badmem static allocation of memory hole

[dimko]

Hello Gentoids,

For years I was struggling with bad ram.(usually compilation issues, and usually with only big apps like chromium, so I reduced amount of threads from 32 to 6 at compile time)
There is one specific area that was causing me trouble which I did not understand:
[ 20.573439] 5555555555555555 bad mem addr 0x0000000f6cf22598 - 0x0000000f6cf225a0 reserved

One of reasons for whatever reasons i had problems with installed version of memtest86+ for whatever reason.
Yesterday I noticed that it worked, and found a faulty area in RAM. Only one. Which makes me believe this is not gonna get worse.(used that ram for many years now)
First I tried disabling using address and mask in grub, whichI immedeatelly failed and made my grub unbootable.(changes reverted, thanks god i had some ubuntu usb flash somewhere)
I used "GRUB_BADRAM=' 0x0000000f6cf22598,0xffffffffffffffff'" in default grub config file. Grub failed to start, i just see "welcome to grub" and nothing else, no kernel choices i have had before.
So I enabled memory testing in kernel for now. memtest=4

Now the question, how can I permanently add ram area it to my current ONLY last kernel in grub?(before I add it somehow to /etc/default/grub for all kernel)?

with 64gig of ram boot times increased by quite a lot.

bonus question: how do I add this ram mask into /etc/default/grub so all kernels I compile in the future can benefit from it?

[NeddySeagoon]

dimko,

Code: Select all

GRUB_BADRAM=' 0x0000000f6cf22598,0xffffffffffffffff'

That fails as its not all RAM.
It covers the firmware and PCI address space too.

Save your good grub.cfg so you can get it back.

Put the GRUB_BADRAM= back and generate a new (broken) grub.cfg.
Look on the kernel command line, in the new (broken) grub.cfg and see what mas added.
Make a note of it.
Restore your good working grub.cfg and undo the damage to the default grub config file.

Now you as back as you were but with you know what command to add to the kernel command line.
Use grubs boot time editor to practice, using the notes from what grub did.

Choose a kernel to boot but press 'e' for edit. Not enter.
Find the linux line and append your new kernel parameter.
Now boot.
This edits the in RAM copy of grub.cnf only, so if horrible things happen, don't do it again :)
A reboot will drop out the in RAM edit.

This lets you test to see if the proposed change has the desired effect.

[dimko]

Choose a kernel to boot but press 'e' for edit. Not enter.
Find the linux line and append your new kernel parameter.
Now boot.
This edits the in RAM copy of grub.cnf only, so if horrible things happen, don't do it again
A reboot will drop out the in RAM edit.

This lets you test to see if the proposed change has the desired effect.

I think i get the gist of what you are saying.
Any idea how I can build a mask?(based on memtest=4 range? ill include it in a sec)

[dimko]

OK, my initial aim was to:
tell grub to tell kernel to block area of memory that memtest86+ has found to be corrupt.

perhaps GRUB_BADRAM ONLY affects grub and not actual kernel?
Grub was never giving me issues anyway, so ram problem for GRUB is not an issue.

I just want to avoid using memtest=4 in kernel settings.(it increases kernel load times)

[dimko]

dmesg |grep -i bad
[ 0.000000] Command line: BOOT_IMAGE=/vmlinuz-6.5.3 root=/dev/nvme0n1p2 ro amd_pstate.shared_mem=1 amd_pstate=active iommu=1 amd_iommu=on iommu=calgary splash GRUB_BADRAM=0x0000000f6cf22598,0xfffffffffffffff8
[ 0.084473] Kernel command line: BOOT_IMAGE=/vmlinuz-6.5.3 root=/dev/nvme0n1p2 ro amd_pstate.shared_mem=1 amd_pstate=active iommu=1 amd_iommu=on iommu=calgary splash GRUB_BADRAM=0x0000000f6cf22598,0xfffffffffffffff8
[ 0.084520] Unknown kernel command line parameters "splash BOOT_IMAGE=/vmlinuz-6.5.3 GRUB_BADRAM=0x0000000f6cf22598,0xfffffffffffffff8", will be passed to user space.
[ 1.490516] GRUB_BADRAM=0x0000000f6cf22598,0xfffffffffffffff8

[NeddySeagoon]

dimko,

Code: Select all

16.3.6 badram

Command: badram addr,mask[,addr,mask...]

    Filter out bad RAM.

    This command notifies the memory manager that specified regions of RAM ought to be filtered out (usually, because they’re damaged). This remains in effect after a payload kernel has been loaded by GRUB, as long as the loaded kernel obtains its memory map from GRUB. Kernels that support this include Linux, GNU Mach, the kernel of FreeBSD and Multiboot kernels in general.

    Syntax is the same as provided by the Memtest86+ utility: a list of address/mask pairs. Given a page-aligned address and a base address / mask pair, if all the bits of the page-aligned address that are enabled by the mask match with the base address, it means this page is to be filtered. This syntax makes it easy to represent patterns that are often result of memory damage, due to physical distribution of memory cells.

    The command is similar to cutmem command.

    Note: The command is not allowed when lockdown is enforced (see Lockdown). This prevents removing EFI memory regions to potentially subvert the security mechanisms provided by the UEFI secure boot. 

I'm no a grub user.

The "as long as the loaded kernel obtains its memory map from GRUB." matters a great deal.

This Ubuntu page is worth reading too as it details how to calculate the mask.

[dimko]

Overall amount of time that i will spend solving this issue is probably more than amount of time i will need to reboot infrequently enough(i "sleep" my PC), until i get new ram, plus memtest=4 does check other ram, so I will be safer from new surprises.

If someone gets similar issue and solves it - please let us know how.

What particularly confuses me, my range of bad ram:
[ 30.885799] ffffffffffffffff bad mem addr 0x0000000f6cf22598 - 0x0000000f6cf225a0 reserved

Now i have 64bit machine with 64gig of ram, but all other documentation I have seen so far has much smaller address and mask numbers.

[dimko]

thanks for the effort @NeddySeagoon

[NeddySeagoon]

dimko,

It looks like you need

Code: Select all

         memmap=nn[KMG]!ss[KMG]
                        [KNL,X86] Mark specific memory as protected.
                        Region of memory to be used, from ss to ss+nn.
                        The memory region may be marked as e820 type 12 (0xc)
                        and is NVDIMM or ADR memory.

Your region

Code: Select all

 bad mem addr 0x0000000f6cf22598 - 0x0000000f6cf225a0

is only 8 bytes but you probably need to protect the whole 4k page.
That will be the 4k starting at 0x0000000f6cf22000, as the kernel works with 4k pages of RAM.