sh: Add ftrace syscall tracing support (was: Re: [ltt-dev] [PATCH] sh: added LTT_DUMP_TABLES support)

sh: Add ftrace syscall tracing support (was: Re: [ltt-dev] [PATCH] sh: added LTT_DUMP_TABLES support)

[Mathieu Desnoyers]

* Giuseppe CAVALLARO (peppe.cavallaro@st.com) wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hi Mathieu,
> 
> Mathieu Desnoyers wrote:
> > So now we only need to fix entry-common.S to get everything working.
> 
> I've just noticed that this has been already done:
> http://git.kernel.org/?p=linux%2Fkernel%2Fgit%2Flethal%2Fsh-2.6.git;a=commitdiff_plain;h=c652d780c9cf7f860141de232b37160fe013feca;hp=c1340c053be7a43d837a3acb352d5008be865a55
> I think it will be necessary to to re-add the _TIF_KERNEL_TRACE against
> this modifications, only.
> 

Hrm, I don't like the approach taken there. It extends the bitmasks for
_TIF_WORK_SYSCALL_MASK and _TIF_ALLWORK_MASK beyond 8 bits, and adds a &
0xff in the assembly whenever needed. This begs for a bug to be
introduced.

If we need something like that on SH, we should probably declare a 8-bit
only bitmask for those assembly sites which does not need the
FTRACE_SYSCALL thread flag, and another mask for sites which need to
test the 2 bytes.

I would also consider other instructions than 2 consecutive "tst" to
extend test with a 2 bytes mask. Something like loading an integer in a
register and testing with the register rather than doing 2 immediate
value tests.

Mathieu


> Regards,
> Peppe
> 
> > 
> > Mathieu
> > 
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
> 
> iEYEARECAAYFAkp39YMACgkQ2Xo3j31MSSJWmQCeJfvmg3cUHsGGy5QNdRcOJbO9
> C1AAnR6gAuKZhaRt4+4UWiWDGucgD2DJ
> =9cBd
> -----END PGP SIGNATURE-----
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

Re: sh: Add ftrace syscall tracing support (was: Re: [ltt-dev] [PATCH] sh: added LTT_DUMP_TABLES support)

[Matt Fleming]

On Tue, Aug 04, 2009 at 09:35:15AM -0400, Mathieu Desnoyers wrote:
> * Giuseppe CAVALLARO (peppe.cavallaro@st.com) wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> > 
> > Hi Mathieu,
> > 
> > Mathieu Desnoyers wrote:
> > > So now we only need to fix entry-common.S to get everything working.
> > 
> > I've just noticed that this has been already done:
> > http://git.kernel.org/?p=linux%2Fkernel%2Fgit%2Flethal%2Fsh-2.6.git;a=commitdiff_plain;h=c652d780c9cf7f860141de232b37160fe013feca;hp=c1340c053be7a43d837a3acb352d5008be865a55
> > I think it will be necessary to to re-add the _TIF_KERNEL_TRACE against
> > this modifications, only.
> > 
> 
> Hrm, I don't like the approach taken there. It extends the bitmasks for
> _TIF_WORK_SYSCALL_MASK and _TIF_ALLWORK_MASK beyond 8 bits, and adds a &
> 0xff in the assembly whenever needed. This begs for a bug to be
> introduced.
> 
> If we need something like that on SH, we should probably declare a 8-bit
> only bitmask for those assembly sites which does not need the
> FTRACE_SYSCALL thread flag, and another mask for sites which need to
> test the 2 bytes.
> 

If I remember correctly, I originally split TIF_WORK_SYSCALL_MASK into a
high 8-bit and low 8-bit chunks. However, I think I remember Paul saying
that wasn't the best way to do it.

I've no issue with doing it either way.