| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAPDyKFr94KXVyazcC93Av0tc32cseR79GAxJ2vAFJa9abvDfjA@mail.gmail.com>
Date: Tue, 12 Jul 2022 12:20:35 +0200
From: Ulf Hansson <ulf.hansson@...aro.org>
To: Vincent Whitchurch <vincent.whitchurch@...s.com>
Cc: kernel <kernel@...s.com>,
"linux-mmc@...r.kernel.org" <linux-mmc@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] mmc: core: Allow speed modes to be adjusted via module param
On Mon, 27 Jun 2022 at 12:08, Vincent Whitchurch
<vincent.whitchurch@...s.com> wrote:
>
> On Thu, Jun 23, 2022 at 03:53:41PM +0200, Ulf Hansson wrote:
> > On Thu, 23 Jun 2022 at 10:00, Vincent Whitchurch
> > <vincent.whitchurch@...s.com> wrote:
> > > During board verification, there is a need to test the various supported
> > > eMMC/SD speed modes. However, since the framework chooses the best mode
> > > supported by the card and the host controller's caps, this currently
> > > necessitates changing the devicetree for every iteration.
> > >
> > > To make changing the modes easier, allow the various host controller
> > > capabilities to be cleared via a module parameter. (A per-controller
> > > debugfs wouldn't work since the controller needs to be re-probed to
> > > trigger re-init of cards. A module parameter is used instead of a
> >
> > I think we could make use of a per-controller debugfs thing, if used
> > in combination with MMC_CAP_AGGRESSIVE_PM and runtime PM.
> >
> > As runtime PM also has sysfs interface for each device, we can control
> > runtime PM for the card's device (to trigger re-initialization of the
> > card). In between runtime suspend/resume of the card's device, we
> > should be able to change the supported speed modes, through debug fs.
> >
> > Would this work for you?
>
> I got it to work with the below commands and the following patch. Note
> that:
>
> (1) MMC_CAP_AGGRESSIVE_PM also needs to be turned on via debugfs to
> avoid having to patch the kernel. The cap is checked on every call
> to runtime_suspend so it (currently) works to set it without
> re-probing the host.
>
> (2) I had to call mmc_select_card_type() even if there is an old card
> since currently it's only called from mmc_decode_ext_csd().
>
> Also, unlike the module parameter, this can't be set from bootargs, but
> that part is not important for my use case.
>
> root@(none):/sys/kernel/debug/mmc0# grep timing ios
> timing spec: 9 (mmc HS200)
>
> // Turn on MMC_CAP_AGGRESSIVE_PM and re-trigger runtime suspend
> root@(none):/sys/kernel/debug/mmc0# echo $(($(cat caps) | (1 << 7))) > caps
> root@(none):/sys/kernel/debug/mmc0# echo on > /sys/bus/mmc/devices/mmc0\:0001/power/control
> root@(none):/sys/kernel/debug/mmc0# echo auto > /sys/bus/mmc/devices/mmc0\:0001/power/control
>
> // MMC_CAP2_HS200_1_8V_SDR
> root@(none):/sys/kernel/debug/mmc0# echo $(($(cat caps2) & ~(1 << 5))) > caps2
> root@(none):/sys/kernel/debug/mmc0# echo on > /sys/bus/mmc/devices/mmc0\:0001/power/control
> root@(none):/sys/kernel/debug/mmc0# grep timing ios
> timing spec: 8 (mmc DDR52)
>
> 8<----
> diff --git a/drivers/mmc/core/debugfs.c b/drivers/mmc/core/debugfs.c
> index 3fdbc801e64a..721925300611 100644
> --- a/drivers/mmc/core/debugfs.c
> +++ b/drivers/mmc/core/debugfs.c
> @@ -12,9 +12,12 @@
> #include <linux/slab.h>
> #include <linux/stat.h>
> #include <linux/fault-inject.h>
> +#include <linux/time.h>
>
> #include <linux/mmc/card.h>
> #include <linux/mmc/host.h>
> +#include <linux/mmc/mmc.h>
> +#include <linux/mmc/sd.h>
>
> #include "core.h"
> #include "card.h"
> @@ -223,6 +226,47 @@ static int mmc_clock_opt_set(void *data, u64 val)
> DEFINE_DEBUGFS_ATTRIBUTE(mmc_clock_fops, mmc_clock_opt_get, mmc_clock_opt_set,
> "%llu\n");
>
> +static int mmc_caps_get(void *data, u64 *val)
> +{
> + *val = *(u32 *)data;
> + return 0;
> +}
> +
> +static int mmc_caps_set(void *data, u64 val)
> +{
> + u32 *caps = data;
> + u32 diff = *caps ^ val;
> + u32 allowed = MMC_CAP_AGGRESSIVE_PM |
> + MMC_CAP_SD_HIGHSPEED |
> + MMC_CAP_MMC_HIGHSPEED |
> + MMC_CAP_UHS |
> + MMC_CAP_DDR;
> +
> + if (diff & ~allowed)
> + return -EINVAL;
> +
> + *caps = val;
> +
> + return 0;
> +}
> +
> +static int mmc_caps2_set(void *data, u64 val)
> +{
> + u32 *caps = data;
> + u32 diff = *caps ^ val;
> + u32 allowed = MMC_CAP2_HSX00_1_8V | MMC_CAP2_HSX00_1_2V;
> +
> + if (diff & ~allowed)
> + return -EINVAL;
> +
> + *caps = val;
> +
> + return 0;
> +}
> +
> +DEFINE_DEBUGFS_ATTRIBUTE(mmc_caps_fops, mmc_caps_get, mmc_caps_set, "0x%08llx\n");
> +DEFINE_DEBUGFS_ATTRIBUTE(mmc_caps2_fops, mmc_caps_get, mmc_caps2_set, "0x%08llx\n");
> +
> void mmc_add_host_debugfs(struct mmc_host *host)
> {
> struct dentry *root;
> @@ -231,8 +275,10 @@ void mmc_add_host_debugfs(struct mmc_host *host)
> host->debugfs_root = root;
>
> debugfs_create_file("ios", S_IRUSR, root, host, &mmc_ios_fops);
> - debugfs_create_x32("caps", S_IRUSR, root, &host->caps);
> - debugfs_create_x32("caps2", S_IRUSR, root, &host->caps2);
> + debugfs_create_file("caps", S_IRUSR | S_IWUSR, root, &host->caps,
> + &mmc_caps_fops);
> + debugfs_create_file("caps2", S_IRUSR | S_IWUSR, root, &host->caps2,
> + &mmc_caps2_fops);
> debugfs_create_file_unsafe("clock", S_IRUSR | S_IWUSR, root, host,
> &mmc_clock_fops);
>
> diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
> index 89cd48fcec79..c79c26add3da 100644
> --- a/drivers/mmc/core/mmc.c
> +++ b/drivers/mmc/core/mmc.c
> @@ -1730,7 +1730,8 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
>
> /* Erase size depends on CSD and Extended CSD */
> mmc_set_erase_size(card);
> - }
> + } else
> + mmc_select_card_type(card);
>
> /* Enable ERASE_GRP_DEF. This bit is lost after a reset or power off. */
> if (card->ext_csd.rev >= 3) {
Overall, the approach seems reasonable to me. Perhaps we can
reorganize the code so mmc_select_card_type() is always called from
mmc_init_card(), rather than from mmc_decode_ext_csd() too.
Kind regards
Uffe
Powered by blists - more mailing lists