[PATCH] fs: semihosting: Use correct variable for error check

[Heinrich Schuchardt]

On 10/17/25 19:08, Sean Anderson wrote:
> On 10/15/25 12:28, Tom Rini wrote:
>> On Thu, Oct 02, 2025 at 10:36:07AM -0400, Sean Anderson wrote:
>>> On 10/2/25 05:52, Heinrich Schuchardt wrote:
>>>> On 10/2/25 11:39, Andrew Goodbody wrote:
>>>>> After calling a function that can return an error, the test to detect
>>>>> that error should use the return value not a different variable. Fix it.
>>>>>
>>>>> This issue was found by Smatch.
>>>>>
>>>>
>>>> Fixes: f676b45151c3 ("fs: Add semihosting filesystem")
>>>>
>>>>> Signed-off-by: Andrew Goodbody <andrew.goodbody at linaro.org>
>>>>> ---
>>>>>    fs/semihostingfs.c | 2 +-
>>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/fs/semihostingfs.c b/fs/semihostingfs.c
>>>>> index 77e39ca407e4d240a1fd573497c5b6b908816454..9d7a136b9ba9b035545b34b31df58e2d65de7db9 100644
>>>>> --- a/fs/semihostingfs.c
>>>>> +++ b/fs/semihostingfs.c
>>>>> @@ -35,7 +35,7 @@ static int smh_fs_read_at(const char *filename, loff_t pos, void *buffer,
>>>>>        }
>>>>>        if (!maxsize) {
>>>>>            size = smh_flen(fd);
>>>>> -        if (ret < 0) {
>>>>> +        if (size < 0) {
>>>>
>>>> The ARM specification (https://cas5-0-urlprotect.trendmicro.com:443/wis/clicktime/v1/query?url=https%3a%2f%2fdeveloper.arm.com%2fdocumentation%2fdui0203%2fj%2fsemihosting%2fsemihosting%2doperations%2fsys%2dflen%2d%2d0x0c%2d&umid=f17ab8fc-2d9d-4cce-91e2-e7e66f117613&rct=1759398729&auth=d807158c60b7d2502abde8a2fc01f40662980862-e0799d2d4d61f93cae033d54733c8fa50ef84918) has:
>>>>
>>>> SYS_FLEN (0x0C)
>>>>
>>>> Returns the length of a specified file.
>>>> On exit, R0 contains:
>>>>      the current length of the file object, if the call is successful
>>>>      -1 if an error occurs.
>>>>
>>>> Please, consider that the file length on 32bit systems may exceed 2^31 You must not consider this as an error.
>>>>
>>>> %s/if (size < 0)/if (size == -1L)/
>>>
>>> This cannot occur because on a 32-bit system SYS_FLEN only returns
>>> 32-bits of information. The host must detect files that exceed 2 GiB in
>>> size and return -1 (simulating EOVERFLOW).
>>
>> OK, but this is for 64-bit systems too, isn't it?
>>
> 
> Sorry, I worded that poorly. What I mean is that the spec says
> 
> | On exit, R0 contains:
> |     the current length of the file object, if the call is successful
> |     -1 if an error occurs.
> 
> which implies that R0 is a signed integer (otherwise they would have

The only thing that is certain here is that the register will be filled 
with (unsigned long)-1L if an error occurs and that any other value is 
not an error.

Everything else is your personal interpretation which does not match the 
existing implementation by Linaro in QEMU.

> specified 0xffffffff or 0xffffffffffffffff as the error value). So
> negative values (other than -1) indicate a negative file size (or are
> unspecified).
> 
> However, both OpenOCD and QEMU just take the return from stat(2) and
> copy it into R0. With a 32-bit host there is no problem since stat(2)
> will itself fail for files over 2 GiB. With a 32-bit target, this
> produces erroneous values for files larger than 4 GiB. Files sizes
> between 2 and 4 GiB could be recovered, but I don't think we should do
> that. On a 64-bit target, there is again no ambiguity (at least until 8

You still don't provide any reason why you don't want to support files 
in the range [2 GiB, 4GiB[ on a 32 bit system.

Instead of restricting U-Boot we should add a fix in QEMU to report an 
error if the filesize exceeds ULONG_MAX - 1.

This is the only way we can ensure that a 5 GiB file is not reported as 
1 GiB in U-Boot as we see it today.

Best regards

Heinrich

> EiB file sizes).
> 
> I don't think we should try to interpret negative values because it's
> ambiguous in the spec, the hosts known to return such values do so
> because of flaws in their implementation, and large files themselves
> have dubious merit.
> 
> --Sean