Attachment 247874 Details for Bug 276421 – Patch v3

[patch] Patch v3

fix.patch (text/plain), 13.56 KB, created by Roger Pau Monné on 2024-01-23 09:12:32 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Roger Pau Monné

Created: 2024-01-23 09:12:32 UTC

Size: 13.56 KB

patch

obsolete

>diff --git a/sys/dev/xen/console/xen_console.c b/sys/dev/xen/console/xen_console.c
>index d232dd4e0c1c..934fa32915bb 100644
>--- a/sys/dev/xen/console/xen_console.c
>+++ b/sys/dev/xen/console/xen_console.c
>@@ -138,6 +138,12 @@ struct putchar_arg {
> 	size_t	n_next;
> };
> 
>+void __weak_symbol
>+xen_emergency_print(const char *str, size_t size)
>+{
>+    HYPERVISOR_console_write(str, size);
>+}
>+
> static void
> putchar(int c, void *arg)
> {
>@@ -150,12 +156,12 @@ putchar(int c, void *arg)
> 		 * We have no buffer, output directly to the
> 		 * console char by char.
> 		 */
>-		HYPERVISOR_console_write((char *)&c, 1);
>+		xen_emergency_print((char *)&c, 1);
> 	} else {
> 		pca->buf[pca->n_next++] = c;
> 		if ((pca->size == pca->n_next) || (c = '\0')) {
> 			/* Flush the buffer */
>-			HYPERVISOR_console_write(pca->buf, pca->n_next);
>+			xen_emergency_print(pca->buf, pca->n_next);
> 			pca->n_next = 0;
> 		}
> 	}
>@@ -185,10 +191,17 @@ xc_printf(const char *fmt, ...)
> 
> #ifdef PRINTF_BUFR_SIZE
> 	if (pca.n_next != 0)
>-		HYPERVISOR_console_write(buf, pca.n_next);
>+		xen_emergency_print(buf, pca.n_next);
> #endif
> }
> 
>+#ifdef EARLY_PRINTF
>+void xen_early_putc(int ch)
>+{
>+	xen_emergency_print((char *)&ch, 1);
>+}
>+#endif
>+
> /*---------------------- Helpers for the console lock -----------------------*/
> /*
>  * The lock is not used when the kernel is panicing as it will never recover
>diff --git a/sys/x86/include/xen/xen-os.h b/sys/x86/include/xen/xen-os.h
>index 12942375be43..d3f21e2a6c45 100644
>--- a/sys/x86/include/xen/xen-os.h
>+++ b/sys/x86/include/xen/xen-os.h
>@@ -52,8 +52,6 @@ extern int xen_disable_pv_disks;
> /* tunable for disabling PV nics */
> extern int xen_disable_pv_nics;
> 
>-extern uint32_t xen_cpuid_base;
>-
> /* compatibility for accessing xen_ulong_t with atomics */
> #define	atomic_clear_xen_ulong		atomic_clear_long
> #define	atomic_set_xen_ulong		atomic_set_long
>@@ -96,6 +94,9 @@ xen_pv_nics_disabled(void)
> 
> bool xen_has_iommu_maps(void);
> 
>+/* (Very) early initialization. */
>+void xen_early_init(void);
>+
> #endif /* !__ASSEMBLY__ */
> 
> #endif /* _MACHINE_X86_XEN_XEN_OS_H_ */
>diff --git a/sys/x86/x86/identcpu.c b/sys/x86/x86/identcpu.c
>index 9c82f9f8a882..221973abdfb9 100644
>--- a/sys/x86/x86/identcpu.c
>+++ b/sys/x86/x86/identcpu.c
>@@ -67,6 +67,8 @@
> #include <x86/isa/icu.h>
> #include <x86/vmware.h>
> 
>+#include <xen/xen-os.h>
>+
> #ifdef __i386__
> #define	IDENTBLUE_CYRIX486	0
> #define	IDENTBLUE_IBMCPU	1
>@@ -1343,8 +1345,13 @@ SYSINIT(hook_tsc_freq, SI_SUB_CONFIGURE, SI_ORDER_ANY, hook_tsc_freq, NULL);
> static struct {
> 	const char	*vm_cpuid;
> 	int		vm_guest;
>+	void		(*init)(void);
> } vm_cpuids[] = {
>-	{ "XenVMMXenVMM",	VM_GUEST_XEN },		/* XEN */
>+	{ "XenVMMXenVMM",	VM_GUEST_XEN,
>+#ifdef XENHVM
>+	  &xen_early_init,
>+#endif
>+	}, 						/* XEN */
> 	{ "Microsoft Hv",	VM_GUEST_HV },		/* Microsoft Hyper-V */
> 	{ "VMwareVMware",	VM_GUEST_VMWARE },	/* VMware VM */
> 	{ "KVMKVMKVM",		VM_GUEST_KVM },		/* KVM */
>@@ -1355,6 +1362,7 @@ static struct {
> static void
> identify_hypervisor_cpuid_base(void)
> {
>+	void (*init_fn)(void);
> 	u_int leaf, regs[4];
> 	int i;
> 
>@@ -1385,10 +1393,13 @@ identify_hypervisor_cpuid_base(void)
> 			regs[0] = leaf + 1;
> 
> 		if (regs[0] >= leaf) {
>+			enum VM_GUEST prev_vm_guest = vm_guest;
>+
> 			for (i = 0; i < nitems(vm_cpuids); i++)
> 				if (strncmp((const char *)&regs[1],
> 				    vm_cpuids[i].vm_cpuid, 12) == 0) {
> 					vm_guest = vm_cpuids[i].vm_guest;
>+					init_fn = vm_cpuids[i].init;
> 					break;
> 				}
> 
>@@ -1397,7 +1408,7 @@ identify_hypervisor_cpuid_base(void)
> 			 * specific hypervisor, record the base, high value,
> 			 * and vendor identifier.
> 			 */
>-			if (vm_guest != VM_GUEST_VM || leaf == 0x40000000) {
>+			if (vm_guest != prev_vm_guest || leaf == 0x40000000) {
> 				hv_base = leaf;
> 				hv_high = regs[0];
> 				((u_int *)&hv_vendor)[0] = regs[1];
>@@ -1409,11 +1420,25 @@ identify_hypervisor_cpuid_base(void)
> 				 * If we found a specific hypervisor, then
> 				 * we are finished.
> 				 */
>-				if (vm_guest != VM_GUEST_VM)
>-					return;
>+				if (vm_guest != VM_GUEST_VM &&
>+				    /*
>+				     * Xen and other hypervisors can expose the
>+				     * HyperV signature in addition to the
>+				     * native one in order to support Viridian
>+				     * extensions for Windows guests.
>+				     *
>+				     * Do the full cpuid scan if HyperV is
>+				     * detected, as the native hypervisor is
>+				     * preferred.
>+				     */
>+				    vm_guest != VM_GUEST_HV)
>+					break;
> 			}
> 		}
> 	}
>+
>+	if (init_fn != NULL)
>+		init_fn();
> }
> 
> void
>diff --git a/sys/x86/xen/hvm.c b/sys/x86/xen/hvm.c
>index b8125e230b6f..e78f10b68168 100644
>--- a/sys/x86/xen/hvm.c
>+++ b/sys/x86/xen/hvm.c
>@@ -41,8 +41,10 @@
> 
> #include <dev/pci/pcivar.h>
> 
>+#include <machine/_inttypes.h>
> #include <machine/cpufunc.h>
> #include <machine/cpu.h>
>+#include <machine/md_var.h>
> #include <machine/smp.h>
> 
> #include <x86/apicreg.h>
>@@ -95,6 +97,14 @@ TUNABLE_INT("hw.xen.disable_pv_nics", &xen_disable_pv_nics);
> 
> /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/
> 
>+void xen_emergency_print(const char *str, size_t size)
>+{
>+    size_t i;
>+
>+    for (i = 0; i < size; i++)
>+	outb(XEN_HVM_DEBUGCONS_IOPORT, str[i]);
>+}
>+
> uint32_t xen_cpuid_base;
> 
> static uint32_t
>@@ -138,7 +148,7 @@ hypervisor_version(void)
> 	uint32_t regs[4];
> 	int major, minor;
> 
>-	do_cpuid(xen_cpuid_base + 1, regs);
>+	do_cpuid(hv_base + 1, regs);
> 
> 	major = regs[0] >> 16;
> 	minor = regs[0] & 0xffff;
>@@ -148,62 +158,88 @@ hypervisor_version(void)
> }
> 
> /*
>- * Allocate and fill in the hypcall page.
>+ * Translate linear to physical address when still running on the bootloader
>+ * created page-tables.
>  */
>-int
>-xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type)
>+static vm_paddr_t __nosanitizeaddress __nosanitizememory
>+early_init_vtop(void *addr)
> {
>-	uint32_t regs[4];
>-
>-	if (xen_cpuid_base != 0)
>-		/* Already setup. */
>-		goto out;
>-
>-	xen_cpuid_base = xen_hvm_cpuid_base();
>-	if (xen_cpuid_base == 0)
>-		return (ENXIO);
>+	return ((uintptr_t)addr - KERNBASE
>+#ifdef __amd64__
>+	    + kernphys - KERNLOAD
>+#endif
>+	    );
>+}
> 
>+static int
>+map_shared_info(void)
>+{
> 	/*
>-	 * Find the hypercall pages.
>+	 * TODO shared info page should be mapped in an unpopulated (IOW:
>+	 * non-RAM) address.  But finding one at this point in boot is
>+	 * complicated, hence re-use a RAM address for the time being.  This
>+	 * sadly causes super-page shattering in the second stage translation
>+	 * page tables.
> 	 */
>-	do_cpuid(xen_cpuid_base + 2, regs);
>-	if (regs[0] != 1)
>-		return (EINVAL);
>+	static shared_info_t shared_page __attribute__((aligned(PAGE_SIZE)));
>+	static struct xen_add_to_physmap xatp = {
>+	    .domid = DOMID_SELF,
>+	    .idx = 0,
>+	    .space = XENMAPSPACE_shared_info,
>+	};
>+	int rc;
> 
>-	wrmsr(regs[1], (init_type == XEN_HVM_INIT_EARLY)
>-	    ? (vm_paddr_t)((uintptr_t)&hypercall_page - KERNBASE)
>-	    : vtophys(&hypercall_page));
>+	if (xatp.gpfn == 0)
>+	    xatp.gpfn = atop(early_init_vtop(&shared_page));
> 
>-out:
>-	hypervisor_version();
>-	return (0);
>+	rc = HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp);
>+	if (rc != 0) {
>+		printf("cannot map shared info page: %d\n", rc);
>+		HYPERVISOR_shared_info = NULL;
>+	} else if (HYPERVISOR_shared_info == NULL)
>+		HYPERVISOR_shared_info = &shared_page;
>+
>+	return (rc);
> }
> 
>-static void
>-xen_hvm_init_shared_info_page(void)
>+/*
>+ * Populate the hypecall and shared pages.
>+ */
>+void
>+xen_early_init(void)
> {
>-	struct xen_add_to_physmap xatp;
>+	uint32_t regs[4];
>+	int rc;
> 
>-	if (xen_pv_domain()) {
>-		/*
>-		 * Already setup in the PV case, shared_info is passed inside
>-		 * of the start_info struct at start of day.
>-		 */
>+#ifdef EARLY_PRINTF
>+	early_putc = &xen_early_putc;
>+#endif
>+
>+	if (hv_high < 2) {
>+		xc_printf("Invalid maximum leaves for hv_base\n");
>+		vm_guest = VM_GUEST_VM;
> 		return;
> 	}
> 
>-	if (HYPERVISOR_shared_info == NULL) {
>-		HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT);
>-		if (HYPERVISOR_shared_info == NULL)
>-			panic("Unable to allocate Xen shared info page");
>+	/*
>+	 * Find the hypercall pages.
>+	 */
>+	do_cpuid(hv_base + 2, regs);
>+	if (regs[0] != 1) {
>+		xc_printf("Invalid number of hypercall pages %" PRIu32 "\n",
>+		    regs[0]);
>+		vm_guest = VM_GUEST_VM;
>+		return;
> 	}
> 
>-	xatp.domid = DOMID_SELF;
>-	xatp.idx = 0;
>-	xatp.space = XENMAPSPACE_shared_info;
>-	xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT;
>-	if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
>-		panic("HYPERVISOR_memory_op failed");
>+	wrmsr(regs[1], early_init_vtop(&hypercall_page));
>+
>+	rc = map_shared_info();
>+	if (rc != 0) {
>+		xc_printf("cannot map shared info page: %d\n", rc);
>+		vm_guest = VM_GUEST_VM;
>+		return;
>+	}
> }
> 
> static int
>@@ -322,32 +358,29 @@ xen_hvm_disable_emulated_devices(void)
> static void
> xen_hvm_init(enum xen_hvm_init_type init_type)
> {
>-	int error;
> 	int i;
> 
> 	if (!xen_domain() ||
> 	    init_type == XEN_HVM_INIT_CANCELLED_SUSPEND)
> 		return;
> 
>-	error = xen_hvm_init_hypercall_stubs(init_type);
>+	hypervisor_version();
> 
> 	switch (init_type) {
> 	case XEN_HVM_INIT_LATE:
>-		if (error != 0)
>-			return;
>-
> 		setup_xen_features();
> #ifdef SMP
> 		cpu_ops = xen_hvm_cpu_ops;
> #endif
> 		break;
> 	case XEN_HVM_INIT_RESUME:
>-		if (error != 0)
>-			panic("Unable to init Xen hypercall stubs on resume");
>-
> 		/* Clear stale vcpu_info. */
> 		CPU_FOREACH(i)
> 			DPCPU_ID_SET(i, vcpu_info, NULL);
>+
>+		if (map_shared_info() != 0)
>+			panic("cannot map shared info page\n");
>+
> 		break;
> 	default:
> 		panic("Unsupported HVM initialization type");
>@@ -357,13 +390,6 @@ xen_hvm_init(enum xen_hvm_init_type init_type)
> 	xen_evtchn_needs_ack = false;
> 	xen_hvm_set_callback(NULL);
> 
>-	/*
>-	 * On (PV)HVM domains we need to request the hypervisor to
>-	 * fill the shared info page, for PVH guest the shared_info page
>-	 * is passed inside the start_info struct and is already set, so this
>-	 * functions are no-ops.
>-	 */
>-	xen_hvm_init_shared_info_page();
> 	xen_hvm_disable_emulated_devices();
> } 
> 
>@@ -412,8 +438,8 @@ xen_hvm_cpu_init(void)
> 	 * Set vCPU ID. If available fetch the ID from CPUID, if not just use
> 	 * the ACPI ID.
> 	 */
>-	KASSERT(xen_cpuid_base != 0, ("Invalid base Xen CPUID leaf"));
>-	cpuid_count(xen_cpuid_base + 4, 0, regs);
>+	KASSERT(hv_base != 0, ("Invalid base Xen CPUID leaf"));
>+	cpuid_count(hv_base + 4, 0, regs);
> 	KASSERT((regs[0] & XEN_HVM_CPUID_VCPU_ID_PRESENT) ||
> 	    !xen_pv_domain(),
> 	    ("Xen PV domain without vcpu_id in cpuid"));
>@@ -443,8 +469,8 @@ xen_has_iommu_maps(void)
> {
> 	uint32_t regs[4];
> 
>-	KASSERT(xen_cpuid_base != 0, ("Invalid base Xen CPUID leaf"));
>-	cpuid_count(xen_cpuid_base + 4, 0, regs);
>+	KASSERT(hv_base != 0, ("Invalid base Xen CPUID leaf"));
>+	cpuid_count(hv_base + 4, 0, regs);
> 
> 	return (regs[0] & XEN_HVM_CPUID_IOMMU_MAPPINGS);
> }
>diff --git a/sys/x86/xen/pv.c b/sys/x86/xen/pv.c
>index 9ea276caee81..7c41332d5b39 100644
>--- a/sys/x86/xen/pv.c
>+++ b/sys/x86/xen/pv.c
>@@ -159,20 +159,8 @@ uint64_t
> hammer_time_xen(vm_paddr_t start_info_paddr)
> {
> 	struct hvm_modlist_entry *mod;
>-	struct xen_add_to_physmap xatp;
> 	uint64_t physfree;
> 	char *kenv;
>-	int rc;
>-
>-	if (isxen()) {
>-		vm_guest = VM_GUEST_XEN;
>-		rc = xen_hvm_init_hypercall_stubs(XEN_HVM_INIT_EARLY);
>-		if (rc) {
>-			xc_printf("ERROR: failed to initialize hypercall page: %d\n",
>-			    rc);
>-			HYPERVISOR_shutdown(SHUTDOWN_crash);
>-		}
>-	}
> 
> 	start_info = (struct hvm_start_info *)(start_info_paddr + KERNBASE);
> 	if (start_info->magic != XEN_HVM_START_MAGIC_VALUE) {
>@@ -211,19 +199,6 @@ hammer_time_xen(vm_paddr_t start_info_paddr)
> 			    PAGE_SIZE), physfree);
> 	}
> 
>-	if (isxen()) {
>-		xatp.domid = DOMID_SELF;
>-		xatp.idx = 0;
>-		xatp.space = XENMAPSPACE_shared_info;
>-		xatp.gpfn = atop(physfree);
>-		if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) {
>-			xc_printf("ERROR: failed to setup shared_info page\n");
>-			HYPERVISOR_shutdown(SHUTDOWN_crash);
>-		}
>-		HYPERVISOR_shared_info = (shared_info_t *)(physfree + KERNBASE);
>-		physfree += PAGE_SIZE;
>-	}
>-
> 	/*
> 	 * Init a static kenv using a free page. The contents will be filled
> 	 * from the parse_preload_data hook.
>diff --git a/sys/x86/xen/xen_apic.c b/sys/x86/xen/xen_apic.c
>index 4f2b03879227..994dc3e0804c 100644
>--- a/sys/x86/xen/xen_apic.c
>+++ b/sys/x86/xen/xen_apic.c
>@@ -343,8 +343,8 @@ xen_setup_cpus(void)
> 	 * that's faster than using event channels because it avoids the VM
> 	 * exit.
> 	 */
>-	KASSERT(xen_cpuid_base != 0, ("Invalid base Xen CPUID leaf"));
>-	cpuid_count(xen_cpuid_base + 4, 0, regs);
>+	KASSERT(hv_base != 0, ("Invalid base Xen CPUID leaf"));
>+	cpuid_count(hv_base + 4, 0, regs);
> 	if ((x2apic_mode && (regs[0] & XEN_HVM_CPUID_X2APIC_VIRT)) ||
> 	    (!x2apic_mode && (regs[0] & XEN_HVM_CPUID_APIC_ACCESS_VIRT)))
> 		return;
>diff --git a/sys/xen/hvm.h b/sys/xen/hvm.h
>index 3ff3ee3c51e6..f6d4292fd785 100644
>--- a/sys/xen/hvm.h
>+++ b/sys/xen/hvm.h
>@@ -96,7 +96,6 @@ enum xen_hvm_init_type {
> 	XEN_HVM_INIT_RESUME,
> };
> 
>-int xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type);
> void xen_hvm_set_callback(device_t);
> void xen_hvm_suspend(void);
> void xen_hvm_resume(bool suspend_cancelled);
>diff --git a/sys/xen/xen-os.h b/sys/xen/xen-os.h
>index d7e9b3ca8632..b1465b112413 100644
>--- a/sys/xen/xen-os.h
>+++ b/sys/xen/xen-os.h
>@@ -151,6 +151,19 @@ int xenmem_free(device_t dev, int res_id, struct resource *res);
> /* Debug/emergency function, prints directly to hypervisor console */
> void xc_printf(const char *, ...) __printflike(1, 2);
> 
>+/*
>+ * Emergency print function, can be defined per-arch, otherwise defaults to
>+ * HYPERVISOR_console_write.  Should not be called directly, use xc_printf
>+ * instead.
>+ */
>+void xen_emergency_print(const char *str, size_t size);
>+
>+/*
>+ * Early printk implementation based on xen_emergency_print.  Only defined if
>+ * EARLY_PRINTF is set.
>+ */
>+void xen_early_putc(int ch);
>+
> #ifndef xen_mb
> #define xen_mb() mb()
> #endif

Actions: View | Diff

Attachments on bug 276421: 247741 | 247748 | 247749 | 247750 | 247769 | 247784 | 247874