xentrace: broken by c/s 23050:4ebba54b666f

xentrace: broken by c/s 23050:4ebba54b666f

[Christoph Egger]

Hi,

C/s 23050:4ebba54b666f broke xentrace:

# xentrace -e 0x8f000 ./xentrace.out
(XEN) Xen trace buffers: cpu 0 p 0000000000000000
(XEN) Xen trace buffers: cpu 1 p 0000000000000000
(XEN) Xen trace buffers: cpu 2 p 0000000000000000
(XEN) Xen trace buffers: cpu 3 p 0000000000000000
(XEN) Xen trace buffers: cpu 4 p 0000000000000000
(XEN) Xen trace buffers: cpu 5 p 0000000000000000
(XEN) Xen trace buffers: cpu 6 p 0000000000000000
(XEN) Xen trace buffers: cpu 7 p 0000000000000000
(XEN) Xen trace buffers: allocation failed! Tracing disabled.
Couldn't enable trace buffers: Invalid argument
#

Christoph

-- 
---to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Einsteinring 24, 85689 Dornach b. Muenchen
Geschaeftsfuehrer: Alberto Bozzo, Andrew Bowd
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632

Re: xentrace: broken by c/s 23050:4ebba54b666f

[Olaf Hering]

On Mon, Mar 21, Christoph Egger wrote:

> 
> Hi,
> 
> C/s 23050:4ebba54b666f broke xentrace:
> 
> # xentrace -e 0x8f000 ./xentrace.out
> (XEN) Xen trace buffers: cpu 0 p 0000000000000000

The system you test on has 8 cpus, like mine.
t_info_pages got too big, so alloc_xenheap_pages() failed.
Can you add a printk to se whats happening in calculate_tbuf_size()?

Olaf

Re: xentrace: broken by c/s 23050:4ebba54b666f

[Olaf Hering]

On Mon, Mar 21, Olaf Hering wrote:

> On Mon, Mar 21, Christoph Egger wrote:
> 
> > 
> > Hi,
> > 
> > C/s 23050:4ebba54b666f broke xentrace:
> > 
> > # xentrace -e 0x8f000 ./xentrace.out
> > (XEN) Xen trace buffers: cpu 0 p 0000000000000000
> 
> The system you test on has 8 cpus, like mine.
> t_info_pages got too big, so alloc_xenheap_pages() failed.
> Can you add a printk to se whats happening in calculate_tbuf_size()?

compile tested patch:


# HG changeset patch
# Parent c81f0ef5a77d90fbf108d3efe489d08df45b63c2
xentrace: print calculated numbers in calculate_tbuf_size()

Print number of pages to allocate for per-cpu tracebuffer and metadata
to ease debugging when allocation fails.

Signed-off-by: Olaf Hering <olaf@aepfle.de>

diff -r c81f0ef5a77d xen/common/trace.c
--- a/xen/common/trace.c	Mon Mar 21 14:52:27 2011 +0000
+++ b/xen/common/trace.c	Mon Mar 21 21:49:52 2011 +0100
@@ -127,6 +127,8 @@
     t_info_pages /= PAGE_SIZE;
     if ( t_info_pages % PAGE_SIZE )
         t_info_pages++;
+    gdprintk(XENLOG_INFO, "requested %u t_info_pages for %u trace pages on %u cpus\n",
+               t_info_pages, pages, num_online_cpus());
     return pages;
 }

Re: xentrace: broken by c/s 23050:4ebba54b666f

[Olaf Hering]

On Mon, Mar 21, Olaf Hering wrote:

> On Mon, Mar 21, Olaf Hering wrote:
> 
> > On Mon, Mar 21, Christoph Egger wrote:
> > 
> > > 
> > > Hi,
> > > 
> > > C/s 23050:4ebba54b666f broke xentrace:
> > > 
> > > # xentrace -e 0x8f000 ./xentrace.out
> > > (XEN) Xen trace buffers: cpu 0 p 0000000000000000
> > 
> > The system you test on has 8 cpus, like mine.
> > t_info_pages got too big, so alloc_xenheap_pages() failed.
> > Can you add a printk to se whats happening in calculate_tbuf_size()?
> 
> compile tested patch:

> +    gdprintk(XENLOG_INFO, "requested %u t_info_pages for %u trace pages on %u cpus\n",
> +               t_info_pages, pages, num_online_cpus());

I can reproduce it now. For some reason the line above will print
nothing. A plain printk gives:

(XEN) requested 0 t_info_pages for 32 trace pages on 8 cpus

I will investigate.

Olaf

Re: xentrace: broken by c/s 23050:4ebba54b666f

[Olaf Hering]

On Mon, Mar 21, Christoph Egger wrote:

> C/s 23050:4ebba54b666f broke xentrace:

Christoph, I sent a series of xentrace patches yesterday. The first one
which checks for t_info_pages becoming zero should fix the issue you are
seeing.

Olaf