Cure kexec() vs. mwait_play_dead() troubles
cover Cure kexec() vs. mwait_play_dead() troubles
Hi!
Ashok observed triple faults when executing kexec() on a kernel which has 'nosmt' on the kernel commandline and HT enabled in the BIOS.
'nosmt' brings up the HT siblings to the point where they initiliazed the CPU and then rolls the bringup back which parks them in mwait_play_dead(). The reason is that all CPUs should have CR4.MCE set. Otherwise a broadcast MCE will immediately shut down the machine.
Some detective work revealed that:
1) The kexec kernel can overwrite text, pagetables, stack and data of the previous kernel.
2) If the kexec kernel writes to the memory which is monitored by an "offline" CPU, that CPU resumes execution. That's obviously doomed when the kexec kernel overwrote text, pagetables, data or stack.
While on my test machine the first kexec() after reset always "worked", the second one reliably ended up in a triple fault.
The following series cures this by:
1) Bringing offline CPUs which are stuck in mwait_play_dead() out of mwait by writing to the monitored cacheline
2) Let the woken up CPUs check the written control word and drop into a HLT loop if the control word requests so.
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably, but there is at least one which prevents the NMI and SMI cause: INIT.
3) If the system uses the regular INIT/STARTUP sequence to wake up secondary CPUS, then "park" all CPUs including the "offline" ones by sending them INIT IPIs.
The INIT IPI brings the CPU into a wait for wakeup state which is not affected by NMI and SMI, but INIT also clears CR4.MCE, so the broadcast MCE problem comes back.
But that's not really any different from a CPU sitting in the HLT loop on the previous kernel. If a broadcast MCE arrives, HLT resumes execution and the CPU tries to handle the MCE on overwritten text, pagetables etc.
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
The series is also available from git:
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git x86/kexec
Thanks,
tglx
--- include/asm/smp.h | 4 + kernel/smp.c | 62 +++++++++++++--------- kernel/smpboot.c | 151 ++++++++++++++++++++++++++++++++++++++++-------------- 3 files changed, 156 insertions(+), 61 deletions(-)
PATCH 1/6 x86/smp: Remove pointless wmb() from native_stop_other_cpus()
The wmb() after the successfull atomic_cmpxchg() is complete voodoo along with the comment stating "sync above data before sending IRQ".
There is no "above" data except for the atomic_t stopping_cpu which has just been acquired. The reboot IPI handler does not check any data and unconditionally disables the CPU.
Remove this cargo cult barrier.
--- arch/x86/kernel/smp.c | 3 --- 1 file changed, 3 deletions(-)
arch/x86/kernel/smp.c +0 -3
@@ -171,9 +171,6 @@ static void native_stop_other_cpus(int wif (atomic_cmpxchg (&stopping_cpu,-1 ,safe_smp_processor_id ()) !=-1 )return ; -/* sync above data before sending IRQ */ -wmb (); -apic_send_IPI_allbutself (REBOOT_VECTOR );/*
PATCH 2/6 x86/smp: Acquire stopping_cpu unconditionally
In reply to Thomas GleixnerThere is no reason to acquire the stopping_cpu atomic_t only when there is more than one online CPU.
Make it unconditional to prepare for fixing the kexec() problem when there are present but "offline" CPUs which play dead in mwait_play_dead().
They need to be brought out of mwait before kexec() as kexec() can overwrite text, pagetables, stacks and the monitored cacheline of the original kernel. The latter causes mwait to resume execution which obviously causes havoc on the kexec kernel which results usually in triple faults.
Move the acquire out of the num_online_cpus() > 1 condition so the upcoming 'kick mwait' fixup is properly protected.
--- arch/x86/kernel/smp.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
arch/x86/kernel/smp.c +4 -4
@@ -152,6 +152,10 @@ static void native_stop_other_cpus(int wif (reboot_force)return ; +/* Only proceed if this is the first CPU to reach this code */ +if (atomic_cmpxchg (&stopping_cpu,-1 ,safe_smp_processor_id ()) !=-1 ) +return ; +/* * Use an own vector here because smp_call_function * does lots of things not suitable in a panic situation.
@@ -167,10 +171,6 @@ static void native_stop_other_cpus(int w* finish their work before we force them off with the NMI. */ if (num_online_cpus () >1 ) { -/* did someone beat us here? */ -if (atomic_cmpxchg (&stopping_cpu,-1 ,safe_smp_processor_id ()) !=-1 ) -return ; -apic_send_IPI_allbutself (REBOOT_VECTOR );/*
PATCH 3/6 x86/smp: Use dedicated cache-line for mwait_play_dead()
In reply to Thomas GleixnerMonitoring idletask::thread_info::flags in mwait_play_dead() has been an obvious choice as all what is needed is a cache line which is not written by other CPUs.
But there is a use case where a "dead" CPU needs to be brought out of that mwait(): kexec().
The CPU needs to be brought out of mwait before kexec() as kexec() can overwrite text, pagetables, stacks and the monitored cacheline of the original kernel. The latter causes mwait to resume execution which obviously causes havoc on the kexec kernel which results usually in triple faults.
Use a dedicated per CPU storage to prepare for that.
--- arch/x86/kernel/smpboot.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-)
arch/x86/kernel/smpboot.c +14 -10
@@ -101,6 +101,17 @@ EXPORT_PER_CPU_SYMBOL(cpu_die_map); DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);EXPORT_PER_CPU_SYMBOL (cpu_info); +struct mwait_cpu_dead { +unsigned int control; +unsigned int status; +}; + +/* +* Cache line aligned data for mwait_play_dead(). Separate on purpose so +* that it's unlikely to be touched by other CPUs. +*/ +static DEFINE_PER_CPU_ALIGNED(struct mwait_cpu_dead, mwait_cpu_dead); +/* Logical package management. We might want to allocate that dynamically */ unsigned int __max_logical_packages__read_mostly ;EXPORT_SYMBOL (__max_logical_packages);
@@ -1758,10 +1769,10 @@ EXPORT_SYMBOL_GPL(cond_wakeup_cpu0);*/ static inline void mwait_play_dead (void ) { +struct mwait_cpu_dead *md =this_cpu_ptr (&mwait_cpu_dead);unsigned int eax, ebx, ecx, edx;unsigned int highest_cstate =0 ;unsigned int highest_subcstate =0 ; -void *mwait_ptr;int i; if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
@@ -1796,13 +1807,6 @@ static inline void mwait_play_dead(void) (highest_subcstate -1 ); } -/* -* This should be a memory location in a cache line which is -* unlikely to be touched by other processors. The actual -* content is immaterial as it is not actually modified in any way. -*/ - mwait_ptr = ¤t_thread_info ()->flags; -wbinvd ();while (1 ) {
@@ -1814,9 +1818,9 @@ static inline void mwait_play_dead(void)* case where we return around the loop. */ mb (); -clflush (mwait_ptr); +clflush (md);mb (); -__monitor (mwait_ptr,0 ,0 ); +__monitor (md,0 ,0 );mb ();__mwait (eax,0 );
PATCH 4/6 x86/smp: Cure kexec() vs. mwait_play_dead() breakage
In reply to Thomas GleixnerTLDR: It's a mess.
When kexec() is executed on a system with "offline" CPUs, which are parked in mwait_play_dead() it can end up in a triple fault during the bootup of the kexec kernel or cause hard to diagnose data corruption.
The reason is that kexec() eventually overwrites the previous kernels text, page tables, data and stack, If it writes to the cache line which is monitored by an previously offlined CPU, MWAIT resumes execution and ends up executing the wrong text, dereferencing overwritten page tables or corrupting the kexec kernels data.
Cure this by bringing the offline CPUs out of MWAIT into HLT.
Write to the monitored cache line of each offline CPU, which makes MWAIT resume execution. The written control word tells the offline CPUs to issue HLT, which does not have the MWAIT problem.
That does not help, if a stray NMI, MCE or SMI hits the offline CPUs as those make it come out of HLT.
A follow up change will put them into INIT, which protects at least against NMI and SMI.
--- arch/x86/include/asm/smp.h | 2 + arch/x86/kernel/smp.c | 21 +++++++--------- arch/x86/kernel/smpboot.c | 59 +++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 71 insertions(+), 11 deletions(-)
arch/x86/include/asm/smp.h +2 -0
@@ -132,6 +132,8 @@ void wbinvd_on_cpu(int cpu);int wbinvd_on_all_cpus (void );void cond_wakeup_cpu0 (void ); +void smp_kick_mwait_play_dead (void ); +void native_smp_send_reschedule (int cpu);void native_send_call_func_ipi (const struct cpumask *mask);void native_send_call_func_single_ipi (int cpu);
arch/x86/kernel/smp.c +10 -11
@@ -21,6 +21,7 @@#include <linux/interrupt.h> #include <linux/cpu.h> #include <linux/gfp.h> +#include <linux/kexec.h> #include <asm/mtrr.h> #include <asm/tlbflush.h>
@@ -156,19 +157,17 @@ static void native_stop_other_cpus(int wif (atomic_cmpxchg (&stopping_cpu,-1 ,safe_smp_processor_id ()) !=-1 )return ; -/* -* Use an own vector here because smp_call_function -* does lots of things not suitable in a panic situation. -*/ +/* For kexec, ensure that offline CPUs are out of MWAIT and in HLT */ +if (kexec_in_progress) +smp_kick_mwait_play_dead ();/* -* We start by using the REBOOT_VECTOR irq. -* The irq is treated as a sync point to allow critical -* regions of code on other cpus to release their spin locks -* and re-enable irqs. Jumping straight to an NMI might -* accidentally cause deadlocks with further shutdown/panic -* code. By syncing, we give the cpus up to one second to -* finish their work before we force them off with the NMI. +* Start by using the REBOOT_VECTOR. That acts as a sync point to +* allow critical regions of code on other cpus to leave their +* critical regions. Jumping straight to an NMI might accidentally +* cause deadlocks with further shutdown code. This gives the CPUs +* up to one second to finish their work before forcing them off +* with the NMI. */ if (num_online_cpus () >1 ) {apic_send_IPI_allbutself (REBOOT_VECTOR );
arch/x86/kernel/smpboot.c +59 -0
93 lines not on this page · raw message
Re: [patch 4/6] x86/smp: Cure kexec() vs. mwait_play_dead() breakage
Quoted 21 lines · Thomas Gleixner
TLDR: It's a mess.
When
kexec()is executed on a system with "offline" CPUs, which are parked inmwait_play_dead()it can end up in a triple fault during the bootup of the kexec kernel or cause hard to diagnose data corruption.The reason is that
kexec()eventually overwrites the previous kernels text, page tables, data and stack, If it writes to the cache line which is monitored by an previously offlined CPU, MWAIT resumes execution and ends up executing the wrong text, dereferencing overwritten page tables or corrupting the kexec kernels data.Cure this by bringing the offline CPUs out of MWAIT into HLT.
Write to the monitored cache line of each offline CPU, which makes MWAIT resume execution. The written control word tells the offline CPUs to issue HLT, which does not have the MWAIT problem.
That does not help, if a stray NMI, MCE or SMI hits the offline CPUs as those make it come out of HLT.
Quoted 6 lines
A follow up change will put them into INIT, which protects at least against NMI and SMI.
Fixes ea53069231f9 ("x86, hotplug: Use mwait to offline a processor, fix the legacy case")Reported-by Ashok Raj <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>
Does this need a stable tag?
Re: [patch 4/6] x86/smp: Cure kexec() vs. mwait_play_dead() breakage
In reply to Thomas GleixnerOn Sat, Jun 03, 2023 at 10:07:01PM +0200, Thomas Gleixner wrote:
[snip]
Quoted 12 lines
Fixes ea53069231f9 ("x86, hotplug: Use mwait to offline a processor, fix the legacy case")Reported-by Ashok Raj <[email protected]>Signed-off-by Thomas Gleixner <[email protected]>--- arch/x86/include/asm/smp.h | 2 + arch/x86/kernel/smp.c | 21 +++++++--------- arch/x86/kernel/smpboot.c | 59 +++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 71 insertions(+), 11 deletions(-)
arch/x86/include/asm/smp.h void wbinvd_on_cpu(int cpu); · line 136
Quoted 5 lines
arch/x86/include/asm/smp.h +2
-0@@ -132,6 +132,8 @@ void wbinvd_on_cpu(int cpu);int wbinvd_on_all_cpus (void );void cond_wakeup_cpu0 (void ); +void smp_kick_mwait_play_dead (void ); +
This seems like its missing prototype for #else for !CONFIG_SMP
#else /* !CONFIG_SMP */
+#define smp_kick_mwait_play_dead(void) { }
Sorry I missed noticing this
Re: [patch 4/6] x86/smp: Cure kexec() vs. mwait_play_dead() breakage
Quoted 9 lines · Ashok Raj
arch/x86/include/asm/smp.h void wbinvd_on_cpu(int cpu); · line 136
Quoted 3 lines · Thomas Gleixner
arch/x86/include/asm/smp.h +2
-0+void smp_kick_mwait_play_dead (void ); +This seems like its missing prototype for #else for !CONFIG_SMP
#else /* !CONFIG_SMP */
+#definesmp_kick_mwait_play_dead(void){ }
Nope. The code which calls this is only compiled when SMP=y
PATCH 5/6 x86/smp: Split sending INIT IPI out into a helper function
In reply to Thomas GleixnerPutting CPUs into INIT is a safer place during kexec() to park CPUs.
Split the INIT assert/deassert sequence out so it can be reused.
--- arch/x86/kernel/smpboot.c | 51 +++++++++++++++++++--------------------------- 1 file changed, 22 insertions(+), 29 deletions(-)
arch/x86/kernel/smpboot.c +22 -29
@@ -853,47 +853,40 @@ wakeup_secondary_cpu_via_nmi(int apicid,return (send_status | accept_status); } -static int -wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +static void send_init_sequence (int phys_apicid) { -unsigned long send_status =0 , accept_status =0 ; -int maxlvt, num_starts, j; - - maxlvt =lapic_get_maxlvt (); +int maxlvt =lapic_get_maxlvt (); -/* -* Be paranoid about clearing APIC errors. -*/ +/* Be paranoid about clearing APIC errors. */ if (APIC_INTEGRATED (boot_cpu_apic_version)) { -if (maxlvt >3 )/* Due to the Pentium erratum 3AP. */ +/* Due to the Pentium erratum 3AP. */ +if (maxlvt >3 )apic_write (APIC_ESR ,0 );apic_read (APIC_ESR ); } -pr_debug ("Asserting INIT\n" ); - -/* -* Turn INIT on target chip -*/ -/* -* Send IPI -*/ -apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , - phys_apicid); - -pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +/* Assert INIT on the target CPU */ +apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle ();udelay (init_udelay); -pr_debug ("Deasserting INIT\n" ); - -/* Target chip */ -/* Send IPI */ +/* Deassert INIT on the target CPU */ apic_icr_write (APIC_INT_LEVELTRIG |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle (); +} + +/* +* Wake up AP by INIT, INIT, STARTUP sequence. +*/ +static int wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +{ +unsigned long send_status =0 , accept_status =0 ; +int maxlvt, num_starts, j; + +preempt_disable (); -pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +send_init_sequence (phys_apicid);mb ();
Re: [patch 5/6] x86/smp: Split sending INIT IPI out into a helper function
Hi,
Quoted 64 lines · Thomas Gleixner
Putting CPUs into INIT is a safer place during
kexec()to park CPUs.Split the INIT assert/deassert sequence out so it can be reused.
Signed-off-by Thomas Gleixner <[email protected]>--- arch/x86/kernel/smpboot.c | 51 +++++++++++++++++++--------------------------- 1 file changed, 22 insertions(+), 29 deletions(-)arch/x86/kernel/smpboot.c +11
-27@@ -853,47 +853,40 @@ wakeup_secondary_cpu_via_nmi(int apicid,return (send_status | accept_status); } -static int -wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +static void send_init_sequence (int phys_apicid) { -unsigned long send_status =0 , accept_status =0 ; -int maxlvt, num_starts, j; - - maxlvt =lapic_get_maxlvt (); +int maxlvt =lapic_get_maxlvt (); -/* -* Be paranoid about clearing APIC errors. -*/ +/* Be paranoid about clearing APIC errors. */ if (APIC_INTEGRATED (boot_cpu_apic_version)) { -if (maxlvt >3 )/* Due to the Pentium erratum 3AP. */ +/* Due to the Pentium erratum 3AP. */ +if (maxlvt >3 )apic_write (APIC_ESR ,0 );apic_read (APIC_ESR ); } -pr_debug ("Asserting INIT\n" ); - -/* -* Turn INIT on target chip -*/ -/* -* Send IPI -*/ -apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , - phys_apicid); - -pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +/* Assert INIT on the target CPU */ +apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle ();udelay (init_udelay); -pr_debug ("Deasserting INIT\n" ); - -/* Target chip */ -/* Send IPI */ +/* Deassert INIT on the target CPU */ apic_icr_write (APIC_INT_LEVELTRIG |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle (); +}
arch/x86/kernel/smpboot.c wakeup_secondary_cpu_via_nmi(int apicid, · line 887
Quoted 10 lines
arch/x86/kernel/smpboot.c +10
-0+ +/* +* Wake up AP by INIT, INIT, STARTUP sequence. +*/ +static int wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +{ +unsigned long send_status =0 , accept_status =0 ; +int maxlvt, num_starts, j; + +preempt_disable ();
This seems like an unbalanced preempt disable..
arch/x86/kernel/smpboot.c wakeup_secondary_cpu_via_nmi(int apicid,
Quoted 8 lines
arch/x86/kernel/smpboot.c +1
-2-pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +send_init_sequence (phys_apicid);mb ();
--Mika
Re: [patch 5/6] x86/smp: Split sending INIT IPI out into a helper function
Hi
Quoted 62 lines · Mika Penttilä
Hi,
Quoted 59 lines · Thomas Gleixner
Putting CPUs into INIT is a safer place during
kexec()to park CPUs.Split the INIT assert/deassert sequence out so it can be reused.
Signed-off-by Thomas Gleixner <[email protected]>--- arch/x86/kernel/smpboot.c | 51 +++++++++++++++++++--------------------------- 1 file changed, 22 insertions(+), 29 deletions(-)arch/x86/kernel/smpboot.c +11
-27@@ -853,47 +853,40 @@ wakeup_secondary_cpu_via_nmi(int apicid,return (send_status | accept_status); } -static int -wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +static void send_init_sequence (int phys_apicid) { -unsigned long send_status =0 , accept_status =0 ; -int maxlvt, num_starts, j; - - maxlvt =lapic_get_maxlvt (); +int maxlvt =lapic_get_maxlvt (); -/* -* Be paranoid about clearing APIC errors. -*/ +/* Be paranoid about clearing APIC errors. */ if (APIC_INTEGRATED (boot_cpu_apic_version)) { -if (maxlvt >3 )/* Due to the Pentium erratum 3AP. */ +/* Due to the Pentium erratum 3AP. */ +if (maxlvt >3 )apic_write (APIC_ESR ,0 );apic_read (APIC_ESR ); } -pr_debug ("Asserting INIT\n" ); - -/* -* Turn INIT on target chip -*/ -/* -* Send IPI -*/ -apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , - phys_apicid); - -pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +/* Assert INIT on the target CPU */ +apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle ();udelay (init_udelay); -pr_debug ("Deasserting INIT\n" ); - -/* Target chip */ -/* Send IPI */ +/* Deassert INIT on the target CPU */ apic_icr_write (APIC_INT_LEVELTRIG |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle (); +}
Quoted 8 lines
arch/x86/kernel/smpboot.c wakeup_secondary_cpu_via_nmi(int apicid, · line 885
Quoted 8 lines
arch/x86/kernel/smpboot.c +8
-0+ +/* +* Wake up AP by INIT, INIT, STARTUP sequence. +*/ +static int wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +{ +unsigned long send_status =0 , accept_status =0 ; +int maxlvt, num_starts, j;
Also maxlvt used uninitialized.
Quoted 14 lines
arch/x86/kernel/smpboot.c wakeup_secondary_cpu_via_nmi(int apicid, · line 882
Quoted 2 lines
arch/x86/kernel/smpboot.c +2
-0+ +preempt_disable ();This seems like an unbalanced preempt disable..
arch/x86/kernel/smpboot.c wakeup_secondary_cpu_via_nmi(int apicid, · line 892
Quoted 5 lines
arch/x86/kernel/smpboot.c +1
-2-pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +send_init_sequence (phys_apicid);mb ();--Mika
Re: [patch 5/6] x86/smp: Split sending INIT IPI out into a helper function
In reply to Mika PenttiläQuoted 12 lines · Mika Penttilä
arch/x86/kernel/smpboot.c wakeup_secondary_cpu_via_nmi(int apicid, · line 887
Quoted 9 lines · Thomas Gleixner
arch/x86/kernel/smpboot.c +9
-0+/* +* Wake up AP by INIT, INIT, STARTUP sequence. +*/ +static int wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +{ +unsigned long send_status =0 , accept_status =0 ; +int maxlvt, num_starts, j; + +preempt_disable ();This seems like an unbalanced preempt disable..
Right. I screwed up a rebase completely....
PATCH v2 5/6 x86/smp: Split sending INIT IPI out into a helper function
In reply to Thomas GleixnerSubject: x86/smp: Split sending INIT IPI out into a helper function
From: Thomas Gleixner <[email protected]>
Date: Fri, 02 Jun 2023 15:04:10 +0200
Putting CPUs into INIT is a safer place during kexec() to park CPUs.
Split the INIT assert/deassert sequence out so it can be reused.
---
V2: Fix rebase screwup
--- arch/x86/kernel/smpboot.c | 49 ++++++++++++++++++---------------------------- 1 file changed, 20 insertions(+), 29 deletions(-)
arch/x86/kernel/smpboot.c +20 -29
@@ -853,47 +853,38 @@ wakeup_secondary_cpu_via_nmi(int apicid,return (send_status | accept_status); } -static int -wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +static void send_init_sequence (int phys_apicid) { -unsigned long send_status =0 , accept_status =0 ; -int maxlvt, num_starts, j; - - maxlvt =lapic_get_maxlvt (); +int maxlvt =lapic_get_maxlvt (); -/* -* Be paranoid about clearing APIC errors. -*/ +/* Be paranoid about clearing APIC errors. */ if (APIC_INTEGRATED (boot_cpu_apic_version)) { -if (maxlvt >3 )/* Due to the Pentium erratum 3AP. */ +/* Due to the Pentium erratum 3AP. */ +if (maxlvt >3 )apic_write (APIC_ESR ,0 );apic_read (APIC_ESR ); } -pr_debug ("Asserting INIT\n" ); - -/* -* Turn INIT on target chip -*/ -/* -* Send IPI -*/ -apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , - phys_apicid); - -pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +/* Assert INIT on the target CPU */ +apic_icr_write (APIC_INT_LEVELTRIG |APIC_INT_ASSERT |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle ();udelay (init_udelay); -pr_debug ("Deasserting INIT\n" ); - -/* Target chip */ -/* Send IPI */ +/* Deassert INIT on the target CPU */ apic_icr_write (APIC_INT_LEVELTRIG |APIC_DM_INIT , phys_apicid); +safe_apic_wait_icr_idle (); +} + +/* +* Wake up AP by INIT, INIT, STARTUP sequence. +*/ +static int wakeup_secondary_cpu_via_init (int phys_apicid,unsigned long start_eip) +{ +unsigned long send_status =0 , accept_status =0 ; +int num_starts, j, maxlvt =lapic_get_maxlvt (); -pr_debug ("Waiting for send to finish...\n" ); - send_status =safe_apic_wait_icr_idle (); +send_init_sequence (phys_apicid);mb ();
PATCH 6/6 x86/smp: Put CPUs into INIT on shutdown if possible
In reply to Thomas GleixnerParking CPUs in a HLT loop is not completely safe vs. kexec() as HLT can resume execution due to NMI, SMI and MCE, which has the same issue as the MWAIT loop.
Kicking the secondary CPUs into INIT makes this safe against NMI and SMI.
A broadcast MCE will take the machine down, but a broadcast MCE which makes HLT resume and execute overwritten text, pagetables or data will end up in a disaster too.
So chose the lesser of two evils and kick the secondary CPUs into INIT unless the system has installed special wakeup mechanisms which are not using INIT.
--- arch/x86/include/asm/smp.h | 2 ++ arch/x86/kernel/smp.c | 38 +++++++++++++++++++++++++++++--------- arch/x86/kernel/smpboot.c | 19 +++++++++++++++++++ 3 files changed, 50 insertions(+), 9 deletions(-)
arch/x86/include/asm/smp.h +2 -0
@@ -139,6 +139,8 @@ void native_send_call_func_ipi(const strvoid native_send_call_func_single_ipi (int cpu);void x86_idle_thread_init (unsigned int cpu,struct task_struct *idle); +bool smp_park_nonboot_cpus_in_init (void ); +void smp_store_boot_cpu_info (void );void smp_store_cpu_info (int id);
arch/x86/kernel/smp.c +29 -9
@@ -130,7 +130,7 @@ static int smp_stop_nmi_callback(unsigne }/* -* this function calls the 'stop' function on all other CPUs in the system. +* Disable virtualization, APIC etc. and park the CPU in a HLT loop */ DEFINE_IDTENTRY_SYSVEC (sysvec_reboot) {
@@ -147,8 +147,7 @@ static int register_stop_handler(void)static void native_stop_other_cpus (int wait) { -unsigned long flags; -unsigned long timeout; +unsigned long flags, timeout;if (reboot_force)return ;
@@ -164,10 +163,10 @@ static void native_stop_other_cpus(int w/* * Start by using the REBOOT_VECTOR. That acts as a sync point to * allow critical regions of code on other cpus to leave their -* critical regions. Jumping straight to an NMI might accidentally -* cause deadlocks with further shutdown code. This gives the CPUs -* up to one second to finish their work before forcing them off -* with the NMI. +* critical regions. Jumping straight to NMI or INIT might +* accidentally cause deadlocks with further shutdown code. This +* gives the CPUs up to one second to finish their work before +* forcing them off with the NMI or INIT. */ if (num_online_cpus () >1 ) {apic_send_IPI_allbutself (REBOOT_VECTOR );
@@ -175,7 +174,7 @@ static void native_stop_other_cpus(int w/* * Don't wait longer than a second for IPI completion. The * wait request is not checked here because that would -* prevent an NMI shutdown attempt in case that not all +* prevent an NMI/INIT shutdown in case that not all * CPUs reach shutdown state. */ timeout =USEC_PER_SEC ;
@@ -183,7 +182,27 @@ static void native_stop_other_cpus(int wudelay (1 ); } -/* if the REBOOT_VECTOR didn't work, try with the NMI */ +/* +* Park all nonboot CPUs in INIT including offline CPUs, if +* possible. That's a safe place where they can't resume execution +* of HLT and then execute the HLT loop from overwritten text or +* page tables. +* +* The only downside is a broadcast MCE, but up to the point where +* the kexec() kernel brought all APs online again an MCE will just +* make HLT resume and handle the MCE. The machine crashs and burns +* due to overwritten text, page tables and data. So there is a +* choice between fire and frying pan. The result is pretty much +* the same. Chose frying pan until x86 provides a sane mechanism +* to park a CPU. +*/ +if (smp_park_nonboot_cpus_in_init ()) + goto done; + +/* +* If park with INIT was not possible and the REBOOT_VECTOR didn't +* take all secondary CPUs offline, try with the NMI. +*/ if (num_online_cpus () >1 ) {/* * If NMI IPI is enabled, try to register the stop handler
@@ -208,6 +227,7 @@ static void native_stop_other_cpus(int wudelay (1 ); } +done:local_irq_save (flags);disable_local_APIC ();mcheck_cpu_clear (this_cpu_ptr (&cpu_info));
arch/x86/kernel/smpboot.c +19 -0
@@ -1467,6 +1467,25 @@ void arch_thaw_secondary_cpus_end(void)cache_aps_init (); } +bool smp_park_nonboot_cpus_in_init (void ) +{ +unsigned int cpu, this_cpu =smp_processor_id (); +unsigned int apicid; + +if (apic->wakeup_secondary_cpu_64 || apic->wakeup_secondary_cpu) +return false; + +for_each_present_cpu (cpu) { +if (cpu == this_cpu) +continue ; + apicid = apic->cpu_present_to_apicid (cpu); +if (apicid ==BAD_APICID ) +continue ; +send_init_sequence (apicid); + } +return true; +} +/* * Early setup to make printk work. */
Re: [patch 6/6] x86/smp: Put CPUs into INIT on shutdown if possible
Quoted 10 lines · Thomas Gleixner
Parking CPUs in a HLT loop is not completely safe vs.
kexec()as HLT can resume execution due to NMI, SMI and MCE, which has the same issue as the MWAIT loop.Kicking the secondary CPUs into INIT makes this safe against NMI and SMI.
A broadcast MCE will take the machine down, but a broadcast MCE which makes HLT resume and execute overwritten text, pagetables or data will end up in a disaster too.
Quoted 5 lines
So chose the lesser of two evils and kick the secondary CPUs into INIT unless the system has installed special wakeup mechanisms which are not using INIT.
Signed-off-by Thomas Gleixner <[email protected]>
For the whole series.
Quoted 9 lines · Thomas Gleixner
Hi!
Ashok observed triple faults when executing
kexec()on a kernel which has 'nosmt' on the kernel commandline and HT enabled in the BIOS.'nosmt' brings up the HT siblings to the point where they initiliazed the CPU and then rolls the bringup back which parks them in
mwait_play_dead(). The reason is that all CPUs should have CR4.MCE set. Otherwise a broadcast MCE will immediately shut down the machine.
...
Quoted 2 lines
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably,
On Intel. On AMD, enabling EFER.SVME and doing CLGI will block everything except single-step #DB (lol) and RESET. #MC handling is implementation-dependent and *might* cause shutdown, but at least there's a chance it will work. And presumably modern CPUs do pend the #MC until GIF=1.
Quoted 10 lines
but there is at least one which prevents the NMI and SMI cause: INIT.
3) If the system uses the regular INIT/STARTUP sequence to wake up secondary CPUS, then "park" all CPUs including the "offline" ones by sending them INIT IPIs.
The INIT IPI brings the CPU into a wait for wakeup state which is not affected by NMI and SMI, but INIT also clears CR4.MCE, so the broadcast MCE problem comes back.
Quoted 6 lines
But that's not really any different from a CPU sitting in the HLT loop on the previous kernel. If a broadcast MCE arrives, HLT resumes execution and the CPU tries to handle the MCE on overwritten text, pagetables etc.
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Why not carve out a page that's hidden across kexec() to hold whatever code+data is needed to safely execute a HLT loop indefinitely? E.g. doesn't the original kernel provide the e820 tables for the post-kexec() kernel? To avoid OOM after many kexec(), reserving a page could be done iff the current kernel wasn't itself kexec()'d.
Quoted 8 lines · Sean Christopherson
Quoted 2 lines · Thomas Gleixner
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably,
On Intel. On AMD, enabling EFER.SVME and doing CLGI will block everything except single-step #DB (lol) and RESET. #MC handling is implementation-dependent and *might* cause shutdown, but at least there's a chance it will work. And presumably modern CPUs do pend the #MC until GIF=1.
Abusing SVME for that is definitely in the realm of creative bonus points, but not necessarily a general purpose solution.
Quoted 5 lines
Quoted 2 lines
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
Quoted 2 lines
Why not carve out a page that's hidden across
kexec()to hold whatever code+data is needed to safely execute a HLT loop indefinitely?
See below.
Quoted 2 lines
E.g. doesn't the original kernel provide the e820 tables for the post-kexec() kernel?
Only for crash kernels if I'm not missing something.
Making this work for regular kexec() including this:
Quoted 2 lines
To avoid OOM after many
kexec(), reserving a page could be done iff the current kernel wasn't itself kexec()'d.
would be possible and I thought about it, but that needs a complete new design of "offline", "shutdown offline" and a non-trivial amount of backwards compatibility magic because you can't assume that the kexec() kernel version is greater or equal to the current one. kexec() is supposed to work both ways, downgrading and upgrading. IOW, that ship sailed long ago.
Thanks,
tglx
Quoted 5 lines · Thomas Gleixner
Quoted 4 lines · Sean Christopherson
Quoted 2 lines · Thomas Gleixner
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably,
Quoted 7 lines
Quoted 4 lines
On Intel. On AMD, enabling EFER.SVME and doing CLGI will block everything except single-step #DB (lol) and RESET. #MC handling is implementation-dependent and *might* cause shutdown, but at least there's a chance it will work. And presumably modern CPUs do pend the #MC until GIF=1.
Abusing SVME for that is definitely in the realm of creative bonus points, but not necessarily a general purpose solution.
Heh, my follow-up ideas for Intel are to abuse XuCode or SEAM ;-)
Quoted 9 lines
Quoted 5 lines
Quoted 2 lines
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
I was thinking of the EDKII code in UefiCpuPkg/PiSmmCpuDxeSmm/MpService.c, e.g. SmmWaitForApArrival(). I've never dug deeply into how EDKII uses SMM, what its timeouts are, etc., I just remember coming across that code when poking around EDKII for other stuff.
Quoted 9 lines
Quoted 2 lines
Why not carve out a page that's hidden across
kexec()to hold whatever code+data is needed to safely execute a HLT loop indefinitely?See below.
Quoted 2 lines
E.g. doesn't the original kernel provide the e820 tables for the post-kexec() kernel?
Only for crash kernels if I'm not missing something.
Ah, drat.
Quoted 11 lines
Making this work for regular
kexec()including this:Quoted 2 lines
To avoid OOM after many
kexec(), reserving a page could be done iff the current kernel wasn't itself kexec()'d.would be possible and I thought about it, but that needs a complete new design of "offline", "shutdown offline" and a non-trivial amount of backwards compatibility magic because you can't assume that the
kexec()kernel version is greater or equal to the current one.kexec()is supposed to work both ways, downgrading and upgrading. IOW, that ship sailed long ago.
Right, but doesn't gaining "full" protection require ruling out unenlightened downgrades? E.g. if someone downgrades to an old kernel, doesn't hide the "offline" CPUs from the kexec() kernel, and boots the old kernel with -nosmt or whatever, then that old kernel will do the naive MWAIT or unprotected HLT and it's hosed again.
If we're relying on the admin to hide the offline CPUs, could we usurp an existing kernel param to hide a small chunk of memory instead?
Quoted 11 lines · Sean Christopherson
Quoted 10 lines · Thomas Gleixner
Quoted 8 lines · Sean Christopherson
Quoted 2 lines · Thomas Gleixner
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably,
On Intel. On AMD, enabling EFER.SVME and doing CLGI will block everything except single-step #DB (lol) and RESET. #MC handling is implementation-dependent and *might* cause shutdown, but at least there's a chance it will work. And presumably modern CPUs do pend the #MC until GIF=1.
Quoted 4 lines
Quoted 2 lines
Abusing SVME for that is definitely in the realm of creative bonus points, but not necessarily a general purpose solution.
Heh, my follow-up ideas for Intel are to abuse XuCode or SEAM ;-)
I feared that :)
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 2 lines
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Quoted 8 lines
Quoted 3 lines
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
I was thinking of the EDKII code in
UefiCpuPkg/PiSmmCpuDxeSmm/MpService.c, e.g.SmmWaitForApArrival(). I've never dug deeply into how EDKII uses SMM, what its timeouts are, etc., I just remember coming across that code when poking around EDKII for other stuff.
There is a comment:
Note the SMI Handlers must ALWAYS take into account the cases that not all APs are available in an SMI run.
Also not all SMIs required global synchronization. But it's all an inpenetrable mess...
Quoted 12 lines
Quoted 11 lines
Making this work for regular
kexec()including this:Quoted 2 lines
To avoid OOM after many
kexec(), reserving a page could be done iff the current kernel wasn't itself kexec()'d.would be possible and I thought about it, but that needs a complete new design of "offline", "shutdown offline" and a non-trivial amount of backwards compatibility magic because you can't assume that the
kexec()kernel version is greater or equal to the current one.kexec()is supposed to work both ways, downgrading and upgrading. IOW, that ship sailed long ago.
Quoted 5 lines
Right, but doesn't gaining "full" protection require ruling out unenlightened downgrades? E.g. if someone downgrades to an old kernel, doesn't hide the "offline" CPUs from the
kexec()kernel, and boots the old kernel with -nosmt or whatever, then that old kernel will do the naive MWAIT or unprotected HLT and it's hosed again.
Of course.
Quoted 2 lines
If we're relying on the admin to hide the offline CPUs, could we usurp an existing kernel param to hide a small chunk of memory instead?
The only "safe" place is below 1M I think. Not sure whether we have some existing command line option to "hide" a range there. Neither am I sure that this would be always the same range.
More questions than answers :)
Thanks
tglx
Quoted 16 lines · Thomas Gleixner
Quoted 8 lines · Sean Christopherson
Quoted 2 lines · Thomas Gleixner
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably,
On Intel. On AMD, enabling EFER.SVME and doing CLGI will block everything except single-step #DB (lol) and RESET. #MC handling is implementation-dependent and *might* cause shutdown, but at least there's a chance it will work. And presumably modern CPUs do pend the #MC until GIF=1.
Abusing SVME for that is definitely in the realm of creative bonus points, but not necessarily a general purpose solution.
Quoted 3 lines
Quoted 2 lines
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Quoted 6 lines
Quoted 2 lines
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
SMM does do the rendezvous of all CPUs, but also has a way to detect the blocked ones, in WFS via some package scoped ubox register. So it knows to skip those. I can find this in internal sources, but they aren't available in the edk2 open reference code. They happen to be documented only in the BWG, which isn't available freely.
I believe its behind the GetSmmDelayedBlockedDisabledCount()->
SmmCpuFeaturesGetSmmRegister()
Quoted 24 lines · Ashok Raj
Quoted 16 lines · Thomas Gleixner
Quoted 8 lines · Sean Christopherson
Quoted 2 lines · Thomas Gleixner
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably,
On Intel. On AMD, enabling EFER.SVME and doing CLGI will block everything except single-step #DB (lol) and RESET. #MC handling is implementation-dependent and *might* cause shutdown, but at least there's a chance it will work. And presumably modern CPUs do pend the #MC until GIF=1.
Abusing SVME for that is definitely in the realm of creative bonus points, but not necessarily a general purpose solution.
Quoted 3 lines
Quoted 2 lines
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Quoted 6 lines
Quoted 2 lines
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
Quoted 5 lines
SMM does do the rendezvous of all CPUs, but also has a way to detect the blocked ones, in WFS via some package scoped ubox register. So it knows to skip those. I can find this in internal sources, but they aren't available in the edk2 open reference code. They happen to be documented only in the BWG, which isn't available freely.
Ah, so putting CPUs into WFS shouldn't result in odd delays. At least not on bare metal. Hmm, and AFAIK the primary use case for SMM in VMs is for secure boot, so taking SMIs after booting and putting CPUs back into WFS should be ok-ish.
Finding a victim to test this in a QEMU VM w/ Secure Boot would be nice to have.
Quoted 25 lines · Sean Christopherson
Quoted 24 lines · Ashok Raj
Quoted 22 lines · Thomas Gleixner
Quoted 8 lines · Sean Christopherson
Quoted 2 lines · Thomas Gleixner
This is only half safe because HLT can resume execution due to NMI, SMI and MCE. Unfortunately there is no real safe mechanism to "park" a CPU reliably,
On Intel. On AMD, enabling EFER.SVME and doing CLGI will block everything except single-step #DB (lol) and RESET. #MC handling is implementation-dependent and *might* cause shutdown, but at least there's a chance it will work. And presumably modern CPUs do pend the #MC until GIF=1.
Abusing SVME for that is definitely in the realm of creative bonus points, but not necessarily a general purpose solution.
Quoted 5 lines
Quoted 2 lines
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
Quoted 8 lines
Quoted 5 lines
SMM does do the rendezvous of all CPUs, but also has a way to detect the blocked ones, in WFS via some package scoped ubox register. So it knows to skip those. I can find this in internal sources, but they aren't available in the edk2 open reference code. They happen to be documented only in the BWG, which isn't available freely.
Ah, so putting CPUs into WFS shouldn't result in odd delays. At least not on bare metal. Hmm, and AFAIK the primary use case for SMM in VMs is for secure
Never knew SMM had any role in VM's.. I thought SMM was always native.
Who owns this SMM for VM's.. from the VirtualBIOS?
Quoted 3 lines
boot, so taking SMIs after booting and putting CPUs back into WFS should be ok-ish.
Finding a victim to test this in a QEMU VM w/ Secure Boot would be nice to have.
I always seem to turn off secureboot installing Ubuntu :-).. I'll try to find someone who might know especially doing SMM In VM.
Can you tell what needs to be validated in the guest? Would doing kexec inside the guest with the new patch set be sufficient?
Or you mean in guest, do a kexec and launch secure boot of new kernel?
If there is a specific test you want done, let me know.
Quoted 19 lines · Ashok Raj
Quoted 18 lines · Sean Christopherson
Quoted 16 lines · Ashok Raj
Quoted 9 lines · Thomas Gleixner
Quoted 5 lines
Quoted 2 lines
So parking them via INIT is not completely solving the problem, but it takes at least NMI and SMI out of the picture.
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely potentially cause problems too?
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
SMM does do the rendezvous of all CPUs, but also has a way to detect the blocked ones, in WFS via some package scoped ubox register. So it knows to skip those. I can find this in internal sources, but they aren't available in the edk2 open reference code. They happen to be documented only in the BWG, which isn't available freely.
Quoted 6 lines
Quoted 2 lines
Ah, so putting CPUs into WFS shouldn't result in odd delays. At least not on bare metal. Hmm, and AFAIK the primary use case for SMM in VMs is for secure
Never knew SMM had any role in VM's.. I thought SMM was always native.
Who owns this SMM for VM's.. from the VirtualBIOS?
Yes?
Quoted 5 lines
Quoted 3 lines
boot, so taking SMIs after booting and putting CPUs back into WFS should be ok-ish.
Finding a victim to test this in a QEMU VM w/ Secure Boot would be nice to have.
I always seem to turn off secureboot installing Ubuntu :-)
Yeah, I don't utilize it in any of my VMs either.
Quoted 6 lines
I'll try to find someone who might know especially doing SMM In VM.
Can you tell what needs to be validated in the guest? Would doing kexec inside the guest with the new patch set be sufficient?
Or you mean in guest, do a kexec and launch secure boot of new kernel?
Yes? I don't actually have hands on experience with such a setup, I'm familiar with it purely through bug reports, e.g. this one
Quoted 1 line
If there is a specific test you want done, let me know.
Smoke testing is all I was thinking. I wouldn't put too much effort into trying to make sure this all works. Like I said earlier, nice to have, but certainly not necessary.
Quoted 11 lines · Sean Christopherson
Yes? I don't actually have hands on experience with such a setup, I'm familiar with it purely through bug reports, e.g. this one
Quoted 1 line
If there is a specific test you want done, let me know.
Smoke testing is all I was thinking. I wouldn't put too much effort into trying to make sure this all works. Like I said earlier, nice to have, but certainly not necessary.
Thanks for the Link Sean. I'll do some followup.
Signature · 2 lines
Cheers, Ashok
Quoted 8 lines · Sean Christopherson
Quoted 1 line
If there is a specific test you want done, let me know.
Smoke testing is all I was thinking. I wouldn't put too much effort into trying to make sure this all works. Like I said earlier, nice to have, but certainly not necessary.
+ Vijay who was helping with testing this inside the VM.
+ Paolo, Laszlo
I haven't found the exact method to test with secure boot/trusted boot yet. But here is what we were able to test thus far.
Vijay was able to get OVMF recompiled with SMM included.
Thanks to Laszlo for pointing me in the right direction. And Paolo for helping with some basic questions.
https://github.com/tianocore/tianocore.github.io/wiki/Testing-SMM-with-QEMU,-KVM-and-libvirt
Surprisingly SMM emulation is sadly damn good :-)
Recipe is to generate SMI by writing to port 0xb2.
- On native, this does generate a broadcast SMI, the SMI_COUNT MSR 0x34 goes up by 1 on all logical CPUs.
- Turn off SMT by #echo off > /sys/devices/system/cpu/smt/control
- Do another port 0xb2, we don't see any hangs
- Bring up SMT by echo on > control, and we can see even the offline CPUs got the SMI as indicated by MSR 0x34. (Which is as expected)
On guest, the only difference was when we turn on HT again, waking the CPUs from INIT, SMI_COUNT has zeroed as opposed to the native. (Which is perfectly fine) All I was looking for was "no hang". And a normal kexec with newly updated code works well inside a guest.
Would this qualify for the smoke test pass? I'll continue to look for a secure boot install if this doesn't close it, just haven't landed at the right spot yet.
Signature · 2 lines
Cheers, Ashok
Quoted 34 lines · Ashok Raj
Quoted 8 lines · Sean Christopherson
Quoted 1 line
If there is a specific test you want done, let me know.
Smoke testing is all I was thinking. I wouldn't put too much effort into trying to make sure this all works. Like I said earlier, nice to have, but certainly not necessary.
+ Vijay who was helping with testing this inside the VM.
+ Paolo, LaszloI haven't found the exact method to test with secure boot/trusted boot yet. But here is what we were able to test thus far.
Vijay was able to get OVMF recompiled with SMM included.
Thanks to Laszlo for pointing me in the right direction. And Paolo for helping with some basic questions.
https://github.com/tianocore/tianocore.github.io/wiki/Testing-SMM-with-QEMU,-KVM-and-libvirt
Surprisingly SMM emulation is sadly damn good :-)
Recipe is to generate SMI by writing to port
0xb2.- On native, this does generate a broadcast SMI, the
SMI_COUNTMSR0x34goes up by 1 on all logical CPUs.
- Turn off SMT by #echo off >/sys/devices/system/cpu/smt/control
- Do another port0xb2, we don't see any hangs
- Bring up SMT by echo on > control, and we can see even the offline CPUs got the SMI as indicated by MSR0x34. (Which is as expected)
Quoted 8 lines
On guest, the only difference was when we turn on HT again, waking the CPUs from INIT,
SMI_COUNThas zeroed as opposed to the native. (Which is perfectly fine) All I was looking for was "no hang". And a normal kexec with newly updated code works well inside a guest.Would this qualify for the smoke test pass? I'll continue to look for a secure boot install if this doesn't close it, just haven't landed at the right spot yet.
Good enough for me, thanks much!
Quoted 5 lines · Sean Christopherson
Quoted 3 lines
Would this qualify for the smoke test pass? I'll continue to look for a secure boot install if this doesn't close it, just haven't landed at the right spot yet.
Good enough for me, thanks much!
Thanks a ton Sean.. if anything you have now scared me there is life for SMM afterall even in a guest :-)... Completely took me by surprise!
Cheers,
Ashok
Quoted 9 lines · Ashok Raj
Quoted 5 lines · Sean Christopherson
Quoted 3 lines
Would this qualify for the smoke test pass? I'll continue to look for a secure boot install if this doesn't close it, just haven't landed at the right spot yet.
Good enough for me, thanks much!
Thanks a ton Sean.. if anything you have now scared me there is life for SMM afterall even in a guest :-)
LOL, you and me both. Thankfully GCE's implementation of Secure Boot for VMs doesn't utilize SMM, so to a large extent I can close my eyes and plug my ears :-)
On 6/7/23 19:33, Sean Christopherson wrote:
Don't most SMM handlers rendezvous all CPUs? I.e. won't blocking SMIs indefinitely
Quoted 12 lines
Quoted 11 lines
Quoted 5 lines
Quoted 1 line
potentially cause problems too?
Not that I'm aware of. If so then this would be a hideous firmware bug as firmware must be aware of CPUs which hang around in INIT independent of this.
SMM does do the rendezvous of all CPUs, but also has a way to detect the blocked ones, in WFS via some package scoped ubox register. So it knows to skip those. I can find this in internal sources, but they aren't available in the edk2 open reference code. They happen to be documented only in the BWG, which isn't available freely.
Quoted 3 lines
Ah, so putting CPUs into WFS shouldn't result in odd delays. At least not on bare metal. Hmm, and AFAIK the primary use case for SMM in VMs is for secure boot, so taking SMIs after booting and putting CPUs back into WFS should be ok-ish.
VMs do not have things like periodic or watchdog SMIs, they only enter SMM in response to IPIs or writes to 0xB1.
The writes to 0xB1 in turn should only happen from UEFI runtime services related to the UEFI variable store. Another possibility could be ACPI bytecode from either DSDT or APEI; not implemented yet and very unlikely to happen in the future, but not impossible either.
Either way they should not happen before the kexec-ed kernel has brought up all CPUs.
Paolo
Quoted 1 line
Finding a victim to test this in a QEMU VM w/ Secure Boot would be nice to have.