linux-kernel - Re: [PATCH v2 3/7] arch/x86/amd/ibs: Fix re-arming IBS Fetch

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-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200910085036.GD35926@hirez.programming.kicks-ass.net>
Date:   Thu, 10 Sep 2020 10:50:36 +0200
From:   peterz@...radead.org
To:     Kim Phillips <kim.phillips@....com>
Cc:     Borislav Petkov <bp@...en8.de>, Borislav Petkov <bp@...e.de>,
        Ingo Molnar <mingo@...nel.org>, Ingo Molnar <mingo@...hat.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Stephane Eranian <eranian@...gle.com>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        "H. Peter Anvin" <hpa@...or.com>, Jiri Olsa <jolsa@...hat.com>,
        Mark Rutland <mark.rutland@....com>,
        Michael Petlan <mpetlan@...hat.com>,
        Namhyung Kim <namhyung@...nel.org>,
        LKML <linux-kernel@...r.kernel.org>, x86 <x86@...nel.org>,
        Stephane Eranian <stephane.eranian@...gle.com>,
        stable@...r.kernel.org
Subject: Re: [PATCH v2 3/7] arch/x86/amd/ibs: Fix re-arming IBS Fetch

On Thu, Sep 10, 2020 at 10:32:23AM +0200, peterz@...radead.org wrote:
> > @@ -363,7 +363,14 @@ perf_ibs_event_update(struct perf_ibs *perf_ibs, struct perf_event *event,
> >  static inline void perf_ibs_enable_event(struct perf_ibs *perf_ibs,
> >  					 struct hw_perf_event *hwc, u64 config)
> >  {
> > -	wrmsrl(hwc->config_base, hwc->config | config | perf_ibs->enable_mask);
> > +	u64 _config = (hwc->config | config) & ~perf_ibs->enable_mask;
> > +
> > +	/* On Fam17h, the periodic fetch counter is set when IbsFetchEn is changed from 0 to 1 */
> > +	if (perf_ibs == &perf_ibs_fetch && boot_cpu_data.x86 >= 0x16 && boot_cpu_data.x86 <= 0x18)
> > +		wrmsrl(hwc->config_base, _config);

> A better option would be to use hwc->flags, you're loading from that
> line already, so it's guaranteed hot and then you only have a single
> branch. Or stick it in perf_ibs near enable_mask, same difference.

I fixed it for you.

---

struct perf_ibs {
	struct pmu                 pmu;                  /*     0   296 */
	/* --- cacheline 4 boundary (256 bytes) was 40 bytes ago --- */
	unsigned int               msr;                  /*   296     4 */

	/* XXX 4 bytes hole, try to pack */

	u64                        config_mask;          /*   304     8 */
	u64                        cnt_mask;             /*   312     8 */
	/* --- cacheline 5 boundary (320 bytes) --- */
	u64                        enable_mask;          /*   320     8 */
	u64                        valid_mask;           /*   328     8 */
	u64                        max_period;           /*   336     8 */
	long unsigned int          offset_mask[1];       /*   344     8 */
	int                        offset_max;           /*   352     4 */
	unsigned int               fetch_count_reset_broken:1; /*   356: 0  4 */

	/* XXX 31 bits hole, try to pack */

	struct cpu_perf_ibs *      pcpu;                 /*   360     8 */
	struct attribute * *       format_attrs;         /*   368     8 */
	struct attribute_group     format_group;         /*   376    40 */
	/* --- cacheline 6 boundary (384 bytes) was 32 bytes ago --- */
	const struct attribute_group  * attr_groups[2];  /*   416    16 */
	u64                        (*get_count)(u64);    /*   432     8 */

	/* size: 440, cachelines: 7, members: 15 */
	/* sum members: 432, holes: 1, sum holes: 4 */
	/* sum bitfield members: 1 bits, bit holes: 1, sum bit holes: 31 bits */
	/* last cacheline: 56 bytes */
};

--- a/arch/x86/events/amd/ibs.c
+++ b/arch/x86/events/amd/ibs.c
@@ -89,6 +89,7 @@ struct perf_ibs {
 	u64				max_period;
 	unsigned long			offset_mask[1];
 	int				offset_max;
+	unsigned int			fetch_count_reset_broken : 1;
 	struct cpu_perf_ibs __percpu	*pcpu;
 
 	struct attribute		**format_attrs;
@@ -370,7 +371,13 @@ perf_ibs_event_update(struct perf_ibs *p
 static inline void perf_ibs_enable_event(struct perf_ibs *perf_ibs,
 					 struct hw_perf_event *hwc, u64 config)
 {
-	wrmsrl(hwc->config_base, hwc->config | config | perf_ibs->enable_mask);
+	u64 _config = (hwc->config | config) & ~perf_ibs->enable_mask;
+
+	if (perf_ibs->fetch_count_reset_broken)
+		wrmsrl(hwc->config_base, _config);
+
+ 	_config |= perf_ibs->enable_mask;
+	wrmsrl(hwc->config_base, _config);
 }
 
 /*
@@ -756,6 +763,13 @@ static __init void perf_event_ibs_init(v
 {
 	struct attribute **attr = ibs_op_format_attrs;
 
+	/*
+	 * Some chips fail to reset the fetch count when it is written; instead
+	 * they need a 0-1 transition of IbsFetchEn.
+	 */
+	if (boot_cpu_data.x86 >= 0x16 && boot_cpu_data.x86 <= 0x18)
+		perf_ibs_fetch.fetch_count_reset_broken = 1;
+
 	perf_ibs_pmu_init(&perf_ibs_fetch, "ibs_fetch");
 
 	if (ibs_caps & IBS_CAPS_OPCNT) {