Debian Patches

Status for gnumach/2:1.8+git20260224-8

Patch Description Author Forwarded Bugs Origin Last update
12_version_suffix.patch =================================================================== no
20_FP_NO.patch diff --git a/i386/include/mach/i386/fp_reg.h b/i386/include/mach/i386/fp_reg.h
index 5673055..f490623 100644
no
50_initrd.patch Jérémie Koenig <jk@jk.fr.eu.org>
Add ramdisk support for d-i.

how to destroy once unused) and rather store data in a task and use the
task libstore class. Also see discussions on the wiki.
no
git-grab_page_nonpriv commit 67c31476ccaf040fbe990bef02dd544296d43706

Page allocation from a CPU pool does not consider vm_page_alloc_paused.

A non-empty CPU page pool always allocates pages even when
vm_page_alloc_paused is set. An unprivileged thread can therefore
acquire a signicant number of pages until the pool is empty. The cycle
can continue for a longer period if the pool is replenished by a VM
privileged thread before the unprivileged thread requests more pages
and in extreme cases can result in complete system memory exhaustion.
Message-ID: <20260228205443.67047-2-mike@weatherwax.co.uk>

diff --git a/vm/vm_page.c b/vm/vm_page.c
index 47eeb042..5cdf0c7b 100644
Mike Kelly <mike@weatherwax.co.uk> no 2026-02-28
git-rebalance commit 74615f69cf6bf95c73443331733677c61ee387b5

Rebalance segments for privileged page allocation

It's possible to allocate all pages in a segment (and higher priority
ones) to privileged threads. This can result in a panic even though
memory is available in lower priority segments which haven't yet been
rebalanced by the pageout daemon. Rebalancing is now done actively
during privileged page allocation.
Message-ID: <20260310212847.153812-2-mike@weatherwax.co.uk>

diff --git a/vm/vm_page.c b/vm/vm_page.c
index 5cdf0c7b..a656aa01 100644
Mike Kelly <mike@weatherwax.co.uk> no 2026-03-10
git-evict-inactive commit 17df0c1968e3cbf69567d8e0ea97da14cd797074

pageout: evict inactive pages from various segments before active ones

Otherwise we can be keeping evicting active pages from the highmem
segment, while we may have a lot of inactive pages in other segments.

diff --git a/vm/vm_page.c b/vm/vm_page.c
index a656aa01..0becb262 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-25
git-i386-8G commit b2f1eea97d36aa0021b109a0c954a5b56df1e468

i386: Increase kernel virtual memory to handle 8GB memory

diff --git a/x86_64/x86_64/vm_param.h b/x86_64/x86_64/vm_param.h
index 056aa52e..38caaeec 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-29
git-x86_64-30G commit a378d126bb545b08f2331b91adc2de9034e05000

x86_64: Increase kernel virtual memory to handle 30GB memory

diff --git a/x86_64/x86_64/vm_param.h b/x86_64/x86_64/vm_param.h
index 38caaeec..63eba559 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-29
git-kmsg-16K commit 38af8366552a95b5825529b3ed2fb0d2829e355e

kmsg: Increase KMSGBUFSIZE to 16K

Boot has become more verbose

diff --git a/device/kmsg.c b/device/kmsg.c
index bb72930d..be22a833 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-29
git-memlimit commit 6ea74eec0beb6e840774ad230f3ce38002410ec7

biosmem: Add MAX_PHYS_END to limit physical memory use

to what was actually tested as working considering the limited amount of
available virtual memory.
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-29
git-vmtophys commit 2077c6523de1005eef73010436d6ff488aa6b25c

linux: Fix vmtophys return type

On PAE we need to be able to return 64b

diff --git a/linux/dev/glue/kmem.c b/linux/dev/glue/kmem.c
index 509229d1..23ef03e8 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-30
git-ahci-pae commit 214078f4903e009291addeef893ecf706fbc9dbb

ahci: Check that the command buffer is within DMA32

This may not be the case with PAE.

diff --git a/linux/dev/drivers/block/ahci.c b/linux/dev/drivers/block/ahci.c
index 751c7ca2..addcc3e5 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-30
git-vmalloc-pae commit 56aa79949c0896d1b1aa3c87d1d8f359254b8d62

linux: Make vmalloc allocate from DMA32

Otherwise ahci cannot use it with PAE kernels.

diff --git a/linux/dev/glue/kmem.c b/linux/dev/glue/kmem.c
index 23ef03e8..afcbfba5 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-30
git-buffer_alloc-pae commit f974c68a797c8caa21db481bd52737800ff372db

linux: Fix buffer allocation for PAE

Asking for DMA32 is not enough, we also need directmap, notably for
read_bsd_label.

diff --git a/linux/dev/glue/block.c b/linux/dev/glue/block.c
index 6db77070..59462a86 100644
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-03-30
git-pageout commit cdba82c27f712624b267eb5dc174294c0b6f1864

vm_page: Add global LRU list

This unfortunately takes room in the vm_page structure, thus reducing
manageable memory.

commit e3690e5874d6cfb13285511a20d925b0c6612d2c

vm_page: Make page eviction use the global LRU list

Otherwise, as long as the highest segment has evictable pages, we will be
focusing on it, and not benefitting from other segments.

This is notably very problematic when the highest segment is small, smaller
than the application working set, but big enough for finding evictable
pages.
Samuel Thibault <samuel.thibault@ens-lyon.org> no 2026-04-12

All known versions for source package 'gnumach'

Links