Message ID | 1317020069-16355-2-git-send-email-egouriou@google.com |
---|---|
State | Accepted, archived |
Headers | show |
On Sun, 25 Sep 2011 23:54:29 -0700, Eric Gouriou <egouriou@google.com> wrote: Seems patch was generated with "git diff -R" > ext4_ext_insert_extent() (respectively ext4_ext_insert_index()) > was using EXT_MAX_EXTENT() (resp. EXT_MAX_INDEX()) to determine > how many entries needed to be moved beyond the insertion point. > In practice this means that (320 - I) * 24 bytes were memmove()'d > when I is the insertion point, rather than (#entries - I) * 24 bytes. > > This patch uses EXT_LAST_EXTENT() (resp. EXT_LAST_INDEX()) instead > to only move existing entries. The code flow is also simplified > slightly to highlight similarities and reduce code duplication in > the insertion logic. > > This patch reduces system CPU consumption by over 25% on a 4kB > synchronous append DIO write workload when used with the > pre-2.6.39 x86_64 memmove() implementation. With the much faster > 2.6.39 memmove() implementation we still see a decrease in > system CPU usage between 2% and 7%. > > Note that the ext_debug() output changes with this patch, splitting > some log information between entries. Users of the ext_debug() output > should note that the "move %d" units changed from reporting the number > of bytes moved to reporting the number of entries moved. > > Signed-off-by: Eric Gouriou <egouriou@google.com> > --- > fs/ext4/extents.c | 85 ++++++++++++++++++++++++++--------------------------- > 1 files changed, 42 insertions(+), 43 deletions(-) > > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c > index 0a7dd85..30dd641 100644 > --- a/fs/ext4/extents.c > +++ b/fs/ext4/extents.c > @@ -751,31 +751,25 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, > return -EIO; > } > > - len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; > if (logical > le32_to_cpu(curp->p_idx->ei_block)) { > /* insert after */ > - if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { > - len = (len - 1) * sizeof(struct ext4_extent_idx); > - len = len < 0 ? 0 : len; > - ext_debug("insert new index %d after: %llu. " > - "move %d from 0x%p to 0x%p\n", > - logical, ptr, len, > - (curp->p_idx + 1), (curp->p_idx + 2)); > - memmove(curp->p_idx + 2, curp->p_idx + 1, len); > - } > + ext_debug("insert new index %d after: %llu\n", logical, ptr); > ix = curp->p_idx + 1; > } else { > /* insert before */ > - len = len * sizeof(struct ext4_extent_idx); > - len = len < 0 ? 0 : len; > - ext_debug("insert new index %d before: %llu. " > - "move %d from 0x%p to 0x%p\n", > - logical, ptr, len, > - curp->p_idx, (curp->p_idx + 1)); > - memmove(curp->p_idx + 1, curp->p_idx, len); > + ext_debug("insert new index %d before: %llu\n", logical, ptr); > ix = curp->p_idx; > } > > + len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; > + BUG_ON(len < 0); > + if (len > 0) { > + ext_debug("insert new index %d: " > + "move %d indices from 0x%p to 0x%p\n", > + logical, len, ix, ix + 1); > + memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); > + } > + > ix->ei_block = cpu_to_le32(logical); > ext4_idx_store_pblock(ix, ptr); > le16_add_cpu(&curp->p_hdr->eh_entries, 1); > @@ -1778,41 +1772,46 @@ has_space: > ext4_ext_pblock(newext), > ext4_ext_is_uninitialized(newext), > ext4_ext_get_actual_len(newext)); > - path[depth].p_ext = EXT_FIRST_EXTENT(eh); > - } else if (le32_to_cpu(newext->ee_block) > + nearex = EXT_FIRST_EXTENT(eh); > + } else { > + if (le32_to_cpu(newext->ee_block) > > le32_to_cpu(nearex->ee_block)) { > -/* BUG_ON(newext->ee_block == nearex->ee_block); */ > - if (nearex != EXT_LAST_EXTENT(eh)) { > - len = EXT_MAX_EXTENT(eh) - nearex; > - len = (len - 1) * sizeof(struct ext4_extent); > - len = len < 0 ? 0 : len; > - ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " > - "move %d from 0x%p to 0x%p\n", > + /* Insert after */ > + ext_debug("insert %d:%llu:[%d]%d %s before: " > + "nearest 0x%p\n" > + le32_to_cpu(newext->ee_block), > + ext4_ext_pblock(newext), > + ext4_ext_is_uninitialized(newext), > + ext4_ext_get_actual_len(newext), > + nearex); > + nearex++; > + } else { > + /* Insert before */ > + BUG_ON(newext->ee_block == nearex->ee_block); > + ext_debug("insert %d:%llu:[%d]%d %s after: " > + "nearest 0x%p\n" > le32_to_cpu(newext->ee_block), > ext4_ext_pblock(newext), > ext4_ext_is_uninitialized(newext), > ext4_ext_get_actual_len(newext), > - nearex, len, nearex + 1, nearex + 2); > - memmove(nearex + 2, nearex + 1, len); > + nearex); > + } > + len = EXT_LAST_EXTENT(eh) - nearex + 1; > + if (len > 0) { > + ext_debug("insert %d:%llu:[%d]%d: " > + "move %d extents from 0x%p to 0x%p\n", > + le32_to_cpu(newext->ee_block), > + ext4_ext_pblock(newext), > + ext4_ext_is_uninitialized(newext), > + ext4_ext_get_actual_len(newext), > + len, nearex, nearex + 1); > + memmove(nearex + 1, nearex, > + len * sizeof(struct ext4_extent)); > } > - path[depth].p_ext = nearex + 1; > - } else { > - BUG_ON(newext->ee_block == nearex->ee_block); > - len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); > - len = len < 0 ? 0 : len; > - ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " > - "move %d from 0x%p to 0x%p\n", > - le32_to_cpu(newext->ee_block), > - ext4_ext_pblock(newext), > - ext4_ext_is_uninitialized(newext), > - ext4_ext_get_actual_len(newext), > - nearex, len, nearex, nearex + 1); > - memmove(nearex + 1, nearex, len); > - path[depth].p_ext = nearex; > } > > le16_add_cpu(&eh->eh_entries, 1); > - nearex = path[depth].p_ext; > + path[depth].p_ext = nearex; > nearex->ee_block = newext->ee_block; > ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); > nearex->ee_len = newext->ee_len; > -- > 1.7.3.1 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Sep 26, 2011 at 05:23, Dmitry Monakhov <dmonakhov@openvz.org> wrote: > > On Sun, 25 Sep 2011 23:54:29 -0700, Eric Gouriou <egouriou@google.com> wrote: > > Seems patch was generated with "git diff -R" It was generated thusly: git format-patch -o upstream-patches/ -s 62d3ac6 I'm still fairly new at the business of submitting patches via mailing lists, so I'm very eager to hear how I could be doing a better job. In this particular case the patch, as sent, looks fine to me. git is not doing a fantastic job at showing the logic changes, however I can't really blame it, the resulting control flow is pretty similar to the pre-existing one, but hopefully cleaner and more efficient. Regards - Eric > > > ext4_ext_insert_extent() (respectively ext4_ext_insert_index()) > > was using EXT_MAX_EXTENT() (resp. EXT_MAX_INDEX()) to determine > > how many entries needed to be moved beyond the insertion point. > > In practice this means that (320 - I) * 24 bytes were memmove()'d > > when I is the insertion point, rather than (#entries - I) * 24 bytes. > > > > This patch uses EXT_LAST_EXTENT() (resp. EXT_LAST_INDEX()) instead > > to only move existing entries. The code flow is also simplified > > slightly to highlight similarities and reduce code duplication in > > the insertion logic. > > > > This patch reduces system CPU consumption by over 25% on a 4kB > > synchronous append DIO write workload when used with the > > pre-2.6.39 x86_64 memmove() implementation. With the much faster > > 2.6.39 memmove() implementation we still see a decrease in > > system CPU usage between 2% and 7%. > > > > Note that the ext_debug() output changes with this patch, splitting > > some log information between entries. Users of the ext_debug() output > > should note that the "move %d" units changed from reporting the number > > of bytes moved to reporting the number of entries moved. > > > > Signed-off-by: Eric Gouriou <egouriou@google.com> > > --- > > fs/ext4/extents.c | 85 ++++++++++++++++++++++++++--------------------------- > > 1 files changed, 42 insertions(+), 43 deletions(-) > > > > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c > > index 0a7dd85..30dd641 100644 > > --- a/fs/ext4/extents.c > > +++ b/fs/ext4/extents.c > > @@ -751,31 +751,25 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, > > return -EIO; > > } > > > > - len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; > > if (logical > le32_to_cpu(curp->p_idx->ei_block)) { > > /* insert after */ > > - if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { > > - len = (len - 1) * sizeof(struct ext4_extent_idx); > > - len = len < 0 ? 0 : len; > > - ext_debug("insert new index %d after: %llu. " > > - "move %d from 0x%p to 0x%p\n", > > - logical, ptr, len, > > - (curp->p_idx + 1), (curp->p_idx + 2)); > > - memmove(curp->p_idx + 2, curp->p_idx + 1, len); > > - } > > + ext_debug("insert new index %d after: %llu\n", logical, ptr); > > ix = curp->p_idx + 1; > > } else { > > /* insert before */ > > - len = len * sizeof(struct ext4_extent_idx); > > - len = len < 0 ? 0 : len; > > - ext_debug("insert new index %d before: %llu. " > > - "move %d from 0x%p to 0x%p\n", > > - logical, ptr, len, > > - curp->p_idx, (curp->p_idx + 1)); > > - memmove(curp->p_idx + 1, curp->p_idx, len); > > + ext_debug("insert new index %d before: %llu\n", logical, ptr); > > ix = curp->p_idx; > > } > > > > + len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; > > + BUG_ON(len < 0); > > + if (len > 0) { > > + ext_debug("insert new index %d: " > > + "move %d indices from 0x%p to 0x%p\n", > > + logical, len, ix, ix + 1); > > + memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); > > + } > > + > > ix->ei_block = cpu_to_le32(logical); > > ext4_idx_store_pblock(ix, ptr); > > le16_add_cpu(&curp->p_hdr->eh_entries, 1); > > @@ -1778,41 +1772,46 @@ has_space: > > ext4_ext_pblock(newext), > > ext4_ext_is_uninitialized(newext), > > ext4_ext_get_actual_len(newext)); > > - path[depth].p_ext = EXT_FIRST_EXTENT(eh); > > - } else if (le32_to_cpu(newext->ee_block) > > + nearex = EXT_FIRST_EXTENT(eh); > > + } else { > > + if (le32_to_cpu(newext->ee_block) > > > le32_to_cpu(nearex->ee_block)) { > > -/* BUG_ON(newext->ee_block == nearex->ee_block); */ > > - if (nearex != EXT_LAST_EXTENT(eh)) { > > - len = EXT_MAX_EXTENT(eh) - nearex; > > - len = (len - 1) * sizeof(struct ext4_extent); > > - len = len < 0 ? 0 : len; > > - ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " > > - "move %d from 0x%p to 0x%p\n", > > + /* Insert after */ > > + ext_debug("insert %d:%llu:[%d]%d %s before: " > > + "nearest 0x%p\n" > > + le32_to_cpu(newext->ee_block), > > + ext4_ext_pblock(newext), > > + ext4_ext_is_uninitialized(newext), > > + ext4_ext_get_actual_len(newext), > > + nearex); > > + nearex++; > > + } else { > > + /* Insert before */ > > + BUG_ON(newext->ee_block == nearex->ee_block); > > + ext_debug("insert %d:%llu:[%d]%d %s after: " > > + "nearest 0x%p\n" > > le32_to_cpu(newext->ee_block), > > ext4_ext_pblock(newext), > > ext4_ext_is_uninitialized(newext), > > ext4_ext_get_actual_len(newext), > > - nearex, len, nearex + 1, nearex + 2); > > - memmove(nearex + 2, nearex + 1, len); > > + nearex); > > + } > > + len = EXT_LAST_EXTENT(eh) - nearex + 1; > > + if (len > 0) { > > + ext_debug("insert %d:%llu:[%d]%d: " > > + "move %d extents from 0x%p to 0x%p\n", > > + le32_to_cpu(newext->ee_block), > > + ext4_ext_pblock(newext), > > + ext4_ext_is_uninitialized(newext), > > + ext4_ext_get_actual_len(newext), > > + len, nearex, nearex + 1); > > + memmove(nearex + 1, nearex, > > + len * sizeof(struct ext4_extent)); > > } > > - path[depth].p_ext = nearex + 1; > > - } else { > > - BUG_ON(newext->ee_block == nearex->ee_block); > > - len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); > > - len = len < 0 ? 0 : len; > > - ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " > > - "move %d from 0x%p to 0x%p\n", > > - le32_to_cpu(newext->ee_block), > > - ext4_ext_pblock(newext), > > - ext4_ext_is_uninitialized(newext), > > - ext4_ext_get_actual_len(newext), > > - nearex, len, nearex, nearex + 1); > > - memmove(nearex + 1, nearex, len); > > - path[depth].p_ext = nearex; > > } > > > > le16_add_cpu(&eh->eh_entries, 1); > > - nearex = path[depth].p_ext; > > + path[depth].p_ext = nearex; > > nearex->ee_block = newext->ee_block; > > ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); > > nearex->ee_len = newext->ee_len; > > -- > > 1.7.3.1 > > > > -- > > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in > > the body of a message to majordomo@vger.kernel.org > > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, Sep 25, 2011 at 11:54:29PM -0700, Eric Gouriou wrote: > ext4_ext_insert_extent() (respectively ext4_ext_insert_index()) > was using EXT_MAX_EXTENT() (resp. EXT_MAX_INDEX()) to determine > how many entries needed to be moved beyond the insertion point. > In practice this means that (320 - I) * 24 bytes were memmove()'d > when I is the insertion point, rather than (#entries - I) * 24 bytes. > > This patch uses EXT_LAST_EXTENT() (resp. EXT_LAST_INDEX()) instead > to only move existing entries. The code flow is also simplified > slightly to highlight similarities and reduce code duplication in > the insertion logic. > > This patch reduces system CPU consumption by over 25% on a 4kB > synchronous append DIO write workload when used with the > pre-2.6.39 x86_64 memmove() implementation. With the much faster > 2.6.39 memmove() implementation we still see a decrease in > system CPU usage between 2% and 7%. > > Note that the ext_debug() output changes with this patch, splitting > some log information between entries. Users of the ext_debug() output > should note that the "move %d" units changed from reporting the number > of bytes moved to reporting the number of entries moved. > > Signed-off-by: Eric Gouriou <egouriou@google.com> Applied, although the patch needed to be tweaked slightly to apply given recent changes to the surrounding code. I think I merged in the patch correctly, but I want to run some extended tests to make sure no problems turn up. - Ted -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c index 0a7dd85..30dd641 100644 --- a/fs/ext4/extents.c +++ b/fs/ext4/extents.c @@ -751,31 +751,25 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, return -EIO; } - len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; if (logical > le32_to_cpu(curp->p_idx->ei_block)) { /* insert after */ - if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { - len = (len - 1) * sizeof(struct ext4_extent_idx); - len = len < 0 ? 0 : len; - ext_debug("insert new index %d after: %llu. " - "move %d from 0x%p to 0x%p\n", - logical, ptr, len, - (curp->p_idx + 1), (curp->p_idx + 2)); - memmove(curp->p_idx + 2, curp->p_idx + 1, len); - } + ext_debug("insert new index %d after: %llu\n", logical, ptr); ix = curp->p_idx + 1; } else { /* insert before */ - len = len * sizeof(struct ext4_extent_idx); - len = len < 0 ? 0 : len; - ext_debug("insert new index %d before: %llu. " - "move %d from 0x%p to 0x%p\n", - logical, ptr, len, - curp->p_idx, (curp->p_idx + 1)); - memmove(curp->p_idx + 1, curp->p_idx, len); + ext_debug("insert new index %d before: %llu\n", logical, ptr); ix = curp->p_idx; } + len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; + BUG_ON(len < 0); + if (len > 0) { + ext_debug("insert new index %d: " + "move %d indices from 0x%p to 0x%p\n", + logical, len, ix, ix + 1); + memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); + } + ix->ei_block = cpu_to_le32(logical); ext4_idx_store_pblock(ix, ptr); le16_add_cpu(&curp->p_hdr->eh_entries, 1); @@ -1778,41 +1772,46 @@ has_space: ext4_ext_pblock(newext), ext4_ext_is_uninitialized(newext), ext4_ext_get_actual_len(newext)); - path[depth].p_ext = EXT_FIRST_EXTENT(eh); - } else if (le32_to_cpu(newext->ee_block) + nearex = EXT_FIRST_EXTENT(eh); + } else { + if (le32_to_cpu(newext->ee_block) > le32_to_cpu(nearex->ee_block)) { -/* BUG_ON(newext->ee_block == nearex->ee_block); */ - if (nearex != EXT_LAST_EXTENT(eh)) { - len = EXT_MAX_EXTENT(eh) - nearex; - len = (len - 1) * sizeof(struct ext4_extent); - len = len < 0 ? 0 : len; - ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, " - "move %d from 0x%p to 0x%p\n", + /* Insert after */ + ext_debug("insert %d:%llu:[%d]%d %s before: " + "nearest 0x%p\n" + le32_to_cpu(newext->ee_block), + ext4_ext_pblock(newext), + ext4_ext_is_uninitialized(newext), + ext4_ext_get_actual_len(newext), + nearex); + nearex++; + } else { + /* Insert before */ + BUG_ON(newext->ee_block == nearex->ee_block); + ext_debug("insert %d:%llu:[%d]%d %s after: " + "nearest 0x%p\n" le32_to_cpu(newext->ee_block), ext4_ext_pblock(newext), ext4_ext_is_uninitialized(newext), ext4_ext_get_actual_len(newext), - nearex, len, nearex + 1, nearex + 2); - memmove(nearex + 2, nearex + 1, len); + nearex); + } + len = EXT_LAST_EXTENT(eh) - nearex + 1; + if (len > 0) { + ext_debug("insert %d:%llu:[%d]%d: " + "move %d extents from 0x%p to 0x%p\n", + le32_to_cpu(newext->ee_block), + ext4_ext_pblock(newext), + ext4_ext_is_uninitialized(newext), + ext4_ext_get_actual_len(newext), + len, nearex, nearex + 1); + memmove(nearex + 1, nearex, + len * sizeof(struct ext4_extent)); } - path[depth].p_ext = nearex + 1; - } else { - BUG_ON(newext->ee_block == nearex->ee_block); - len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); - len = len < 0 ? 0 : len; - ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, " - "move %d from 0x%p to 0x%p\n", - le32_to_cpu(newext->ee_block), - ext4_ext_pblock(newext), - ext4_ext_is_uninitialized(newext), - ext4_ext_get_actual_len(newext), - nearex, len, nearex, nearex + 1); - memmove(nearex + 1, nearex, len); - path[depth].p_ext = nearex; } le16_add_cpu(&eh->eh_entries, 1); - nearex = path[depth].p_ext; + path[depth].p_ext = nearex; nearex->ee_block = newext->ee_block; ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); nearex->ee_len = newext->ee_len;
ext4_ext_insert_extent() (respectively ext4_ext_insert_index()) was using EXT_MAX_EXTENT() (resp. EXT_MAX_INDEX()) to determine how many entries needed to be moved beyond the insertion point. In practice this means that (320 - I) * 24 bytes were memmove()'d when I is the insertion point, rather than (#entries - I) * 24 bytes. This patch uses EXT_LAST_EXTENT() (resp. EXT_LAST_INDEX()) instead to only move existing entries. The code flow is also simplified slightly to highlight similarities and reduce code duplication in the insertion logic. This patch reduces system CPU consumption by over 25% on a 4kB synchronous append DIO write workload when used with the pre-2.6.39 x86_64 memmove() implementation. With the much faster 2.6.39 memmove() implementation we still see a decrease in system CPU usage between 2% and 7%. Note that the ext_debug() output changes with this patch, splitting some log information between entries. Users of the ext_debug() output should note that the "move %d" units changed from reporting the number of bytes moved to reporting the number of entries moved. Signed-off-by: Eric Gouriou <egouriou@google.com> --- fs/ext4/extents.c | 85 ++++++++++++++++++++++++++--------------------------- 1 files changed, 42 insertions(+), 43 deletions(-)