Re: [PATCH] treat AHCI controllers under VMD as part of VMD

Hi Mariusz,

Thank you for the feedback.  I'm working up a replacement patch based
on your comments and should have it ready in the next couple days.  It
works in a different way and has a different commit message, so go
ahead and reject this patch if there's a process for that.

On Tue, Feb 7, 2023 at 3:44 AM Mariusz Tkaczyk
[off-list ref] wrote:

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 [off-list ref] wrote:

quoted

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 <redacted>
---
 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.

quoted

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

quoted

+
 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

quoted

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