This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Fix i386 memmove issue [BZ #22644]
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: Andrew Senkevich <andrew dot n dot senkevich at gmail dot com>
- Cc: Florian Weimer <fweimer at redhat dot com>, Szabolcs Nagy <szabolcs dot nagy at arm dot com>, Andreas Schwab <schwab at suse dot de>, nd <nd at arm dot com>, libc-alpha <libc-alpha at sourceware dot org>, Max Horn <max at quendi dot de>, thomas at grindinggear dot com
- Date: Mon, 19 Mar 2018 13:50:16 -0700
- Subject: Re: [PATCH] Fix i386 memmove issue [BZ #22644]
- References: <CAMXFM3s5b2a6o_CMS0cJ35oE3SuJUHssqLAUY94LSZhEEtG71A@mail.gmail.com> <mvm606ts2w6.fsf@suse.de> <CAMXFM3vqw7eiUT1XYVZWyW=eJkgEpjNc74A8mbmdKuR_hywN3w@mail.gmail.com> <CAMe9rOo4Ys=ScEKKnuv5XvYqTHjvFxSkXjch8ya8sX7dfP+q4Q@mail.gmail.com> <CAMXFM3vA=-LSK5VtKhhbs0g9faBdG0GjoAbNb6OPEbSt8dAA7A@mail.gmail.com> <mvmlgeojjmi.fsf@suse.de> <597954a1-85da-e524-1454-27c46af57413@redhat.com> <e7d81f18-315c-4434-a734-41029c0d387f@arm.com> <8c8ab285-bddc-90e6-a50e-7be89703367a@redhat.com> <CAMXFM3s39wwjHt=5PBp=Np8Rp=mXHfPwMq6_evtfx9MV08-mtQ@mail.gmail.com> <CAMe9rOoCe+qiOFK8=dwxm+eB=_wSoHbXdid0RKSUHqOtM7Wwfg@mail.gmail.com> <CAMXFM3vWVVbe0unZaueF+u62KJ-Sxa6ttMdZRaf_UesxMU=ZKA@mail.gmail.com> <CAMe9rOqsVpCV24f7g=kpFeOAQ_rkHuj-ZN5TDqs7BZ814BDQkg@mail.gmail.com> <CAMXFM3ujXmdgVyew8mY7h=tMhEH+qdALnke6iavecPnX-rVz_Q@mail.gmail.com>
On Mon, Mar 19, 2018 at 1:32 PM, Andrew Senkevich
<andrew.n.senkevich@gmail.com> wrote:
> 2018-03-19 20:38 GMT+01:00 H.J. Lu <hjl.tools@gmail.com>:
>> On Mon, Mar 19, 2018 at 12:29 PM, Andrew Senkevich
>> <andrew.n.senkevich@gmail.com> wrote:
>>> 2018-03-19 18:57 GMT+01:00 H.J. Lu <hjl.tools@gmail.com>:
>>>> On Mon, Mar 19, 2018 at 10:52 AM, Andrew Senkevich
>>>> <andrew.n.senkevich@gmail.com> wrote:
>>>>> 2018-03-19 16:33 GMT+01:00 Florian Weimer <fweimer@redhat.com>:
>>>>>> On 03/19/2018 03:25 PM, Szabolcs Nagy wrote:
>>>>>>>
>>>>>>> i thought not using MAP_FIXED is the 'non-overriding MAP_FIXED variant'
>>>>>
>>>>> Hmm, I tried so and had got much bigger address. But now I see it
>>>>> works (with size just 0x20000000 and 0x70000000 as a hint) in both 64
>>>>> and 32 bit cases, so proposing the following:
>>>>>
>>>>> diff --git a/string/test-memmove.c b/string/test-memmove.c
>>>>> index edc7a4c..fa4037d 100644
>>>>> --- a/string/test-memmove.c
>>>>> +++ b/string/test-memmove.c
>>>>> @@ -24,6 +24,7 @@
>>>>> # define TEST_NAME "memmove"
>>>>> #endif
>>>>> #include "test-string.h"
>>>>> +#include <support/test-driver.h>
>>>>>
>>>>> char *simple_memmove (char *, const char *, size_t);
>>>>>
>>>>> @@ -245,6 +246,59 @@ do_random_tests (void)
>>>>> }
>>>>> }
>>>>>
>>>>> +static void
>>>>> +do_test2 (void)
>>>>> +{
>>>>> + size_t size = 0x20000000;
>>>>> + uint32_t * large_buf;
>>>>> +
>>>>> + large_buf = mmap ((void*)0x70000000, size, PROT_READ | PROT_WRITE,
>>>>> + MAP_PRIVATE | MAP_ANON, -1, 0);
>>>>> +
>>>>> + if (large_buf == MAP_FAILED)
>>>>> + error (EXIT_UNSUPPORTED, errno, "Large mmap failed");
>>>>> +
>>>>> + if ((uintptr_t)large_buf > 0x80000000 - 128)
>>>>
>>>> Don't you need to test if address is too low such that
>>>> 0x80000000 - (uintptr_t)large_buf > 0x20000000?
>>>
>>> Indeed, thanks. Updated below.
>>>
>>> diff --git a/string/test-memmove.c b/string/test-memmove.c
>>> index edc7a4c..9f1437d 100644
>>> --- a/string/test-memmove.c
>>> +++ b/string/test-memmove.c
>>> @@ -24,6 +24,7 @@
>>> # define TEST_NAME "memmove"
>>> #endif
>>> #include "test-string.h"
>>> +#include <support/test-driver.h>
>>>
>>> char *simple_memmove (char *, const char *, size_t);
>>>
>>> @@ -245,6 +246,60 @@ do_random_tests (void)
>>> }
>>> }
>>>
>>> +static void
>>> +do_test2 (void)
>>> +{
>>> + size_t size = 0x20000000;
>>> + uint32_t * large_buf;
>>> +
>>> + large_buf = mmap ((void*)0x70000000, size, PROT_READ | PROT_WRITE,
>>> + MAP_PRIVATE | MAP_ANON, -1, 0);
>>> +
>>> + if (large_buf == MAP_FAILED)
>>> + error (EXIT_UNSUPPORTED, errno, "Large mmap failed");
>>> +
>>> + if ((uintptr_t)large_buf > 0x80000000 - 128
>>> + || 0x80000000 - (uintptr_t)large_buf > 0x20000000)
>>> + {
>>> + error (0, 0,"Large mmap allocated improperly");
>>> + ret = EXIT_UNSUPPORTED;
>>> + munmap((void *)large_buf, size);
>>> + return;
>>> + }
>>> +
>>> + size_t bytes_move = 0x80000000 - (uintptr_t)large_buf;
>>> + size_t arr_size = bytes_move / sizeof(uint32_t);
>>> + size_t i;
>>> +
>>> + FOR_EACH_IMPL (impl, 0)
>>> + {
>>> + for (i = 0; i < arr_size; i++)
>>> + large_buf[i] = i;
>>> +
>>> + uint32_t * dst = &large_buf[33];
>>> +
>>> +#ifdef TEST_BCOPY
>>> + CALL (impl, (char *)large_buf, (char *)dst, bytes_move);
>>> +#else
>>> + CALL (impl, (char *)dst, (char *)large_buf, bytes_move);
>>> +#endif
>>> +
>>> + for (i = 0; i < arr_size; i++)
>>> + {
>>> + if (dst[i] != i)
>>> + {
>>> + error (0, 0,
>>> + "Wrong result in function %s dst \"%p\" src \"%p\" offset \"%d\"",
>>> + impl->name, dst, large_buf, i);
>>> + ret = 1;
>>> + break;
>>> + }
>>> + }
>>> + }
>>> +
>>> + munmap((void *)large_buf, size);
>>> +}
>>> +
>>> int
>>> test_main (void)
>>> {
>>> @@ -284,6 +339,9 @@ test_main (void)
>>> }
>>>
>>> do_random_tests ();
>>> +
>>> + do_test2 ();
>>> +
>>> return ret;
>>> }
>>>
>>>
>>
>> Look good.
>>
>> Please submit the updated patch with the proper commit message.
>
> [BZ #22644] Fix i386 memmove issue.
>
>
> * sysdeps/i386/i686/multiarch/memcpy-sse2-unaligned.S: Fixed
> branch conditions.
> * string/test-memmove.c (do_test2): New testcase.
>
OK.
Thanks.
--
H.J.