Re: [PATCH v4] hw/rtc/mc146818rtc: Make this rtc device target independent
On 02/01/2023 17.47, Bernhard Beschow wrote: Am 2. Januar 2023 16:09:08 UTC schrieb Thomas Huth : On 02/01/2023 14.36, Thomas Huth wrote: On 31/12/2022 00.45, Bernhard Beschow wrote: Am 29. Dezember 2022 10:58:48 UTC schrieb Thomas Huth : [...] static uint32_t rtc_periodic_clock_ticks(RTCState *s) { @@ -922,14 +911,15 @@ static void rtc_realizefn(DeviceState *dev, Error **errp) rtc_set_date_from_host(isadev); switch (s->lost_tick_policy) { -#ifdef TARGET_I386 - case LOST_TICK_POLICY_SLEW: - s->coalesced_timer = - timer_new_ns(rtc_clock, rtc_coalesced_timer, s); - break; -#endif case LOST_TICK_POLICY_DISCARD: break; + case LOST_TICK_POLICY_SLEW: + /* Slew tick policy is only available on x86 */ + if (arch_type == QEMU_ARCH_I386) { This reflects the intention much better than before, which is nice. How does `arch_type` play together with qemu-system-all? IIUC it should be possible to load all arch backends simultaneously while `arch_type` is an external symbol defined by each arch backend differently. So this seems to conflict. I assume that there still will be a main arch_type for the current selected machine? ... not sure how this will exactly work, though ... Can we just add a property such as "slew-tick-policy-available" instead? It should default to false and all x86 machines would need to opt in explicitly. Sounds like a good idea, it's certainly better than checking arch_type here ... I'll give it a try, thanks! I've now had a look at this, and it's also getting ugly: Since the property has to be set before realize() is done, the setting of the property has to be added to the mc146818_rtc_init() function. Thus this function would need a new parameter - and it then needs to be changed all over the place, i.e. also for all the non-x86 machines, defeating the idea of a default value... Maybe it makes more sense to check for a TYPE_X86_MACHINE machine type instead? Maybe you could base your patch on https://lists.gnu.org/archive/html/qemu-devel/2022-12/msg03795.html ? That would help, indeed. ... OK, then let's postpone my clean-up until your series has landed. Thomas
Re: [PATCH v4] hw/rtc/mc146818rtc: Make this rtc device target independent
Am 2. Januar 2023 16:09:08 UTC schrieb Thomas Huth : >On 02/01/2023 14.36, Thomas Huth wrote: >> On 31/12/2022 00.45, Bernhard Beschow wrote: >>> >>> Am 29. Dezember 2022 10:58:48 UTC schrieb Thomas Huth : >[...] static uint32_t rtc_periodic_clock_ticks(RTCState *s) { @@ -922,14 +911,15 @@ static void rtc_realizefn(DeviceState *dev, Error **errp) rtc_set_date_from_host(isadev); switch (s->lost_tick_policy) { -#ifdef TARGET_I386 - case LOST_TICK_POLICY_SLEW: - s->coalesced_timer = - timer_new_ns(rtc_clock, rtc_coalesced_timer, s); - break; -#endif case LOST_TICK_POLICY_DISCARD: break; + case LOST_TICK_POLICY_SLEW: + /* Slew tick policy is only available on x86 */ + if (arch_type == QEMU_ARCH_I386) { >>> >>> This reflects the intention much better than before, which is nice. >>> >>> How does `arch_type` play together with qemu-system-all? IIUC it should be >>> possible to load all arch backends simultaneously while `arch_type` is an >>> external symbol defined by each arch backend differently. So this seems to >>> conflict. >> >> I assume that there still will be a main arch_type for the current selected >> machine? ... not sure how this will exactly work, though ... >> >>> Can we just add a property such as "slew-tick-policy-available" instead? It >>> should default to false and all x86 machines would need to opt in >>> explicitly. >> >> Sounds like a good idea, it's certainly better than checking arch_type here >> ... I'll give it a try, thanks! > >I've now had a look at this, and it's also getting ugly: Since the property >has to be set before realize() is done, the setting of the property has to be >added to the mc146818_rtc_init() function. Thus this function would need a new >parameter - and it then needs to be changed all over the place, i.e. also for >all the non-x86 machines, defeating the idea of a default value... > >Maybe it makes more sense to check for a TYPE_X86_MACHINE machine type instead? Maybe you could base your patch on https://lists.gnu.org/archive/html/qemu-devel/2022-12/msg03795.html ? This patch looks like it should get you covered and is part of my PIIX consolidation series [1] which seems to be ready to be queued into Phil's mips-next tree. My series is currently just blocked by a MIPS-related regression. Best regards, Bernhard [1] https://lists.gnu.org/archive/html/qemu-devel/2022-12/msg03788.html > > Thomas >
Re: [PATCH v4] hw/rtc/mc146818rtc: Make this rtc device target independent
On 02/01/2023 14.36, Thomas Huth wrote: On 31/12/2022 00.45, Bernhard Beschow wrote: Am 29. Dezember 2022 10:58:48 UTC schrieb Thomas Huth : [...] static uint32_t rtc_periodic_clock_ticks(RTCState *s) { @@ -922,14 +911,15 @@ static void rtc_realizefn(DeviceState *dev, Error **errp) rtc_set_date_from_host(isadev); switch (s->lost_tick_policy) { -#ifdef TARGET_I386 - case LOST_TICK_POLICY_SLEW: - s->coalesced_timer = - timer_new_ns(rtc_clock, rtc_coalesced_timer, s); - break; -#endif case LOST_TICK_POLICY_DISCARD: break; + case LOST_TICK_POLICY_SLEW: + /* Slew tick policy is only available on x86 */ + if (arch_type == QEMU_ARCH_I386) { This reflects the intention much better than before, which is nice. How does `arch_type` play together with qemu-system-all? IIUC it should be possible to load all arch backends simultaneously while `arch_type` is an external symbol defined by each arch backend differently. So this seems to conflict. I assume that there still will be a main arch_type for the current selected machine? ... not sure how this will exactly work, though ... Can we just add a property such as "slew-tick-policy-available" instead? It should default to false and all x86 machines would need to opt in explicitly. Sounds like a good idea, it's certainly better than checking arch_type here ... I'll give it a try, thanks! I've now had a look at this, and it's also getting ugly: Since the property has to be set before realize() is done, the setting of the property has to be added to the mc146818_rtc_init() function. Thus this function would need a new parameter - and it then needs to be changed all over the place, i.e. also for all the non-x86 machines, defeating the idea of a default value... Maybe it makes more sense to check for a TYPE_X86_MACHINE machine type instead? Thomas
Re: [PATCH v4] hw/rtc/mc146818rtc: Make this rtc device target independent
On 31/12/2022 00.45, Bernhard Beschow wrote: Am 29. Dezember 2022 10:58:48 UTC schrieb Thomas Huth : The only reason for this code being target dependent is the apic-related code in rtc_policy_slew_deliver_irq(). Since these apic functions are rather simple, we can easily move them into a new, separate file (apic_irqcount.c) which will always be compiled and linked if either APIC or the mc146818 device are required. This way we can get rid of the #ifdef TARGET_I386 switches in mc146818rtc.c and declare it in the softmmu_ss instead of specific_ss, so that the code only gets compiled once for all targets. Reviewed-by: Philippe Mathieu-Daudé Reviewed-by: Mark Cave-Ayland Signed-off-by: Thomas Huth --- v4: Check for QEMU_ARCH_I386 instead of looking for an APIC Can we find a more appropriate name for the helpers than "apic" then? If the slew tick policy is a workaround for (x86-) KVM I propose to do s/apic/kvm/ while still compiling for every target. Yes, since the IRQ-counting is also used by the old i8259 PIC, it likely makes sense to rename the functions. static uint32_t rtc_periodic_clock_ticks(RTCState *s) { @@ -922,14 +911,15 @@ static void rtc_realizefn(DeviceState *dev, Error **errp) rtc_set_date_from_host(isadev); switch (s->lost_tick_policy) { -#ifdef TARGET_I386 -case LOST_TICK_POLICY_SLEW: -s->coalesced_timer = -timer_new_ns(rtc_clock, rtc_coalesced_timer, s); -break; -#endif case LOST_TICK_POLICY_DISCARD: break; +case LOST_TICK_POLICY_SLEW: +/* Slew tick policy is only available on x86 */ +if (arch_type == QEMU_ARCH_I386) { This reflects the intention much better than before, which is nice. How does `arch_type` play together with qemu-system-all? IIUC it should be possible to load all arch backends simultaneously while `arch_type` is an external symbol defined by each arch backend differently. So this seems to conflict. I assume that there still will be a main arch_type for the current selected machine? ... not sure how this will exactly work, though ... Can we just add a property such as "slew-tick-policy-available" instead? It should default to false and all x86 machines would need to opt in explicitly. Sounds like a good idea, it's certainly better than checking arch_type here ... I'll give it a try, thanks! Thomas
Re: [PATCH v4] hw/rtc/mc146818rtc: Make this rtc device target independent
Am 29. Dezember 2022 10:58:48 UTC schrieb Thomas Huth : >The only reason for this code being target dependent is the apic-related >code in rtc_policy_slew_deliver_irq(). Since these apic functions are rather >simple, we can easily move them into a new, separate file (apic_irqcount.c) >which will always be compiled and linked if either APIC or the mc146818 device >are required. This way we can get rid of the #ifdef TARGET_I386 switches in >mc146818rtc.c and declare it in the softmmu_ss instead of specific_ss, so >that the code only gets compiled once for all targets. > >Reviewed-by: Philippe Mathieu-Daudé >Reviewed-by: Mark Cave-Ayland >Signed-off-by: Thomas Huth >--- > v4: Check for QEMU_ARCH_I386 instead of looking for an APIC Can we find a more appropriate name for the helpers than "apic" then? If the slew tick policy is a workaround for (x86-) KVM I propose to do s/apic/kvm/ while still compiling for every target. > > include/hw/i386/apic.h | 1 + > include/hw/i386/apic_internal.h | 1 - > include/hw/rtc/mc146818rtc.h| 1 + > hw/intc/apic_common.c | 27 - > hw/intc/apic_irqcount.c | 53 + > hw/rtc/mc146818rtc.c| 26 +--- > hw/intc/meson.build | 6 +++- > hw/rtc/meson.build | 3 +- > 8 files changed, 69 insertions(+), 49 deletions(-) > create mode 100644 hw/intc/apic_irqcount.c > >diff --git a/include/hw/i386/apic.h b/include/hw/i386/apic.h >index da1d2fe155..96b9939425 100644 >--- a/include/hw/i386/apic.h >+++ b/include/hw/i386/apic.h >@@ -9,6 +9,7 @@ int apic_accept_pic_intr(DeviceState *s); > void apic_deliver_pic_intr(DeviceState *s, int level); > void apic_deliver_nmi(DeviceState *d); > int apic_get_interrupt(DeviceState *s); >+void apic_report_irq_delivered(int delivered); > void apic_reset_irq_delivered(void); > int apic_get_irq_delivered(void); > void cpu_set_apic_base(DeviceState *s, uint64_t val); >diff --git a/include/hw/i386/apic_internal.h b/include/hw/i386/apic_internal.h >index c175e7e718..e61ad04769 100644 >--- a/include/hw/i386/apic_internal.h >+++ b/include/hw/i386/apic_internal.h >@@ -199,7 +199,6 @@ typedef struct VAPICState { > > extern bool apic_report_tpr_access; > >-void apic_report_irq_delivered(int delivered); > bool apic_next_timer(APICCommonState *s, int64_t current_time); > void apic_enable_tpr_access_reporting(DeviceState *d, bool enable); > void apic_enable_vapic(DeviceState *d, hwaddr paddr); >diff --git a/include/hw/rtc/mc146818rtc.h b/include/hw/rtc/mc146818rtc.h >index 1db0fcee92..45bcd6f040 100644 >--- a/include/hw/rtc/mc146818rtc.h >+++ b/include/hw/rtc/mc146818rtc.h >@@ -55,5 +55,6 @@ ISADevice *mc146818_rtc_init(ISABus *bus, int base_year, > qemu_irq intercept_irq); > void rtc_set_memory(ISADevice *dev, int addr, int val); > int rtc_get_memory(ISADevice *dev, int addr); >+void qmp_rtc_reset_reinjection(Error **errp); > > #endif /* HW_RTC_MC146818RTC_H */ >diff --git a/hw/intc/apic_common.c b/hw/intc/apic_common.c >index 2a20982066..b0f85f9384 100644 >--- a/hw/intc/apic_common.c >+++ b/hw/intc/apic_common.c >@@ -33,7 +33,6 @@ > #include "hw/sysbus.h" > #include "migration/vmstate.h" > >-static int apic_irq_delivered; > bool apic_report_tpr_access; > > void cpu_set_apic_base(DeviceState *dev, uint64_t val) >@@ -122,32 +121,6 @@ void apic_handle_tpr_access_report(DeviceState *dev, >target_ulong ip, > vapic_report_tpr_access(s->vapic, CPU(s->cpu), ip, access); > } > >-void apic_report_irq_delivered(int delivered) >-{ >-apic_irq_delivered += delivered; >- >-trace_apic_report_irq_delivered(apic_irq_delivered); >-} >- >-void apic_reset_irq_delivered(void) >-{ >-/* Copy this into a local variable to encourage gcc to emit a plain >- * register for a sys/sdt.h marker. For details on this workaround, see: >- * https://sourceware.org/bugzilla/show_bug.cgi?id=13296 >- */ >-volatile int a_i_d = apic_irq_delivered; >-trace_apic_reset_irq_delivered(a_i_d); >- >-apic_irq_delivered = 0; >-} >- >-int apic_get_irq_delivered(void) >-{ >-trace_apic_get_irq_delivered(apic_irq_delivered); >- >-return apic_irq_delivered; >-} >- > void apic_deliver_nmi(DeviceState *dev) > { > APICCommonState *s = APIC_COMMON(dev); >diff --git a/hw/intc/apic_irqcount.c b/hw/intc/apic_irqcount.c >new file mode 100644 >index 00..0aadef1cb5 >--- /dev/null >+++ b/hw/intc/apic_irqcount.c >@@ -0,0 +1,53 @@ >+/* >+ * APIC support - functions for counting the delivered IRQs. >+ * (this code is in a separate file since it is used from the >+ * mc146818rtc code on targets without APIC, too) >+ * >+ * Copyright (c) 2011 Jan Kiszka, Siemens AG >+ * >+ * This library is free software; you can redistribute it and/or >+ * modify it under the terms of the GNU Lesser General Public >+ * License as published by the Free Software Foundation; either >+ * version 2.1 of the License, or (at your