qed: Revisit chain implementation
RoCE driver is going to need a 32-bit chain [current chain implementation
for qed* currently supports only 16-bit producer/consumer chains].
This patch adds said support, as well as doing other slight tweaks and
modifications to qed's chain API.
Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index e9ce6a7..151173d 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -17,6 +17,7 @@
#include <linux/pci.h>
#include <linux/slab.h>
#include <linux/string.h>
+#include <linux/vmalloc.h>
#include <linux/etherdevice.h>
#include <linux/qed/qed_chain.h>
#include <linux/qed/qed_if.h>
@@ -1779,92 +1780,285 @@
qed_iov_free_hw_info(cdev);
}
+static void qed_chain_free_next_ptr(struct qed_dev *cdev,
+ struct qed_chain *p_chain)
+{
+ void *p_virt = p_chain->p_virt_addr, *p_virt_next = NULL;
+ dma_addr_t p_phys = p_chain->p_phys_addr, p_phys_next = 0;
+ struct qed_chain_next *p_next;
+ u32 size, i;
+
+ if (!p_virt)
+ return;
+
+ size = p_chain->elem_size * p_chain->usable_per_page;
+
+ for (i = 0; i < p_chain->page_cnt; i++) {
+ if (!p_virt)
+ break;
+
+ p_next = (struct qed_chain_next *)((u8 *)p_virt + size);
+ p_virt_next = p_next->next_virt;
+ p_phys_next = HILO_DMA_REGPAIR(p_next->next_phys);
+
+ dma_free_coherent(&cdev->pdev->dev,
+ QED_CHAIN_PAGE_SIZE, p_virt, p_phys);
+
+ p_virt = p_virt_next;
+ p_phys = p_phys_next;
+ }
+}
+
+static void qed_chain_free_single(struct qed_dev *cdev,
+ struct qed_chain *p_chain)
+{
+ if (!p_chain->p_virt_addr)
+ return;
+
+ dma_free_coherent(&cdev->pdev->dev,
+ QED_CHAIN_PAGE_SIZE,
+ p_chain->p_virt_addr, p_chain->p_phys_addr);
+}
+
+static void qed_chain_free_pbl(struct qed_dev *cdev, struct qed_chain *p_chain)
+{
+ void **pp_virt_addr_tbl = p_chain->pbl.pp_virt_addr_tbl;
+ u32 page_cnt = p_chain->page_cnt, i, pbl_size;
+ u8 *p_pbl_virt = p_chain->pbl.p_virt_table;
+
+ if (!pp_virt_addr_tbl)
+ return;
+
+ if (!p_chain->pbl.p_virt_table)
+ goto out;
+
+ for (i = 0; i < page_cnt; i++) {
+ if (!pp_virt_addr_tbl[i])
+ break;
+
+ dma_free_coherent(&cdev->pdev->dev,
+ QED_CHAIN_PAGE_SIZE,
+ pp_virt_addr_tbl[i],
+ *(dma_addr_t *)p_pbl_virt);
+
+ p_pbl_virt += QED_CHAIN_PBL_ENTRY_SIZE;
+ }
+
+ pbl_size = page_cnt * QED_CHAIN_PBL_ENTRY_SIZE;
+ dma_free_coherent(&cdev->pdev->dev,
+ pbl_size,
+ p_chain->pbl.p_virt_table, p_chain->pbl.p_phys_table);
+out:
+ vfree(p_chain->pbl.pp_virt_addr_tbl);
+}
+
+void qed_chain_free(struct qed_dev *cdev, struct qed_chain *p_chain)
+{
+ switch (p_chain->mode) {
+ case QED_CHAIN_MODE_NEXT_PTR:
+ qed_chain_free_next_ptr(cdev, p_chain);
+ break;
+ case QED_CHAIN_MODE_SINGLE:
+ qed_chain_free_single(cdev, p_chain);
+ break;
+ case QED_CHAIN_MODE_PBL:
+ qed_chain_free_pbl(cdev, p_chain);
+ break;
+ }
+}
+
+static int
+qed_chain_alloc_sanity_check(struct qed_dev *cdev,
+ enum qed_chain_cnt_type cnt_type,
+ size_t elem_size, u32 page_cnt)
+{
+ u64 chain_size = ELEMS_PER_PAGE(elem_size) * page_cnt;
+
+ /* The actual chain size can be larger than the maximal possible value
+ * after rounding up the requested elements number to pages, and after
+ * taking into acount the unusuable elements (next-ptr elements).
+ * The size of a "u16" chain can be (U16_MAX + 1) since the chain
+ * size/capacity fields are of a u32 type.
+ */
+ if ((cnt_type == QED_CHAIN_CNT_TYPE_U16 &&
+ chain_size > 0x10000) ||
+ (cnt_type == QED_CHAIN_CNT_TYPE_U32 &&
+ chain_size > 0x100000000ULL)) {
+ DP_NOTICE(cdev,
+ "The actual chain size (0x%llx) is larger than the maximal possible value\n",
+ chain_size);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int
+qed_chain_alloc_next_ptr(struct qed_dev *cdev, struct qed_chain *p_chain)
+{
+ void *p_virt = NULL, *p_virt_prev = NULL;
+ dma_addr_t p_phys = 0;
+ u32 i;
+
+ for (i = 0; i < p_chain->page_cnt; i++) {
+ p_virt = dma_alloc_coherent(&cdev->pdev->dev,
+ QED_CHAIN_PAGE_SIZE,
+ &p_phys, GFP_KERNEL);
+ if (!p_virt) {
+ DP_NOTICE(cdev, "Failed to allocate chain memory\n");
+ return -ENOMEM;
+ }
+
+ if (i == 0) {
+ qed_chain_init_mem(p_chain, p_virt, p_phys);
+ qed_chain_reset(p_chain);
+ } else {
+ qed_chain_init_next_ptr_elem(p_chain, p_virt_prev,
+ p_virt, p_phys);
+ }
+
+ p_virt_prev = p_virt;
+ }
+ /* Last page's next element should point to the beginning of the
+ * chain.
+ */
+ qed_chain_init_next_ptr_elem(p_chain, p_virt_prev,
+ p_chain->p_virt_addr,
+ p_chain->p_phys_addr);
+
+ return 0;
+}
+
+static int
+qed_chain_alloc_single(struct qed_dev *cdev, struct qed_chain *p_chain)
+{
+ dma_addr_t p_phys = 0;
+ void *p_virt = NULL;
+
+ p_virt = dma_alloc_coherent(&cdev->pdev->dev,
+ QED_CHAIN_PAGE_SIZE, &p_phys, GFP_KERNEL);
+ if (!p_virt) {
+ DP_NOTICE(cdev, "Failed to allocate chain memory\n");
+ return -ENOMEM;
+ }
+
+ qed_chain_init_mem(p_chain, p_virt, p_phys);
+ qed_chain_reset(p_chain);
+
+ return 0;
+}
+
+static int qed_chain_alloc_pbl(struct qed_dev *cdev, struct qed_chain *p_chain)
+{
+ u32 page_cnt = p_chain->page_cnt, size, i;
+ dma_addr_t p_phys = 0, p_pbl_phys = 0;
+ void **pp_virt_addr_tbl = NULL;
+ u8 *p_pbl_virt = NULL;
+ void *p_virt = NULL;
+
+ size = page_cnt * sizeof(*pp_virt_addr_tbl);
+ pp_virt_addr_tbl = vmalloc(size);
+ if (!pp_virt_addr_tbl) {
+ DP_NOTICE(cdev,
+ "Failed to allocate memory for the chain virtual addresses table\n");
+ return -ENOMEM;
+ }
+ memset(pp_virt_addr_tbl, 0, size);
+
+ /* The allocation of the PBL table is done with its full size, since it
+ * is expected to be successive.
+ * qed_chain_init_pbl_mem() is called even in a case of an allocation
+ * failure, since pp_virt_addr_tbl was previously allocated, and it
+ * should be saved to allow its freeing during the error flow.
+ */
+ size = page_cnt * QED_CHAIN_PBL_ENTRY_SIZE;
+ p_pbl_virt = dma_alloc_coherent(&cdev->pdev->dev,
+ size, &p_pbl_phys, GFP_KERNEL);
+ qed_chain_init_pbl_mem(p_chain, p_pbl_virt, p_pbl_phys,
+ pp_virt_addr_tbl);
+ if (!p_pbl_virt) {
+ DP_NOTICE(cdev, "Failed to allocate chain pbl memory\n");
+ return -ENOMEM;
+ }
+
+ for (i = 0; i < page_cnt; i++) {
+ p_virt = dma_alloc_coherent(&cdev->pdev->dev,
+ QED_CHAIN_PAGE_SIZE,
+ &p_phys, GFP_KERNEL);
+ if (!p_virt) {
+ DP_NOTICE(cdev, "Failed to allocate chain memory\n");
+ return -ENOMEM;
+ }
+
+ if (i == 0) {
+ qed_chain_init_mem(p_chain, p_virt, p_phys);
+ qed_chain_reset(p_chain);
+ }
+
+ /* Fill the PBL table with the physical address of the page */
+ *(dma_addr_t *)p_pbl_virt = p_phys;
+ /* Keep the virtual address of the page */
+ p_chain->pbl.pp_virt_addr_tbl[i] = p_virt;
+
+ p_pbl_virt += QED_CHAIN_PBL_ENTRY_SIZE;
+ }
+
+ return 0;
+}
+
int qed_chain_alloc(struct qed_dev *cdev,
enum qed_chain_use_mode intended_use,
enum qed_chain_mode mode,
- u16 num_elems,
- size_t elem_size,
- struct qed_chain *p_chain)
+ enum qed_chain_cnt_type cnt_type,
+ u32 num_elems, size_t elem_size, struct qed_chain *p_chain)
{
- dma_addr_t p_pbl_phys = 0;
- void *p_pbl_virt = NULL;
- dma_addr_t p_phys = 0;
- void *p_virt = NULL;
- u16 page_cnt = 0;
- size_t size;
+ u32 page_cnt;
+ int rc = 0;
if (mode == QED_CHAIN_MODE_SINGLE)
page_cnt = 1;
else
page_cnt = QED_CHAIN_PAGE_CNT(num_elems, elem_size, mode);
- size = page_cnt * QED_CHAIN_PAGE_SIZE;
- p_virt = dma_alloc_coherent(&cdev->pdev->dev,
- size, &p_phys, GFP_KERNEL);
- if (!p_virt) {
- DP_NOTICE(cdev, "Failed to allocate chain mem\n");
+ rc = qed_chain_alloc_sanity_check(cdev, cnt_type, elem_size, page_cnt);
+ if (rc) {
+ DP_NOTICE(cdev,
+ "Cannot allocate a chain with the given arguments:\n"
+ "[use_mode %d, mode %d, cnt_type %d, num_elems %d, elem_size %zu]\n",
+ intended_use, mode, cnt_type, num_elems, elem_size);
+ return rc;
+ }
+
+ qed_chain_init_params(p_chain, page_cnt, (u8) elem_size, intended_use,
+ mode, cnt_type);
+
+ switch (mode) {
+ case QED_CHAIN_MODE_NEXT_PTR:
+ rc = qed_chain_alloc_next_ptr(cdev, p_chain);
+ break;
+ case QED_CHAIN_MODE_SINGLE:
+ rc = qed_chain_alloc_single(cdev, p_chain);
+ break;
+ case QED_CHAIN_MODE_PBL:
+ rc = qed_chain_alloc_pbl(cdev, p_chain);
+ break;
+ }
+ if (rc)
goto nomem;
- }
-
- if (mode == QED_CHAIN_MODE_PBL) {
- size = page_cnt * QED_CHAIN_PBL_ENTRY_SIZE;
- p_pbl_virt = dma_alloc_coherent(&cdev->pdev->dev,
- size, &p_pbl_phys,
- GFP_KERNEL);
- if (!p_pbl_virt) {
- DP_NOTICE(cdev, "Failed to allocate chain pbl mem\n");
- goto nomem;
- }
-
- qed_chain_pbl_init(p_chain, p_virt, p_phys, page_cnt,
- (u8)elem_size, intended_use,
- p_pbl_phys, p_pbl_virt);
- } else {
- qed_chain_init(p_chain, p_virt, p_phys, page_cnt,
- (u8)elem_size, intended_use, mode);
- }
return 0;
nomem:
- dma_free_coherent(&cdev->pdev->dev,
- page_cnt * QED_CHAIN_PAGE_SIZE,
- p_virt, p_phys);
- dma_free_coherent(&cdev->pdev->dev,
- page_cnt * QED_CHAIN_PBL_ENTRY_SIZE,
- p_pbl_virt, p_pbl_phys);
-
- return -ENOMEM;
+ qed_chain_free(cdev, p_chain);
+ return rc;
}
-void qed_chain_free(struct qed_dev *cdev,
- struct qed_chain *p_chain)
-{
- size_t size;
-
- if (!p_chain->p_virt_addr)
- return;
-
- if (p_chain->mode == QED_CHAIN_MODE_PBL) {
- size = p_chain->page_cnt * QED_CHAIN_PBL_ENTRY_SIZE;
- dma_free_coherent(&cdev->pdev->dev, size,
- p_chain->pbl.p_virt_table,
- p_chain->pbl.p_phys_table);
- }
-
- size = p_chain->page_cnt * QED_CHAIN_PAGE_SIZE;
- dma_free_coherent(&cdev->pdev->dev, size,
- p_chain->p_virt_addr,
- p_chain->p_phys_addr);
-}
-
-int qed_fw_l2_queue(struct qed_hwfn *p_hwfn,
- u16 src_id, u16 *dst_id)
+int qed_fw_l2_queue(struct qed_hwfn *p_hwfn, u16 src_id, u16 *dst_id)
{
if (src_id >= RESC_NUM(p_hwfn, QED_L2_QUEUE)) {
u16 min, max;
- min = (u16)RESC_START(p_hwfn, QED_L2_QUEUE);
+ min = (u16) RESC_START(p_hwfn, QED_L2_QUEUE);
max = min + RESC_NUM(p_hwfn, QED_L2_QUEUE);
DP_NOTICE(p_hwfn,
"l2_queue id [%d] is not valid, available indices [%d - %d]\n",