From patchwork Thu Sep 26 13:53:32 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chao Gao X-Patchwork-Id: 11162799 Return-Path: Received: from mail.kernel.org (pdx-korg-mail-1.web.codeaurora.org [172.30.200.123]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 3F53417D4 for ; Thu, 26 Sep 2019 13:51:28 +0000 (UTC) Received: from lists.xenproject.org (lists.xenproject.org [192.237.175.120]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id 198EA207FF for ; Thu, 26 Sep 2019 13:51:28 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 198EA207FF Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=intel.com Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=xen-devel-bounces@lists.xenproject.org Received: from localhost ([127.0.0.1] helo=lists.xenproject.org) by lists.xenproject.org with esmtp (Exim 4.89) (envelope-from ) id 1iDU9h-0002Bb-R0; Thu, 26 Sep 2019 13:50:01 +0000 Received: from us1-rack-iad1.inumbo.com ([172.99.69.81]) by lists.xenproject.org with esmtp (Exim 4.89) (envelope-from ) id 1iDU9g-00027m-8T for xen-devel@lists.xenproject.org; Thu, 26 Sep 2019 13:50:00 +0000 X-Inumbo-ID: 80e4a5a6-e064-11e9-8628-bc764e2007e4 Received: from mga04.intel.com (unknown [192.55.52.120]) by localhost (Halon) with ESMTPS id 80e4a5a6-e064-11e9-8628-bc764e2007e4; Thu, 26 Sep 2019 13:49:47 +0000 (UTC) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga008.fm.intel.com ([10.253.24.58]) by fmsmga104.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 26 Sep 2019 06:49:47 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.64,552,1559545200"; d="scan'208";a="189126045" Received: from gao-cwp.sh.intel.com ([10.239.159.26]) by fmsmga008.fm.intel.com with ESMTP; 26 Sep 2019 06:49:44 -0700 From: Chao Gao To: xen-devel@lists.xenproject.org Date: Thu, 26 Sep 2019 21:53:32 +0800 Message-Id: <1569506015-26938-5-git-send-email-chao.gao@intel.com> X-Mailer: git-send-email 1.9.1 In-Reply-To: <1569506015-26938-1-git-send-email-chao.gao@intel.com> References: <1569506015-26938-1-git-send-email-chao.gao@intel.com> Subject: [Xen-devel] [PATCH v11 4/7] x86/microcode: Synchronize late microcode loading X-BeenThere: xen-devel@lists.xenproject.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: Xen developer discussion List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Cc: Sergey Dyasli , Kevin Tian , Borislav Petkov , Ashok Raj , Wei Liu , Jun Nakajima , Andrew Cooper , Jan Beulich , Thomas Gleixner , Chao Gao , =?utf-8?q?Roger_Pau_Monn=C3=A9?= MIME-Version: 1.0 Errors-To: xen-devel-bounces@lists.xenproject.org Sender: "Xen-devel" This patch ports microcode improvement patches from linux kernel. Before you read any further: the early loading method is still the preferred one and you should always do that. The following patch is improving the late loading mechanism for long running jobs and cloud use cases. Gather all cores and serialize the microcode update on them by doing it one-by-one to make the late update process as reliable as possible and avoid potential issues caused by the microcode update. Signed-off-by: Chao Gao Tested-by: Chao Gao [linux commit: a5321aec6412b20b5ad15db2d6b916c05349dbff] [linux commit: bb8c13d61a629276a162c1d2b1a20a815cbcfbb7] Cc: Kevin Tian Cc: Jun Nakajima Cc: Ashok Raj Cc: Borislav Petkov Cc: Thomas Gleixner Cc: Andrew Cooper Cc: Jan Beulich Reviewed-by: Jan Beulich --- Changes in v11: - Use the sample code of wait_for_state() provided by Jan - make wait_cpu_call{in,out} take unsigned int to avoid type casting - do assignment in while clause in control_thread_fn() to eliminate duplication. Changes in v10: - introduce wait_for_state() and set_state() helper functions - make wait_for_condition() return bool and take const void * - disable/enable watchdog in control thread - rename "master" and "slave" thread to "primary" and "secondary" Changes in v9: - log __buildin_return_address(0) when timeout - divide CPUs into three logical sets and they will call different functions during ucode loading. The 'control thread' is chosen to coordinate ucode loading on all CPUs. Since only control thread would set 'loading_state', we can get rid of 'cmpxchg' stuff in v8. - s/rep_nop/cpu_relax - each thread updates its revision number itself - add XENLOG_ERR prefix for each line of multi-line log messages Changes in v8: - to support blocking #NMI handling during loading ucode * introduce a flag, 'loading_state', to mark the start or end of ucode loading. * use a bitmap for cpu callin since if cpu may stay in #NMI handling, there are two places for a cpu to call in. bitmap won't be counted twice. * don't wait for all CPUs callout, just wait for CPUs that perform the update. We have to do this because some threads may be stuck in NMI handling (where cannot reach the rendezvous). - emit a warning if the system stays in stop_machine context for more than 1s - comment that rdtsc is fine while loading an update - use cmpxchg() to avoid panic being called on multiple CPUs - Propagate revision number to other threads - refine comments and prompt messages Changes in v7: - Check whether 'timeout' is 0 rather than "<=0" since it is unsigned int. - reword the comment above microcode_update_cpu() to clearly state that one thread per core should do the update. --- xen/arch/x86/microcode.c | 297 ++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 267 insertions(+), 30 deletions(-) diff --git a/xen/arch/x86/microcode.c b/xen/arch/x86/microcode.c index 9c0e5c4..6c23879 100644 --- a/xen/arch/x86/microcode.c +++ b/xen/arch/x86/microcode.c @@ -30,18 +30,52 @@ #include #include #include +#include #include #include #include +#include +#include #include #include #include #include +/* + * Before performing a late microcode update on any thread, we + * rendezvous all cpus in stop_machine context. The timeout for + * waiting for cpu rendezvous is 30ms. It is the timeout used by + * live patching + */ +#define MICROCODE_CALLIN_TIMEOUT_US 30000 + +/* + * Timeout for each thread to complete update is set to 1s. It is a + * conservative choice considering all possible interference. + */ +#define MICROCODE_UPDATE_TIMEOUT_US 1000000 + static module_t __initdata ucode_mod; static signed int __initdata ucode_mod_idx; static bool_t __initdata ucode_mod_forced; +static unsigned int nr_cores; + +/* + * These states help to coordinate CPUs during loading an update. + * + * The semantics of each state is as follow: + * - LOADING_PREPARE: initial state of 'loading_state'. + * - LOADING_CALLIN: CPUs are allowed to callin. + * - LOADING_ENTER: all CPUs have called in. Initiate ucode loading. + * - LOADING_EXIT: ucode loading is done or aborted. + */ +static enum { + LOADING_PREPARE, + LOADING_CALLIN, + LOADING_ENTER, + LOADING_EXIT, +} loading_state; /* * If we scan the initramfs.cpio for the early microcode code @@ -190,6 +224,16 @@ static DEFINE_SPINLOCK(microcode_mutex); DEFINE_PER_CPU(struct cpu_signature, cpu_sig); /* + * Count the CPUs that have entered, exited the rendezvous and succeeded in + * microcode update during late microcode update respectively. + * + * Note that a bitmap is used for callin to allow cpu to set a bit multiple + * times. It is required to do busy-loop in #NMI handling. + */ +static cpumask_t cpu_callin_map; +static atomic_t cpu_out, cpu_updated; + +/* * Return a patch that covers current CPU. If there are multiple patches, * return the one with the highest revision number. Return error If no * patch is found and an error occurs during the parsing process. Otherwise @@ -231,6 +275,34 @@ static bool microcode_update_cache(struct microcode_patch *patch) return true; } +/* Wait for a condition to be met with a timeout (us). */ +static int wait_for_condition(bool (*func)(unsigned int data), + unsigned int data, unsigned int timeout) +{ + while ( !func(data) ) + { + if ( !timeout-- ) + { + printk("CPU%u: Timeout in %pS\n", + smp_processor_id(), __builtin_return_address(0)); + return -EBUSY; + } + udelay(1); + } + + return 0; +} + +static bool wait_cpu_callin(unsigned int nr) +{ + return cpumask_weight(&cpu_callin_map) >= nr; +} + +static bool wait_cpu_callout(unsigned int nr) +{ + return atomic_read(&cpu_out) >= nr; +} + /* * Load a microcode update to current CPU. * @@ -264,40 +336,150 @@ static int microcode_update_cpu(const struct microcode_patch *patch) return err; } -static long do_microcode_update(void *patch) +static bool wait_for_state(typeof(loading_state) state) { - unsigned int cpu; - int ret = microcode_update_cpu(patch); + typeof(loading_state) cur_state; - /* Store the patch after a successful loading */ - if ( !ret && patch ) + while ( (cur_state = ACCESS_ONCE(loading_state)) != state ) { - spin_lock(µcode_mutex); - microcode_update_cache(patch); - spin_unlock(µcode_mutex); - patch = NULL; + if ( cur_state == LOADING_EXIT ) + return false; + cpu_relax(); } - if ( microcode_ops->end_update_percpu ) - microcode_ops->end_update_percpu(); + return true; +} + +static void set_state(unsigned int state) +{ + loading_state = state; + smp_wmb(); +} + +static int secondary_thread_fn(void) +{ + unsigned int primary = cpumask_first(this_cpu(cpu_sibling_mask)); + + if ( !wait_for_state(LOADING_CALLIN) ) + return -EBUSY; + + cpumask_set_cpu(smp_processor_id(), &cpu_callin_map); + + if ( !wait_for_state(LOADING_EXIT) ) + return -EBUSY; + + /* Copy update revision from the primary thread. */ + this_cpu(cpu_sig).rev = per_cpu(cpu_sig, primary).rev; + + return 0; +} + +static int primary_thread_fn(const struct microcode_patch *patch) +{ + int ret = 0; + + if ( !wait_for_state(LOADING_CALLIN) ) + return -EBUSY; + + cpumask_set_cpu(smp_processor_id(), &cpu_callin_map); + + if ( !wait_for_state(LOADING_ENTER) ) + return -EBUSY; + + ret = microcode_ops->apply_microcode(patch); + if ( !ret ) + atomic_inc(&cpu_updated); + atomic_inc(&cpu_out); + + return ret; +} + +static int control_thread_fn(const struct microcode_patch *patch) +{ + unsigned int cpu = smp_processor_id(), done; + unsigned long tick; + int ret; /* - * Each thread tries to load ucode. Only the first thread of a core - * would succeed while other threads would encounter -EINVAL which - * indicates current ucode revision is equal to or newer than the - * given patch. It is actually expected; so ignore this error. + * We intend to keep interrupt disabled for a long time, which may lead to + * watchdog timeout. */ - if ( ret == -EINVAL ) - ret = 0; + watchdog_disable(); - cpu = cpumask_next(smp_processor_id(), &cpu_online_map); - if ( cpu < nr_cpu_ids ) - return continue_hypercall_on_cpu(cpu, do_microcode_update, patch) ?: - ret; + /* Allow threads to call in */ + set_state(LOADING_CALLIN); - /* Free the patch if no CPU has loaded it successfully. */ - if ( patch ) - microcode_free_patch(patch); + cpumask_set_cpu(cpu, &cpu_callin_map); + + /* Waiting for all threads calling in */ + ret = wait_for_condition(wait_cpu_callin, num_online_cpus(), + MICROCODE_CALLIN_TIMEOUT_US); + if ( ret ) + { + set_state(LOADING_EXIT); + return ret; + } + + /* Let primary threads load the given ucode update */ + set_state(LOADING_ENTER); + + ret = microcode_ops->apply_microcode(patch); + if ( !ret ) + atomic_inc(&cpu_updated); + atomic_inc(&cpu_out); + + tick = rdtsc_ordered(); + /* Wait for primary threads finishing update */ + while ( (done = atomic_read(&cpu_out)) != nr_cores ) + { + /* + * During each timeout interval, at least a CPU is expected to + * finish its update. Otherwise, something goes wrong. + * + * Note that RDTSC (in wait_for_condition()) is safe for threads to + * execute while waiting for completion of loading an update. + */ + if ( wait_for_condition(wait_cpu_callout, (done + 1), + MICROCODE_UPDATE_TIMEOUT_US) ) + panic("Timeout when finished updating microcode (finished %u/%u)", + done, nr_cores); + + /* Print warning message once if long time is spent here */ + if ( tick && rdtsc_ordered() - tick >= cpu_khz * 1000 ) + { + printk(XENLOG_WARNING + "WARNING: UPDATING MICROCODE HAS CONSUMED MORE THAN 1 SECOND!\n"); + tick = 0; + } + } + + /* Mark loading is done to unblock other threads */ + set_state(LOADING_EXIT); + + watchdog_enable(); + + return ret; +} + +static int do_microcode_update(void *patch) +{ + unsigned int cpu = smp_processor_id(); + int ret; + + /* + * The control thread set state to coordinate ucode loading. Primary + * threads load the given ucode patch. Secondary threads just wait for + * the completion of the ucode loading process. + */ + if ( cpu == cpumask_first(&cpu_online_map) ) + ret = control_thread_fn(patch); + else if ( cpu == cpumask_first(this_cpu(cpu_sibling_mask)) ) + ret = primary_thread_fn(patch); + else + ret = secondary_thread_fn(); + + if ( microcode_ops->end_update_percpu ) + microcode_ops->end_update_percpu(); return ret; } @@ -306,6 +488,7 @@ int microcode_update(XEN_GUEST_HANDLE_PARAM(const_void) buf, unsigned long len) { int ret; void *buffer; + unsigned int cpu, updated; struct microcode_patch *patch; if ( len != (uint32_t)len ) @@ -325,30 +508,84 @@ int microcode_update(XEN_GUEST_HANDLE_PARAM(const_void) buf, unsigned long len) return -EFAULT; } + /* cpu_online_map must not change during update */ + if ( !get_cpu_maps() ) + { + xfree(buffer); + return -EBUSY; + } + patch = parse_blob(buffer, len); xfree(buffer); if ( IS_ERR(patch) ) { ret = PTR_ERR(patch); printk(XENLOG_WARNING "Parsing microcode blob error %d\n", ret); - return ret; + goto put; } if ( !patch ) - return -ENOENT; + { + ret = -ENOENT; + goto put; + } if ( microcode_ops->start_update ) { ret = microcode_ops->start_update(); - if ( ret != 0 ) + if ( ret ) { microcode_free_patch(patch); - return ret; + goto put; } } - return continue_hypercall_on_cpu(cpumask_first(&cpu_online_map), - do_microcode_update, patch); + cpumask_clear(&cpu_callin_map); + atomic_set(&cpu_out, 0); + atomic_set(&cpu_updated, 0); + loading_state = LOADING_PREPARE; + + /* Calculate the number of online CPU core */ + nr_cores = 0; + for_each_online_cpu(cpu) + if ( cpu == cpumask_first(per_cpu(cpu_sibling_mask, cpu)) ) + nr_cores++; + + printk(XENLOG_INFO "%u cores are to update their microcode\n", nr_cores); + + /* + * Late loading dance. Why the heavy-handed stop_machine effort? + * + * - HT siblings must be idle and not execute other code while the other + * sibling is loading microcode in order to avoid any negative + * interactions cause by the loading. + * + * - In addition, microcode update on the cores must be serialized until + * this requirement can be relaxed in the future. Right now, this is + * conservative and good. + */ + ret = stop_machine_run(do_microcode_update, patch, NR_CPUS); + + updated = atomic_read(&cpu_updated); + if ( updated > 0 ) + { + spin_lock(µcode_mutex); + microcode_update_cache(patch); + spin_unlock(µcode_mutex); + } + else + microcode_free_patch(patch); + + if ( updated && updated != nr_cores ) + printk(XENLOG_ERR "ERROR: Updating microcode succeeded on %u cores and failed\n" + XENLOG_ERR "on other %u cores. A system with differing microcode\n" + XENLOG_ERR "revisions is considered unstable. Please reboot and do not\n" + XENLOG_ERR "load the microcode that triggers this warning!\n", + updated, nr_cores - updated); + + put: + put_cpu_maps(); + return ret; } static int __init microcode_init(void)