Skip to contentSkip to conversation

/dev/hpet - fixes and cleanup

PATCH In review David Brownell 4 participants 17 messages Jul 22, 2008 – Jul 31, 2008

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 /dev/hpet - fixes and cleanup

From: David Brownell <[email protected]>

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) bitwidth, not *timer* bitwidths (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
+ only open comparators that have an interrupt, and can thus perform "real work"
+ 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.

I get the feeling few folk use this userspace interface, else those nonfunctional IRQs would be causing more trouble.

---
I CC'd everyone who MAINTAINERS says maintains HPET. Odd to have four entries!!

And re that kernel interface: IMO, worth having a relatively generic interface for hardware timers instead of inventing a new interface for each new bit of hardware. Embedded systems tend to have at least half a dozen SOC timers available...

 Documentation/hpet.txt |   43 ++++++++++++-------------
 arch/x86/kernel/hpet.c |    6 ++-
 drivers/char/hpet.c    |   82 +++++++++++--------------------------------------
 include/linux/hpet.h   |   16 ++-------
 4 files changed, 50 insertions(+), 97 deletions(-)
Documentation/hpet.txt +21 -22
@@ -1,21 +1,32 @@
 		High Precision Event Timer Driver for Linux
 
-The High Precision Event Timer (HPET) hardware is the future replacement
-for the 8254 and Real Time Clock (RTC) periodic timer functionality.
-Each HPET can have up to 32 timers.  It is possible to configure the
-first two timers as legacy replacements for 8254 and RTC periodic timers.
-A specification done by Intel and Microsoft can be found at
-<http://www.intel.com/technology/architecture/hpetspec.htm>.
+The High Precision Event Timer (HPET) hardware follows a specification
+by Intel and Microsoft which can be found at
+
+	http://www.intel.com/technology/architecture/hpetspec.htm
+
+Each HPET has one fixed-rate counter (at 10+ MHz, hence "High Precision")
+and up to 32 comparators.  Normally three or more comparators are provided,
+each of which can generate oneshot interupts and at least one of which has
+additional hardware to support periodic interrupts.  The comparators are
+also called "timers", which can be misleading since usually timers are
+independent of each other ... these share a counter, complicating resets.
+
+HPET devices can support two interrupt routing modes.  In one mode, the
+comparators are additional interrupt sources with no particular system
+role.  Many x86 BIOS writers don't route HPET interrupts at all, which
+prevents use of that mode.  They support the other "legacy replacement"
+mode where the first two comparators block interrupts from 8254 timers
+and from the RTC.
 
 The driver supports detection of HPET driver allocation and initialization
 of the HPET before the driver module_init routine is called.  This enables
 platform code which uses timer 0 or 1 as the main timer to intercept HPET
 initialization.  An example of this initialization can be found in
-arch/i386/kernel/time_hpet.c.
+arch/x86/kernel/hpet.c.
 
-The driver provides two APIs which are very similar to the API found in
-the rtc.c driver.  There is a user space API and a kernel space API.
-An example user space program is provided below.
+The driver provides a userspace API which resembles the API found in the
+RTC driver framework.  An example user space program is provided below.
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -286,15 +297,3 @@ out:
 
 	return;
 }
-
-The kernel API has three interfaces exported from the driver:
-
-	hpet_register(struct hpet_task *tp, int periodic)
-	hpet_unregister(struct hpet_task *tp)
-	hpet_control(struct hpet_task *tp, unsigned int cmd, unsigned long arg)
-
-The kernel module using this interface fills in the ht_func and ht_data
-members of the hpet_task structure before calling hpet_register.
-hpet_control simply vectors to the hpet_ioctl routine and has the same
-commands and respective arguments as the user API.  hpet_unregister
-is used to terminate usage of the HPET timer reserved by hpet_register.
arch/x86/kernel/hpet.c +5 -1
@@ -127,13 +127,17 @@ static void hpet_reserve_platform_timers
 	hd.hd_phys_address = hpet_address;
 	hd.hd_address = hpet;
 	hd.hd_nirqs = nrtimers;
-	hd.hd_flags = HPET_DATA_PLATFORM;
 	hpet_reserve_timer(&hd, 0);
 
 #ifdef CONFIG_HPET_EMULATE_RTC
 	hpet_reserve_timer(&hd, 1);
 #endif
 
+	/*
+	 * 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;
 
drivers/char/hpet.c +19 -63

136 lines not on this page · raw message

include/linux/hpet.h +1 -10
@@ -91,23 +91,14 @@ struct hpet {
  * exported interfaces
  */
 
-struct hpet_task {
-	void (*ht_func) (void *);
-	void *ht_data;
-	void *ht_opaque;
-};
-
 struct hpet_data {
 	unsigned long hd_phys_address;
 	void __iomem *hd_address;
 	unsigned short hd_nirqs;
-	unsigned short hd_flags;
 	unsigned int hd_state;	/* timer allocated */
 	unsigned int hd_irq[HPET_MAX_TIMERS];
 };
 
-#define	HPET_DATA_PLATFORM	0x0001	/* platform call to hpet_alloc */
-
 static inline void hpet_reserve_timer(struct hpet_data *hd, int timer)
 {
 	hd->hd_state |= (1 << timer);
@@ -125,7 +119,7 @@ struct hpet_info {
 	unsigned short hi_timer;
 };
 
-#define	HPET_INFO_PERIODIC	0x0001	/* timer is periodic */
+#define HPET_INFO_PERIODIC	0x0010	/* periodic-capable comparator */
 
 #define	HPET_IE_ON	_IO('h', 0x01)	/* interrupt on */
 #define	HPET_IE_OFF	_IO('h', 0x02)	/* interrupt off */
maintainer
#1 lore raw
Quoted 4 lines · David Brownell

* Tighten and correct the userspace interface code
...
+ only open comparators that have an interrupt, and can thus perform "real work"

This prevents userspace applications from doing mmap() on /dev/hpet even if there is no interrupt.

This seems to be the only part of the userspace interface that is used in practice. Because of the availability of POSIX timers, it might make sense to deprecate the HPET ioctl interface.

Quoted 3 lines

And re that kernel interface: IMO, worth having a relatively generic interface for hardware timers instead of inventing a new interface for each new bit of hardware.

AFAIK it was intended to be similar to the RTC callback interface, but that was before the generic high-precision timer stuff was introduced.

Regards,
Clemens

#2 lore raw
Quoted 8 lines · Clemens Ladisch
Quoted 4 lines · David Brownell

* Tighten and correct the userspace interface code
...
+ only open comparators that have an interrupt, and can thus perform "real work"

This prevents userspace applications from doing mmap() on /dev/hpet even if there is no interrupt.

OK, I removed that ... HPET_IE_ON will already fail.

I didn't find any software using /dev/hpet, except for the example in Documentation/hpet.txt ... presumably I was looking in the wrong place. I'd surely have retched at anything mmapping hardware registers though. ;)

Quoted 3 lines

This seems to be the only part of the userspace interface that is used in practice. Because of the availability of POSIX timers, it might make sense to deprecate the HPET ioctl interface.

I'll leave that part up to someone else. If POSIX timers are a sufficient userspace interface, great ... then that mmap son't really be needed either! In fact, would the /dev/hpet support even be needed? It's not very portable...

Quoted 6 lines
Quoted 3 lines

And re that kernel interface: IMO, worth having a relatively generic interface for hardware timers instead of inventing a new interface for each new bit of hardware.

AFAIK it was intended to be similar to the RTC callback interface, but that was before the generic high-precision timer stuff was introduced.

People seem to want access to the hardware backed timers too, and not necessarily coupled to a task like the RTC stuff assumes. Folk who want to drive that issue will need to sort out requirements... the requests I've heard seem to focus on talking to kernel code.

But since HPET "timers" are really weak compared to what most embedded platforms offer, I'm quite sure any API designed to its capabilities would be underfeatured! Example, there are no external inputs (like clocks or event pulses) or outputs (PWM or whatever) with HPET, and likewise no flexibility about inputs (which internal clock, or maybe external clocks).

- Dave

#3 lore raw
Quoted 11 lines · David Brownell
Quoted 8 lines · Clemens Ladisch
Quoted 4 lines · David Brownell

  * Tighten and correct the userspace interface code
      ...
      + only open comparators that have an interrupt, and can thus perform "real work"

This prevents userspace applications from doing mmap() on /dev/hpet even if there is no interrupt.

OK, I removed that ... HPET_IE_ON will already fail.

This change will be included in the next version of this patch, which I'll send after other responses get more of a chance to trickle in. (That line already changed because of the hd_task elimination.)

drivers/char/hpet.c +1 -2
@@ -189,8 +189,7 @@ static int hpet_open(struct inode *inode
 
 	for (devp = NULL, hpetp = hpets; hpetp && !devp; hpetp = hpetp->hp_next)
 		for (i = 0; i < hpetp->hp_ntimer; i++)
-			if (hpetp->hp_dev[i].hd_flags & HPET_OPEN
-					|| !hpetp->hp_dev[i].hd_hdwirq)
+			if (hpetp->hp_dev[i].hd_flags & HPET_OPEN)
 				continue;
 			else {
 				devp = &hpetp->hp_dev[i];
maintainer
#4 lore raw
In reply to David Brownell
Quoted 4 lines · David Brownell

I didn't find any software using /dev/hpet, except for the example in Documentation/hpet.txt ... presumably I was looking in the wrong place. I'd surely have retched at anything mmapping hardware registers though. ;)

http://subversion.jackaudio.org/jack/trunk/jack/config/os/gnu-linux/time.c

Quoted 7 lines
Quoted 3 lines

This seems to be the only part of the userspace interface that is used in practice. Because of the availability of POSIX timers, it might make sense to deprecate the HPET ioctl interface.

I'll leave that part up to someone else. If POSIX timers are a sufficient userspace interface, great ... then that mmap son't really be needed either!

The idea is to be able to get a high-precision timer value without doing a syscall. (Whether the syscall overhead actually matters in a specific application is another question.)

Regards,
Clemens

#5 lore raw
Quoted 6 lines · Clemens Ladisch
Quoted 4 lines · David Brownell
Quoted 3 lines

This seems to be the only part of the userspace interface that is used in practice. Because of the availability of POSIX timers, it might make sense to deprecate the HPET ioctl interface.

Quoted 7 lines
Quoted 3 lines

I'll leave that part up to someone else. If POSIX timers are a sufficient userspace interface, great ... then that mmap son't really be needed either!

The idea is to be able to get a high-precision timer value without doing a syscall. (Whether the syscall overhead actually matters in a specific application is another question.)

On x86_64 the vsyscall stuff should kick in for gettimeofday() when HPET is in use, eliminating the need for /dev/hpet mmapping.

I don't really know vsyscalls ... but if that could be done on i386 systems, the last argument for /dev/hpet would seem gone...

#6 lore raw
Quoted 7 lines · David Brownell

On x86_64 the vsyscall stuff should kick in for gettimeofday() when HPET is in use, eliminating the need for /dev/hpet mmapping.

I don't really know vsyscalls ... but if that could be done on i386 systems, the last argument for /dev/hpet would seem gone...

i386 already uses vsyscalls.

-hpa

#7 lore raw

PATCH 1/1 /dev/hpet - fixes and cleanup

In reply to Clemens Ladisch

From: David Brownell <[email protected]>

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) bitwidth, not *timer* bitwidths (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.

I get the feeling few folk use this userspace interface, else those nonfunctional IRQs would be causing more trouble. (And I'm told the main current user is probably Jack, through mmap, for what it seems the gettimeofday vsyscall will do on x86_64.)

---
Refreshed to address the issue Clemens noted, and to cope with a recent partial fix for the dead code issue.

 Documentation/hpet.txt |   43 ++++++++++++------------
 arch/x86/kernel/hpet.c |    6 ++-
 drivers/char/hpet.c    |   85 ++++++++++---------------------------------------
 include/linux/hpet.h   |   11 ------
 4 files changed, 46 insertions(+), 99 deletions(-)
Documentation/hpet.txt +21 -22
@@ -1,21 +1,32 @@
 		High Precision Event Timer Driver for Linux
 
-The High Precision Event Timer (HPET) hardware is the future replacement
-for the 8254 and Real Time Clock (RTC) periodic timer functionality.
-Each HPET can have up to 32 timers.  It is possible to configure the
-first two timers as legacy replacements for 8254 and RTC periodic timers.
-A specification done by Intel and Microsoft can be found at
-<http://www.intel.com/technology/architecture/hpetspec.htm>.
+The High Precision Event Timer (HPET) hardware follows a specification
+by Intel and Microsoft which can be found at
+
+	http://www.intel.com/technology/architecture/hpetspec.htm
+
+Each HPET has one fixed-rate counter (at 10+ MHz, hence "High Precision")
+and up to 32 comparators.  Normally three or more comparators are provided,
+each of which can generate oneshot interupts and at least one of which has
+additional hardware to support periodic interrupts.  The comparators are
+also called "timers", which can be misleading since usually timers are
+independent of each other ... these share a counter, complicating resets.
+
+HPET devices can support two interrupt routing modes.  In one mode, the
+comparators are additional interrupt sources with no particular system
+role.  Many x86 BIOS writers don't route HPET interrupts at all, which
+prevents use of that mode.  They support the other "legacy replacement"
+mode where the first two comparators block interrupts from 8254 timers
+and from the RTC.
 
 The driver supports detection of HPET driver allocation and initialization
 of the HPET before the driver module_init routine is called.  This enables
 platform code which uses timer 0 or 1 as the main timer to intercept HPET
 initialization.  An example of this initialization can be found in
-arch/i386/kernel/time_hpet.c.
+arch/x86/kernel/hpet.c.
 
-The driver provides two APIs which are very similar to the API found in
-the rtc.c driver.  There is a user space API and a kernel space API.
-An example user space program is provided below.
+The driver provides a userspace API which resembles the API found in the
+RTC driver framework.  An example user space program is provided below.
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -286,15 +297,3 @@ out:
 
 	return;
 }
-
-The kernel API has three interfaces exported from the driver:
-
-	hpet_register(struct hpet_task *tp, int periodic)
-	hpet_unregister(struct hpet_task *tp)
-	hpet_control(struct hpet_task *tp, unsigned int cmd, unsigned long arg)
-
-The kernel module using this interface fills in the ht_func and ht_data
-members of the hpet_task structure before calling hpet_register.
-hpet_control simply vectors to the hpet_ioctl routine and has the same
-commands and respective arguments as the user API.  hpet_unregister
-is used to terminate usage of the HPET timer reserved by hpet_register.
arch/x86/kernel/hpet.c +5 -1
@@ -115,13 +115,17 @@ static void hpet_reserve_platform_timers
 	hd.hd_phys_address = hpet_address;
 	hd.hd_address = hpet;
 	hd.hd_nirqs = nrtimers;
-	hd.hd_flags = HPET_DATA_PLATFORM;
 	hpet_reserve_timer(&hd, 0);
 
 #ifdef CONFIG_HPET_EMULATE_RTC
 	hpet_reserve_timer(&hd, 1);
 #endif
 
+	/*
+	 * 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;
 
drivers/char/hpet.c +19 -66

139 lines not on this page · raw message

include/linux/hpet.h +1 -10
@@ -91,23 +91,14 @@ struct hpet {
  * exported interfaces
  */
 
-struct hpet_task {
-	void (*ht_func) (void *);
-	void *ht_data;
-	void *ht_opaque;
-};
-
 struct hpet_data {
 	unsigned long hd_phys_address;
 	void __iomem *hd_address;
 	unsigned short hd_nirqs;
-	unsigned short hd_flags;
 	unsigned int hd_state;	/* timer allocated */
 	unsigned int hd_irq[HPET_MAX_TIMERS];
 };
 
-#define	HPET_DATA_PLATFORM	0x0001	/* platform call to hpet_alloc */
-
 static inline void hpet_reserve_timer(struct hpet_data *hd, int timer)
 {
 	hd->hd_state |= (1 << timer);
@@ -125,7 +116,7 @@ struct hpet_info {
 	unsigned short hi_timer;
 };
 
-#define	HPET_INFO_PERIODIC	0x0001	/* timer is periodic */
+#define HPET_INFO_PERIODIC	0x0010	/* periodic-capable comparator */
 
 #define	HPET_IE_ON	_IO('h', 0x01)	/* interrupt on */
 #define	HPET_IE_OFF	_IO('h', 0x02)	/* interrupt off */
maintainer
#8 lore raw

drivers/char/hpet.c static int hpet_ioctl_ieon(struct hpet_d · line 456

Quoted 12 lines · David Brownell

...
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.
...

drivers/char/hpet.c +6 -0
+
+		/* FIXME this may trash both the system clocksource and
+		 * the current clock event device!  Use HPET_TN_SETVAL
+		 * instead, like arch/x86/kernel/hpet.c does ... never
+		 * modify the counter, ever.
+		 */
 		m = read_counter(&hpet->hpet_mc);
 		write_counter(t + m + hpetp->hp_delta, &timer->hpet_compare);

This comment seems to assume that the code below modifies the main counter, which it doesn't. Additionally, HPET_TN_SETVAL has the same value as Tn_VAL_SET_CNF_MASK (from <linux/hpet.h>), which _is_ used.

Regards,
Clemens

#9 lore raw
Quoted 10 lines · Clemens Ladisch
Quoted 7 lines

+             /* FIXME this may trash both the system clocksource and +              * the current clock event device!  Use HPET_TN_SETVAL +              * instead, like arch/x86/kernel/hpet.c does ... never +              * modify the counter, ever.
+              */
              m = read_counter(&hpet->hpet_mc);
              write_counter(t + m + hpetp->hp_delta, &timer->hpet_compare);

This comment seems to assume that the code below modifies the main counter, which it doesn't.

There's only one counter. How could it not modify that?

Oh ... I see. It's called write_counter() but doesn't actually write the counter. Likewise, read_counter() is not actually reading the counter. Gaack ...

So that's not really an issue (good!). I'll strike that comment, except to comment that it's explicitly not modifying any counter (just a hidden write-only accumulator) ... but those silly function names should really be changed so they have a less tenuous connection with reality. In fact, most places would be better off just hard-wiring 32-bit access...

Quoted 2 lines

Additionally, HPET_TN_SETVAL has the same
value as Tn_VAL_SET_CNF_MASK (from <linux/hpet.h>), which _is_ used.

OK, I can see that too. This HPET stuff is really a lot dirtier than I had expected ... there's no reason at all to have two separate headers with two incompatible sets of definitions for the same registers!!

Any comments on the rest?

- Dave

maintainer
#10 lore raw
Quoted 3 lines · David Brownell

Oh ... I see. It's called write_counter() but doesn't actually write the counter. Likewise, read_counter() is not actually reading the counter. Gaack ...

A more correct name would be something like write_a_register_that_has_the_same_size_as_a_counter_register().

Quoted 2 lines

In fact, most places would be better off just hard-wiring 32-bit access...

... and the definition of read/write_counter() depends on the word size of the CPU architecture and has nothing to do with the actual size of the HPET registers ...

I'm hoping that I can deprecate and delete much of this code without having to clean it up first.

Quoted 1 line

Any comments on the rest?

Looks fine. I would have preferred a series of smaller patches, but my feelings on this are not strong enough to split it myself or to ask you to do it.

Regards,
Clemens

#11 lore raw
Quoted 7 lines · Clemens Ladisch
Quoted 3 lines · David Brownell

Oh ... I see. It's called write_counter() but doesn't actually write the counter. Likewise, read_counter() is not actually reading the counter. Gaack ...

A more correct name would be something like write_a_register_that_has_the_same_size_as_a_counter_register().

writel() would be my choice ... :)

I don't think anything in that code really needs to modify the upper 32 bits of those registers. And nothing except the ia64 HPET clocksource needs to read them either.

Quoted 2 lines

I'm hoping that I can deprecate and delete much of this code without having to clean it up first.

I'm not sure how practical that goal is, but I agree that there seems to be no point to /dev/hpet ... certainly, you could start by deprecating it and making sure distroes don't enable it.

Quoted 5 lines
Quoted 1 line

Any comments on the rest?

Looks fine. I would have preferred a series of smaller patches, but my feelings on this are not strong enough to split it myself or to ask you to do it.

OK, I'll send along the updated patch (against 2.6.27-rc1).

- Dave

#12 lore raw

PATCH 1/1 /dev/hpet - fixes and cleanup

In reply to Clemens Ladisch

From: David Brownell <[email protected]>

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.

---
I CC'd everyone who MAINTAINERS says maintains HPET. Odd to have four entries!!

And re that kernel interface ... IMO, worth having a relatively generic interface for hardware timers instead of inventing a new interface for each new bit of hardware. Embedded systems tend to have at least half a dozen timers to spare.

 Documentation/hpet.txt |   43 +++++++++++------------
 arch/x86/kernel/hpet.c |    6 ++-
 drivers/char/hpet.c    |   90 +++++++++++++------------------------------------
 include/linux/hpet.h   |   11 -----
 4 files changed, 51 insertions(+), 99 deletions(-)
Documentation/hpet.txt +21 -22
@@ -1,21 +1,32 @@
 		High Precision Event Timer Driver for Linux
 
-The High Precision Event Timer (HPET) hardware is the future replacement
-for the 8254 and Real Time Clock (RTC) periodic timer functionality.
-Each HPET can have up to 32 timers.  It is possible to configure the
-first two timers as legacy replacements for 8254 and RTC periodic timers.
-A specification done by Intel and Microsoft can be found at
-<http://www.intel.com/technology/architecture/hpetspec.htm>.
+The High Precision Event Timer (HPET) hardware follows a specification
+by Intel and Microsoft which can be found at
+
+	http://www.intel.com/technology/architecture/hpetspec.htm
+
+Each HPET has one fixed-rate counter (at 10+ MHz, hence "High Precision")
+and up to 32 comparators.  Normally three or more comparators are provided,
+each of which can generate oneshot interupts and at least one of which has
+additional hardware to support periodic interrupts.  The comparators are
+also called "timers", which can be misleading since usually timers are
+independent of each other ... these share a counter, complicating resets.
+
+HPET devices can support two interrupt routing modes.  In one mode, the
+comparators are additional interrupt sources with no particular system
+role.  Many x86 BIOS writers don't route HPET interrupts at all, which
+prevents use of that mode.  They support the other "legacy replacement"
+mode where the first two comparators block interrupts from 8254 timers
+and from the RTC.
 
 The driver supports detection of HPET driver allocation and initialization
 of the HPET before the driver module_init routine is called.  This enables
 platform code which uses timer 0 or 1 as the main timer to intercept HPET
 initialization.  An example of this initialization can be found in
-arch/i386/kernel/time_hpet.c.
+arch/x86/kernel/hpet.c.
 
-The driver provides two APIs which are very similar to the API found in
-the rtc.c driver.  There is a user space API and a kernel space API.
-An example user space program is provided below.
+The driver provides a userspace API which resembles the API found in the
+RTC driver framework.  An example user space program is provided below.
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -286,15 +297,3 @@ out:
 
 	return;
 }
-
-The kernel API has three interfaces exported from the driver:
-
-	hpet_register(struct hpet_task *tp, int periodic)
-	hpet_unregister(struct hpet_task *tp)
-	hpet_control(struct hpet_task *tp, unsigned int cmd, unsigned long arg)
-
-The kernel module using this interface fills in the ht_func and ht_data
-members of the hpet_task structure before calling hpet_register.
-hpet_control simply vectors to the hpet_ioctl routine and has the same
-commands and respective arguments as the user API.  hpet_unregister
-is used to terminate usage of the HPET timer reserved by hpet_register.
arch/x86/kernel/hpet.c +5 -1
@@ -115,13 +115,17 @@ static void hpet_reserve_platform_timers
 	hd.hd_phys_address = hpet_address;
 	hd.hd_address = hpet;
 	hd.hd_nirqs = nrtimers;
-	hd.hd_flags = HPET_DATA_PLATFORM;
 	hpet_reserve_timer(&hd, 0);
 
 #ifdef CONFIG_HPET_EMULATE_RTC
 	hpet_reserve_timer(&hd, 1);
 #endif
 
+	/*
+	 * 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;
 
drivers/char/hpet.c +24 -66

150 lines not on this page · raw message

include/linux/hpet.h +1 -10
@@ -91,23 +91,14 @@ struct hpet {
  * exported interfaces
  */
 
-struct hpet_task {
-	void (*ht_func) (void *);
-	void *ht_data;
-	void *ht_opaque;
-};
-
 struct hpet_data {
 	unsigned long hd_phys_address;
 	void __iomem *hd_address;
 	unsigned short hd_nirqs;
-	unsigned short hd_flags;
 	unsigned int hd_state;	/* timer allocated */
 	unsigned int hd_irq[HPET_MAX_TIMERS];
 };
 
-#define	HPET_DATA_PLATFORM	0x0001	/* platform call to hpet_alloc */
-
 static inline void hpet_reserve_timer(struct hpet_data *hd, int timer)
 {
 	hd->hd_state |= (1 << timer);
@@ -125,7 +116,7 @@ struct hpet_info {
 	unsigned short hi_timer;
 };
 
-#define	HPET_INFO_PERIODIC	0x0001	/* timer is periodic */
+#define HPET_INFO_PERIODIC	0x0010	/* periodic-capable comparator */
 
 #define	HPET_IE_ON	_IO('h', 0x01)	/* interrupt on */
 #define	HPET_IE_OFF	_IO('h', 0x02)	/* interrupt off */
#13 lore raw
Quoted 37 lines · David Brownell

From: David Brownell <[email protected]>

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.

Quoted 8 lines

---
I CC'd everyone who MAINTAINERS says maintains HPET. Odd to have four entries!!

And re that kernel interface ... IMO, worth having a relatively generic interface for hardware timers instead of inventing a new interface for each new bit of hardware. Embedded systems tend to have at least half a dozen timers to spare.

i've asked Clemens of how to merge this and we'll do it via tip/timers/hpet - so i queued it up there with the ACK of Clemens.

I also merged it up to latest -git. (the documentation bits already moved, etc.)

I suspect this is a 2.6.27 change, given its fix nature, but it will need some cooking in linux-next (via auto-timers-next) before this can be sent to Linus.

Ingo

#14 lore raw
Quoted 3 lines · Ingo Molnar

I suspect this is a 2.6.27 change, given its fix nature, but it will need some cooking in linux-next (via auto-timers-next) before this can be sent to Linus.

-tip testing found a build failure caused by this patch:

drivers/built-in.o: In function `hpet_alloc':
: undefined reference to `__udivdi3'
drivers/built-in.o: In function `hpet_alloc':
: undefined reference to `__umoddi3'

with:

http://redhat.com/~mingo/misc/config-Thu_Jul_31_18_46_31_CEST_2008.bad

Please send a delta fix, i've already applied your previous patch and pushed it out, you can get that tree via:

http://people.redhat.com/mingo/tip.git/README

... and do something like:

git-checkout tip/timers/hpet

The last commit is yours.

Thanks,

Ingo

#15 lore raw
Quoted 4 lines · Ingo Molnar

  drivers/built-in.o: In function `hpet_alloc':
  : undefined reference to `__udivdi3'
  drivers/built-in.o: In function `hpet_alloc':
  : undefined reference to `__umoddi3'

Verified on 64 and 32 bit builds.

drivers/char/hpet.c +4 -2
@@ -701,6 +701,7 @@ int hpet_alloc(struct hpet_data *hdp)
 	static struct hpets *last = NULL;
 	unsigned long period;
 	unsigned long long temp;
+	u32 remainder;
 
 	/*
 	 * hpet_alloc can be called by platform dependent code.
@@ -764,12 +765,13 @@ int hpet_alloc(struct hpet_data *hdp)
 		printk("%s %d", i > 0 ? "," : "", hdp->hd_irq[i]);
 	printk("\n");
 
+	temp = hpetp->hp_tick_freq;
+	remainder = do_div(temp, 1000000);
 	printk(KERN_INFO
 		"hpet%u: %u comparators, %d-bit %u.%06u MHz counter\n",
 		hpetp->hp_which, hpetp->hp_ntimer,
 		cap & HPET_COUNTER_SIZE_MASK ? 64 : 32,
-		(unsigned) (hpetp->hp_tick_freq / 1000000),
-		(unsigned) (hpetp->hp_tick_freq % 1000000));
+		(unsigned) temp, remainder);
 
 	mcfg = readq(&hpet->hpet_config);
 	if ((mcfg & HPET_ENABLE_CNF_MASK) == 0) {
Applied
#16 lore raw
Quoted 7 lines · David Brownell
Quoted 4 lines · Ingo Molnar

  drivers/built-in.o: In function `hpet_alloc':
  : undefined reference to `__udivdi3'
  drivers/built-in.o: In function `hpet_alloc':
  : undefined reference to `__umoddi3'

Verified on 64 and 32 bit builds.

applied to tip/timers/hpet - thanks David!

Ingo