Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC + timer.it_value.tv_usec * UM_NSEC_PER_USEC; nsecs += os_nsecs(); Not looking at the rest of the code, the first thing that comes to my mind is that this very easy overflows if 32 bit multiplications are used (tv_sec is usually 32bit and dictates the first multiplication unless typecast is used). Stian Skjelstad -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Hurray, Houston we have ignition. We now have working userspace timers. It is still schizophrenic - userspace is HZ, kernel is NOHZ because the userpace has to keep checking did the kernel timer fire yet at a HZ interval. However, even that is a major progress compared to having userspace timer behavior determined by the phase of the moon, the position of a black goat relative to a silver knife, etc. It is now spot on - you set HZ=100 in the .config, you get 100. Before you used to get something... like 39-45 depending on the weather. The userspace is now significantly more responsive and snappy (that is expected as it now gets decent clock). Kernel behavior on timers in first instance also looks correct and NOHZ-ish (traffic shapers work). I am going to hit it with the torture suite now to see if there is significant difference with relation to other known bugs like the ext4 writeout (my original patch versions seemed to aggravate it). I will try to get around to restore my virtual desktop setup over X to see what difference does it make. Judging by the way userspace behaves after the changes it should be better than before. A. On 10/05/15 15:34, Thomas Meyer wrote: Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, which seems to be the natural thing for posix timers. I will send this next patch as something that should be includable into the kernel, i.e. With correct description and signed off line and so on. But feel free to have a look at v6 and give feedback. With kind regards Thomas diff --git a/arch/um/Makefile b/arch/um/Makefile index 17d4460..a4a434f 100644 --- a/arch/um/Makefile +++ b/arch/um/Makefile @@ -130,7 +130,7 @@ export LDS_ELF_FORMAT := $(ELF_FORMAT) # The wrappers will select whether using malloc or the kernel allocator. LINK_WRAPS = -Wl,--wrap,malloc -Wl,--wrap,free -Wl,--wrap,calloc -LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) +LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) -lrt # Used by link-vmlinux.sh which has special support for um link export CFLAGS_vmlinux := $(LINK-y) $(LINK_WRAPS) $(LD_FLAGS_CMDLINE) diff --git a/arch/um/include/asm/irq.h b/arch/um/include/asm/irq.h index 4a2037f..0f2a5b1 100644 --- a/arch/um/include/asm/irq.h +++ b/arch/um/include/asm/irq.h @@ -16,8 +16,9 @@ #define TELNETD_IRQ12 #define XTERM_IRQ 13 #define RANDOM_IRQ 14 +#define HRTIMER_IRQ15 -#define LAST_IRQ RANDOM_IRQ +#define LAST_IRQ HRTIMER_IRQ #define NR_IRQS (LAST_IRQ + 1) #endif diff --git a/arch/um/include/shared/as-layout.h b/arch/um/include/shared/as-layout.h index ca1843e..798aa6e 100644 --- a/arch/um/include/shared/as-layout.h +++ b/arch/um/include/shared/as-layout.h @@ -17,7 +17,7 @@ /* Some constant macros are used in both assembler and * C code. Therefore we cannot annotate them always with - * 'UL' and other type specifiers unilaterally. We + * 'UL' and other type specifiers unilaterally. We * use the following macros to deal with this. */ @@ -28,6 +28,13 @@ #define _UML_AC(X, Y) __UML_AC(X, Y) #endif +/** + * userspace stub address space layout: + * Below macros define the layout of the stub code and data + * which are mapped in each userspace process: + * - one page of code located at 0x10 followed by + * - one page of data + */ #define STUB_START _UML_AC(, 0x10) #define STUB_CODE _UML_AC((unsigned long), STUB_START) #define STUB_DATA _UML_AC((unsigned long), STUB_CODE + UM_KERN_PAGE_SIZE) diff --git a/arch/um/include/shared/kern_util.h b/arch/um/include/shared/kern_util.h index 83a91f9..0282b36 100644 --- a/arch/um/include/shared/kern_util.h +++ b/arch/um/include/shared/kern_util.h @@ -37,6 +37,7 @@ extern void initial_thread_cb(void (*proc)(void *), void *arg); extern int is_syscall(unsigned long addr); extern void timer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); +extern void hrtimer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); extern int start_uml(void); extern void paging_init(void); diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h index d824528..7f7368b 100644 --- a/arch/um/include/shared/os.h +++ b/arch/um/include/shared/os.h @@ -217,7 +217,8 @@ extern int set_umid(char *name); extern char *get_umid(void); /* signal.c */ -extern void timer_init(void); +extern void uml_timer_set_signal_handler(void); +extern
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Hi Thomas, hi Richard, It is now possible to reproducibly hang it. I have not been able to concoct a synthetic test (yet), but a non-synthetic one, namely installing an update to base-files on Debian is a guaranteed hang. So IO on itself does not hang it, CPU on itself does not, a mix of two does. It hangs in userspace, spinning at 100% CPU on that thread. If you whack the offending thread with -11 from the host, UML recovers, killing the affected process. I cannot look at this in detail for a few days though - the earliest I can pick it up is on Sat (in my free time). On the positive side - the behavior we are getting now is better, so we just need to figure out the root cause for the hang(s) and stabilize it. A. On 11/05/15 13:52, Anton Ivanov wrote: Hurray, Houston we have ignition. We now have working userspace timers. It is still schizophrenic - userspace is HZ, kernel is NOHZ because the userpace has to keep checking did the kernel timer fire yet at a HZ interval. However, even that is a major progress compared to having userspace timer behavior determined by the phase of the moon, the position of a black goat relative to a silver knife, etc. It is now spot on - you set HZ=100 in the .config, you get 100. Before you used to get something... like 39-45 depending on the weather. The userspace is now significantly more responsive and snappy (that is expected as it now gets decent clock). Kernel behavior on timers in first instance also looks correct and NOHZ-ish (traffic shapers work). I am going to hit it with the torture suite now to see if there is significant difference with relation to other known bugs like the ext4 writeout (my original patch versions seemed to aggravate it). I will try to get around to restore my virtual desktop setup over X to see what difference does it make. Judging by the way userspace behaves after the changes it should be better than before. A. On 10/05/15 15:34, Thomas Meyer wrote: Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, which seems to be the natural thing for posix timers. I will send this next patch as something that should be includable into the kernel, i.e. With correct description and signed off line and so on. But feel free to have a look at v6 and give feedback. With kind regards Thomas diff --git a/arch/um/Makefile b/arch/um/Makefile index 17d4460..a4a434f 100644 --- a/arch/um/Makefile +++ b/arch/um/Makefile @@ -130,7 +130,7 @@ export LDS_ELF_FORMAT := $(ELF_FORMAT) # The wrappers will select whether using malloc or the kernel allocator. LINK_WRAPS = -Wl,--wrap,malloc -Wl,--wrap,free -Wl,--wrap,calloc -LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) +LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) -lrt # Used by link-vmlinux.sh which has special support for um link export CFLAGS_vmlinux := $(LINK-y) $(LINK_WRAPS) $(LD_FLAGS_CMDLINE) diff --git a/arch/um/include/asm/irq.h b/arch/um/include/asm/irq.h index 4a2037f..0f2a5b1 100644 --- a/arch/um/include/asm/irq.h +++ b/arch/um/include/asm/irq.h @@ -16,8 +16,9 @@ #define TELNETD_IRQ12 #define XTERM_IRQ 13 #define RANDOM_IRQ 14 +#define HRTIMER_IRQ15 -#define LAST_IRQ RANDOM_IRQ +#define LAST_IRQ HRTIMER_IRQ #define NR_IRQS (LAST_IRQ + 1) #endif diff --git a/arch/um/include/shared/as-layout.h b/arch/um/include/shared/as-layout.h index ca1843e..798aa6e 100644 --- a/arch/um/include/shared/as-layout.h +++ b/arch/um/include/shared/as-layout.h @@ -17,7 +17,7 @@ /* Some constant macros are used in both assembler and * C code. Therefore we cannot annotate them always with - * 'UL' and other type specifiers unilaterally. We + * 'UL' and other type specifiers unilaterally. We * use the following macros to deal with this. */ @@ -28,6 +28,13 @@ #define _UML_AC(X, Y) __UML_AC(X, Y) #endif +/** + * userspace stub address space layout: + * Below macros define the layout of the stub code and data + * which are mapped in each userspace process: + * - one page of code located at 0x10 followed by + * - one page of data + */ #define STUB_START _UML_AC(, 0x10) #define STUB_CODE _UML_AC((unsigned long), STUB_START) #define STUB_DATA _UML_AC((unsigned long), STUB_CODE + UM_KERN_PAGE_SIZE) diff --git a/arch/um/include/shared/kern_util.h b/arch/um/include/shared/kern_util.h index 83a91f9..0282b36 100644 ---
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
The likely suspect is arch/um/os-Linux/skas/process.c It is spinning in the while(1) loop. However, the current code looks correct and the original code does not make any sense at least to me: Original code gets the _VALUE_ of the current userspace itimer() and it ensures that the signal is delivered the first time (only the first time) it skips vtalrm signals until that time has lapsed. So far so good - we are approximating real clock using a clock whose value depends on CPU. if (getitimer(ITIMER_VIRTUAL, timer)) printk(UM_KERN_ERR Failed to get itimer, errno = %d\n, errno); nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC + timer.it_value.tv_usec * UM_NSEC_PER_USEC; nsecs += os_nsecs(); However, instead of resetting the next check value to the _INTERVAL_ value which would have been the obvious thing to do in the check it resets it by incrementing it with the _VALUE_ nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC + timer.it_value.tv_usec * UM_NSEC_PER_USEC; nsecs += os_nsecs(); This is inside the while(1) loop so there is no re-adjustment of the values. So in fact, in the original code it fires at some feedback loop rate depending on CPU usage by this UML instance. Weird. In any case, in order to figure out the correct replacement here, we need to understand the logic (or the bug) in the original. Quite clearly the logical replacement where the timer fires exactly when it is expected to fire does not quite work. So what is the idea here? A. On 11/05/15 16:05, Anton Ivanov wrote: Hi Thomas, hi Richard, It is now possible to reproducibly hang it. I have not been able to concoct a synthetic test (yet), but a non-synthetic one, namely installing an update to base-files on Debian is a guaranteed hang. So IO on itself does not hang it, CPU on itself does not, a mix of two does. It hangs in userspace, spinning at 100% CPU on that thread. If you whack the offending thread with -11 from the host, UML recovers, killing the affected process. I cannot look at this in detail for a few days though - the earliest I can pick it up is on Sat (in my free time). On the positive side - the behavior we are getting now is better, so we just need to figure out the root cause for the hang(s) and stabilize it. A. On 11/05/15 13:52, Anton Ivanov wrote: Hurray, Houston we have ignition. We now have working userspace timers. It is still schizophrenic - userspace is HZ, kernel is NOHZ because the userpace has to keep checking did the kernel timer fire yet at a HZ interval. However, even that is a major progress compared to having userspace timer behavior determined by the phase of the moon, the position of a black goat relative to a silver knife, etc. It is now spot on - you set HZ=100 in the .config, you get 100. Before you used to get something... like 39-45 depending on the weather. The userspace is now significantly more responsive and snappy (that is expected as it now gets decent clock). Kernel behavior on timers in first instance also looks correct and NOHZ-ish (traffic shapers work). I am going to hit it with the torture suite now to see if there is significant difference with relation to other known bugs like the ext4 writeout (my original patch versions seemed to aggravate it). I will try to get around to restore my virtual desktop setup over X to see what difference does it make. Judging by the way userspace behaves after the changes it should be better than before. A. On 10/05/15 15:34, Thomas Meyer wrote: Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, which seems to be the natural thing for posix timers. I will send this next patch as something that should be includable into the kernel, i.e. With correct description and signed off line and so on. But feel free to have a look at v6 and give feedback. With kind regards Thomas diff --git a/arch/um/Makefile b/arch/um/Makefile index 17d4460..a4a434f 100644 --- a/arch/um/Makefile +++ b/arch/um/Makefile @@ -130,7 +130,7 @@ export LDS_ELF_FORMAT := $(ELF_FORMAT) # The wrappers will select whether using malloc or the kernel allocator. LINK_WRAPS = -Wl,--wrap,malloc -Wl,--wrap,free -Wl,--wrap,calloc -LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) +LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) -lrt # Used by link-vmlinux.sh which has
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
On 11/05/15 18:00, Thomas Meyer wrote: Hi, maybe there is a bug in how the timers are created for all user space processes. In the latest patch I use os__timer_remain for the initial interval. The idea was to launch all timers at the same time. But I now think this can never work using relative times, especially when os__timer_remain returns 0, then the new timer is never launched. That is not a bad idea, just requires a more complex check in setting up interval so it is reset to default when a zero value or value HZ is passed. That may explain the hangs you see. You could try to replace os__timer_remain with the current HZ value in nanoseconds. I am going to adjust time.c instead as above. A. Kind regards Thomad A [snip] A. -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Hi, maybe there is a bug in how the timers are created for all user space processes. In the latest patch I use os__timer_remain for the initial interval. The idea was to launch all timers at the same time. But I now think this can never work using relative times, especially when os__timer_remain returns 0, then the new timer is never launched. That may explain the hangs you see. You could try to replace os__timer_remain with the current HZ value in nanoseconds. Kind regards Thomad Am 11.05.2015 5:43 nachm. schrieb Anton Ivanov anton.iva...@kot-begemot.co.uk: The likely suspect is arch/um/os-Linux/skas/process.c It is spinning in the while(1) loop. However, the current code looks correct and the original code does not make any sense at least to me: Original code gets the _VALUE_ of the current userspace itimer() and it ensures that the signal is delivered the first time (only the first time) it skips vtalrm signals until that time has lapsed. So far so good - we are approximating real clock using a clock whose value depends on CPU. if (getitimer(ITIMER_VIRTUAL, timer)) printk(UM_KERN_ERR Failed to get itimer, errno = %d\n, errno); nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC + timer.it_value.tv_usec * UM_NSEC_PER_USEC; nsecs += os_nsecs(); However, instead of resetting the next check value to the _INTERVAL_ value which would have been the obvious thing to do in the check it resets it by incrementing it with the _VALUE_ nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC + timer.it_value.tv_usec * UM_NSEC_PER_USEC; nsecs += os_nsecs(); This is inside the while(1) loop so there is no re-adjustment of the values. So in fact, in the original code it fires at some feedback loop rate depending on CPU usage by this UML instance. Weird. In any case, in order to figure out the correct replacement here, we need to understand the logic (or the bug) in the original. Quite clearly the logical replacement where the timer fires exactly when it is expected to fire does not quite work. So what is the idea here? A. On 11/05/15 16:05, Anton Ivanov wrote: Hi Thomas, hi Richard, It is now possible to reproducibly hang it. I have not been able to concoct a synthetic test (yet), but a non-synthetic one, namely installing an update to base-files on Debian is a guaranteed hang. So IO on itself does not hang it, CPU on itself does not, a mix of two does. It hangs in userspace, spinning at 100% CPU on that thread. If you whack the offending thread with -11 from the host, UML recovers, killing the affected process. I cannot look at this in detail for a few days though - the earliest I can pick it up is on Sat (in my free time). On the positive side - the behavior we are getting now is better, so we just need to figure out the root cause for the hang(s) and stabilize it. A. On 11/05/15 13:52, Anton Ivanov wrote: Hurray, Houston we have ignition. We now have working userspace timers. It is still schizophrenic - userspace is HZ, kernel is NOHZ because the userpace has to keep checking did the kernel timer fire yet at a HZ interval. However, even that is a major progress compared to having userspace timer behavior determined by the phase of the moon, the position of a black goat relative to a silver knife, etc. It is now spot on - you set HZ=100 in the .config, you get 100. Before you used to get something... like 39-45 depending on the weather. The userspace is now significantly more responsive and snappy (that is expected as it now gets decent clock). Kernel behavior on timers in first instance also looks correct and NOHZ-ish (traffic shapers work). I am going to hit it with the torture suite now to see if there is significant difference with relation to other known bugs like the ext4 writeout (my original patch versions seemed to aggravate it). I will try to get around to restore my virtual desktop setup over X to see what difference does it make. Judging by the way userspace behaves after the changes it should be better than before. A. On 10/05/15 15:34, Thomas Meyer wrote: Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, which seems to be the natural
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
On 11/05/15 18:41, Thomas Meyer wrote: Am 11.05.2015 7:31 nachm. schrieb Anton Ivanov anton.iva...@kot-begemot.co.uk: On 11/05/15 18:20, Anton Ivanov wrote: On 11/05/15 18:00, Thomas Meyer wrote: Hi, maybe there is a bug in how the timers are created for all user space processes. In the latest patch I use os__timer_remain for the initial interval. The idea was to launch all timers at the same time. But I now think this can never work using relative times, especially when os__timer_remain returns 0, then the new timer is never launched. That is not a bad idea, just requires a more complex check in setting up interval so it is reset to default when a zero value or value HZ is passed. That may explain the hangs you see. You could try to replace os__timer_remain with the current HZ value in nanoseconds. I am going to adjust time.c instead as above. Both approaches fail - the check (that is expected actually, I did not think properly here - it does not cover the stub in kernel/skas) and the setting of data to the correct initial values. Also, if it was just not starting timers it would have been possible to beat the process with USR2 on the head until its morale improves. Well, while true; do kill -USR2 PID ; done does not change the behavior, it still hangs and is still possible to terminate the errant process by sending SIGSEGV to the UML thread which is shown as 100% CPU and in R state. Yes, sending USR2 should do the same. Correct. Problem is somewhere else. I am surprised it worked correctly in the original one as there for 100% flat CPU usage it should have converged to a behavior which is similar to what we have now. Mhh. Strange. Any hints how I can reproduce this? I need to see the same behaviour, i.e. the hang on my machine to begin to understand what is going on here. Start off with a Debian image of let's say wheezy and hit it with a dist-upgrade to the next release. This produces enough IO+CPU at the same time to trigger it. I get it 100% reproducible every time it tries to unpack the first couple of packages. I tried to trigger backtraces but could not get anything informative. The old trick of hitting the kernel with SIGILL which worked very nicely in older kernel versions no longer works :( Thanks for testing Anton. I wish I had more time at the moment to debug it and fix it :( We are nearly there and it is worth it. A. A. A. Kind regards Thomad A [snip] A. -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Am 11.05.2015 9:42 nachm. schrieb Anton Ivanov anton.iva...@kot-begemot.co.uk: On 11/05/15 18:41, Thomas Meyer wrote: Am 11.05.2015 7:31 nachm. schrieb Anton Ivanov anton.iva...@kot-begemot.co.uk: On 11/05/15 18:20, Anton Ivanov wrote: On 11/05/15 18:00, Thomas Meyer wrote: Hi, maybe there is a bug in how the timers are created for all user space processes. In the latest patch I use os__timer_remain for the initial interval. The idea was to launch all timers at the same time. But I now think this can never work using relative times, especially when os__timer_remain returns 0, then the new timer is never launched. That is not a bad idea, just requires a more complex check in setting up interval so it is reset to default when a zero value or value HZ is passed. That may explain the hangs you see. You could try to replace os__timer_remain with the current HZ value in nanoseconds. I am going to adjust time.c instead as above. Both approaches fail - the check (that is expected actually, I did not think properly here - it does not cover the stub in kernel/skas) and the setting of data to the correct initial values. Also, if it was just not starting timers it would have been possible to beat the process with USR2 on the head until its morale improves. Well, while true; do kill -USR2 PID ; done does not change the behavior, it still hangs and is still possible to terminate the errant process by sending SIGSEGV to the UML thread which is shown as 100% CPU and in R state. Yes, sending USR2 should do the same. Correct. Problem is somewhere else. I am surprised it worked correctly in the original one as there for 100% flat CPU usage it should have converged to a behavior which is similar to what we have now. Mhh. Strange. Any hints how I can reproduce this? I need to see the same behaviour, i.e. the hang on my machine to begin to understand what is going on here. Start off with a Debian image of let's say wheezy and hit it with a dist-upgrade to the next release. This produces enough IO+CPU at the same time to trigger it. I get it 100% reproducible every time it tries to unpack the first couple of packages. Okay, I'll try that! I tried to trigger backtraces but could not get anything informative. The old trick of hitting the kernel with SIGILL which worked very nicely in older kernel versions no longer works :( You may can trigger an backtrace via uml_console. Thanks for testing Anton. I wish I had more time at the moment to debug it and fix it :( We are nearly there and it is worth it. Yes, same here. But we will finish this! A. A. A. Kind regards Thomad A [snip] A. -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Am 10.05.2015 um 16:34 schrieb Thomas Meyer: Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, which seems to be the natural thing for posix timers. I will send this next patch as something that should be includable into the kernel, i.e. With correct description and signed off line and so on. But feel free to have a look at v6 and give feedback. The timer stuff looks generally okay but please submit again as a clean patch series with one logical change per patch such that I can review it in depth. :-) Thanks, //richard -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
On 11/05/15 18:20, Anton Ivanov wrote: On 11/05/15 18:00, Thomas Meyer wrote: Hi, maybe there is a bug in how the timers are created for all user space processes. In the latest patch I use os__timer_remain for the initial interval. The idea was to launch all timers at the same time. But I now think this can never work using relative times, especially when os__timer_remain returns 0, then the new timer is never launched. That is not a bad idea, just requires a more complex check in setting up interval so it is reset to default when a zero value or value HZ is passed. That may explain the hangs you see. You could try to replace os__timer_remain with the current HZ value in nanoseconds. I am going to adjust time.c instead as above. Both approaches fail - the check (that is expected actually, I did not think properly here - it does not cover the stub in kernel/skas) and the setting of data to the correct initial values. Also, if it was just not starting timers it would have been possible to beat the process with USR2 on the head until its morale improves. Well, while true; do kill -USR2 PID ; done does not change the behavior, it still hangs and is still possible to terminate the errant process by sending SIGSEGV to the UML thread which is shown as 100% CPU and in R state. Problem is somewhere else. I am surprised it worked correctly in the original one as there for 100% flat CPU usage it should have converged to a behavior which is similar to what we have now. A. A. Kind regards Thomad A [snip] A. -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Am 11.05.2015 7:31 nachm. schrieb Anton Ivanov anton.iva...@kot-begemot.co.uk: On 11/05/15 18:20, Anton Ivanov wrote: On 11/05/15 18:00, Thomas Meyer wrote: Hi, maybe there is a bug in how the timers are created for all user space processes. In the latest patch I use os__timer_remain for the initial interval. The idea was to launch all timers at the same time. But I now think this can never work using relative times, especially when os__timer_remain returns 0, then the new timer is never launched. That is not a bad idea, just requires a more complex check in setting up interval so it is reset to default when a zero value or value HZ is passed. That may explain the hangs you see. You could try to replace os__timer_remain with the current HZ value in nanoseconds. I am going to adjust time.c instead as above. Both approaches fail - the check (that is expected actually, I did not think properly here - it does not cover the stub in kernel/skas) and the setting of data to the correct initial values. Also, if it was just not starting timers it would have been possible to beat the process with USR2 on the head until its morale improves. Well, while true; do kill -USR2 PID ; done does not change the behavior, it still hangs and is still possible to terminate the errant process by sending SIGSEGV to the UML thread which is shown as 100% CPU and in R state. Yes, sending USR2 should do the same. Correct. Problem is somewhere else. I am surprised it worked correctly in the original one as there for 100% flat CPU usage it should have converged to a behavior which is similar to what we have now. Mhh. Strange. Any hints how I can reproduce this? I need to see the same behaviour, i.e. the hang on my machine to begin to understand what is going on here. Thanks for testing Anton. A. A. Kind regards Thomad A [snip] A. -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Am 10.05.2015 um 16:34 schrieb Thomas Meyer: Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. Ahh, ok! I was a bit confused as it contains beside the new features also new comments to unrelated code, changes coding style, etc... I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, which seems to be the natural thing for posix timers. I will send this next patch as something that should be includable into the kernel, i.e. With correct description and signed off line and so on. But feel free to have a look at v6 and give feedback. I will. :-) Thanks, //richard -- One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y ___ User-mode-linux-devel mailing list User-mode-linux-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote:
Hi,
Changes:
- also create posix timer in stub_clone_handler()
- incorporated antons remarks
Hm, this patch does a *lot* more than the changelog says.
diff --git a/arch/um/Makefile b/arch/um/Makefile
index 17d4460..a4a434f 100644
--- a/arch/um/Makefile
+++ b/arch/um/Makefile
@@ -130,7 +130,7 @@ export LDS_ELF_FORMAT := $(ELF_FORMAT)
# The wrappers will select whether using malloc or the kernel allocator.
LINK_WRAPS = -Wl,--wrap,malloc -Wl,--wrap,free -Wl,--wrap,calloc
-LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt))
+LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) -lrt
# Used by link-vmlinux.sh which has special support for um link
export CFLAGS_vmlinux := $(LINK-y) $(LINK_WRAPS) $(LD_FLAGS_CMDLINE)
diff --git a/arch/um/include/asm/irq.h b/arch/um/include/asm/irq.h
index 4a2037f..0f2a5b1 100644
--- a/arch/um/include/asm/irq.h
+++ b/arch/um/include/asm/irq.h
@@ -16,8 +16,9 @@
#define TELNETD_IRQ12
#define XTERM_IRQ 13
#define RANDOM_IRQ 14
+#define HRTIMER_IRQ15
-#define LAST_IRQ RANDOM_IRQ
+#define LAST_IRQ HRTIMER_IRQ
#define NR_IRQS (LAST_IRQ + 1)
#endif
diff --git a/arch/um/include/shared/as-layout.h
b/arch/um/include/shared/as-layout.h
index ca1843e..798aa6e 100644
--- a/arch/um/include/shared/as-layout.h
+++ b/arch/um/include/shared/as-layout.h
@@ -17,7 +17,7 @@
/* Some constant macros are used in both assembler and
* C code. Therefore we cannot annotate them always with
- * 'UL' and other type specifiers unilaterally. We
+ * 'UL' and other type specifiers unilaterally. We
* use the following macros to deal with this.
*/
@@ -28,6 +28,13 @@
#define _UML_AC(X, Y) __UML_AC(X, Y)
#endif
+/**
+ * userspace stub address space layout:
+ * Below macros define the layout of the stub code and data
+ * which are mapped in each userspace process:
+ * - one page of code located at 0x10 followed by
+ * - one page of data
+ */
#define STUB_START _UML_AC(, 0x10)
#define STUB_CODE _UML_AC((unsigned long), STUB_START)
#define STUB_DATA _UML_AC((unsigned long), STUB_CODE + UM_KERN_PAGE_SIZE)
diff --git a/arch/um/include/shared/kern_util.h
b/arch/um/include/shared/kern_util.h
index 83a91f9..0282b36 100644
--- a/arch/um/include/shared/kern_util.h
+++ b/arch/um/include/shared/kern_util.h
@@ -37,6 +37,7 @@ extern void initial_thread_cb(void (*proc)(void *), void
*arg);
extern int is_syscall(unsigned long addr);
extern void timer_handler(int sig, struct siginfo *unused_si, struct
uml_pt_regs *regs);
+extern void hrtimer_handler(int sig, struct siginfo *unused_si, struct
uml_pt_regs *regs);
extern int start_uml(void);
extern void paging_init(void);
diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h
index d824528..7f7368b 100644
--- a/arch/um/include/shared/os.h
+++ b/arch/um/include/shared/os.h
@@ -217,7 +217,8 @@ extern int set_umid(char *name);
extern char *get_umid(void);
/* signal.c */
-extern void timer_init(void);
+extern void uml_timer_set_signal_handler(void);
+extern void uml_hrtimer_set_signal_handler(void);
extern void set_sigstack(void *sig_stack, int size);
extern void remove_sigstack(void);
extern void set_handler(int sig);
@@ -238,12 +239,16 @@ extern void um_early_printk(const char *s, unsigned int
n);
extern void os_fix_helper_signals(void);
/* time.c */
-extern void idle_sleep(unsigned long long nsecs);
-extern int set_interval(void);
-extern int timer_one_shot(int ticks);
-extern long long disable_timer(void);
+extern void os_idle_sleep(unsigned long long nsecs);
+extern int os_timer_create(void* timer);
+extern int os_timer_set_interval(void* timer, void* its);
+extern int os_timer_one_shot(int ticks);
+extern long long os_timer_disable(void);
+extern long os_timer_remain(void* timer);
extern void uml_idle_timer(void);
+extern long long os_persistent_clock_emulation(void);
extern long long os_nsecs(void);
+extern long long os_vnsecs(void);
/* skas/mem.c */
extern long run_syscall_stub(struct mm_id * mm_idp,
diff --git a/arch/um/include/shared/skas/stub-data.h
b/arch/um/include/shared/skas/stub-data.h
index f6ed92c..f98b9e2 100644
--- a/arch/um/include/shared/skas/stub-data.h
+++ b/arch/um/include/shared/skas/stub-data.h
@@ -6,12 +6,12 @@
#ifndef __STUB_DATA_H
#define __STUB_DATA_H
-#include sys/time.h
+#include time.h
struct stub_data {
- long offset;
+ unsigned long offset;
int fd;
- struct itimerval timer;
+ struct itimerspec timer;
long err;
};
diff --git a/arch/um/include/shared/timer-internal.h
b/arch/um/include/shared/timer-internal.h
new file mode 100644
index 000..afdc6dc
--- /dev/null
+++ b/arch/um/include/shared/timer-internal.h
@@ -0,0 +1,18 @@
+/*
+ * Copyright (C) 2012 - 2014 Cisco Systems
+ *
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
On 10/05/15 15:34, Thomas Meyer wrote: Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. Looks good, does it test out good - I will tell you during the week. I already have most of the testcases set up from the period when I did only the kernel part. The best test is running any of the kernel queueing disciplines. Stock UML is completely unable to do any of them correctly. I can probably concoct something related to timers and timeouts in protocols too to demo the breakage with the current timer state in these. I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, If you mean SIGALRM that had some issues. I tried that originally and gave up - that is why the patch uses SIGUSR2 - it is something nobody uses and it did not interfere with the existing use of ALRM and VTALRM. A. which seems to be the natural thing for posix timers. I will send this next patch as something that should be includable into the kernel, i.e. With correct description and signed off line and so on. But feel free to have a look at v6 and give feedback. With kind regards Thomas diff --git a/arch/um/Makefile b/arch/um/Makefile index 17d4460..a4a434f 100644 --- a/arch/um/Makefile +++ b/arch/um/Makefile @@ -130,7 +130,7 @@ export LDS_ELF_FORMAT := $(ELF_FORMAT) # The wrappers will select whether using malloc or the kernel allocator. LINK_WRAPS = -Wl,--wrap,malloc -Wl,--wrap,free -Wl,--wrap,calloc -LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) +LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) -lrt # Used by link-vmlinux.sh which has special support for um link export CFLAGS_vmlinux := $(LINK-y) $(LINK_WRAPS) $(LD_FLAGS_CMDLINE) diff --git a/arch/um/include/asm/irq.h b/arch/um/include/asm/irq.h index 4a2037f..0f2a5b1 100644 --- a/arch/um/include/asm/irq.h +++ b/arch/um/include/asm/irq.h @@ -16,8 +16,9 @@ #define TELNETD_IRQ12 #define XTERM_IRQ 13 #define RANDOM_IRQ 14 +#define HRTIMER_IRQ15 -#define LAST_IRQ RANDOM_IRQ +#define LAST_IRQ HRTIMER_IRQ #define NR_IRQS (LAST_IRQ + 1) #endif diff --git a/arch/um/include/shared/as-layout.h b/arch/um/include/shared/as-layout.h index ca1843e..798aa6e 100644 --- a/arch/um/include/shared/as-layout.h +++ b/arch/um/include/shared/as-layout.h @@ -17,7 +17,7 @@ /* Some constant macros are used in both assembler and * C code. Therefore we cannot annotate them always with - * 'UL' and other type specifiers unilaterally. We + * 'UL' and other type specifiers unilaterally. We * use the following macros to deal with this. */ @@ -28,6 +28,13 @@ #define _UML_AC(X, Y) __UML_AC(X, Y) #endif +/** + * userspace stub address space layout: + * Below macros define the layout of the stub code and data + * which are mapped in each userspace process: + * - one page of code located at 0x10 followed by + * - one page of data + */ #define STUB_START _UML_AC(, 0x10) #define STUB_CODE _UML_AC((unsigned long), STUB_START) #define STUB_DATA _UML_AC((unsigned long), STUB_CODE + UM_KERN_PAGE_SIZE) diff --git a/arch/um/include/shared/kern_util.h b/arch/um/include/shared/kern_util.h index 83a91f9..0282b36 100644 --- a/arch/um/include/shared/kern_util.h +++ b/arch/um/include/shared/kern_util.h @@ -37,6 +37,7 @@ extern void initial_thread_cb(void (*proc)(void *), void *arg); extern int is_syscall(unsigned long addr); extern void timer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); +extern void hrtimer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); extern int start_uml(void); extern void paging_init(void); diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h index d824528..7f7368b 100644 --- a/arch/um/include/shared/os.h +++ b/arch/um/include/shared/os.h @@ -217,7 +217,8 @@ extern int set_umid(char *name); extern char *get_umid(void); /* signal.c */ -extern void timer_init(void); +extern void uml_timer_set_signal_handler(void); +extern void uml_hrtimer_set_signal_handler(void); extern void set_sigstack(void *sig_stack, int size); extern void remove_sigstack(void); extern void set_handler(int sig); @@ -238,12 +239,16 @@ extern void um_early_printk(const char *s, unsigned int n); extern void os_fix_helper_signals(void); /* time.c */ -extern void idle_sleep(unsigned long long nsecs); -extern int set_interval(void); -extern int timer_one_shot(int ticks); -extern long long disable_timer(void); +extern void
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
On 10/05/15 13:35, Richard Weinberger wrote: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Richard, I think this is what it takes to fix the timer subsystem and the patch does all of it: 1. Replace kernel itimer source with posix timers: 1.1 os portion (new timer routines and a sleep routine). 1.2. kernel portion - new IRQ routines for the posix timer 1.3. A sprinkle of changes to support the fact that we now have one more IRQ. 2. Changes to userspace to support new timer source 3. Changes to stub invocation (added by Thomas in most recent revision). 4. Changes to idle. I would love it to be smaller so we could change just the timer source, but I do not think this is possible. There are lots of moving parts which setup timer intervals in UML directly and they all need to be fixed for a proper timer replacement. The result is worth it though - lots of things that did not quite work in UML (f.e. network qos, etc) start working. I am going to give it some more testing this week, though based on reading it, it should just work (I have ran the bits from 1.1-1.3 for 4 years now so they have quite a bit of mileage). A. diff --git a/arch/um/Makefile b/arch/um/Makefile index 17d4460..a4a434f 100644 --- a/arch/um/Makefile +++ b/arch/um/Makefile @@ -130,7 +130,7 @@ export LDS_ELF_FORMAT := $(ELF_FORMAT) # The wrappers will select whether using malloc or the kernel allocator. LINK_WRAPS = -Wl,--wrap,malloc -Wl,--wrap,free -Wl,--wrap,calloc -LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) +LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) -lrt # Used by link-vmlinux.sh which has special support for um link export CFLAGS_vmlinux := $(LINK-y) $(LINK_WRAPS) $(LD_FLAGS_CMDLINE) diff --git a/arch/um/include/asm/irq.h b/arch/um/include/asm/irq.h index 4a2037f..0f2a5b1 100644 --- a/arch/um/include/asm/irq.h +++ b/arch/um/include/asm/irq.h @@ -16,8 +16,9 @@ #define TELNETD_IRQ12 #define XTERM_IRQ 13 #define RANDOM_IRQ 14 +#define HRTIMER_IRQ15 -#define LAST_IRQ RANDOM_IRQ +#define LAST_IRQ HRTIMER_IRQ #define NR_IRQS (LAST_IRQ + 1) #endif diff --git a/arch/um/include/shared/as-layout.h b/arch/um/include/shared/as-layout.h index ca1843e..798aa6e 100644 --- a/arch/um/include/shared/as-layout.h +++ b/arch/um/include/shared/as-layout.h @@ -17,7 +17,7 @@ /* Some constant macros are used in both assembler and * C code. Therefore we cannot annotate them always with - * 'UL' and other type specifiers unilaterally. We + * 'UL' and other type specifiers unilaterally. We * use the following macros to deal with this. */ @@ -28,6 +28,13 @@ #define _UML_AC(X, Y) __UML_AC(X, Y) #endif +/** + * userspace stub address space layout: + * Below macros define the layout of the stub code and data + * which are mapped in each userspace process: + * - one page of code located at 0x10 followed by + * - one page of data + */ #define STUB_START _UML_AC(, 0x10) #define STUB_CODE _UML_AC((unsigned long), STUB_START) #define STUB_DATA _UML_AC((unsigned long), STUB_CODE + UM_KERN_PAGE_SIZE) diff --git a/arch/um/include/shared/kern_util.h b/arch/um/include/shared/kern_util.h index 83a91f9..0282b36 100644 --- a/arch/um/include/shared/kern_util.h +++ b/arch/um/include/shared/kern_util.h @@ -37,6 +37,7 @@ extern void initial_thread_cb(void (*proc)(void *), void *arg); extern int is_syscall(unsigned long addr); extern void timer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); +extern void hrtimer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); extern int start_uml(void); extern void paging_init(void); diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h index d824528..7f7368b 100644 --- a/arch/um/include/shared/os.h +++ b/arch/um/include/shared/os.h @@ -217,7 +217,8 @@ extern int set_umid(char *name); extern char *get_umid(void); /* signal.c */ -extern void timer_init(void); +extern void uml_timer_set_signal_handler(void); +extern void uml_hrtimer_set_signal_handler(void); extern void set_sigstack(void *sig_stack, int size); extern void remove_sigstack(void); extern void set_handler(int sig); @@ -238,12 +239,16 @@ extern void um_early_printk(const char *s, unsigned int n); extern void os_fix_helper_signals(void); /* time.c */ -extern void idle_sleep(unsigned long long nsecs); -extern int set_interval(void); -extern int timer_one_shot(int ticks); -extern long long disable_timer(void); +extern void os_idle_sleep(unsigned long long nsecs); +extern int os_timer_create(void* timer); +extern int os_timer_set_interval(void* timer, void* its); +extern int
Re: [uml-devel] [PATCH v6] um: Add a high resolution timer subsystem
Am 10.05.2015 um 14:35 schrieb Richard Weinberger richard.weinber...@gmail.com: On Sun, May 10, 2015 at 1:14 AM, Thomas Meyer tho...@m3y3r.de wrote: Hi, Changes: - also create posix timer in stub_clone_handler() - incorporated antons remarks Hm, this patch does a *lot* more than the changelog says. Hi, yes PATCH was probably the wrong keyword in the subject line. It should have been RFC. I just wanted to have feedback of the current state of this patch/work. I'm currently working on cleaning up the patch and switch from SIGUSR2 to SIGNALRM, which seems to be the natural thing for posix timers. I will send this next patch as something that should be includable into the kernel, i.e. With correct description and signed off line and so on. But feel free to have a look at v6 and give feedback. With kind regards Thomas diff --git a/arch/um/Makefile b/arch/um/Makefile index 17d4460..a4a434f 100644 --- a/arch/um/Makefile +++ b/arch/um/Makefile @@ -130,7 +130,7 @@ export LDS_ELF_FORMAT := $(ELF_FORMAT) # The wrappers will select whether using malloc or the kernel allocator. LINK_WRAPS = -Wl,--wrap,malloc -Wl,--wrap,free -Wl,--wrap,calloc -LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) +LD_FLAGS_CMDLINE = $(foreach opt,$(LDFLAGS),-Wl,$(opt)) -lrt # Used by link-vmlinux.sh which has special support for um link export CFLAGS_vmlinux := $(LINK-y) $(LINK_WRAPS) $(LD_FLAGS_CMDLINE) diff --git a/arch/um/include/asm/irq.h b/arch/um/include/asm/irq.h index 4a2037f..0f2a5b1 100644 --- a/arch/um/include/asm/irq.h +++ b/arch/um/include/asm/irq.h @@ -16,8 +16,9 @@ #define TELNETD_IRQ12 #define XTERM_IRQ 13 #define RANDOM_IRQ 14 +#define HRTIMER_IRQ15 -#define LAST_IRQ RANDOM_IRQ +#define LAST_IRQ HRTIMER_IRQ #define NR_IRQS (LAST_IRQ + 1) #endif diff --git a/arch/um/include/shared/as-layout.h b/arch/um/include/shared/as-layout.h index ca1843e..798aa6e 100644 --- a/arch/um/include/shared/as-layout.h +++ b/arch/um/include/shared/as-layout.h @@ -17,7 +17,7 @@ /* Some constant macros are used in both assembler and * C code. Therefore we cannot annotate them always with - * 'UL' and other type specifiers unilaterally. We + * 'UL' and other type specifiers unilaterally. We * use the following macros to deal with this. */ @@ -28,6 +28,13 @@ #define _UML_AC(X, Y) __UML_AC(X, Y) #endif +/** + * userspace stub address space layout: + * Below macros define the layout of the stub code and data + * which are mapped in each userspace process: + * - one page of code located at 0x10 followed by + * - one page of data + */ #define STUB_START _UML_AC(, 0x10) #define STUB_CODE _UML_AC((unsigned long), STUB_START) #define STUB_DATA _UML_AC((unsigned long), STUB_CODE + UM_KERN_PAGE_SIZE) diff --git a/arch/um/include/shared/kern_util.h b/arch/um/include/shared/kern_util.h index 83a91f9..0282b36 100644 --- a/arch/um/include/shared/kern_util.h +++ b/arch/um/include/shared/kern_util.h @@ -37,6 +37,7 @@ extern void initial_thread_cb(void (*proc)(void *), void *arg); extern int is_syscall(unsigned long addr); extern void timer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); +extern void hrtimer_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs); extern int start_uml(void); extern void paging_init(void); diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h index d824528..7f7368b 100644 --- a/arch/um/include/shared/os.h +++ b/arch/um/include/shared/os.h @@ -217,7 +217,8 @@ extern int set_umid(char *name); extern char *get_umid(void); /* signal.c */ -extern void timer_init(void); +extern void uml_timer_set_signal_handler(void); +extern void uml_hrtimer_set_signal_handler(void); extern void set_sigstack(void *sig_stack, int size); extern void remove_sigstack(void); extern void set_handler(int sig); @@ -238,12 +239,16 @@ extern void um_early_printk(const char *s, unsigned int n); extern void os_fix_helper_signals(void); /* time.c */ -extern void idle_sleep(unsigned long long nsecs); -extern int set_interval(void); -extern int timer_one_shot(int ticks); -extern long long disable_timer(void); +extern void os_idle_sleep(unsigned long long nsecs); +extern int os_timer_create(void* timer); +extern int os_timer_set_interval(void* timer, void* its); +extern int os_timer_one_shot(int ticks); +extern long long os_timer_disable(void); +extern long os_timer_remain(void* timer); extern void uml_idle_timer(void); +extern long long os_persistent_clock_emulation(void); extern long long os_nsecs(void); +extern long long os_vnsecs(void); /* skas/mem.c */ extern long run_syscall_stub(struct mm_id * mm_idp, diff --git a/arch/um/include/shared/skas/stub-data.h b/arch/um/include/shared/skas/stub-data.h index f6ed92c..f98b9e2 100644 --- a/arch/um/include/shared/skas/stub-data.h +++
