GitLab

Do free and fclose when error occur.

Change-Id: Ifc611579cf919eeb3fbbf1dff70e7f851a26e083
Signed-off-by: Daegeun Song <daegeun.song@lge.com>

This reverts commit 4bf06a51.

Bug: 17360804
Bug: 17332389
Change-Id: I7e4d55215f391f5b3f71388943e9d95e5eff6f81

see comment above pagemap_read() from the kernel

Bug: 17332389
Change-Id: Iaec9a2c8d2783f5c1e7ba06d9c7156305abe0453
Signed-off-by: Daniel Rosenberg <drosen@google.com>

Use uint64_t and lseek64 to handle 64-bit virtual addresses
when libpagemap is compiled as a 32-bit library.

Change-Id: Ie4b6c7ef05aac604011f3ee28b059d9dfcd63edb

libpagemap was storing a virtual pfn in an int, which works on arm64
with 39 bits of virtual address space but fails on x86_64.  Use an
unsigned long instead.

Fixes errors when running procrank on x86_64:
warning: could not read usage for 1

Change-Id: I171c8ee49faa51accf3c1bb69059d549aee04979

/proc/pid/maps may report a zero-length memory region for a 4kB
PROT_GROWSDOWN region because it subtracts 4kB for the guard page.
Return 0 instead of -1 when this occurs, and set range_out to NULL
and len to 0.  All existing callers of pm_process_pagemap_range
will not dereference range_out if len is 0.

Bug: 14683277
Change-Id: If405651ad034dda780b93fab2dc616e177a0b917

procrank/librank not impacted because they end quickly. But other
programs that use libpagemap and last for long time can easily see this
memory leak.

Change-Id: I8c9e9444555bef9145c9d89850987a29f15a9b3b
Signed-off-by: Carton He <carton.he@marvell.com>

Add pm_process_usage flags to get memory usage by a process, only
counting pages with specified flags set.

Change-Id: I900b673ddbb5ae92312773a8670dd59769617268

Mappings that are not from a file do not have a name.  The sscanf
will read all of the fields up to the name, and then leave name
untouched.  This causes the previous name to be reused.  Reset
name to an empty string before each call to sscanf.

Change-Id: Ib146732983eb074d0d4773be094efa0b672f5ed2

Add pm_process_usage flags to get memory usage by a process, only
counting pages with specified flags set.

Change-Id: I900b673ddbb5ae92312773a8670dd59769617268

Mappings that are not from a file do not have a name.  The sscanf
will read all of the fields up to the name, and then leave name
untouched.  This causes the previous name to be reused.  Reset
name to an empty string before each call to sscanf.

Change-Id: Ib146732983eb074d0d4773be094efa0b672f5ed2

Fixed these problems:
1. Page swapped bit was not extracted correctly.
2. Pages were ignored when page present bit is not set.
3. Bit operations were not correct.
4. There was a compiler warning about unsigned/signed comparision.
5. Line limit was too short for the map file. This was causing procrank
   to generate a warning and remove the related process from results.

Change-Id: Ifed3913a49b15f59010cfa842090a13228074df9

Instead of handling maps with the name "[vectors]" specially,
silently ignore EOF when reading from /proc/<pid>/pagemap, which
occurs any time a a mapping is outside of the userspace range.

Change-Id: I674ade1eab6fd7732c6d9e120d0750cca5415b25

In kernel 2.6.37, the vector page was added to /proc/<pid>/maps,
but because it is located above TASK_SIZE (usually 0xbf000000),
it is considered to be in the kernel's address space, not the
process', so it doesn't show up in /proc/<pid>/pagemap.

When the vector page is detected, using the name "[vectors]",
remove it from the map.

Change-Id: I5f0758bbe5d2b927056fa9fee684fea63dd0fa4b