lkml.org 
[lkml]   [2008]   [Nov]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch 05/24] perfmon: X86 generic code (x86)
On Wed, Nov 26, 2008 at 12:42:09AM -0800, eranian@googlemail.com wrote:
> + * set cannot be NULL. Context is locked. Interrupts are masked.
> + *
> + * Caller has already restored all PMD and PMC registers, if
> + * necessary (i.e., lazy restore scheme).
> + *
> + * On x86, the only common code just needs to unsecure RDPMC if necessary

What is insecure about RDPMC? (except perhaps when secure
computing mode is on)

I think it should be enabled by default BTW because on Core2+ you
can always read the fixed counters with it.

> + */
> + if (using_nmi)
> + iip = __get_cpu_var(real_iip);

Call it real_rip perhaps?

> + /*
> + * only NMI related calls
> + */
> + if (val != DIE_NMI_IPI)
> + return NOTIFY_DONE;
> +
> + /*
> + * perfmon not using NMI
> + */
> + if (!__get_cpu_var(pfm_using_nmi))
> + return NOTIFY_DONE;

It should not register in this case. die notifiers are costly
because they make a lot of exceptions slower.

> + /*
> + * we need to register our NMI handler when the kernels boots
> + * to avoid a deadlock condition with the NMI watchdog or Oprofile

What deadlock?

> + * if we were to try and register/unregister on-demand.
> + */
> + register_die_notifier(&pfm_nmi_nb);
> + return 0;
> +
> +/*
> + * arch-specific user visible interface definitions
> + */
> +
> +#define PFM_ARCH_MAX_PMCS (256+64) /* 256 HW 64 SW */
> +#define PFM_ARCH_MAX_PMDS (256+64) /* 256 HW 64 SW */

A little excessive for current x86s?

> +#define _ASM_X86_PERFMON_KERN_H_
> +
> +#ifdef CONFIG_PERFMON
> +#include <linux/unistd.h>
> +#ifdef CONFIG_4KSTACKS
> +#define PFM_ARCH_STK_ARG 8
> +#else
> +#define PFM_ARCH_STK_ARG 16
> +#endif

Very fancy. But is it really worth it?

> + * bits as this may cause crash on some processors.
> + */
> + if (pfm_pmu_conf->pmd_desc[cnum].type & PFM_REG_C64)
> + value = (value | ~pfm_pmu_conf->ovfl_mask)
> + & ~pfm_pmu_conf->pmd_desc[cnum].rsvd_msk;
> +
> + PFM_DBG_ovfl("pfm_arch_write_pmd(0x%lx, 0x%Lx)",
> + pfm_pmu_conf->pmd_desc[cnum].hw_addr,
> + (unsigned long long) value);
> +
> + wrmsrl(pfm_pmu_conf->pmd_desc[cnum].hw_addr, value);

Not sure how well error handling would fit in here, but it's
normally a good idea to make at least the first wrmsrl to
these counters a checking_wrmsrl because sometimes simulators
or hypervisors don't implement them.

> + */
> +static inline void pfm_arch_unload_context(struct pfm_context *ctx)

In general a lot of these inlines seem rather large. If they are
called more than once consider out of lining for better code size.

> + * x86 does not need extra alignment requirements for the sampling buffer
> + */
> +#define PFM_ARCH_SMPL_ALIGN_SIZE 0
> +
> +asmlinkage void pmu_interrupt(void);
> +
> +static inline void pfm_arch_bv_copy(u64 *a, u64 *b, int nbits)

All these bitmap wrappers just seem like unnecessary obfuscation.
Could you just drop them and call the standard functions directly?


-Andi

--
ak@linux.intel.com


\
 
 \ /
  Last update: 2008-11-26 12:25    [W:0.094 / U:1.416 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site