pthread_create: TLS alignment and cleanup

This patch replaces 2 function calls from musl with
2 custom ones and introduces an additional size
when allocating the map for a thread, `libc.tls_align`.
It is needed because Unikraft demands that the tls be aligned.

Signed-off-by: Dragos Iulian Argint <dragosargint21@gmail.com>
Reviewed-by: Florin Postolache <florin.postolache80@gmail.com>
Reviewed-by: Simon Kuenzer <simon@unikraft.io>
Reviewed-by: Cezar Craciunoiu <cezar.craciunoiu@unikraft.io>
Reviewed-by: Razvan Deaconescu <razvand@unikraft.io>
Reviewed-by: Robert Kuban <robert.kuban@opensynergy.com>
Approved-by: Simon Kuenzer <simon@unikraft.io>
Tested-by: Unikraft CI <monkey@unikraft.io>
GitHub-Closes: #9

diff --git a/patches/0014-Change-some-things-in-pthread_create.c.patch b/patches/0014-Change-some-things-in-pthread_create.c.patch
new file mode 100644 (file)
index 0000000..fe3318d
--- /dev/null
+++ b/patches/0014-Change-some-things-in-pthread_create.c.patch
@@ -0,0 +1,89 @@
+From cab8414305bdbb07a9adfa72682d110fd7365da9 Mon Sep 17 00:00:00 2001
+From: Dragos Iulian Argint <dragosargint21@gmail.com>
+Date: Sun, 14 Aug 2022 20:14:33 +0300
+Subject: [PATCH] Change some things in pthread_create.c
+
+This patch replaces 2 function calls from musl with
+2 custom ones and introduces an additional size
+when allocating the map for a thread, `libc.tls_align`.
+It is needed because Unikraft demands that the tls be aligned.
+
+Signed-off-by: Dragos Iulian Argint <dragosargint21@gmail.com>
+---
+ src/thread/pthread_create.c | 28 +++++++++++++++++++++-------
+ 1 file changed, 21 insertions(+), 7 deletions(-)
+
+diff --git a/src/thread/pthread_create.c b/src/thread/pthread_create.c
+index 439ee36..d7d5da9 100644
+--- a/src/thread/pthread_create.c
++++ b/src/thread/pthread_create.c
+@@ -10,6 +10,11 @@ void *__mmap(void *, size_t, int, int, int, off_t);
+ int __munmap(void *, size_t);
+ int __mprotect(void *, size_t, int);
+ 
++/*
++ * Unikraft version of `__unmapself()`. See `__uk_unmapself.c`
++ */
++void __uk_unmapself(void *base, size_t size);
++
+ static void dummy_0()
+ {
+ }
+@@ -108,9 +113,15 @@ _Noreturn void __pthread_exit(void *result)
+                * explicitly wait for vmlock holders first. */
+               __vm_wait();
+ 
+-              /* The following call unmaps the thread's stack mapping
+-               * and then exits without touching the stack. */
+-              __unmapself(self->map_base, self->map_size);
++              /*
++               * In the original code `__unampself()` was called here.
++               * That call unmaps the thread's stack mapping and then
++               * exits without touching the stack. However, we cannot
++               * do this since we don't have a kernel stack. We use
++               * a different approach and let the `idle` thread do the
++               * cleaning. Please refer to the `__uk_unampself.c` file.
++               */
++              __uk_unmapself(self->map_base, self->map_size);
+       }
+ 
+       for (;;) __syscall(SYS_exit, 0);
+@@ -181,7 +192,10 @@ static void init_file_lock(FILE *f)
+       if (f && f->lock<0) f->lock = 0;
+ }
+ 
+-void *__copy_tls(unsigned char *);
++/*
++ * Unikraft version of `__copy_tls()`
++ */
++void *__uk_copy_tls(unsigned char *);
+ 
+ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict attrp, void *(*entry)(void *), void *restrict arg)
+ {
+@@ -237,7 +251,7 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
+       } else {
+               guard = ROUND(attr._a_guardsize);
+               size = guard + ROUND(attr._a_stacksize
+-                      + libc.tls_size +  __pthread_tsd_size);
++                      + libc.tls_size  + libc.tls_align +  __pthread_tsd_size);
+       }
+ 
+       if (!tsd) {
+@@ -255,12 +269,12 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
+               }
+               tsd = map + size - __pthread_tsd_size;
+               if (!stack) {
+-                      stack = tsd - libc.tls_size;
++                      stack = tsd - libc.tls_size - libc.tls_align;
+                       stack_limit = map + guard;
+               }
+       }
+ 
+-      new = __copy_tls(tsd - libc.tls_size);
++      new = __uk_copy_tls(tsd - libc.tls_size);
+       new->map_base = map;
+       new->map_size = size;
+       new->stack = stack;
+-- 
+2.17.1
+