| Message ID | 20161122092258.32514-1-andrew.donnellan@au1.ibm.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
Andrew Donnellan <andrew.donnellan@au1.ibm.com> writes: > skiboot now supports "fast reboot", a reboot procedure where skiboot > reinitialises hardware and loads a new kernel without re-IPLing the > machine. At present, fast reboot support is still experimental and is not > enabled by default, however it is intended that it will be enabled by > default in a near-future release. > > There may be some circumstances where the user wants to force a full IPL > reboot rather than using fast reboot. Add support for the > OPAL_REBOOT_FULL_IPL reboot type, enabled by writing 1 to > /sys/firmware/opal/force_full_ipl_reboot. On versions of skiboot that > implement the OPAL_REBOOT_FULL_IPL reboot type, this will force an IPL. On > versions that do not, print an error message on reboot and proceed with a > regular reboot (which could be a full IPL or a fast reboot). Hi Andrew, Can I suggest: - Fix OPAL - Fix OPAL - Fix OPAL ... Oh OK, it's firmware :D There's existing logic in kernel/reboot.c to handle a reboot= command line parameter, which can set the reboot_mode, so my preference would be that we use that. Currently we completely ignore the reboot_mode, so there's no backward compatibility issue. It looks like the default is REBOOT_COLD, whatever that means. So I think we could define that on powernv REBOOT_HARD means "do a full IPL". Or we could use "force", which already exists, though I can't quite see what that actually means on x86. cheers
On 22/11/16 21:27, Michael Ellerman wrote: > Can I suggest: > - Fix OPAL > - Fix OPAL > - Fix OPAL > > ... > > Oh OK, it's firmware :D Yep :D > There's existing logic in kernel/reboot.c to handle a reboot= command > line parameter, which can set the reboot_mode, so my preference would be > that we use that. > > Currently we completely ignore the reboot_mode, so there's no backward > compatibility issue. > > It looks like the default is REBOOT_COLD, whatever that means. So I > think we could define that on powernv REBOOT_HARD means "do a full IPL". Sounds good. > Or we could use "force", which already exists, though I can't quite see > what that actually means on x86. Hmm.
On 23/11/16 12:37, Andrew Donnellan wrote: >> There's existing logic in kernel/reboot.c to handle a reboot= command >> line parameter, which can set the reboot_mode, so my preference would be >> that we use that. >> >> Currently we completely ignore the reboot_mode, so there's no backward >> compatibility issue. >> >> It looks like the default is REBOOT_COLD, whatever that means. So I >> think we could define that on powernv REBOOT_HARD means "do a full IPL". > > Sounds good. Thinking about this more - you can't change that at runtime. :( > >> Or we could use "force", which already exists, though I can't quite see >> what that actually means on x86. > > Hmm. It looks like that on x86, if reboot_force = 1, we don't call machine_shutdown() and go straight to __machine_emergency_restart()?
Andrew Donnellan <andrew.donnellan@au1.ibm.com> writes: > On 23/11/16 12:37, Andrew Donnellan wrote: >>> There's existing logic in kernel/reboot.c to handle a reboot= command >>> line parameter, which can set the reboot_mode, so my preference would be >>> that we use that. >>> >>> Currently we completely ignore the reboot_mode, so there's no backward >>> compatibility issue. >>> >>> It looks like the default is REBOOT_COLD, whatever that means. So I >>> think we could define that on powernv REBOOT_HARD means "do a full IPL". >> >> Sounds good. > > Thinking about this more - you can't change that at runtime. :( Yeah. Is that a problem? From my POV this is basically a "my firmware is broken" workaround, so you're going to want to set it permanently until you update your firmware. >>> Or we could use "force", which already exists, though I can't quite see >>> what that actually means on x86. >> >> Hmm. > > It looks like that on x86, if reboot_force = 1, we don't call > machine_shutdown() and go straight to __machine_emergency_restart()? I think so. But I don't know what that actually means from a user's point of view. cheers
Michael Ellerman <mpe@ellerman.id.au> writes: > Andrew Donnellan <andrew.donnellan@au1.ibm.com> writes: > >> On 23/11/16 12:37, Andrew Donnellan wrote: >>>> There's existing logic in kernel/reboot.c to handle a reboot= command >>>> line parameter, which can set the reboot_mode, so my preference would be >>>> that we use that. >>>> >>>> Currently we completely ignore the reboot_mode, so there's no backward >>>> compatibility issue. >>>> >>>> It looks like the default is REBOOT_COLD, whatever that means. So I >>>> think we could define that on powernv REBOOT_HARD means "do a full IPL". >>> >>> Sounds good. >> >> Thinking about this more - you can't change that at runtime. :( > > Yeah. Is that a problem? From my POV this is basically a "my firmware is > broken" workaround, so you're going to want to set it permanently until > you update your firmware. The only real world application I can see wanting to do that is perhaps HTX - where it chooses what kind of bootme run it wants. So, well, for certain definitions of "real world".
Andrew Donnellan <andrew.donnellan@au1.ibm.com> writes: > skiboot now supports "fast reboot", a reboot procedure where skiboot > reinitialises hardware and loads a new kernel without re-IPLing the > machine. At present, fast reboot support is still experimental and is not > enabled by default, however it is intended that it will be enabled by > default in a near-future release. > > There may be some circumstances where the user wants to force a full IPL > reboot rather than using fast reboot. Add support for the > OPAL_REBOOT_FULL_IPL reboot type, enabled by writing 1 to > /sys/firmware/opal/force_full_ipl_reboot. On versions of skiboot that > implement the OPAL_REBOOT_FULL_IPL reboot type, this will force an IPL. On > versions that do not, print an error message on reboot and proceed with a > regular reboot (which could be a full IPL or a fast reboot). > > Cc: Stewart Smith <stewart@linux.vnet.ibm.com> > Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> > Signed-off-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> > > --- > > Corresponding skiboot patch: http://patchwork.ozlabs.org/patch/697601/ FWIW I've just merged the skiboot patch.
diff --git a/arch/powerpc/include/asm/opal-api.h b/arch/powerpc/include/asm/opal-api.h index 0e2e57b..4de1f63 100644 --- a/arch/powerpc/include/asm/opal-api.h +++ b/arch/powerpc/include/asm/opal-api.h @@ -918,6 +918,7 @@ enum OpalSysCooling { enum { OPAL_REBOOT_NORMAL = 0, OPAL_REBOOT_PLATFORM_ERROR = 1, + OPAL_REBOOT_FULL_IPL = 2, }; /* Argument to OPAL_PCI_TCE_KILL */ diff --git a/arch/powerpc/include/asm/opal.h b/arch/powerpc/include/asm/opal.h index e958b70..a5e4065 100644 --- a/arch/powerpc/include/asm/opal.h +++ b/arch/powerpc/include/asm/opal.h @@ -270,6 +270,7 @@ extern void opal_platform_dump_init(void); extern void opal_sys_param_init(void); extern void opal_msglog_init(void); extern void opal_msglog_sysfs_init(void); +extern void opal_reboot_sysfs_init(void); extern int opal_async_comp_init(void); extern int opal_sensor_init(void); extern int opal_hmi_handler_init(void); diff --git a/arch/powerpc/platforms/powernv/opal.c b/arch/powerpc/platforms/powernv/opal.c index 6c9a65b..c619d8d 100644 --- a/arch/powerpc/platforms/powernv/opal.c +++ b/arch/powerpc/platforms/powernv/opal.c @@ -750,6 +750,8 @@ static int __init opal_init(void) opal_sys_param_init(); /* Setup message log sysfs interface. */ opal_msglog_sysfs_init(); + /* Setup reboot sysfs interface. */ + opal_reboot_sysfs_init(); } /* Initialize platform devices: IPMI backend, PRD & flash interface */ diff --git a/arch/powerpc/platforms/powernv/setup.c b/arch/powerpc/platforms/powernv/setup.c index efe8b6b..d2460fc 100644 --- a/arch/powerpc/platforms/powernv/setup.c +++ b/arch/powerpc/platforms/powernv/setup.c @@ -28,6 +28,7 @@ #include <linux/bug.h> #include <linux/pci.h> #include <linux/cpufreq.h> +#include <linux/sysfs.h> #include <asm/machdep.h> #include <asm/firmware.h> @@ -118,6 +119,53 @@ static void pnv_prepare_going_down(void) opal_flash_term_callback(); } +static bool force_full_ipl_reboot = false; + +static ssize_t force_full_ipl_reboot_show(struct kobject *k, + struct kobj_attribute *attr, + char *buf) +{ + return sprintf(buf, "%d\n", (int) force_full_ipl_reboot); +} + +static ssize_t force_full_ipl_reboot_store(struct kobject *k, + struct kobj_attribute *attr, + const char *buf, + size_t count) +{ + if (count != 2) /* including trailing NUL */ + return -EINVAL; + + switch (buf[0]) { + case '0': + force_full_ipl_reboot = false; + return count; + case '1': + force_full_ipl_reboot = true; + return count; + default: + return -EINVAL; + } +} + +static struct kobj_attribute opal_force_full_ipl_reboot_attr = + __ATTR(force_full_ipl_reboot, 0644, force_full_ipl_reboot_show, + force_full_ipl_reboot_store); + +void opal_reboot_sysfs_init(void) { + int rc = 0; + + if (!opal_kobj) { + pr_warn("setup: opal kobject is not available"); + return; + } + + rc = sysfs_create_file(opal_kobj, &opal_force_full_ipl_reboot_attr.attr); + if (rc) + pr_err("setup: unable to create sysfs file " + "force_full_ipl_reboot (%d)", rc); +} + static void __noreturn pnv_restart(char *cmd) { long rc = OPAL_BUSY; @@ -125,7 +173,20 @@ static void __noreturn pnv_restart(char *cmd) pnv_prepare_going_down(); while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) { - rc = opal_cec_reboot(); + if (force_full_ipl_reboot && opal_check_token(OPAL_CEC_REBOOT2)) { + rc = opal_cec_reboot2(OPAL_REBOOT_FULL_IPL, + "Full IPL reboot requested"); + + if (rc == OPAL_UNSUPPORTED) { + pr_err("Firmware doesn't support forcing full " + "IPL reboot"); + force_full_ipl_reboot = false; + rc = opal_cec_reboot(); + } + } else { + rc = opal_cec_reboot(); + } + if (rc == OPAL_BUSY_EVENT) opal_poll_events(NULL); else
skiboot now supports "fast reboot", a reboot procedure where skiboot reinitialises hardware and loads a new kernel without re-IPLing the machine. At present, fast reboot support is still experimental and is not enabled by default, however it is intended that it will be enabled by default in a near-future release. There may be some circumstances where the user wants to force a full IPL reboot rather than using fast reboot. Add support for the OPAL_REBOOT_FULL_IPL reboot type, enabled by writing 1 to /sys/firmware/opal/force_full_ipl_reboot. On versions of skiboot that implement the OPAL_REBOOT_FULL_IPL reboot type, this will force an IPL. On versions that do not, print an error message on reboot and proceed with a regular reboot (which could be a full IPL or a fast reboot). Cc: Stewart Smith <stewart@linux.vnet.ibm.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> --- Corresponding skiboot patch: http://patchwork.ozlabs.org/patch/697601/ This seemed like a good idea at the time. A sysfs-based flag was the best option that came to mind; other suggestions are welcome. Is "full IPL" a reasonable term to use or should I try to think of something a bit less IBM-ese? --- arch/powerpc/include/asm/opal-api.h | 1 + arch/powerpc/include/asm/opal.h | 1 + arch/powerpc/platforms/powernv/opal.c | 2 ++ arch/powerpc/platforms/powernv/setup.c | 63 +++++++++++++++++++++++++++++++++- 4 files changed, 66 insertions(+), 1 deletion(-)