diff options
| author | Linus Torvalds <torvalds@linux-foundation.org> | 2026-08-19 10:09:22 -0700 |
|---|---|---|
| committer | Linus Torvalds <torvalds@linux-foundation.org> | 2026-08-19 10:09:22 -0700 |
| commit | f4cdf7ca9a1fdcca413157df19753f388a5a224e (patch) | |
| tree | 60a31895b2446cd7fa27514e8c2843534a0de0da /drivers/iommu/intel/perf.c | |
| download | linux-stable-f4cdf7ca9a1fdcca413157df19753f388a5a224e.tar.gz linux-stable-f4cdf7ca9a1fdcca413157df19753f388a5a224e.zip | |
Merge tag 'media/v7.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-mediagrafted
Pull media updates from Mauro Carvalho Chehab:
- v4l2-core: added ISP statistics support and per-block validation
- v4l2-core: Allow unknown HDR10 white point and luminance
- New camera sensors: Sony IMX678 and IMX471m, Himax HM1092 IR sensor
- New codec: Milos: VPU v2.0 codec support
- isp driver: gained support for Dreamchip RPPX1 ISP framework
- vsp1 driver: gained support for RZ/T2H and RZ/N2H
- Novalake driver: gained CVS support for new NVL hardware
- dvb-core: fix feed leak on failed DMX_ADD_PID
- several driver fixes, cleanups and minor improvements
* tag 'media/v7.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media: (308 commits)
media: ipu-bridge: check all DMI entries when overriding sensor rotation
media: v4l2-async: avoid deleting unlinked ASC entry on link error
media: rzg2l-cru: Align bytesperline to hardware DMA stride requirement
media: intel/ipu6: fix async notifier cleanup leak on parse error
media: staging/ipu7: fix async notifier UAF on probe error path
media: amd: isp4: fix self-deadlock in isp4sd_pwron_and_init() error path
media: amd: isp4: release partial allocations in isp4if_alloc_fw_gpumem()
media: rcar-isp: Fix VSPX reference leaks
media: rcar-isp: Release ISPCORE resources
media: i2c: imx415: Release runtime PM reference on VBLANK error
media: i2c: imx415: Return test pattern write errors
media: renesas: vsp1: Declare index variables in for loop statement
media: renesas: vsp1: Make reset control optional to support platforms without a reset line
media: dt-bindings: media: renesas,vsp1: Document RZ/T2H and RZ/N2H SoCs
media: dt-bindings: media: renesas,fcp: Document RZ/T2H and RZ/N2H SoCs
media: nxp: imx8-isi: Add additional 32-bit RGB format support
media: nxp: imx8-isi: Add 16-bit raw Bayer format support
media: nxp: imx8-isi: Implement per-stream reference counting for multiplexed streams
media: nxp: imx8-isi: Use BIT_ULL() for 64-bit stream masks
media: nxp: imx8-isi: Correct color map between V4L2 and ISI
...
Diffstat (limited to 'drivers/iommu/intel/perf.c')
| -rw-r--r-- | drivers/iommu/intel/perf.c | 164 |
1 files changed, 164 insertions, 0 deletions
diff --git a/drivers/iommu/intel/perf.c b/drivers/iommu/intel/perf.c new file mode 100644 index 000000000..bec98dcbb --- /dev/null +++ b/drivers/iommu/intel/perf.c @@ -0,0 +1,164 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * perf.c - performance monitor + * + * Copyright (C) 2021 Intel Corporation + * + * Author: Lu Baolu <baolu.lu@linux.intel.com> + * Fenghua Yu <fenghua.yu@intel.com> + */ + +#include <linux/spinlock.h> + +#include "iommu.h" +#include "perf.h" + +static DEFINE_SPINLOCK(latency_lock); + +bool dmar_latency_enabled(struct intel_iommu *iommu, enum latency_type type) +{ + struct latency_statistic *lstat = iommu->perf_statistic; + + return lstat && lstat[type].enabled; +} + +int dmar_latency_enable(struct intel_iommu *iommu, enum latency_type type) +{ + struct latency_statistic *lstat; + unsigned long flags; + int ret = -EBUSY; + + if (dmar_latency_enabled(iommu, type)) + return 0; + + spin_lock_irqsave(&latency_lock, flags); + if (!iommu->perf_statistic) { + iommu->perf_statistic = kzalloc_objs(*lstat, DMAR_LATENCY_NUM, + GFP_ATOMIC); + if (!iommu->perf_statistic) { + ret = -ENOMEM; + goto unlock_out; + } + } + + lstat = iommu->perf_statistic; + + if (!lstat[type].enabled) { + lstat[type].enabled = true; + lstat[type].counter[COUNTS_MIN] = UINT_MAX; + ret = 0; + } +unlock_out: + spin_unlock_irqrestore(&latency_lock, flags); + + return ret; +} + +void dmar_latency_disable(struct intel_iommu *iommu, enum latency_type type) +{ + struct latency_statistic *lstat = iommu->perf_statistic; + unsigned long flags; + + if (!dmar_latency_enabled(iommu, type)) + return; + + spin_lock_irqsave(&latency_lock, flags); + memset(&lstat[type], 0, sizeof(*lstat)); + spin_unlock_irqrestore(&latency_lock, flags); +} + +void dmar_latency_update(struct intel_iommu *iommu, enum latency_type type, u64 latency) +{ + struct latency_statistic *lstat = iommu->perf_statistic; + unsigned long flags; + u64 min, max; + + if (!dmar_latency_enabled(iommu, type)) + return; + + spin_lock_irqsave(&latency_lock, flags); + if (latency < 100) + lstat[type].counter[COUNTS_10e2]++; + else if (latency < 1000) + lstat[type].counter[COUNTS_10e3]++; + else if (latency < 10000) + lstat[type].counter[COUNTS_10e4]++; + else if (latency < 100000) + lstat[type].counter[COUNTS_10e5]++; + else if (latency < 1000000) + lstat[type].counter[COUNTS_10e6]++; + else if (latency < 10000000) + lstat[type].counter[COUNTS_10e7]++; + else + lstat[type].counter[COUNTS_10e8_plus]++; + + min = lstat[type].counter[COUNTS_MIN]; + max = lstat[type].counter[COUNTS_MAX]; + lstat[type].counter[COUNTS_MIN] = min_t(u64, min, latency); + lstat[type].counter[COUNTS_MAX] = max_t(u64, max, latency); + lstat[type].counter[COUNTS_SUM] += latency; + lstat[type].samples++; + spin_unlock_irqrestore(&latency_lock, flags); +} + +static char *latency_counter_names[] = { + " <0.1us", + " 0.1us-1us", " 1us-10us", " 10us-100us", + " 100us-1ms", " 1ms-10ms", " >=10ms", + " min(us)", " max(us)", " average(us)" +}; + +static char *latency_type_names[] = { + " inv_iotlb", " inv_devtlb", " inv_iec", + " svm_prq" +}; + +void dmar_latency_snapshot(struct intel_iommu *iommu, char *str, size_t size) +{ + struct latency_statistic *lstat = iommu->perf_statistic; + unsigned long flags; + int bytes = 0, i, j; + + memset(str, 0, size); + + for (i = 0; i < COUNTS_NUM; i++) + bytes += scnprintf(str + bytes, size - bytes, + "%s", latency_counter_names[i]); + + spin_lock_irqsave(&latency_lock, flags); + for (i = 0; i < DMAR_LATENCY_NUM; i++) { + if (!dmar_latency_enabled(iommu, i)) + continue; + + bytes += scnprintf(str + bytes, size - bytes, + "\n%s", latency_type_names[i]); + + for (j = 0; j < COUNTS_NUM; j++) { + u64 val = lstat[i].counter[j]; + + switch (j) { + case COUNTS_MIN: + if (val == UINT_MAX) + val = 0; + else + val = div_u64(val, 1000); + break; + case COUNTS_MAX: + val = div_u64(val, 1000); + break; + case COUNTS_SUM: + if (lstat[i].samples) + val = div_u64(val, (lstat[i].samples * 1000)); + else + val = 0; + break; + default: + break; + } + + bytes += scnprintf(str + bytes, size - bytes, + "%12lld", val); + } + } + spin_unlock_irqrestore(&latency_lock, flags); +} |
