lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240827203804.4989-1-Ashish.Kalra@amd.com>
Date: Tue, 27 Aug 2024 20:38:04 +0000
From: Ashish Kalra <Ashish.Kalra@....com>
To: <seanjc@...gle.com>, <pbonzini@...hat.com>, <dave.hansen@...ux.intel.com>,
	<tglx@...utronix.de>, <mingo@...hat.com>, <bp@...en8.de>, <x86@...nel.org>
CC: <hpa@...or.com>, <peterz@...radead.org>, <linux-kernel@...r.kernel.org>,
	<kvm@...r.kernel.org>, <thomas.lendacky@....com>, <michael.roth@....com>,
	<kexec@...ts.infradead.org>, <linux-coco@...ts.linux.dev>
Subject: [PATCH] x86/sev: Fix host kdump support for SNP

From: Ashish Kalra <ashish.kalra@....com>

With active SNP VMs, SNP_SHUTDOWN_EX invoked during panic notifiers causes
crashkernel boot failure with the following signature:

[  563.497112] sysrq: Trigger a crash
[  563.508415] Kernel panic - not syncing: sysrq triggered crash
[  563.522002] CPU: 10 UID: 0 PID: 4661 Comm: bash Kdump: loaded Not tainted 6.11.0-rc3-next-20240813-snp-host-f2a41ff576cc-dirty #61
[  563.549762] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS RXM100AB 10/17/2022
[  563.566266] Call Trace:
[  563.576430]  <TASK>
[  563.585932]  dump_stack_lvl+0x2b/0x90
[  563.597244]  dump_stack+0x14/0x20
[  563.608141]  panic+0x3b9/0x400
[  563.618801]  ? srso_alias_return_thunk+0x5/0xfbef5
[  563.631271]  sysrq_handle_crash+0x19/0x20
[  563.642696]  __handle_sysrq+0xf9/0x290
[  563.653691]  ? srso_alias_return_thunk+0x5/0xfbef5
[  563.666126]  write_sysrq_trigger+0x60/0x80
...
...
[  564.186804] in panic
[  564.194287] in panic_other_cpus_shutdown
[  564.203674] kexec: in crash_smp_send_stop
[  564.213205] kexec: in kdump_nmi_shootdown_cpus
[  564.224338] Kernel Offset: 0x35a00000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
[  564.282209] in snp_shutdown_on_panic after decommision, wbinvd + df_flush required
[  564.462217] ccp 0000:23:00.1: SEV-SNP DF_FLUSH failed with error 14
[  564.676920] kexec: in native_machine_crash_shutdown
early console in extract_kernel
input_data: 0x000000007410d2cc
input_len: 0x0000000000ce98b2
output: 0x0000000071600000
output_len: 0x000000000379eb8c
kernel_total_size: 0x0000000002c30000
needed_size: 0x0000000003800000
trampoline_32bit: 0x0000000000000000

Invalid physical address chosen!

Physical KASLR disabled: no suitable memory region!

Virtual KASLR using RDRAND RDTSC...

Decompressing Linux... Parsing ELF... Performing relocations... done.
Booting the kernel (entry_offset: 0x0000000000000bda).
[    0.000000] Linux version 6.11.0-rc3-next-20240813-snp-host-f2a41ff576cc-dirty (amd@...anolx7e2ehost) (gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0, GNU ld (GNU Binutils) 2.40) #61 SMP Mon Aug 19 19:59:02 UTC 2024
[    0.000000] Command line: BOOT_IMAGE=/vmlinuz-6.11.0-rc3-next-20240813-snp-host-f2a41ff576cc-dirty root=UUID=4b87a03b-0e78-42ca-a8ad-997e63bba4e0 ro console=tty0 console=ttyS0,115200n8 earlyprintk=ttyS0,115200n8 amd_iommu_dump=1 reset_devices systemd.unit=kdump-tools-dump.service nr_cpus=1 irqpoll nousb elfcorehdr=1916276K
[    0.000000] KERNEL supported cpus:
...
...
[    1.671804] AMD-Vi: Using global IVHD EFR:0x841f77e022094ace, EFR2:0x0
[    1.679835] AMD-Vi: Translation is already enabled - trying to copy translation structures
[    1.689363] AMD-Vi: Copied DEV table from previous kernel.
[    1.864369] AMD-Vi: Completion-Wait loop timed out
[    2.038289] AMD-Vi: Completion-Wait loop timed out
[    2.212215] AMD-Vi: Completion-Wait loop timed out
[    2.386141] AMD-Vi: Completion-Wait loop timed out
[    2.560068] AMD-Vi: Completion-Wait loop timed out
[    2.733997] AMD-Vi: Completion-Wait loop timed out
[    2.907927] AMD-Vi: Completion-Wait loop timed out
[    3.081855] AMD-Vi: Completion-Wait loop timed out
[    3.225500] AMD-Vi: Completion-Wait loop timed out
[    3.231083] ..TIMER: vector=0x30 apic1=0 pin1=2 apic2=-1 pin2=-1
d out
[    3.579592] AMD-Vi: Completion-Wait loop timed out
[    3.753164] AMD-Vi: Completion-Wait loop timed out
[    3.815762] Kernel panic - not syncing: timer doesn't work through Interrupt-remapped IO-APIC
[    3.825347] CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Not tainted 6.11.0-rc3-next-20240813-snp-host-f2a41ff576cc-dirty #61
[    3.837188] Hardware name: AMD Corporation ETHANOL_X/ETHANOL_X, BIOS RXM100AB 10/17/2022
[    3.846215] Call Trace:
[    3.848939]  <TASK>
[    3.851277]  dump_stack_lvl+0x2b/0x90
[    3.855354]  dump_stack+0x14/0x20
[    3.859050]  panic+0x3b9/0x400
[    3.862454]  panic_if_irq_remap+0x21/0x30
[    3.866925]  setup_IO_APIC+0x8aa/0xa50
[    3.871106]  ? __pfx_amd_iommu_enable_faulting+0x10/0x10
[    3.877032]  ? __cpuhp_setup_state+0x5e/0xd0
[    3.881793]  apic_intr_mode_init+0x6a/0xf0
[    3.886360]  x86_late_time_init+0x28/0x40
[    3.890832]  start_kernel+0x6a8/0xb50
[    3.894914]  x86_64_start_reservations+0x1c/0x30
[    3.900064]  x86_64_start_kernel+0xbf/0x110
[    3.904729]  ? setup_ghcb+0x12/0x130
[    3.908716]  common_startup_64+0x13e/0x141
[    3.913283]  </TASK>
[    3.915715] in panic
[    3.918149] in panic_other_cpus_shutdown
[    3.922523] ---[ end Kernel panic - not syncing: timer doesn't work through Interrupt-remapped IO-APIC ]---

This happens as SNP_SHUTDOWN_EX fails when SNP VMs are active as the
firmware checks every encryption-capable ASID to verify that it is
not in use by a guest and a DF_FLUSH is not required. If a DF_FLUSH
is required, the firmware returns DFFLUSH_REQUIRED.

To fix this, added support to do SNP_DECOMMISION of all active SNP VMs
in the panic notifier before doing SNP_SHUTDOWN_EX, but then
SNP_DECOMMISION tags all CPUs on which guest has been activated to do
a WBINVD. This causes SNP_DF_FLUSH command failure with the following
flow: SNP_DECOMMISION -> SNP_SHUTDOWN_EX -> SNP_DF_FLUSH ->
failure with WBINVD_REQUIRED.

When panic notifier is invoked all other CPUs have already been
shutdown, so it is not possible to do a wbinvd_on_all_cpus() after
SNP_DECOMMISION has been executed. This eventually causes SNP_SHUTDOWN_EX
to fail after SNP_DECOMMISION.

Adding fix to do SNP_DECOMMISION and subsequent WBINVD on all CPUs
during NMI shutdown of CPUs as part of disabling virtualization on
all CPUs via cpu_emergency_disable_virtualization ->
svm_emergency_disable().

This fixes and enables crashkernel/kdump on SNP host.

Fixes: c3b86e61b756 ("x86/cpufeatures: Enable/unmask SEV-SNP CPU feature")
Signed-off-by: Ashish Kalra <ashish.kalra@....com>
---
 arch/x86/kvm/svm/sev.c | 39 ++++++++++++++++++
 arch/x86/kvm/svm/svm.c | 91 ++++++++++++++++++++++++++++++++++++++++++
 arch/x86/kvm/svm/svm.h |  3 +-
 3 files changed, 132 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index 0b851ef937f2..34ddea43c4e6 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -89,6 +89,7 @@ static unsigned int nr_asids;
 static unsigned long *sev_asid_bitmap;
 static unsigned long *sev_reclaim_asid_bitmap;
 
+static void **snp_asid_to_gctx_pages_map;
 static int snp_decommission_context(struct kvm *kvm);
 
 struct enc_region {
@@ -2248,6 +2249,9 @@ static int snp_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
 		goto e_free_context;
 	}
 
+	if (snp_asid_to_gctx_pages_map)
+		snp_asid_to_gctx_pages_map[sev_get_asid(kvm)] = sev->snp_context;
+
 	return 0;
 
 e_free_context:
@@ -2884,9 +2888,35 @@ static int snp_decommission_context(struct kvm *kvm)
 	snp_free_firmware_page(sev->snp_context);
 	sev->snp_context = NULL;
 
+	if (snp_asid_to_gctx_pages_map)
+		snp_asid_to_gctx_pages_map[sev_get_asid(kvm)] = NULL;
+
 	return 0;
 }
 
+/*
+ * NOTE: called in NMI context from sev_emergency_disable().
+ */
+void snp_decommision_all(void)
+{
+	struct sev_data_snp_addr data = {};
+	int ret, asid;
+
+	if (!snp_asid_to_gctx_pages_map)
+		return;
+
+	for (asid = 1; asid < min_sev_asid; asid++) {
+		if (snp_asid_to_gctx_pages_map[asid]) {
+			data.address = __sme_pa(snp_asid_to_gctx_pages_map[asid]);
+			ret = sev_do_cmd(SEV_CMD_SNP_DECOMMISSION, &data, NULL);
+			if (!ret) {
+				snp_free_firmware_page(snp_asid_to_gctx_pages_map[asid]);
+				snp_asid_to_gctx_pages_map[asid] = NULL;
+			}
+		}
+	}
+}
+
 void sev_vm_destroy(struct kvm *kvm)
 {
 	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
@@ -3052,6 +3082,13 @@ void __init sev_hardware_setup(void)
 	sev_es_supported = true;
 	sev_snp_supported = sev_snp_enabled && cc_platform_has(CC_ATTR_HOST_SEV_SNP);
 
+	if (sev_snp_supported) {
+		snp_asid_to_gctx_pages_map = kmalloc_array(min_sev_asid,
+							   sizeof(void *),
+							   GFP_KERNEL | __GFP_ZERO);
+		if (!snp_asid_to_gctx_pages_map)
+			pr_warn("Could not allocate SNP asid to guest context map\n");
+	}
 out:
 	if (boot_cpu_has(X86_FEATURE_SEV))
 		pr_info("SEV %s (ASIDs %u - %u)\n",
@@ -3094,6 +3131,8 @@ void sev_hardware_unsetup(void)
 
 	misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
 	misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
+
+	kfree(snp_asid_to_gctx_pages_map);
 }
 
 int sev_cpu_init(struct svm_cpu_data *sd)
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index e13c54d93964..a8f64a1710c2 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -17,6 +17,7 @@
 #include <linux/highmem.h>
 #include <linux/amd-iommu.h>
 #include <linux/sched.h>
+#include <linux/delay.h>
 #include <linux/trace_events.h>
 #include <linux/slab.h>
 #include <linux/hashtable.h>
@@ -248,6 +249,8 @@ static unsigned long iopm_base;
 
 DEFINE_PER_CPU(struct svm_cpu_data, svm_data);
 
+static DEFINE_SPINLOCK(snp_decommision_lock);
+
 /*
  * Only MSR_TSC_AUX is switched via the user return hook.  EFER is switched via
  * the VMCB, and the SYSCALL/SYSENTER MSRs are handled by VMLOAD/VMSAVE.
@@ -594,9 +597,97 @@ static inline void kvm_cpu_svm_disable(void)
 
 static void svm_emergency_disable(void)
 {
+	static atomic_t waiting_for_cpus_synchronized;
+	static bool synchronize_cpus_initiated;
+	static bool snp_decommision_handled;
+	static atomic_t cpus_synchronized;
+
 	kvm_rebooting = true;
 
 	kvm_cpu_svm_disable();
+
+	if (!cc_platform_has(CC_ATTR_HOST_SEV_SNP))
+		return;
+
+	/*
+	 * SNP_SHUTDOWN_EX fails when SNP VMs are active as the firmware checks
+	 * every encryption-capable ASID to verify that it is not in use by a
+	 * guest and a DF_FLUSH is not required. If a DF_FLUSH is required,
+	 * the firmware returns DFFLUSH_REQUIRED. To address this, SNP_DECOMMISION
+	 * is required to shutdown all active SNP VMs, but SNP_DECOMMISION tags all
+	 * CPUs that guest was activated on to do a WBINVD. When panic notifier
+	 * is invoked all other CPUs have already been shutdown, so it is not
+	 * possible to do a wbinvd_on_all_cpus() after SNP_DECOMMISION has been
+	 * executed. This eventually causes SNP_SHUTDOWN_EX to fail after
+	 * SNP_DECOMMISION. To fix this, do SNP_DECOMMISION and subsequent WBINVD
+	 * on all CPUs during NMI shutdown of CPUs as part of disabling
+	 * virtualization on all CPUs via cpu_emergency_disable_virtualization().
+	 */
+
+	spin_lock(&snp_decommision_lock);
+
+	/*
+	 * exit early for call from native_machine_crash_shutdown()
+	 * as SNP_DECOMMISSION has already been done as part of
+	 * NMI shutdown of the CPUs.
+	 */
+	if (snp_decommision_handled) {
+		spin_unlock(&snp_decommision_lock);
+		return;
+	}
+
+	/*
+	 * Synchronize all CPUs handling NMI before issuing
+	 * SNP_DECOMMISSION.
+	 */
+	if (!synchronize_cpus_initiated) {
+		/*
+		 * one CPU handling panic, the other CPU is initiator for
+		 * CPU synchronization.
+		 */
+		atomic_set(&waiting_for_cpus_synchronized, num_online_cpus() - 2);
+		synchronize_cpus_initiated = true;
+		/*
+		 * Ensure CPU synchronization parameters are setup before dropping
+		 * the lock to let other CPUs continue to reach synchronization.
+		 */
+		wmb();
+
+		spin_unlock(&snp_decommision_lock);
+
+		/*
+		 * This will not cause system to hang forever as the CPU
+		 * handling panic waits for maximum one second for
+		 * other CPUs to stop in nmi_shootdown_cpus().
+		 */
+		while (atomic_read(&waiting_for_cpus_synchronized) > 0)
+		       mdelay(1);
+
+		/* Reacquire the lock once CPUs are synchronized */
+		spin_lock(&snp_decommision_lock);
+
+		atomic_set(&cpus_synchronized, 1);
+	} else {
+		atomic_dec(&waiting_for_cpus_synchronized);
+		/*
+		 * drop the lock to let other CPUs contiune to reach
+		 * synchronization.
+		 */
+		spin_unlock(&snp_decommision_lock);
+
+		while (atomic_read(&cpus_synchronized) == 0)
+		       mdelay(1);
+
+		/* Try to re-acquire lock after CPUs are synchronized */
+		spin_lock(&snp_decommision_lock);
+	}
+
+	if (!snp_decommision_handled) {
+		snp_decommision_all();
+		snp_decommision_handled = true;
+	}
+	spin_unlock(&snp_decommision_lock);
+	wbinvd();
 }
 
 static void svm_hardware_disable(void)
diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
index 76107c7d0595..2f933b941b8d 100644
--- a/arch/x86/kvm/svm/svm.h
+++ b/arch/x86/kvm/svm/svm.h
@@ -749,6 +749,7 @@ void sev_snp_init_protected_guest_state(struct kvm_vcpu *vcpu);
 int sev_gmem_prepare(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn, int max_order);
 void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end);
 int sev_private_max_mapping_level(struct kvm *kvm, kvm_pfn_t pfn);
+void snp_decommision_all(void);
 #else
 static inline struct page *snp_safe_alloc_page_node(int node, gfp_t gfp)
 {
@@ -779,7 +780,7 @@ static inline int sev_private_max_mapping_level(struct kvm *kvm, kvm_pfn_t pfn)
 {
 	return 0;
 }
-
+static void snp_decommision_all(void);
 #endif
 
 /* vmenter.S */
-- 
2.34.1


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ