From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754403AbdKJXfW (ORCPT ); Fri, 10 Nov 2017 18:35:22 -0500 Received: from mx2.suse.de ([195.135.220.15]:42049 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751038AbdKJXfU (ORCPT ); Fri, 10 Nov 2017 18:35:20 -0500 Date: Sat, 11 Nov 2017 00:35:15 +0100 From: "Luis R. Rodriguez" To: Pali =?iso-8859-1?Q?Roh=E1r?= Cc: "Luis R. Rodriguez" , Ming Lei , Greg Kroah-Hartman , Kalle Valo , David Gnedt , Michal Kazior , Daniel Wagner , Tony Lindgren , Sebastian Reichel , Pavel Machek , Ivaylo Dimitrov , Aaro Koskinen , Grazvydas Ignotas , linux-kernel@vger.kernel.org, linux-wireless@vger.kernel.org, netdev@vger.kernel.org Subject: Re: [PATCH v2 5/6] firmware: Add request_firmware_prefer_user() function Message-ID: <20171110233515.GU22894@wotan.suse.de> References: <1482598381-16513-1-git-send-email-pali.rohar@gmail.com> <1510270708-14377-1-git-send-email-pali.rohar@gmail.com> <1510270708-14377-6-git-send-email-pali.rohar@gmail.com> <20171110202601.GR22894@wotan.suse.de> <20171110210819.ogupue5wlzm2ar3i@pali> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: <20171110210819.ogupue5wlzm2ar3i@pali> User-Agent: Mutt/1.6.0 (2016-04-01) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, Nov 10, 2017 at 10:08:19PM +0100, Pali Rohár wrote: > On Friday 10 November 2017 21:26:01 Luis R. Rodriguez wrote: > > On Fri, Nov 10, 2017 at 12:38:27AM +0100, Pali Rohár wrote: > > > This function works pretty much like request_firmware(), but it prefer > > > usermode helper. If usermode helper fails then it fallback to direct > > > access. Useful for dynamic or model specific firmware data. > > > > > > Signed-off-by: Pali Rohár > > > --- > > > drivers/base/firmware_class.c | 45 +++++++++++++++++++++++++++++++++++++++-- > > > include/linux/firmware.h | 9 +++++++++ > > > 2 files changed, 52 insertions(+), 2 deletions(-) > > > > > > diff --git a/drivers/base/firmware_class.c b/drivers/base/firmware_class.c > > > index 4b57cf5..c3a9fe5 100644 > > > --- a/drivers/base/firmware_class.c > > > +++ b/drivers/base/firmware_class.c > > > @@ -195,6 +195,11 @@ static int __fw_state_check(struct fw_state *fw_st, enum fw_status status) > > > #endif > > > #define FW_OPT_NO_WARN (1U << 3) > > > #define FW_OPT_NOCACHE (1U << 4) > > > +#ifdef CONFIG_FW_LOADER_USER_HELPER > > > +#define FW_OPT_PREFER_USER (1U << 5) > > > +#else > > > +#define FW_OPT_PREFER_USER 0 > > > +#endif > > > > I've been cleaning these up these flags [0], which I'll shortly respin based > > on feedback, so this sort of stuff should be avoided at all costs. > > > > Regardless of this even if you *leave* the flag in place and a driver required > > this, but the kernel was compiled without CONFIG_FW_LOADER_USER_HELPER then > > calling fw_load_from_user_helper would just already return -ENOENT, as such it > > would in turn fallback to direct fs loading so the #ifef'ery seems to be not > > needed. > > > > [0] https://lkml.kernel.org/r/20170914225422.31034-1-mcgrof@kernel.org > > > > > struct firmware_cache { > > > /* firmware_buf instance will be added into the below list */ > > > @@ -1214,13 +1219,26 @@ static void fw_abort_batch_reqs(struct firmware *fw) > > > if (ret <= 0) /* error or already assigned */ > > > goto out; > > > > > > - ret = fw_get_filesystem_firmware(device, fw->priv); > > > + if (opt_flags & FW_OPT_PREFER_USER) { > > > + ret = fw_load_from_user_helper(fw, name, device, opt_flags, timeout); > > > + if (ret && !(opt_flags & FW_OPT_NO_WARN)) { > > > + dev_warn(device, > > > + "User helper firmware load for %s failed with error %d\n", > > > + name, ret); > > > + dev_warn(device, "Falling back to direct firmware load\n"); > > > > As I had noted before, the usermode helper was really not well designed, > > as such extending further use of it is something we should shy away unless we > > determine its completely necessary. > > > > So what's wrong with this driver failing at direct access, which should be fast, > > and relying on a uevent to then work using the current fallback mechanisms? > > > > The commit log in no way documents any of the justifications for further > > extending use of the usermode helper. > > Hi! See patch 6/6. It is needed to avoid direct access and wl1251 on > Nokia N900 needs to use userspace helper which prepares firmware data. My point is your commit log in no way describes the shortcomings of the current affairs for device drivers which only can access the data it needs using the firmware fallback mechanism. In order for a change to go in, specially if its extending use of the fallback mechanism through sysfs now as primary citizen, the justification should be well documented on the commit log. For instance you may want to highlight that what I documented here: https://www.kernel.org/doc/html/latest/driver-api/firmware/fallback-mechanisms.html "CONFIG_FW_LOADER_USER_HELPER: enables building the firmware fallback mechanism. Most distributions enable this option today. If enabled but CONFIG_FW_LOADER_USER_HELPER_FALLBACK is disabled, only the custom fallback mechanism is available and for the request_firmware_nowait() call." Since most distros disable CONFIG_FW_LOADER_USER_HELPER_FALLBACK, in practice this means the fallback mechanism is never actually triggered, and the only way to do that in practice this for most distros is to use the custom fallback mechanism, and the only difference there is that the custom fallback mechanism has an infinite timeout since we have no clear way to know what it is or when it will complete, other than when it actually does its work. That begs the question, why cant you just use request_firmware_nowait() with the custom fallback mechanism? Your commit log should explain the shortcomings of the current API. Also note that "usermode helper" refers to kernel/umh.c, and the only code being used from that UMH API is the usermodehelper_read_lock_wait() on async, or usermodehelper_read_trylock() on sync. Nothing else. The rest of the firmware uploading is done via a sysfs file upload mechanism. Its precisely why I documented the fallback mechanism instead as: "Firmware sysfs loading facility" Adding any other flag which reflects "UMH" only will confuse people further. If you are going to add a new flag, perhaps "FW_OPT_PREFER_SYSFS_LOAD" is much more suitable. > Direct access is just fallback when userspace helper is not available. > Without userspace helper on devices where wl1251 do not have own eeprom > memory, wl1251 cannot work. That is not an explanation as to why request_firmware_nowait() could not be used with the custom fallback interface. > I know that usermode helper is not well designed, but it is the best > option what we can do for wl1251. It does not seem like you have tested the custom fallback mechanism, nor have you documented properly the justification for extending and making use of the sysfs loading facility a first option for loading firmware. Luis