linux/arch/x86/kernel/hpet.c

1338 lines
31 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
#include <linux/clockchips.h>
#include <linux/interrupt.h>
#include <linux/export.h>
#include <linux/delay.h>
#include <linux/hpet.h>
#include <linux/cpu.h>
#include <linux/irq.h>
#include <asm/hpet.h>
#include <asm/time.h>
#undef pr_fmt
#define pr_fmt(fmt) "hpet: " fmt
struct hpet_dev {
struct clock_event_device evt;
unsigned int num;
int cpu;
unsigned int irq;
unsigned int flags;
char name[10];
};
enum hpet_mode {
HPET_MODE_UNUSED,
HPET_MODE_LEGACY,
HPET_MODE_CLOCKEVT,
HPET_MODE_DEVICE,
};
struct hpet_channel {
unsigned int num;
unsigned int irq;
enum hpet_mode mode;
unsigned int boot_cfg;
};
struct hpet_base {
unsigned int nr_channels;
unsigned int boot_cfg;
struct hpet_channel *channels;
};
#define HPET_MASK CLOCKSOURCE_MASK(32)
#define HPET_DEV_USED_BIT 2
#define HPET_DEV_USED (1 << HPET_DEV_USED_BIT)
#define HPET_DEV_VALID 0x8
#define HPET_DEV_FSB_CAP 0x1000
#define HPET_DEV_PERI_CAP 0x2000
#define HPET_MIN_CYCLES 128
#define HPET_MIN_PROG_DELTA (HPET_MIN_CYCLES + (HPET_MIN_CYCLES >> 1))
/*
* HPET address is set in acpi/boot.c, when an ACPI entry exists
*/
unsigned long hpet_address;
u8 hpet_blockid; /* OS timer block num */
bool hpet_msi_disable;
#ifdef CONFIG_PCI_MSI
static struct hpet_dev *hpet_devs;
static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
static struct irq_domain *hpet_domain;
#endif
static void __iomem *hpet_virt_address;
static struct hpet_base hpet_base;
static bool hpet_legacy_int_enabled;
static unsigned long hpet_freq;
bool boot_hpet_disable;
bool hpet_force_user;
static bool hpet_verbose;
static struct clock_event_device hpet_clockevent;
static inline
struct hpet_dev *clockevent_to_channel(struct clock_event_device *evt)
{
return container_of(evt, struct hpet_dev, evt);
}
inline unsigned int hpet_readl(unsigned int a)
{
return readl(hpet_virt_address + a);
}
static inline void hpet_writel(unsigned int d, unsigned int a)
{
writel(d, hpet_virt_address + a);
}
static inline void hpet_set_mapping(void)
{
hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
}
static inline void hpet_clear_mapping(void)
{
iounmap(hpet_virt_address);
hpet_virt_address = NULL;
}
/*
* HPET command line enable / disable
*/
static int __init hpet_setup(char *str)
{
while (str) {
char *next = strchr(str, ',');
if (next)
*next++ = 0;
if (!strncmp("disable", str, 7))
boot_hpet_disable = true;
if (!strncmp("force", str, 5))
hpet_force_user = true;
if (!strncmp("verbose", str, 7))
hpet_verbose = true;
str = next;
}
return 1;
}
__setup("hpet=", hpet_setup);
static int __init disable_hpet(char *str)
{
boot_hpet_disable = true;
return 1;
}
__setup("nohpet", disable_hpet);
static inline int is_hpet_capable(void)
{
return !boot_hpet_disable && hpet_address;
}
/**
* is_hpet_enabled - Check whether the legacy HPET timer interrupt is enabled
*/
int is_hpet_enabled(void)
{
return is_hpet_capable() && hpet_legacy_int_enabled;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(is_hpet_enabled);
static void _hpet_print_config(const char *function, int line)
{
u32 i, id, period, cfg, status, channels, l, h;
pr_info("%s(%d):\n", function, line);
id = hpet_readl(HPET_ID);
period = hpet_readl(HPET_PERIOD);
pr_info("ID: 0x%x, PERIOD: 0x%x\n", id, period);
cfg = hpet_readl(HPET_CFG);
status = hpet_readl(HPET_STATUS);
pr_info("CFG: 0x%x, STATUS: 0x%x\n", cfg, status);
l = hpet_readl(HPET_COUNTER);
h = hpet_readl(HPET_COUNTER+4);
pr_info("COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
channels = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
for (i = 0; i < channels; i++) {
l = hpet_readl(HPET_Tn_CFG(i));
h = hpet_readl(HPET_Tn_CFG(i)+4);
pr_info("T%d: CFG_l: 0x%x, CFG_h: 0x%x\n", i, l, h);
l = hpet_readl(HPET_Tn_CMP(i));
h = hpet_readl(HPET_Tn_CMP(i)+4);
pr_info("T%d: CMP_l: 0x%x, CMP_h: 0x%x\n", i, l, h);
l = hpet_readl(HPET_Tn_ROUTE(i));
h = hpet_readl(HPET_Tn_ROUTE(i)+4);
pr_info("T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n", i, l, h);
}
}
#define hpet_print_config() \
do { \
if (hpet_verbose) \
_hpet_print_config(__func__, __LINE__); \
} while (0)
/*
* When the HPET driver (/dev/hpet) is enabled, we need to reserve
* timer 0 and timer 1 in case of RTC emulation.
*/
#ifdef CONFIG_HPET
x86: using HPET in MSI mode and setting up per CPU HPET timers, fix On Sat, Sep 06, 2008 at 06:03:53AM -0700, Ingo Molnar wrote: > > it crashes two testsystems, the fault on a NULL pointer in hpet init, > with: > > initcall print_all_ICs+0x0/0x520 returned 0 after 26 msecs > calling hpet_late_init+0x0/0x1c0 > BUG: unable to handle kernel NULL pointer dereference at 000000000000008c > IP: [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > PGD 0 > Oops: 0000 [1] SMP > CPU 0 > Modules linked in: > Pid: 1, comm: swapper Not tainted 2.6.27-rc5 #29725 > RIP: 0010:[<ffffffff80d228be>] [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > RSP: 0018:ffff88003fa07dd0 EFLAGS: 00010246 > RAX: 0000000000000000 RBX: 0000000000000003 RCX: 0000000000000000 > RDX: ffffc20000000160 RSI: 0000000000000000 RDI: 0000000000000003 > RBP: ffff88003fa07e90 R08: 0000000000000000 R09: ffff88003fa07dd0 > R10: 0000000000000001 R11: 0000000000000000 R12: ffff88003fa07dd0 > R13: 0000000000000002 R14: ffffc20000000000 R15: 000000006f57e511 > FS: 0000000000000000(0000) GS:ffffffff80cf6a80(0000) knlGS:0000000000000000 > CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b > CR2: 000000000000008c CR3: 0000000000201000 CR4: 00000000000006e0 > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 > Process swapper (pid: 1, threadinfo ffff88003fa06000, task ffff88003fa08000) > Stack: 00000000fed00000 ffffc20000000000 0000000100000003 0000000800000002 > 0000000000000000 0000000000000000 0000000000000000 0000000000000000 > 0000000000000000 0000000000000000 0000000000000000 0000000000000000 > Call Trace: > [<ffffffff80d227c0>] ? hpet_late_init+0x0/0x1c0 > [<ffffffff80209045>] do_one_initcall+0x45/0x190 > [<ffffffff80296f39>] ? register_irq_proc+0x19/0xe0 > [<ffffffff80d0d140>] ? early_idt_handler+0x0/0x73 > [<ffffffff80d0dabc>] kernel_init+0x14c/0x1b0 > [<ffffffff80942ac1>] ? trace_hardirqs_on_thunk+0x3a/0x3f > [<ffffffff8020dbd9>] child_rip+0xa/0x11 > [<ffffffff8020ceee>] ? restore_args+0x0/0x30 > [<ffffffff80d0d970>] ? kernel_init+0x0/0x1b0 > [<ffffffff8020dbcf>] ? child_rip+0x0/0x11 > Code: 20 48 83 c1 01 48 39 f1 75 e3 44 89 e8 4c 8b 05 29 29 22 00 31 f6 48 8d 78 01 66 66 90 89 f0 48 8d 04 80 48 c1 e0 05 4a 8d 0c 00 <f6> 81 8c 00 00 00 08 74 26 8b 81 80 00 00 00 8b 91 88 00 00 00 > RIP [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > RSP <ffff88003fa07dd0> > CR2: 000000000000008c > Kernel panic - not syncing: Fatal exception There was one code path, with CONFIG_PCI_MSI disabled, where we were accessing hpet_devs without initialization. That resulted in the above crash. The change below adds a check for hpet_devs. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Shaohua Li <shaohua.li@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-08 17:18:40 +00:00
static void hpet_reserve_msi_timers(struct hpet_data *hd);
x86: using HPET in MSI mode and setting up per CPU HPET timers, fix On Sat, Sep 06, 2008 at 06:03:53AM -0700, Ingo Molnar wrote: > > it crashes two testsystems, the fault on a NULL pointer in hpet init, > with: > > initcall print_all_ICs+0x0/0x520 returned 0 after 26 msecs > calling hpet_late_init+0x0/0x1c0 > BUG: unable to handle kernel NULL pointer dereference at 000000000000008c > IP: [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > PGD 0 > Oops: 0000 [1] SMP > CPU 0 > Modules linked in: > Pid: 1, comm: swapper Not tainted 2.6.27-rc5 #29725 > RIP: 0010:[<ffffffff80d228be>] [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > RSP: 0018:ffff88003fa07dd0 EFLAGS: 00010246 > RAX: 0000000000000000 RBX: 0000000000000003 RCX: 0000000000000000 > RDX: ffffc20000000160 RSI: 0000000000000000 RDI: 0000000000000003 > RBP: ffff88003fa07e90 R08: 0000000000000000 R09: ffff88003fa07dd0 > R10: 0000000000000001 R11: 0000000000000000 R12: ffff88003fa07dd0 > R13: 0000000000000002 R14: ffffc20000000000 R15: 000000006f57e511 > FS: 0000000000000000(0000) GS:ffffffff80cf6a80(0000) knlGS:0000000000000000 > CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b > CR2: 000000000000008c CR3: 0000000000201000 CR4: 00000000000006e0 > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 > Process swapper (pid: 1, threadinfo ffff88003fa06000, task ffff88003fa08000) > Stack: 00000000fed00000 ffffc20000000000 0000000100000003 0000000800000002 > 0000000000000000 0000000000000000 0000000000000000 0000000000000000 > 0000000000000000 0000000000000000 0000000000000000 0000000000000000 > Call Trace: > [<ffffffff80d227c0>] ? hpet_late_init+0x0/0x1c0 > [<ffffffff80209045>] do_one_initcall+0x45/0x190 > [<ffffffff80296f39>] ? register_irq_proc+0x19/0xe0 > [<ffffffff80d0d140>] ? early_idt_handler+0x0/0x73 > [<ffffffff80d0dabc>] kernel_init+0x14c/0x1b0 > [<ffffffff80942ac1>] ? trace_hardirqs_on_thunk+0x3a/0x3f > [<ffffffff8020dbd9>] child_rip+0xa/0x11 > [<ffffffff8020ceee>] ? restore_args+0x0/0x30 > [<ffffffff80d0d970>] ? kernel_init+0x0/0x1b0 > [<ffffffff8020dbcf>] ? child_rip+0x0/0x11 > Code: 20 48 83 c1 01 48 39 f1 75 e3 44 89 e8 4c 8b 05 29 29 22 00 31 f6 48 8d 78 01 66 66 90 89 f0 48 8d 04 80 48 c1 e0 05 4a 8d 0c 00 <f6> 81 8c 00 00 00 08 74 26 8b 81 80 00 00 00 8b 91 88 00 00 00 > RIP [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > RSP <ffff88003fa07dd0> > CR2: 000000000000008c > Kernel panic - not syncing: Fatal exception There was one code path, with CONFIG_PCI_MSI disabled, where we were accessing hpet_devs without initialization. That resulted in the above crash. The change below adds a check for hpet_devs. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Shaohua Li <shaohua.li@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-08 17:18:40 +00:00
static void __init hpet_reserve_platform_timers(void)
{
struct hpet_data hd;
unsigned int i;
memset(&hd, 0, sizeof(hd));
hd.hd_phys_address = hpet_address;
hd.hd_address = hpet_virt_address;
hd.hd_nirqs = hpet_base.nr_channels;
hpet_reserve_timer(&hd, 0);
#ifdef CONFIG_HPET_EMULATE_RTC
hpet_reserve_timer(&hd, 1);
#endif
hpet: /dev/hpet - fixes and cleanup Minor /dev/hpet updates and bugfixes: * Remove dead code, mostly remnants of an incomplete/unusable kernel interface ... noted when addressing "sparse" warnings: + hpet_unregister() and a routine it calls + hpet_task and all references, including hpet_task_lock + hpet_data.hd_flags (and HPET_DATA_PLATFORM) * Correct and improve boot message: + displays *counter* (shared between comparators) bit width, not *timer* bit widths (which are often mixed) + relabel "timers" as "comparators"; this is less confusing, they are not independent like normal timers are (sigh) + display MHz not Hz; it's never less than 10 MHz. * Tighten and correct the userspace interface code + don't accidentally program comparators in 64-bit mode using 32-bit values ... always force comparators into 32-bit mode + provide the correct bit definition flagging comparators with periodic capability ... the ABI is unchanged * Update Documentation/hpet.txt + be more correct and current + expand description a bit + don't mention that now-gone kernel interface Plus, add a FIXME comment for something that could cause big trouble on systems with more capable HPETs than at least Intel seems to ship. It seems that few folk use this userspace interface; it's not very usable given the general lack of HPET IRQ routing. I'm told that the only real point of it any more is to mmap for fast timestamps; IMO that's handled better through the gettimeofday() vsyscall. Signed-off-by: David Brownell <dbrownell@users.sourceforge.net> Acked-by: Clemens Ladisch <clemens@ladisch.de> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-07-29 19:47:38 +00:00
/*
* NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
* is wrong for i8259!) not the output IRQ. Many BIOS writers
* don't bother configuring *any* comparator interrupts.
*/
hd.hd_irq[0] = HPET_LEGACY_8254;
hd.hd_irq[1] = HPET_LEGACY_RTC;
for (i = 2; i < hpet_base.nr_channels; i++)
hd.hd_irq[i] = hpet_base.channels[i].irq;
x86: using HPET in MSI mode and setting up per CPU HPET timers, fix On Sat, Sep 06, 2008 at 06:03:53AM -0700, Ingo Molnar wrote: > > it crashes two testsystems, the fault on a NULL pointer in hpet init, > with: > > initcall print_all_ICs+0x0/0x520 returned 0 after 26 msecs > calling hpet_late_init+0x0/0x1c0 > BUG: unable to handle kernel NULL pointer dereference at 000000000000008c > IP: [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > PGD 0 > Oops: 0000 [1] SMP > CPU 0 > Modules linked in: > Pid: 1, comm: swapper Not tainted 2.6.27-rc5 #29725 > RIP: 0010:[<ffffffff80d228be>] [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > RSP: 0018:ffff88003fa07dd0 EFLAGS: 00010246 > RAX: 0000000000000000 RBX: 0000000000000003 RCX: 0000000000000000 > RDX: ffffc20000000160 RSI: 0000000000000000 RDI: 0000000000000003 > RBP: ffff88003fa07e90 R08: 0000000000000000 R09: ffff88003fa07dd0 > R10: 0000000000000001 R11: 0000000000000000 R12: ffff88003fa07dd0 > R13: 0000000000000002 R14: ffffc20000000000 R15: 000000006f57e511 > FS: 0000000000000000(0000) GS:ffffffff80cf6a80(0000) knlGS:0000000000000000 > CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b > CR2: 000000000000008c CR3: 0000000000201000 CR4: 00000000000006e0 > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 > Process swapper (pid: 1, threadinfo ffff88003fa06000, task ffff88003fa08000) > Stack: 00000000fed00000 ffffc20000000000 0000000100000003 0000000800000002 > 0000000000000000 0000000000000000 0000000000000000 0000000000000000 > 0000000000000000 0000000000000000 0000000000000000 0000000000000000 > Call Trace: > [<ffffffff80d227c0>] ? hpet_late_init+0x0/0x1c0 > [<ffffffff80209045>] do_one_initcall+0x45/0x190 > [<ffffffff80296f39>] ? register_irq_proc+0x19/0xe0 > [<ffffffff80d0d140>] ? early_idt_handler+0x0/0x73 > [<ffffffff80d0dabc>] kernel_init+0x14c/0x1b0 > [<ffffffff80942ac1>] ? trace_hardirqs_on_thunk+0x3a/0x3f > [<ffffffff8020dbd9>] child_rip+0xa/0x11 > [<ffffffff8020ceee>] ? restore_args+0x0/0x30 > [<ffffffff80d0d970>] ? kernel_init+0x0/0x1b0 > [<ffffffff8020dbcf>] ? child_rip+0x0/0x11 > Code: 20 48 83 c1 01 48 39 f1 75 e3 44 89 e8 4c 8b 05 29 29 22 00 31 f6 48 8d 78 01 66 66 90 89 f0 48 8d 04 80 48 c1 e0 05 4a 8d 0c 00 <f6> 81 8c 00 00 00 08 74 26 8b 81 80 00 00 00 8b 91 88 00 00 00 > RIP [<ffffffff80d228be>] hpet_late_init+0xfe/0x1c0 > RSP <ffff88003fa07dd0> > CR2: 000000000000008c > Kernel panic - not syncing: Fatal exception There was one code path, with CONFIG_PCI_MSI disabled, where we were accessing hpet_devs without initialization. That resulted in the above crash. The change below adds a check for hpet_devs. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Shaohua Li <shaohua.li@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-08 17:18:40 +00:00
hpet_reserve_msi_timers(&hd);
hpet_alloc(&hd);
}
static void __init hpet_select_device_channel(void)
{
int i;
for (i = 0; i < hpet_base.nr_channels; i++) {
struct hpet_channel *hc = hpet_base.channels + i;
/* Associate the first unused channel to /dev/hpet */
if (hc->mode == HPET_MODE_UNUSED) {
hc->mode = HPET_MODE_DEVICE;
return;
}
}
}
#else
static inline void hpet_reserve_platform_timers(void) { }
static inline void hpet_select_device_channel(void) {}
#endif
/* Common HPET functions */
static void hpet_stop_counter(void)
{
u32 cfg = hpet_readl(HPET_CFG);
cfg &= ~HPET_CFG_ENABLE;
hpet_writel(cfg, HPET_CFG);
}
static void hpet_reset_counter(void)
{
hpet_writel(0, HPET_COUNTER);
hpet_writel(0, HPET_COUNTER + 4);
}
static void hpet_start_counter(void)
{
unsigned int cfg = hpet_readl(HPET_CFG);
cfg |= HPET_CFG_ENABLE;
hpet_writel(cfg, HPET_CFG);
}
static void hpet_restart_counter(void)
{
hpet_stop_counter();
hpet_reset_counter();
hpet_start_counter();
}
static void hpet_resume_device(void)
{
force_hpet_resume();
}
static void hpet_resume_counter(struct clocksource *cs)
{
hpet_resume_device();
hpet_restart_counter();
}
static void hpet_enable_legacy_int(void)
{
unsigned int cfg = hpet_readl(HPET_CFG);
cfg |= HPET_CFG_LEGACY;
hpet_writel(cfg, HPET_CFG);
hpet_legacy_int_enabled = true;
}
static void hpet_legacy_clockevent_register(void)
{
/* Start HPET legacy interrupts */
hpet_enable_legacy_int();
/*
* Start HPET with the boot CPU's cpumask and make it global after
* the IO_APIC has been initialized.
*/
hpet_clockevent.cpumask = cpumask_of(boot_cpu_data.cpu_index);
clockevents_config_and_register(&hpet_clockevent, hpet_freq,
HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
global_clock_event = &hpet_clockevent;
pr_debug("Clockevent registered\n");
}
static int hpet_set_periodic(struct clock_event_device *evt, int timer)
{
unsigned int cfg, cmp, now;
uint64_t delta;
hpet_stop_counter();
delta = ((uint64_t)(NSEC_PER_SEC / HZ)) * evt->mult;
delta >>= evt->shift;
now = hpet_readl(HPET_COUNTER);
cmp = now + (unsigned int)delta;
cfg = hpet_readl(HPET_Tn_CFG(timer));
cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC | HPET_TN_SETVAL |
HPET_TN_32BIT;
hpet_writel(cfg, HPET_Tn_CFG(timer));
hpet_writel(cmp, HPET_Tn_CMP(timer));
udelay(1);
/*
* HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
* cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
* bit is automatically cleared after the first write.
* (See AMD-8111 HyperTransport I/O Hub Data Sheet,
* Publication # 24674)
*/
hpet_writel((unsigned int)delta, HPET_Tn_CMP(timer));
hpet_start_counter();
hpet_print_config();
return 0;
}
static int hpet_set_oneshot(struct clock_event_device *evt, int timer)
{
unsigned int cfg;
cfg = hpet_readl(HPET_Tn_CFG(timer));
cfg &= ~HPET_TN_PERIODIC;
cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
hpet_writel(cfg, HPET_Tn_CFG(timer));
return 0;
}
static int hpet_shutdown(struct clock_event_device *evt, int timer)
{
unsigned int cfg;
cfg = hpet_readl(HPET_Tn_CFG(timer));
cfg &= ~HPET_TN_ENABLE;
hpet_writel(cfg, HPET_Tn_CFG(timer));
return 0;
}
x86/hpet: Cure interface abuse in the resume path The HPET resume path abuses irq_domain_[de]activate_irq() to restore the MSI message in the HPET chip for the boot CPU on resume and it relies on an implementation detail of the interrupt core code, which magically makes the HPET unmask call invoked via a irq_disable/enable pair. This worked as long as the irq code did unconditionally invoke the unmask() callback. With the recent changes which keep track of the masked state to avoid expensive hardware access, this does not longer work. As a consequence the HPET timer interrupts are not unmasked which breaks resume as the boot CPU waits forever that a timer interrupt arrives. Make the restore of the MSI message explicit and invoke the unmask() function directly. While at it get rid of the pointless affinity setting as nothing can change the affinity of the interrupt and the vector across suspend/resume. The restore of the MSI message reestablishes the previous affinity setting which is the correct one. Fixes: bf22ff45bed6 ("genirq: Avoid unnecessary low level irq function calls") Reported-and-tested-by: Tomi Sarvela <tomi.p.sarvela@intel.com> Reported-by: Martin Peres <martin.peres@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: jeffy.chen@rock-chips.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Marc Zyngier <marc.zyngier@arm.com> Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1707312158590.2287@nanos
2017-07-31 20:07:09 +00:00
static int hpet_resume(struct clock_event_device *evt)
{
hpet_enable_legacy_int();
hpet_print_config();
return 0;
}
static int hpet_next_event(unsigned long delta, int channel)
{
u32 cnt;
x86: Hpet: Avoid the comparator readback penalty Due to the overly intelligent design of HPETs, we need to workaround the problem that the compare value which we write is already behind the actual counter value at the point where the value hits the real compare register. This happens for two reasons: 1) We read out the counter, add the delta and write the result to the compare register. When a NMI or SMI hits between the read out and the write then the counter can be ahead of the event already 2) The write to the compare register is delayed by up to two HPET cycles in certain chipsets. We worked around this by reading back the compare register to make sure that the written value has hit the hardware. For certain ICH9+ chipsets this can require two readouts, as the first one can return the previous compare register value. That's bad performance wise for the normal case where the event is far enough in the future. As we already know that the write can be delayed by up to two cycles we can avoid the read back of the compare register completely if we make the decision whether the delta has elapsed already or not based on the following calculation: cmp = event - actual_count; If cmp is less than 8 HPET clock cycles, then we decide that the event has happened already and return -ETIME. That covers the above #1 and #2 problems which would cause a wait for HPET wraparound (~306 seconds). Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Nix <nix@esperi.org.uk> Tested-by: Artur Skawina <art.08.09@gmail.com> Cc: Damien Wyart <damien.wyart@free.fr> Tested-by: John Drescher <drescherjm@gmail.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: Andreas Herrmann <andreas.herrmann3@amd.com> Tested-by: Borislav Petkov <borislav.petkov@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> LKML-Reference: <alpine.LFD.2.00.1009151500060.2416@localhost6.localdomain6>
2010-09-15 13:11:57 +00:00
s32 res;
cnt = hpet_readl(HPET_COUNTER);
cnt += (u32) delta;
hpet_writel(cnt, HPET_Tn_CMP(channel));
/*
x86: Hpet: Avoid the comparator readback penalty Due to the overly intelligent design of HPETs, we need to workaround the problem that the compare value which we write is already behind the actual counter value at the point where the value hits the real compare register. This happens for two reasons: 1) We read out the counter, add the delta and write the result to the compare register. When a NMI or SMI hits between the read out and the write then the counter can be ahead of the event already 2) The write to the compare register is delayed by up to two HPET cycles in certain chipsets. We worked around this by reading back the compare register to make sure that the written value has hit the hardware. For certain ICH9+ chipsets this can require two readouts, as the first one can return the previous compare register value. That's bad performance wise for the normal case where the event is far enough in the future. As we already know that the write can be delayed by up to two cycles we can avoid the read back of the compare register completely if we make the decision whether the delta has elapsed already or not based on the following calculation: cmp = event - actual_count; If cmp is less than 8 HPET clock cycles, then we decide that the event has happened already and return -ETIME. That covers the above #1 and #2 problems which would cause a wait for HPET wraparound (~306 seconds). Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Nix <nix@esperi.org.uk> Tested-by: Artur Skawina <art.08.09@gmail.com> Cc: Damien Wyart <damien.wyart@free.fr> Tested-by: John Drescher <drescherjm@gmail.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: Andreas Herrmann <andreas.herrmann3@amd.com> Tested-by: Borislav Petkov <borislav.petkov@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> LKML-Reference: <alpine.LFD.2.00.1009151500060.2416@localhost6.localdomain6>
2010-09-15 13:11:57 +00:00
* HPETs are a complete disaster. The compare register is
* based on a equal comparison and neither provides a less
* than or equal functionality (which would require to take
* the wraparound into account) nor a simple count down event
* mode. Further the write to the comparator register is
* delayed internally up to two HPET clock cycles in certain
* chipsets (ATI, ICH9,10). Some newer AMD chipsets have even
* longer delays. We worked around that by reading back the
* compare register, but that required another workaround for
* ICH9,10 chips where the first readout after write can
* return the old stale value. We already had a minimum
* programming delta of 5us enforced, but a NMI or SMI hitting
x86: Hpet: Avoid the comparator readback penalty Due to the overly intelligent design of HPETs, we need to workaround the problem that the compare value which we write is already behind the actual counter value at the point where the value hits the real compare register. This happens for two reasons: 1) We read out the counter, add the delta and write the result to the compare register. When a NMI or SMI hits between the read out and the write then the counter can be ahead of the event already 2) The write to the compare register is delayed by up to two HPET cycles in certain chipsets. We worked around this by reading back the compare register to make sure that the written value has hit the hardware. For certain ICH9+ chipsets this can require two readouts, as the first one can return the previous compare register value. That's bad performance wise for the normal case where the event is far enough in the future. As we already know that the write can be delayed by up to two cycles we can avoid the read back of the compare register completely if we make the decision whether the delta has elapsed already or not based on the following calculation: cmp = event - actual_count; If cmp is less than 8 HPET clock cycles, then we decide that the event has happened already and return -ETIME. That covers the above #1 and #2 problems which would cause a wait for HPET wraparound (~306 seconds). Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Nix <nix@esperi.org.uk> Tested-by: Artur Skawina <art.08.09@gmail.com> Cc: Damien Wyart <damien.wyart@free.fr> Tested-by: John Drescher <drescherjm@gmail.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: Andreas Herrmann <andreas.herrmann3@amd.com> Tested-by: Borislav Petkov <borislav.petkov@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> LKML-Reference: <alpine.LFD.2.00.1009151500060.2416@localhost6.localdomain6>
2010-09-15 13:11:57 +00:00
* between the counter readout and the comparator write can
* move us behind that point easily. Now instead of reading
* the compare register back several times, we make the ETIME
* decision based on the following: Return ETIME if the
* counter value after the write is less than HPET_MIN_CYCLES
x86: Hpet: Avoid the comparator readback penalty Due to the overly intelligent design of HPETs, we need to workaround the problem that the compare value which we write is already behind the actual counter value at the point where the value hits the real compare register. This happens for two reasons: 1) We read out the counter, add the delta and write the result to the compare register. When a NMI or SMI hits between the read out and the write then the counter can be ahead of the event already 2) The write to the compare register is delayed by up to two HPET cycles in certain chipsets. We worked around this by reading back the compare register to make sure that the written value has hit the hardware. For certain ICH9+ chipsets this can require two readouts, as the first one can return the previous compare register value. That's bad performance wise for the normal case where the event is far enough in the future. As we already know that the write can be delayed by up to two cycles we can avoid the read back of the compare register completely if we make the decision whether the delta has elapsed already or not based on the following calculation: cmp = event - actual_count; If cmp is less than 8 HPET clock cycles, then we decide that the event has happened already and return -ETIME. That covers the above #1 and #2 problems which would cause a wait for HPET wraparound (~306 seconds). Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Nix <nix@esperi.org.uk> Tested-by: Artur Skawina <art.08.09@gmail.com> Cc: Damien Wyart <damien.wyart@free.fr> Tested-by: John Drescher <drescherjm@gmail.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: Andreas Herrmann <andreas.herrmann3@amd.com> Tested-by: Borislav Petkov <borislav.petkov@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> LKML-Reference: <alpine.LFD.2.00.1009151500060.2416@localhost6.localdomain6>
2010-09-15 13:11:57 +00:00
* away from the event or if the counter is already ahead of
* the event. The minimum programming delta for the generic
* clockevents code is set to 1.5 * HPET_MIN_CYCLES.
*/
x86: Hpet: Avoid the comparator readback penalty Due to the overly intelligent design of HPETs, we need to workaround the problem that the compare value which we write is already behind the actual counter value at the point where the value hits the real compare register. This happens for two reasons: 1) We read out the counter, add the delta and write the result to the compare register. When a NMI or SMI hits between the read out and the write then the counter can be ahead of the event already 2) The write to the compare register is delayed by up to two HPET cycles in certain chipsets. We worked around this by reading back the compare register to make sure that the written value has hit the hardware. For certain ICH9+ chipsets this can require two readouts, as the first one can return the previous compare register value. That's bad performance wise for the normal case where the event is far enough in the future. As we already know that the write can be delayed by up to two cycles we can avoid the read back of the compare register completely if we make the decision whether the delta has elapsed already or not based on the following calculation: cmp = event - actual_count; If cmp is less than 8 HPET clock cycles, then we decide that the event has happened already and return -ETIME. That covers the above #1 and #2 problems which would cause a wait for HPET wraparound (~306 seconds). Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Nix <nix@esperi.org.uk> Tested-by: Artur Skawina <art.08.09@gmail.com> Cc: Damien Wyart <damien.wyart@free.fr> Tested-by: John Drescher <drescherjm@gmail.com> Cc: Venkatesh Pallipadi <venki@google.com> Cc: Arjan van de Ven <arjan@linux.intel.com> Cc: Andreas Herrmann <andreas.herrmann3@amd.com> Tested-by: Borislav Petkov <borislav.petkov@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> LKML-Reference: <alpine.LFD.2.00.1009151500060.2416@localhost6.localdomain6>
2010-09-15 13:11:57 +00:00
res = (s32)(cnt - hpet_readl(HPET_COUNTER));
return res < HPET_MIN_CYCLES ? -ETIME : 0;
}
static int hpet_legacy_shutdown(struct clock_event_device *evt)
{
return hpet_shutdown(evt, 0);
}
static int hpet_legacy_set_oneshot(struct clock_event_device *evt)
{
return hpet_set_oneshot(evt, 0);
}
static int hpet_legacy_set_periodic(struct clock_event_device *evt)
{
return hpet_set_periodic(evt, 0);
}
static int hpet_legacy_resume(struct clock_event_device *evt)
{
x86/hpet: Cure interface abuse in the resume path The HPET resume path abuses irq_domain_[de]activate_irq() to restore the MSI message in the HPET chip for the boot CPU on resume and it relies on an implementation detail of the interrupt core code, which magically makes the HPET unmask call invoked via a irq_disable/enable pair. This worked as long as the irq code did unconditionally invoke the unmask() callback. With the recent changes which keep track of the masked state to avoid expensive hardware access, this does not longer work. As a consequence the HPET timer interrupts are not unmasked which breaks resume as the boot CPU waits forever that a timer interrupt arrives. Make the restore of the MSI message explicit and invoke the unmask() function directly. While at it get rid of the pointless affinity setting as nothing can change the affinity of the interrupt and the vector across suspend/resume. The restore of the MSI message reestablishes the previous affinity setting which is the correct one. Fixes: bf22ff45bed6 ("genirq: Avoid unnecessary low level irq function calls") Reported-and-tested-by: Tomi Sarvela <tomi.p.sarvela@intel.com> Reported-by: Martin Peres <martin.peres@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: jeffy.chen@rock-chips.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Marc Zyngier <marc.zyngier@arm.com> Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1707312158590.2287@nanos
2017-07-31 20:07:09 +00:00
return hpet_resume(evt);
}
static int hpet_legacy_next_event(unsigned long delta,
struct clock_event_device *evt)
{
return hpet_next_event(delta, 0);
}
/*
* The HPET clock event device
*/
static struct clock_event_device hpet_clockevent = {
.name = "hpet",
.features = CLOCK_EVT_FEAT_PERIODIC |
CLOCK_EVT_FEAT_ONESHOT,
.set_state_periodic = hpet_legacy_set_periodic,
.set_state_oneshot = hpet_legacy_set_oneshot,
.set_state_shutdown = hpet_legacy_shutdown,
.tick_resume = hpet_legacy_resume,
.set_next_event = hpet_legacy_next_event,
.irq = 0,
.rating = 50,
};
/*
* HPET MSI Support
*/
#ifdef CONFIG_PCI_MSI
void hpet_msi_unmask(struct irq_data *data)
{
struct hpet_dev *hdev = irq_data_get_irq_handler_data(data);
unsigned int cfg;
/* unmask it */
cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
cfg |= HPET_TN_ENABLE | HPET_TN_FSB;
hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
}
void hpet_msi_mask(struct irq_data *data)
{
struct hpet_dev *hdev = irq_data_get_irq_handler_data(data);
unsigned int cfg;
/* mask it */
cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
cfg &= ~(HPET_TN_ENABLE | HPET_TN_FSB);
hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
}
void hpet_msi_write(struct hpet_dev *hdev, struct msi_msg *msg)
{
hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
}
static int hpet_msi_shutdown(struct clock_event_device *evt)
{
return hpet_shutdown(evt, clockevent_to_channel(evt)->num);
}
static int hpet_msi_set_oneshot(struct clock_event_device *evt)
{
return hpet_set_oneshot(evt, clockevent_to_channel(evt)->num);
}
static int hpet_msi_set_periodic(struct clock_event_device *evt)
{
return hpet_set_periodic(evt, clockevent_to_channel(evt)->num);
}
static int hpet_msi_resume(struct clock_event_device *evt)
{
struct hpet_dev *hdev = clockevent_to_channel(evt);
x86/hpet: Cure interface abuse in the resume path The HPET resume path abuses irq_domain_[de]activate_irq() to restore the MSI message in the HPET chip for the boot CPU on resume and it relies on an implementation detail of the interrupt core code, which magically makes the HPET unmask call invoked via a irq_disable/enable pair. This worked as long as the irq code did unconditionally invoke the unmask() callback. With the recent changes which keep track of the masked state to avoid expensive hardware access, this does not longer work. As a consequence the HPET timer interrupts are not unmasked which breaks resume as the boot CPU waits forever that a timer interrupt arrives. Make the restore of the MSI message explicit and invoke the unmask() function directly. While at it get rid of the pointless affinity setting as nothing can change the affinity of the interrupt and the vector across suspend/resume. The restore of the MSI message reestablishes the previous affinity setting which is the correct one. Fixes: bf22ff45bed6 ("genirq: Avoid unnecessary low level irq function calls") Reported-and-tested-by: Tomi Sarvela <tomi.p.sarvela@intel.com> Reported-by: Martin Peres <martin.peres@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: jeffy.chen@rock-chips.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Marc Zyngier <marc.zyngier@arm.com> Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1707312158590.2287@nanos
2017-07-31 20:07:09 +00:00
struct irq_data *data = irq_get_irq_data(hdev->irq);
struct msi_msg msg;
x86/hpet: Cure interface abuse in the resume path The HPET resume path abuses irq_domain_[de]activate_irq() to restore the MSI message in the HPET chip for the boot CPU on resume and it relies on an implementation detail of the interrupt core code, which magically makes the HPET unmask call invoked via a irq_disable/enable pair. This worked as long as the irq code did unconditionally invoke the unmask() callback. With the recent changes which keep track of the masked state to avoid expensive hardware access, this does not longer work. As a consequence the HPET timer interrupts are not unmasked which breaks resume as the boot CPU waits forever that a timer interrupt arrives. Make the restore of the MSI message explicit and invoke the unmask() function directly. While at it get rid of the pointless affinity setting as nothing can change the affinity of the interrupt and the vector across suspend/resume. The restore of the MSI message reestablishes the previous affinity setting which is the correct one. Fixes: bf22ff45bed6 ("genirq: Avoid unnecessary low level irq function calls") Reported-and-tested-by: Tomi Sarvela <tomi.p.sarvela@intel.com> Reported-by: Martin Peres <martin.peres@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com> Cc: jeffy.chen@rock-chips.com Cc: Peter Zijlstra <peterz@infradead.org> Cc: Marc Zyngier <marc.zyngier@arm.com> Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1707312158590.2287@nanos
2017-07-31 20:07:09 +00:00
/* Restore the MSI msg and unmask the interrupt */
irq_chip_compose_msi_msg(data, &msg);
hpet_msi_write(hdev, &msg);
hpet_msi_unmask(data);
return 0;
}
static int hpet_msi_next_event(unsigned long delta,
struct clock_event_device *evt)
{
return hpet_next_event(delta, clockevent_to_channel(evt)->num);
}
static irqreturn_t hpet_interrupt_handler(int irq, void *data)
{
struct hpet_dev *dev = data;
struct clock_event_device *evt = &dev->evt;
if (!evt->event_handler) {
pr_info("Spurious interrupt HPET timer %d\n", dev->num);
return IRQ_HANDLED;
}
evt->event_handler(evt);
return IRQ_HANDLED;
}
static int hpet_setup_irq(struct hpet_dev *dev)
{
if (request_irq(dev->irq, hpet_interrupt_handler,
IRQF_TIMER | IRQF_NOBALANCING,
dev->name, dev))
return -1;
disable_irq(dev->irq);
irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
enable_irq(dev->irq);
pr_debug("%s irq %d for MSI\n", dev->name, dev->irq);
return 0;
}
static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
{
struct clock_event_device *evt = &hdev->evt;
if (!(hdev->flags & HPET_DEV_VALID))
return;
hdev->cpu = cpu;
per_cpu(cpu_hpet_dev, cpu) = hdev;
evt->name = hdev->name;
hpet_setup_irq(hdev);
evt->irq = hdev->irq;
evt->rating = 110;
evt->features = CLOCK_EVT_FEAT_ONESHOT;
if (hdev->flags & HPET_DEV_PERI_CAP) {
evt->features |= CLOCK_EVT_FEAT_PERIODIC;
evt->set_state_periodic = hpet_msi_set_periodic;
}
evt->set_state_shutdown = hpet_msi_shutdown;
evt->set_state_oneshot = hpet_msi_set_oneshot;
evt->tick_resume = hpet_msi_resume;
evt->set_next_event = hpet_msi_next_event;
evt->cpumask = cpumask_of(hdev->cpu);
clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
0x7FFFFFFF);
}
static struct hpet_dev *hpet_get_unused_timer(void)
{
int i;
if (!hpet_devs)
return NULL;
for (i = 0; i < hpet_base.nr_channels; i++) {
struct hpet_dev *hdev = &hpet_devs[i];
if (!(hdev->flags & HPET_DEV_VALID))
continue;
if (test_and_set_bit(HPET_DEV_USED_BIT,
(unsigned long *)&hdev->flags))
continue;
return hdev;
}
return NULL;
}
static int hpet_cpuhp_online(unsigned int cpu)
{
struct hpet_dev *hdev = hpet_get_unused_timer();
if (hdev)
init_one_hpet_msi_clockevent(hdev, cpu);
return 0;
}
static int hpet_cpuhp_dead(unsigned int cpu)
{
struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
if (!hdev)
return 0;
free_irq(hdev->irq, hdev);
hdev->flags &= ~HPET_DEV_USED;
per_cpu(cpu_hpet_dev, cpu) = NULL;
return 0;
}
#ifdef CONFIG_HPET
/* Reserve at least one timer for userspace (/dev/hpet) */
#define RESERVE_TIMERS 1
#else
#define RESERVE_TIMERS 0
#endif
static void __init hpet_msi_capability_lookup(unsigned int start_timer)
{
unsigned int num_timers;
unsigned int num_timers_used = 0;
int i, irq;
if (hpet_msi_disable)
return;
if (boot_cpu_has(X86_FEATURE_ARAT))
return;
num_timers = hpet_base.nr_channels;
hpet_print_config();
hpet_domain = hpet_create_irq_domain(hpet_blockid);
if (!hpet_domain)
return;
treewide: kzalloc() -> kcalloc() The kzalloc() function has a 2-factor argument form, kcalloc(). This patch replaces cases of: kzalloc(a * b, gfp) with: kcalloc(a * b, gfp) as well as handling cases of: kzalloc(a * b * c, gfp) with: kzalloc(array3_size(a, b, c), gfp) as it's slightly less ugly than: kzalloc_array(array_size(a, b), c, gfp) This does, however, attempt to ignore constant size factors like: kzalloc(4 * 1024, gfp) though any constants defined via macros get caught up in the conversion. Any factors with a sizeof() of "unsigned char", "char", and "u8" were dropped, since they're redundant. The Coccinelle script used for this was: // Fix redundant parens around sizeof(). @@ type TYPE; expression THING, E; @@ ( kzalloc( - (sizeof(TYPE)) * E + sizeof(TYPE) * E , ...) | kzalloc( - (sizeof(THING)) * E + sizeof(THING) * E , ...) ) // Drop single-byte sizes and redundant parens. @@ expression COUNT; typedef u8; typedef __u8; @@ ( kzalloc( - sizeof(u8) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(__u8) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(char) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(unsigned char) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(u8) * COUNT + COUNT , ...) | kzalloc( - sizeof(__u8) * COUNT + COUNT , ...) | kzalloc( - sizeof(char) * COUNT + COUNT , ...) | kzalloc( - sizeof(unsigned char) * COUNT + COUNT , ...) ) // 2-factor product with sizeof(type/expression) and identifier or constant. @@ type TYPE; expression THING; identifier COUNT_ID; constant COUNT_CONST; @@ ( - kzalloc + kcalloc ( - sizeof(TYPE) * (COUNT_ID) + COUNT_ID, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * COUNT_ID + COUNT_ID, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * (COUNT_CONST) + COUNT_CONST, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * COUNT_CONST + COUNT_CONST, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * (COUNT_ID) + COUNT_ID, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * COUNT_ID + COUNT_ID, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * (COUNT_CONST) + COUNT_CONST, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * COUNT_CONST + COUNT_CONST, sizeof(THING) , ...) ) // 2-factor product, only identifiers. @@ identifier SIZE, COUNT; @@ - kzalloc + kcalloc ( - SIZE * COUNT + COUNT, SIZE , ...) // 3-factor product with 1 sizeof(type) or sizeof(expression), with // redundant parens removed. @@ expression THING; identifier STRIDE, COUNT; type TYPE; @@ ( kzalloc( - sizeof(TYPE) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(TYPE) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(TYPE) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(TYPE) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(THING) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kzalloc( - sizeof(THING) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kzalloc( - sizeof(THING) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kzalloc( - sizeof(THING) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) ) // 3-factor product with 2 sizeof(variable), with redundant parens removed. @@ expression THING1, THING2; identifier COUNT; type TYPE1, TYPE2; @@ ( kzalloc( - sizeof(TYPE1) * sizeof(TYPE2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | kzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | kzalloc( - sizeof(THING1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | kzalloc( - sizeof(THING1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | kzalloc( - sizeof(TYPE1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) | kzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) ) // 3-factor product, only identifiers, with redundant parens removed. @@ identifier STRIDE, SIZE, COUNT; @@ ( kzalloc( - (COUNT) * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - (COUNT) * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - (COUNT) * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - (COUNT) * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) ) // Any remaining multi-factor products, first at least 3-factor products, // when they're not all constants... @@ expression E1, E2, E3; constant C1, C2, C3; @@ ( kzalloc(C1 * C2 * C3, ...) | kzalloc( - (E1) * E2 * E3 + array3_size(E1, E2, E3) , ...) | kzalloc( - (E1) * (E2) * E3 + array3_size(E1, E2, E3) , ...) | kzalloc( - (E1) * (E2) * (E3) + array3_size(E1, E2, E3) , ...) | kzalloc( - E1 * E2 * E3 + array3_size(E1, E2, E3) , ...) ) // And then all remaining 2 factors products when they're not all constants, // keeping sizeof() as the second factor argument. @@ expression THING, E1, E2; type TYPE; constant C1, C2, C3; @@ ( kzalloc(sizeof(THING) * C2, ...) | kzalloc(sizeof(TYPE) * C2, ...) | kzalloc(C1 * C2 * C3, ...) | kzalloc(C1 * C2, ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * (E2) + E2, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * E2 + E2, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * (E2) + E2, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * E2 + E2, sizeof(THING) , ...) | - kzalloc + kcalloc ( - (E1) * E2 + E1, E2 , ...) | - kzalloc + kcalloc ( - (E1) * (E2) + E1, E2 , ...) | - kzalloc + kcalloc ( - E1 * E2 + E1, E2 , ...) ) Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-12 21:03:40 +00:00
hpet_devs = kcalloc(num_timers, sizeof(struct hpet_dev), GFP_KERNEL);
if (!hpet_devs)
return;
for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
struct hpet_dev *hdev = &hpet_devs[num_timers_used];
unsigned int cfg = hpet_base.channels[i].boot_cfg;
/* Only consider HPET timer with MSI support */
if (!(cfg & HPET_TN_FSB_CAP))
continue;
hdev->flags = 0;
if (cfg & HPET_TN_PERIODIC_CAP)
hdev->flags |= HPET_DEV_PERI_CAP;
sprintf(hdev->name, "hpet%d", i);
hdev->num = i;
irq = hpet_assign_irq(hpet_domain, hdev, hdev->num);
if (irq <= 0)
continue;
hdev->irq = irq;
hdev->flags |= HPET_DEV_FSB_CAP;
hdev->flags |= HPET_DEV_VALID;
num_timers_used++;
if (num_timers_used == num_possible_cpus())
break;
}
pr_info("%d channels of %d reserved for per-cpu timers\n",
num_timers, num_timers_used);
}
#ifdef CONFIG_HPET
static void __init hpet_reserve_msi_timers(struct hpet_data *hd)
{
int i;
if (!hpet_devs)
return;
for (i = 0; i < hpet_base.nr_channels; i++) {
struct hpet_dev *hdev = &hpet_devs[i];
if (!(hdev->flags & HPET_DEV_VALID))
continue;
hd->hd_irq[hdev->num] = hdev->irq;
hpet_reserve_timer(hd, hdev->num);
}
}
#endif
#else
static inline void hpet_msi_capability_lookup(unsigned int start_timer) { }
#ifdef CONFIG_HPET
static inline void hpet_reserve_msi_timers(struct hpet_data *hd) { }
#endif
#define hpet_cpuhp_online NULL
#define hpet_cpuhp_dead NULL
#endif
[PATCH] clocksource init adjustments (fix bug #7426) This patch resolves the issue found here: http://bugme.osdl.org/show_bug.cgi?id=7426 The basic summary is: Currently we register most of i386/x86_64 clocksources at module_init time. Then we enable clocksource selection at late_initcall time. This causes some problems for drivers that use gettimeofday for init calibration routines (specifically the es1968 driver in this case), where durring module_init, the only clocksource available is the low-res jiffies clocksource. This may cause slight calibration errors, due to the small sampling time used. It should be noted that drivers that require fine grained time may not function on architectures that do not have better then jiffies resolution timekeeping (there are a few). However, this does not discount the reasonable need for such fine-grained timekeeping at init time. Thus the solution here is to register clocksources earlier (ideally when the hardware is being initialized), and then we enable clocksource selection at fs_initcall (before device_initcall). This patch should probably get some testing time in -mm, since clocksource selection is one of the most important issues for correct timekeeping, and I've only been able to test this on a few of my own boxes. Signed-off-by: John Stultz <johnstul@us.ibm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: "David S. Miller" <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-05 08:30:50 +00:00
/*
* Clock source related code
*/
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
#if defined(CONFIG_SMP) && defined(CONFIG_64BIT)
/*
* Reading the HPET counter is a very slow operation. If a large number of
* CPUs are trying to access the HPET counter simultaneously, it can cause
* massive delays and slow down system performance dramatically. This may
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
* happen when HPET is the default clock source instead of TSC. For a
* really large system with hundreds of CPUs, the slowdown may be so
* severe, that it can actually crash the system because of a NMI watchdog
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
* soft lockup, for example.
*
* If multiple CPUs are trying to access the HPET counter at the same time,
* we don't actually need to read the counter multiple times. Instead, the
* other CPUs can use the counter value read by the first CPU in the group.
*
* This special feature is only enabled on x86-64 systems. It is unlikely
* that 32-bit x86 systems will have enough CPUs to require this feature
* with its associated locking overhead. We also need 64-bit atomic read.
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
*
* The lock and the HPET value are stored together and can be read in a
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
* single atomic 64-bit read. It is explicitly assumed that arch_spinlock_t
* is 32 bits in size.
*/
union hpet_lock {
struct {
arch_spinlock_t lock;
u32 value;
};
u64 lockval;
};
static union hpet_lock hpet __cacheline_aligned = {
{ .lock = __ARCH_SPIN_LOCK_UNLOCKED, },
};
static u64 read_hpet(struct clocksource *cs)
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
{
unsigned long flags;
union hpet_lock old, new;
BUILD_BUG_ON(sizeof(union hpet_lock) != 8);
/*
* Read HPET directly if in NMI.
*/
if (in_nmi())
return (u64)hpet_readl(HPET_COUNTER);
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
/*
* Read the current state of the lock and HPET value atomically.
*/
old.lockval = READ_ONCE(hpet.lockval);
if (arch_spin_is_locked(&old.lock))
goto contended;
local_irq_save(flags);
if (arch_spin_trylock(&hpet.lock)) {
new.value = hpet_readl(HPET_COUNTER);
/*
* Use WRITE_ONCE() to prevent store tearing.
*/
WRITE_ONCE(hpet.value, new.value);
arch_spin_unlock(&hpet.lock);
local_irq_restore(flags);
return (u64)new.value;
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
}
local_irq_restore(flags);
contended:
/*
* Contended case
* --------------
* Wait until the HPET value change or the lock is free to indicate
* its value is up-to-date.
*
* It is possible that old.value has already contained the latest
* HPET value while the lock holder was in the process of releasing
* the lock. Checking for lock state change will enable us to return
* the value immediately instead of waiting for the next HPET reader
* to come along.
*/
do {
cpu_relax();
new.lockval = READ_ONCE(hpet.lockval);
} while ((new.value == old.value) && arch_spin_is_locked(&new.lock));
return (u64)new.value;
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
}
#else
/*
* For UP or 32-bit.
*/
static u64 read_hpet(struct clocksource *cs)
[PATCH] clocksource init adjustments (fix bug #7426) This patch resolves the issue found here: http://bugme.osdl.org/show_bug.cgi?id=7426 The basic summary is: Currently we register most of i386/x86_64 clocksources at module_init time. Then we enable clocksource selection at late_initcall time. This causes some problems for drivers that use gettimeofday for init calibration routines (specifically the es1968 driver in this case), where durring module_init, the only clocksource available is the low-res jiffies clocksource. This may cause slight calibration errors, due to the small sampling time used. It should be noted that drivers that require fine grained time may not function on architectures that do not have better then jiffies resolution timekeeping (there are a few). However, this does not discount the reasonable need for such fine-grained timekeeping at init time. Thus the solution here is to register clocksources earlier (ideally when the hardware is being initialized), and then we enable clocksource selection at fs_initcall (before device_initcall). This patch should probably get some testing time in -mm, since clocksource selection is one of the most important issues for correct timekeeping, and I've only been able to test this on a few of my own boxes. Signed-off-by: John Stultz <johnstul@us.ibm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: "David S. Miller" <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-05 08:30:50 +00:00
{
return (u64)hpet_readl(HPET_COUNTER);
[PATCH] clocksource init adjustments (fix bug #7426) This patch resolves the issue found here: http://bugme.osdl.org/show_bug.cgi?id=7426 The basic summary is: Currently we register most of i386/x86_64 clocksources at module_init time. Then we enable clocksource selection at late_initcall time. This causes some problems for drivers that use gettimeofday for init calibration routines (specifically the es1968 driver in this case), where durring module_init, the only clocksource available is the low-res jiffies clocksource. This may cause slight calibration errors, due to the small sampling time used. It should be noted that drivers that require fine grained time may not function on architectures that do not have better then jiffies resolution timekeeping (there are a few). However, this does not discount the reasonable need for such fine-grained timekeeping at init time. Thus the solution here is to register clocksources earlier (ideally when the hardware is being initialized), and then we enable clocksource selection at fs_initcall (before device_initcall). This patch should probably get some testing time in -mm, since clocksource selection is one of the most important issues for correct timekeeping, and I've only been able to test this on a few of my own boxes. Signed-off-by: John Stultz <johnstul@us.ibm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: "David S. Miller" <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-05 08:30:50 +00:00
}
x86/hpet: Reduce HPET counter read contention On a large system with many CPUs, using HPET as the clock source can have a significant impact on the overall system performance because of the following reasons: 1) There is a single HPET counter shared by all the CPUs. 2) HPET counter reading is a very slow operation. Using HPET as the default clock source may happen when, for example, the TSC clock calibration exceeds the allowable tolerance. Something the performance slowdown can be so severe that the system may crash because of a NMI watchdog soft lockup, for example. During the TSC clock calibration process, the default clock source will be set temporarily to HPET. For systems with many CPUs, it is possible that NMI watchdog soft lockup may occur occasionally during that short time period where HPET clocking is active as is shown in the kernel log below: [ 71.646504] hpet0: 8 comparators, 64-bit 14.318180 MHz counter [ 71.655313] Switching to clocksource hpet [ 95.679135] BUG: soft lockup - CPU#144 stuck for 23s! [swapper/144:0] [ 95.693363] BUG: soft lockup - CPU#145 stuck for 23s! [swapper/145:0] [ 95.695580] BUG: soft lockup - CPU#582 stuck for 23s! [swapper/582:0] [ 95.698128] BUG: soft lockup - CPU#357 stuck for 23s! [swapper/357:0] This patch addresses the above issues by reducing HPET read contention using the fact that if more than one CPUs are trying to access HPET at the same time, it will be more efficient when only one CPU in the group reads the HPET counter and shares it with the rest of the group instead of each group member trying to read the HPET counter individually. This is done by using a combination quadword that contains a 32-bit stored HPET value and a 32-bit spinlock. The CPU that gets the lock will be responsible for reading the HPET counter and storing it in the quadword. The others will monitor the change in HPET value and lock status and grab the latest stored HPET value accordingly. This change is only enabled on 64-bit SMP configuration. On a 4-socket Haswell-EX box with 144 threads (HT on), running the AIM7 compute workload (1500 users) on a 4.8-rc1 kernel (HZ=1000) with and without the patch has the following performance numbers (with HPET or TSC as clock source): TSC = 1042431 jobs/min HPET w/o patch = 798068 jobs/min HPET with patch = 1029445 jobs/min The perf profile showed a reduction of the %CPU time consumed by read_hpet from 11.19% without patch to 1.24% with patch. [ tglx: It's really sad that we need to have such hacks just to deal with the fact that cpu vendors have not managed to fix the TSC wreckage within 15+ years. Were They Forgetting? ] Signed-off-by: Waiman Long <Waiman.Long@hpe.com> Tested-by: Prarit Bhargava <prarit@redhat.com> Cc: Scott J Norton <scott.norton@hpe.com> Cc: Douglas Hatch <doug.hatch@hpe.com> Cc: Randy Wright <rwright@hpe.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@suse.de> Link: http://lkml.kernel.org/r/1473182530-29175-1-git-send-email-Waiman.Long@hpe.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-09-06 17:22:10 +00:00
#endif
[PATCH] clocksource init adjustments (fix bug #7426) This patch resolves the issue found here: http://bugme.osdl.org/show_bug.cgi?id=7426 The basic summary is: Currently we register most of i386/x86_64 clocksources at module_init time. Then we enable clocksource selection at late_initcall time. This causes some problems for drivers that use gettimeofday for init calibration routines (specifically the es1968 driver in this case), where durring module_init, the only clocksource available is the low-res jiffies clocksource. This may cause slight calibration errors, due to the small sampling time used. It should be noted that drivers that require fine grained time may not function on architectures that do not have better then jiffies resolution timekeeping (there are a few). However, this does not discount the reasonable need for such fine-grained timekeeping at init time. Thus the solution here is to register clocksources earlier (ideally when the hardware is being initialized), and then we enable clocksource selection at fs_initcall (before device_initcall). This patch should probably get some testing time in -mm, since clocksource selection is one of the most important issues for correct timekeeping, and I've only been able to test this on a few of my own boxes. Signed-off-by: John Stultz <johnstul@us.ibm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: "David S. Miller" <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-05 08:30:50 +00:00
static struct clocksource clocksource_hpet = {
.name = "hpet",
.rating = 250,
.read = read_hpet,
.mask = HPET_MASK,
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
.resume = hpet_resume_counter,
[PATCH] clocksource init adjustments (fix bug #7426) This patch resolves the issue found here: http://bugme.osdl.org/show_bug.cgi?id=7426 The basic summary is: Currently we register most of i386/x86_64 clocksources at module_init time. Then we enable clocksource selection at late_initcall time. This causes some problems for drivers that use gettimeofday for init calibration routines (specifically the es1968 driver in this case), where durring module_init, the only clocksource available is the low-res jiffies clocksource. This may cause slight calibration errors, due to the small sampling time used. It should be noted that drivers that require fine grained time may not function on architectures that do not have better then jiffies resolution timekeeping (there are a few). However, this does not discount the reasonable need for such fine-grained timekeeping at init time. Thus the solution here is to register clocksources earlier (ideally when the hardware is being initialized), and then we enable clocksource selection at fs_initcall (before device_initcall). This patch should probably get some testing time in -mm, since clocksource selection is one of the most important issues for correct timekeeping, and I've only been able to test this on a few of my own boxes. Signed-off-by: John Stultz <johnstul@us.ibm.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: "David S. Miller" <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-05 08:30:50 +00:00
};
/*
* AMD SB700 based systems with spread spectrum enabled use a SMM based
* HPET emulation to provide proper frequency setting.
*
* On such systems the SMM code is initialized with the first HPET register
* access and takes some time to complete. During this time the config
* register reads 0xffffffff. We check for max 1000 loops whether the
* config register reads a non-0xffffffff value to make sure that the
* HPET is up and running before we proceed any further.
*
* A counting loop is safe, as the HPET access takes thousands of CPU cycles.
*
* On non-SB700 based machines this check is only done once and has no
* side effects.
*/
static bool __init hpet_cfg_working(void)
{
int i;
for (i = 0; i < 1000; i++) {
if (hpet_readl(HPET_CFG) != 0xFFFFFFFF)
return true;
}
pr_warn("Config register invalid. Disabling HPET\n");
return false;
}
static bool __init hpet_counting(void)
{
u64 start, now, t1;
hpet_restart_counter();
t1 = hpet_readl(HPET_COUNTER);
start = rdtsc();
/*
* We don't know the TSC frequency yet, but waiting for
* 200000 TSC cycles is safe:
* 4 GHz == 50us
* 1 GHz == 200us
*/
do {
if (t1 != hpet_readl(HPET_COUNTER))
return true;
now = rdtsc();
} while ((now - start) < 200000UL);
pr_warn("Counter not counting. HPET disabled\n");
return false;
}
/**
* hpet_enable - Try to setup the HPET timer. Returns 1 on success.
*/
int __init hpet_enable(void)
{
u32 hpet_period, cfg, id, irq;
unsigned int i, channels;
struct hpet_channel *hc;
u64 freq;
if (!is_hpet_capable())
return 0;
hpet_set_mapping();
if (!hpet_virt_address)
return 0;
/* Validate that the config register is working */
if (!hpet_cfg_working())
goto out_nohpet;
/* Validate that the counter is counting */
if (!hpet_counting())
goto out_nohpet;
/*
* Read the period and check for a sane value:
*/
hpet_period = hpet_readl(HPET_PERIOD);
if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
goto out_nohpet;
/* The period is a femtoseconds value. Convert it to a frequency. */
freq = FSEC_PER_SEC;
do_div(freq, hpet_period);
hpet_freq = freq;
/*
* Read the HPET ID register to retrieve the IRQ routing
* information and the number of channels
*/
id = hpet_readl(HPET_ID);
hpet_print_config();
/* This is the HPET channel number which is zero based */
channels = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
/*
* The legacy routing mode needs at least two channels, tick timer
* and the rtc emulation channel.
*/
if (IS_ENABLED(CONFIG_HPET_EMULATE_RTC) && channels < 2)
goto out_nohpet;
hc = kcalloc(channels, sizeof(*hc), GFP_KERNEL);
if (!hc) {
pr_warn("Disabling HPET.\n");
goto out_nohpet;
}
hpet_base.channels = hc;
hpet_base.nr_channels = channels;
/* Read, store and sanitize the global configuration */
cfg = hpet_readl(HPET_CFG);
hpet_base.boot_cfg = cfg;
cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
hpet_writel(cfg, HPET_CFG);
if (cfg)
pr_warn("Global config: Unknown bits %#x\n", cfg);
/* Read, store and sanitize the per channel configuration */
for (i = 0; i < channels; i++, hc++) {
hc->num = i;
cfg = hpet_readl(HPET_Tn_CFG(i));
hc->boot_cfg = cfg;
irq = (cfg & Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
hc->irq = irq;
cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
hpet_writel(cfg, HPET_Tn_CFG(i));
cfg &= ~(HPET_TN_PERIODIC | HPET_TN_PERIODIC_CAP
| HPET_TN_64BIT_CAP | HPET_TN_32BIT | HPET_TN_ROUTE
| HPET_TN_FSB | HPET_TN_FSB_CAP);
if (cfg)
pr_warn("Channel #%u config: Unknown bits %#x\n", i, cfg);
}
hpet_print_config();
clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
if (id & HPET_ID_LEGSUP) {
hpet_legacy_clockevent_register();
hpet_base.channels[0].mode = HPET_MODE_LEGACY;
if (IS_ENABLED(CONFIG_HPET_EMULATE_RTC))
hpet_base.channels[1].mode = HPET_MODE_LEGACY;
return 1;
}
return 0;
out_nohpet:
kfree(hpet_base.channels);
hpet_base.channels = NULL;
hpet_base.nr_channels = 0;
hpet_clear_mapping();
hpet_address = 0;
return 0;
}
/*
* The late initialization runs after the PCI quirks have been invoked
* which might have detected a system on which the HPET can be enforced.
*/
static __init int hpet_late_init(void)
{
int ret;
if (!hpet_address) {
if (!force_hpet_address)
return -ENODEV;
hpet_address = force_hpet_address;
hpet_enable();
}
if (!hpet_virt_address)
return -ENODEV;
if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
hpet_msi_capability_lookup(2);
else
hpet_msi_capability_lookup(0);
hpet_reserve_platform_timers();
hpet_print_config();
if (hpet_msi_disable)
return 0;
if (boot_cpu_has(X86_FEATURE_ARAT))
return 0;
ret = cpuhp_setup_state(CPUHP_AP_X86_HPET_ONLINE, "x86/hpet:online",
hpet_cpuhp_online, NULL);
if (ret)
return ret;
ret = cpuhp_setup_state(CPUHP_X86_HPET_DEAD, "x86/hpet:dead", NULL,
hpet_cpuhp_dead);
if (ret)
goto err_cpuhp;
return 0;
err_cpuhp:
cpuhp_remove_state(CPUHP_AP_X86_HPET_ONLINE);
return ret;
}
fs_initcall(hpet_late_init);
void hpet_disable(void)
{
unsigned int i;
u32 cfg;
if (!is_hpet_capable() || !hpet_virt_address)
return;
/* Restore boot configuration with the enable bit cleared */
cfg = hpet_base.boot_cfg;
cfg &= ~HPET_CFG_ENABLE;
hpet_writel(cfg, HPET_CFG);
/* Restore the channel boot configuration */
for (i = 0; i < hpet_base.nr_channels; i++)
hpet_writel(hpet_base.channels[i].boot_cfg, HPET_Tn_CFG(i));
/* If the HPET was enabled at boot time, reenable it */
if (hpet_base.boot_cfg & HPET_CFG_ENABLE)
hpet_writel(hpet_base.boot_cfg, HPET_CFG);
}
#ifdef CONFIG_HPET_EMULATE_RTC
/*
* HPET in LegacyReplacement mode eats up the RTC interrupt line. When HPET
* is enabled, we support RTC interrupt functionality in software.
*
* RTC has 3 kinds of interrupts:
*
* 1) Update Interrupt - generate an interrupt, every second, when the
* RTC clock is updated
* 2) Alarm Interrupt - generate an interrupt at a specific time of day
* 3) Periodic Interrupt - generate periodic interrupt, with frequencies
* 2Hz-8192Hz (2Hz-64Hz for non-root user) (all frequencies in powers of 2)
*
* (1) and (2) above are implemented using polling at a frequency of 64 Hz:
* DEFAULT_RTC_INT_FREQ.
*
* The exact frequency is a tradeoff between accuracy and interrupt overhead.
*
* For (3), we use interrupts at 64 Hz, or the user specified periodic frequency,
* if it's higher.
*/
#include <linux/mc146818rtc.h>
#include <linux/rtc.h>
#define DEFAULT_RTC_INT_FREQ 64
#define DEFAULT_RTC_SHIFT 6
#define RTC_NUM_INTS 1
static unsigned long hpet_rtc_flags;
static int hpet_prev_update_sec;
static struct rtc_time hpet_alarm_time;
static unsigned long hpet_pie_count;
static u32 hpet_t1_cmp;
static u32 hpet_default_delta;
static u32 hpet_pie_delta;
static unsigned long hpet_pie_limit;
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
static rtc_irq_handler irq_handler;
/*
* Check that the HPET counter c1 is ahead of c2
*/
static inline int hpet_cnt_ahead(u32 c1, u32 c2)
{
return (s32)(c2 - c1) < 0;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
/*
* Registers a IRQ handler.
*/
int hpet_register_irq_handler(rtc_irq_handler handler)
{
if (!is_hpet_enabled())
return -ENODEV;
if (irq_handler)
return -EBUSY;
irq_handler = handler;
return 0;
}
EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
/*
* Deregisters the IRQ handler registered with hpet_register_irq_handler()
* and does cleanup.
*/
void hpet_unregister_irq_handler(rtc_irq_handler handler)
{
if (!is_hpet_enabled())
return;
irq_handler = NULL;
hpet_rtc_flags = 0;
}
EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
/*
* Timer 1 for RTC emulation. We use one shot mode, as periodic mode
* is not supported by all HPET implementations for timer 1.
*
* hpet_rtc_timer_init() is called when the rtc is initialized.
*/
int hpet_rtc_timer_init(void)
{
unsigned int cfg, cnt, delta;
unsigned long flags;
if (!is_hpet_enabled())
return 0;
if (!hpet_default_delta) {
uint64_t clc;
clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
hpet_default_delta = clc;
}
if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
delta = hpet_default_delta;
else
delta = hpet_pie_delta;
local_irq_save(flags);
cnt = delta + hpet_readl(HPET_COUNTER);
hpet_writel(cnt, HPET_T1_CMP);
hpet_t1_cmp = cnt;
cfg = hpet_readl(HPET_T1_CFG);
cfg &= ~HPET_TN_PERIODIC;
cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
hpet_writel(cfg, HPET_T1_CFG);
local_irq_restore(flags);
return 1;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
static void hpet_disable_rtc_channel(void)
{
u32 cfg = hpet_readl(HPET_T1_CFG);
cfg &= ~HPET_TN_ENABLE;
hpet_writel(cfg, HPET_T1_CFG);
}
/*
* The functions below are called from rtc driver.
* Return 0 if HPET is not being used.
* Otherwise do the necessary changes and return 1.
*/
int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
{
if (!is_hpet_enabled())
return 0;
hpet_rtc_flags &= ~bit_mask;
if (unlikely(!hpet_rtc_flags))
hpet_disable_rtc_channel();
return 1;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
int hpet_set_rtc_irq_bit(unsigned long bit_mask)
{
unsigned long oldbits = hpet_rtc_flags;
if (!is_hpet_enabled())
return 0;
hpet_rtc_flags |= bit_mask;
if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
hpet_prev_update_sec = -1;
if (!oldbits)
hpet_rtc_timer_init();
return 1;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
int hpet_set_alarm_time(unsigned char hrs, unsigned char min, unsigned char sec)
{
if (!is_hpet_enabled())
return 0;
hpet_alarm_time.tm_hour = hrs;
hpet_alarm_time.tm_min = min;
hpet_alarm_time.tm_sec = sec;
return 1;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
int hpet_set_periodic_freq(unsigned long freq)
{
uint64_t clc;
if (!is_hpet_enabled())
return 0;
if (freq <= DEFAULT_RTC_INT_FREQ) {
hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
} else {
clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
do_div(clc, freq);
clc >>= hpet_clockevent.shift;
hpet_pie_delta = clc;
x86, hpet: Fix bug in RTC emulation We think there exists a bug in the HPET code that emulates the RTC. In the normal case, when the RTC frequency is set, the rtc driver tells the hpet code about it here: int hpet_set_periodic_freq(unsigned long freq) { uint64_t clc; if (!is_hpet_enabled()) return 0; if (freq <= DEFAULT_RTC_INT_FREQ) hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq; else { clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC; do_div(clc, freq); clc >>= hpet_clockevent.shift; hpet_pie_delta = (unsigned long) clc; } return 1; } If freq is set to 64Hz (DEFAULT_RTC_INT_FREQ) or lower, then hpet_pie_limit (a static) is set to non-zero. Then, on every one-shot HPET interrupt, hpet_rtc_timer_reinit is called to compute the next timeout. Well, that function has this logic: if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit) delta = hpet_default_delta; else delta = hpet_pie_delta; Since hpet_pie_limit is not 0, hpet_default_delta is used. That corresponds to 64Hz. Now, if you set a different rtc frequency, you'll take the else path through hpet_set_periodic_freq, but unfortunately no one resets hpet_pie_limit back to 0. Boom....now you are stuck with 64Hz RTC interrupts forever. The patch below just resets the hpet_pie_limit value when requested freq is greater than DEFAULT_RTC_INT_FREQ, which we think fixes this problem. Signed-off-by: Alok N Kataria <akataria@vmware.com> LKML-Reference: <201003112200.o2BM0Hre012875@imap1.linux-foundation.org> Signed-off-by: Daniel Hecht <dhecht@vmware.com> Cc: Venkatesh Pallipadi <venkatesh.pallipadi@gmail.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
2010-03-11 22:00:16 +00:00
hpet_pie_limit = 0;
}
return 1;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
int hpet_rtc_dropped_irq(void)
{
return is_hpet_enabled();
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
static void hpet_rtc_timer_reinit(void)
{
unsigned int delta;
int lost_ints = -1;
if (unlikely(!hpet_rtc_flags))
hpet_disable_rtc_channel();
if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
delta = hpet_default_delta;
else
delta = hpet_pie_delta;
/*
* Increment the comparator value until we are ahead of the
* current count.
*/
do {
hpet_t1_cmp += delta;
hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
lost_ints++;
} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
if (lost_ints) {
if (hpet_rtc_flags & RTC_PIE)
hpet_pie_count += lost_ints;
if (printk_ratelimit())
pr_warn("Lost %d RTC interrupts\n", lost_ints);
}
}
irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
{
struct rtc_time curr_time;
unsigned long rtc_int_flag = 0;
hpet_rtc_timer_reinit();
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
memset(&curr_time, 0, sizeof(struct rtc_time));
if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
x86/hpet: Fix /dev/rtc breakage caused by RTC cleanup Ville Syrjälä reports "The first time I run hwclock after rebooting I get this: open("/dev/rtc", O_RDONLY) = 3 ioctl(3, PHN_SET_REGS or RTC_UIE_ON, 0) = 0 select(4, [3], NULL, NULL, {10, 0}) = 0 (Timeout) ioctl(3, PHN_NOT_OH or RTC_UIE_OFF, 0) = 0 close(3) = 0 On all subsequent runs I get this: open("/dev/rtc", O_RDONLY) = 3 ioctl(3, PHN_SET_REGS or RTC_UIE_ON, 0) = -1 EINVAL (Invalid argument) ioctl(3, RTC_RD_TIME, 0x7ffd76b3ae70) = -1 EINVAL (Invalid argument) close(3) = 0" This was caused by a stupid typo in a patch that should have been a simple rename to move around contents of a header file, but accidentally wrote zeroes into the rtc rather than reading from it: 463a86304cae ("char/genrtc: x86: remove remnants of asm/rtc.h") Reported-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Tested-by: Jarkko Nikula <jarkko.nikula@linux.intel.com> Tested-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Arnd Bergmann <arnd@arndb.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: Alexandre Belloni <alexandre.belloni@free-electrons.com> Cc: Borislav Petkov <bp@suse.de> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: rtc-linux@googlegroups.com Fixes: 463a86304cae ("char/genrtc: x86: remove remnants of asm/rtc.h") Link: http://lkml.kernel.org/r/20160809195528.1604312-1-arnd@arndb.de Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-09 19:54:53 +00:00
mc146818_get_time(&curr_time);
if (hpet_rtc_flags & RTC_UIE &&
curr_time.tm_sec != hpet_prev_update_sec) {
if (hpet_prev_update_sec >= 0)
rtc_int_flag = RTC_UF;
hpet_prev_update_sec = curr_time.tm_sec;
}
if (hpet_rtc_flags & RTC_PIE && ++hpet_pie_count >= hpet_pie_limit) {
rtc_int_flag |= RTC_PF;
hpet_pie_count = 0;
}
if (hpet_rtc_flags & RTC_AIE &&
(curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
(curr_time.tm_min == hpet_alarm_time.tm_min) &&
(curr_time.tm_hour == hpet_alarm_time.tm_hour))
rtc_int_flag |= RTC_AF;
if (rtc_int_flag) {
rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
if (irq_handler)
irq_handler(rtc_int_flag, dev_id);
}
return IRQ_HANDLED;
}
x86, rtc: make CONFIG_HPET_EMULATE_RTC usable from modules enabled, then interrupts don't work for the rtc-cmos driver which results in RTC_AIE*, RTC_PIE* and RTC_ALM being unusable. This affects hwclock from util-linux-ng at least on i386 since that uses RTC_PIE_ON. (For x86-64, a polling method is used for unknown reasons.) This patch series now 1. export the functions from arch/x86/kernel/hpet.c that the old char/rtc driver uses to work around that problem, 2. makes it possible to compile the old rtc driver as module, while still having CONFIG_HPET_EMULATE_RTC enabled and 3. makes use of the exported functions in (1) in the new rtc-cmos driver. This patch: This patch makes the RTC emulation functions in arch/x86/kernel/hpet.c usable for kernel modules. It - exports the functions (EXPORT_SYMBOL_GPL()), - adds an interface to register the interrupt callback function instead of using only a fixed callback function and - replaces the rtc_get_rtc_time() function which depends on CONFIG_RTC with a call to get_rtc_time() which is defined in include/asm-generic/rtc.h. The only dependency to CONFIG_RTC is the call to rtc_interrupt() which is removed by the next patch. After this, there's no (code) dependency of this functions to CONFIG_RTC=y any more. Signed-off-by: Bernhard Walle <bwalle@suse.de> Cc: Alessandro Zummo <a.zummo@towertech.it> Cc: David Brownell <david-b@pacbell.net> Cc: Andi Kleen <ak@suse.de> Cc: john stultz <johnstul@us.ibm.com> Cc: Robert Picco <Robert.Picco@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:28 +00:00
EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
#endif