pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
Subsystems (3)
v2 cover pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
PAPR v2.12 defines a new hypercall, H_WATCHDOG. This patch series adds support for this hypercall to powerpc/pseries kernels and introduces a new watchdog driver, "pseries-wdt", for the virtual timers exposed by the hypercall.
This series is preceded by the following:
RFC v1: https://lore.kernel.org/linux-watchdog/[email protected]/
RFC v2: https://lore.kernel.org/linux-watchdog/[email protected]/
PATCH v1: https://lore.kernel.org/linux-watchdog/[email protected]/
Changes of note from PATCH v1:
- Trim down the large comment documenting the H_WATCHDOG hypercall. The comment is likely to rot, so remove anything we aren't using and anything overly obvious.
- Remove any preprocessor definitions not actually used in the module right now. If we want to use other features offered by the hypercall we can add them in later. They're just clutter until then.
- Simplify the "action" module parameter. The value is now an index into an array of possible timeoutAction values. This design removes the need for the custom get/set methods used in PATCH v1.
Now we merely need to check that the "action" value is a valid index during pseries_wdt_probe(). Easy.
- Make the timeoutAction a member of pseries_wdt, "action". This eliminates the use of a global variable during pseries_wdt_start().
- Use watchdog_init_timeout() idiomatically. Check its return value and error out of pseries_wdt_probe() if it fails.
PATCH v2 1/4 powerpc/pseries: hvcall.h: add H_WATCHDOG opcode, H_NOOP return code
PAPR v2.12 defines a new hypercall, H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers.
Add the opcode for the H_WATCHDOG hypercall to hvcall.h. While here, add a definition for H_NOOP, a possible return code for H_WATCHDOG.
--- arch/powerpc/include/asm/hvcall.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/powerpc/include/asm/hvcall.h +3 -1
@@ -87,6 +87,7 @@#define H_P7 -60#define H_P8 -61#define H_P9 -62 +#define H_NOOP -63#define H_TOO_BIG -64#define H_UNSUPPORTED -67#define H_OVERLAP -68
@@ -324,7 +325,8 @@#define H_RPT_INVALIDATE 0x448#define H_SCM_FLUSH 0x44C#define H_GET_ENERGY_SCALE_INFO 0x450 -#define MAX_HCALL_OPCODE H_GET_ENERGY_SCALE_INFO +#define H_WATCHDOG 0x45C +#define MAX_HCALL_OPCODE H_WATCHDOG/* Scope args for H_SCM_UNBIND_ALL */ #define H_UNBIND_SCOPE_ALL (0x1)
Signature · 1 line
2.27.0
Re: [PATCH v2 1/4] powerpc/pseries: hvcall.h: add H_WATCHDOG opcode, H_NOOP return code
Quoted 16 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers.Add the opcode for the
H_WATCHDOGhypercall tohvcall.h. While here, add a definition forH_NOOP, a possible return code forH_WATCHDOG.Signed-off-by Scott Cheloha <[email protected]>--- arch/powerpc/include/asm/hvcall.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)arch/powerpc/include/asm/hvcall.h +0
-0
arch/powerpc/include/asm/hvcall.h line 93
Quoted 7 lines
arch/powerpc/include/asm/hvcall.h +1
-0@@ -87,6 +87,7 @@#define H_P7 -60#define H_P8 -61#define H_P9 -62 +#define H_NOOP -63#define H_TOO_BIG -64#define H_UNSUPPORTED -67#define H_OVERLAP -68
Not a problem to fix in your series, but I guess these should be parenthesized i.e.
#define H_P7 (-60)
#define H_P8 (-61)
#define H_P9 (-62)
#define H_NOOP (-63)
arch/powerpc/include/asm/hvcall.h line 330
Quoted 8 lines
arch/powerpc/include/asm/hvcall.h +2
-1@@ -324,7 +325,8 @@#define H_RPT_INVALIDATE 0x448#define H_SCM_FLUSH 0x44C#define H_GET_ENERGY_SCALE_INFO 0x450 -#define MAX_HCALL_OPCODE H_GET_ENERGY_SCALE_INFO +#define H_WATCHDOG 0x45C +#define MAX_HCALL_OPCODE H_WATCHDOG
Looks fine.
Re: [PATCH v2 1/4] powerpc/pseries: hvcall.h: add H_WATCHDOG opcode, H_NOOP return code
Hi!
Quoted 7 lines · Nathan Lynch
Not a problem to fix in your series, but I guess these should be parenthesized i.e.
#define
H_P7(-60)
#defineH_P8(-61)
#defineH_P9(-62)
#defineH_NOOP(-63)
Why? It does not change the semantics of any correct code. For what incorrect code will it make the diagnostics clearer?
Segher
Re: [PATCH v2 1/4] powerpc/pseries: hvcall.h: add H_WATCHDOG opcode, H_NOOP return code
Quoted 6 lines · Segher Boessenkool
Hi!
Quoted 3 lines · Nathan Lynch
Not a problem to fix in your series, but I guess these should be parenthesized i.e.
Quoted 7 lines
Quoted 4 lines
#define
H_P7(-60)
#defineH_P8(-61)
#defineH_P9(-62)
#defineH_NOOP(-63)Why? It does not change the semantics of any correct code. For what incorrect code will it make the diagnostics clearer?
Yeah never mind, I was confused.
PATCH v2 2/4 powerpc/pseries: add FW_FEATURE_WATCHDOG flag
In reply to Scott ChelohaPAPR v2.12 specifies a new optional function set, "hcall-watchdog", for the /rtas/ibm,hypertas-functions property. The presence of this function set indicates support for the H_WATCHDOG hypercall.
Check for this function set and, if present, set the new FW_FEATURE_WATCHDOG flag.
--- arch/powerpc/include/asm/firmware.h | 3 ++- arch/powerpc/platforms/pseries/firmware.c | 1 + 2 files changed, 3 insertions(+), 1 deletion(-)
arch/powerpc/include/asm/firmware.h +2 -1
@@ -55,6 +55,7 @@#define FW_FEATURE_RPT_INVALIDATE ASM_CONST(0x0000010000000000)#define FW_FEATURE_FORM2_AFFINITY ASM_CONST(0x0000020000000000)#define FW_FEATURE_ENERGY_SCALE_INFO ASM_CONST(0x0000040000000000) +#define FW_FEATURE_WATCHDOG ASM_CONST(0x0000080000000000)#ifndef __ASSEMBLY__
@@ -76,7 +77,7 @@ enum {FW_FEATURE_DRC_INFO |FW_FEATURE_BLOCK_REMOVE |FW_FEATURE_PAPR_SCM |FW_FEATURE_ULTRAVISOR |FW_FEATURE_RPT_INVALIDATE |FW_FEATURE_FORM2_AFFINITY | -FW_FEATURE_ENERGY_SCALE_INFO , +FW_FEATURE_ENERGY_SCALE_INFO |FW_FEATURE_WATCHDOG ,FW_FEATURE_PSERIES_ALWAYS =0 ,FW_FEATURE_POWERNV_POSSIBLE =FW_FEATURE_OPAL |FW_FEATURE_ULTRAVISOR , FW_FEATURE_POWERNV_ALWAYS = 0,
arch/powerpc/platforms/pseries/firmware.c +1 -0
@@ -67,6 +67,7 @@ hypertas_fw_features_table[] = { {FW_FEATURE_PAPR_SCM, "hcall-scm"}, {FW_FEATURE_RPT_INVALIDATE, "hcall-rpt-invalidate"}, {FW_FEATURE_ENERGY_SCALE_INFO, "hcall-energy-scale-info"}, + {FW_FEATURE_WATCHDOG, "hcall-watchdog"}, };/* Build up the firmware features bitmask using the contents of
Signature · 1 line
2.27.0
Re: [PATCH v2 2/4] powerpc/pseries: add FW_FEATURE_WATCHDOG flag
Quoted 8 lines · Scott Cheloha
PAPR v2.12 specifies a new optional function set, "hcall-watchdog", for the /rtas/ibm,hypertas-functions property. The presence of this function set indicates support for the
H_WATCHDOGhypercall.Check for this function set and, if present, set the new
FW_FEATURE_WATCHDOGflag.Signed-off-by Scott Cheloha <[email protected]>
...
Quoted 26 lines
arch/powerpc/include/asm/firmware.h +2
-1@@ -55,6 +55,7 @@#define FW_FEATURE_RPT_INVALIDATE ASM_CONST(0x0000010000000000)#define FW_FEATURE_FORM2_AFFINITY ASM_CONST(0x0000020000000000)#define FW_FEATURE_ENERGY_SCALE_INFO ASM_CONST(0x0000040000000000) +#define FW_FEATURE_WATCHDOG ASM_CONST(0x0000080000000000)#ifndef __ASSEMBLY__@@ -76,7 +77,7 @@ enum {FW_FEATURE_DRC_INFO |FW_FEATURE_BLOCK_REMOVE |FW_FEATURE_PAPR_SCM |FW_FEATURE_ULTRAVISOR |FW_FEATURE_RPT_INVALIDATE |FW_FEATURE_FORM2_AFFINITY | -FW_FEATURE_ENERGY_SCALE_INFO , +FW_FEATURE_ENERGY_SCALE_INFO |FW_FEATURE_WATCHDOG ,FW_FEATURE_PSERIES_ALWAYS =0 ,FW_FEATURE_POWERNV_POSSIBLE =FW_FEATURE_OPAL |FW_FEATURE_ULTRAVISOR , FW_FEATURE_POWERNV_ALWAYS = 0,arch/powerpc/platforms/pseries/firmware.c +0
-0
arch/powerpc/platforms/pseries/firmware.c hypertas_fw_features_table[] = { · line 71
Quoted 5 lines
arch/powerpc/platforms/pseries/firmware.c +1
-0@@ -67,6 +67,7 @@ hypertas_fw_features_table[] = { {FW_FEATURE_PAPR_SCM, "hcall-scm"}, {FW_FEATURE_RPT_INVALIDATE, "hcall-rpt-invalidate"}, {FW_FEATURE_ENERGY_SCALE_INFO, "hcall-energy-scale-info"}, + {FW_FEATURE_WATCHDOG, "hcall-watchdog"}, };
All looks correct.
PATCH v2 3/4 powerpc/pseries: register pseries-wdt device with platform bus
In reply to Scott ChelohaPAPR v2.12 defines a new hypercall, H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers.
These timers do not conform to PowerPC device conventions. They are not affixed to any extant bus, nor do they have full representation in the device tree.
As a workaround we represent them as platform devices.
This patch registers a single platform device, "pseries-wdt", with the platform bus if the FW_FEATURE_WATCHDOG flag is set.
A driver for this device, "pseries-wdt", will be introduced in a subsequent patch.
--- arch/powerpc/platforms/pseries/setup.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)
arch/powerpc/platforms/pseries/setup.c +17 -0
@@ -14,6 +14,7 @@#include <linux/cpu.h> #include <linux/errno.h> +#include <linux/platform_device.h> #include <linux/sched.h> #include <linux/kernel.h> #include <linux/mm.h>
@@ -169,6 +170,22 @@ static void __init fwnmi_init(void)#endif } +/* +* Affix a device for the first timer to the platform bus if +* we have firmware support for the H_WATCHDOG hypercall. +*/ +static struct platform_device *pseries_wdt_pdev; + +static __init int pseries_wdt_init (void ) +{ +if (!firmware_has_feature (FW_FEATURE_WATCHDOG )) +return 0 ; + pseries_wdt_pdev =platform_device_register_simple ("pseries-wdt" , +0 ,NULL ,0 ); +return 0 ; +} +machine_subsys_initcall (pseries, pseries_wdt_init); +static void pseries_8259_cascade (struct irq_desc *desc) {struct irq_chip *chip =irq_desc_get_chip (desc);
Signature · 1 line
2.27.0
Re: [PATCH v2 3/4] powerpc/pseries: register pseries-wdt device with platform bus
Quoted 42 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers.These timers do not conform to PowerPC device conventions. They are not affixed to any extant bus, nor do they have full representation in the device tree.
As a workaround we represent them as platform devices.
This patch registers a single platform device, "pseries-wdt", with the platform bus if the
FW_FEATURE_WATCHDOGflag is set.A driver for this device, "pseries-wdt", will be introduced in a subsequent patch.
Signed-off-by Scott Cheloha <[email protected]>--- arch/powerpc/platforms/pseries/setup.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)arch/powerpc/platforms/pseries/setup.c +7
-0@@ -14,6 +14,7 @@#include <linux/cpu.h> #include <linux/errno.h> +#include <linux/platform_device.h> #include <linux/sched.h> #include <linux/kernel.h> #include <linux/mm.h> @@ -169,6 +170,22 @@ static void __init fwnmi_init(void)#endif } +/* +* Affix a device for the first timer to the platform bus if +* we have firmware support for the H_WATCHDOG hypercall. +*/ +static struct platform_device *pseries_wdt_pdev; +
arch/powerpc/platforms/pseries/setup.c static void __init fwnmi_init(void) · line 188
Quoted 10 lines
arch/powerpc/platforms/pseries/setup.c +10
-0+static __init int pseries_wdt_init (void ) +{ +if (!firmware_has_feature (FW_FEATURE_WATCHDOG )) +return 0 ; + pseries_wdt_pdev =platform_device_register_simple ("pseries-wdt" , +0 ,NULL ,0 ); +return 0 ; +} +machine_subsys_initcall (pseries, pseries_wdt_init); +
Seems like we don't need pseries_wdt_pdev as it's unused elsewhere? But that's quite minor.
Michael Ellerman
reviewerRe: [PATCH v2 3/4] powerpc/pseries: register pseries-wdt device with platform bus
Quoted 3 lines · Nathan Lynch
Quoted 2 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers.
...
Quoted 3 lines
Seems like we don't need
pseries_wdt_pdevas it's unused elsewhere? But that's quite minor.
It's minor but please drop it in the next version.
cheers
Re: [PATCH v2 3/4] powerpc/pseries: register pseries-wdt device with platform bus
Quoted 10 lines · Michael Ellerman
Quoted 3 lines · Nathan Lynch
Quoted 2 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers....
Quoted 3 lines
Seems like we don't need
pseries_wdt_pdevas it's unused elsewhere? But that's quite minor.It's minor but please drop it in the next version.
Dropped.
PATCH v2 4/4 watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
In reply to Scott ChelohaPAPR v2.12 defines a new hypercall, H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers. The timers have millisecond granularity. The guest is terminated when a timer expires.
This patch adds a watchdog driver for these timers, "pseries-wdt".
pseries_wdt_probe() currently assumes the existence of only one platform device and always assigns it watchdogNumber 1. If we ever expose more than one timer to userspace we will need to devise a way to assign a distinct watchdogNumber to each platform device at device registration time.
--- .../watchdog/watchdog-parameters.rst | 12 + drivers/watchdog/Kconfig | 8 + drivers/watchdog/Makefile | 1 + drivers/watchdog/pseries-wdt.c | 264 ++++++++++++++++++ 4 files changed, 285 insertions(+) create mode 100644 drivers/watchdog/pseries-wdt.c
Documentation/watchdog/watchdog-parameters.rst +12 -0
@@ -425,6 +425,18 @@ pnx833x_wdt: ------------------------------------------------- +pseries-wdt: + action: + Action taken when watchdog expires: 0 (power off), 1 (restart), + 2 (dump and restart). (default=1) + timeout: + Initial watchdog timeout in seconds. (default=60) + nowayout: + Watchdog cannot be stopped once started. + (default=kernel config parameter) + +------------------------------------------------- + rc32434_wdt: timeout: Watchdog timeout value, in seconds (default=20)
drivers/watchdog/Kconfig +8 -0
@@ -1932,6 +1932,14 @@ config MEN_A21_WDT # PPC64 Architecture +config PSERIES_WDT + tristate "POWER Architecture Platform Watchdog Timer" + depends on PPC_PSERIES + select WATCHDOG_CORE + help + Driver for virtual watchdog timers provided by PAPR + hypervisors (e.g. PowerVM, KVM). + config WATCHDOG_RTAS tristate "RTAS watchdog" depends on PPC_RTAS
drivers/watchdog/Makefile +1 -0
@@ -184,6 +184,7 @@ obj-$(CONFIG_BOOKE_WDT) += booke_wdt.o obj-$(CONFIG_MEN_A21_WDT) += mena21_wdt.o # PPC64 Architecture +obj-$(CONFIG_PSERIES_WDT) += pseries-wdt.o obj-$(CONFIG_WATCHDOG_RTAS) += wdrtas.o # S390 Architecture
drivers/watchdog/pseries-wdt.c new file +264 -0
264 lines not on this page · raw message
Signature · 1 line
2.27.0
Re: [PATCH v2 4/4] watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
Quoted 7 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers. The timers have millisecond granularity. The guest is terminated when a timer expires.This patch adds a watchdog driver for these timers, "pseries-wdt".
Quoted 7 lines
pseries_wdt_probe()currently assumes the existence of only one platform device and always assigns it watchdogNumber 1. If we ever expose more than one timer to userspace we will need to devise a way to assign a distinct watchdogNumber to each platform device at device registration time.Signed-off-by Scott Cheloha <[email protected]>
Besides the patch ordering and 0444 vs. 0644 (which is up to the PPC maintainer to decide anyway :) ), looks good to me.
Quoted 332 lines
--- .../watchdog/watchdog-parameters.rst | 12 + drivers/watchdog/Kconfig | 8 + drivers/watchdog/Makefile | 1 + drivers/watchdog/pseries-wdt.c | 264 ++++++++++++++++++ 4 files changed, 285 insertions(+)create mode 100644
drivers/watchdog/pseries-wdt.cDocumentation/watchdog/watchdog-parameters.rst +12
-0@@ -425,6 +425,18 @@ pnx833x_wdt: ------------------------------------------------- +pseries-wdt: + action: + Action taken when watchdog expires: 0 (power off), 1 (restart), + 2 (dump and restart). (default=1) + timeout: + Initial watchdog timeout in seconds. (default=60) + nowayout: + Watchdog cannot be stopped once started. + (default=kernel config parameter) + +------------------------------------------------- + rc32434_wdt: timeout: Watchdog timeout value, in seconds (default=20)drivers/watchdog/Kconfig +8
-0@@ -1932,6 +1932,14 @@ config MEN_A21_WDT # PPC64 Architecture +config PSERIES_WDT + tristate "POWER Architecture Platform Watchdog Timer" + depends on PPC_PSERIES + select WATCHDOG_CORE + help + Driver for virtual watchdog timers provided by PAPR + hypervisors (e.g. PowerVM, KVM). + config WATCHDOG_RTAS tristate "RTAS watchdog" depends on PPC_RTASdrivers/watchdog/Makefile +1
-0@@ -184,6 +184,7 @@ obj-$(CONFIG_BOOKE_WDT) += booke_wdt.o obj-$(CONFIG_MEN_A21_WDT) += mena21_wdt.o # PPC64 Architecture +obj-$(CONFIG_PSERIES_WDT) += pseries-wdt.o obj-$(CONFIG_WATCHDOG_RTAS) += wdrtas.o # S390 Architecturedrivers/watchdog/pseries-wdt.c new file +264
-0@@ -0,0 +1,264 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* Copyright (c) 2022 International Business Machines, Inc. +*/ + +#include <linux/bitops.h> +#include <linux/kernel.h> +#include <linux/limits.h> +#include <linux/math.h> +#include <linux/mod_devicetable.h> +#include <linux/module.h> +#include <linux/moduleparam.h> +#include <linux/platform_device.h> +#include <linux/watchdog.h> + +#define DRV_NAME "pseries-wdt" + +/* +* The PAPR's MSB->LSB bit ordering is 0->63. These macros simplify +* defining bitfields as described in the PAPR without needing to +* transpose values to the more C-like 63->0 ordering. +*/ +#define SETFIELD (_v, _b, _e) \ + (((unsigned long)(_v) << PPC_BITLSHIFT(_e)) & PPC_BITMASK((_b), (_e))) +#define GETFIELD (_v, _b, _e) \ + (((unsigned long)(_v) & PPC_BITMASK((_b), (_e))) >> PPC_BITLSHIFT(_e)) + +/* +* The H_WATCHDOG hypercall first appears in PAPR v2.12 and is +* described fully in sections 14.5 and 14.15.6. +* +* +* H_WATCHDOG Input +* +* R4: "flags": +* +* Bits 48-55: "operation" +* +* 0x01 Start Watchdog +* 0x02 Stop Watchdog +* 0x03 Query Watchdog Capabilities +*/ +#define PSERIES_WDTF_OP (op) SETFIELD((op), 48, 55) +#define PSERIES_WDTF_OP_START PSERIES_WDTF_OP(0x1) +#define PSERIES_WDTF_OP_STOP PSERIES_WDTF_OP(0x2) +#define PSERIES_WDTF_OP_QUERY PSERIES_WDTF_OP(0x3) + +/* +* Bits 56-63: "timeoutAction" (for "Start Watchdog" only) +* +* 0x01 Hard poweroff +* 0x02 Hard restart +* 0x03 Dump restart +*/ +#define PSERIES_WDTF_ACTION (ac) SETFIELD(ac, 56, 63) +#define PSERIES_WDTF_ACTION_HARD_POWEROFF PSERIES_WDTF_ACTION(0x1) +#define PSERIES_WDTF_ACTION_HARD_RESTART PSERIES_WDTF_ACTION(0x2) +#define PSERIES_WDTF_ACTION_DUMP_RESTART PSERIES_WDTF_ACTION(0x3) + +/* +* H_WATCHDOG Output +* +* R3: Return code +* +* H_SUCCESS The operation completed. +* +* H_BUSY The hypervisor is too busy; retry the operation. +* +* H_PARAMETER The given "flags" are somehow invalid. Either the +* "operation" or "timeoutAction" is invalid, or a +* reserved bit is set. +* +* H_P2 The given "watchdogNumber" is zero or exceeds the +* supported maximum value. +* +* H_P3 The given "timeoutInMs" is below the supported +* minimum value. +* +* H_NOOP The given "watchdogNumber" is already stopped. +* +* H_HARDWARE The operation failed for ineffable reasons. +* +* H_FUNCTION The H_WATCHDOG hypercall is not supported by this +* hypervisor. +* +* R4: +* +* - For the "Query Watchdog Capabilities" operation, a 64-bit +* value structured as follows: +* +* Bits 0-15: The minimum supported timeout in milliseconds. +* Bits 16-31: The number of watchdogs supported. +* Bits 32-63: Reserved. +*/ +#define PSERIES_WDTQ_MIN_TIMEOUT (cap) GETFIELD((cap), 0, 15) +#define PSERIES_WDTQ_MAX_NUMBER (cap) GETFIELD((cap), 16, 31) + +static const unsigned long pseries_wdt_action[] = { + [0 ] =PSERIES_WDTF_ACTION_HARD_POWEROFF , + [1 ] =PSERIES_WDTF_ACTION_HARD_RESTART , + [2 ] =PSERIES_WDTF_ACTION_DUMP_RESTART , +}; + +#define WATCHDOG_ACTION 1 +static unsigned int action =WATCHDOG_ACTION ; +module_param (action, uint,0444 ); +MODULE_PARM_DESC(action, "Action taken when watchdog expires (default=" + __MODULE_STRING(WATCHDOG_ACTION) ")"); + +static bool nowayout =WATCHDOG_NOWAYOUT ; +module_param (nowayout, bool,0444 ); +MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default=" + __MODULE_STRING(WATCHDOG_NOWAYOUT) ")"); + +#define WATCHDOG_TIMEOUT 60 +static unsigned int timeout =WATCHDOG_TIMEOUT ; +module_param (timeout, uint,0444 ); +MODULE_PARM_DESC(timeout, "Initial watchdog timeout in seconds (default=" + __MODULE_STRING(WATCHDOG_TIMEOUT) ")"); + +struct pseries_wdt { +struct watchdog_device wd; +unsigned long action; +unsigned long num;/* Watchdog numbers are 1-based */ +}; + +static int pseries_wdt_start (struct watchdog_device *wdd) +{ +struct device *dev = wdd->parent; +struct pseries_wdt *pw =watchdog_get_drvdata (wdd); +unsigned long flags, msecs; +long rc; + + flags = pw->action |PSERIES_WDTF_OP_START ; + msecs = wdd->timeout *1000UL ; + rc =plpar_hcall_norets (H_WATCHDOG , flags, pw->num, msecs); +if (rc !=H_SUCCESS ) { +dev_crit (dev,"H_WATCHDOG: %ld: failed to start timer %lu" , + rc, pw->num); +return -EIO ; + } +return 0 ; +} + +static int pseries_wdt_stop (struct watchdog_device *wdd) +{ +struct device *dev = wdd->parent; +struct pseries_wdt *pw =watchdog_get_drvdata (wdd); +long rc; + + rc =plpar_hcall_norets (H_WATCHDOG ,PSERIES_WDTF_OP_STOP , pw->num); +if (rc !=H_SUCCESS && rc !=H_NOOP ) { +dev_crit (dev,"H_WATCHDOG: %ld: failed to stop timer %lu" , + rc, pw->num); +return -EIO ; + } +return 0 ; +} + +static struct watchdog_info pseries_wdt_info = { + .identity =DRV_NAME , + .options =WDIOF_KEEPALIVEPING |WDIOF_MAGICCLOSE |WDIOF_SETTIMEOUT + |WDIOF_PRETIMEOUT , +}; + +static const struct watchdog_ops pseries_wdt_ops = { + .owner =THIS_MODULE , + .start = pseries_wdt_start, + .stop = pseries_wdt_stop, +}; + +static int pseries_wdt_probe (struct platform_device *pdev) +{ +unsigned long ret[PLPAR_HCALL_BUFSIZE ] = {0 }; +unsigned long cap; +long rc; +struct pseries_wdt *pw; +int err; + + rc =plpar_hcall (H_WATCHDOG , ret,PSERIES_WDTF_OP_QUERY ); +if (rc ==H_FUNCTION ) +return -ENODEV ; +if (rc !=H_SUCCESS ) +return -EIO ; + cap = ret[0 ]; + + pw =devm_kzalloc (&pdev->dev,sizeof (*pw),GFP_KERNEL ); +if (!pw) +return -ENOMEM ; + +/* +* Assume watchdogNumber 1 for now. If we ever support +* multiple timers we will need to devise a way to choose a +* distinct watchdogNumber for each platform device at device +* registration time. +*/ + pw->num =1 ; + +if (action >=ARRAY_SIZE (pseries_wdt_action)) +return -EINVAL ; + pw->action = pseries_wdt_action[action]; + + pw->wd.parent = &pdev->dev; + pw->wd.info = &pseries_wdt_info; + pw->wd.ops = &pseries_wdt_ops; + pw->wd.min_timeout =DIV_ROUND_UP (PSERIES_WDTQ_MIN_TIMEOUT (cap),1000 ); + pw->wd.max_timeout =UINT_MAX /1000 ; + pw->wd.timeout = timeout; +if (watchdog_init_timeout (&pw->wd,0 ,NULL )) +return -EINVAL ; +watchdog_set_nowayout (&pw->wd, nowayout); +watchdog_stop_on_reboot (&pw->wd); +watchdog_stop_on_unregister (&pw->wd); +watchdog_set_drvdata (&pw->wd, pw); + + err =devm_watchdog_register_device (&pdev->dev, &pw->wd); +if (err) +return err; + +platform_set_drvdata (pdev, &pw->wd); + +return 0 ; +} + +static int pseries_wdt_suspend (struct platform_device *pdev,pm_message_t state) +{ +struct watchdog_device *wd =platform_get_drvdata (pdev); + +if (watchdog_active (wd)) +return pseries_wdt_stop (wd); +return 0 ; +} + +static int pseries_wdt_resume (struct platform_device *pdev) +{ +struct watchdog_device *wd =platform_get_drvdata (pdev); + +if (watchdog_active (wd)) +return pseries_wdt_start (wd); +return 0 ; +} + +static const struct platform_device_id pseries_wdt_id[] = { + { .name ="pseries-wdt" }, + {} +}; +MODULE_DEVICE_TABLE (platform, pseries_wdt_id); + +static struct platform_driver pseries_wdt_driver = { + .driver = { + .name =DRV_NAME , + .owner =THIS_MODULE , + }, + .id_table = pseries_wdt_id, + .probe = pseries_wdt_probe, + .resume = pseries_wdt_resume, + .suspend = pseries_wdt_suspend, +}; +module_platform_driver (pseries_wdt_driver); + +MODULE_AUTHOR ("Alexey Kardashevskiy <[email protected]>" ); +MODULE_AUTHOR ("Scott Cheloha <[email protected]>" ); +MODULE_DESCRIPTION ("POWER Architecture Platform Watchdog Driver" ); +MODULE_LICENSE ("GPL" );
Signature · 1 line
Alexey
Re: [PATCH v2 4/4] watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
In reply to Scott ChelohaQuoted 7 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers. The timers have millisecond granularity. The guest is terminated when a timer expires.This patch adds a watchdog driver for these timers, "pseries-wdt".
Quoted 7 lines
pseries_wdt_probe()currently assumes the existence of only one platform device and always assigns it watchdogNumber 1. If we ever expose more than one timer to userspace we will need to devise a way to assign a distinct watchdogNumber to each platform device at device registration time.Signed-off-by Scott Cheloha <[email protected]>
Michael Ellerman
reviewerRe: [PATCH v2 4/4] watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
In reply to Scott ChelohaHi Scott,
A few comments below ...
Quoted 27 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. The hypercall permits guest control of one or more virtual watchdog timers. The timers have millisecond granularity. The guest is terminated when a timer expires.This patch adds a watchdog driver for these timers, "pseries-wdt".
pseries_wdt_probe()currently assumes the existence of only one platform device and always assigns it watchdogNumber 1. If we ever expose more than one timer to userspace we will need to devise a way to assign a distinct watchdogNumber to each platform device at device registration time.Signed-off-by Scott Cheloha <[email protected]>--- .../watchdog/watchdog-parameters.rst | 12 + drivers/watchdog/Kconfig | 8 + drivers/watchdog/Makefile | 1 + drivers/watchdog/pseries-wdt.c | 264 ++++++++++++++++++ 4 files changed, 285 insertions(+) create mode 100644 drivers/watchdog/pseries-wdt.cDocumentation/watchdog/watchdog-parameters.rst +0
-0
Documentation/watchdog/watchdog-parameters.rst pnx833x_wdt: · line 431
Quoted 7 lines
Documentation/watchdog/watchdog-parameters.rst +4
-0@@ -425,6 +425,18 @@ pnx833x_wdt: ------------------------------------------------- +pseries-wdt: + action: + Action taken when watchdog expires: 0 (power off), 1 (restart), + 2 (dump and restart). (default=1)
I doesn't look like these values match what other drivers use to any great extent.
So why not use the values from PAPR directly?
ie. 1 = power off, 2 = hard reset, 3 = dump & restart.
It seems like it would be easier to follow if the values map directly.
It's possible in future PAPR adds 247 to mean something, in which case maybe we'd want to map that to a less silly value, but at least for now the PAPR values are sensible enough.
Documentation/watchdog/watchdog-parameters.rst pnx833x_wdt: · line 433
Quoted 2 lines
Documentation/watchdog/watchdog-parameters.rst +2
-0+ timeout: + Initial watchdog timeout in seconds. (default=60)
That seems like a pretty common value, I don't see any guidance in PAPR. Do we have any input from PowerVM on whether that's a good value?
Quoted 22 lines
drivers/watchdog/pseries-wdt.c new file +16
-0@@ -0,0 +1,264 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* Copyright (c) 2022 International Business Machines, Inc. +*/ + +#include <linux/bitops.h> +#include <linux/kernel.h> +#include <linux/limits.h> +#include <linux/math.h> +#include <linux/mod_devicetable.h> +#include <linux/module.h> +#include <linux/moduleparam.h> +#include <linux/platform_device.h> +#include <linux/watchdog.h> + +#define DRV_NAME "pseries-wdt"
drivers/watchdog/pseries-wdt.c line 26
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+ +/* +* The PAPR's MSB->LSB bit ordering is 0->63. These macros simplify +* defining bitfields as described in the PAPR without needing to +* transpose values to the more C-like 63->0 ordering. +*/ +#define SETFIELD (_v, _b, _e) \ + (((unsigned long)(_v) << PPC_BITLSHIFT(_e)) & PPC_BITMASK((_b), (_e))) +#define GETFIELD (_v, _b, _e) \ + (((unsigned long)(_v) & PPC_BITMASK((_b), (_e))) >> PPC_BITLSHIFT(_e))
This will probably sound like a cranky maintainer rant, but ..., I really dislike these GETFIELD/SETFIELD macros.
I know you didn't invent them, but I would be much happier if you didn't use them.
I know they (slightly) simplify things when you're transcribing values from PAPR into the source, but that happens only once.
And then for the rest of eternity the source is harder to read because there's this ridiculous level of indirection through insane macros just to define some constants.
Anyone trying to use a debugger against this code will see a value in memory like 0x200 and have to sit down and work out which SETFIELD() macro it corresponds to.
Quoted 9 lines
drivers/watchdog/pseries-wdt.c new file +9
-0+/* +* The H_WATCHDOG hypercall first appears in PAPR v2.12 and is +* described fully in sections 14.5 and 14.15.6. +* +* +* H_WATCHDOG Input +* +* R4: "flags": +*
drivers/watchdog/pseries-wdt.c line 46
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+* Bits 48-55: "operation" +* +* 0x01 Start Watchdog +* 0x02 Stop Watchdog +* 0x03 Query Watchdog Capabilities +*/ +#define PSERIES_WDTF_OP (op) SETFIELD((op), 48, 55) +#define PSERIES_WDTF_OP_START PSERIES_WDTF_OP(0x1) +#define PSERIES_WDTF_OP_STOP PSERIES_WDTF_OP(0x2) +#define PSERIES_WDTF_OP_QUERY PSERIES_WDTF_OP(0x3)
eg, IMHO these are much more reader friendly:
#define PSERIES_WDTF_OP_START (1 << 8)
#define PSERIES_WDTF_OP_STOP (2 << 8)
#define PSERIES_WDTF_OP_QUERY (3 << 8)
drivers/watchdog/pseries-wdt.c line 58
Quoted 11 lines
drivers/watchdog/pseries-wdt.c new file +11
-0+/* +* Bits 56-63: "timeoutAction" (for "Start Watchdog" only) +* +* 0x01 Hard poweroff +* 0x02 Hard restart +* 0x03 Dump restart +*/ +#define PSERIES_WDTF_ACTION (ac) SETFIELD(ac, 56, 63) +#define PSERIES_WDTF_ACTION_HARD_POWEROFF PSERIES_WDTF_ACTION(0x1) +#define PSERIES_WDTF_ACTION_HARD_RESTART PSERIES_WDTF_ACTION(0x2) +#define PSERIES_WDTF_ACTION_DUMP_RESTART PSERIES_WDTF_ACTION(0x3)
These are a slam dunk:
#define PSERIES_WDTF_ACTION_HARD_POWEROFF 1
#define PSERIES_WDTF_ACTION_HARD_RESTART 2
#define PSERIES_WDTF_ACTION_DUMP_RESTART 3
Quoted 27 lines
drivers/watchdog/pseries-wdt.c new file +27
-0+ +/* +* H_WATCHDOG Output +* +* R3: Return code +* +* H_SUCCESS The operation completed. +* +* H_BUSY The hypervisor is too busy; retry the operation. +* +* H_PARAMETER The given "flags" are somehow invalid. Either the +* "operation" or "timeoutAction" is invalid, or a +* reserved bit is set. +* +* H_P2 The given "watchdogNumber" is zero or exceeds the +* supported maximum value. +* +* H_P3 The given "timeoutInMs" is below the supported +* minimum value. +* +* H_NOOP The given "watchdogNumber" is already stopped. +* +* H_HARDWARE The operation failed for ineffable reasons. +* +* H_FUNCTION The H_WATCHDOG hypercall is not supported by this +* hypervisor. +*
drivers/watchdog/pseries-wdt.c line 95
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+* R4: +* +* - For the "Query Watchdog Capabilities" operation, a 64-bit +* value structured as follows: +* +* Bits 0-15: The minimum supported timeout in milliseconds. +* Bits 16-31: The number of watchdogs supported. +* Bits 32-63: Reserved. +*/ +#define PSERIES_WDTQ_MIN_TIMEOUT (cap) GETFIELD((cap), 0, 15)
This one is less obviously better, but I still think it's clearer as all the logic is there in front of you, rather than hidden in the macro. It is clearer that we're only returning a 16-bit value.
#define PSERIES_WDTQ_MIN_TIMEOUT(cap) (((cap) >> 48) & 0xffff)
drivers/watchdog/pseries-wdt.c line 96
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+#define PSERIES_WDTQ_MAX_NUMBER (cap) GETFIELD((cap), 16, 31)
That's unused.
I guess we're assuming at least one timer is always supported? Seems reasonable.
drivers/watchdog/pseries-wdt.c line 102
Quoted 6 lines
drivers/watchdog/pseries-wdt.c new file +6
-0+ +static const unsigned long pseries_wdt_action[] = { + [0 ] =PSERIES_WDTF_ACTION_HARD_POWEROFF , + [1 ] =PSERIES_WDTF_ACTION_HARD_RESTART , + [2 ] =PSERIES_WDTF_ACTION_DUMP_RESTART , +};
If we used the PAPR values we wouldn't need that ^
drivers/watchdog/pseries-wdt.c line 104
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+#define WATCHDOG_ACTION 1
DEFAULT_ACTION ?
drivers/watchdog/pseries-wdt.c line 115
Quoted 11 lines
drivers/watchdog/pseries-wdt.c new file +11
-0+static unsigned int action =WATCHDOG_ACTION ; +module_param (action, uint,0444 ); +MODULE_PARM_DESC(action, "Action taken when watchdog expires (default=" + __MODULE_STRING(WATCHDOG_ACTION) ")"); + +static bool nowayout =WATCHDOG_NOWAYOUT ; +module_param (nowayout, bool,0444 ); +MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default=" + __MODULE_STRING(WATCHDOG_NOWAYOUT) ")"); + +#define WATCHDOG_TIMEOUT 60
DEFAULT_TIMEOUT ?
drivers/watchdog/pseries-wdt.c line 124
Quoted 9 lines
drivers/watchdog/pseries-wdt.c new file +9
-0+static unsigned int timeout =WATCHDOG_TIMEOUT ; +module_param (timeout, uint,0444 ); +MODULE_PARM_DESC(timeout, "Initial watchdog timeout in seconds (default=" + __MODULE_STRING(WATCHDOG_TIMEOUT) ")"); + +struct pseries_wdt { +struct watchdog_device wd; +unsigned long action; +unsigned long num;/* Watchdog numbers are 1-based */
num can just be an int.
But do we even need it, do we anticipate supporting multiple timers? Should we just hard code '1' ?
drivers/watchdog/pseries-wdt.c line 134
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+}; + +static int pseries_wdt_start (struct watchdog_device *wdd) +{ +struct device *dev = wdd->parent; +struct pseries_wdt *pw =watchdog_get_drvdata (wdd); +unsigned long flags, msecs; +long rc; + + flags = pw->action |PSERIES_WDTF_OP_START ;
We set pw->action at probe time based on the module param action, but this is the only place we use it.
If we use the PAPR values, this could just be:
flags = (pw->action << 8) | PSERIES_WDTF_OP_START;
And is there any benefit in storing action in pseries_wdt, we could just use the module param value here.
drivers/watchdog/pseries-wdt.c line 135
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+ msecs = wdd->timeout *1000UL ;
Using MSEC_PER_SEC makes it clearer what that conversion is doing.
Quoted 18 lines
drivers/watchdog/pseries-wdt.c new file +18
-0+ rc =plpar_hcall_norets (H_WATCHDOG , flags, pw->num, msecs); +if (rc !=H_SUCCESS ) { +dev_crit (dev,"H_WATCHDOG: %ld: failed to start timer %lu" , + rc, pw->num); +return -EIO ; + } +return 0 ; +} + +static int pseries_wdt_stop (struct watchdog_device *wdd) +{ +struct device *dev = wdd->parent; +struct pseries_wdt *pw =watchdog_get_drvdata (wdd); +long rc; + + rc =plpar_hcall_norets (H_WATCHDOG ,PSERIES_WDTF_OP_STOP , pw->num); +if (rc !=H_SUCCESS && rc !=H_NOOP ) { + dev_crit(dev, "H_WATCHDOG: %ld: failed to stop timer %lu",
drivers/watchdog/pseries-wdt.c line 163
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+ rc, pw->num); +return -EIO ; + } +return 0 ; +} + +static struct watchdog_info pseries_wdt_info = { + .identity =DRV_NAME , + .options =WDIOF_KEEPALIVEPING |WDIOF_MAGICCLOSE |WDIOF_SETTIMEOUT + |WDIOF_PRETIMEOUT ,
I don't know the watchdog code to know if those make sense.
Quoted 5 lines
drivers/watchdog/pseries-wdt.c new file +5
-0+}; + +static const struct watchdog_ops pseries_wdt_ops = { + .owner =THIS_MODULE , + .start = pseries_wdt_start,
drivers/watchdog/pseries-wdt.c line 178
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+ .stop = pseries_wdt_stop, +}; + +static int pseries_wdt_probe (struct platform_device *pdev) +{ +unsigned long ret[PLPAR_HCALL_BUFSIZE ] = {0 }; +unsigned long cap; +long rc; +struct pseries_wdt *pw; +int err;
Try to use reverse xmas tree for new code please.
Quoted 17 lines
drivers/watchdog/pseries-wdt.c new file +17
-0+ rc =plpar_hcall (H_WATCHDOG , ret,PSERIES_WDTF_OP_QUERY ); +if (rc ==H_FUNCTION ) +return -ENODEV ; +if (rc !=H_SUCCESS ) +return -EIO ; + cap = ret[0 ]; + + pw =devm_kzalloc (&pdev->dev,sizeof (*pw),GFP_KERNEL ); +if (!pw) +return -ENOMEM ; + +/* +* Assume watchdogNumber 1 for now. If we ever support +* multiple timers we will need to devise a way to choose a +* distinct watchdogNumber for each platform device at device +* registration time. +*/
drivers/watchdog/pseries-wdt.c line 206
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+ pw->num =1 ; + +if (action >=ARRAY_SIZE (pseries_wdt_action)) +return -EINVAL ; + pw->action = pseries_wdt_action[action]; + + pw->wd.parent = &pdev->dev; + pw->wd.info = &pseries_wdt_info; + pw->wd.ops = &pseries_wdt_ops; + pw->wd.min_timeout =DIV_ROUND_UP (PSERIES_WDTQ_MIN_TIMEOUT (cap),1000 );
MSEC_TO_SEC again?
drivers/watchdog/pseries-wdt.c line 207
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+ pw->wd.max_timeout =UINT_MAX /1000 ;
Where does that value come from?
drivers/watchdog/pseries-wdt.c line 210
Quoted 3 lines
drivers/watchdog/pseries-wdt.c new file +3
-0+ pw->wd.timeout = timeout; +if (watchdog_init_timeout (&pw->wd,0 ,NULL )) +return -EINVAL ;
It's late so maybe I'm misreading it, but does watchdog_init_timeout() actually clamp the values if we don't pass a timeout?
It looks like basically a nop when we pass timeout_param=0 and dev=NULL.
Which makes me think we aren't checking anywhere that the timeout we are using >= what firmware will accept.
Quoted 37 lines
drivers/watchdog/pseries-wdt.c new file +37
-0+watchdog_set_nowayout (&pw->wd, nowayout); +watchdog_stop_on_reboot (&pw->wd); +watchdog_stop_on_unregister (&pw->wd); +watchdog_set_drvdata (&pw->wd, pw); + + err =devm_watchdog_register_device (&pdev->dev, &pw->wd); +if (err) +return err; + +platform_set_drvdata (pdev, &pw->wd); + +return 0 ; +} + +static int pseries_wdt_suspend (struct platform_device *pdev,pm_message_t state) +{ +struct watchdog_device *wd =platform_get_drvdata (pdev); + +if (watchdog_active (wd)) +return pseries_wdt_stop (wd); +return 0 ; +} + +static int pseries_wdt_resume (struct platform_device *pdev) +{ +struct watchdog_device *wd =platform_get_drvdata (pdev); + +if (watchdog_active (wd)) +return pseries_wdt_start (wd); +return 0 ; +} + +static const struct platform_device_id pseries_wdt_id[] = { + { .name ="pseries-wdt" }, + {} +}; +MODULE_DEVICE_TABLE (platform, pseries_wdt_id);
drivers/watchdog/pseries-wdt.c line 257
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+ +static struct platform_driver pseries_wdt_driver = { + .driver = { + .name =DRV_NAME , + .owner =THIS_MODULE , + }, + .id_table = pseries_wdt_id, + .probe = pseries_wdt_probe, + .resume = pseries_wdt_resume, + .suspend = pseries_wdt_suspend,
I don't see any handling of the possible requirement to suspend timers across LPM. I don't think just wiring these up is enough?
drivers/watchdog/pseries-wdt.c line 262
Quoted 5 lines
drivers/watchdog/pseries-wdt.c new file +5
-0+}; +module_platform_driver (pseries_wdt_driver); + +MODULE_AUTHOR ("Alexey Kardashevskiy <[email protected]>" ); +MODULE_AUTHOR ("Scott Cheloha <[email protected]>" );
I'd prefer the module authors were just the names, email addresses inevitably bitrot.
Your email address is in the change log.
cheers
Quoted 4 lines
drivers/watchdog/pseries-wdt.c new file +2
-0+MODULE_DESCRIPTION ("POWER Architecture Platform Watchdog Driver" ); +MODULE_LICENSE ("GPL" );Signature · 1 line
2.27.0
Re: [PATCH v2 4/4] watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
Quoted 10 lines · Michael Ellerman
Quoted 8 lines · Scott Cheloha
+ * - For the "Query Watchdog Capabilities" operation, a 64-bit
+ * value structured as follows:
+ *+ * Bits 0-15: The minimum supported timeout in milliseconds. + * Bits 16-31: The number of watchdogs supported. + * Bits 32-63: Reserved.+ */
+#definePSERIES_WDTQ_MIN_TIMEOUT(cap)GETFIELD((cap), 0, 15)
Quoted 5 lines
This one is less obviously better, but I still think it's clearer as all the logic is there in front of you, rather than hidden in the macro. It is clearer that we're only returning a 16-bit value.
#define
PSERIES_WDTQ_MIN_TIMEOUT(cap)(((cap) >> 48) &0xffff)
Or even
((cap) >> 48)
since it is a 64-bit value. If you want better defences you should not use macros here at all, anyway (but inline functions, instead).
I could rant about the 1000UL being meaningless and/or misleading, or that 0x1 is just silly, but it is a sunny day :-)
Segher
Re: [PATCH v2 4/4] watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
In reply to Michael EllermanQuoted 26 lines · Michael Ellerman
Hi Scott,
A few comments below ...
Quoted 8 lines · Scott Cheloha
[...]
Documentation/watchdog/watchdog-parameters.rst +0
-0Documentation/watchdog/watchdog-parameters.rst pnx833x_wdt: · line 431
Quoted 7 lines
Documentation/watchdog/watchdog-parameters.rst +4
-0@@ -425,6 +425,18 @@ pnx833x_wdt: ------------------------------------------------- +pseries-wdt: + action: + Action taken when watchdog expires: 0 (power off), 1 (restart), + 2 (dump and restart). (default=1)I doesn't look like these values match what other drivers use to any great extent.
So why not use the values from PAPR directly?
Quoted 7 lines
ie. 1 = power off, 2 = hard reset, 3 = dump & restart.
It seems like it would be easier to follow if the values map directly.
It's possible in future PAPR adds 247 to mean something, in which case maybe we'd want to map that to a less silly value, but at least for now the PAPR values are sensible enough.
I tried using 1-2-3 in Patch v1 but Guenter objected and we switched:
https://lore.kernel.org/linux-watchdog/[email protected]/
I think the code is fine to read as-is. We're not expecting the administrator to read the PAPR, right? So 1-2-3 is not any more intuitive for the user than 0-1-2.
Given that it's all arbitrary and there aren't any hard rules for module parameters outside of general programmer "that seems fine"-ness, I would really like to leave the numbers as-is.
Quoted 5 lines
Documentation/watchdog/watchdog-parameters.rst pnx833x_wdt: · line 433
Quoted 2 lines
Documentation/watchdog/watchdog-parameters.rst +2
-0+ timeout: + Initial watchdog timeout in seconds. (default=60)That seems like a pretty common value, I don't see any guidance in PAPR. Do we have any input from PowerVM on whether that's a good value?
Currently the minimum timeout is 500ms on all the builds I've tried. I doubt the minimum will ever be anywhere near as large as 60s on a practical H_WATCHDOG implementation, so I don't think there is any risk of the driver failing to probe.
Real software using the watchdog API will set a timeout to a smaller value if it needs to.
60 seconds gives userland ample time to reconfigure the watchdog without risk of it expiring in the midst of a bunch of ioctl(2) calls before they reach the main loop.
Quoted 42 lines
Quoted 22 lines
drivers/watchdog/pseries-wdt.c new file +16
-0@@ -0,0 +1,264 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* Copyright (c) 2022 International Business Machines, Inc. +*/ + +#include <linux/bitops.h> +#include <linux/kernel.h> +#include <linux/limits.h> +#include <linux/math.h> +#include <linux/mod_devicetable.h> +#include <linux/module.h> +#include <linux/moduleparam.h> +#include <linux/platform_device.h> +#include <linux/watchdog.h> + +#define DRV_NAME "pseries-wdt"drivers/watchdog/pseries-wdt.c line 26
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+ +/* +* The PAPR's MSB->LSB bit ordering is 0->63. These macros simplify +* defining bitfields as described in the PAPR without needing to +* transpose values to the more C-like 63->0 ordering. +*/ +#define SETFIELD (_v, _b, _e) \ + (((unsigned long)(_v) << PPC_BITLSHIFT(_e)) & PPC_BITMASK((_b), (_e))) +#define GETFIELD (_v, _b, _e) \ + (((unsigned long)(_v) & PPC_BITMASK((_b), (_e))) >> PPC_BITLSHIFT(_e))This will probably sound like a cranky maintainer rant, but ..., I really dislike these GETFIELD/SETFIELD macros.
I know you didn't invent them, but I would be much happier if you didn't use them.
I know they (slightly) simplify things when you're transcribing values from PAPR into the source, but that happens only once.
Quoted 7 lines
And then for the rest of eternity the source is harder to read because there's this ridiculous level of indirection through insane macros just to define some constants.
Anyone trying to use a debugger against this code will see a value in memory like
0x200and have to sit down and work out whichSETFIELD()macro it corresponds to.
Don't look at me, I never would have come up with them. I got them from Alexey :)
I will drop them.
Quoted 40 lines
Quoted 9 lines
drivers/watchdog/pseries-wdt.c new file +9
-0+/* +* The H_WATCHDOG hypercall first appears in PAPR v2.12 and is +* described fully in sections 14.5 and 14.15.6. +* +* +* H_WATCHDOG Input +* +* R4: "flags": +* drivers/watchdog/pseries-wdt.c line 46
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+* Bits 48-55: "operation" +* +* 0x01 Start Watchdog +* 0x02 Stop Watchdog +* 0x03 Query Watchdog Capabilities +*/ +#define PSERIES_WDTF_OP (op) SETFIELD((op), 48, 55) +#define PSERIES_WDTF_OP_START PSERIES_WDTF_OP(0x1) +#define PSERIES_WDTF_OP_STOP PSERIES_WDTF_OP(0x2) +#define PSERIES_WDTF_OP_QUERY PSERIES_WDTF_OP(0x3)eg, IMHO these are much more reader friendly:
#define
PSERIES_WDTF_OP_START(1 << 8)
#definePSERIES_WDTF_OP_STOP(2 << 8)
#definePSERIES_WDTF_OP_QUERY(3 << 8)drivers/watchdog/pseries-wdt.c line 58
Quoted 11 lines
drivers/watchdog/pseries-wdt.c new file +11
-0+/* +* Bits 56-63: "timeoutAction" (for "Start Watchdog" only) +* +* 0x01 Hard poweroff +* 0x02 Hard restart +* 0x03 Dump restart +*/ +#define PSERIES_WDTF_ACTION (ac) SETFIELD(ac, 56, 63) +#define PSERIES_WDTF_ACTION_HARD_POWEROFF PSERIES_WDTF_ACTION(0x1) +#define PSERIES_WDTF_ACTION_HARD_RESTART PSERIES_WDTF_ACTION(0x2) +#define PSERIES_WDTF_ACTION_DUMP_RESTART PSERIES_WDTF_ACTION(0x3)These are a slam dunk:
Quoted 3 lines
#define
PSERIES_WDTF_ACTION_HARD_POWEROFF1
#definePSERIES_WDTF_ACTION_HARD_RESTART2
#definePSERIES_WDTF_ACTION_DUMP_RESTART3
Yes, yes they are.
Quoted 42 lines
Quoted 27 lines
drivers/watchdog/pseries-wdt.c new file +27
-0+ +/* +* H_WATCHDOG Output +* +* R3: Return code +* +* H_SUCCESS The operation completed. +* +* H_BUSY The hypervisor is too busy; retry the operation. +* +* H_PARAMETER The given "flags" are somehow invalid. Either the +* "operation" or "timeoutAction" is invalid, or a +* reserved bit is set. +* +* H_P2 The given "watchdogNumber" is zero or exceeds the +* supported maximum value. +* +* H_P3 The given "timeoutInMs" is below the supported +* minimum value. +* +* H_NOOP The given "watchdogNumber" is already stopped. +* +* H_HARDWARE The operation failed for ineffable reasons. +* +* H_FUNCTION The H_WATCHDOG hypercall is not supported by this +* hypervisor. +* drivers/watchdog/pseries-wdt.c line 95
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+* R4: +* +* - For the "Query Watchdog Capabilities" operation, a 64-bit +* value structured as follows: +* +* Bits 0-15: The minimum supported timeout in milliseconds. +* Bits 16-31: The number of watchdogs supported. +* Bits 32-63: Reserved. +*/ +#define PSERIES_WDTQ_MIN_TIMEOUT (cap) GETFIELD((cap), 0, 15)This one is less obviously better, but I still think it's clearer as all the logic is there in front of you, rather than hidden in the macro. It is clearer that we're only returning a 16-bit value.
Quoted 8 lines
#define
PSERIES_WDTQ_MIN_TIMEOUT(cap)(((cap) >> 48) &0xffff)drivers/watchdog/pseries-wdt.c line 96
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+#define PSERIES_WDTQ_MAX_NUMBER (cap) GETFIELD((cap), 16, 31)That's unused.
I guess we're assuming at least one timer is always supported? Seems reasonable.
There is a distinction between "we have support for this hypercall" and "you have a timer available to you". We should double-check.
I can't imagine it ever being an issue on a practical, working implementation, but it might save us some debugging if there is ever a hypervisor bug where somehow they allocate us zero timers to work with.
Quoted 9 lines
drivers/watchdog/pseries-wdt.c line 102
Quoted 6 lines
drivers/watchdog/pseries-wdt.c new file +6
-0+ +static const unsigned long pseries_wdt_action[] = { + [0 ] =PSERIES_WDTF_ACTION_HARD_POWEROFF , + [1 ] =PSERIES_WDTF_ACTION_HARD_RESTART , + [2 ] =PSERIES_WDTF_ACTION_DUMP_RESTART , +};If we used the PAPR values we wouldn't need that ^
Quoted 3 lines
drivers/watchdog/pseries-wdt.c line 104
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+#define WATCHDOG_ACTION 1
DEFAULT_ACTION?
The idiom for the default timeout is "WATCHDOG_TIMEOUT" so I went with "WATCHDOG_ACTION".
Quoted 5 lines
Quoted 5 lines
drivers/watchdog/pseries-wdt.c new file +5
-0+static unsigned int action =WATCHDOG_ACTION ; +module_param (action, uint,0444 ); +MODULE_PARM_DESC(action, "Action taken when watchdog expires (default=" + __MODULE_STRING(WATCHDOG_ACTION) ")"); +
Quoted 8 lines
drivers/watchdog/pseries-wdt.c line 115
Quoted 6 lines
drivers/watchdog/pseries-wdt.c new file +6
-0+static bool nowayout =WATCHDOG_NOWAYOUT ; +module_param (nowayout, bool,0444 ); +MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default=" + __MODULE_STRING(WATCHDOG_NOWAYOUT) ")"); + +#define WATCHDOG_TIMEOUT 60
DEFAULT_TIMEOUT?
"WATCHDOG_TIMEOUT" is the idiomatic name for the default timeout in drivers/watchdog/.
Quoted 11 lines
drivers/watchdog/pseries-wdt.c line 124
Quoted 9 lines
drivers/watchdog/pseries-wdt.c new file +9
-0+static unsigned int timeout =WATCHDOG_TIMEOUT ; +module_param (timeout, uint,0444 ); +MODULE_PARM_DESC(timeout, "Initial watchdog timeout in seconds (default=" + __MODULE_STRING(WATCHDOG_TIMEOUT) ")"); + +struct pseries_wdt { +struct watchdog_device wd; +unsigned long action; +unsigned long num;/* Watchdog numbers are 1-based */ num can just be an int.
It's an argument to the hypercall, which takes an unsigned long. Do we need to save 4 bytes?
I guess if we wanted to be precise it should be a 16-bit value.
Quoted 2 lines
But do we even need it, do we anticipate supporting multiple timers? Should we just hard code '1' ?
We have not had a serious discussion about whether more timers in userspace make sense. This code let's us experiment with it, though.
Quoted 14 lines
drivers/watchdog/pseries-wdt.c line 134
Quoted 10 lines
drivers/watchdog/pseries-wdt.c new file +10
-0+}; + +static int pseries_wdt_start (struct watchdog_device *wdd) +{ +struct device *dev = wdd->parent; +struct pseries_wdt *pw =watchdog_get_drvdata (wdd); +unsigned long flags, msecs; +long rc; + + flags = pw->action |PSERIES_WDTF_OP_START ;We set
pw->actionat probe time based on the module param action, but this is the only place we use it.
Quoted 6 lines
If we use the PAPR values, this could just be:
flags = (pw->action << 8) | PSERIES_WDTF_OP_START;And is there any benefit in storing action in
pseries_wdt, we could just use the module param value here.
That was Guenter's idea and I went with it.
Quoted 3 lines
drivers/watchdog/pseries-wdt.c line 135
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+ msecs = wdd->timeout *1000UL ;Using
MSEC_PER_SECmakes it clearer what that conversion is doing.
Done.
Quoted 22 lines
Quoted 22 lines
drivers/watchdog/pseries-wdt.c new file +22
-0+ rc =plpar_hcall_norets (H_WATCHDOG , flags, pw->num, msecs); +if (rc !=H_SUCCESS ) { +dev_crit (dev,"H_WATCHDOG: %ld: failed to start timer %lu" , + rc, pw->num); +return -EIO ; + } +return 0 ; +} + +static int pseries_wdt_stop (struct watchdog_device *wdd) +{ +struct device *dev = wdd->parent; +struct pseries_wdt *pw =watchdog_get_drvdata (wdd); +long rc; + + rc =plpar_hcall_norets (H_WATCHDOG ,PSERIES_WDTF_OP_STOP , pw->num); +if (rc !=H_SUCCESS && rc !=H_NOOP ) { +dev_crit (dev,"H_WATCHDOG: %ld: failed to stop timer %lu" , + rc, pw->num); +return -EIO ; + } +return 0 ;
Quoted 8 lines
drivers/watchdog/pseries-wdt.c line 163
Quoted 6 lines
drivers/watchdog/pseries-wdt.c new file +6
-0+} + +static struct watchdog_info pseries_wdt_info = { + .identity =DRV_NAME , + .options =WDIOF_KEEPALIVEPING |WDIOF_MAGICCLOSE |WDIOF_SETTIMEOUT + |WDIOF_PRETIMEOUT ,I don't know the watchdog code to know if those make sense.
It makes sense.
Quoted 9 lines
Quoted 9 lines
drivers/watchdog/pseries-wdt.c new file +9
-0+}; + +static const struct watchdog_ops pseries_wdt_ops = { + .owner =THIS_MODULE , + .start = pseries_wdt_start, + .stop = pseries_wdt_stop, +}; + +static int pseries_wdt_probe (struct platform_device *pdev)
Quoted 8 lines
drivers/watchdog/pseries-wdt.c line 178
Quoted 6 lines
drivers/watchdog/pseries-wdt.c new file +6
-0+{ +unsigned long ret[PLPAR_HCALL_BUFSIZE ] = {0 }; +unsigned long cap; +long rc; +struct pseries_wdt *pw; +int err;Try to use reverse xmas tree for new code please.
Is it not good practice to keep declarations of a particular type adjacent?
It feels... correct-ish to keep the longs together.
In this case there is no downside to doing "reverse xmas tree" because sizeof(long) is the same as sizeof(void *), but this looks odd to me:
unsigned long ret[PLPAR_HCALL_BUFSIZE] = { 0 };
struct pseries_wdt *pw;
unsigned long cap;
long rc;
int err;Quoted 21 lines
Quoted 21 lines
drivers/watchdog/pseries-wdt.c new file +21
-0+ rc =plpar_hcall (H_WATCHDOG , ret,PSERIES_WDTF_OP_QUERY ); +if (rc ==H_FUNCTION ) +return -ENODEV ; +if (rc !=H_SUCCESS ) +return -EIO ; + cap = ret[0 ]; + + pw =devm_kzalloc (&pdev->dev,sizeof (*pw),GFP_KERNEL ); +if (!pw) +return -ENOMEM ; + +/* +* Assume watchdogNumber 1 for now. If we ever support +* multiple timers we will need to devise a way to choose a +* distinct watchdogNumber for each platform device at device +* registration time. +*/ + pw->num =1 ; + +if (action >=ARRAY_SIZE (pseries_wdt_action)) +return -EINVAL ;
Quoted 8 lines
drivers/watchdog/pseries-wdt.c line 206
Quoted 6 lines
drivers/watchdog/pseries-wdt.c new file +6
-0+ pw->action = pseries_wdt_action[action]; + + pw->wd.parent = &pdev->dev; + pw->wd.info = &pseries_wdt_info; + pw->wd.ops = &pseries_wdt_ops; + pw->wd.min_timeout =DIV_ROUND_UP (PSERIES_WDTQ_MIN_TIMEOUT (cap),1000 );
MSEC_TO_SECagain?
MSEC_PER_SEC, sure.
Quoted 3 lines
drivers/watchdog/pseries-wdt.c line 207
Quoted 1 line
drivers/watchdog/pseries-wdt.c new file +1
-0+ pw->wd.max_timeout =UINT_MAX /1000 ;Where does that value come from?
It's just the maximum value the watchdog framework will accept. It's in one of the watchdog headers.
Quoted 11 lines
drivers/watchdog/pseries-wdt.c line 210
Quoted 3 lines
drivers/watchdog/pseries-wdt.c new file +3
-0+ pw->wd.timeout = timeout; +if (watchdog_init_timeout (&pw->wd,0 ,NULL )) +return -EINVAL ;It's late so maybe I'm misreading it, but does
watchdog_init_timeout()actually clamp the values if we don't pass a timeout?It looks like basically a nop when we pass timeout_param=0 and dev=NULL.
Which makes me think we aren't checking anywhere that the timeout we are using >= what firmware will accept.
No, watchdog_init_timeout() checks that
min_timeout <= timeout <= max_timeout
and returns an error if not. If somehow the minimum timeout exceeds the default 60 seconds we will catch it here and fail the probe.
Quoted 42 lines
Quoted 42 lines
drivers/watchdog/pseries-wdt.c new file +42
-0+watchdog_set_nowayout (&pw->wd, nowayout); +watchdog_stop_on_reboot (&pw->wd); +watchdog_stop_on_unregister (&pw->wd); +watchdog_set_drvdata (&pw->wd, pw); + + err =devm_watchdog_register_device (&pdev->dev, &pw->wd); +if (err) +return err; + +platform_set_drvdata (pdev, &pw->wd); + +return 0 ; +} + +static int pseries_wdt_suspend (struct platform_device *pdev,pm_message_t state) +{ +struct watchdog_device *wd =platform_get_drvdata (pdev); + +if (watchdog_active (wd)) +return pseries_wdt_stop (wd); +return 0 ; +} + +static int pseries_wdt_resume (struct platform_device *pdev) +{ +struct watchdog_device *wd =platform_get_drvdata (pdev); + +if (watchdog_active (wd)) +return pseries_wdt_start (wd); +return 0 ; +} + +static const struct platform_device_id pseries_wdt_id[] = { + { .name ="pseries-wdt" }, + {} +}; +MODULE_DEVICE_TABLE (platform, pseries_wdt_id); + +static struct platform_driver pseries_wdt_driver = { + .driver = { + .name =DRV_NAME , + .owner =THIS_MODULE ,
Quoted 8 lines
drivers/watchdog/pseries-wdt.c line 257
Quoted 5 lines
drivers/watchdog/pseries-wdt.c new file +5
-0+ }, + .id_table = pseries_wdt_id, + .probe = pseries_wdt_probe, + .resume = pseries_wdt_resume, + .suspend = pseries_wdt_suspend,I don't see any handling of the possible requirement to suspend timers across LPM. I don't think just wiring these up is enough?
I talked to Brian King about this and we decided that leaving the watchdog running across an LPM might lead to some potentially confusing behavior.
For example, if the watchdog expires while we're suspeded and the machine is hard reset the instant we come out of it on the other side.
Unless there is an ask by downstream software to actually leave the timer running over an LPM I think it is safest to err on the side of caution and unconditionally stop running timers before suspend.
Quoted 10 lines
drivers/watchdog/pseries-wdt.c line 262
Quoted 5 lines
drivers/watchdog/pseries-wdt.c new file +5
-0+}; +module_platform_driver (pseries_wdt_driver); + +MODULE_AUTHOR ("Alexey Kardashevskiy <[email protected]>" ); +MODULE_AUTHOR ("Scott Cheloha <[email protected]>" );I'd prefer the module authors were just the names, email addresses inevitably bitrot.
Your email address is in the change log.
Dropped.
Michael Ellerman
reviewerRe: [PATCH v2 4/4] watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
In reply to Scott ChelohaScott Cheloha <[email protected]> writes:
...
drivers/watchdog/pseries-wdt.c line 252
Quoted 5 lines
drivers/watchdog/pseries-wdt.c new file +5
-0+ +static struct platform_driver pseries_wdt_driver = { + .driver = { + .name =DRV_NAME , + .owner =THIS_MODULE ,
That owner assignment is not required.
It's set for you by platform_driver_register() via module_platform_driver().
cheers
Re: [PATCH v2 4/4] watchdog/pseries-wdt: initial support for H_WATCHDOG-based watchdog timers
Quoted 10 lines · Michael Ellerman
Scott Cheloha <[email protected]> writes:
...drivers/watchdog/pseries-wdt.c line 252
Quoted 5 lines
drivers/watchdog/pseries-wdt.c new file +5
-0+ +static struct platform_driver pseries_wdt_driver = { + .driver = { + .name =DRV_NAME , + .owner =THIS_MODULE ,That owner assignment is not required.
Quoted 2 lines
It's set for you by
platform_driver_register()viamodule_platform_driver().
Great, removed.
Hi,
I tried this series out with mainline QEMU built with Alexey's patch [1] and I wasn't able to get it to work. I'm using a simple QEMU command line booting a fedora36 guest in a Power9 boston host:
sudo ./qemu-system-ppc64 \
-M pseries,cap-cfpc=broken,cap-sbbc=broken,cap-ibs=broken,cap-ccf-assist=off,ic-mode=dual \
-m 4G -accel kvm -cpu POWER9 -smp 1,maxcpus=1,threads=1,cores=1,sockets=1 \
-device virtio-scsi-pci,id=scsi0,bus=pci.0,addr=0x2 \
-drive file=/home/danielhb/fedora36.qcow2,if=none,id=drive-scsi0-0-0-0,format=qcow2,cache=none \
-device scsi-hd,bus=scsi0.0,channel=0,scsi-id=0,lun=0,drive=drive-scsi0-0-0-0,id=scsi0-0-0-0,bootindex=2 \
-device qemu-xhci,id=usb,bus=pci.0,addr=0x4 -nographic -display none
Guest is running v5.19-rc2 with this series applied. Kernel config consists of 'pseries_le_defconfig' plus the following 'watchdog' related changes:
[root@fedora ~]# cat linux/.config | grep PSERIES_WDT CONFIG_PSERIES_WDT=y
[root@fedora ~]# cat linux/.config | grep -i watchdog CONFIG_PPC_WATCHDOG=y CONFIG_HAVE_NMI_WATCHDOG=yCONFIG_WATCHDOG=yCONFIG_WATCHDOG_CORE=y CONFIG_WATCHDOG_NOWAYOUT=y CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED=y CONFIG_WATCHDOG_OPEN_TIMEOUT=0
# CONFIG_WATCHDOG_SYSFS is not set
# CONFIG_WATCHDOG_HRTIMER_PRETIMEOUT is not set # Watchdog Pretimeout Governors
# CONFIG_WATCHDOG_PRETIMEOUT_GOV is not set
# Watchdog Device Drivers
# CONFIG_SOFT_WATCHDOG is not set
# CONFIG_XILINX_WATCHDOG is not set
# CONFIG_ZIIRAVE_WATCHDOG is not set
# CONFIG_CADENCE_WATCHDOG is not set
# CONFIG_DW_WATCHDOG is not set
# CONFIG_MAX63XX_WATCHDOG is not set CONFIG_WATCHDOG_RTAS=y
# PCI-based Watchdog Cards
# CONFIG_PCIPCWATCHDOG is not set
# USB-based Watchdog Cards
# CONFIG_USBPCWATCHDOG is not set
# CONFIG_WQ_WATCHDOG is not set
[root@fedora ~]#
Kernel command line:
[root@fedora ~]# cat /proc/cmdline BOOT_IMAGE=(ieee1275/disk,msdos2)/vmlinuz-5.19.0-rc2-00054-g12ede8ffb103 \
root=/dev/mapper/fedora_fedora-root ro rd.lvm.lv=fedora_fedora/root \
pseries-wdt.timeout=60 pseries-wdt.nowayout=1 pseries-wdt.action=2
With all that, executing
echo V > /dev/watchdog0
Does nothing. dmesg is clean and the guest doesn't reboot after the 60 sec timeout. I also tried with PSERIES_WDT being compiled as a module instead of built-in. Same results.
What am I missing?
[1] https://patchwork.ozlabs.org/project/qemu-ppc/patch/[email protected]/
Thanks,
Daniel
Quoted 35 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. This patch series adds support for this hypercall to powerpc/pseries kernels and introduces a new watchdog driver, "pseries-wdt", for the virtual timers exposed by the hypercall.This series is preceded by the following:
RFC v1: https://lore.kernel.org/linux-watchdog/[email protected]/
RFC v2: https://lore.kernel.org/linux-watchdog/[email protected]/
PATCH v1: https://lore.kernel.org/linux-watchdog/[email protected]/Changes of note from PATCH v1:
- Trim down the large comment documenting the
H_WATCHDOGhypercall. The comment is likely to rot, so remove anything we aren't using and anything overly obvious.- Remove any preprocessor definitions not actually used in the module right now. If we want to use other features offered by the hypercall we can add them in later. They're just clutter until then.
- Simplify the "action" module parameter. The value is now an index into an array of possible timeoutAction values. This design removes the need for the custom get/set methods used in PATCH v1.
Now we merely need to check that the "action" value is a valid index during
pseries_wdt_probe(). Easy.- Make the timeoutAction a member of
pseries_wdt, "action". This eliminates the use of a global variable duringpseries_wdt_start().- Use
watchdog_init_timeout()idiomatically. Check its return value and error out ofpseries_wdt_probe()if it fails.
Quoted 5 lines · Daniel Henrique Barboza
Hi,
I tried this series out with mainline QEMU built with Alexey's patch [1] and I wasn't able to get it to work. I'm using a simple QEMU command line booting a fedora36 guest in a Power9 boston host:
I would assume the H_WATCHDOG hypercall is not implemented by the qemu pseries machine type. It is a very new hypercall.
-Tyrel
Quoted 127 lines
sudo ./qemu-system-ppc64 \
-M pseries,cap-cfpc=broken,cap-sbbc=broken,cap-ibs=broken,cap-ccf-assist=off,ic-mode=dual \(quoted diff) +0
-3-m 4G -accel kvm -cpu POWER9 -smp 1,maxcpus=1,threads=1,cores=1,sockets=1 \ -device virtio-scsi-pci,id=scsi0,bus=pci.0,addr=0x2 \ -drivefile=/home/danielhb/fedora36.qcow2,if=none,id=drive-scsi0-0-0-0,format=qcow2,cache=none \
-device scsi-hd,bus=scsi0.0,channel=0,scsi-id=0,lun=0,drive=drive-scsi0-0-0-0,id=scsi0-0-0-0,bootindex=2 \
-device qemu-xhci,id=usb,bus=pci.0,addr=0x4 -nographic -display noneGuest is running v5.19-rc2 with this series applied. Kernel config consists of '
pseries_le_defconfig' plus the following 'watchdog' related changes:[root@fedora ~]# cat linux/.config | grep
PSERIES_WDTCONFIG_PSERIES_WDT=y[root@fedora ~]# cat linux/.config | grep -i watchdog
CONFIG_PPC_WATCHDOG=yCONFIG_HAVE_NMI_WATCHDOG=yCONFIG_WATCHDOG=yCONFIG_WATCHDOG_CORE=yCONFIG_WATCHDOG_NOWAYOUT=yCONFIG_WATCHDOG_HANDLE_BOOT_ENABLED=yCONFIG_WATCHDOG_OPEN_TIMEOUT=0
#CONFIG_WATCHDOG_SYSFSis not set
#CONFIG_WATCHDOG_HRTIMER_PRETIMEOUTis not set # Watchdog Pretimeout Governors
#CONFIG_WATCHDOG_PRETIMEOUT_GOVis not set
# Watchdog Device Drivers
#CONFIG_SOFT_WATCHDOGis not set
#CONFIG_XILINX_WATCHDOGis not set
#CONFIG_ZIIRAVE_WATCHDOGis not set
#CONFIG_CADENCE_WATCHDOGis not set
#CONFIG_DW_WATCHDOGis not set
#CONFIG_MAX63XX_WATCHDOGis not setCONFIG_WATCHDOG_RTAS=y
# PCI-based Watchdog Cards
#CONFIG_PCIPCWATCHDOGis not set
# USB-based Watchdog Cards
#CONFIG_USBPCWATCHDOGis not set
#CONFIG_WQ_WATCHDOGis not set
[root@fedora ~]#Kernel command line:
[root@fedora ~]# cat
/proc/cmdlineBOOT_IMAGE=(ieee1275/disk,msdos2)/vmlinuz-5.19.0-rc2-00054-g12ede8ffb103 \
root=/dev/mapper/fedora_fedora-root ro rd.lvm.lv=fedora_fedora/root \
pseries-wdt.timeout=60 pseries-wdt.nowayout=1 pseries-wdt.action=2With all that, executing
echo V >
/dev/watchdog0Does nothing. dmesg is clean and the guest doesn't reboot after the 60 sec timeout. I also tried with
PSERIES_WDTbeing compiled as a module instead of built-in. Same results.What am I missing?
[1]
https://patchwork.ozlabs.org/project/qemu-ppc/patch/[email protected]/Thanks,
Daniel
Quoted 41 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall, H_WATCHDOG. This patch series adds support for this hypercall to powerpc/pseries kernels and introduces a new watchdog driver, "pseries-wdt", for the virtual timers exposed by the hypercall.
This series is preceded by the following:
RFC v1:
https://lore.kernel.org/linux-watchdog/[email protected]/RFC v2:
https://lore.kernel.org/linux-watchdog/[email protected]/PATCH v1:
https://lore.kernel.org/linux-watchdog/[email protected]/Changes of note from PATCH v1:
- Trim down the large comment documenting the
H_WATCHDOGhypercall. The comment is likely to rot, so remove anything we aren't using and anything overly obvious.- Remove any preprocessor definitions not actually used in the module right now. If we want to use other features offered by the hypercall we can add them in later. They're just clutter until then.
- Simplify the "action" module parameter. The value is now an index into an array of possible timeoutAction values. This design removes the need for the custom get/set methods used in PATCH v1.
Now we merely need to check that the "action" value is a valid index during pseries_wdt_probe(). Easy.
- Make the timeoutAction a member of
pseries_wdt, "action". This eliminates the use of a global variable duringpseries_wdt_start().- Use
watchdog_init_timeout()idiomatically. Check its return value and error out ofpseries_wdt_probe()if it fails.
Quoted 9 lines · Tyrel Datwyler
Quoted 5 lines · Daniel Henrique Barboza
Hi,
I tried this series out with mainline QEMU built with Alexey's patch [1] and I wasn't able to get it to work. I'm using a simple QEMU command line booting a fedora36 guest in a Power9 boston host:
I would assume the
H_WATCHDOGhypercall is not implemented by the qemu pseries machine type. It is a very new hypercall.
Alexey implemented QEMU support for this hypercall here:
"[qemu] ppc/spapr: Implement H_WATCHDOG"
https://patchwork.ozlabs.org/project/qemu-ppc/patch/[email protected]/
It is under review in the QEMU mailing list. I tried it out with this series and wasn't able to get it to work.
Thanks,
Daniel
Quoted 131 lines
-Tyrel
Quoted 127 lines
sudo ./qemu-system-ppc64 \
-M pseries,cap-cfpc=broken,cap-sbbc=broken,cap-ibs=broken,cap-ccf-assist=off,ic-mode=dual \(quoted diff) +0
-3-m 4G -accel kvm -cpu POWER9 -smp 1,maxcpus=1,threads=1,cores=1,sockets=1 \ -device virtio-scsi-pci,id=scsi0,bus=pci.0,addr=0x2 \ -drivefile=/home/danielhb/fedora36.qcow2,if=none,id=drive-scsi0-0-0-0,format=qcow2,cache=none \
-device scsi-hd,bus=scsi0.0,channel=0,scsi-id=0,lun=0,drive=drive-scsi0-0-0-0,id=scsi0-0-0-0,bootindex=2 \
-device qemu-xhci,id=usb,bus=pci.0,addr=0x4 -nographic -display noneGuest is running v5.19-rc2 with this series applied. Kernel config consists of '
pseries_le_defconfig' plus the following 'watchdog' related changes:[root@fedora ~]# cat linux/.config | grep
PSERIES_WDTCONFIG_PSERIES_WDT=y[root@fedora ~]# cat linux/.config | grep -i watchdog
CONFIG_PPC_WATCHDOG=yCONFIG_HAVE_NMI_WATCHDOG=yCONFIG_WATCHDOG=yCONFIG_WATCHDOG_CORE=yCONFIG_WATCHDOG_NOWAYOUT=yCONFIG_WATCHDOG_HANDLE_BOOT_ENABLED=yCONFIG_WATCHDOG_OPEN_TIMEOUT=0
#CONFIG_WATCHDOG_SYSFSis not set
#CONFIG_WATCHDOG_HRTIMER_PRETIMEOUTis not set # Watchdog Pretimeout Governors
#CONFIG_WATCHDOG_PRETIMEOUT_GOVis not set
# Watchdog Device Drivers
#CONFIG_SOFT_WATCHDOGis not set
#CONFIG_XILINX_WATCHDOGis not set
#CONFIG_ZIIRAVE_WATCHDOGis not set
#CONFIG_CADENCE_WATCHDOGis not set
#CONFIG_DW_WATCHDOGis not set
#CONFIG_MAX63XX_WATCHDOGis not setCONFIG_WATCHDOG_RTAS=y
# PCI-based Watchdog Cards
#CONFIG_PCIPCWATCHDOGis not set
# USB-based Watchdog Cards
#CONFIG_USBPCWATCHDOGis not set
#CONFIG_WQ_WATCHDOGis not set
[root@fedora ~]#Kernel command line:
[root@fedora ~]# cat
/proc/cmdlineBOOT_IMAGE=(ieee1275/disk,msdos2)/vmlinuz-5.19.0-rc2-00054-g12ede8ffb103 \
root=/dev/mapper/fedora_fedora-root ro rd.lvm.lv=fedora_fedora/root \
pseries-wdt.timeout=60 pseries-wdt.nowayout=1 pseries-wdt.action=2With all that, executing
echo V >
/dev/watchdog0Does nothing. dmesg is clean and the guest doesn't reboot after the 60 sec timeout. I also tried with
PSERIES_WDTbeing compiled as a module instead of built-in. Same results.What am I missing?
[1]
https://patchwork.ozlabs.org/project/qemu-ppc/patch/[email protected]/Thanks,
Daniel
Quoted 41 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall, H_WATCHDOG. This patch series adds support for this hypercall to powerpc/pseries kernels and introduces a new watchdog driver, "pseries-wdt", for the virtual timers exposed by the hypercall.
This series is preceded by the following:
RFC v1:
https://lore.kernel.org/linux-watchdog/[email protected]/RFC v2:
https://lore.kernel.org/linux-watchdog/[email protected]/PATCH v1:
https://lore.kernel.org/linux-watchdog/[email protected]/Changes of note from PATCH v1:
- Trim down the large comment documenting the
H_WATCHDOGhypercall. The comment is likely to rot, so remove anything we aren't using and anything overly obvious.- Remove any preprocessor definitions not actually used in the module right now. If we want to use other features offered by the hypercall we can add them in later. They're just clutter until then.
- Simplify the "action" module parameter. The value is now an index into an array of possible timeoutAction values. This design removes the need for the custom get/set methods used in PATCH v1.
Now we merely need to check that the "action" value is a valid index during pseries_wdt_probe(). Easy.
- Make the timeoutAction a member of
pseries_wdt, "action". This eliminates the use of a global variable duringpseries_wdt_start().- Use
watchdog_init_timeout()idiomatically. Check its return value and error out ofpseries_wdt_probe()if it fails.
Hi,
Update: checking out 'dmesg' more carefully I found out that the module probe is failing with the following message:
[ 186.298424][ T811] pseries-wdt: probe of pseries-wdt.0 failed with error -5
This fail is consistent. If I remove the module and modprobe it again the same error happens.
The message is being throw by pseries_wdt_probe() (patch 4/4). Back in QEMU, in Alexey's H_WATCHDOG implementation [1], I see that h_watchdog is returning H_PARAMETER because the retrieved 'watchdogNumber' is zero.
Also, the pseries_wdt module still appears in 'lsmod' despite this probe error. Not sure if this is a bug:
[root@fedora ~]# rmmod pseries_wdt
[root@fedora ~]# modprobe pseries_wdt
[ 1792.846769][ T865] pseries-wdt: probe of pseries-wdt.0 failed with error -5
[root@fedora ~]# lsmod | grep pseriespseries_wdt 262144 0
[root@fedora ~]#
For reference this is all the output of 'lsmod' in the guest:
[root@fedora ~]# lsmod
Module Size Used by pseries_wdt 262144 0 nfnetlink 262144 1 evdev 327680 1 input_leds 262144 0 led_class 262144 1 input_leds fuse 458752 1 xfs 1835008 2 libcrc32c 262144 1 xfs virtio_scsi 327680 2 virtio_pci 262144 0 virtio 327680 2 virtio_scsi,virtio_pci vmx_crypto 262144 0 gf128mul 262144 1 vmx_crypto crc32c_vpmsum 327680 1 virtio_ring 327680 3 virtio_scsi,virtio_pci,virtio virtio_pci_legacy_dev 262144 1 virtio_pci virtio_pci_modern_dev 262144 1 virtio_pci
autofs4 327680 2
Given that the error is being thrown by Alexey's QEMU code, I'll bring it up in the QEMU mailing list in [1] and follow it up from there.
[1] https://patchwork.ozlabs.org/project/qemu-ppc/patch/[email protected]/
Thanks,
Daniel
Quoted 118 lines · Daniel Henrique Barboza
Hi,
I tried this series out with mainline QEMU built with Alexey's patch [1] and I wasn't able to get it to work. I'm using a simple QEMU command line booting a fedora36 guest in a Power9 boston host:
sudo ./qemu-system-ppc64 \
(quoted diff) +0
-6-M pseries,cap-cfpc=broken,cap-sbbc=broken,cap-ibs=broken,cap-ccf-assist=off,ic-mode=dual \ -m 4G -accel kvm -cpu POWER9 -smp 1,maxcpus=1,threads=1,cores=1,sockets=1 \ -device virtio-scsi-pci,id=scsi0,bus=pci.0,addr=0x2 \ -drive file=/home/danielhb/fedora36.qcow2,if=none,id=drive-scsi0-0-0-0,format=qcow2,cache=none \ -device scsi-hd,bus=scsi0.0,channel=0,scsi-id=0,lun=0,drive=drive-scsi0-0-0-0,id=scsi0-0-0-0,bootindex=2 \ -device qemu-xhci,id=usb,bus=pci.0,addr=0x4 -nographic -display noneGuest is running v5.19-rc2 with this series applied. Kernel config consists of '
pseries_le_defconfig' plus the following 'watchdog' related changes:[root@fedora ~]# cat linux/.config | grep
PSERIES_WDTCONFIG_PSERIES_WDT=y[root@fedora ~]# cat linux/.config | grep -i watchdog
CONFIG_PPC_WATCHDOG=yCONFIG_HAVE_NMI_WATCHDOG=yCONFIG_WATCHDOG=yCONFIG_WATCHDOG_CORE=yCONFIG_WATCHDOG_NOWAYOUT=yCONFIG_WATCHDOG_HANDLE_BOOT_ENABLED=yCONFIG_WATCHDOG_OPEN_TIMEOUT=0
#CONFIG_WATCHDOG_SYSFSis not set
#CONFIG_WATCHDOG_HRTIMER_PRETIMEOUTis not set # Watchdog Pretimeout Governors
#CONFIG_WATCHDOG_PRETIMEOUT_GOVis not set
# Watchdog Device Drivers
#CONFIG_SOFT_WATCHDOGis not set
#CONFIG_XILINX_WATCHDOGis not set
#CONFIG_ZIIRAVE_WATCHDOGis not set
#CONFIG_CADENCE_WATCHDOGis not set
#CONFIG_DW_WATCHDOGis not set
#CONFIG_MAX63XX_WATCHDOGis not setCONFIG_WATCHDOG_RTAS=y
# PCI-based Watchdog Cards
#CONFIG_PCIPCWATCHDOGis not set
# USB-based Watchdog Cards
#CONFIG_USBPCWATCHDOGis not set
#CONFIG_WQ_WATCHDOGis not set
[root@fedora ~]#Kernel command line:
[root@fedora ~]# cat
/proc/cmdlineBOOT_IMAGE=(ieee1275/disk,msdos2)/vmlinuz-5.19.0-rc2-00054-g12ede8ffb103 \
root=/dev/mapper/fedora_fedora-root ro rd.lvm.lv=fedora_fedora/root \
pseries-wdt.timeout=60 pseries-wdt.nowayout=1 pseries-wdt.action=2With all that, executing
echo V >
/dev/watchdog0Does nothing. dmesg is clean and the guest doesn't reboot after the 60 sec timeout. I also tried with
PSERIES_WDTbeing compiled as a module instead of built-in. Same results.What am I missing?
[1] https://patchwork.ozlabs.org/project/qemu-ppc/patch/[email protected]/
Thanks,
Daniel
Quoted 35 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall, H_WATCHDOG. This patch series adds support for this hypercall to powerpc/pseries kernels and introduces a new watchdog driver, "pseries-wdt", for the virtual timers exposed by the hypercall.
This series is preceded by the following:
RFC v1: https://lore.kernel.org/linux-watchdog/[email protected]/
RFC v2: https://lore.kernel.org/linux-watchdog/[email protected]/
PATCH v1: https://lore.kernel.org/linux-watchdog/[email protected]/Changes of note from PATCH v1:
- Trim down the large comment documenting the
H_WATCHDOGhypercall. The comment is likely to rot, so remove anything we aren't using and anything overly obvious.- Remove any preprocessor definitions not actually used in the module right now. If we want to use other features offered by the hypercall we can add them in later. They're just clutter until then.
- Simplify the "action" module parameter. The value is now an index into an array of possible timeoutAction values. This design removes the need for the custom get/set methods used in PATCH v1.
Now we merely need to check that the "action" value is a valid index during pseries_wdt_probe(). Easy.
- Make the timeoutAction a member of
pseries_wdt, "action". This eliminates the use of a global variable duringpseries_wdt_start().- Use
watchdog_init_timeout()idiomatically. Check its return value and error out ofpseries_wdt_probe()if it fails.
Quoted 10 lines · Scott Cheloha
PAPR v2.12 defines a new hypercall,
H_WATCHDOG. This patch series adds support for this hypercall to powerpc/pseries kernels and introduces a new watchdog driver, "pseries-wdt", for the virtual timers exposed by the hypercall.This series is preceded by the following:
RFC v1: https://lore.kernel.org/linux-watchdog/[email protected]/
RFC v2: https://lore.kernel.org/linux-watchdog/[email protected]/
PATCH v1: https://lore.kernel.org/linux-watchdog/[email protected]/
Tested successfully with mainline QEMU plus Alexey's new h_watchdog patches in
https://patchwork.ozlabs.org/project/qemu-ppc/list/?series=305226.
All patches of this series:
Thanks,
Daniel
Quoted 25 lines
Changes of note from PATCH v1:
- Trim down the large comment documenting the
H_WATCHDOGhypercall. The comment is likely to rot, so remove anything we aren't using and anything overly obvious.- Remove any preprocessor definitions not actually used in the module right now. If we want to use other features offered by the hypercall we can add them in later. They're just clutter until then.
- Simplify the "action" module parameter. The value is now an index into an array of possible timeoutAction values. This design removes the need for the custom get/set methods used in PATCH v1.
Now we merely need to check that the "action" value is a valid index during
pseries_wdt_probe(). Easy.- Make the timeoutAction a member of
pseries_wdt, "action". This eliminates the use of a global variable duringpseries_wdt_start().- Use
watchdog_init_timeout()idiomatically. Check its return value and error out ofpseries_wdt_probe()if it fails.