Skip to contentSkip to conversation

direct_io: diotest4: fix mips MAP_SHARED/MAP_FIXED mmap

PATCH New Jan Stancek 2 participants 5 messages Jul 4, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 direct_io: diotest4: fix mips MAP_SHARED/MAP_FIXED mmap

Hongzhi reports that this test is failing on mips64 with 5.1+:
diotest4 10 TBROK : diotest4.c:368: can't mmap file: Invalid argument diotest4 11 TBROK : diotest4.c:368: Remaining cases broken
I could reproduce it on emulated 5kc-malta, running 5.2.0-rc7.

Test is trying to map in area immediately following heap as MAP_SHARED, but it's using wrong alignment, because MMAP_GRANULARITY is always defined as single page - lapi/mmap.h is not including config.h.

Usage of MAP_FIXED in test seems unnecessary, so drop that too and let the kernel pick an address.

---
 include/lapi/mmap.h                      | 2 ++
 testcases/kernel/io/direct_io/diotest4.c | 8 ++------
 2 files changed, 4 insertions(+), 6 deletions(-)
include/lapi/mmap.h +2 -0
index 18547c0d47ed..98b6ade1e2ab 100644
@@ -19,6 +19,8 @@
 #ifndef LAPI_MMAP_H__
 #define LAPI_MMAP_H__
 
+#include "config.h"
+
 #ifndef MAP_HUGETLB
 # define MAP_HUGETLB 0x40000
 #endif
testcases/kernel/io/direct_io/diotest4.c +2 -6
index e4616e400abd..bf200cd41a27 100644
@@ -352,18 +352,14 @@ int main(int argc, char *argv[])
 	total++;
 
 	/* Test-10: read, write to a mmaped file */
-	shm_base = (char *)(((long)sbrk(0) + (shmsz - 1)) & ~(shmsz - 1));
-	if (shm_base == NULL) {
-		tst_brkm(TBROK, cleanup, "sbrk failed: %s", strerror(errno));
-	}
 	offset = 4096;
 	count = bufsize;
 	if ((fd = open(filename, O_DIRECT | O_RDWR)) < 0) {
 		tst_brkm(TBROK, cleanup, "can't open %s: %s",
 			 filename, strerror(errno));
 	}
-	shm_base = mmap(shm_base, 0x100000, PROT_READ | PROT_WRITE,
-			MAP_SHARED | MAP_FIXED, fd, 0);
+	shm_base = mmap(0, 0x100000, PROT_READ | PROT_WRITE,
+			MAP_SHARED, fd, 0);
 	if (shm_base == (caddr_t) - 1) {
 		tst_brkm(TBROK, cleanup, "can't mmap file: %s",
 			 strerror(errno));
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/1 direct_io: diotest4: fix mips MAP_SHARED/MAP_FIXED mmap

Hi!

include/lapi/mmap.h line 26

Quoted 13 lines
include/lapi/mmap.h +2 -0
index 18547c0d47ed..98b6ade1e2ab 100644
@@ -19,6 +19,8 @@
 #ifndef LAPI_MMAP_H__
 #define LAPI_MMAP_H__
 
+#include "config.h"
+
 #ifndef MAP_HUGETLB
 # define MAP_HUGETLB 0x40000
 #endif

Huh, this looks completely unrelevant to the patch itself, or do I miss something?

Quoted 15 lines
testcases/kernel/io/direct_io/diotest4.c +0 -4
index e4616e400abd..bf200cd41a27 100644
@@ -352,18 +352,14 @@ int main(int argc, char *argv[])
 	total++;
 
 	/* Test-10: read, write to a mmaped file */
-	shm_base = (char *)(((long)sbrk(0) + (shmsz - 1)) & ~(shmsz - 1));
-	if (shm_base == NULL) {
-		tst_brkm(TBROK, cleanup, "sbrk failed: %s", strerror(errno));
-	}
 	offset = 4096;
 	count = bufsize;
 	if ((fd = open(filename, O_DIRECT | O_RDWR)) < 0) {

testcases/kernel/io/direct_io/diotest4.c int main(int argc, char *argv[]) · line 365

Quoted 10 lines
testcases/kernel/io/direct_io/diotest4.c +2 -2
 		tst_brkm(TBROK, cleanup, "can't open %s: %s",
 			 filename, strerror(errno));
 	}
-	shm_base = mmap(shm_base, 0x100000, PROT_READ | PROT_WRITE,
-			MAP_SHARED | MAP_FIXED, fd, 0);
+	shm_base = mmap(0, 0x100000, PROT_READ | PROT_WRITE,
+			MAP_SHARED, fd, 0);
 	if (shm_base == (caddr_t) - 1) {
 		tst_brkm(TBROK, cleanup, "can't mmap file: %s",
 			 strerror(errno));

Agree here, the part that computes the mmap addres based on the sbrk() output looks completely bogus to me.

Signature · 2 lines
Cyril Hrubis
[email protected]
#2 lore raw

PATCH 1/1 direct_io: diotest4: fix mips MAP_SHARED/MAP_FIXED mmap

----- Original Message -----

Quoted 9 lines

Hi!

Quoted 8 lines
include/lapi/mmap.h +0 -0
index 18547c0d47ed..98b6ade1e2ab 100644
@@ -19,6 +19,8 @@
 #ifndef LAPI_MMAP_H__
 #define LAPI_MMAP_H__
 
Quoted 8 lines

include/lapi/mmap.h line 26

Quoted 5 lines
include/lapi/mmap.h +2 -0
+#include "config.h"
+
 #ifndef MAP_HUGETLB
 # define MAP_HUGETLB 0x40000
 #endif

Huh, this looks completely unrelevant to the patch itself, or do I miss something?

This fixes the original problem of bad alignment. Hunk below improves on it by removing MAP_FIXED altogether.

I could split it into 2 patches, if you prefer that.

Quoted 35 lines
Quoted 17 lines
testcases/kernel/io/direct_io/diotest4.c +0 -4
index e4616e400abd..bf200cd41a27 100644
@@ -352,18 +352,14 @@ int main(int argc, char *argv[])
 	total++;
 
 	/* Test-10: read, write to a mmaped file */
-	shm_base = (char *)(((long)sbrk(0) + (shmsz - 1)) & ~(shmsz - 1));
-	if (shm_base == NULL) {
-		tst_brkm(TBROK, cleanup, "sbrk failed: %s", strerror(errno));
-	}
 	offset = 4096;
 	count = bufsize;
 	if ((fd = open(filename, O_DIRECT | O_RDWR)) < 0) {

testcases/kernel/io/direct_io/diotest4.c int main(int argc, char *argv[]) · line 365

Quoted 10 lines
testcases/kernel/io/direct_io/diotest4.c +2 -2
 		tst_brkm(TBROK, cleanup, "can't open %s: %s",
 			 filename, strerror(errno));
 	}
-	shm_base = mmap(shm_base, 0x100000, PROT_READ | PROT_WRITE,
-			MAP_SHARED | MAP_FIXED, fd, 0);
+	shm_base = mmap(0, 0x100000, PROT_READ | PROT_WRITE,
+			MAP_SHARED, fd, 0);
 	if (shm_base == (caddr_t) - 1) {
 		tst_brkm(TBROK, cleanup, "can't mmap file: %s",
 			 strerror(errno));

Agree here, the part that computes the mmap addres based on the sbrk() output looks completely bogus to me.

--
Cyril Hrubis
[email protected]

#3 lore raw

PATCH 1/1 direct_io: diotest4: fix mips MAP_SHARED/MAP_FIXED mmap

Hi!

Quoted 17 lines
Quoted 8 lines
Quoted 8 lines
include/lapi/mmap.h +0 -0
index 18547c0d47ed..98b6ade1e2ab 100644
@@ -19,6 +19,8 @@
 #ifndef LAPI_MMAP_H__
 #define LAPI_MMAP_H__
 
Quoted 8 lines

include/lapi/mmap.h line 26

Quoted 5 lines
include/lapi/mmap.h +2 -0
+#include "config.h"
+
 #ifndef MAP_HUGETLB
 # define MAP_HUGETLB 0x40000
 #endif

Huh, this looks completely unrelevant to the patch itself, or do I miss something?

Quoted 2 lines

This fixes the original problem of bad alignment. Hunk below improves on it by removing MAP_FIXED altogether.

Ah, that is needed for the HAVE_SYS_SHM_H.

Quoted 1 line

I could split it into 2 patches, if you prefer that.

Yes please, both are acked, so just push them.

Quoted 34 lines
Quoted 34 lines
Quoted 16 lines
testcases/kernel/io/direct_io/diotest4.c +0 -4
index e4616e400abd..bf200cd41a27 100644
@@ -352,18 +352,14 @@ int main(int argc, char *argv[])
 	total++;
 
 	/* Test-10: read, write to a mmaped file */
-	shm_base = (char *)(((long)sbrk(0) + (shmsz - 1)) & ~(shmsz - 1));
-	if (shm_base == NULL) {
-		tst_brkm(TBROK, cleanup, "sbrk failed: %s", strerror(errno));
-	}
 	offset = 4096;
 	count = bufsize;
 	if ((fd = open(filename, O_DIRECT | O_RDWR)) < 0) {

testcases/kernel/io/direct_io/diotest4.c int main(int argc, char *argv[]) · line 365

Quoted 10 lines
testcases/kernel/io/direct_io/diotest4.c +2 -2
 		tst_brkm(TBROK, cleanup, "can't open %s: %s",
 			 filename, strerror(errno));
 	}
-	shm_base = mmap(shm_base, 0x100000, PROT_READ | PROT_WRITE,
-			MAP_SHARED | MAP_FIXED, fd, 0);
+	shm_base = mmap(0, 0x100000, PROT_READ | PROT_WRITE,
+			MAP_SHARED, fd, 0);
 	if (shm_base == (caddr_t) - 1) {
 		tst_brkm(TBROK, cleanup, "can't mmap file: %s",
 			 strerror(errno));

Agree here, the part that computes the mmap addres based on the sbrk() output looks completely bogus to me.

--
Cyril Hrubis
[email protected]

Signature · 2 lines
Cyril Hrubis
[email protected]
#4 lore raw

PATCH 1/1 direct_io: diotest4: fix mips MAP_SHARED/MAP_FIXED mmap

----- Original Message -----

Quoted 18 lines

Hi!

Quoted 17 lines
Quoted 16 lines

include/lapi/mmap.h line 26

Quoted 13 lines
include/lapi/mmap.h +2 -0
index 18547c0d47ed..98b6ade1e2ab 100644
@@ -19,6 +19,8 @@
 #ifndef LAPI_MMAP_H__
 #define LAPI_MMAP_H__
 
+#include "config.h"
+
 #ifndef MAP_HUGETLB
 # define MAP_HUGETLB 0x40000
 #endif

Huh, this looks completely unrelevant to the patch itself, or do I miss something?

Quoted 8 lines
Quoted 2 lines

This fixes the original problem of bad alignment. Hunk below improves on it by removing MAP_FIXED altogether.

Ah, that is needed for the HAVE_SYS_SHM_H.

Quoted 1 line

I could split it into 2 patches, if you prefer that.

Yes please, both are acked, so just push them.

Pushed as 2 patches.