From cf9648cb71d6f1a463553d2fcd4c8137587dc060 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:42 +0800 Subject: scsi: vmw_pscsi: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Ewan D. Milne Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/vmw_pvscsi.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/scsi/vmw_pvscsi.c b/drivers/scsi/vmw_pvscsi.c index ecee4b3ff073..d71abd416eb4 100644 --- a/drivers/scsi/vmw_pvscsi.c +++ b/drivers/scsi/vmw_pvscsi.c @@ -335,7 +335,7 @@ static void pvscsi_create_sg(struct pvscsi_ctx *ctx, BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT); sge = &ctx->sgl->sge[0]; - for (i = 0; i < count; i++, sg++) { + for (i = 0; i < count; i++, sg = sg_next(sg)) { sge[i].addr = sg_dma_address(sg); sge[i].length = sg_dma_len(sg); sge[i].flags = 0; -- cgit v1.2.3 From c0d0d81ad34a41e2aa562bc59f22bed3e9e98080 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:43 +0800 Subject: scsi: advansys: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Ewan D. Milne Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/advansys.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/scsi/advansys.c b/drivers/scsi/advansys.c index d37584403c33..b87de8d3d844 100644 --- a/drivers/scsi/advansys.c +++ b/drivers/scsi/advansys.c @@ -7714,7 +7714,7 @@ adv_get_sglist(struct asc_board *boardp, adv_req_t *reqp, sg_block->sg_ptr = 0L; /* Last ADV_SG_BLOCK in list. */ return ADV_SUCCESS; } - slp++; + slp = sg_next(slp); } sg_block->sg_cnt = NO_OF_SG_PER_BLOCK; prev_sg_block = sg_block; -- cgit v1.2.3 From 46e8e475a160be5e31e99171b7c0c8a21eb4d6ad Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:44 +0800 Subject: scsi: lpfc: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed by: Ewan D. Milne Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/lpfc/lpfc_nvmet.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/drivers/scsi/lpfc/lpfc_nvmet.c b/drivers/scsi/lpfc/lpfc_nvmet.c index d74bfd264495..877d1fb92754 100644 --- a/drivers/scsi/lpfc/lpfc_nvmet.c +++ b/drivers/scsi/lpfc/lpfc_nvmet.c @@ -2662,8 +2662,7 @@ lpfc_nvmet_prep_fcp_wqe(struct lpfc_hba *phba, nvmewqe->drvrTimeout = (phba->fc_ratov * 3) + LPFC_DRVR_TIMEOUT; nvmewqe->context1 = ndlp; - for (i = 0; i < rsp->sg_cnt; i++) { - sgel = &rsp->sg[i]; + for_each_sg(rsp->sg, sgel, rsp->sg_cnt, i) { physaddr = sg_dma_address(sgel); cnt = sg_dma_len(sgel); sgl->addr_hi = putPaddrHigh(physaddr); -- cgit v1.2.3 From 3c1a30df6d9c21c3235b6af5f25da2765b19d05b Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:45 +0800 Subject: scsi: mvumi: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message and folded in build fix reported by zeroday] Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Reviewed-by: Ewan D. Milne Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/mvumi.c | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) diff --git a/drivers/scsi/mvumi.c b/drivers/scsi/mvumi.c index a5410615edac..53f3563aca22 100644 --- a/drivers/scsi/mvumi.c +++ b/drivers/scsi/mvumi.c @@ -211,23 +211,22 @@ static int mvumi_make_sgl(struct mvumi_hba *mhba, struct scsi_cmnd *scmd, unsigned int sgnum = scsi_sg_count(scmd); dma_addr_t busaddr; - sg = scsi_sglist(scmd); - *sg_count = dma_map_sg(&mhba->pdev->dev, sg, sgnum, + *sg_count = dma_map_sg(&mhba->pdev->dev, scsi_sglist(scmd), sgnum, scmd->sc_data_direction); if (*sg_count > mhba->max_sge) { dev_err(&mhba->pdev->dev, "sg count[0x%x] is bigger than max sg[0x%x].\n", *sg_count, mhba->max_sge); - dma_unmap_sg(&mhba->pdev->dev, sg, sgnum, + dma_unmap_sg(&mhba->pdev->dev, scsi_sglist(scmd), sgnum, scmd->sc_data_direction); return -1; } - for (i = 0; i < *sg_count; i++) { - busaddr = sg_dma_address(&sg[i]); + scsi_for_each_sg(scmd, sg, *sg_count, i) { + busaddr = sg_dma_address(sg); m_sg->baseaddr_l = cpu_to_le32(lower_32_bits(busaddr)); m_sg->baseaddr_h = cpu_to_le32(upper_32_bits(busaddr)); m_sg->flags = 0; - sgd_setsz(mhba, m_sg, cpu_to_le32(sg_dma_len(&sg[i]))); + sgd_setsz(mhba, m_sg, cpu_to_le32(sg_dma_len(sg))); if ((i + 1) == *sg_count) m_sg->flags |= 1U << mhba->eot_flag; -- cgit v1.2.3 From c71ae886d1321e74f524c7c023933cf87768915d Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:46 +0800 Subject: scsi: ipr: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/ipr.c | 29 ++++++++++++++++------------- 1 file changed, 16 insertions(+), 13 deletions(-) diff --git a/drivers/scsi/ipr.c b/drivers/scsi/ipr.c index 6d053e220153..bf17540affbc 100644 --- a/drivers/scsi/ipr.c +++ b/drivers/scsi/ipr.c @@ -3915,22 +3915,23 @@ static int ipr_copy_ucode_buffer(struct ipr_sglist *sglist, u8 *buffer, u32 len) { int bsize_elem, i, result = 0; - struct scatterlist *scatterlist; + struct scatterlist *sg; void *kaddr; /* Determine the actual number of bytes per element */ bsize_elem = PAGE_SIZE * (1 << sglist->order); - scatterlist = sglist->scatterlist; + sg = sglist->scatterlist; - for (i = 0; i < (len / bsize_elem); i++, buffer += bsize_elem) { - struct page *page = sg_page(&scatterlist[i]); + for (i = 0; i < (len / bsize_elem); i++, sg = sg_next(sg), + buffer += bsize_elem) { + struct page *page = sg_page(sg); kaddr = kmap(page); memcpy(kaddr, buffer, bsize_elem); kunmap(page); - scatterlist[i].length = bsize_elem; + sg->length = bsize_elem; if (result != 0) { ipr_trace; @@ -3939,13 +3940,13 @@ static int ipr_copy_ucode_buffer(struct ipr_sglist *sglist, } if (len % bsize_elem) { - struct page *page = sg_page(&scatterlist[i]); + struct page *page = sg_page(sg); kaddr = kmap(page); memcpy(kaddr, buffer, len % bsize_elem); kunmap(page); - scatterlist[i].length = len % bsize_elem; + sg->length = len % bsize_elem; } sglist->buffer_len = len; @@ -3966,6 +3967,7 @@ static void ipr_build_ucode_ioadl64(struct ipr_cmnd *ipr_cmd, struct ipr_ioarcb *ioarcb = &ipr_cmd->ioarcb; struct ipr_ioadl64_desc *ioadl64 = ipr_cmd->i.ioadl64; struct scatterlist *scatterlist = sglist->scatterlist; + struct scatterlist *sg; int i; ipr_cmd->dma_use_sg = sglist->num_dma_sg; @@ -3974,10 +3976,10 @@ static void ipr_build_ucode_ioadl64(struct ipr_cmnd *ipr_cmd, ioarcb->ioadl_len = cpu_to_be32(sizeof(struct ipr_ioadl64_desc) * ipr_cmd->dma_use_sg); - for (i = 0; i < ipr_cmd->dma_use_sg; i++) { + for_each_sg(scatterlist, sg, ipr_cmd->dma_use_sg, i) { ioadl64[i].flags = cpu_to_be32(IPR_IOADL_FLAGS_WRITE); - ioadl64[i].data_len = cpu_to_be32(sg_dma_len(&scatterlist[i])); - ioadl64[i].address = cpu_to_be64(sg_dma_address(&scatterlist[i])); + ioadl64[i].data_len = cpu_to_be32(sg_dma_len(sg)); + ioadl64[i].address = cpu_to_be64(sg_dma_address(sg)); } ioadl64[i-1].flags |= cpu_to_be32(IPR_IOADL_FLAGS_LAST); @@ -3997,6 +3999,7 @@ static void ipr_build_ucode_ioadl(struct ipr_cmnd *ipr_cmd, struct ipr_ioarcb *ioarcb = &ipr_cmd->ioarcb; struct ipr_ioadl_desc *ioadl = ipr_cmd->i.ioadl; struct scatterlist *scatterlist = sglist->scatterlist; + struct scatterlist *sg; int i; ipr_cmd->dma_use_sg = sglist->num_dma_sg; @@ -4006,11 +4009,11 @@ static void ipr_build_ucode_ioadl(struct ipr_cmnd *ipr_cmd, ioarcb->ioadl_len = cpu_to_be32(sizeof(struct ipr_ioadl_desc) * ipr_cmd->dma_use_sg); - for (i = 0; i < ipr_cmd->dma_use_sg; i++) { + for_each_sg(scatterlist, sg, ipr_cmd->dma_use_sg, i) { ioadl[i].flags_and_data_len = - cpu_to_be32(IPR_IOADL_FLAGS_WRITE | sg_dma_len(&scatterlist[i])); + cpu_to_be32(IPR_IOADL_FLAGS_WRITE | sg_dma_len(sg)); ioadl[i].address = - cpu_to_be32(sg_dma_address(&scatterlist[i])); + cpu_to_be32(sg_dma_address(sg)); } ioadl[i-1].flags_and_data_len |= -- cgit v1.2.3 From 74eb7446eda5e348dbb90981e91e9caf54f2fde4 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:47 +0800 Subject: scsi: pmcraid: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/pmcraid.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/drivers/scsi/pmcraid.c b/drivers/scsi/pmcraid.c index e338d7a4f571..286cac59cb5f 100644 --- a/drivers/scsi/pmcraid.c +++ b/drivers/scsi/pmcraid.c @@ -3270,7 +3270,7 @@ static int pmcraid_copy_sglist( int direction ) { - struct scatterlist *scatterlist; + struct scatterlist *sg; void *kaddr; int bsize_elem; int i; @@ -3279,10 +3279,10 @@ static int pmcraid_copy_sglist( /* Determine the actual number of bytes per element */ bsize_elem = PAGE_SIZE * (1 << sglist->order); - scatterlist = sglist->scatterlist; + sg = sglist->scatterlist; - for (i = 0; i < (len / bsize_elem); i++, buffer += bsize_elem) { - struct page *page = sg_page(&scatterlist[i]); + for (i = 0; i < (len / bsize_elem); i++, sg = sg_next(sg), buffer += bsize_elem) { + struct page *page = sg_page(sg); kaddr = kmap(page); if (direction == DMA_TO_DEVICE) @@ -3297,11 +3297,11 @@ static int pmcraid_copy_sglist( return -EFAULT; } - scatterlist[i].length = bsize_elem; + sg->length = bsize_elem; } if (len % bsize_elem) { - struct page *page = sg_page(&scatterlist[i]); + struct page *page = sg_page(sg); kaddr = kmap(page); @@ -3312,7 +3312,7 @@ static int pmcraid_copy_sglist( kunmap(page); - scatterlist[i].length = len % bsize_elem; + sg->length = len % bsize_elem; } if (rc) { -- cgit v1.2.3 From 1194b5ce57d27c2ad49ed26f8cec98757c7c23ec Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:48 +0800 Subject: scsi: usb: image: microtek: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Cc: Oliver Neukum Cc: Greg Kroah-Hartman Cc: linux-usb@vger.kernel.org Reviewed-by: Bart Van Assche Reviewed-by: Christoph Hellwig Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/usb/image/microtek.c | 20 ++++++++------------ drivers/usb/image/microtek.h | 2 +- 2 files changed, 9 insertions(+), 13 deletions(-) diff --git a/drivers/usb/image/microtek.c b/drivers/usb/image/microtek.c index 607be1f4fe27..0a57c2cc8e5a 100644 --- a/drivers/usb/image/microtek.c +++ b/drivers/usb/image/microtek.c @@ -488,7 +488,6 @@ static void mts_command_done( struct urb *transfer ) static void mts_do_sg (struct urb* transfer) { - struct scatterlist * sg; int status = transfer->status; MTS_INT_INIT(); @@ -500,13 +499,12 @@ static void mts_do_sg (struct urb* transfer) mts_transfer_cleanup(transfer); } - sg = scsi_sglist(context->srb); - context->fragment++; + context->curr_sg = sg_next(context->curr_sg); mts_int_submit_urb(transfer, context->data_pipe, - sg_virt(&sg[context->fragment]), - sg[context->fragment].length, - context->fragment + 1 == scsi_sg_count(context->srb) ? + sg_virt(context->curr_sg), + context->curr_sg->length, + sg_is_last(context->curr_sg) ? mts_data_done : mts_do_sg); } @@ -526,22 +524,20 @@ static void mts_build_transfer_context(struct scsi_cmnd *srb, struct mts_desc* desc) { int pipe; - struct scatterlist * sg; - + MTS_DEBUG_GOT_HERE(); desc->context.instance = desc; desc->context.srb = srb; - desc->context.fragment = 0; if (!scsi_bufflen(srb)) { desc->context.data = NULL; desc->context.data_length = 0; return; } else { - sg = scsi_sglist(srb); - desc->context.data = sg_virt(&sg[0]); - desc->context.data_length = sg[0].length; + desc->context.curr_sg = scsi_sglist(srb); + desc->context.data = sg_virt(desc->context.curr_sg); + desc->context.data_length = desc->context.curr_sg->length; } diff --git a/drivers/usb/image/microtek.h b/drivers/usb/image/microtek.h index 66685e59241a..7bd5f4639c4a 100644 --- a/drivers/usb/image/microtek.h +++ b/drivers/usb/image/microtek.h @@ -21,7 +21,7 @@ struct mts_transfer_context void *data; unsigned data_length; int data_pipe; - int fragment; + struct scatterlist *curr_sg; u8 *scsi_status; /* status returned from ep_response after command completion */ }; -- cgit v1.2.3 From da5567369fb686eb050da15364f11ef561f9a03e Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:49 +0800 Subject: scsi: staging: unisys: visorhba: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Cc: devel@driverdev.osuosl.org Cc: Greg Kroah-Hartman Acked-by: Greg Kroah-Hartman Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/staging/unisys/visorhba/visorhba_main.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/drivers/staging/unisys/visorhba/visorhba_main.c b/drivers/staging/unisys/visorhba/visorhba_main.c index 2dad36a05518..dd979ee4dcf1 100644 --- a/drivers/staging/unisys/visorhba/visorhba_main.c +++ b/drivers/staging/unisys/visorhba/visorhba_main.c @@ -871,12 +871,11 @@ static void do_scsi_nolinuxstat(struct uiscmdrsp *cmdrsp, return; } - sg = scsi_sglist(scsicmd); - for (i = 0; i < scsi_sg_count(scsicmd); i++) { - this_page_orig = kmap_atomic(sg_page(sg + i)); + scsi_for_each_sg(scsicmd, sg, scsi_sg_count(scsicmd), i) { + this_page_orig = kmap_atomic(sg_page(sg)); this_page = (void *)((unsigned long)this_page_orig | - sg[i].offset); - memcpy(this_page, buf + bufind, sg[i].length); + sg->offset); + memcpy(this_page, buf + bufind, sg->length); kunmap_atomic(this_page_orig); } kfree(buf); -- cgit v1.2.3 From 013be03840c2ba3a2717f9cee457f01fdc4d8436 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:51 +0800 Subject: scsi: s390: zfcp_fc: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Cc: Steffen Maier Cc: Benjamin Block Cc: Martin Schwidefsky Cc: Heiko Carstens Cc: linux-s390@vger.kernel.org Acked-by: Benjamin Block Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/s390/scsi/zfcp_fc.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/s390/scsi/zfcp_fc.c b/drivers/s390/scsi/zfcp_fc.c index 33eddb02ee30..b018b61bd168 100644 --- a/drivers/s390/scsi/zfcp_fc.c +++ b/drivers/s390/scsi/zfcp_fc.c @@ -620,7 +620,7 @@ static void zfcp_fc_sg_free_table(struct scatterlist *sg, int count) { int i; - for (i = 0; i < count; i++, sg++) + for (i = 0; i < count; i++, sg = sg_next(sg)) if (sg) free_page((unsigned long) sg_virt(sg)); else @@ -641,7 +641,7 @@ static int zfcp_fc_sg_setup_table(struct scatterlist *sg, int count) int i; sg_init_table(sg, count); - for (i = 0; i < count; i++, sg++) { + for (i = 0; i < count; i++, sg = sg_next(sg)) { addr = (void *) get_zeroed_page(GFP_KERNEL); if (!addr) { zfcp_fc_sg_free_table(sg, i); -- cgit v1.2.3 From a7a253ba6c26490fb132ee682661c03a6c454fd5 Mon Sep 17 00:00:00 2001 From: Finn Thain Date: Tue, 18 Jun 2019 09:37:52 +0800 Subject: scsi: aha152x: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. Finn added the change to replace SCp.buffers_residual with sg_is_last() for fixing updating it, and the similar change has been applied on NCR5380.c [mkp: clarified commit message] Signed-off-by: Finn Thain Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/aha152x.c | 46 +++++++++++++++++++++++----------------------- 1 file changed, 23 insertions(+), 23 deletions(-) diff --git a/drivers/scsi/aha152x.c b/drivers/scsi/aha152x.c index 97872838b983..f07f3fa9b58d 100644 --- a/drivers/scsi/aha152x.c +++ b/drivers/scsi/aha152x.c @@ -948,7 +948,6 @@ static int aha152x_internal_queue(struct scsi_cmnd *SCpnt, SCp.ptr : buffer pointer SCp.this_residual : buffer length SCp.buffer : next buffer - SCp.buffers_residual : left buffers in list SCp.phase : current state of the command */ if ((phase & resetting) || !scsi_sglist(SCpnt)) { @@ -956,13 +955,11 @@ static int aha152x_internal_queue(struct scsi_cmnd *SCpnt, SCpnt->SCp.this_residual = 0; scsi_set_resid(SCpnt, 0); SCpnt->SCp.buffer = NULL; - SCpnt->SCp.buffers_residual = 0; } else { scsi_set_resid(SCpnt, scsi_bufflen(SCpnt)); SCpnt->SCp.buffer = scsi_sglist(SCpnt); SCpnt->SCp.ptr = SG_ADDRESS(SCpnt->SCp.buffer); SCpnt->SCp.this_residual = SCpnt->SCp.buffer->length; - SCpnt->SCp.buffers_residual = scsi_sg_count(SCpnt) - 1; } DO_LOCK(flags); @@ -2030,10 +2027,9 @@ static void datai_run(struct Scsi_Host *shpnt) } if (CURRENT_SC->SCp.this_residual == 0 && - CURRENT_SC->SCp.buffers_residual > 0) { + !sg_is_last(CURRENT_SC->SCp.buffer)) { /* advance to next buffer */ - CURRENT_SC->SCp.buffers_residual--; - CURRENT_SC->SCp.buffer++; + CURRENT_SC->SCp.buffer = sg_next(CURRENT_SC->SCp.buffer); CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer); CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length; } @@ -2136,10 +2132,10 @@ static void datao_run(struct Scsi_Host *shpnt) CMD_INC_RESID(CURRENT_SC, -2 * data_count); } - if(CURRENT_SC->SCp.this_residual==0 && CURRENT_SC->SCp.buffers_residual>0) { + if (CURRENT_SC->SCp.this_residual == 0 && + !sg_is_last(CURRENT_SC->SCp.buffer)) { /* advance to next buffer */ - CURRENT_SC->SCp.buffers_residual--; - CURRENT_SC->SCp.buffer++; + CURRENT_SC->SCp.buffer = sg_next(CURRENT_SC->SCp.buffer); CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer); CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length; } @@ -2158,22 +2154,26 @@ static void datao_run(struct Scsi_Host *shpnt) static void datao_end(struct Scsi_Host *shpnt) { if(TESTLO(DMASTAT, DFIFOEMP)) { - int data_count = (DATA_LEN - scsi_get_resid(CURRENT_SC)) - - GETSTCNT(); + u32 datao_cnt = GETSTCNT(); + int datao_out = DATA_LEN - scsi_get_resid(CURRENT_SC); + int done; + struct scatterlist *sg = scsi_sglist(CURRENT_SC); - CMD_INC_RESID(CURRENT_SC, data_count); + CMD_INC_RESID(CURRENT_SC, datao_out - datao_cnt); - data_count -= CURRENT_SC->SCp.ptr - - SG_ADDRESS(CURRENT_SC->SCp.buffer); - while(data_count>0) { - CURRENT_SC->SCp.buffer--; - CURRENT_SC->SCp.buffers_residual++; - data_count -= CURRENT_SC->SCp.buffer->length; + done = scsi_bufflen(CURRENT_SC) - scsi_get_resid(CURRENT_SC); + /* Locate the first SG entry not yet sent */ + while (done > 0 && !sg_is_last(sg)) { + if (done < sg->length) + break; + done -= sg->length; + sg = sg_next(sg); } - CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer) - - data_count; - CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length + - data_count; + + CURRENT_SC->SCp.buffer = sg; + CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer) + done; + CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length - + done; } SETPORT(SXFRCTL0, CH1|CLRCH1|CLRSTCNT); @@ -2501,7 +2501,7 @@ static void get_command(struct seq_file *m, struct scsi_cmnd * ptr) seq_printf(m, "); resid=%d; residual=%d; buffers=%d; phase |", scsi_get_resid(ptr), ptr->SCp.this_residual, - ptr->SCp.buffers_residual); + sg_nents(ptr->SCp.buffer) - 1); if (ptr->SCp.phase & not_issued) seq_puts(m, "not issued|"); -- cgit v1.2.3 From 79da19b48fc188f6b5186e9ed605e10e5bef3ad6 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:53 +0800 Subject: scsi: imm: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/imm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/scsi/imm.c b/drivers/scsi/imm.c index 64ae418d29f3..56d29f157749 100644 --- a/drivers/scsi/imm.c +++ b/drivers/scsi/imm.c @@ -686,7 +686,7 @@ static int imm_completion(struct scsi_cmnd *cmd) if (cmd->SCp.buffer && !cmd->SCp.this_residual) { /* if scatter/gather, advance to the next segment */ if (cmd->SCp.buffers_residual--) { - cmd->SCp.buffer++; + cmd->SCp.buffer = sg_next(cmd->SCp.buffer); cmd->SCp.this_residual = cmd->SCp.buffer->length; cmd->SCp.ptr = sg_virt(cmd->SCp.buffer); -- cgit v1.2.3 From 1b3a4640106603ce6b3ad261ef4d20ea60e9790b Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:54 +0800 Subject: scsi: pcmcia: nsp_cs: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/pcmcia/nsp_cs.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/scsi/pcmcia/nsp_cs.c b/drivers/scsi/pcmcia/nsp_cs.c index a81748e6e8fb..97416e1dcc5b 100644 --- a/drivers/scsi/pcmcia/nsp_cs.c +++ b/drivers/scsi/pcmcia/nsp_cs.c @@ -789,7 +789,7 @@ static void nsp_pio_read(struct scsi_cmnd *SCpnt) SCpnt->SCp.buffers_residual != 0 ) { //nsp_dbg(NSP_DEBUG_DATA_IO, "scatterlist next timeout=%d", time_out); SCpnt->SCp.buffers_residual--; - SCpnt->SCp.buffer++; + SCpnt->SCp.buffer = sg_next(SCpnt->SCp.buffer); SCpnt->SCp.ptr = BUFFER_ADDR; SCpnt->SCp.this_residual = SCpnt->SCp.buffer->length; time_out = 1000; @@ -887,7 +887,7 @@ static void nsp_pio_write(struct scsi_cmnd *SCpnt) SCpnt->SCp.buffers_residual != 0 ) { //nsp_dbg(NSP_DEBUG_DATA_IO, "scatterlist next"); SCpnt->SCp.buffers_residual--; - SCpnt->SCp.buffer++; + SCpnt->SCp.buffer = sg_next(SCpnt->SCp.buffer); SCpnt->SCp.ptr = BUFFER_ADDR; SCpnt->SCp.this_residual = SCpnt->SCp.buffer->length; time_out = 1000; -- cgit v1.2.3 From 57ef4e510939e305ad905cc54b8031401ff5450e Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:55 +0800 Subject: scsi: ppa: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/ppa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/scsi/ppa.c b/drivers/scsi/ppa.c index 35213082e933..a406cc825426 100644 --- a/drivers/scsi/ppa.c +++ b/drivers/scsi/ppa.c @@ -590,7 +590,7 @@ static int ppa_completion(struct scsi_cmnd *cmd) if (cmd->SCp.buffer && !cmd->SCp.this_residual) { /* if scatter/gather, advance to the next segment */ if (cmd->SCp.buffers_residual--) { - cmd->SCp.buffer++; + cmd->SCp.buffer = sg_next(cmd->SCp.buffer); cmd->SCp.this_residual = cmd->SCp.buffer->length; cmd->SCp.ptr = sg_virt(cmd->SCp.buffer); -- cgit v1.2.3 From c3c0fd9b108f6360ceaade790571be097df1f3ef Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Tue, 18 Jun 2019 09:37:56 +0800 Subject: scsi: wd33c93: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/wd33c93.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/scsi/wd33c93.c b/drivers/scsi/wd33c93.c index 74be04f2357c..ae5935c0a149 100644 --- a/drivers/scsi/wd33c93.c +++ b/drivers/scsi/wd33c93.c @@ -744,7 +744,7 @@ transfer_bytes(const wd33c93_regs regs, struct scsi_cmnd *cmd, * source or destination for THIS transfer. */ if (!cmd->SCp.this_residual && cmd->SCp.buffers_residual) { - ++cmd->SCp.buffer; + cmd->SCp.buffer = sg_next(cmd->SCp.buffer); --cmd->SCp.buffers_residual; cmd->SCp.this_residual = cmd->SCp.buffer->length; cmd->SCp.ptr = sg_virt(cmd->SCp.buffer); -- cgit v1.2.3 From 0e9fdd2b315c0fde535336ea09499a927e415566 Mon Sep 17 00:00:00 2001 From: Finn Thain Date: Tue, 18 Jun 2019 09:37:57 +0800 Subject: scsi: NCR5380: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Cc: Michael Schmitz Reviewed-by: Michael Schmitz Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Signed-off-by: Finn Thain Signed-off-by: Martin K. Petersen --- drivers/scsi/NCR5380.c | 41 ++++++++++++++++++----------------------- 1 file changed, 18 insertions(+), 23 deletions(-) diff --git a/drivers/scsi/NCR5380.c b/drivers/scsi/NCR5380.c index fe0535affc14..4ef44fafe6ca 100644 --- a/drivers/scsi/NCR5380.c +++ b/drivers/scsi/NCR5380.c @@ -149,12 +149,10 @@ static inline void initialize_SCp(struct scsi_cmnd *cmd) if (scsi_bufflen(cmd)) { cmd->SCp.buffer = scsi_sglist(cmd); - cmd->SCp.buffers_residual = scsi_sg_count(cmd) - 1; cmd->SCp.ptr = sg_virt(cmd->SCp.buffer); cmd->SCp.this_residual = cmd->SCp.buffer->length; } else { cmd->SCp.buffer = NULL; - cmd->SCp.buffers_residual = 0; cmd->SCp.ptr = NULL; cmd->SCp.this_residual = 0; } @@ -163,6 +161,17 @@ static inline void initialize_SCp(struct scsi_cmnd *cmd) cmd->SCp.Message = 0; } +static inline void advance_sg_buffer(struct scsi_cmnd *cmd) +{ + struct scatterlist *s = cmd->SCp.buffer; + + if (!cmd->SCp.this_residual && s && !sg_is_last(s)) { + cmd->SCp.buffer = sg_next(s); + cmd->SCp.ptr = sg_virt(cmd->SCp.buffer); + cmd->SCp.this_residual = cmd->SCp.buffer->length; + } +} + /** * NCR5380_poll_politely2 - wait for two chip register values * @hostdata: host private data @@ -1672,12 +1681,7 @@ static void NCR5380_information_transfer(struct Scsi_Host *instance) sun3_dma_setup_done != cmd) { int count; - if (!cmd->SCp.this_residual && cmd->SCp.buffers_residual) { - ++cmd->SCp.buffer; - --cmd->SCp.buffers_residual; - cmd->SCp.this_residual = cmd->SCp.buffer->length; - cmd->SCp.ptr = sg_virt(cmd->SCp.buffer); - } + advance_sg_buffer(cmd); count = sun3scsi_dma_xfer_len(hostdata, cmd); @@ -1727,15 +1731,11 @@ static void NCR5380_information_transfer(struct Scsi_Host *instance) * scatter-gather list, move onto the next one. */ - if (!cmd->SCp.this_residual && cmd->SCp.buffers_residual) { - ++cmd->SCp.buffer; - --cmd->SCp.buffers_residual; - cmd->SCp.this_residual = cmd->SCp.buffer->length; - cmd->SCp.ptr = sg_virt(cmd->SCp.buffer); - dsprintk(NDEBUG_INFORMATION, instance, "%d bytes and %d buffers left\n", - cmd->SCp.this_residual, - cmd->SCp.buffers_residual); - } + advance_sg_buffer(cmd); + dsprintk(NDEBUG_INFORMATION, instance, + "this residual %d, sg ents %d\n", + cmd->SCp.this_residual, + sg_nents(cmd->SCp.buffer)); /* * The preferred transfer method is going to be @@ -2136,12 +2136,7 @@ static void NCR5380_reselect(struct Scsi_Host *instance) if (sun3_dma_setup_done != tmp) { int count; - if (!tmp->SCp.this_residual && tmp->SCp.buffers_residual) { - ++tmp->SCp.buffer; - --tmp->SCp.buffers_residual; - tmp->SCp.this_residual = tmp->SCp.buffer->length; - tmp->SCp.ptr = sg_virt(tmp->SCp.buffer); - } + advance_sg_buffer(tmp); count = sun3scsi_dma_xfer_len(hostdata, tmp); -- cgit v1.2.3 From ee5a1dbfec57cc1ffdedf2bd767c84d5e0498ed8 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Thu, 6 Jun 2019 16:34:10 +0800 Subject: scsi: esp: use sg helper to iterate over scatterlist Unlike the legacy I/O path, scsi-mq preallocates a large array to hold the scatterlist for each request. This static allocation can consume substantial amounts of memory on modern controllers which support a large number of concurrently outstanding requests. To facilitate a switch to a smaller static allocation combined with a dynamic allocation for requests that need it, we need to make sure all SCSI drivers handle chained scatterlists correctly. Convert remaining drivers that directly dereference the scatterlist array to using the iterator functions. [mkp: clarified commit message] Cc: Christoph Hellwig Cc: Bart Van Assche Cc: Ewan D. Milne Cc: Hannes Reinecke Cc: Finn Thain Cc: Guenter Roeck Reported-by: Guenter Roeck Tested-by: Guenter Roeck Reviewed-by: Finn Thain Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/esp_scsi.c | 20 +++++++++++++------- drivers/scsi/esp_scsi.h | 2 ++ 2 files changed, 15 insertions(+), 7 deletions(-) diff --git a/drivers/scsi/esp_scsi.c b/drivers/scsi/esp_scsi.c index 76fd02ccbf49..1e3ea9dc74d2 100644 --- a/drivers/scsi/esp_scsi.c +++ b/drivers/scsi/esp_scsi.c @@ -370,6 +370,7 @@ static void esp_map_dma(struct esp *esp, struct scsi_cmnd *cmd) struct esp_cmd_priv *spriv = ESP_CMD_PRIV(cmd); struct scatterlist *sg = scsi_sglist(cmd); int total = 0, i; + struct scatterlist *s; if (cmd->sc_data_direction == DMA_NONE) return; @@ -380,16 +381,18 @@ static void esp_map_dma(struct esp *esp, struct scsi_cmnd *cmd) * a dma address, so perform an identity mapping. */ spriv->num_sg = scsi_sg_count(cmd); - for (i = 0; i < spriv->num_sg; i++) { - sg[i].dma_address = (uintptr_t)sg_virt(&sg[i]); - total += sg_dma_len(&sg[i]); + + scsi_for_each_sg(cmd, s, spriv->num_sg, i) { + s->dma_address = (uintptr_t)sg_virt(s); + total += sg_dma_len(s); } } else { spriv->num_sg = scsi_dma_map(cmd); - for (i = 0; i < spriv->num_sg; i++) - total += sg_dma_len(&sg[i]); + scsi_for_each_sg(cmd, s, spriv->num_sg, i) + total += sg_dma_len(s); } spriv->cur_residue = sg_dma_len(sg); + spriv->prv_sg = NULL; spriv->cur_sg = sg; spriv->tot_residue = total; } @@ -443,7 +446,8 @@ static void esp_advance_dma(struct esp *esp, struct esp_cmd_entry *ent, p->tot_residue = 0; } if (!p->cur_residue && p->tot_residue) { - p->cur_sg++; + p->prv_sg = p->cur_sg; + p->cur_sg = sg_next(p->cur_sg); p->cur_residue = sg_dma_len(p->cur_sg); } } @@ -464,6 +468,7 @@ static void esp_save_pointers(struct esp *esp, struct esp_cmd_entry *ent) return; } ent->saved_cur_residue = spriv->cur_residue; + ent->saved_prv_sg = spriv->prv_sg; ent->saved_cur_sg = spriv->cur_sg; ent->saved_tot_residue = spriv->tot_residue; } @@ -478,6 +483,7 @@ static void esp_restore_pointers(struct esp *esp, struct esp_cmd_entry *ent) return; } spriv->cur_residue = ent->saved_cur_residue; + spriv->prv_sg = ent->saved_prv_sg; spriv->cur_sg = ent->saved_cur_sg; spriv->tot_residue = ent->saved_tot_residue; } @@ -1646,7 +1652,7 @@ static int esp_msgin_process(struct esp *esp) spriv = ESP_CMD_PRIV(ent->cmd); if (spriv->cur_residue == sg_dma_len(spriv->cur_sg)) { - spriv->cur_sg--; + spriv->cur_sg = spriv->prv_sg; spriv->cur_residue = 1; } else spriv->cur_residue++; diff --git a/drivers/scsi/esp_scsi.h b/drivers/scsi/esp_scsi.h index aa87a6b72dcc..91b32f2a1a1b 100644 --- a/drivers/scsi/esp_scsi.h +++ b/drivers/scsi/esp_scsi.h @@ -251,6 +251,7 @@ struct esp_cmd_priv { int num_sg; int cur_residue; + struct scatterlist *prv_sg; struct scatterlist *cur_sg; int tot_residue; }; @@ -273,6 +274,7 @@ struct esp_cmd_entry { struct scsi_cmnd *cmd; unsigned int saved_cur_residue; + struct scatterlist *saved_prv_sg; struct scatterlist *saved_cur_sg; unsigned int saved_tot_residue; -- cgit v1.2.3 From 4635873c561ac57b66adfcc2487c38106b1c916c Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Sun, 28 Apr 2019 15:39:30 +0800 Subject: scsi: lib/sg_pool.c: improve APIs for allocating sg pool sg_alloc_table_chained() currently allows the caller to provide one preallocated SGL and returns if the requested number isn't bigger than size of that SGL. This is used to inline an SGL for an IO request. However, scattergather code only allows that size of the 1st preallocated SGL to be SG_CHUNK_SIZE(128). This means a substantial amount of memory (4KB) is claimed for the SGL for each IO request. If the I/O is small, it would be prudent to allocate a smaller SGL. Introduce an extra parameter to sg_alloc_table_chained() and sg_free_table_chained() for specifying size of the preallocated SGL. Both __sg_free_table() and __sg_alloc_table() assume that each SGL has the same size except for the last one. Change the code to allow both functions to accept a variable size for the 1st preallocated SGL. [mkp: attempted to clarify commit desc] Cc: Christoph Hellwig Cc: Bart Van Assche Cc: Ewan D. Milne Cc: Hannes Reinecke Cc: Sagi Grimberg Cc: Chuck Lever Cc: netdev@vger.kernel.org Cc: linux-nvme@lists.infradead.org Suggested-by: Christoph Hellwig Signed-off-by: Ming Lei Reviewed-by: Christoph Hellwig Signed-off-by: Martin K. Petersen --- drivers/nvme/host/fc.c | 7 ++++--- drivers/nvme/host/rdma.c | 7 ++++--- drivers/nvme/target/loop.c | 4 ++-- drivers/scsi/scsi_lib.c | 10 ++++++---- include/linux/scatterlist.h | 11 +++++++---- lib/scatterlist.c | 36 +++++++++++++++++++++++------------- lib/sg_pool.c | 37 +++++++++++++++++++++++++++---------- net/sunrpc/xprtrdma/svc_rdma_rw.c | 5 +++-- 8 files changed, 76 insertions(+), 41 deletions(-) diff --git a/drivers/nvme/host/fc.c b/drivers/nvme/host/fc.c index dd8169bbf0d2..46811caac9d2 100644 --- a/drivers/nvme/host/fc.c +++ b/drivers/nvme/host/fc.c @@ -2112,7 +2112,8 @@ nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq, freq->sg_table.sgl = freq->first_sgl; ret = sg_alloc_table_chained(&freq->sg_table, - blk_rq_nr_phys_segments(rq), freq->sg_table.sgl); + blk_rq_nr_phys_segments(rq), freq->sg_table.sgl, + SG_CHUNK_SIZE); if (ret) return -ENOMEM; @@ -2122,7 +2123,7 @@ nvme_fc_map_data(struct nvme_fc_ctrl *ctrl, struct request *rq, freq->sg_cnt = fc_dma_map_sg(ctrl->lport->dev, freq->sg_table.sgl, op->nents, dir); if (unlikely(freq->sg_cnt <= 0)) { - sg_free_table_chained(&freq->sg_table, true); + sg_free_table_chained(&freq->sg_table, SG_CHUNK_SIZE); freq->sg_cnt = 0; return -EFAULT; } @@ -2148,7 +2149,7 @@ nvme_fc_unmap_data(struct nvme_fc_ctrl *ctrl, struct request *rq, nvme_cleanup_cmd(rq); - sg_free_table_chained(&freq->sg_table, true); + sg_free_table_chained(&freq->sg_table, SG_CHUNK_SIZE); freq->sg_cnt = 0; } diff --git a/drivers/nvme/host/rdma.c b/drivers/nvme/host/rdma.c index f383146e7d0f..f7ea19b45798 100644 --- a/drivers/nvme/host/rdma.c +++ b/drivers/nvme/host/rdma.c @@ -1133,7 +1133,7 @@ static void nvme_rdma_unmap_data(struct nvme_rdma_queue *queue, WRITE ? DMA_TO_DEVICE : DMA_FROM_DEVICE); nvme_cleanup_cmd(rq); - sg_free_table_chained(&req->sg_table, true); + sg_free_table_chained(&req->sg_table, SG_CHUNK_SIZE); } static int nvme_rdma_set_sg_null(struct nvme_command *c) @@ -1248,7 +1248,8 @@ static int nvme_rdma_map_data(struct nvme_rdma_queue *queue, req->sg_table.sgl = req->first_sgl; ret = sg_alloc_table_chained(&req->sg_table, - blk_rq_nr_phys_segments(rq), req->sg_table.sgl); + blk_rq_nr_phys_segments(rq), req->sg_table.sgl, + SG_CHUNK_SIZE); if (ret) return -ENOMEM; @@ -1288,7 +1289,7 @@ out_unmap_sg: req->nents, rq_data_dir(rq) == WRITE ? DMA_TO_DEVICE : DMA_FROM_DEVICE); out_free_table: - sg_free_table_chained(&req->sg_table, true); + sg_free_table_chained(&req->sg_table, SG_CHUNK_SIZE); return ret; } diff --git a/drivers/nvme/target/loop.c b/drivers/nvme/target/loop.c index 9e211ad6bdd3..b16dc3981c69 100644 --- a/drivers/nvme/target/loop.c +++ b/drivers/nvme/target/loop.c @@ -77,7 +77,7 @@ static void nvme_loop_complete_rq(struct request *req) struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req); nvme_cleanup_cmd(req); - sg_free_table_chained(&iod->sg_table, true); + sg_free_table_chained(&iod->sg_table, SG_CHUNK_SIZE); nvme_complete_rq(req); } @@ -157,7 +157,7 @@ static blk_status_t nvme_loop_queue_rq(struct blk_mq_hw_ctx *hctx, iod->sg_table.sgl = iod->first_sgl; if (sg_alloc_table_chained(&iod->sg_table, blk_rq_nr_phys_segments(req), - iod->sg_table.sgl)) + iod->sg_table.sgl, SG_CHUNK_SIZE)) return BLK_STS_RESOURCE; iod->req.sg = iod->sg_table.sgl; diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index 0916bd6d22b0..acc0f7080f18 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -541,9 +541,9 @@ static void scsi_uninit_cmd(struct scsi_cmnd *cmd) static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd) { if (cmd->sdb.table.nents) - sg_free_table_chained(&cmd->sdb.table, true); + sg_free_table_chained(&cmd->sdb.table, SG_CHUNK_SIZE); if (scsi_prot_sg_count(cmd)) - sg_free_table_chained(&cmd->prot_sdb->table, true); + sg_free_table_chained(&cmd->prot_sdb->table, SG_CHUNK_SIZE); } static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd) @@ -976,7 +976,8 @@ static blk_status_t scsi_init_sgtable(struct request *req, * If sg table allocation fails, requeue request later. */ if (unlikely(sg_alloc_table_chained(&sdb->table, - blk_rq_nr_phys_segments(req), sdb->table.sgl))) + blk_rq_nr_phys_segments(req), sdb->table.sgl, + SG_CHUNK_SIZE))) return BLK_STS_RESOURCE; /* @@ -1030,7 +1031,8 @@ blk_status_t scsi_init_io(struct scsi_cmnd *cmd) ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio); if (sg_alloc_table_chained(&prot_sdb->table, ivecs, - prot_sdb->table.sgl)) { + prot_sdb->table.sgl, + SG_CHUNK_SIZE)) { ret = BLK_STS_RESOURCE; goto out_free_sgtables; } diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h index 30a9a55c28ba..6eec50fb36c8 100644 --- a/include/linux/scatterlist.h +++ b/include/linux/scatterlist.h @@ -266,10 +266,11 @@ int sg_split(struct scatterlist *in, const int in_mapped_nents, typedef struct scatterlist *(sg_alloc_fn)(unsigned int, gfp_t); typedef void (sg_free_fn)(struct scatterlist *, unsigned int); -void __sg_free_table(struct sg_table *, unsigned int, bool, sg_free_fn *); +void __sg_free_table(struct sg_table *, unsigned int, unsigned int, + sg_free_fn *); void sg_free_table(struct sg_table *); int __sg_alloc_table(struct sg_table *, unsigned int, unsigned int, - struct scatterlist *, gfp_t, sg_alloc_fn *); + struct scatterlist *, unsigned int, gfp_t, sg_alloc_fn *); int sg_alloc_table(struct sg_table *, unsigned int, gfp_t); int __sg_alloc_table_from_pages(struct sg_table *sgt, struct page **pages, unsigned int n_pages, unsigned int offset, @@ -331,9 +332,11 @@ size_t sg_zero_buffer(struct scatterlist *sgl, unsigned int nents, #endif #ifdef CONFIG_SG_POOL -void sg_free_table_chained(struct sg_table *table, bool first_chunk); +void sg_free_table_chained(struct sg_table *table, + unsigned nents_first_chunk); int sg_alloc_table_chained(struct sg_table *table, int nents, - struct scatterlist *first_chunk); + struct scatterlist *first_chunk, + unsigned nents_first_chunk); #endif /* diff --git a/lib/scatterlist.c b/lib/scatterlist.c index 739dc9fe2c55..77ec8eec3fd0 100644 --- a/lib/scatterlist.c +++ b/lib/scatterlist.c @@ -181,7 +181,8 @@ static void sg_kfree(struct scatterlist *sg, unsigned int nents) * __sg_free_table - Free a previously mapped sg table * @table: The sg table header to use * @max_ents: The maximum number of entries per single scatterlist - * @skip_first_chunk: don't free the (preallocated) first scatterlist chunk + * @nents_first_chunk: Number of entries int the (preallocated) first + * scatterlist chunk, 0 means no such preallocated first chunk * @free_fn: Free function * * Description: @@ -191,9 +192,10 @@ static void sg_kfree(struct scatterlist *sg, unsigned int nents) * **/ void __sg_free_table(struct sg_table *table, unsigned int max_ents, - bool skip_first_chunk, sg_free_fn *free_fn) + unsigned int nents_first_chunk, sg_free_fn *free_fn) { struct scatterlist *sgl, *next; + unsigned curr_max_ents = nents_first_chunk ?: max_ents; if (unlikely(!table->sgl)) return; @@ -209,9 +211,9 @@ void __sg_free_table(struct sg_table *table, unsigned int max_ents, * sg_size is then one less than alloc size, since the last * element is the chain pointer. */ - if (alloc_size > max_ents) { - next = sg_chain_ptr(&sgl[max_ents - 1]); - alloc_size = max_ents; + if (alloc_size > curr_max_ents) { + next = sg_chain_ptr(&sgl[curr_max_ents - 1]); + alloc_size = curr_max_ents; sg_size = alloc_size - 1; } else { sg_size = alloc_size; @@ -219,11 +221,12 @@ void __sg_free_table(struct sg_table *table, unsigned int max_ents, } table->orig_nents -= sg_size; - if (skip_first_chunk) - skip_first_chunk = false; + if (nents_first_chunk) + nents_first_chunk = 0; else free_fn(sgl, alloc_size); sgl = next; + curr_max_ents = max_ents; } table->sgl = NULL; @@ -246,6 +249,8 @@ EXPORT_SYMBOL(sg_free_table); * @table: The sg table header to use * @nents: Number of entries in sg list * @max_ents: The maximum number of entries the allocator returns per call + * @nents_first_chunk: Number of entries int the (preallocated) first + * scatterlist chunk, 0 means no such preallocated chunk provided by user * @gfp_mask: GFP allocation mask * @alloc_fn: Allocator to use * @@ -262,10 +267,13 @@ EXPORT_SYMBOL(sg_free_table); **/ int __sg_alloc_table(struct sg_table *table, unsigned int nents, unsigned int max_ents, struct scatterlist *first_chunk, - gfp_t gfp_mask, sg_alloc_fn *alloc_fn) + unsigned int nents_first_chunk, gfp_t gfp_mask, + sg_alloc_fn *alloc_fn) { struct scatterlist *sg, *prv; unsigned int left; + unsigned curr_max_ents = nents_first_chunk ?: max_ents; + unsigned prv_max_ents; memset(table, 0, sizeof(*table)); @@ -281,8 +289,8 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents, do { unsigned int sg_size, alloc_size = left; - if (alloc_size > max_ents) { - alloc_size = max_ents; + if (alloc_size > curr_max_ents) { + alloc_size = curr_max_ents; sg_size = alloc_size - 1; } else sg_size = alloc_size; @@ -316,7 +324,7 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents, * If this is not the first mapping, chain previous part. */ if (prv) - sg_chain(prv, max_ents, sg); + sg_chain(prv, prv_max_ents, sg); else table->sgl = sg; @@ -327,6 +335,8 @@ int __sg_alloc_table(struct sg_table *table, unsigned int nents, sg_mark_end(&sg[sg_size - 1]); prv = sg; + prv_max_ents = curr_max_ents; + curr_max_ents = max_ents; } while (left); return 0; @@ -349,9 +359,9 @@ int sg_alloc_table(struct sg_table *table, unsigned int nents, gfp_t gfp_mask) int ret; ret = __sg_alloc_table(table, nents, SG_MAX_SINGLE_ALLOC, - NULL, gfp_mask, sg_kmalloc); + NULL, 0, gfp_mask, sg_kmalloc); if (unlikely(ret)) - __sg_free_table(table, SG_MAX_SINGLE_ALLOC, false, sg_kfree); + __sg_free_table(table, SG_MAX_SINGLE_ALLOC, 0, sg_kfree); return ret; } diff --git a/lib/sg_pool.c b/lib/sg_pool.c index d1c1e6388eaa..b3b8cf62ff49 100644 --- a/lib/sg_pool.c +++ b/lib/sg_pool.c @@ -69,18 +69,27 @@ static struct scatterlist *sg_pool_alloc(unsigned int nents, gfp_t gfp_mask) /** * sg_free_table_chained - Free a previously mapped sg table * @table: The sg table header to use - * @first_chunk: was first_chunk not NULL in sg_alloc_table_chained? + * @nents_first_chunk: size of the first_chunk SGL passed to + * sg_alloc_table_chained * * Description: * Free an sg table previously allocated and setup with * sg_alloc_table_chained(). * + * @nents_first_chunk has to be same with that same parameter passed + * to sg_alloc_table_chained(). + * **/ -void sg_free_table_chained(struct sg_table *table, bool first_chunk) +void sg_free_table_chained(struct sg_table *table, + unsigned nents_first_chunk) { - if (first_chunk && table->orig_nents <= SG_CHUNK_SIZE) + if (table->orig_nents <= nents_first_chunk) return; - __sg_free_table(table, SG_CHUNK_SIZE, first_chunk, sg_pool_free); + + if (nents_first_chunk == 1) + nents_first_chunk = 0; + + __sg_free_table(table, SG_CHUNK_SIZE, nents_first_chunk, sg_pool_free); } EXPORT_SYMBOL_GPL(sg_free_table_chained); @@ -89,31 +98,39 @@ EXPORT_SYMBOL_GPL(sg_free_table_chained); * @table: The sg table header to use * @nents: Number of entries in sg list * @first_chunk: first SGL + * @nents_first_chunk: number of the SGL of @first_chunk * * Description: * Allocate and chain SGLs in an sg table. If @nents@ is larger than - * SG_CHUNK_SIZE a chained sg table will be setup. + * @nents_first_chunk a chained sg table will be setup. * **/ int sg_alloc_table_chained(struct sg_table *table, int nents, - struct scatterlist *first_chunk) + struct scatterlist *first_chunk, unsigned nents_first_chunk) { int ret; BUG_ON(!nents); - if (first_chunk) { - if (nents <= SG_CHUNK_SIZE) { + if (first_chunk && nents_first_chunk) { + if (nents <= nents_first_chunk) { table->nents = table->orig_nents = nents; sg_init_table(table->sgl, nents); return 0; } } + /* User supposes that the 1st SGL includes real entry */ + if (nents_first_chunk == 1) { + first_chunk = NULL; + nents_first_chunk = 0; + } + ret = __sg_alloc_table(table, nents, SG_CHUNK_SIZE, - first_chunk, GFP_ATOMIC, sg_pool_alloc); + first_chunk, nents_first_chunk, + GFP_ATOMIC, sg_pool_alloc); if (unlikely(ret)) - sg_free_table_chained(table, (bool)first_chunk); + sg_free_table_chained(table, nents_first_chunk); return ret; } EXPORT_SYMBOL_GPL(sg_alloc_table_chained); diff --git a/net/sunrpc/xprtrdma/svc_rdma_rw.c b/net/sunrpc/xprtrdma/svc_rdma_rw.c index 2121c9b4d275..48fe3b16b0d9 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_rw.c +++ b/net/sunrpc/xprtrdma/svc_rdma_rw.c @@ -73,7 +73,8 @@ svc_rdma_get_rw_ctxt(struct svcxprt_rdma *rdma, unsigned int sges) ctxt->rw_sg_table.sgl = ctxt->rw_first_sgl; if (sg_alloc_table_chained(&ctxt->rw_sg_table, sges, - ctxt->rw_sg_table.sgl)) { + ctxt->rw_sg_table.sgl, + SG_CHUNK_SIZE)) { kfree(ctxt); ctxt = NULL; } @@ -84,7 +85,7 @@ out: static void svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma, struct svc_rdma_rw_ctxt *ctxt) { - sg_free_table_chained(&ctxt->rw_sg_table, true); + sg_free_table_chained(&ctxt->rw_sg_table, SG_CHUNK_SIZE); spin_lock(&rdma->sc_rw_ctxt_lock); list_add(&ctxt->rw_list, &rdma->sc_rw_ctxts); -- cgit v1.2.3 From 92524fa12312d1f082a473e14c590c48b4ef3fe5 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Sun, 28 Apr 2019 15:39:31 +0800 Subject: scsi: core: avoid preallocating big SGL for protection information scsi_mq_setup_tags() currently preallocates a big buffer for protection SGL entries. scsi_mq_sgl_size() is used to determine the size for both data and protection information scatterlists but the protection buffer is usually much smaller. For example, one 512-byte sector needs 8 bytes of protection information. Given that the maximum number of sectors for one request is 2560 (BLK_DEF_MAX_SECTORS) sectors, the max protection information buffer size is just 20K. The protection information segment count generally matches the number of bios in the request. As a result, the typical actual number of segments won't be very big. And should the need arise, allocating a bigger SGL from slab is fast enough. Pre-allocate only one SGL entry for protection information and switch to runtime allocation in case that the protection information segment number is bigger than 1. This reduces memory tied up by static command allocations. For example, 500+ MB is saved on single lpfc HBA. [mkp: attempted to clarify commit desc] Cc: Christoph Hellwig Cc: Bart Van Assche Cc: Ewan D. Milne Cc: Hannes Reinecke Reviewed-by: Christoph Hellwig Signed-off-by: Ming Lei Reviewed-by: Bart Van Assche Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index acc0f7080f18..c115458c835e 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -39,6 +39,12 @@ #include "scsi_priv.h" #include "scsi_logging.h" +/* + * Size of integrity metadata is usually small, 1 inline sg should + * cover normal cases. + */ +#define SCSI_INLINE_PROT_SG_CNT 1 + static struct kmem_cache *scsi_sdb_cache; static struct kmem_cache *scsi_sense_cache; static struct kmem_cache *scsi_sense_isadma_cache; @@ -543,7 +549,8 @@ static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd) if (cmd->sdb.table.nents) sg_free_table_chained(&cmd->sdb.table, SG_CHUNK_SIZE); if (scsi_prot_sg_count(cmd)) - sg_free_table_chained(&cmd->prot_sdb->table, SG_CHUNK_SIZE); + sg_free_table_chained(&cmd->prot_sdb->table, + SCSI_INLINE_PROT_SG_CNT); } static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd) @@ -1032,7 +1039,7 @@ blk_status_t scsi_init_io(struct scsi_cmnd *cmd) if (sg_alloc_table_chained(&prot_sdb->table, ivecs, prot_sdb->table.sgl, - SG_CHUNK_SIZE)) { + SCSI_INLINE_PROT_SG_CNT)) { ret = BLK_STS_RESOURCE; goto out_free_sgtables; } @@ -1824,7 +1831,8 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) sgl_size = scsi_mq_sgl_size(shost); cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size; if (scsi_host_get_prot(shost)) - cmd_size += sizeof(struct scsi_data_buffer) + sgl_size; + cmd_size += sizeof(struct scsi_data_buffer) + + sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT; memset(&shost->tag_set, 0, sizeof(shost->tag_set)); shost->tag_set.ops = &scsi_mq_ops; -- cgit v1.2.3 From 3dccdf53c2f38399b11085ded4447ce1467f006c Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Sun, 28 Apr 2019 15:39:32 +0800 Subject: scsi: core: avoid preallocating big SGL for data scsi_mq_setup_tags() preallocates a big buffer for the IO SGL. The size is based on scsi_mq_sgl_size() which is determined based on shost->sg_tablesize and SG_CHUNK_SIZE. Modern DMA engines are often capable of dealing with very big segments so the resulting scsi_mq_sgl_size() is often too big. SG_CHUNK_SIZE results in a static 4KB SGL allocation per command. If an HBA has lots of deep queues, preallocation for the sg list can consume substantial amounts of memory. For lpfc, nr_hw_queues can be 70 and each queue's depth 3781. This means the resulting preallocation for the data SGL is 70*3781*2K = 517MB. Switch to runtime allocation for SGL for lists longer than 2 entries. This is the approach used by NVMe PCI so it should be reasonable for SCSI as well. Runtime SGL allocation has always been the case for the legacy I/O path so this is nothing new. [mkp: attempted to clarify commit desc] Cc: Christoph Hellwig Cc: Bart Van Assche Cc: Ewan D. Milne Cc: Hannes Reinecke Reviewed-by: Christoph Hellwig Signed-off-by: Ming Lei Reviewed-by: Bart Van Assche Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index c115458c835e..45b43e97505c 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -45,6 +45,8 @@ */ #define SCSI_INLINE_PROT_SG_CNT 1 +#define SCSI_INLINE_SG_CNT 2 + static struct kmem_cache *scsi_sdb_cache; static struct kmem_cache *scsi_sense_cache; static struct kmem_cache *scsi_sense_isadma_cache; @@ -547,7 +549,8 @@ static void scsi_uninit_cmd(struct scsi_cmnd *cmd) static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd) { if (cmd->sdb.table.nents) - sg_free_table_chained(&cmd->sdb.table, SG_CHUNK_SIZE); + sg_free_table_chained(&cmd->sdb.table, + SCSI_INLINE_SG_CNT); if (scsi_prot_sg_count(cmd)) sg_free_table_chained(&cmd->prot_sdb->table, SCSI_INLINE_PROT_SG_CNT); @@ -984,7 +987,7 @@ static blk_status_t scsi_init_sgtable(struct request *req, */ if (unlikely(sg_alloc_table_chained(&sdb->table, blk_rq_nr_phys_segments(req), sdb->table.sgl, - SG_CHUNK_SIZE))) + SCSI_INLINE_SG_CNT))) return BLK_STS_RESOURCE; /* @@ -1550,9 +1553,9 @@ static int scsi_dispatch_cmd(struct scsi_cmnd *cmd) } /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */ -static unsigned int scsi_mq_sgl_size(struct Scsi_Host *shost) +static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost) { - return min_t(unsigned int, shost->sg_tablesize, SG_CHUNK_SIZE) * + return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) * sizeof(struct scatterlist); } @@ -1734,7 +1737,7 @@ static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq, if (scsi_host_get_prot(shost)) { sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size; - cmd->prot_sdb = (void *)sg + scsi_mq_sgl_size(shost); + cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost); } return 0; @@ -1828,7 +1831,7 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) { unsigned int cmd_size, sgl_size; - sgl_size = scsi_mq_sgl_size(shost); + sgl_size = scsi_mq_inline_sgl_size(shost); cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size; if (scsi_host_get_prot(shost)) cmd_size += sizeof(struct scsi_data_buffer) + -- cgit v1.2.3 From b79d9a09ae23c7047bdce3a15e284398334198ea Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Thu, 6 Jun 2019 16:34:08 +0800 Subject: scsi: lib/sg_pool.c: clear 'first_chunk' in case of no preallocation If user doesn't ask to preallocate by passing zero 'nents_first_chunk' to sg_alloc_table_chained, we need to make sure that 'first_chunk' is cleared. Otherwise, __sg_alloc_table() still may think that the 1st SGL should be from the preallocation. Fixes the issue by clearing 'first_chunk' in sg_alloc_table_chained() if 'nents_first_chunk' is zero. Cc: Christoph Hellwig Cc: Bart Van Assche Cc: Ewan D. Milne Cc: Hannes Reinecke Cc: Guenter Roeck Reported-by: Guenter Roeck Tested-by: Guenter Roeck Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- lib/sg_pool.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/lib/sg_pool.c b/lib/sg_pool.c index b3b8cf62ff49..f1cc8372df67 100644 --- a/lib/sg_pool.c +++ b/lib/sg_pool.c @@ -102,7 +102,9 @@ EXPORT_SYMBOL_GPL(sg_free_table_chained); * * Description: * Allocate and chain SGLs in an sg table. If @nents@ is larger than - * @nents_first_chunk a chained sg table will be setup. + * @nents_first_chunk a chained sg table will be setup. @first_chunk is + * ignored if nents_first_chunk <= 1 because user expects the SGL points + * non-chain SGL. * **/ int sg_alloc_table_chained(struct sg_table *table, int nents, @@ -121,7 +123,7 @@ int sg_alloc_table_chained(struct sg_table *table, int nents, } /* User supposes that the 1st SGL includes real entry */ - if (nents_first_chunk == 1) { + if (nents_first_chunk <= 1) { first_chunk = NULL; nents_first_chunk = 0; } -- cgit v1.2.3 From 3e99b3b13a1fc8f7354edaee4c04f73a07faba69 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Thu, 6 Jun 2019 16:34:09 +0800 Subject: scsi: core: don't preallocate small SGL in case of NO_SG_CHAIN The preallocated small SGL depends on SG_CHAIN so if the ARCH doesn't support SG_CHAIN, preallocation of small SGL can't work at all. Fix this issue by not using small preallocation in case of NO_SG_CHAIN. Cc: Christoph Hellwig Cc: Bart Van Assche Cc: Ewan D. Milne Cc: Hannes Reinecke Cc: Guenter Roeck Reported-by: Guenter Roeck Reviewed-by: Christoph Hellwig Reviewed-by: Bart Van Assche Tested-by: Guenter Roeck Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index 45b43e97505c..30cbea57ea08 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -43,9 +43,13 @@ * Size of integrity metadata is usually small, 1 inline sg should * cover normal cases. */ +#ifdef CONFIG_ARCH_NO_SG_CHAIN +#define SCSI_INLINE_PROT_SG_CNT 0 +#define SCSI_INLINE_SG_CNT 0 +#else #define SCSI_INLINE_PROT_SG_CNT 1 - #define SCSI_INLINE_SG_CNT 2 +#endif static struct kmem_cache *scsi_sdb_cache; static struct kmem_cache *scsi_sense_cache; -- cgit v1.2.3