_vm.flags = VM_ALLOC;
-		console_remap_vm.addr = VMALLOC_START;
+		console_remap_vm.addr = (void *)VMALLOC_START;
 		console_remap_vm.size = vaddr - VMALLOC_START;
 		vmlist = &console_remap_vm;
 	}
arch/alpha/kernel/time.c +23 -30

89 lines not on this page · raw message

arch/alpha/kernel/traps.c +1 -0
@@ -299,6 +299,7 @@ do_entIF(unsigned long type, unsigned lo
 	      case 3: /* FEN fault */
 	      case 5: /* illoc */
 	      default: /* unexpected instruction-fault type */
+		break;
 	}
 	send_sig(SIGILL, current, 1);
 }
arch/alpha/kernel/irq_alpha.c +1 -3
@@ -18,8 +18,6 @@
 unsigned long __irq_attempt[NR_IRQS];
 #endif
 
-extern unsigned long irq_err_count;
-
 /* Hack minimum IPL during interrupt processing for broken hardware.  */
 #ifdef CONFIG_ALPHA_BROKEN_IRQ_MASK
 int __min_ipl;
@@ -151,7 +149,7 @@ process_mcheck_info(unsigned long vector
 #endif
 
 	if (expected) {
-		int cpu = smp_processor_id();
+		int cpu __attribute__((unused)) = smp_processor_id();
 		mcheck_expected(cpu) = 0;
 		mcheck_taken(cpu) = 1;
 		return;