From mboxrd@z Thu Jan 1 00:00:00 1970 From: Huang Ying Subject: Re: ERST: how to avoid a dynamic memory allocation in panic case Date: Wed, 23 Jan 2013 08:39:54 +0800 Message-ID: <1358901594.7408.85.camel@yhuang-dev> References: Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from mga11.intel.com ([192.55.52.93]:29438 "EHLO mga11.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751077Ab3AWAkR (ORCPT ); Tue, 22 Jan 2013 19:40:17 -0500 In-Reply-To: Sender: linux-acpi-owner@vger.kernel.org List-Id: linux-acpi@vger.kernel.org To: Seiji Aguchi Cc: "rjw@sisk.pl" , "linux-kernel@vger.kernel.org" , "linux-acpi@vger.kernel.org" , "H. Peter Anvin (hpa@zytor.com)" , "robert.moore@intel.com" , "trenn@suse.de" , "myron.stowe@redhat.com" On Tue, 2013-01-22 at 23:30 +0000, Seiji Aguchi wrote: > [Issue] > > Current erst driver is kicked in panic case. > On the other hand, a dynamic memory allocation seems to run in it > with a following code path. > > erst_writer -> erst_write -> __erst_write_to_storage -> many of apei_exec_run -> > ctx->ins_table[entry->instruction].run() > > which are functions defined in erst_ins_type table i.e: > > apei_exec_read_register -> apei_read -> acpi_os_read_memory64 -> acpi_os_ioremap -> ioremap_cache > > which possibly involve IOMMU allocator. > > I may cause a failure of an erst driver if the panic happens in an interrupt context. > > [Idea] > > If we can remove ioremap_cache() from acpi_os_read_memory() as follows, > It is easy to fix this issue. > But I'm not sure if it is feasible because I can't see the reason of tying acpi_os_ioremap() from a git log... > > Any comment? > > > @@ -918,10 +918,7 @@ acpi_os_read_memory(acpi_physical_address phys_addr, u64 *value, u32 width) > virt_addr = acpi_map_vaddr_lookup(phys_addr, size); > if (!virt_addr) { > rcu_read_unlock(); > - virt_addr = acpi_os_ioremap(phys_addr, size); > - if (!virt_addr) > - return AE_BAD_ADDRESS; > - unmap = true; > + return AE_BAD_ADDRESS; No. We can not do that. Because some users rely on acpi_os_read_memory to do ioremap for them. The correct fixing should be pre-map the io-memory that may be accessed in erst code patch with acpi_map(). Best Regards, Huang Ying > } > > if (!value) > > > Seiji