IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Pull RAS updates from Borislav Petkov:
- Add initial support for RAS hardware found on AMD server GPUs (MI200).
Those GPUs and CPUs are connected together through the coherent
fabric and the GPU memory controllers report errors through x86's MCA
so EDAC needs to support them. The amd64_edac driver supports now HBM
(High Bandwidth Memory) and thus such heterogeneous memory controller
systems
- Other small cleanups and improvements
* tag 'ras_core_for_v6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
EDAC/amd64: Cache and use GPU node map
EDAC/amd64: Add support for AMD heterogeneous Family 19h Model 30h-3Fh
EDAC/amd64: Document heterogeneous system enumeration
x86/MCE/AMD, EDAC/mce_amd: Decode UMC_V2 ECC errors
x86/amd_nb: Re-sort and re-indent PCI defines
x86/amd_nb: Add MI200 PCI IDs
ras/debugfs: Fix error checking for debugfs_create_dir()
x86/MCE: Check a hw error's address to determine proper recovery action
AMD systems have historically provided an "AMD Node ID" that is a unique
identifier for each die in a multi-die package. This was associated with
a unique instance of the AMD Northbridge on a legacy system. And now it
is associated with a unique instance of the AMD Data Fabric on modern
systems. Each instance is referred to as a "Node"; this is an
AMD-specific term not to be confused with NUMA nodes.
The data fabric provides a number of interfaces accessible through a set
of functions in a single PCI device. There is one PCI device per Data
Fabric (AMD Node), and multi-die systems will see multiple such PCI
devices. The AMD Node ID matches a Node's position in the PCI hierarchy.
For example, the Node 0 is accessed using the first PCI device, Node 1
is accessed using the second, and so on. A logical CPU can find its AMD
Node ID using CPUID. Furthermore, the AMD Node ID is used within the
hardware fabric, so it is not purely a logical value.
Heterogeneous AMD systems, with a CPU Data Fabric connected to GPU data
fabrics, follow a similar convention. Each CPU and GPU die has a unique
AMD Node ID value, and each Node ID corresponds to PCI devices in
sequential order.
However, there are two caveats:
1) GPUs are not x86, and they don't have CPUID to read their AMD Node ID
like on CPUs. This means the value is more implicit and based on PCI
enumeration and hardware-specifics.
2) There is a gap in the hardware values for AMD Node IDs. Values 0-7
are for CPUs and values 8-15 are for GPUs.
For example, a system with one CPU die and two GPUs dies will have the
following values:
CPU0 -> AMD Node 0
GPU0 -> AMD Node 8
GPU1 -> AMD Node 9
EDAC is the only subsystem where this has a practical effect. Memory
errors on AMD systems are commonly reported through MCA to a CPU on the
local AMD Node. The error information is passed along to EDAC where the
AMD EDAC modules use the AMD Node ID of reporting logical CPU to access
AMD Node information.
However, memory errors from a GPU die will be reported to the CPU die.
Therefore, the logical CPU's AMD Node ID can't be used since it won't
match the AMD Node ID of the GPU die. The AMD Node ID of the GPU die is
provided as part of the MCA information, and the value will match the
hardware enumeration (e.g. 8-15).
Handle this situation by discovering GPU dies the same way as CPU dies
in the AMD NB code. But do a "node id" fixup in AMD64 EDAC where it's
needed.
The GPU data fabrics provide a register with the base AMD Node ID for
their local "type", i.e. GPU data fabric. This value is the same for all
fabrics of the same type in a system.
Read and cache the base AMD Node ID from one of the GPU devices during
module initialization. Use this to fixup the "node id" when reporting
memory errors at runtime.
[ bp: Squash a fix making gpu_node_map static as reported by
Tom Rix <trix@redhat.com>.
Link: https://lore.kernel.org/r/20230610210930.174074-1-trix@redhat.com ]
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Co-developed-by: Muralidhara M K <muralidhara.mk@amd.com>
Signed-off-by: Muralidhara M K <muralidhara.mk@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230515113537.1052146-6-muralimk@amd.com
The AMD64 EDAC module version information is not exposed through ABI
like MODULE_VERSION(). Instead it is printed during module init.
Version numbers can be confusing in cases where module updates are
partly backported resulting in a difference between upstream and
backported module versions.
Remove the AMD64 EDAC module version information to avoid user
confusion.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230410190959.3367528-1-yazen.ghannam@amd.com
GPU Nodes will use a different method to determine the chip select
and channel of an error. A function pointer should be used rather than
introduce another branching condition.
Prepare for this by adding get_err_info() to pvt->ops. This function is
only called from the modern code path, so a legacy function is not
defined.
Make sure to call this after MCA_STATUS[SyndV] is checked, since the
csrow value is found in MCA_SYND.
[ Yazen: rebased/reworked patch and reworded commit message. ]
Signed-off-by: Muralidhara M K <muralidhara.mk@amd.com>
Co-developed-by: Naveen Krishna Chatradhi <naveenkrishna.chatradhi@amd.com>
Signed-off-by: Naveen Krishna Chatradhi <naveenkrishna.chatradhi@amd.com>
Co-developed-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230127170419.1824692-23-yazen.ghannam@amd.com
The bulk of system-specific information is gathered at init time with
hw_info_get(). This function calls a number of helper functions, and
many of these helper functions are split between a modern UMC/DF path
and a legacy DCT path.
Split hw_info_get() into legacy and modern versions. This creates two
separate code paths early on, and legacy and modern helper functions can
be called directly in the appropriate code path.
Also, simplify hw_info_put() and share it between legacy and modern
systems. NULL pointer checks are done in pci_dev_put() and kfree(), so
they can be called unconditionally.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230127170419.1824692-12-yazen.ghannam@amd.com
Future AMD systems will support heterogeneous "AMD Node" types, e.g.
CPU and GPU types. Therefore, a global family type shared across all
AMD nodes is no longer appropriate.
Move struct low_ops routines and members of struct amd64_family_type
to struct amd64_pvt.
Currently, there are many code branches that split between "modern" and
"legacy" systems. Another code branch will be needed in order to cover
GPU cases. However, rather than introduce another branching case in
multiple functions, the current branching code should be switched to a
set of function pointers. This change makes the code more readable and
simplifies adding support for new families/models.
In order to reuse code, define two sets of function pointers. Use one
for modern systems (Family 17h and later). This will not change between
current CPU families. Use another set of function pointers for legacy
systems (before Family 17h). Use the Family 16h versions as default
for the legacy ops since these are the latest, and adjust the function
pointers as needed for older families.
[ Yazen: rebased/reworked patch and reworded commit message. ]
[ bp: Fix rev8 or later check. ]
Signed-off-by: Muralidhara M K <muralidhara.mk@amd.com>
Co-developed-by: Naveen Krishna Chatradhi <naveenkrishna.chatradhi@amd.com>
Signed-off-by: Naveen Krishna Chatradhi <naveenkrishna.chatradhi@amd.com>
Co-developed-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230127170419.1824692-11-yazen.ghannam@amd.com
The early_channel_count() function seems to have been useful in the past
for knowing how many EDAC mci structures to populate. However, this is no
longer needed as the maximum channel count for a system is used instead.
Remove the early_channel_count() helper functions and related code. Use the
size of the channel layer when iterating over channel structures.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230127170419.1824692-6-yazen.ghannam@amd.com
PCI Function 0 is used on Family 17h and later only to read the "dhar"
value. This value is printed and provided through a module-specific
debug sysfs file. The value is not used for any Family 17h and later
code, and it does not have any apparent debug value on these systems.
Remove "dhar", Function 0 PCI IDs, and all related code.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230127170419.1824692-5-yazen.ghannam@amd.com
The scrub registers on AMD Family 17h and later may be inaccessible to
the OS. Furthermore, hardware designers recommend that the scrubbing
feature is managed by the firmware.
Remove support for the sdram_scrub_rate interface for AMD Family 17h
systems and later by not setting the scrub function pointers. The EDAC MC
core will then not expose the scrub files in sysfs.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20230127170419.1824692-3-yazen.ghannam@amd.com
Introduce a "family flags" bitmask that can be used to indicate any
special behavior needed on a per-family basis.
Add a flag to indicate a system uses the new register offsets introduced
with Family 19h Model 10h.
Use this flag to account for register offset changes, a new bitfield
indicating DDR5 use on a memory controller, and to set the proper number
of chip select masks.
Rework f17_addr_mask_to_cs_size() to properly handle the change in chip
select masks. And update code comments to reflect the updated Chip
Select, DIMM, and Mask relationships.
[uninitialized variable warning]
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: William Roche <william.roche@oracle.com>
Link: https://lore.kernel.org/r/20220202144307.2678405-3-yazen.ghannam@amd.com
Current AMD systems allow mixing of DIMM types within a system. However,
DIMMs within a channel, i.e. managed by a single Unified Memory
Controller (UMC), must be of the same type.
Handle this possible configuration by checking and setting the memory
type for each individual "UMC" structure.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: William Roche <william.roche@oracle.com>
Link: https://lore.kernel.org/r/20220202144307.2678405-2-yazen.ghannam@amd.com
There's no need for them to be in a separate file so merge them into the
main driver compilation unit like the other EDAC drivers do.
Drop now-unneeded function export, make the function static and shorten
static function names.
No functional changes.
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Yazen Ghannam <yazen.ghannam@amd.com>
Link: https://lkml.kernel.org/r/20201215110517.5215-1-bp@alien8.de
Future AMD systems will support asymmetric dual-rank DIMMs. These are
DIMMs where the ranks are of different sizes.
The even rank will use the Primary Even Chip Select registers and the
odd rank will use the Secondary Odd Chip Select registers.
Recognize if a Secondary Odd Chip Select is being used. Use the
Secondary Odd Address Mask when calculating the chip select size.
[ bp: move csrow_sec_enabled() to the header, fix CS_ODD define and
tone-down the capitalized words spelling. ]
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: "linux-edac@vger.kernel.org" <linux-edac@vger.kernel.org>
Cc: James Morse <james.morse@arm.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Link: https://lkml.kernel.org/r/20190821235938.118710-8-Yazen.Ghannam@amd.com
The struct chip_select array that's used for saving chip select bases
and masks is fixed at length of two. There should be one struct
chip_select for each controller, so this array should be increased to
support systems that may have more than two controllers.
Increase the size of the struct chip_select array to eight, which is the
largest number of controllers per die currently supported on AMD
systems.
Fix number of DIMMs and Chip Select bases/masks on Family17h, because
AMD Family 17h systems support 2 DIMMs, 4 CS bases, and 2 CS masks per
channel.
Also, carve out the Family 17h+ reading of the bases/masks into a
separate function. This effectively reverts the original bases/masks
reading code to before Family 17h support was added.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: "linux-edac@vger.kernel.org" <linux-edac@vger.kernel.org>
Cc: James Morse <james.morse@arm.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Link: https://lkml.kernel.org/r/20190821235938.118710-2-Yazen.Ghannam@amd.com
The struct chip_select array that's used for saving chip select bases
and masks is fixed at length of two. There should be one struct
chip_select for each controller, so this array should be increased to
support systems that may have more than two controllers.
Increase the size of the struct chip_select array to eight, which is the
largest number of controllers per die currently supported on AMD
systems.
Also, carve out the Family 17h+ reading of the bases/masks into a
separate function. This effectively reverts the original bases/masks
reading code to before Family 17h support was added.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Kim Phillips <kim.phillips@amd.com>
Cc: James Morse <james.morse@arm.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: https://lkml.kernel.org/r/20190228153558.127292-5-Yazen.Ghannam@amd.com
Future AMD systems may support x16 symbol sizes.
Recognize if a system is using x16 symbol size. Also, simplify the print
statement.
Note that a x16 syndrome vector table is not necessary like with x4 or
x8 syndromes. This is because systems that support x16 symbol sizes are
SMCA systems and in that case, the syndrome can be directly extracted
from the MCA_SYND[Syndrome] field.
[ bp: massage. ]
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Kim Phillips <kim.phillips@amd.com>
Cc: James Morse <james.morse@arm.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: https://lkml.kernel.org/r/20190228153558.127292-4-Yazen.Ghannam@amd.com
The first few models of Family 17h all had 2 Unified Memory Controllers
per Die, so this was treated as a fixed value. However, future systems
may have more Unified Memory Controllers per Die.
Related to this, the channel number and base address of a Unified Memory
Controller were found by matching on fixed, known values. However,
current and future systems follow this pattern for the channel number
and base address of a Unified Memory Controller: 0xYXXXXX, where Y is
the channel number. So matching on hardcoded values is not necessary.
Set the number of Unified Memory Controllers at driver init time based
on the family/model. Also, update the functions that find the channel
number and base address of a Unified Memory Controller to support more
than two.
[ bp: Move num_umcs into the .c file and simplify comment. ]
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Kim Phillips <kim.phillips@amd.com>
Cc: James Morse <james.morse@arm.com>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: https://lkml.kernel.org/r/20190228153558.127292-3-Yazen.Ghannam@amd.com
Last time we did that was when we enabled Bulldozer. Now, we enabled Zen
so it is only natural ... :-)
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Yazen Ghannam <Yazen.Ghannam@amd.com>
Prefix the warn and error macros with the respective string so that
callers don't have to say "Error" or "Warning". We save us string length
this way in the actual calls.
While at it, shorten the calls in reserve_mc_sibling_devs().
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Yazen Ghannam <Yazen.Ghannam@amd.com>
- remove homegrown instances counting.
- take F3 PCI device from amd_nb caching instead of F2 which was used with the
PCI core.
With those changes, the driver doesn't need to register a PCI driver and
relies on the northbridges caching which we do anyway on AMD.
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Yazen Ghannam <yazen.ghannam@amd.com>
The scrub rate control register has moved to function 2 in PCI config
space and is at a different offset on family 0x15, models 0x60 and
later. The minimum recommended scrub rate has also changed. (Refer to
D18F2x1c9_dct[1:0][DramScrub] in Fam15hM60h BKDG).
Adjust set_scrub_rate() and get_scrub_rate() functions to accommodate
this.
Tested on F15hM60h, Fam15h, models 00h-0fh and Fam10h systems.
Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@amd.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: http://lkml.kernel.org/r/1443440593-2316-2-git-send-email-Aravind.Gopalakrishnan@amd.com
[ Cleanup conditionals. ]
Signed-off-by: Borislav Petkov <bp@suse.de>
Instead of calling device_create_file() and device_remove_file()
manually, pass the static attribute groups with the new
edac_mc_add_mc_with_groups(). The conditional creation of inject sysfs
files is done by a proper is_visible callback.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Link: http://lkml.kernel.org/r/1423046938-18111-4-git-send-email-tiwai@suse.de
Signed-off-by: Borislav Petkov <bp@suse.de>
This patch adds support for ECC error decoding for F15h M60h processor.
Aside from the usual changes, the patch adds support for some new features
in the processor:
- DDR4(unbuffered, registered); LRDIMM DDR3 support
- relevant debug messages have been modified/added to report these
memory types
- new dbam_to_cs mappers
- if (F15h M60h && LRDIMM); we need a 'multiplier' value to find
cs_size. This multiplier value is obtained from the per-dimm
DCSM register. So, change the interface to accept a 'cs_mask_nr'
value to facilitate this calculation
- switch-casing determine_memory_type()
- done to cleanse the function of too many if-else statements
and improve readability
- This is now called early in read_mc_regs() to cache dram_type
Misc cleanup:
- amd64_pci_table[] is condensed by using PCI_VDEVICE macro.
Testing details:
Tested the patch by injecting 'ECC' type errors using mce_amd_inj
and error decoding works fine.
Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@amd.com>
Link: http://lkml.kernel.org/r/1414617483-4941-1-git-send-email-Aravind.Gopalakrishnan@amd.com
[ Boris: determine_memory_type() cleanups ]
Signed-off-by: Borislav Petkov <bp@suse.de>
Rationale behind this change:
- F2x1xx addresses were stopped from being mapped explicitly to DCT1
from F15h (OR) onwards. They use _dct[0:1] mechanism to access the
registers. So we should move away from using address ranges to select
DCT for these families.
- On newer processors, the address ranges used to indicate DCT1 (0x140,
0x1a0) have different meanings than what is assumed currently.
Changes introduced:
- amd64_read_dct_pci_cfg() now takes in dct value and uses it for
'selecting the dct'
- Update usage of the function. Keep in mind that different families
have specific handling requirements
- Remove [k8|f10]_read_dct_pci_cfg() as they don't do much different
from amd64_read_pci_cfg()
- Move the k8 specific check to amd64_read_pci_cfg
- Remove f15_read_dct_pci_cfg() and move logic to amd64_read_dct_pci_cfg()
- Remove now needless .read_dct_pci_cfg
Testing:
- Tested on Fam 10h; Fam15h Models: 00h, 30h; Fam16h using 'EDAC_DEBUG'
and mce_amd_inj
- driver obtains info from F2x registers and caches it in pvt
structures correctly
- ECC decoding works fine
Signed-off-by: Aravind Gopalakrishnan <aravind.gopalakrishnan@amd.com>
Link: http://lkml.kernel.org/r/1410799058-3149-1-git-send-email-aravind.gopalakrishnan@amd.com
Signed-off-by: Borislav Petkov <bp@suse.de>
