Message ID | 20190806152611.280389-1-vsementsov@virtuozzo.com |
---|---|
State | New |
Headers | show |
Series | [v2] util/hbitmap: strict hbitmap_reset | expand |
On 06.08.19 17:26, Vladimir Sementsov-Ogievskiy wrote: > hbitmap_reset has an unobvious property: it rounds requested region up. > It may provoke bugs, like in recently fixed write-blocking mode of > mirror: user calls reset on unaligned region, not keeping in mind that > there are possible unrelated dirty bytes, covered by rounded-up region > and information of this unrelated "dirtiness" will be lost. > > Make hbitmap_reset strict: assert that arguments are aligned, allowing > only one exception when @start + @count == hb->orig_size. It's needed > to comfort users of hbitmap_next_dirty_area, which cares about > hb->orig_size. > > Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> > --- > > v2 based on Max's https://github.com/XanClic/qemu.git block > which will be merged soon to 4.1, and this patch goes to 4.2 > Based-on: https://github.com/XanClic/qemu.git block > > v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand > we all agreed to just assert alignment instead of aligning down > automatically. > > include/qemu/hbitmap.h | 5 +++++ > tests/test-hbitmap.c | 2 +- > util/hbitmap.c | 4 ++++ > 3 files changed, 10 insertions(+), 1 deletion(-) > > diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h > index 4afbe6292e..7865e819ca 100644 > --- a/include/qemu/hbitmap.h > +++ b/include/qemu/hbitmap.h > @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); > * @count: Number of bits to reset. > * > * Reset a consecutive range of bits in an HBitmap. > + * @start and @count must be aligned to bitmap granularity. The only exception > + * is resetting the tail of the bitmap: @count may be equal to @start + > + * hb->orig_size, s/@start + hb->orig_size/hb->orig_size - @start/, I think. > in this case @count may be not aligned. @start + @count +are With those fixed: Reviewed-by: Max Reitz <mreitz@redhat.com> > + * allowed to be greater than hb->orig_size, but only if @start < hb->orig_size > + * and @start + @count = ALIGN_UP(hb->orig_size, granularity). > */ > void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count); > > diff --git a/tests/test-hbitmap.c b/tests/test-hbitmap.c > index 592d8219db..2be56d1597 100644 > --- a/tests/test-hbitmap.c > +++ b/tests/test-hbitmap.c > @@ -423,7 +423,7 @@ static void test_hbitmap_granularity(TestHBitmapData *data, > hbitmap_test_check(data, 0); > hbitmap_test_set(data, 0, 3); > g_assert_cmpint(hbitmap_count(data->hb), ==, 4); > - hbitmap_test_reset(data, 0, 1); > + hbitmap_test_reset(data, 0, 2); > g_assert_cmpint(hbitmap_count(data->hb), ==, 2); > } > > diff --git a/util/hbitmap.c b/util/hbitmap.c > index bcc0acdc6a..586920cb52 100644 > --- a/util/hbitmap.c > +++ b/util/hbitmap.c > @@ -476,6 +476,10 @@ void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count) > /* Compute range in the last layer. */ > uint64_t first; > uint64_t last = start + count - 1; > + uint64_t gran = 1ULL << hb->granularity; > + > + assert(!(start & (gran - 1))); > + assert(!(count & (gran - 1)) || (start + count == hb->orig_size)); > > trace_hbitmap_reset(hb, start, count, > start >> hb->granularity, last >> hb->granularity); >
06.08.2019 19:09, Max Reitz wrote: > On 06.08.19 17:26, Vladimir Sementsov-Ogievskiy wrote: >> hbitmap_reset has an unobvious property: it rounds requested region up. >> It may provoke bugs, like in recently fixed write-blocking mode of >> mirror: user calls reset on unaligned region, not keeping in mind that >> there are possible unrelated dirty bytes, covered by rounded-up region >> and information of this unrelated "dirtiness" will be lost. >> >> Make hbitmap_reset strict: assert that arguments are aligned, allowing >> only one exception when @start + @count == hb->orig_size. It's needed >> to comfort users of hbitmap_next_dirty_area, which cares about >> hb->orig_size. >> >> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> >> --- >> >> v2 based on Max's https://github.com/XanClic/qemu.git block >> which will be merged soon to 4.1, and this patch goes to 4.2 >> Based-on: https://github.com/XanClic/qemu.git block >> >> v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand >> we all agreed to just assert alignment instead of aligning down >> automatically. >> >> include/qemu/hbitmap.h | 5 +++++ >> tests/test-hbitmap.c | 2 +- >> util/hbitmap.c | 4 ++++ >> 3 files changed, 10 insertions(+), 1 deletion(-) >> >> diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h >> index 4afbe6292e..7865e819ca 100644 >> --- a/include/qemu/hbitmap.h >> +++ b/include/qemu/hbitmap.h >> @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); >> * @count: Number of bits to reset. >> * >> * Reset a consecutive range of bits in an HBitmap. >> + * @start and @count must be aligned to bitmap granularity. The only exception >> + * is resetting the tail of the bitmap: @count may be equal to @start + >> + * hb->orig_size, > > s/@start + hb->orig_size/hb->orig_size - @start/, I think. Ha, I wanted to say start + count equal to orig_size. Yours is OK too of course. > >> in this case @count may be not aligned. @start + @count > > +are > > With those fixed: > > Reviewed-by: Max Reitz <mreitz@redhat.com> Thanks! > >> + * allowed to be greater than hb->orig_size, but only if @start < hb->orig_size >> + * and @start + @count = ALIGN_UP(hb->orig_size, granularity). >> */ >> void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count); >> >> diff --git a/tests/test-hbitmap.c b/tests/test-hbitmap.c >> index 592d8219db..2be56d1597 100644 >> --- a/tests/test-hbitmap.c >> +++ b/tests/test-hbitmap.c >> @@ -423,7 +423,7 @@ static void test_hbitmap_granularity(TestHBitmapData *data, >> hbitmap_test_check(data, 0); >> hbitmap_test_set(data, 0, 3); >> g_assert_cmpint(hbitmap_count(data->hb), ==, 4); >> - hbitmap_test_reset(data, 0, 1); >> + hbitmap_test_reset(data, 0, 2); >> g_assert_cmpint(hbitmap_count(data->hb), ==, 2); >> } >> >> diff --git a/util/hbitmap.c b/util/hbitmap.c >> index bcc0acdc6a..586920cb52 100644 >> --- a/util/hbitmap.c >> +++ b/util/hbitmap.c >> @@ -476,6 +476,10 @@ void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count) >> /* Compute range in the last layer. */ >> uint64_t first; >> uint64_t last = start + count - 1; >> + uint64_t gran = 1ULL << hb->granularity; >> + >> + assert(!(start & (gran - 1))); >> + assert(!(count & (gran - 1)) || (start + count == hb->orig_size)); >> >> trace_hbitmap_reset(hb, start, count, >> start >> hb->granularity, last >> hb->granularity); >> > >
On 8/6/19 12:19 PM, Vladimir Sementsov-Ogievskiy wrote: > 06.08.2019 19:09, Max Reitz wrote: >> On 06.08.19 17:26, Vladimir Sementsov-Ogievskiy wrote: >>> hbitmap_reset has an unobvious property: it rounds requested region up. >>> It may provoke bugs, like in recently fixed write-blocking mode of >>> mirror: user calls reset on unaligned region, not keeping in mind that >>> there are possible unrelated dirty bytes, covered by rounded-up region >>> and information of this unrelated "dirtiness" will be lost. >>> >>> Make hbitmap_reset strict: assert that arguments are aligned, allowing >>> only one exception when @start + @count == hb->orig_size. It's needed >>> to comfort users of hbitmap_next_dirty_area, which cares about >>> hb->orig_size. >>> >>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> >>> --- >>> >>> v2 based on Max's https://github.com/XanClic/qemu.git block >>> which will be merged soon to 4.1, and this patch goes to 4.2 >>> Based-on: https://github.com/XanClic/qemu.git block >>> >>> v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand >>> we all agreed to just assert alignment instead of aligning down >>> automatically. >>> >>> include/qemu/hbitmap.h | 5 +++++ >>> tests/test-hbitmap.c | 2 +- >>> util/hbitmap.c | 4 ++++ >>> 3 files changed, 10 insertions(+), 1 deletion(-) >>> >>> diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h >>> index 4afbe6292e..7865e819ca 100644 >>> --- a/include/qemu/hbitmap.h >>> +++ b/include/qemu/hbitmap.h >>> @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); >>> * @count: Number of bits to reset. >>> * >>> * Reset a consecutive range of bits in an HBitmap. >>> + * @start and @count must be aligned to bitmap granularity. The only exception >>> + * is resetting the tail of the bitmap: @count may be equal to @start + >>> + * hb->orig_size, >> >> s/@start + hb->orig_size/hb->orig_size - @start/, I think. > > Ha, I wanted to say start + count equal to orig_size. Yours is OK too of course. > >> >>> in this case @count may be not aligned. @start + @count >> >> +are >> >> With those fixed: >> >> Reviewed-by: Max Reitz <mreitz@redhat.com> > > Thanks! > I'll add this to the pile for 4.2, after I fix the rebase conflicts that arose from 4.1-rc4. --js
07.08.2019 19:27, John Snow wrote: > > > On 8/6/19 12:19 PM, Vladimir Sementsov-Ogievskiy wrote: >> 06.08.2019 19:09, Max Reitz wrote: >>> On 06.08.19 17:26, Vladimir Sementsov-Ogievskiy wrote: >>>> hbitmap_reset has an unobvious property: it rounds requested region up. >>>> It may provoke bugs, like in recently fixed write-blocking mode of >>>> mirror: user calls reset on unaligned region, not keeping in mind that >>>> there are possible unrelated dirty bytes, covered by rounded-up region >>>> and information of this unrelated "dirtiness" will be lost. >>>> >>>> Make hbitmap_reset strict: assert that arguments are aligned, allowing >>>> only one exception when @start + @count == hb->orig_size. It's needed >>>> to comfort users of hbitmap_next_dirty_area, which cares about >>>> hb->orig_size. >>>> >>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> >>>> --- >>>> >>>> v2 based on Max's https://github.com/XanClic/qemu.git block >>>> which will be merged soon to 4.1, and this patch goes to 4.2 >>>> Based-on: https://github.com/XanClic/qemu.git block >>>> >>>> v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand >>>> we all agreed to just assert alignment instead of aligning down >>>> automatically. >>>> >>>> include/qemu/hbitmap.h | 5 +++++ >>>> tests/test-hbitmap.c | 2 +- >>>> util/hbitmap.c | 4 ++++ >>>> 3 files changed, 10 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h >>>> index 4afbe6292e..7865e819ca 100644 >>>> --- a/include/qemu/hbitmap.h >>>> +++ b/include/qemu/hbitmap.h >>>> @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); >>>> * @count: Number of bits to reset. >>>> * >>>> * Reset a consecutive range of bits in an HBitmap. >>>> + * @start and @count must be aligned to bitmap granularity. The only exception >>>> + * is resetting the tail of the bitmap: @count may be equal to @start + >>>> + * hb->orig_size, >>> >>> s/@start + hb->orig_size/hb->orig_size - @start/, I think. >> >> Ha, I wanted to say start + count equal to orig_size. Yours is OK too of course. >> >>> >>>> in this case @count may be not aligned. @start + @count >>> >>> +are >>> >>> With those fixed: >>> >>> Reviewed-by: Max Reitz <mreitz@redhat.com> >> >> Thanks! >> > > I'll add this to the pile for 4.2, after I fix the rebase conflicts that > arose from 4.1-rc4. > Hi! Didn't you forget, or should I resend?
On 9/11/19 11:13 AM, Vladimir Sementsov-Ogievskiy wrote: > 07.08.2019 19:27, John Snow wrote: >> >> >> On 8/6/19 12:19 PM, Vladimir Sementsov-Ogievskiy wrote: >>> 06.08.2019 19:09, Max Reitz wrote: >>>> On 06.08.19 17:26, Vladimir Sementsov-Ogievskiy wrote: >>>>> hbitmap_reset has an unobvious property: it rounds requested region up. >>>>> It may provoke bugs, like in recently fixed write-blocking mode of >>>>> mirror: user calls reset on unaligned region, not keeping in mind that >>>>> there are possible unrelated dirty bytes, covered by rounded-up region >>>>> and information of this unrelated "dirtiness" will be lost. >>>>> >>>>> Make hbitmap_reset strict: assert that arguments are aligned, allowing >>>>> only one exception when @start + @count == hb->orig_size. It's needed >>>>> to comfort users of hbitmap_next_dirty_area, which cares about >>>>> hb->orig_size. >>>>> >>>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> >>>>> --- >>>>> >>>>> v2 based on Max's https://github.com/XanClic/qemu.git block >>>>> which will be merged soon to 4.1, and this patch goes to 4.2 >>>>> Based-on: https://github.com/XanClic/qemu.git block >>>>> >>>>> v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand >>>>> we all agreed to just assert alignment instead of aligning down >>>>> automatically. >>>>> >>>>> include/qemu/hbitmap.h | 5 +++++ >>>>> tests/test-hbitmap.c | 2 +- >>>>> util/hbitmap.c | 4 ++++ >>>>> 3 files changed, 10 insertions(+), 1 deletion(-) >>>>> >>>>> diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h >>>>> index 4afbe6292e..7865e819ca 100644 >>>>> --- a/include/qemu/hbitmap.h >>>>> +++ b/include/qemu/hbitmap.h >>>>> @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); >>>>> * @count: Number of bits to reset. >>>>> * >>>>> * Reset a consecutive range of bits in an HBitmap. >>>>> + * @start and @count must be aligned to bitmap granularity. The only exception >>>>> + * is resetting the tail of the bitmap: @count may be equal to @start + >>>>> + * hb->orig_size, >>>> >>>> s/@start + hb->orig_size/hb->orig_size - @start/, I think. >>> >>> Ha, I wanted to say start + count equal to orig_size. Yours is OK too of course. >>> >>>> >>>>> in this case @count may be not aligned. @start + @count >>>> >>>> +are >>>> >>>> With those fixed: >>>> >>>> Reviewed-by: Max Reitz <mreitz@redhat.com> >>> >>> Thanks! >>> >> >> I'll add this to the pile for 4.2, after I fix the rebase conflicts that >> arose from 4.1-rc4. >> > > Hi! > > Didn't you forget, or should I resend? > > I must have dropped the patch by accident during the rebasing. As an apology, I squashed in Max's suggestions from the list. Check that they look OK, please? Thanks, applied to my bitmaps tree: https://github.com/jnsnow/qemu/commits/bitmaps https://github.com/jnsnow/qemu.git --js
11.09.2019 20:59, John Snow wrote: > > > On 9/11/19 11:13 AM, Vladimir Sementsov-Ogievskiy wrote: >> 07.08.2019 19:27, John Snow wrote: >>> >>> >>> On 8/6/19 12:19 PM, Vladimir Sementsov-Ogievskiy wrote: >>>> 06.08.2019 19:09, Max Reitz wrote: >>>>> On 06.08.19 17:26, Vladimir Sementsov-Ogievskiy wrote: >>>>>> hbitmap_reset has an unobvious property: it rounds requested region up. >>>>>> It may provoke bugs, like in recently fixed write-blocking mode of >>>>>> mirror: user calls reset on unaligned region, not keeping in mind that >>>>>> there are possible unrelated dirty bytes, covered by rounded-up region >>>>>> and information of this unrelated "dirtiness" will be lost. >>>>>> >>>>>> Make hbitmap_reset strict: assert that arguments are aligned, allowing >>>>>> only one exception when @start + @count == hb->orig_size. It's needed >>>>>> to comfort users of hbitmap_next_dirty_area, which cares about >>>>>> hb->orig_size. >>>>>> >>>>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> >>>>>> --- >>>>>> >>>>>> v2 based on Max's https://github.com/XanClic/qemu.git block >>>>>> which will be merged soon to 4.1, and this patch goes to 4.2 >>>>>> Based-on: https://github.com/XanClic/qemu.git block >>>>>> >>>>>> v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand >>>>>> we all agreed to just assert alignment instead of aligning down >>>>>> automatically. >>>>>> >>>>>> include/qemu/hbitmap.h | 5 +++++ >>>>>> tests/test-hbitmap.c | 2 +- >>>>>> util/hbitmap.c | 4 ++++ >>>>>> 3 files changed, 10 insertions(+), 1 deletion(-) >>>>>> >>>>>> diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h >>>>>> index 4afbe6292e..7865e819ca 100644 >>>>>> --- a/include/qemu/hbitmap.h >>>>>> +++ b/include/qemu/hbitmap.h >>>>>> @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); >>>>>> * @count: Number of bits to reset. >>>>>> * >>>>>> * Reset a consecutive range of bits in an HBitmap. >>>>>> + * @start and @count must be aligned to bitmap granularity. The only exception >>>>>> + * is resetting the tail of the bitmap: @count may be equal to @start + >>>>>> + * hb->orig_size, >>>>> >>>>> s/@start + hb->orig_size/hb->orig_size - @start/, I think. >>>> >>>> Ha, I wanted to say start + count equal to orig_size. Yours is OK too of course. >>>> >>>>> >>>>>> in this case @count may be not aligned. @start + @count >>>>> >>>>> +are >>>>> >>>>> With those fixed: >>>>> >>>>> Reviewed-by: Max Reitz <mreitz@redhat.com> >>>> >>>> Thanks! >>>> >>> >>> I'll add this to the pile for 4.2, after I fix the rebase conflicts that >>> arose from 4.1-rc4. >>> >> >> Hi! >> >> Didn't you forget, or should I resend? >> >> > > I must have dropped the patch by accident during the rebasing. As an > apology, I squashed in Max's suggestions from the list. Check that they > look OK, please? > > Thanks, applied to my bitmaps tree: > > https://github.com/jnsnow/qemu/commits/bitmaps > https://github.com/jnsnow/qemu.git > Thanks! Still: Quote from your branch: > * Reset a consecutive range of bits in an HBitmap. > + * @start and @count must be aligned to bitmap granularity. The only exception > + * is resetting the tail of the bitmap: @count may be equal to hb->orig_size - > + * start, in this case @count may be not aligned. @start + @count are s/start/@start/ (corresponds to Max's comment, too) Also, I'm not sure about "are" suggested by Max. "are" is for plural, but here I meant one object: sum of @start and @count. So, you may use exactly "Sum of @start and @count is" or "(@start + @count) sum is" or just "(@start + @count) is", whichever you like more. > + * allowed to be greater than hb->orig_size, but only if @start < hb->orig_size > + * and @start + @count = ALIGN_UP(hb->orig_size, granularity). > */ > void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count);
On 9/12/19 4:20 AM, Vladimir Sementsov-Ogievskiy wrote: > 11.09.2019 20:59, John Snow wrote: >> >> >> On 9/11/19 11:13 AM, Vladimir Sementsov-Ogievskiy wrote: >>> 07.08.2019 19:27, John Snow wrote: >>>> >>>> >>>> On 8/6/19 12:19 PM, Vladimir Sementsov-Ogievskiy wrote: >>>>> 06.08.2019 19:09, Max Reitz wrote: >>>>>> On 06.08.19 17:26, Vladimir Sementsov-Ogievskiy wrote: >>>>>>> hbitmap_reset has an unobvious property: it rounds requested region up. >>>>>>> It may provoke bugs, like in recently fixed write-blocking mode of >>>>>>> mirror: user calls reset on unaligned region, not keeping in mind that >>>>>>> there are possible unrelated dirty bytes, covered by rounded-up region >>>>>>> and information of this unrelated "dirtiness" will be lost. >>>>>>> >>>>>>> Make hbitmap_reset strict: assert that arguments are aligned, allowing >>>>>>> only one exception when @start + @count == hb->orig_size. It's needed >>>>>>> to comfort users of hbitmap_next_dirty_area, which cares about >>>>>>> hb->orig_size. >>>>>>> >>>>>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> >>>>>>> --- >>>>>>> >>>>>>> v2 based on Max's https://github.com/XanClic/qemu.git block >>>>>>> which will be merged soon to 4.1, and this patch goes to 4.2 >>>>>>> Based-on: https://github.com/XanClic/qemu.git block >>>>>>> >>>>>>> v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand >>>>>>> we all agreed to just assert alignment instead of aligning down >>>>>>> automatically. >>>>>>> >>>>>>> include/qemu/hbitmap.h | 5 +++++ >>>>>>> tests/test-hbitmap.c | 2 +- >>>>>>> util/hbitmap.c | 4 ++++ >>>>>>> 3 files changed, 10 insertions(+), 1 deletion(-) >>>>>>> >>>>>>> diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h >>>>>>> index 4afbe6292e..7865e819ca 100644 >>>>>>> --- a/include/qemu/hbitmap.h >>>>>>> +++ b/include/qemu/hbitmap.h >>>>>>> @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); >>>>>>> * @count: Number of bits to reset. >>>>>>> * >>>>>>> * Reset a consecutive range of bits in an HBitmap. >>>>>>> + * @start and @count must be aligned to bitmap granularity. The only exception >>>>>>> + * is resetting the tail of the bitmap: @count may be equal to @start + >>>>>>> + * hb->orig_size, >>>>>> >>>>>> s/@start + hb->orig_size/hb->orig_size - @start/, I think. >>>>> >>>>> Ha, I wanted to say start + count equal to orig_size. Yours is OK too of course. >>>>> >>>>>> >>>>>>> in this case @count may be not aligned. @start + @count >>>>>> >>>>>> +are >>>>>> >>>>>> With those fixed: >>>>>> >>>>>> Reviewed-by: Max Reitz <mreitz@redhat.com> >>>>> >>>>> Thanks! >>>>> >>>> >>>> I'll add this to the pile for 4.2, after I fix the rebase conflicts that >>>> arose from 4.1-rc4. >>>> >>> >>> Hi! >>> >>> Didn't you forget, or should I resend? >>> >>> >> >> I must have dropped the patch by accident during the rebasing. As an >> apology, I squashed in Max's suggestions from the list. Check that they >> look OK, please? >> >> Thanks, applied to my bitmaps tree: >> >> https://github.com/jnsnow/qemu/commits/bitmaps >> https://github.com/jnsnow/qemu.git >> > > Thanks! Still: > > Quote from your branch: > > > * Reset a consecutive range of bits in an HBitmap. > > + * @start and @count must be aligned to bitmap granularity. The only exception > > + * is resetting the tail of the bitmap: @count may be equal to hb->orig_size - > > + * start, in this case @count may be not aligned. @start + @count are > > s/start/@start/ (corresponds to Max's comment, too) > OK, you got it. > Also, I'm not sure about "are" suggested by Max. "are" is for plural, but here I meant > one object: sum of @start and @count. > There's not great agreement universally about how to treat things like collective nouns. Sometimes "Data" is singular, but sometimes it's plural. "It depends." In this case, "start + count" refers to one sum, but two constituent pieces, so it's functioning like a collective noun. We might say "a + b (together) /are/ ..." but also "the sum of a + b /is/". > So, you may use exactly "Sum of @start and @count is" or "(@start + @count) sum is" or > just "(@start + @count) is", whichever you like more. > I like using "the sum of @x and @y is" for being grammatically unambiguous. updated and pushed. (Sorry about my language again! --js)
Am 13.09.2019 um 20:49 hat John Snow geschrieben: > On 9/12/19 4:20 AM, Vladimir Sementsov-Ogievskiy wrote: > > Also, I'm not sure about "are" suggested by Max. "are" is for plural, but here I meant > > one object: sum of @start and @count. > > > > There's not great agreement universally about how to treat things like > collective nouns. Sometimes "Data" is singular, but sometimes it's > plural. "It depends." > > In this case, "start + count" refers to one sum, but two constituent > pieces, so it's functioning like a collective noun. > > We might say "a + b (together) /are/ ..." but also "the sum of a + b /is/". > > > So, you may use exactly "Sum of @start and @count is" or "(@start + @count) sum is" or > > just "(@start + @count) is", whichever you like more. > > > > I like using "the sum of @x and @y is" for being grammatically unambiguous. > > updated and pushed. > > (Sorry about my language again! --js) Time to revive https://patchwork.kernel.org/patch/8725621/? ;-) Kevin
On 9/16/19 4:00 AM, Kevin Wolf wrote: > Am 13.09.2019 um 20:49 hat John Snow geschrieben: >> On 9/12/19 4:20 AM, Vladimir Sementsov-Ogievskiy wrote: >>> Also, I'm not sure about "are" suggested by Max. "are" is for plural, but here I meant >>> one object: sum of @start and @count. >>> >> >> There's not great agreement universally about how to treat things like >> collective nouns. Sometimes "Data" is singular, but sometimes it's >> plural. "It depends." >> >> In this case, "start + count" refers to one sum, but two constituent >> pieces, so it's functioning like a collective noun. >> >> We might say "a + b (together) /are/ ..." but also "the sum of a + b /is/". >> >>> So, you may use exactly "Sum of @start and @count is" or "(@start + @count) sum is" or >>> just "(@start + @count) is", whichever you like more. >>> >> >> I like using "the sum of @x and @y is" for being grammatically unambiguous. >> >> updated and pushed. >> >> (Sorry about my language again! --js) > > Time to revive https://patchwork.kernel.org/patch/8725621/? ;-) > > Kevin > Ja, bitte.
diff --git a/include/qemu/hbitmap.h b/include/qemu/hbitmap.h index 4afbe6292e..7865e819ca 100644 --- a/include/qemu/hbitmap.h +++ b/include/qemu/hbitmap.h @@ -132,6 +132,11 @@ void hbitmap_set(HBitmap *hb, uint64_t start, uint64_t count); * @count: Number of bits to reset. * * Reset a consecutive range of bits in an HBitmap. + * @start and @count must be aligned to bitmap granularity. The only exception + * is resetting the tail of the bitmap: @count may be equal to @start + + * hb->orig_size, in this case @count may be not aligned. @start + @count + * allowed to be greater than hb->orig_size, but only if @start < hb->orig_size + * and @start + @count = ALIGN_UP(hb->orig_size, granularity). */ void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count); diff --git a/tests/test-hbitmap.c b/tests/test-hbitmap.c index 592d8219db..2be56d1597 100644 --- a/tests/test-hbitmap.c +++ b/tests/test-hbitmap.c @@ -423,7 +423,7 @@ static void test_hbitmap_granularity(TestHBitmapData *data, hbitmap_test_check(data, 0); hbitmap_test_set(data, 0, 3); g_assert_cmpint(hbitmap_count(data->hb), ==, 4); - hbitmap_test_reset(data, 0, 1); + hbitmap_test_reset(data, 0, 2); g_assert_cmpint(hbitmap_count(data->hb), ==, 2); } diff --git a/util/hbitmap.c b/util/hbitmap.c index bcc0acdc6a..586920cb52 100644 --- a/util/hbitmap.c +++ b/util/hbitmap.c @@ -476,6 +476,10 @@ void hbitmap_reset(HBitmap *hb, uint64_t start, uint64_t count) /* Compute range in the last layer. */ uint64_t first; uint64_t last = start + count - 1; + uint64_t gran = 1ULL << hb->granularity; + + assert(!(start & (gran - 1))); + assert(!(count & (gran - 1)) || (start + count == hb->orig_size)); trace_hbitmap_reset(hb, start, count, start >> hb->granularity, last >> hb->granularity);
hbitmap_reset has an unobvious property: it rounds requested region up. It may provoke bugs, like in recently fixed write-blocking mode of mirror: user calls reset on unaligned region, not keeping in mind that there are possible unrelated dirty bytes, covered by rounded-up region and information of this unrelated "dirtiness" will be lost. Make hbitmap_reset strict: assert that arguments are aligned, allowing only one exception when @start + @count == hb->orig_size. It's needed to comfort users of hbitmap_next_dirty_area, which cares about hb->orig_size. Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> --- v2 based on Max's https://github.com/XanClic/qemu.git block which will be merged soon to 4.1, and this patch goes to 4.2 Based-on: https://github.com/XanClic/qemu.git block v1 was "[PATCH] util/hbitmap: fix unaligned reset", and as I understand we all agreed to just assert alignment instead of aligning down automatically. include/qemu/hbitmap.h | 5 +++++ tests/test-hbitmap.c | 2 +- util/hbitmap.c | 4 ++++ 3 files changed, 10 insertions(+), 1 deletion(-)