Skip to contentSkip to conversation

pages already locked on ia64

DISCUSSION Christian Hinkelbein 3 participants 4 messages Nov 14, 2003 – Nov 24, 2003

lore
Share & tools
lore
#0 lore raw

Hi,
i have the need to lock malloc'ed user mem between kernel calls.

On ia32 i succesfully do a lock_page() on every page returned by get_user_pages(), lateron i do unlock_page().
On ia64 sometimes (on ia32 seemingly never) the first page is already locked, just after malloc(). For a quick hack i check all pages before trying to lock them, so that i am later just unlocking the not previously locked ones.

I am quite shure that my approach is totally ill. Could someone point me to some piece of information that helps me get cured?

christian

#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Christian Hinkelbein

Christian> Hi, i have the need to lock malloc'ed user mem between Christian> kernel calls.

Christian> On ia32 i succesfully do a lock_page() on every page Christian> returned by get_user_pages(), lateron i do unlock_page(). Christian> On ia64 sometimes (on ia32 seemingly never) the first Christian> page is already locked, just after malloc(). For a quick Christian> hack i check all pages before trying to lock them, so Christian> that i am later just unlocking the not previously locked Christian> ones.

Umh, what page-size did you use?

In any case, a minimal test-program that reproduces the problem would probably help you get an answer much faster.

Thanks,

	--david
#2 lore raw
In reply to Christian Hinkelbein
Quoted 8 lines · David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Christian Hinkelbein

Christian> On ia32 i succesfully do a lock_page() on every page Christian> returned by get_user_pages(), lateron i do unlock_page(). Christian> On ia64 sometimes (on ia32 seemingly never) the first Christian> page is already locked, just after malloc(). For a quick Christian> hack ......

Quoted 4 lines

Umh, what page-size did you use?

In any case, a minimal test-program that reproduces the problem would probably help you get an answer much faster.

Uuuh, thanks for the hint. It was the bigger page-size (16k) that triggered an old bug in my driver.

christian

#3 lore raw
In reply to Christian Hinkelbein
Quoted 11 lines · Christian Hinkelbein
Quoted 8 lines · David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Christian Hinkelbein

Christian> On ia32 i succesfully do a lock_page() on every page Christian> returned by get_user_pages(), lateron i do unlock_page(). Christian> On ia64 sometimes (on ia32 seemingly never) the first Christian> page is already locked, just after malloc(). For a quick Christian> hack ......

Quoted 8 lines
Quoted 4 lines

Umh, what page-size did you use?

In any case, a minimal test-program that reproduces the problem would probably help you get an answer much faster.

Uuuh, thanks for the hint. It was the bigger page-size (16k) that triggered an old bug in my driver.

Your driver is still buggy - you usually cannot lock more that one page at a time. Trivial case: mmap the same page of a file at two consecutive addresses and ask your driver to access that 2*PAGE_SIZE chunk of memory. It will deadlock.

If you wish to make sure that the pages don't go away then elevating their refcount is sufficient.