Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758112Ab2EPGPu (ORCPT ); Wed, 16 May 2012 02:15:50 -0400 Received: from mail-pz0-f46.google.com ([209.85.210.46]:40532 "EHLO mail-pz0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751335Ab2EPGPs (ORCPT ); Wed, 16 May 2012 02:15:48 -0400 Date: Tue, 15 May 2012 23:14:17 -0700 From: Anton Vorontsov To: Kees Cook Cc: Greg Kroah-Hartman , Colin Cross , Arnd Bergmann , John Stultz , arve@android.com, Rebecca Schultz Zavin , Jesper Juhl , Randy Dunlap , Stephen Boyd , Thomas Meyer , Andrew Morton , Marco Stornelli , WANG Cong , linux-kernel@vger.kernel.org, devel@driverdev.osuosl.org, linaro-kernel@lists.linaro.org, patches@linaro.org, kernel-team@android.com Subject: Re: [PATCH 10/11] pstore/ram: Switch to persistent_ram routines Message-ID: <20120516061416.GB18058@lizard> References: <20120512001506.GA8653@lizard> <20120512001840.GJ14782@lizard> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 4346 Lines: 112 Hello Kees, On Mon, May 14, 2012 at 03:21:17PM -0700, Kees Cook wrote: [...] > > -       buf = cxt->virt_addr + (id * cxt->record_size); > > -       memset(buf, '\0', cxt->record_size); > > +       persistent_ram_free_old(cxt->przs[id]); > > Hm, I don't think persistent_ram_free_old() is what's wanted here. > That appears to entirely release the region? I want to make sure the > memory is cleared first. And will this area come back on a write, or > does it stay released? It just releases ECC-restored memory region (a copy). The original (persistent) region is still fully reusable after that call. (It is a pity that pstore internals can't use the restored copy directly, as pstore expects that it will release the region itself after pstore_mkfile(), so we somewhat duplicate the memory during psi->read(). We'd better fix it some day, but it's a minor issue so far.) > > > >        return 0; > >  } > > @@ -200,6 +203,7 @@ static int __init ramoops_probe(struct platform_device *pdev) > >        struct ramoops_platform_data *pdata = pdev->dev.platform_data; > >        struct ramoops_context *cxt = &oops_cxt; > >        int err = -EINVAL; > > +       int i; > > > >        /* Only a single ramoops area allowed at a time, so fail extra > >         * probes. > > @@ -237,32 +241,37 @@ static int __init ramoops_probe(struct platform_device *pdev) > >        cxt->record_size = pdata->record_size; > >        cxt->dump_oops = pdata->dump_oops; > > > > +       cxt->przs = kzalloc(sizeof(*cxt->przs) * cxt->max_count, GFP_KERNEL); > > +       if (!cxt->przs) { > > +               pr_err("failed to initialize a prz array\n"); > > +               goto fail_przs; > > This should be fail_out. Thanks, will fix all of these error handling negligences. > > +       } > > + > > +       for (i = 0; i < cxt->max_count; i++) { > > +               size_t sz = cxt->record_size; > > +               phys_addr_t start = cxt->phys_addr + sz * i; > > + > > +               cxt->przs[i] = persistent_ram_new(start, sz, 0); > > persistent_ram_new() is marked as __init, so this is unsafe to call if > built as a module. I think persistent_ram_new() will need to lose the > __init marking, or I'm misunderstanding something. Um. ramoops' probe routine is also __init. persistent_ram_new is a part of ramoops module, so their __init functions will be discarded at the same time. ram_console can't be a module, so it is also fine. So I think it's all fine. > > +               if (IS_ERR(cxt->przs[i])) { > > +                       err = PTR_ERR(cxt->przs[i]); > > +                       pr_err("failed to initialize a prz\n"); > > Since neither persistent_ram_new() nor persistent_ram_buffer_map() > report the location of the failure, I'd like to keep the error report > (removed below "pr_err("request mem region (0x%lx@0x%llx) > failed\n",...") for failures, so there is something actionable in > dmesg when the platform data is mismatched for the hardware. Sure thing, will do. I'll also start using dev_err() for new code, that way it's more clearer which module reported the error. [...] > >        cxt->pstore.data = cxt; > > -       cxt->pstore.bufsize = cxt->record_size; > > -       cxt->pstore.buf = kmalloc(cxt->pstore.bufsize, GFP_KERNEL); > >        spin_lock_init(&cxt->pstore.buf_lock); > > +       cxt->pstore.bufsize = cxt->przs[0]->buffer_size; > > +       cxt->pstore.buf = kmalloc(cxt->pstore.bufsize, GFP_KERNEL); > > I don't see a reason to re-order these (nothing can use buf yet > because we haven't registered it with pstore yet). Yeah, this is a left over. Thank for catching. [...] > > +fail_przs: > > +       for (i = 0; cxt->przs[i]; i++) > > +               persistent_ram_free(cxt->przs[i]); > > This can lead to a BUG, since persistent_ram_free() doesn't handle > NULL arguments. The for loop has 'cxt->przs[i]' condition. :-) Thanks for the review! -- Anton Vorontsov Email: cbouatmailru@gmail.com -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/