mirror of
https://github.com/termux/termux-packages.git
synced 2025-05-12 11:23:30 +00:00
e21578f662
Patched `rts` heap reservation logic on Android:
* Need for patch
- There seems to be a bug on Android. Sometimes simple commands like `ghc --help` usage 90% - 100% cpu and hangs.
- It doesn't happen every time, but ~25% of the times (at least in my
testing).
* Cause of the bug
- The function `osReserveHeapMemory` tries to allocate virtual space starting
from `0x4200000000`. The allocated space has to start at an address >= this address.
- If the kernel doesn't allocate inside this space, it keeps on repeating the
`mmap` call with increasing starting point.
- Now, on Android the kernel sometimes return an address above (as in counting)
this `hint` address. It repeatedly returns the same address for subsequent calls.
- Thus, an infinite loop occurs.
References:
- 383be28ffd/rts/posix/OSMem.c (L461)
- https://github.com/termux/termux-packages/pull/22991#issuecomment-2759137291
* Solution (proposed by Robert Kirkman):
- It introduces a new helper function `osTryReserveHeapMemoryRecursive`.
This transforms the heap reservation logic into a recursive one.
- `osTryReserveHeapMemory()` is run multiple times without unmapping the
undesired addresses. Thus, forcing the kernel to map subsequent calls of
`mmap` to a new, unique address until an address above the `0x4200000000`
mark is obtained.
- After which each recursive call unmaps its undesired address before returning
the desired address (in order from last mapped to first mapped).
References:
- https://gitlab.haskell.org/ghc/ghc/-/merge_requests/14164
- https://github.com/termux/termux-packages/pull/22991#issuecomment-2761325484
Co-authored-by: Robert Kirkman <rkirkman@termux.dev>
Signed-off-by: Aditya Alok <alok@termux.dev>
47 lines
2.4 KiB
Diff
47 lines
2.4 KiB
Diff
--- ghc-9.12.2/hadrian/src/Settings/Packages.hs 2025-03-29 15:32:47.000000000 +0530
|
|
+++ ghc-9.12.2.mod/hadrian/src/Settings/Packages.hs 2025-04-07 15:06:41.398290889 +0530
|
|
@@ -85,8 +85,8 @@
|
|
-- (#14335) and completely untested in CI for cross
|
|
-- backends at the moment, so we might as well disable it
|
|
-- for cross GHC.
|
|
- [ andM [expr ghcWithInterpreter, notStage0, notCross] `cabalFlag` "internal-interpreter"
|
|
- , notM cross `cabalFlag` "terminfo"
|
|
+ [ andM [expr ghcWithInterpreter, notStage0] `cabalFlag` "internal-interpreter"
|
|
+ , notStage0 `cabalFlag` "terminfo"
|
|
, arg "-build-tool-depends"
|
|
, flag UseLibzstd `cabalFlag` "with-libzstd"
|
|
-- ROMES: While the boot compiler is not updated wrt -this-unit-id
|
|
@@ -121,7 +121,7 @@
|
|
|
|
-------------------------------- ghcPkg --------------------------------
|
|
, package ghcPkg ?
|
|
- builder (Cabal Flags) ? notM cross `cabalFlag` "terminfo"
|
|
+ builder (Cabal Flags) ? notStage0 `cabalFlag` "terminfo"
|
|
|
|
-------------------------------- ghcPrim -------------------------------
|
|
, package ghcPrim ? mconcat
|
|
@@ -213,7 +213,7 @@
|
|
-- dependencies.
|
|
-- TODO: Perhaps the user should rather be responsible for this?
|
|
, package haskeline ?
|
|
- builder (Cabal Flags) ? notM cross `cabalFlag` "terminfo"
|
|
+ builder (Cabal Flags) ? notStage0 `cabalFlag` "terminfo"
|
|
|
|
-------------------------------- terminfo ------------------------------
|
|
, package terminfo ?
|
|
@@ -374,10 +374,10 @@
|
|
|
|
, input "**/RtsUtils.c" ? pure
|
|
[ "-DProjectVersion=" ++ show projectVersion
|
|
- , "-DHostPlatform=" ++ show hostPlatform
|
|
- , "-DHostArch=" ++ show hostArch
|
|
- , "-DHostOS=" ++ show hostOs
|
|
- , "-DHostVendor=" ++ show hostVendor
|
|
+ , "-DHostPlatform=" ++ show targetPlatform
|
|
+ , "-DHostArch=" ++ show targetArch
|
|
+ , "-DHostOS=" ++ show targetOs
|
|
+ , "-DHostVendor=" ++ show targetVendor
|
|
, "-DBuildPlatform=" ++ show buildPlatform
|
|
, "-DBuildArch=" ++ show buildArch
|
|
, "-DBuildOS=" ++ show buildOs
|