Attachment 875905 Details for Bug 1227301

[patch] Debug patch V2

v2-0001-xen-make-multicall-debug-boot-time-selectable.patch (text/plain), 7.21 KB, created by Jürgen Groß on 2024-07-05 08:10:43 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Jürgen Groß

Created: 2024-07-05 08:10:43 UTC

Size: 7.21 KB

patch

obsolete

>From e259eab2b465bb4490cf286f3147b340ce35def9 Mon Sep 17 00:00:00 2001
>From: Juergen Gross <jgross@suse.com>
>To: x86@kernel.org
>To: linux-doc@vger.kernel.org
>To: linux-kernel@vger.kernel.org
>Cc: Jonathan Corbet <corbet@lwn.net>
>Cc: Juergen Gross <jgross@suse.com>
>Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
>Cc: Thomas Gleixner <tglx@linutronix.de>
>Cc: Ingo Molnar <mingo@redhat.com>
>Cc: Borislav Petkov <bp@alien8.de>
>Cc: Dave Hansen <dave.hansen@linux.intel.com>
>Cc: "H. Peter Anvin" <hpa@zytor.com>
>Cc: xen-devel@lists.xenproject.org
>Date: Wed, 3 Jul 2024 08:36:56 +0200
>Subject: [PATCH v2] xen: make multicall debug boot time selectable
>
>Today Xen multicall debugging needs to be enabled via modifying a
>define in a source file for getting debug data of multicall errors
>encountered by users.
>
>Switch multicall debugging to depend on a boot parameter "xen_mc_debug"
>instead, enabling affected users to boot with the new parameter set in
>order to get better diagnostics.
>
>With debugging enabled print the following information in case at least
>one of the batched calls failed:
>- all calls of the batch with operation, result and caller
>- all parameters of each call
>- all parameters stored in the multicall data for each call
>
>Signed-off-by: Juergen Gross <jgross@suse.com>
>---
>V2:
>- fixed argument printing, added parameter printing
>- in debug case print entries without error, too
>---
> .../admin-guide/kernel-parameters.txt         |   6 +
> arch/x86/xen/multicalls.c                     | 131 +++++++++++++++---
> 2 files changed, 114 insertions(+), 23 deletions(-)
>
>diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
>index 27ec49af1bf2..b33d048e01d8 100644
>--- a/Documentation/admin-guide/kernel-parameters.txt
>+++ b/Documentation/admin-guide/kernel-parameters.txt
>@@ -7427,6 +7427,12 @@
> 			Crash from Xen panic notifier, without executing late
> 			panic() code such as dumping handler.
> 
>+	xen_mc_debug	[X86,XEN,EARLY]
>+			Enable multicall debugging when running as a Xen PV guest.
>+			Enabling this feature will reduce performance a little
>+			bit, so it should only be enabled for obtaining extended
>+			debug data in case of multicall errors.
>+
> 	xen_msr_safe=	[X86,XEN,EARLY]
> 			Format: <bool>
> 			Select whether to always use non-faulting (safe) MSR
>diff --git a/arch/x86/xen/multicalls.c b/arch/x86/xen/multicalls.c
>index 07054572297f..4e6d53a8408c 100644
>--- a/arch/x86/xen/multicalls.c
>+++ b/arch/x86/xen/multicalls.c
>@@ -23,6 +23,8 @@
> #include <linux/percpu.h>
> #include <linux/hardirq.h>
> #include <linux/debugfs.h>
>+#include <linux/jump_label.h>
>+#include <linux/printk.h>
> 
> #include <asm/xen/hypercall.h>
> 
>@@ -31,18 +33,12 @@
> 
> #define MC_BATCH	32
> 
>-#define MC_DEBUG	0
>-
> #define MC_ARGS		(MC_BATCH * 16)
> 
> 
> struct mc_buffer {
> 	unsigned mcidx, argidx, cbidx;
> 	struct multicall_entry entries[MC_BATCH];
>-#if MC_DEBUG
>-	struct multicall_entry debug[MC_BATCH];
>-	void *caller[MC_BATCH];
>-#endif
> 	unsigned char args[MC_ARGS];
> 	struct callback {
> 		void (*fn)(void *);
>@@ -50,13 +46,104 @@ struct mc_buffer {
> 	} callbacks[MC_BATCH];
> };
> 
>+struct mc_debug_data {
>+	struct multicall_entry debug[MC_BATCH];
>+	void *caller[MC_BATCH];
>+	size_t argsz[MC_BATCH];
>+	unsigned long *args[MC_BATCH];
>+};
>+
> static DEFINE_PER_CPU(struct mc_buffer, mc_buffer);
>+static struct mc_debug_data __percpu *mc_debug_data;
>+static struct mc_debug_data mc_debug_data_early __initdata;
> DEFINE_PER_CPU(unsigned long, xen_mc_irq_flags);
> 
>+static struct static_key mc_debug __ro_after_init;
>+static bool mc_debug_enabled __initdata;
>+
>+static int __init xen_parse_mc_debug(char *arg)
>+{
>+	mc_debug_enabled = true;
>+	static_key_slow_inc(&mc_debug);
>+
>+	return 0;
>+}
>+early_param("xen_mc_debug", xen_parse_mc_debug);
>+
>+static int __init mc_debug_enable(void)
>+{
>+	struct mc_debug_data __percpu *mcdb;
>+	unsigned long flags;
>+
>+	if (!mc_debug_enabled)
>+		return 0;
>+
>+	mcdb = alloc_percpu(struct mc_debug_data);
>+	if (!mcdb) {
>+		pr_err("xen_mc_debug inactive\n");
>+		static_key_slow_dec(&mc_debug);
>+		return -ENOMEM;
>+	}
>+
>+	/* Be careful when switching to percpu debug data. */
>+	local_irq_save(flags);
>+	xen_mc_flush();
>+	mc_debug_data = mcdb;
>+	local_irq_restore(flags);
>+
>+	pr_info("xen_mc_debug active\n");
>+
>+	return 0;
>+}
>+early_initcall(mc_debug_enable);
>+
>+/* Number of parameters of hypercalls used via multicalls. */
>+static const uint8_t hpcpars[] = {
>+	[__HYPERVISOR_mmu_update] = 4,
>+	[__HYPERVISOR_stack_switch] = 2,
>+	[__HYPERVISOR_fpu_taskswitch] = 1,
>+	[__HYPERVISOR_update_descriptor] = 2,
>+	[__HYPERVISOR_update_va_mapping] = 3,
>+	[__HYPERVISOR_mmuext_op] = 4,
>+};
>+
>+static void print_debug_data(struct mc_buffer *b, struct mc_debug_data *mcdb,
>+			     int idx)
>+{
>+	unsigned int arg;
>+	unsigned int opidx = mcdb->debug[idx].op & 0xff;
>+	unsigned int pars = 0;
>+
>+	pr_err("  call %2d: op=%lu result=%ld caller=%pS ", idx + 1,
>+	       mcdb->debug[idx].op, b->entries[idx].result, mcdb->caller[idx]);
>+	if (opidx < ARRAY_SIZE(hpcpars))
>+		pars = hpcpars[opidx];
>+	if (pars) {
>+		pr_cont("pars=");
>+		for (arg = 0; arg < pars; arg++)
>+			pr_cont("%lx ", mcdb->debug[idx].args[arg]);
>+	}
>+	if (mcdb->argsz[idx]) {
>+		pr_cont("args=");
>+		for (arg = 0; arg < mcdb->argsz[idx] / 8; arg++)
>+			pr_cont("%lx ", mcdb->args[idx][arg]);
>+	}
>+	pr_cont("\n");
>+}
>+
>+static struct mc_debug_data * __ref get_mc_debug_ptr(void)
>+{
>+	if (likely(mc_debug_data))
>+		return this_cpu_ptr(mc_debug_data);
>+
>+	return &mc_debug_data_early;
>+}
>+
> void xen_mc_flush(void)
> {
> 	struct mc_buffer *b = this_cpu_ptr(&mc_buffer);
> 	struct multicall_entry *mc;
>+	struct mc_debug_data *mcdb = NULL;
> 	int ret = 0;
> 	unsigned long flags;
> 	int i;
>@@ -69,10 +156,11 @@ void xen_mc_flush(void)
> 
> 	trace_xen_mc_flush(b->mcidx, b->argidx, b->cbidx);
> 
>-#if MC_DEBUG
>-	memcpy(b->debug, b->entries,
>-	       b->mcidx * sizeof(struct multicall_entry));
>-#endif
>+	if (static_key_false(&mc_debug)) {
>+		mcdb = get_mc_debug_ptr();
>+		memcpy(mcdb->debug, b->entries,
>+		       b->mcidx * sizeof(struct multicall_entry));
>+	}
> 
> 	switch (b->mcidx) {
> 	case 0:
>@@ -103,21 +191,14 @@ void xen_mc_flush(void)
> 		pr_err("%d of %d multicall(s) failed: cpu %d\n",
> 		       ret, b->mcidx, smp_processor_id());
> 		for (i = 0; i < b->mcidx; i++) {
>-			if (b->entries[i].result < 0) {
>-#if MC_DEBUG
>-				pr_err("  call %2d: op=%lu arg=[%lx] result=%ld\t%pS\n",
>-				       i + 1,
>-				       b->debug[i].op,
>-				       b->debug[i].args[0],
>-				       b->entries[i].result,
>-				       b->caller[i]);
>-#else
>+			if (static_key_false(&mc_debug)) {
>+				print_debug_data(b, mcdb, i);
>+			} else if (b->entries[i].result < 0) {
> 				pr_err("  call %2d: op=%lu arg=[%lx] result=%ld\n",
> 				       i + 1,
> 				       b->entries[i].op,
> 				       b->entries[i].args[0],
> 				       b->entries[i].result);
>-#endif
> 			}
> 		}
> 	}
>@@ -155,9 +236,13 @@ struct multicall_space __xen_mc_entry(size_t args)
> 	}
> 
> 	ret.mc = &b->entries[b->mcidx];
>-#if MC_DEBUG
>-	b->caller[b->mcidx] = __builtin_return_address(0);
>-#endif
>+	if (static_key_false(&mc_debug)) {
>+		struct mc_debug_data *mcdb = get_mc_debug_ptr();
>+
>+		mcdb->caller[b->mcidx] = __builtin_return_address(0);
>+		mcdb->argsz[b->mcidx] = args;
>+		mcdb->args[b->mcidx] = (unsigned long *)(&b->args[argidx]);
>+	}
> 	b->mcidx++;
> 	ret.args = &b->args[argidx];
> 	b->argidx = argidx + args;
>-- 
>2.35.3
>

Actions: View | Diff

Attachments on bug 1227301: 875832 | 875833 | 875846 | 875854 | 875905 | 875916 | 875936 | 875954 | 875955