[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH] xentrace: dynamic tracebuffer size allocation


  • To: Olaf Hering <olaf@xxxxxxxxx>
  • From: George Dunlap <George.Dunlap@xxxxxxxxxxxxx>
  • Date: Wed, 16 Mar 2011 11:32:24 +0000
  • Cc: xen-devel@xxxxxxxxxxxxxxxxxxx, Keir Fraser <keir@xxxxxxx>
  • Delivery-date: Wed, 16 Mar 2011 04:33:18 -0700
  • Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:sender:in-reply-to:references:date :x-google-sender-auth:message-id:subject:from:to:cc:content-type :content-transfer-encoding; b=iSQbH2bHKl1jDnRLq/f2LwVIsT2waYJsP6yxgzCkRtH11Pxh4Qs1+C+kDTycrsLrbL bD3uYtpEN4qaOyzzVcCnJEWLj2WoRx12hPMGEhTneyVUoXNRX3SzdkVE6aGH5D01n/UW JIMubgKxCoPOtcfiWZWhT5JT207y5ecAbN4kY=
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>

Olaf,

Your mailer appears to be mucking around with the whitespace; the
patch as downloaded won't apply due to mismatches.  Can you refresh to
tip and send it again?

To keep your mailer from screwing up the patch, you can either:
* Send it as an attachment (simple but makes it harder to comment on)
* {Use a mailer that | configure your mailer so that it} doesn't muck
with the whitlespaces
* Use "hg email" (which is what I do).

 -George

On Mon, Mar 14, 2011 at 5:33 PM, Olaf Hering <olaf@xxxxxxxxx> wrote:
> On Mon, Feb 07, George Dunlap wrote:
>
>> On Sun, Feb 6, 2011 at 1:39 PM, Olaf Hering <olaf@xxxxxxxxx> wrote:
>> > @@ -85,20 +100,30 @@ static void calc_tinfo_first_offset(void
>> >  }
>> >
>> >  /**
>> > - * check_tbuf_size - check to make sure that the proposed size will fit
>> > + * calculate_tbuf_size - check to make sure that the proposed size will 
>> > fit
>> >  * in the currently sized struct t_info and allows prod and cons to
>> >  * reach double the value without overflow.
>> >  */
>> > -static int check_tbuf_size(u32 pages)
>> > +static int calculate_tbuf_size(unsigned int pages)
>> >  {
>> >     struct t_buf dummy;
>> > -    typeof(dummy.prod) size;
>> > -
>> > -    size = ((typeof(dummy.prod))pages)  * PAGE_SIZE;
>> > -
>> > -    return (size / PAGE_SIZE != pages)
>> > -           || (size + size < size)
>> > -           || (num_online_cpus() * pages + t_info_first_offset > 
>> > T_INFO_SIZE / sizeof(uint32_t));
>> > +    typeof(dummy.prod) size = -1;
>> > +
>> > +    /* max size holds up to n pages */
>> > +    size /= PAGE_SIZE;
>>
>> size=-1, then size /= PAGE_SIZE?  Is this a clever way of finding the
>> maximum buffer size able to be pointed to?  If so, it needs a comment
>> explaining why it works; I'm not convinced just by looking at it this
>> is will work properly.
>
> George,
>
> are you ok with the attached version?
>
>
>
> Allocate tracebuffers dynamically, based on the requested buffer size.
> Calculate t_info_size from requested t_buf size.
> Fix allocation failure path, free pages outside the spinlock.
> Remove casts for rawbuf, it can be a void pointer since no math is done.
>
> Signed-off-by: Olaf Hering <olaf@xxxxxxxxx>
>
> ---
> v3:
>  add comments to calculate_tbuf_size for side effects and max value
> v2:
>  if per_cpu allocation fails, free also t_info to allow a retry with a
>  smaller tbuf_size
>
>  xen/common/trace.c |  249 
> ++++++++++++++++++++++-------------------------------
>  1 file changed, 104 insertions(+), 145 deletions(-)
>
> Index: xen-unstable.hg-4.1.23033/xen/common/trace.c
> ===================================================================
> --- xen-unstable.hg-4.1.23033.orig/xen/common/trace.c
> +++ xen-unstable.hg-4.1.23033/xen/common/trace.c
> @@ -42,14 +42,14 @@ CHECK_t_buf;
>  #define compat_t_rec t_rec
>  #endif
>
> -/* opt_tbuf_size: trace buffer size (in pages) */
> -static unsigned int opt_tbuf_size = 0;
> +/* opt_tbuf_size: trace buffer size (in pages) for each cpu */
> +static unsigned int opt_tbuf_size;
>  integer_param("tbuf_size", opt_tbuf_size);
>
>  /* Pointers to the meta-data objects for all system trace buffers */
>  static struct t_info *t_info;
> -#define T_INFO_PAGES 2  /* Size fixed at 2 pages for now. */
> -#define T_INFO_SIZE ((T_INFO_PAGES)*(PAGE_SIZE))
> +static unsigned int t_info_pages;
> +
>  static DEFINE_PER_CPU_READ_MOSTLY(struct t_buf *, t_bufs);
>  static DEFINE_PER_CPU_READ_MOSTLY(unsigned char *, t_data);
>  static DEFINE_PER_CPU_READ_MOSTLY(spinlock_t, t_lock);
> @@ -78,6 +78,21 @@ static u32 tb_event_mask = TRC_ALL;
>  * i.e., sizeof(_type) * ans >= _x. */
>  #define fit_to_type(_type, _x) (((_x)+sizeof(_type)-1) / sizeof(_type))
>
> +static int cpu_callback(
> +    struct notifier_block *nfb, unsigned long action, void *hcpu)
> +{
> +    unsigned int cpu = (unsigned long)hcpu;
> +
> +    if ( action == CPU_UP_PREPARE )
> +        spin_lock_init(&per_cpu(t_lock, cpu));
> +
> +    return NOTIFY_DONE;
> +}
> +
> +static struct notifier_block cpu_nfb = {
> +    .notifier_call = cpu_callback
> +};
> +
>  static void calc_tinfo_first_offset(void)
>  {
>     int offset_in_bytes = offsetof(struct t_info, mfn_offset[NR_CPUS]);
> @@ -85,20 +100,34 @@ static void calc_tinfo_first_offset(void
>  }
>
>  /**
> - * check_tbuf_size - check to make sure that the proposed size will fit
> + * calculate_tbuf_size - check to make sure that the proposed size will fit
>  * in the currently sized struct t_info and allows prod and cons to
>  * reach double the value without overflow.
> + * Initialize t_info_pages based on number of trace pages.
>  */
> -static int check_tbuf_size(u32 pages)
> +static int calculate_tbuf_size(unsigned int pages)
>  {
>     struct t_buf dummy;
>     typeof(dummy.prod) size;
> -
> -    size = ((typeof(dummy.prod))pages)  * PAGE_SIZE;
> -
> -    return (size / PAGE_SIZE != pages)
> -           || (size + size < size)
> -           || (num_online_cpus() * pages + t_info_first_offset > T_INFO_SIZE 
> / sizeof(uint32_t));
> +
> +    /* force maximum value for an unsigned type */
> +    size = -1;
> +
> +    /* max size holds up to n pages */
> +    size /= PAGE_SIZE;
> +    if ( pages > size )
> +    {
> +        gdprintk(XENLOG_INFO, "%s: requested number of %u pages reduced to 
> %u\n",
> +               __func__, pages, (unsigned int)size);
> +        pages = size;
> +    }
> +
> +    t_info_pages = num_online_cpus() * pages + t_info_first_offset;
> +    t_info_pages *= sizeof(uint32_t);
> +    t_info_pages /= PAGE_SIZE;
> +    if ( t_info_pages % PAGE_SIZE )
> +        t_info_pages++;
> +    return pages;
>  }
>
>  /**
> @@ -111,47 +140,28 @@ static int check_tbuf_size(u32 pages)
>  * This function may also be called later when enabling trace buffers
>  * via the SET_SIZE hypercall.
>  */
> -static int alloc_trace_bufs(void)
> +static int alloc_trace_bufs(unsigned int pages)
>  {
> -    int           i, cpu, order;
> -    unsigned long nr_pages;
> +    int i, cpu, order;
>     /* Start after a fixed-size array of NR_CPUS */
>     uint32_t *t_info_mfn_list;
>     int offset;
>
> -    if ( opt_tbuf_size == 0 )
> -        return -EINVAL;
> +    if ( t_info )
> +        return -EBUSY;
>
> -    if ( check_tbuf_size(opt_tbuf_size) )
> -    {
> -        printk("Xen trace buffers: tb size %d too large. "
> -               "Tracing disabled.\n",
> -               opt_tbuf_size);
> +    if ( pages == 0 )
>         return -EINVAL;
> -    }
>
> -    /* t_info size is fixed for now. Currently this works great, so there
> -     * seems to be no need to make it dynamic. */
> -    t_info = alloc_xenheap_pages(get_order_from_pages(T_INFO_PAGES), 0);
> -    if ( t_info == NULL )
> -    {
> -        printk("Xen trace buffers: t_info allocation failed! "
> -               "Tracing disabled.\n");
> -        return -ENOMEM;
> -    }
> -
> -    for ( i = 0; i < T_INFO_PAGES; i++ )
> -        share_xen_page_with_privileged_guests(
> -            virt_to_page(t_info) + i, XENSHARE_readonly);
> -
> -    t_info_mfn_list = (uint32_t *)t_info;
> -    offset = t_info_first_offset;
> +    /* Calculate offset in u32 of first mfn */
> +    calc_tinfo_first_offset();
>
> -    t_info->tbuf_size = opt_tbuf_size;
> -    printk(XENLOG_INFO "tbuf_size %d\n", t_info->tbuf_size);
> +    pages = calculate_tbuf_size(pages);
> +    order = get_order_from_pages(pages);
>
> -    nr_pages = opt_tbuf_size;
> -    order = get_order_from_pages(nr_pages);
> +    t_info = alloc_xenheap_pages(get_order_from_pages(t_info_pages), 0);
> +    if ( t_info == NULL )
> +        goto out_dealloc;
>
>     /*
>      * First, allocate buffers for all of the cpus.  If any
> @@ -159,27 +169,29 @@ static int alloc_trace_bufs(void)
>      */
>     for_each_online_cpu(cpu)
>     {
> -        int flags;
> -        char         *rawbuf;
> +        void *rawbuf;
>         struct t_buf *buf;
>
>         if ( (rawbuf = alloc_xenheap_pages(
>                 order, MEMF_bits(32 + PAGE_SHIFT))) == NULL )
>         {
> -            printk("Xen trace buffers: memory allocation failed\n");
> -            opt_tbuf_size = 0;
> +            printk("Xen trace buffers: memory allocation failed on cpu 
> %d\n", cpu);
>             goto out_dealloc;
>         }
>
> -        spin_lock_irqsave(&per_cpu(t_lock, cpu), flags);
> -
> -        per_cpu(t_bufs, cpu) = buf = (struct t_buf *)rawbuf;
> +        per_cpu(t_bufs, cpu) = buf = rawbuf;
>         buf->cons = buf->prod = 0;
>         per_cpu(t_data, cpu) = (unsigned char *)(buf + 1);
> +    }
>
> -        spin_unlock_irqrestore(&per_cpu(t_lock, cpu), flags);
> +    offset = t_info_first_offset;
> +    t_info_mfn_list = (uint32_t *)t_info;
>
> -    }
> +    for(i = 0; i < t_info_pages; i++)
> +        share_xen_page_with_privileged_guests(
> +            virt_to_page(t_info) + i, XENSHARE_readonly);
> +
> +    t_info->tbuf_size = pages;
>
>     /*
>      * Now share the pages to xentrace can map them, and write them in
> @@ -188,89 +200,75 @@ static int alloc_trace_bufs(void)
>     for_each_online_cpu(cpu)
>     {
>         /* Share pages so that xentrace can map them. */
> -        char         *rawbuf;
> +        void *rawbuf = per_cpu(t_bufs, cpu);
> +        struct page_info *p = virt_to_page(rawbuf);
> +        uint32_t mfn = virt_to_mfn(rawbuf);
>
> -        if ( (rawbuf = (char *)per_cpu(t_bufs, cpu)) )
> +        for ( i = 0; i < pages; i++ )
>         {
> -            struct page_info *p = virt_to_page(rawbuf);
> -            uint32_t mfn = virt_to_mfn(rawbuf);
> +            share_xen_page_with_privileged_guests(p + i, XENSHARE_writable);
>
> -            for ( i = 0; i < nr_pages; i++ )
> -            {
> -                share_xen_page_with_privileged_guests(
> -                    p + i, XENSHARE_writable);
> -
> -                t_info_mfn_list[offset + i]=mfn + i;
> -            }
> -            /* Write list first, then write per-cpu offset. */
> -            wmb();
> -            t_info->mfn_offset[cpu]=offset;
> -            printk(XENLOG_INFO "p%d mfn %"PRIx32" offset %d\n",
> -                   cpu, mfn, offset);
> -            offset+=i;
> +            t_info_mfn_list[offset + i]=mfn + i;
>         }
> +        t_info->mfn_offset[cpu]=offset;
> +        printk(XENLOG_INFO "p%d mfn %"PRIx32" offset %d\n",
> +               cpu, mfn, offset);
> +        offset+=i;
> +
> +        spin_lock_init(&per_cpu(t_lock, cpu));
>     }
>
> -    data_size  = (opt_tbuf_size * PAGE_SIZE - sizeof(struct t_buf));
> +    data_size  = (pages * PAGE_SIZE - sizeof(struct t_buf));
>     t_buf_highwater = data_size >> 1; /* 50% high water */
> +    opt_tbuf_size = pages;
> +
> +    register_cpu_notifier(&cpu_nfb);
> +
> +    printk("Xen trace buffers: initialised\n");
> +    wmb(); /* above must be visible before tb_init_done flag set */
> +    tb_init_done = 1;
>
>     return 0;
> +
>  out_dealloc:
>     for_each_online_cpu(cpu)
>     {
> -        int flags;
> -        char * rawbuf;
> -
> -        spin_lock_irqsave(&per_cpu(t_lock, cpu), flags);
> -        if ( (rawbuf = (char *)per_cpu(t_bufs, cpu)) )
> +        void *rawbuf = per_cpu(t_bufs, cpu);
> +        per_cpu(t_bufs, cpu) = NULL;
> +        printk("Xen trace buffers: cpu %d p %p\n", cpu, rawbuf);
> +        if ( rawbuf )
>         {
> -            per_cpu(t_bufs, cpu) = NULL;
>             ASSERT(!(virt_to_page(rawbuf)->count_info & PGC_allocated));
>             free_xenheap_pages(rawbuf, order);
>         }
> -        spin_unlock_irqrestore(&per_cpu(t_lock, cpu), flags);
>     }
> -
> +    free_xenheap_pages(t_info, get_order_from_pages(t_info_pages));
> +    t_info = NULL;
> +    printk("Xen trace buffers: allocation failed! Tracing disabled.\n");
>     return -ENOMEM;
>  }
>
>
>  /**
> - * tb_set_size - handle the logic involved with dynamically
> - * allocating and deallocating tbufs
> + * tb_set_size - handle the logic involved with dynamically allocating tbufs
>  *
>  * This function is called when the SET_SIZE hypercall is done.
>  */
> -static int tb_set_size(int size)
> +static int tb_set_size(unsigned int pages)
>  {
>     /*
>      * Setting size is a one-shot operation. It can be done either at
>      * boot time or via control tools, but not by both. Once buffers
>      * are created they cannot be destroyed.
>      */
> -    int ret = 0;
> -
> -    if ( opt_tbuf_size != 0 )
> +    if ( opt_tbuf_size && pages != opt_tbuf_size )
>     {
> -        if ( size != opt_tbuf_size )
> -            gdprintk(XENLOG_INFO, "tb_set_size from %d to %d not 
> implemented\n",
> -                     opt_tbuf_size, size);
> +        gdprintk(XENLOG_INFO, "tb_set_size from %d to %d not implemented\n",
> +                     opt_tbuf_size, pages);
>         return -EINVAL;
>     }
>
> -    if ( size <= 0 )
> -        return -EINVAL;
> -
> -    opt_tbuf_size = size;
> -
> -    if ( (ret = alloc_trace_bufs()) != 0 )
> -    {
> -        opt_tbuf_size = 0;
> -        return ret;
> -    }
> -
> -    printk("Xen trace buffers: initialized\n");
> -    return 0;
> +    return alloc_trace_bufs(pages);
>  }
>
>  int trace_will_trace_event(u32 event)
> @@ -299,21 +297,6 @@ int trace_will_trace_event(u32 event)
>     return 1;
>  }
>
> -static int cpu_callback(
> -    struct notifier_block *nfb, unsigned long action, void *hcpu)
> -{
> -    unsigned int cpu = (unsigned long)hcpu;
> -
> -    if ( action == CPU_UP_PREPARE )
> -        spin_lock_init(&per_cpu(t_lock, cpu));
> -
> -    return NOTIFY_DONE;
> -}
> -
> -static struct notifier_block cpu_nfb = {
> -    .notifier_call = cpu_callback
> -};
> -
>  /**
>  * init_trace_bufs - performs initialization of the per-cpu trace buffers.
>  *
> @@ -323,37 +306,13 @@ static struct notifier_block cpu_nfb = {
>  */
>  void __init init_trace_bufs(void)
>  {
> -    int i;
> -
> -    /* Calculate offset in u32 of first mfn */
> -    calc_tinfo_first_offset();
> -
> -    /* Per-cpu t_lock initialisation. */
> -    for_each_online_cpu ( i )
> -        spin_lock_init(&per_cpu(t_lock, i));
> -    register_cpu_notifier(&cpu_nfb);
> -
> -    if ( opt_tbuf_size == 0 )
> -    {
> -        printk("Xen trace buffers: disabled\n");
> -        goto fail;
> -    }
> -
> -    if ( alloc_trace_bufs() != 0 )
> +    if ( opt_tbuf_size && alloc_trace_bufs(opt_tbuf_size) )
>     {
> -        dprintk(XENLOG_INFO, "Xen trace buffers: "
> -                "allocation size %d failed, disabling\n",
> -                opt_tbuf_size);
> -        goto fail;
> +        gdprintk(XENLOG_INFO, "Xen trace buffers: "
> +                 "allocation size %d failed, disabling\n",
> +                 opt_tbuf_size);
> +        opt_tbuf_size = 0;
>     }
> -
> -    printk("Xen trace buffers: initialised\n");
> -    wmb(); /* above must be visible before tb_init_done flag set */
> -    tb_init_done = 1;
> -    return;
> -
> - fail:
> -    opt_tbuf_size = 0;
>  }
>
>  /**
> @@ -372,7 +331,7 @@ int tb_control(xen_sysctl_tbuf_op_t *tbc
>     case XEN_SYSCTL_TBUFOP_get_info:
>         tbc->evt_mask   = tb_event_mask;
>         tbc->buffer_mfn = t_info ? virt_to_mfn(t_info) : 0;
> -        tbc->size = T_INFO_PAGES * PAGE_SIZE;
> +        tbc->size = t_info_pages * PAGE_SIZE;
>         break;
>     case XEN_SYSCTL_TBUFOP_set_cpu_mask:
>         rc = xenctl_cpumap_to_cpumask(&tb_cpu_mask, &tbc->cpu_mask);
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel
>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.