MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <18368.6678.827830.689229@domain.hid> Date: Sat, 23 Feb 2008 14:05:26 +0100 In-Reply-To: <47C01300.9070204@domain.hid> References: <47BEA5C3.9050100@domain.hid> <18367.7523.207507.331701@domain.hid> <47C01300.9070204@domain.hid> From: Gilles Chanteperdrix Subject: Re: [Adeos-main] [PATCH] tracer: use per-cpu trace buffers List-Id: General discussion about Adeos List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Jan Kiszka Cc: Jan Kiszka , adeos-main , Philippe Gerum Jan Kiszka wrote: > Gilles Chanteperdrix wrote: > > Jan Kiszka wrote: > > > This patch addresses two issues: > > > > > > - fixes boot hang with CONFIG_IPIPE_TRACE, !CONFIG_IPIPE_TRACE_VMALLOC, > > > and large CONFIG_NR_CPUS (due to the huge trace buffers this implies, > > > even if the system has only a few CPUs in fact) > > > > This is something I observed even on UP systems: the kernel does not > > boot on ARM if CONFIG_IPIPE_TRACE_VMALLOC is not selected. > > > > Yeah, well known, that's why we have vmalloc'ed trace buffers. > > I just wondered again if this is an arch-specific issue, or if it > depends on the RAM size. If it is arch-specific, you are welcome to > change the default of IPIPE_TRACE_VMALLOC ("default y if ARM || ..."). > Or reduce IPIPE_TRACE_SHIFT for ARM, given there is a reasonable value > (/wrt to the tracer) your boards still boot with. I will do some tests and propose something like that. I had this problem with linux 2.6.15 where the initial image (I guess text + data + bss) must fit in 3 MB. I have not tested since then. -- Gilles Chanteperdrix.