From: Mariusz Tkaczyk <mariusz.tkaczyk@linux.intel.com>
To: Kevin Friedberg <kev.friedberg@gmail.com>
Cc: linux-raid@vger.kernel.org
Subject: Re: [PATCH] treat AHCI controllers under VMD as part of VMD
Date: Tue, 7 Feb 2023 09:44:23 +0100 [thread overview]
Message-ID: <20230207094423.00001a30@linux.intel.com> (raw)
In-Reply-To: <20230126021659.3801-1-kev.friedberg@gmail.com>
Hi Kevin,
I found time to take a look into it closer. I think that it is not complete
solution. Please see my comments.
On Wed, 25 Jan 2023 21:16:59 -0500
Kevin Friedberg <kev.friedberg@gmail.com> wrote:
> Detect when a SATA controller has been mapped under Intel Alderlake RST
> VMD and list it as part of the domain, instead of independently, so that
> it can use the VMD controller's RAID capabilities.
>
> Signed-off-by: Kevin Friedberg <kev.friedberg@gmail.com>
> ---
> platform-intel.c | 15 +++++++++------
> super-intel.c | 25 ++++++++++++++++++++++++-
> 2 files changed, 33 insertions(+), 7 deletions(-)
>
> diff --git a/platform-intel.c b/platform-intel.c
> index 757f0b1b..859bf743 100644
> --- a/platform-intel.c
> +++ b/platform-intel.c
> @@ -64,10 +64,12 @@ struct sys_dev *find_driver_devices(const char *bus,
> const char *driver)
> if (strcmp(driver, "isci") == 0)
> type = SYS_DEV_SAS;
> - else if (strcmp(driver, "ahci") == 0)
> + else if (strcmp(driver, "ahci") == 0) {
> + /* if looking for sata devs, ignore vmd */
> + vmd = find_driver_devices("pci", "vmd");
> type = SYS_DEV_SATA;
> - else if (strcmp(driver, "nvme") == 0) {
> - /* if looking for nvme devs, first look for vmd */
> + } else if (strcmp(driver, "nvme") == 0) {
> + /* if looking for nvme devs, also look for vmd */
> vmd = find_driver_devices("pci", "vmd");
> type = SYS_DEV_NVME;
> } else if (strcmp(driver, "vmd") == 0)
> @@ -104,8 +106,8 @@ struct sys_dev *find_driver_devices(const char *bus,
> const char *driver) sprintf(path, "/sys/bus/%s/drivers/%s/%s",
> bus, driver, de->d_name);
>
> - /* if searching for nvme - skip vmd connected one */
> - if (type == SYS_DEV_NVME) {
> + /* if searching for nvme or ahci - skip vmd connected one */
> + if (type == SYS_DEV_NVME || type == SYS_DEV_SATA) {
> struct sys_dev *dev;
> char *rp = realpath(path, NULL);
> for (dev = vmd; dev; dev = dev->next) {
> @@ -166,7 +168,8 @@ struct sys_dev *find_driver_devices(const char *bus,
> const char *driver) }
> closedir(driver_dir);
>
> - if (vmd) {
> + /* VMD adopts multiple types but should only be listed once */
> + if (vmd && type == SYS_DEV_NVME) {
> if (list)
> list->next = vmd;
> else
The SATA behind VMD deserves own type, let say SYS_DEV_SATA_VMD. We cannot use
SYS_DEV_VMD because it will allow to use NVME devices behind VMD in SATA Raid
array. It means that if you have them connected, like:
VMD___ NVME0
|_ NVME1
|_ SATA___SATA0
|__SATA1
You will be able to mix SATA and NVME drives together in RAID. Mdmonitor
could mix them too (if appropriate policy is set). That is not allowed from at
least VROC requirements PoV.
> diff --git a/super-intel.c b/super-intel.c
> index 89fac626..4ef8f0d8 100644
> --- a/super-intel.c
> +++ b/super-intel.c
> @@ -2680,6 +2680,8 @@ static void print_imsm_capability_export(const struct
> imsm_orom *orom) printf("IMSM_MAX_VOLUMES_PER_CONTROLLER=%d\n",orom->vphba);
> }
>
> +#define PCI_CLASS_AHCI_CNTRL "0x010601"
This should be defined in platform-intel.h
> +
> static int detail_platform_imsm(int verbose, int enumerate_only, char
> *controller_path) {
> /* There are two components to imsm platform support, the ahci SATA
> @@ -2752,11 +2754,32 @@ static int detail_platform_imsm(int verbose, int
> enumerate_only, char *controlle for (hba = list; hba; hba = hba->next) {
> if (hba->type == SYS_DEV_VMD) {
> char buf[PATH_MAX];
> + struct dirent *ent;
> + DIR *dir;
> +
> printf(" I/O Controller : %s (%s)\n",
> vmd_domain_to_controller(hba,
> buf), get_sys_dev_type(hba->type));
> + dir = opendir(hba->path);
> + for (ent = readdir(dir); ent; ent =
> readdir(dir)) {
> + char ent_path[PATH_MAX];
> +
> + sprintf(ent_path, "%s/%s",
> hba->path, ent->d_name);
> + devpath_to_char(ent_path,
> "class", buf, sizeof(buf), 0);
> + if (strcmp(buf,
> PCI_CLASS_AHCI_CNTRL) == 0) {
> + host_base =
> ahci_get_port_count(ent_path, &port_count);
> + if
> (ahci_enumerate_ports(ent_path, port_count, host_base, verbose)) {
> + if (verbose
> > 0)
> +
> pr_err("failed to enumerate ports on VMD SATA controller at %s.\n",
> +
> hba->pci_id);
> + result |= 2;
> + }
> + }
> + }
> + closedir(dir);
> +
> if (print_nvme_info(hba)) {
> if (verbose > 0)
> - pr_err("failed to
> get devices attached to VMD domain.\n");
> + pr_err("failed to
> get NVMe devices attached to VMD domain.\n"); result |= 2;
> }
> }
Similar logic is already there:
https://git.kernel.org/pub/scm/utils/mdadm/mdadm.git/tree/super-intel.c#n2780
I would like to reuse it instead of adding new code branch. Could you please
extract similar parts to new function and use it in both places?
Thanks,
Mariusz
next prev parent reply other threads:[~2023-02-07 8:44 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-26 2:16 [PATCH] treat AHCI controllers under VMD as part of VMD Kevin Friedberg
2023-02-01 8:54 ` Mariusz Tkaczyk
2023-02-05 6:23 ` Kevin Friedberg
2023-02-07 8:44 ` Mariusz Tkaczyk [this message]
2023-02-09 0:39 ` Xiao Ni
2023-02-10 10:17 ` Mariusz Tkaczyk
2023-02-13 7:02 ` Kevin Friedberg
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230207094423.00001a30@linux.intel.com \
--to=mariusz.tkaczyk@linux.intel.com \
--cc=kev.friedberg@gmail.com \
--cc=linux-raid@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).