Re: [PATCH v3 04/16] platform/x86/intel/ifs: Remove memory allocation from load path

From: Hans de Goede
Date: Thu Nov 17 2022 - 13:02:15 EST


Hi Jithu,

On 11/17/22 18:29, Jithu Joseph wrote:
> IFS requires tests to be authenticated once for each CPU socket
> on a system.
>
> scan_chunks_sanity_check() was dynamically allocating memory
> to store the state of whether tests have been authenticated on
> each socket for every load operation.
>
> Move the memory allocation to init path and store the pointer
> in ifs_data struct.
>
> Reviewed-by: Tony Luck <tony.luck@xxxxxxxxx>
> Suggested-by: Borislav Petkov <bp@xxxxxxxxx>
> Signed-off-by: Jithu Joseph <jithu.joseph@xxxxxxxxx>
> ---
> - Replaced global pkg_auth pointer to struct ifs_data (Hans)
> - With this change there are conflicts in patches 11 and 12 (I will
> post the updated 11 and 12 if this is satisfactory)
>
> drivers/platform/x86/intel/ifs/ifs.h | 2 ++
> drivers/platform/x86/intel/ifs/core.c | 12 ++++++++++--
> drivers/platform/x86/intel/ifs/load.c | 14 ++++----------
> 3 files changed, 16 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/platform/x86/intel/ifs/ifs.h b/drivers/platform/x86/intel/ifs/ifs.h
> index 3ff1d9aaeaa9..8de1952a1b7b 100644
> --- a/drivers/platform/x86/intel/ifs/ifs.h
> +++ b/drivers/platform/x86/intel/ifs/ifs.h
> @@ -191,6 +191,7 @@ union ifs_status {
> * struct ifs_data - attributes related to intel IFS driver
> * @integrity_cap_bit: MSR_INTEGRITY_CAPS bit enumerating this test
> * @loaded_version: stores the currently loaded ifs image version.
> + * @pkg_auth: array of bool storing per package auth status
> * @loaded: If a valid test binary has been loaded into the memory
> * @loading_error: Error occurred on another CPU while loading image
> * @valid_chunks: number of chunks which could be validated.
> @@ -199,6 +200,7 @@ union ifs_status {
> */
> struct ifs_data {
> int integrity_cap_bit;
> + bool *pkg_auth;
> int loaded_version;
> bool loaded;
> bool loading_error;
> diff --git a/drivers/platform/x86/intel/ifs/core.c b/drivers/platform/x86/intel/ifs/core.c
> index 5fb7f655c291..6980a31e9786 100644
> --- a/drivers/platform/x86/intel/ifs/core.c
> +++ b/drivers/platform/x86/intel/ifs/core.c
> @@ -4,6 +4,7 @@
> #include <linux/module.h>
> #include <linux/kdev_t.h>
> #include <linux/semaphore.h>
> +#include <linux/slab.h>
>
> #include <asm/cpu_device_id.h>
>
> @@ -51,8 +52,14 @@ static int __init ifs_init(void)
> ifs_device.misc.groups = ifs_get_groups();
>
> if ((msrval & BIT(ifs_device.data.integrity_cap_bit)) &&
> - !misc_register(&ifs_device.misc))
> - return 0;
> + !misc_register(&ifs_device.misc)) {
> + ifs_device.data.pkg_auth = kmalloc_array(topology_max_packages(),
> + sizeof(bool), GFP_KERNEL);

Thank you for the new version, but as I mentioned in my review, this kmalloc
must be done *before* the misc_register(&ifs_device.misc), because as soon
as that is done the other code may get triggered creating a race condition.

More in general && the misc_register to gether with the integrity_cap_bit
is not really nice. If someone does not pay close attention they may
mis that the check of the if has the pretty big side-effect of
registering the actual misc device.

Generally speaking test-conditions for if-s should not have side
effects if possible.

> + if (!ifs_device.data.pkg_auth)
> + return -ENOMEM;
> + else
> + return 0;
> + }
>
> return -ENODEV;
> }

This also makes me realize that you have your -ENODEV error exit and
your normal success exit paths switched around from what is normal.

Why not just write the above as (can be done as part of this
patch since you need to touch it all anyways):

if (!(msrval & BIT(ifs_device.data.integrity_cap_bit))
return -ENODEV;

ifs_device.data.pkg_auth = kmalloc_array(topology_max_packages(), sizeof(bool), GFP_KERNEL);
if (!ifs_device.data.pkg_auth)
return -ENOMEM

ret = misc_register(&ifs_device.misc);
if (ret) {
kfree(ifs_device.data.pkg_auth);
return ret;
}

return 0;
}

That makes this all look much more like a normal probe() function
with the success 0 return at the end.

Where as your version has the success 0 return nested 2 levels
deep in the else of a kmalloc() error check...

Regards,

Hans



> @@ -60,6 +67,7 @@ static int __init ifs_init(void)
> static void __exit ifs_exit(void)
> {
> misc_deregister(&ifs_device.misc);
> + kfree(ifs_device.data.pkg_auth);
> }
>
> module_init(ifs_init);
> diff --git a/drivers/platform/x86/intel/ifs/load.c b/drivers/platform/x86/intel/ifs/load.c
> index 89ce265887ea..8423c486d11b 100644
> --- a/drivers/platform/x86/intel/ifs/load.c
> +++ b/drivers/platform/x86/intel/ifs/load.c
> @@ -3,7 +3,6 @@
>
> #include <linux/firmware.h>
> #include <asm/cpu.h>
> -#include <linux/slab.h>
> #include <asm/microcode_intel.h>
>
> #include "ifs.h"
> @@ -118,16 +117,12 @@ static void copy_hashes_authenticate_chunks(struct work_struct *work)
> */
> static int scan_chunks_sanity_check(struct device *dev)
> {
> - int metadata_size, curr_pkg, cpu, ret = -ENOMEM;
> + int metadata_size, curr_pkg, cpu, ret;
> struct ifs_data *ifsd = ifs_get_data(dev);
> - bool *package_authenticated;
> struct ifs_work local_work;
> char *test_ptr;
>
> - package_authenticated = kcalloc(topology_max_packages(), sizeof(bool), GFP_KERNEL);
> - if (!package_authenticated)
> - return ret;
> -
> + memset(ifsd->pkg_auth, 0, (topology_max_packages() * sizeof(bool)));
> metadata_size = ifs_header_ptr->metadata_size;
>
> /* Spec says that if the Meta Data Size = 0 then it should be treated as 2000 */
> @@ -150,7 +145,7 @@ static int scan_chunks_sanity_check(struct device *dev)
> cpus_read_lock();
> for_each_online_cpu(cpu) {
> curr_pkg = topology_physical_package_id(cpu);
> - if (package_authenticated[curr_pkg])
> + if (ifsd->pkg_auth[curr_pkg])
> continue;
> reinit_completion(&ifs_done);
> local_work.dev = dev;
> @@ -161,12 +156,11 @@ static int scan_chunks_sanity_check(struct device *dev)
> ret = -EIO;
> goto out;
> }
> - package_authenticated[curr_pkg] = 1;
> + ifsd->pkg_auth[curr_pkg] = 1;
> }
> ret = 0;
> out:
> cpus_read_unlock();
> - kfree(package_authenticated);
>
> return ret;
> }