Merge branch 'parisc-4.9-4' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:
 "On parisc we were still seeing occasional random segmentation faults
  and memory corruption on SMP machines. Dave Anglin then looked again
  at the TLB related code and found two issues in the PCI DMA and
  generic TLB flush functions.

  Then, in our startup code we had some timing of the cache and TLB
  functions to calculate a threshold when to use a complete TLB/cache
  flush or just to flush a specific range. This code produced a race
  with newly started CPUs and thus lead to occasional kernel crashes
  (due to stale TLB/cache entries). The patch by Dave fixes this issue
  by flushing the local caches before starting secondary CPUs and by
  removing the race.

  The last problem fixed by this series is that we quite often suffered
  from hung tasks and self-detected stalls on the CPUs. It was somehow
  clear that this was related to the (in v4.7) newly introduced cr16
  clocksource and the own implementation of sched_clock(). I replaced
  the open-coded sched_clock() function and switched to the generic
  sched_clock() implementation which seems to have fixed this isse as
  well.

  All patches have been sucessfully tested on a variety of machines,
  including our debian buildd servers.

  All patches (beside the small pr_cont fix) are tagged for stable
  releases"

* 'parisc-4.9-4' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Also flush data TLB in flush_icache_page_asm
  parisc: Fix race in pci-dma.c
  parisc: Switch to generic sched_clock implementation
  parisc: Fix races in parisc_setup_cache_timing()
  parisc: Fix printk continuations in system detection

arch/parisc/Kconfig

@@ -34,7 +34,9 @@ config PARISC
 	select HAVE_ARCH_HASH
 	select HAVE_ARCH_SECCOMP_FILTER
 	select HAVE_ARCH_TRACEHOOK
-	select HAVE_UNSTABLE_SCHED_CLOCK if (SMP || !64BIT)
+	select GENERIC_SCHED_CLOCK
+	select HAVE_UNSTABLE_SCHED_CLOCK if SMP
+	select GENERIC_CLOCKEVENTS
 	select ARCH_NO_COHERENT_DMA_MMAP
 	select CPU_NO_EFFICIENT_FFS
 

arch/parisc/kernel/cache.c

@@ -369,6 +369,7 @@ void __init parisc_setup_cache_timing(void)
 {
 	unsigned long rangetime, alltime;
 	unsigned long size, start;
+	unsigned long threshold;
 
 	alltime = mfctl(16);
 	flush_data_cache();
@@ -382,17 +383,12 @@ void __init parisc_setup_cache_timing(void)
 	printk(KERN_DEBUG "Whole cache flush %lu cycles, flushing %lu bytes %lu cycles\n",
 		alltime, size, rangetime);
 
-	/* Racy, but if we see an intermediate value, it's ok too... */
-	parisc_cache_flush_threshold = size * alltime / rangetime;
-
-	parisc_cache_flush_threshold = L1_CACHE_ALIGN(parisc_cache_flush_threshold);
-	if (!parisc_cache_flush_threshold)
-		parisc_cache_flush_threshold = FLUSH_THRESHOLD;
-
-	if (parisc_cache_flush_threshold > cache_info.dc_size)
-		parisc_cache_flush_threshold = cache_info.dc_size;
-
-	printk(KERN_INFO "Setting cache flush threshold to %lu kB\n",
+	threshold = L1_CACHE_ALIGN(size * alltime / rangetime);
+	if (threshold > cache_info.dc_size)
+		threshold = cache_info.dc_size;
+	if (threshold)
+		parisc_cache_flush_threshold = threshold;
+	printk(KERN_INFO "Cache flush threshold set to %lu KiB\n",
 		parisc_cache_flush_threshold/1024);
 
 	/* calculate TLB flush threshold */
@@ -401,7 +397,7 @@ void __init parisc_setup_cache_timing(void)
 	flush_tlb_all();
 	alltime = mfctl(16) - alltime;
 
-	size = PAGE_SIZE;
+	size = 0;
 	start = (unsigned long) _text;
 	rangetime = mfctl(16);
 	while (start < (unsigned long) _end) {
@@ -414,13 +410,10 @@ void __init parisc_setup_cache_timing(void)
 	printk(KERN_DEBUG "Whole TLB flush %lu cycles, flushing %lu bytes %lu cycles\n",
 		alltime, size, rangetime);
 
-	parisc_tlb_flush_threshold = size * alltime / rangetime;
-	parisc_tlb_flush_threshold *= num_online_cpus();
-	parisc_tlb_flush_threshold = PAGE_ALIGN(parisc_tlb_flush_threshold);
-	if (!parisc_tlb_flush_threshold)
-		parisc_tlb_flush_threshold = FLUSH_TLB_THRESHOLD;
-
-	printk(KERN_INFO "Setting TLB flush threshold to %lu kB\n",
+	threshold = PAGE_ALIGN(num_online_cpus() * size * alltime / rangetime);
+	if (threshold)
+		parisc_tlb_flush_threshold = threshold;
+	printk(KERN_INFO "TLB flush threshold set to %lu KiB\n",
 		parisc_tlb_flush_threshold/1024);
 }
 

arch/parisc/kernel/inventory.c

@@ -58,7 +58,7 @@ void __init setup_pdc(void)
 	status = pdc_system_map_find_mods(&module_result, &module_path, 0);
 	if (status == PDC_OK) {
 		pdc_type = PDC_TYPE_SYSTEM_MAP;
-		printk("System Map.\n");
+		pr_cont("System Map.\n");
 		return;
 	}
 
@@ -77,7 +77,7 @@ void __init setup_pdc(void)
 	status = pdc_pat_cell_get_number(&cell_info);
 	if (status == PDC_OK) {
 		pdc_type = PDC_TYPE_PAT;
-		printk("64 bit PAT.\n");
+		pr_cont("64 bit PAT.\n");
 		return;
 	}
 #endif
@@ -97,12 +97,12 @@ void __init setup_pdc(void)
 	case 0xC:		/* 715/64, at least */
 
 		pdc_type = PDC_TYPE_SNAKE;
-		printk("Snake.\n");
+		pr_cont("Snake.\n");
 		return;
 
 	default:		/* Everything else */
 
-		printk("Unsupported.\n");
+		pr_cont("Unsupported.\n");
 		panic("If this is a 64-bit machine, please try a 64-bit kernel.\n");
 	}
 }

arch/parisc/kernel/pacache.S

@@ -96,7 +96,7 @@ fitmanyloop:					/* Loop if LOOP >= 2 */
 
 fitmanymiddle:					/* Loop if LOOP >= 2 */
 	addib,COND(>)		-1, %r31, fitmanymiddle	/* Adjusted inner loop decr */
-	pitlbe		0(%sr1, %r28)
+	pitlbe		%r0(%sr1, %r28)
 	pitlbe,m	%arg1(%sr1, %r28)	/* Last pitlbe and addr adjust */
 	addib,COND(>)		-1, %r29, fitmanymiddle	/* Middle loop decr */
 	copy		%arg3, %r31		/* Re-init inner loop count */
@@ -139,7 +139,7 @@ fdtmanyloop:					/* Loop if LOOP >= 2 */
 
 fdtmanymiddle:					/* Loop if LOOP >= 2 */
 	addib,COND(>)		-1, %r31, fdtmanymiddle	/* Adjusted inner loop decr */
-	pdtlbe		0(%sr1, %r28)
+	pdtlbe		%r0(%sr1, %r28)
 	pdtlbe,m	%arg1(%sr1, %r28)	/* Last pdtlbe and addr adjust */
 	addib,COND(>)		-1, %r29, fdtmanymiddle	/* Middle loop decr */
 	copy		%arg3, %r31		/* Re-init inner loop count */
@@ -626,12 +626,12 @@ ENTRY_CFI(copy_user_page_asm)
 	/* Purge any old translations */
 
 #ifdef CONFIG_PA20
-	pdtlb,l		0(%r28)
-	pdtlb,l		0(%r29)
+	pdtlb,l		%r0(%r28)
+	pdtlb,l		%r0(%r29)
 #else
 	tlb_lock	%r20,%r21,%r22
-	pdtlb		0(%r28)
-	pdtlb		0(%r29)
+	pdtlb		%r0(%r28)
+	pdtlb		%r0(%r29)
 	tlb_unlock	%r20,%r21,%r22
 #endif
 
@@ -774,10 +774,10 @@ ENTRY_CFI(clear_user_page_asm)
 	/* Purge any old translation */
 
 #ifdef CONFIG_PA20
-	pdtlb,l		0(%r28)
+	pdtlb,l		%r0(%r28)
 #else
 	tlb_lock	%r20,%r21,%r22
-	pdtlb		0(%r28)
+	pdtlb		%r0(%r28)
 	tlb_unlock	%r20,%r21,%r22
 #endif
 
@@ -858,10 +858,10 @@ ENTRY_CFI(flush_dcache_page_asm)
 	/* Purge any old translation */
 
 #ifdef CONFIG_PA20
-	pdtlb,l		0(%r28)
+	pdtlb,l		%r0(%r28)
 #else
 	tlb_lock	%r20,%r21,%r22
-	pdtlb		0(%r28)
+	pdtlb		%r0(%r28)
 	tlb_unlock	%r20,%r21,%r22
 #endif
 
@@ -898,10 +898,10 @@ ENTRY_CFI(flush_dcache_page_asm)
 	sync
 
 #ifdef CONFIG_PA20
-	pdtlb,l		0(%r25)
+	pdtlb,l		%r0(%r25)
 #else
 	tlb_lock	%r20,%r21,%r22
-	pdtlb		0(%r25)
+	pdtlb		%r0(%r25)
 	tlb_unlock	%r20,%r21,%r22
 #endif
 
@@ -931,13 +931,18 @@ ENTRY_CFI(flush_icache_page_asm)
 	depwi		0, 31,PAGE_SHIFT, %r28	/* Clear any offset bits */
 #endif
 
-	/* Purge any old translation */
+	/* Purge any old translation.  Note that the FIC instruction
+	 * may use either the instruction or data TLB.  Given that we
+	 * have a flat address space, it's not clear which TLB will be
+	 * used.  So, we purge both entries.  */
 
 #ifdef CONFIG_PA20
+	pdtlb,l		%r0(%r28)
 	pitlb,l         %r0(%sr4,%r28)
 #else
 	tlb_lock        %r20,%r21,%r22
-	pitlb           (%sr4,%r28)
+	pdtlb		%r0(%r28)
+	pitlb           %r0(%sr4,%r28)
 	tlb_unlock      %r20,%r21,%r22
 #endif
 
@@ -976,10 +981,12 @@ ENTRY_CFI(flush_icache_page_asm)
 	sync
 
 #ifdef CONFIG_PA20
+	pdtlb,l		%r0(%r28)
 	pitlb,l         %r0(%sr4,%r25)
 #else
 	tlb_lock        %r20,%r21,%r22
-	pitlb           (%sr4,%r25)
+	pdtlb		%r0(%r28)
+	pitlb           %r0(%sr4,%r25)
 	tlb_unlock      %r20,%r21,%r22
 #endif
 

arch/parisc/kernel/pci-dma.c

@@ -95,8 +95,8 @@ static inline int map_pte_uncached(pte_t * pte,
 
 		if (!pte_none(*pte))
 			printk(KERN_ERR "map_pte_uncached: page already exists\n");
-		set_pte(pte, __mk_pte(*paddr_ptr, PAGE_KERNEL_UNC));
 		purge_tlb_start(flags);
+		set_pte(pte, __mk_pte(*paddr_ptr, PAGE_KERNEL_UNC));
 		pdtlb_kernel(orig_vaddr);
 		purge_tlb_end(flags);
 		vaddr += PAGE_SIZE;

arch/parisc/kernel/setup.c

@@ -334,6 +334,10 @@ static int __init parisc_init(void)
 	/* tell PDC we're Linux. Nevermind failure. */
 	pdc_stable_write(0x40, &osid, sizeof(osid));
 	
+	/* start with known state */
+	flush_cache_all_local();
+	flush_tlb_all_local(NULL);
+
 	processor_init();
 #ifdef CONFIG_SMP
 	pr_info("CPU(s): %d out of %d %s at %d.%06d MHz online\n",

arch/parisc/kernel/time.c

@@ -14,6 +14,7 @@
 #include <linux/module.h>
 #include <linux/rtc.h>
 #include <linux/sched.h>
+#include <linux/sched_clock.h>
 #include <linux/kernel.h>
 #include <linux/param.h>
 #include <linux/string.h>
@@ -39,18 +40,6 @@
 
 static unsigned long clocktick __read_mostly;	/* timer cycles per tick */
 
-#ifndef CONFIG_64BIT
-/*
- * The processor-internal cycle counter (Control Register 16) is used as time
- * source for the sched_clock() function.  This register is 64bit wide on a
- * 64-bit kernel and 32bit on a 32-bit kernel. Since sched_clock() always
- * requires a 64bit counter we emulate on the 32-bit kernel the higher 32bits
- * with a per-cpu variable which we increase every time the counter
- * wraps-around (which happens every ~4 secounds).
- */
-static DEFINE_PER_CPU(unsigned long, cr16_high_32_bits);
-#endif
-
 /*
  * We keep time on PA-RISC Linux by using the Interval Timer which is
  * a pair of registers; one is read-only and one is write-only; both
@@ -121,12 +110,6 @@ irqreturn_t __irq_entry timer_interrupt(int irq, void *dev_id)
 	 */
 	mtctl(next_tick, 16);
 
-#if !defined(CONFIG_64BIT)
-	/* check for overflow on a 32bit kernel (every ~4 seconds). */
-	if (unlikely(next_tick < now))
-		this_cpu_inc(cr16_high_32_bits);
-#endif
-
 	/* Skip one clocktick on purpose if we missed next_tick.
 	 * The new CR16 must be "later" than current CR16 otherwise
 	 * itimer would not fire until CR16 wrapped - e.g 4 seconds
@@ -208,7 +191,7 @@ EXPORT_SYMBOL(profile_pc);
 
 /* clock source code */
 
-static cycle_t read_cr16(struct clocksource *cs)
+static cycle_t notrace read_cr16(struct clocksource *cs)
 {
 	return get_cycles();
 }
@@ -287,26 +270,9 @@ void read_persistent_clock(struct timespec *ts)
 }
 
 
-/*
- * sched_clock() framework
- */
-
-static u32 cyc2ns_mul __read_mostly;
-static u32 cyc2ns_shift __read_mostly;
-
-u64 sched_clock(void)
+static u64 notrace read_cr16_sched_clock(void)
 {
-	u64 now;
-
-	/* Get current cycle counter (Control Register 16). */
-#ifdef CONFIG_64BIT
-	now = mfctl(16);
-#else
-	now = mfctl(16) + (((u64) this_cpu_read(cr16_high_32_bits)) << 32);
-#endif
-
-	/* return the value in ns (cycles_2_ns) */
-	return mul_u64_u32_shr(now, cyc2ns_mul, cyc2ns_shift);
+	return get_cycles();
 }
 
 
@@ -316,17 +282,16 @@ u64 sched_clock(void)
 
 void __init time_init(void)
 {
-	unsigned long current_cr16_khz;
+	unsigned long cr16_hz;
 
-	current_cr16_khz = PAGE0->mem_10msec/10;  /* kHz */
 	clocktick = (100 * PAGE0->mem_10msec) / HZ;
-
-	/* calculate mult/shift values for cr16 */
-	clocks_calc_mult_shift(&cyc2ns_mul, &cyc2ns_shift, current_cr16_khz,
-				NSEC_PER_MSEC, 0);
-
 	start_cpu_itimer();	/* get CPU 0 started */
 
+	cr16_hz = 100 * PAGE0->mem_10msec;  /* Hz */
+
 	/* register at clocksource framework */
-	clocksource_register_khz(&clocksource_cr16, current_cr16_khz);
+	clocksource_register_hz(&clocksource_cr16, cr16_hz);
+
+	/* register as sched_clock source */
+	sched_clock_register(read_cr16_sched_clock, BITS_PER_LONG, cr16_hz);
 }