Re: [PATCH 10/13] x86/tdx: Convert shared memory back to private on kexec

From: Kalra, Ashish
Date: Thu Oct 05 2023 - 14:41:52 EST


Hello Kirill,

On 10/5/2023 8:13 AM, Kirill A. Shutemov wrote:
TDX guests allocate shared buffers to perform I/O. It is done by
allocating pages normally from the buddy allocator and converting them
to shared with set_memory_decrypted().

The target kernel has no idea what memory is converted this way. It only
sees E820_TYPE_RAM.

Accessing shared memory via private mapping is fatal. It leads to
unrecoverable TD exit.

On TD shutdown (also covers kexec), walk direct mapping and convert all
shared memory back to private. It makes all RAM private again and target
kernel may use it normally.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
---
arch/x86/Kconfig | 1 +
arch/x86/coco/tdx/kexec.c | 0
arch/x86/coco/tdx/tdx.c | 137 +++++++++++++++++++++++++++++++++++++-
3 files changed, 136 insertions(+), 2 deletions(-)
create mode 100644 arch/x86/coco/tdx/kexec.c

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 7368d254d01f..b5acf9fb4c70 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -884,6 +884,7 @@ config INTEL_TDX_GUEST
select X86_MEM_ENCRYPT
select X86_MCE
select UNACCEPTED_MEMORY
+ select EMERGENCY_VIRT_CALLBACK
help
Support running as a guest under Intel TDX. Without this support,
the guest kernel can not boot or run under TDX.
diff --git a/arch/x86/coco/tdx/kexec.c b/arch/x86/coco/tdx/kexec.c
new file mode 100644
index 000000000000..e69de29bb2d1
diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
index 56e152126f20..ac0745303983 100644
--- a/arch/x86/coco/tdx/tdx.c
+++ b/arch/x86/coco/tdx/tdx.c
@@ -6,6 +6,7 @@
#include <linux/cpufeature.h>
#include <linux/debugfs.h>
+#include <linux/delay.h>
#include <linux/export.h>
#include <linux/io.h>
#include <asm/coco.h>
@@ -14,6 +15,8 @@
#include <asm/insn.h>
#include <asm/insn-eval.h>
#include <asm/pgtable.h>
+#include <asm/reboot.h>
+#include <asm/set_memory.h>
/* MMIO direction */
#define EPT_READ 0
@@ -40,6 +43,9 @@
static atomic_long_t nr_shared;
+static atomic_t conversions_in_progress;
+static bool conversion_allowed = true;
+
static inline bool pte_decrypted(pte_t pte)
{
return cc_mkdec(pte_val(pte)) == pte_val(pte);
@@ -704,6 +710,14 @@ static bool tdx_tlb_flush_required(bool private)
static bool tdx_cache_flush_required(void)
{
+ /*
+ * Avoid issuing CLFLUSH on set_memory_decrypted() if conversions
+ * stopped. Otherwise it can race with unshare_all_memory() and trigger
+ * implicit conversion to shared.
+ */
+ if (!conversion_allowed)
+ return false;
+
/*
* AMD SME/SEV can avoid cache flushing if HW enforces cache coherence.
* TDX doesn't have such capability.
@@ -787,12 +801,25 @@ static bool tdx_enc_status_changed(unsigned long vaddr, int numpages, bool enc)
static int tdx_enc_status_change_prepare(unsigned long vaddr, int numpages,
bool enc)
{
+ atomic_inc(&conversions_in_progress);
+
+ /*
+ * Check after bumping conversions_in_progress to serialize
+ * against tdx_shutdown().
+ */
+ if (!conversion_allowed) {
+ atomic_dec(&conversions_in_progress);
+ return -EBUSY;
+ }
+
/*
* Only handle shared->private conversion here.
* See the comment in tdx_early_init().
*/
- if (enc && !tdx_enc_status_changed(vaddr, numpages, enc))
+ if (enc && !tdx_enc_status_changed(vaddr, numpages, enc)) {
+ atomic_dec(&conversions_in_progress);
return -EIO;
+ }
return 0;
}
@@ -804,17 +831,115 @@ static int tdx_enc_status_change_finish(unsigned long vaddr, int numpages,
* Only handle private->shared conversion here.
* See the comment in tdx_early_init().
*/
- if (!enc && !tdx_enc_status_changed(vaddr, numpages, enc))
+ if (!enc && !tdx_enc_status_changed(vaddr, numpages, enc)) {
+ atomic_dec(&conversions_in_progress);
return -EIO;
+ }
if (enc)
atomic_long_sub(numpages, &nr_shared);
else
atomic_long_add(numpages, &nr_shared);
+ atomic_dec(&conversions_in_progress);
+
return 0;
}
+static void unshare_all_memory(bool unmap)
+{
+ unsigned long addr, end;
+ long found = 0, shared;
+
+ /*
+ * Walk direct mapping and convert all shared memory back to private,
+ */
+
+ addr = PAGE_OFFSET;
+ end = PAGE_OFFSET + get_max_mapped();
+
+ while (addr < end) {
+ unsigned long size;
+ unsigned int level;
+ pte_t *pte;
+
+ pte = lookup_address(addr, &level);

IIRC, you were earlier walking the direct mapping using walk_page_range_novma(), any particular reason to use lookup_address() instead ?

+ size = page_level_size(level);
+
+ if (pte && pte_decrypted(*pte)) {

Additionally need to add check for pte_none() here to handle physical memory holes in direct mapping.

+ int pages = size / PAGE_SIZE;
+
+ /*
+ * Touching memory with shared bit set triggers implicit
+ * conversion to shared.
+ *
+ * Make sure nobody touches the shared range from
+ * now on.
+ *
+ * Bypass unmapping for crash scenario. Unmapping
+ * requires sleepable context, but in crash case kernel
+ * hits the code path with interrupts disabled.

In case of SNP we will need to temporarily enable interrupts during this unsharing as we invoke set_memory_encrypted() which then hits a BUG_ON() in cpa_flush() if interrupts are disabled.

Thanks,
Ashish

+ * It shouldn't be a problem as all secondary CPUs are
+ * down and kernel runs with interrupts disabled, so
+ * there is no room for race.
+ */
+ if (unmap)
+ set_memory_np(addr, pages);
+
+ if (!tdx_enc_status_changed(addr, pages, true)) {
+ pr_err("Failed to unshare range %#lx-%#lx\n",
+ addr, addr + size);
+ }
+
+ found += pages;
+ }
+
+ addr += size;
+ }
+
+ shared = atomic_long_read(&nr_shared);
+ if (shared != found) {
+ pr_err("shared page accounting is off\n");
+ pr_err("nr_shared = %ld, nr_found = %ld\n", shared, found);
+ }
+}
+
+static void tdx_shutdown(void)
+{
+ unsigned long timeout;
+
+ /*
+ * Stop new private<->shared conversions and wait for in-flight
+ * conversions to complete.
+ *
+ * Do not wait more than 30 seconds.
+ */
+ timeout = 30 * USEC_PER_SEC;
+ conversion_allowed = false;
+ while (atomic_read(&conversions_in_progress) && timeout--)
+ udelay(1);
+
+ if (!timeout)
+ pr_warn("Failed to finish shared<->private conversions\n");
+
+ unshare_all_memory(true);
+
+ native_machine_shutdown();
+}
+
+static void tdx_crash_shutdown(void)
+{
+ /*
+ * Crash can race with private<->shared conversion.
+ *
+ * There's no clean way out: report and proceed.
+ */
+ if (atomic_read(&conversions_in_progress))
+ pr_warn("Failed to finish shared<->private conversions\n");
+
+ unshare_all_memory(false);
+}
+
void __init tdx_early_init(void)
{
struct tdx_module_args args = {
@@ -882,6 +1007,14 @@ void __init tdx_early_init(void)
*/
x86_cpuinit.parallel_bringup = false;
+ machine_ops.shutdown = tdx_shutdown;
+
+ /*
+ * KVM overrides machine_ops.crash_shutdown, use emergency
+ * virt callback instead.
+ */
+ cpu_emergency_register_virt_callback(tdx_crash_shutdown);
+
pr_info("Guest detected\n");
}