| 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
| ||
|
Message-ID: <20170112175223.qsui53kln7vpmwop@pd.tnic>
Date: Thu, 12 Jan 2017 18:52:23 +0100
From: Borislav Petkov <bp@...en8.de>
To: Suravee Suthikulpanit <Suravee.Suthikulpanit@....com>
Cc: linux-kernel@...r.kernel.org, iommu@...ts.linux-foundation.org,
joro@...tes.org, peterz@...radead.org, mingo@...hat.com
Subject: Re: [PATCH v7 7/7] perf/amd/iommu: Enable support for multiple IOMMUs
On Mon, Jan 09, 2017 at 09:33:47PM -0600, Suravee Suthikulpanit wrote:
> This patch adds multi-IOMMU support for perf by exposing
> an AMD IOMMU PMU for each IOMMU found in the system via:
>
> /sys/device/amd_iommu_x /* where x is the IOMMU index. */
Straight into the top-level devices hierarchy?
Why not add a proper hierarchy to
/sys/devices/system/iommu/ ...
?
Jörg, don't you have a sane iommu hierarchy already in sysfs?
> This allows users to specify different events to be programed
> onto performance counters of each IOMMU.
>
> Cc: Peter Zijlstra <peterz@...radead.org>
> Cc: Borislav Petkov <bp@...en8.de>
> Signed-off-by: Suravee Suthikulpanit <Suravee.Suthikulpanit@....com>
> ---
> arch/x86/events/amd/iommu.c | 119 ++++++++++++++++++++++++--------------------
> 1 file changed, 64 insertions(+), 55 deletions(-)
>
> diff --git a/arch/x86/events/amd/iommu.c b/arch/x86/events/amd/iommu.c
> index 2403c78..5fd97b5 100644
> --- a/arch/x86/events/amd/iommu.c
> +++ b/arch/x86/events/amd/iommu.c
> @@ -35,10 +35,13 @@
> #define _GET_PASID_MASK(ev) ((ev->hw.extra_reg.config >> 16) & 0xFFFFULL)
> #define _GET_DOMID_MASK(ev) ((ev->hw.extra_reg.config >> 32) & 0xFFFFULL)
>
> -static struct perf_amd_iommu __perf_iommu;
> +#define PERF_AMD_IOMMU_NAME_SZ 16
>
> struct perf_amd_iommu {
> + struct list_head list;
> struct pmu pmu;
> + uint idx;
This uint thing has crept in:
$ git grep "uint " arch/x86/
arch/x86/crypto/crc32-pclmul_asm.S:101: * uint crc32_pclmul_le_16(unsigned char const *buffer,
arch/x86/crypto/crc32-pclmul_asm.S:102: * size_t len, uint crc32)
arch/x86/events/amd/iommu.h:32:extern u8 amd_iommu_pc_get_max_banks(uint idx);
arch/x86/events/amd/iommu.h:34:extern u8 amd_iommu_pc_get_max_counters(uint idx);
arch/x86/events/amd/iommu.h:36:extern int amd_iommu_pc_set_reg(uint idx, u16 devid, u8 bank, u8 cntr,
arch/x86/events/amd/iommu.h:39:extern int amd_iommu_pc_set_counter(uint idx, u8 bank, u8 cntr, u64 *value);
arch/x86/events/amd/iommu.h:41:extern int amd_iommu_pc_get_counter(uint idx, u8 bank, u8 cntr, u64 *value);
"unsigned int" please.
> + char name[PERF_AMD_IOMMU_NAME_SZ];
> u8 max_banks;
> u8 max_counters;
> u64 cntr_assign_mask;
> @@ -46,6 +49,8 @@ struct perf_amd_iommu {
> const struct attribute_group *attr_groups[4];
> };
>
> +LIST_HEAD(perf_amd_iommu_list);
static
> +
> #define format_group attr_groups[0]
> #define cpumask_group attr_groups[1]
> #define events_group attr_groups[2]
> @@ -204,8 +209,7 @@ static int clear_avail_iommu_bnk_cntr(struct perf_amd_iommu *perf_iommu,
> static int perf_iommu_event_init(struct perf_event *event)
> {
> struct hw_perf_event *hwc = &event->hw;
> - struct perf_amd_iommu *perf_iommu;
> - u64 config, config1;
> + struct perf_amd_iommu *pi = container_of(event->pmu, struct perf_amd_iommu, pmu);
Do this deref when you actually need it:
struct perf_amd_iommu *pi;
because you might exit earlier and end up doing it for nothing.
>
> /* test the event attr type check for PMU enumeration */
> if (event->attr.type != event->pmu->type)
> @@ -227,27 +231,17 @@ static int perf_iommu_event_init(struct perf_event *event)
> if (event->cpu < 0)
> return -EINVAL;
>
> - perf_iommu = &__perf_iommu;
> -
> - if (event->pmu != &perf_iommu->pmu)
> - return -ENOENT;
> -
> - if (perf_iommu) {
> - config = event->attr.config;
> - config1 = event->attr.config1;
> - } else {
> - return -EINVAL;
> - }
> -
> /* update the hw_perf_event struct with the iommu config data */
> - hwc->config = config;
> - hwc->extra_reg.config = config1;
---> here:
pi = container_of(event->pmu, struct perf_amd_iommu, pmu);
> + hwc->idx = pi->idx;
> + hwc->config = event->attr.config;
> + hwc->extra_reg.config = event->attr.config1;
Align vertically on the = sign.
>
> return 0;
> }
>
> static void perf_iommu_enable_event(struct perf_event *ev)
> {
> + struct hw_perf_event *hwc = &ev->hw;
> u8 csource = _GET_CSOURCE(ev);
> u16 devid = _GET_DEVID(ev);
> u8 bank = _GET_BANK(ev);
...
> static void perf_iommu_disable_event(struct perf_event *event)
> {
> + struct hw_perf_event *hwc = &event->hw;
> u64 reg = 0ULL;
>
> - amd_iommu_pc_set_reg(0, _GET_DEVID(event), _GET_BANK(event),
> + amd_iommu_pc_set_reg(hwc->idx, _GET_DEVID(event), _GET_BANK(event),
> _GET_CNTR(event), IOMMU_PC_COUNTER_SRC_REG, ®);
> }
>
> @@ -301,7 +296,7 @@ static void perf_iommu_start(struct perf_event *event, int flags)
>
> val = local64_read(&hwc->prev_count);
>
> - amd_iommu_pc_set_counter(0, _GET_BANK(event), _GET_CNTR(event), &val);
> + amd_iommu_pc_set_counter(hwc->idx, _GET_BANK(event), _GET_CNTR(event), &val);
> enable:
> perf_iommu_enable_event(event);
> perf_event_update_userpage(event);
> @@ -314,7 +309,7 @@ static void perf_iommu_read(struct perf_event *event)
> s64 delta;
> struct hw_perf_event *hwc = &event->hw;
>
> - if (amd_iommu_pc_get_counter(0, _GET_BANK(event), _GET_CNTR(event), &cnt))
> + if (amd_iommu_pc_get_counter(hwc->idx, _GET_BANK(event), _GET_CNTR(event), &cnt))
> return;
>
> /* IOMMU pc counter register is only 48 bits */
> @@ -417,14 +412,20 @@ static __init int _init_events_attrs(struct perf_amd_iommu *perf_iommu)
>
> static __init void amd_iommu_pc_exit(void)
> {
> - if (__perf_iommu.events_group != NULL) {
> - kfree(__perf_iommu.events_group);
> - __perf_iommu.events_group = NULL;
> + struct perf_amd_iommu *pi, *next;
> +
> + list_for_each_entry_safe(pi, next, &perf_amd_iommu_list, list) {
> + list_del(&pi->list);
> +
> + kfree(pi->events_group);
> + pi->events_group = NULL;
Why that assignment if you're going to free it in the next line?
> +
> + kfree(pi);
> }
> }
>
> -static __init int _init_perf_amd_iommu(
> - struct perf_amd_iommu *perf_iommu, char *name)
> +static __init int
> +init_one_perf_amd_iommu(struct perf_amd_iommu *perf_iommu, uint idx)
> {
> int ret;
>
> @@ -441,54 +442,62 @@ static __init int _init_perf_amd_iommu(
> if (_init_events_attrs(perf_iommu) != 0)
> pr_err("Only support raw events.\n");
Here's also
/* Init cpumask attributes to only core 0 */
cpumask_set_cpu(0, &iommu_cpumask);
is that still correct for all IOMMUs?
>
> - perf_iommu->max_banks = amd_iommu_pc_get_max_banks(0);
> - perf_iommu->max_counters = amd_iommu_pc_get_max_counters(0);
> + snprintf(perf_iommu->name, PERF_AMD_IOMMU_NAME_SZ, "amd_iommu_%u", idx);
Do this ^^
> + perf_iommu->idx = idx;
> + perf_iommu->max_banks = amd_iommu_pc_get_max_banks(idx);
> + perf_iommu->max_counters = amd_iommu_pc_get_max_counters(idx);
> if (!perf_iommu->max_banks || !perf_iommu->max_counters)
> return -EINVAL;
... here, after the test has passed.
> /* Init null attributes */
> perf_iommu->null_group = NULL;
No need for this - you're getting zeroed memory from kzalloc().
> + /* Setting up PMU */
> + perf_iommu->pmu.event_init = perf_iommu_event_init,
> + perf_iommu->pmu.add = perf_iommu_add,
> + perf_iommu->pmu.del = perf_iommu_del,
> + perf_iommu->pmu.start = perf_iommu_start,
> + perf_iommu->pmu.stop = perf_iommu_stop,
> + perf_iommu->pmu.read = perf_iommu_read,
> + perf_iommu->pmu.task_ctx_nr = perf_invalid_context;
> perf_iommu->pmu.attr_groups = perf_iommu->attr_groups;
Align vertically on the =
> - ret = perf_pmu_register(&perf_iommu->pmu, name, -1);
> + ret = perf_pmu_register(&perf_iommu->pmu, perf_iommu->name, -1);
> if (ret) {
> pr_err("Error initializing AMD IOMMU perf counters.\n");
> amd_iommu_pc_exit();
> } else {
> - pr_info("perf: amd_iommu: Detected. (%d banks, %d counters/bank)\n",
> - amd_iommu_pc_get_max_banks(0),
> - amd_iommu_pc_get_max_counters(0));
> + pr_info("Detected %s, w/ %d banks, %d counters/bank\n",
I'd prefer:
"Detected AMD IOMMU #%d (%d banks, %d counters/bank)"
where the "#%d" is idx.
> static __init int amd_iommu_pc_init(void)
> {
> + uint i;
> +
> /* Make sure the IOMMU PC resource is available */
> if (!amd_iommu_pc_supported())
> return -ENODEV;
>
> - _init_perf_amd_iommu(&__perf_iommu, "amd_iommu");
> + for (i = 0 ; i < amd_iommu_get_num_iommus(); i++) {
> + int ret;
> + struct perf_amd_iommu *pi;
> +
> + pi = kzalloc(sizeof(struct perf_amd_iommu), GFP_KERNEL);
> + if (!pi)
> + return -ENOMEM;
> +
> + ret = init_one_perf_amd_iommu(pi, i);
> + if (ret)
> + return ret;
You're leaking here memory if some IOMMUs fail and that function returns
-EINVAL from the max_banks and counters test above.
So you need to unwind all that setup init_one_perf_amd_iommu() did
for the iommus from 0 up to i - 1; IOW; you need to move the call to
amd_iommu_pc_exit() here and not in the init function.
That'll be all for now. :-)
Thanks.
--
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.
Powered by blists - more mailing lists