build error on x86, file dmaSendFd.h, line 85

[chenm001]

The addr is 32bits in x86 platform, the sg_phys() return dma_addr_t, it is mismatch to 'long'

[hanw]

can you post the error message?

[jankcorn]

Are you sure that addr is 32 bits on x86? we always run 64-bit versions,
so seems peculiar.

In addition, I am accustomed to sizeof(long) == sizeof(void *), for
example, see the last line of the next-to-last paragraph of:
http://www.makelinux.net/ldd3/chp-11-sect-1
( was looking for this in the c std, but can't find similar text... :-( )

Thanks
jca

On Sun, Jan 17, 2016 at 9:00 AM, Han Wang [email protected] wrote:

can you post the error message?

—
Reply to this email directly or view it on GitHub
#106 (comment)
.

[chenm001]

Below is error message, the problem is Shift 32 bits and OR with addr
btw: the dma_addr_t defined in Linux header, so it maybe 32 or 64bits depends OS configure.

---

prepare_bin_target
ubuntu.exe
cc1plus: warnings being treated as errors
In file included from /chen/connectal/cpp/dmaManager.c:35:
/chen/connectal/cpp/dmaSendFd.h: In function int send_fd_to_portal(PortalInternal_, int, int, int):
/chen/connectal/cpp/dmaSendFd.h:85: error: left shift count >= width of type
make[2]: *_* [ubuntu.exe] Error 1
make[1]: *** [ubuntu.exe] Error 2
make: *** [build.bluesim] Error 2
rhel:/connectal/examples/echo#
************* END **************************************

At 2016-01-18 01:18:10,jankcorn [email protected] wrote:
Are you sure that addr is 32 bits on x86? we always run 64-bit versions,
so seems peculiar.

In addition, I am accustomed to sizeof(long) == sizeof(void *), for
example, see the last line of the next-to-last paragraph of:
http://www.makelinux.net/ldd3/chp-11-sect-1
( was looking for this in the c std, but can't find similar text... :-( )

Thanks
jca

On Sun, Jan 17, 2016 at 9:00 AM, Han Wang [email protected] wrote:

can you post the error message?

—
Reply to this email directly or view it on GitHub
#106 (comment)
.

—
Reply to this email directly or view it on GitHub.

[jankcorn]

I see. This is code only for simulations.
It appears that simulation builds (xsim, bluesim, modelsim, verilator) have
to be run on 64-bit architectures (which, in practice, are always x86).

I would highly recommend switching to a machine with a 64-bit OS. (since
the 32 bit os can't address much memory anyway for simulation...)

Is that possible?
jca

On Sun, Jan 17, 2016 at 9:54 PM, chenm001 [email protected] wrote:

Below is error message, the problem is Shift 32 bits and OR with addr
btw: the dma_addr_t defined in Linux header, so it maybe 32 or 64bits
depends OS configure.

---

prepare_bin_target
ubuntu.exe
cc1plus: warnings being treated as errors
In file included from /chen/connectal/cpp/dmaManager.c:35:
/chen/connectal/cpp/dmaSendFd.h: In function int
send_fd_to_portal(PortalInternal_, int, int, int):
/chen/connectal/cpp/dmaSendFd.h:85: error: left shift count >= width of
type
make[2]: *_* [ubuntu.exe] Error 1
make[1]: *** [ubuntu.exe] Error 2
make: *** [build.bluesim] Error 2
rhel:/connectal/examples/echo#
************* END **************************************

At 2016-01-18 01:18:10,jankcorn [email protected] wrote:
Are you sure that addr is 32 bits on x86? we always run 64-bit versions,
so seems peculiar.

In addition, I am accustomed to sizeof(long) == sizeof(void *), for
example, see the last line of the next-to-last paragraph of:
http://www.makelinux.net/ldd3/chp-11-sect-1
( was looking for this in the c std, but can't find similar text... :-( )

Thanks
jca

On Sun, Jan 17, 2016 at 9:00 AM, Han Wang [email protected]
wrote:

can you post the error message?

—
Reply to this email directly or view it on GitHub
<
#106 (comment)

.

—
Reply to this email directly or view it on GitHub.

—
Reply to this email directly or view it on GitHub
#106 (comment)
.

[jameyhicks]

I feel like I'm hearing half of a phone conversation.

@chenm001, it would be possible to put together a simulation configuration that supports 32-bit x86, but I agree with @jankcorn that it would be better to switch to 64-bit x86. Because who knows how many other places we assumed x86 was 64bit.

[chenm001]

Agree, for really world, we may use x64 OS.
In simluate enviorment, I simple workaround by modify local to 'uint64_t addr;'

[jameyhicks]

If that change works, I think we could update dmaSendFd.h accordingly.

We do not have any 32-bit x86 systems to test with, so please let us know
if any other changes are required.

On Fri, Jan 22, 2016 at 7:38 AM chenm001 [email protected] wrote:

Agree, for really world, we may use x64 OS.
In simluate enviorment, I simple workaround by modify local to 'uint64_t
addr;'

—
Reply to this email directly or view it on GitHub
#106 (comment)
.

[chenm001]

I will do more test and let you know later

At 2016-01-22 22:22:07,"Jamey Hicks" [email protected] wrote:
If that change works, I think we could update dmaSendFd.h accordingly.

We do not have any 32-bit x86 systems to test with, so please let us know
if any other changes are required.

On Fri, Jan 22, 2016 at 7:38 AM chenm001 [email protected] wrote:

Agree, for really world, we may use x64 OS.
In simluate enviorment, I simple workaround by modify local to 'uint64_t
addr;'

—
Reply to this email directly or view it on GitHub
#106 (comment)
.

—
Reply to this email directly or view it on GitHub.